DOCTORAT DE L’UNIVERSITÉ BORDEAUX 2 · Université Victor Segalen Bordeaux 2 Année 2010 Thèse n°1 724 THÈSE pour le DOCTORAT DE L’UNIVERSITÉ BORDEAUX 2 Mention : Science,

Université Victor Segalen Bordeaux 2

Année 2010

Thèse n°1 724

THÈSE

pour le

DOCTORAT DE L’UNIVERSITÉ BORDEAUX 2

Mention : Science, Technologies, Santé

Option : Epidémiologie et Santé Publique

Présentée et soutenue publiquement

Le 27 septembre 2010 Par Junaid Ahmad BHATTI Né(e) le 15/01/1980 à Rawalpindi (Pakistan)

Les facteurs environnementaux dans les accidents de la circulation sur des routes interurbaines dans les pays en

développement

Situational factors involved in traffic crashes on interurban roads in developing countries

Membres du Jury Monsieur le Pr. Pierre PHILIP ................................................................Président du jury Monsieur le Dr. Pierre VAN ELSLANDE..............................................Rapporteur Madame le Pr. María SEGUÍ-GÓMEZ...................................................Rapporteur Monsieur le Dr. Junaid Abdul RAZZAK................................................Examinateur Monsieur le Pr. Louis-Rachid SALMI....................................................Directeur

2

3

Contents Abbreviations ............................................................................................................................. 8 Résumé....................................................................................................................................... 9 Abstract .................................................................................................................................... 10 1. Introduction .......................................................................................................................... 11 1. Introduction .......................................................................................................................... 13 2. Background .......................................................................................................................... 14

2.1 Road injury burden in LMICs ........................................................................................ 14 2.2 Road safety is a key to development in LMICs ............................................................. 14 2.3 Risk factors..................................................................................................................... 15 2.4 Multiple factors involved in traffic crashes.................................................................... 16 2.5 Implications of interactions for highway safety............................................................. 16 2.6 Factors limiting implementation of engineering measures in LMICs............................ 17 2.7 Interurban road safety research gaps in LMICs ............................................................. 20

3. Objectives............................................................................................................................. 20 4. Descriptive studies ............................................................................................................... 21

4.1 Study I: Traffic crash and injury burden on Yaoundé-Douala road section, Cameroon 21 Objectives......................................................................................................................... 22 Methods............................................................................................................................ 22 Results .............................................................................................................................. 24

4.2 Study II: Differences in police, ambulance, and emergency department reporting of traffic injuries on Karachi-Hala road section, Pakistan........................................................ 29

Objectives......................................................................................................................... 29 Methods............................................................................................................................ 30 Results .............................................................................................................................. 31

5. Analytical Studies ................................................................................................................ 36 5.1 Study III: Situational factors at traffic crash sites: a case-control study on Yaoundé-Douala road section, Cameroon ........................................................................................... 36


5.2 Study IV: Burden and factors associated with highway work zone crashes, Karachi-Hala road section, Pakistan .................................................................................................. 41


5.3 Study V: Road hazard perception at high-risk crash sites in voluntary Pakistani drivers.............................................................................................................................................. 47


6. Discussion ............................................................................................................................ 56 6.1 Originality of studies...................................................................................................... 56 6.2 Comparison with published literature ............................................................................ 56

7. Limitations and perspectives................................................................................................ 61 8. Conclusion............................................................................................................................ 62 References ................................................................................................................................ 63

4

Publications (peer-reviewed).................................................................................................... 71 Related to thesis ................................................................................................................... 71 Other articles related to traffic injuries................................................................................. 71 Related to other injuries ....................................................................................................... 72

Appendices ............................................................................................................................... 73

5

Index of tables Table 1. The Haddon Matrix applied to highway crashes involving speed (examples of intervention) ............................................................................................................................. 15 Table 2. Traffic fatalities according to road network type in France, 2004............................. 17 Table 3 Burden of road traffic crashes and injuries, according to vehicle type on Yaoundé-Douala road section (2004-2007)............................................................................................. 24 Table 4. Crash types, causes, and situational factors on the Yaoundé-Douala road section (2004-2007).............................................................................................................................. 27 Table 5. Traffic injury outcome, according to road-user characteristics, on the Yaoundé-Douala road section (2004-2007)............................................................................................. 28 Table 6. Traffic injuries reported to police, ambulance, and emergency department on Karachi-Hala road section (2008). ........................................................................................... 34 Table 7. Differences in outcome of traffic injury among police, ambulance, and emergency department for same patient on Karachi-Hala road section, 2008 (N=108) ............................ 35 Table 8. Ascertainment of police, ambulance, and emergency department records for traffic fatalities and injuries on Karachi-Hala road section (N=1 214)............................................... 35 Table 9. Situational variables at case and control sites on Yaoundé-Douala road section, Cameroon ................................................................................................................................. 40 Table 10. Road crash fatality and injury risk per 109 vehicle-km on the Karachi-Hala road section, Pakistan (2006-08) ...................................................................................................... 45 Table 11. Highway work zone crash fatality and injury risk per 109 vehicle-km on 50-km long sub-section on Karachi-Hala road, Pakistan (2006-08) ........................................................... 45 Table 12. Factors associated with work-zone crashes on the 196-km-long Karachi-Hala road section, Pakistan (2006-08) ...................................................................................................... 46 Table 13. Characteristics of high- and low-risk sites on Yaoundé-Douala and Karachi-Hala road sections............................................................................................................................. 52 Table 14. Characteristics of Pakistani drivers included in sample (N=100). ........................... 53 Table 15. Differences in hazard perception, and reported preferred speeds for high- and low-risk site pairs on Yaoundé-Douala and Karachi-Hala road sections........................................ 54 Table 16. Factors associated with hazard perception of high- and low-risk sites on Yaoundé-Douala and Karachi-Hala road sections ................................................................................... 55 Table 17. Analytical studies of traffic crash and injury risk on interurban road sections in developing countries. ............................................................................................................... 79 Table 18. Traffic injury intervention studies on interurban road sections in developing countries. .................................................................................................................................. 80 Table 19. Traffic fatalities and injuries according to crash types and causes on Yaoundé-Douala road section (2004-3007)............................................................................................. 87 Table 20. Driver age, sex, and vehicles driven on Karachi-Hala road section (July 2009) ... 137 Table 21. Situational factors at high- and low- risk site pairs on Yaoundé-Douala and Karachi-Hala road sections .................................................................................................................. 138 Table 22. Driver-related factors associated with hazard perception of sites on Yaoundé-Douala and Karachi-Hala road sections ................................................................................. 139 Table 23. Situational factors associated with hazard perception of sites on Yaoundé-Douala and Karachi-Hala road sections.............................................................................................. 140

6

Index of figures Figure 1. Traffic fatality per 100 000 inhabitants in various countries [2] .............................. 14 Figure 2. Traffic fatalities according to road user groups in different countries...................... 15 Figure 3. Contribution of risk factors in road traffic crashes (adapted from [55]) .................. 16 Figure 4. Percentage difference of crash fatalities between official reported and estimated figures (adapted from WHO, 2009 [2]).................................................................................... 18 Figure 5. Causes of road crashes as determined by the police in developing countries (adapted from Wootton and Jacobs 1996 [7])......................................................................................... 19 Figure 6. Traffic injury outcome, according to road user group on Yaoundé-Douala road section (2004-2007).................................................................................................................. 25 Figure 7. Monthly trend of traffic fatalities and injuries on the Yaoundé-Douala road section (Jan 2004 to May 2007) ........................................................................................................... 26 Figure 8. Month-wise police, ambulance, and emergency department reporting of traffic injuries on Karachi-Hala road section (2008). ......................................................................... 32 Figure 9. Outcome of traffic injuries reported to emergency department according to New Injury Severity Score (NISS) on Karachi-Hala road section (2008)........................................ 33 Figure 10. Traffic injuries reported to police, ambulance service, and emergency departments on Karachi-Hala road section in 2008 (N=1 214) .................................................................... 35 Figure 11. Injury crash site density along 25-km stretches of Yaoundé-Douala road section. 39 Figure 12. Karachi-Hala Road Section, province of Sindh, Pakistan ...................................... 42 Figure 13. Examples of normal traffic zone (A) and work zone (B) on interurban road section in the province of Sindh, Pakistan............................................................................................ 43 Figure 14. Picture extracted of a high-risk site video and related questions, from the Karachi-Hala road section...................................................................................................................... 50 Figure 15. Literature available for assessing research needs on interurban traffic safety (1995-2009)......................................................................................................................................... 75 Figure 16. Weekly pattern of traffic fatalities and injuries on Yaoundé-Douala road section (2004-2007).............................................................................................................................. 86 Figure 17. Hourly pattern of traffic crashes and fatalities on Yaoundé-Douala road section (2004-2007).............................................................................................................................. 86

7

Acknowledgements I am thankful to Allah for the courage he has bestowed upon me; to my parents for being role models; to my wife for her unconditional love and affection; to my daughter for giving my life a sense; to my teachers; and friends for their support to complete this work This thesis was not possible without the guidance and attention that I had received from my supervisor, Pr. L. Rachid SALMI. His keen interest and patience throughout my training in Masters and then in PhD were invaluable contribution to this whole work. I am grateful to Dr. Emmanuel LAGARDE, who had played a very vital role in this thesis and helped us advance this work when it was not that evident. I express my gratitude to Dr. Junaid A. RAZZAK, his encouragement in the field helped me endure the difficult work conditions and to come up with useful questions and data for this thesis. In France, I would like to thank all the members of the research team PPCT, in particular Aymery CONSTANT, Benjamin CONTRAND, and Ludivine ORRIOLS. At ISPED, I am grateful to all the teachers and in particular to Mme. Marthe-Aline JUTAND and Pr. Ahmadou ALIOUM for their guidance. Special thanks to Dr. Jean-François TESSIER for all the support he has given during our stay in France. I am also grateful to my colleagues in the Master’s program, in particular Mohammad BERRAHO (Morocco) for helping me with my studies. I am also thankful to Mr. Zaheer SATTI (Paris) to help us during stay at different occasions. In Pakistan, I would like to thank Dr. Aftab PATHAN (NHMP), Mr. Irshad SODHAR (NHMP), Mr. Naeem-ul-lah SHIEKH (NHMP), Eng. Ali Bin Usman SHAH (NHA), Mr. Ameer HUSSAIN (RTIRP, JPMC), Mr. Faisal EDHI (Edhi foundation), Mr. Javed SHAH (AKU), Dr. Sanaullah BASHIR (DUHS), and Dr. Kiran EJAZ (AKU) for their support in data collection. Special thanks to my Uncle and Aunt in Karachi for helping me during my stay. I would like to acknowledge Dr. Jöelle SOBNGWI and other co-investigators, for their support in completing road safety studies in Cameroon. This thesis was completed with the financial support of the Higher Education Commission of Pakistan. I am also grateful to Pr. Georges PIERRON (SFERE) and Mme France LAMISCARRE at SFERE (Paris) for their support during my studies. I would like to thank all the drivers who participated in the hazard perception study and the owners of transport agencies who provided us the space to conduct the interviews.

8

Abbreviations AKU Aga Khan University AVCI Années de Vie Corrigées du facteur Incapacité BMI Body Mass Index EAS Edhi Ambulance Service ESS Epworth Sleepiness Scale DALY Disability-Adjusted Life Year DWI Driving While Intoxicated ED Emergency Department GDP Gross Domestic Product GPS Global Positioning System HIC High Income Country HWZ Highway Work Zones IRF International Road Federation JPMC Jinnah Post Graduate Medical Centre LMIC Low- and Middle-Income Country NHA National Highway Authority, Pakistan NHMP National Highway and Motorway Police, Pakistan NISS New Injury Severity Score OR Odds Ratio PIB Produit Intérieur Brut PKR Pakistani Rupee PRBM Pays à Revenu Bas et Moyen PRE Pays à Revenu Élevé RR Relative Risk RTC Road Traffic Crash RTI Road Traffic Injury RTIRP Road Traffic Injury Research & Prevention Centre USA United States of America US$ US Dollar WHO World Health Organization

9

Résumé Introduction : La sécurité routière sur le réseau interurbain est un problème majeur de santé publique dans les Pays à Revenu Bas et Moyen (PRBM) mais peu d'attention y a été consacrée. Les objectifs de cette thèse étaient d’évaluer le fardeau des traumatismes en relation avec le trafic interurbain, la déclaration des usagers blessés dans des bases de données différentes, d’analyser l’association entre les facteurs situationnels (caractéristiques physiques et circonstances environnementales) et les sites des accidents et la perception de la dangerosité des tronçons accidentogènes dans les PRBM. Méthodes et résultats : Pour répondre à ces objectifs, cinq études spécifiques ont été réalisées dans deux PRBM, le Cameroun et le Pakistan. L’étude I a évalué le nombre de tués par véhicules-km parcourus et les facteurs qui leur étaient associés, en utilisant les rapports de police entre 2004 et 2007 sur l’axe Yaoundé-Douala, Cameroun. Le taux de mortalité était de 73 par 100 millions véhicules km parcourus, un taux 35 fois plus élevé que sur un même type de route en pays à revenu élevé. La mortalité était plus élevée pour les accidents impliquant des usagers vulnérables, les véhicules roulant en sens opposé et ceux dus à une défaillance mécanique, y compris un éclatement de pneu. L’étude II a évalué les différences de déclaration d’accidents faites par les services de police, d’ambulance et des urgences en 2008 sur l’axe Karachi-Hala, Pakistan. La mortalité était de 53 par 109 véhicules-km parcourus ; le taux de mortalité était 13 fois plus élevé sur cet axe par rapport à un même type de route en France. La police a déclaré un mort sur cinq et un blessé grave sur dix. Les usagers de la route vulnérables, y compris les piétons et deux-roues ont été deux fois moins déclarés par la police que par les services d'ambulance ou des urgences. L’étude III a étudié les facteurs situationnels associés aux sites des accidents sur l’axe Yaoundé-Douala par une approche de type cas-témoins. Les facteurs tels que le profil routier plat (rapport de cotes [RC] ajusté =1,52 ; intervalle de confiance à 95 % [IC95 %]=1,15-2,04), les surfaces irrégulières (RC=1,43 ; IC95 %=1,04-1,99), les obstacles à proximité (RC=1,99 ; IC95 %=1,09-3,63) et les intersections à trois (RC=3,11 ; IC95 %=1,15-8,39) ou à quatre directions (RC=3,23 ; IC95 %=1,51-6,92) étaient significativement associés à des sites d’accidents corporels. De plus, la probabilité des accidents augmentait dans des zones urbaines situées dans des régions de plaine (RC=2,23 ; IC95 %=1,97-2,77). L’étude IV a étudié le fardeau des traumatismes dus aux accidents ainsi que les facteurs associés dans des zones en travaux sur l’axe Karachi-Hala en utilisant les méthodes de cohorte historique. Un tiers de la mortalité routière était survenu dans des zones en travaux et le risque de mortalité était quatre fois plus élevé dans ces zones que dans les autres zones. Un accident sur deux a eu lieu entre des véhicules roulant en sens opposé dans ces zones. L’étude V a étudié la perception de la dangerosité des tronçons accidentogènes (au moins 3 accidents sur 3 ans) et non accidentogènes (aucun accident déclaré) sur les deux axes des précédentes études, en montrant leurs vidéos à des conducteurs volontaires pakistanais. Les conducteurs n’ont perçu comme dangereux que la moitié des tronçons accidentogènes. La perception de la dangerosité des tronçons plats et droits était plus faible par rapport aux tronçons en courbes et avec une pente. La perception de la dangerosité en zone urbaine d’un tronçon accidentogène était significativement moins élevée (RC=0,58 ; IC95 %=0,51-0,68) que celle d’un tronçon non accidentogène ayant la même caractéristique (RC=2,04 ; IC95 %=1,51-2,74). La perception de la dangerosité d’un tronçon accidentogène avec panneau de signalisation était significativement plus élevée (RC=2,75 ; IC95 %=2,38-3,16) par rapport à des tronçons non accidentogènes ayant la même caractéristique (RC=0,50 ; IC95 %=0,34-0,72). Conclusion : Cette thèse montre combien des méthodes épidémiologiques simples, mais novatrices, peuvent être utiles pour évaluer le fardeau des traumatismes par accidents et leurs facteurs de risques dans les PRBM. Ces pays sont confrontés à un énorme fardeau de morbidité routière qui est souvent sous-déclarée dans les données de la police. Un système de surveillance fiable et valide est nécessaire dans les PRBM. De plus, la politique de prévention pourrait être améliorée par une meilleure communication d’information entre les autorités routières et policières concernant les facteurs situationnels. De la même façon, les mesures de sécurité dans les zones en travaux devraient être contrôlées par un système dédié. Enfin, la sécurité routière sur les routes interurbaines dans les PRBM pourrait être améliorée en rendant les routes plus « informant », en particulier avec l’application de mesures peu couteuses telles que les panneaux de signalisations sur les tronçons accidentogènes. Mots Clés: Accidents de la circulation; pays en développement ; trauma; usagers vulnérables.

10

Abstract Background: Interurban traffic safety is a major public health problem, but has received little attention in Low- and Middle-Income Countries (LMICs). The objectives of this thesis were to assess the burden of injury related to interurban traffic, and reporting of these injuries in different datasets, to analyze situational factors (physical characteristics and environmental circumstances) associated with crash sites, and road hazard perception of high-risk crash sites in LMICs. Methods and results: These objectives were assessed in five specific studies conducted in two LMICs, Cameroon and Pakistan. In study I, traffic fatality per vehicle-km and associated crash factors were assessed using police reports for years 2004 to 2007, on the two-lane Yaoundé-Douala road section in Cameroon. Traffic fatality was 73 per 100 million vehicle-km, a rate 35 times higher than a similar road in a high-income country. Fatality was higher for crashes involving vulnerable road users, crashes between oppositely-moving vehicles, and those due to mechanical failure including tyre burst. In study II , traffic injury reporting to police, ambulance, and Emergency Department (ED) in 2008 was assessed, on the four-lane Karachi-Hala road section in Pakistan. Crash fatality was over 53 per 109 vehicle-km, a rate 13 times higher than a similar road in France. Police reported only one out of five fatalities and one out of ten severe injuries. Vulnerable road users were two times less reported in police data than ambulance or ED data. In study III , situational factors associated with injury crash sites were assessed on the Yaoundé-Douala road section, using case-control methods. Factors such as flat road profiles (adjusted Odds Ratios [OR]=1.52; 95% Confidence Interval [95%CI]=1.15-2.01), irregular surface conditions (OR=1.43; 95%CI=1.04-1.99), nearby road obstacles (OR=1.99; 95%CI=1.09-3.63), and three- (OR=3.11; 95%CI=1.15-8.39) or four-legged (OR=3.23; 95%CI= 1.51-6.92) intersections were significantly associated with injury crash sites. Furthermore, the likelihood of crash increased with built-up areas situated in plain regions (OR=2.33; 95%CI=1.97-2.77). In study IV , traffic injury burden and factors associated with Highway Work Zones (HWZs) crashes were assessed on the Karachi-Hala road section, using historical cohort methods. HWZs accounted for one third of traffic fatalities, and fatality per vehicle-km was four times higher in HWZs than other zones. One out of two HWZ crashes occurred between oppositely moving vehicles. In study V, hazard perception of high-risk (with ≥ 3 crashes in 3 years) and low-risk sites (no crash reported) from the two above road sections was assessed by showing videos to voluntary Pakistani drivers. Drivers were able to identify only half of the high-risk sites as hazardous. Sites with a flat and straight road profile had a lower hazard perception compared to those with curved and slope road profile. High-risk sites situated in built-up areas were perceived less hazardous (OR = 0.58; 95%CI=0.51-0.68) compared to low-risk sites (OR = 2.04; 95%CI=1.51-2.74) with same road situation. Further, high-risk sites with vertical road signs were more likely to be perceived hazardous (OR = 2.75; 95%CI=2.38-3.16) than low-risk sites (OR = 0.50; 95%CI=0.34-0.72) with such signs. Conclusion: This thesis illustrates how innovative yet simple epidemiological methods can be useful in assessing the injury burden and specific risk factors in LMICs. These countries face a high burden of interurban road injuries, mostly under-reported in police data. A reliable and accurate injury surveillance system is needed in these countries. Moreover, prevention policy can be improved by better information transfer between road and police authorities regarding situational factors. Similarly, a monitoring system is required to examine the HWZ safety interventions in these countries. Lastly, interurban road safety can be improved by making roads self-explaining, especially by implementing low-cost interventions such as vertical signs at high-risk sites. Keywords: Developing country; highway safety; injury; prevention; vulnerable road users.

11

1. Introduction Les traumatismes routiers sont un problème majeur et pourtant très négligé de la santé publique dans les Pays à Revenu Bas et Moyen (PRBM) [1]. Une enquête récente sur la sécurité routière dans 178 pays a montré que chaque année plus de 90 % des 1,2 million de tués sur les routes surviennent dans ces pays [2]. De plus, ces traumatismes sont la principale cause de pertes des Années de Vie Corrigées du facteur Incapacité (AVCI) dans les PRBM, car de nombreux enfants et des hommes en âge de production souffrent de ces blessures [3]. Les traumatismes pourraient coûter jusqu'à 1 à1,5 % du Produit Intérieur Brut (PIB) de ces pays [4]. On estime que la mortalité routière augmenterait de 80 % entre 1990 et 2020 dans les PRBM, à moins que des mesures appropriées soient mises en œuvre [5]. Pourtant, le transport routier est un facteur essentiel de développement dans les PRBM [6]. Près de 90 % des voyageurs et du fret dans ces pays sont transportés par le réseau routier urbain et interurbain [6, 7]. La sécurité routière sur ces routes devient donc un élément stratégique du processus d’accroissement du développement [8]. Même dans les pays développés, cette catégorie de routes contribue considérablement à la mortalité routière et à des blessures graves [9]. Par exemple au Pakistan, plus de 27% des accidents mortels surviennent sur les routes interurbaines alors qu'elles représentent moins de 5 % de l'ensemble du réseau [10]. Les avantages potentiels de la mise en œuvre des mesures de sécurité routière sur ces routes sont potentiellement énormes, comme cela a été montré par des études dans les pays développés [11]. La recherche joue un rôle central dans la mise en œuvre des interventions sur la circulation [12]. Les conditions routières relativement sûres dans les Pays à Revenu Élevé (PRE) doivent beaucoup aux recherches sur la sécurité routière menées dans les années 1960 et 1970 [13]. Par exemple en Suède, il a été démontré que la recherche sur la gestion de la vitesse en zones urbaines a largement contribué à réduire la mortalité et la morbidité routières, avec un bon rapport coût-bénéfice [14]. Malheureusement, la recherche sur la prévention et la prise en charge des traumatismes routiers reste encore rudimentaire dans les PRBM [15]. La Banque Mondiale a indiqué que des interventions à l'efficacité prouvée existent, mais leur mise en œuvre dans les PRBM est entravée par le manque de recherche pour documenter et comprendre les problèmes spécifiques et locaux des traumatismes [16]. Si les interventions ne sont pas adaptées à la situation locale, elles peuvent ne pas produire le même succès, comme en témoignent les études dans les PRE [17]. Le manque de recherche dans les PRBM est illustré par notre revue de la littérature sur le fardeau des traumatismes routiers et les facteurs de risque associés aux routes interurbaines (Appendix 1) : les indicateurs comparables de la mortalité routier sont rarement évalués et rapportés ; certaines études mentionnent une certaine spécificité, comme la sur-implication des piétons et des occupants de transports collectifs dans les accidents sur des routes interurbaines, mais la répartition réelle des usagers de la route impliqués dans ces accidents n’est pas connue ; la plupart des études épidémiologiques se focalise sur les comportements routiers à risque alors que les facteurs situationnels (caractéristiques physiques et circonstances de l'environnement) n’ont presque jamais été étudiés [18-48]. Des recherches antérieures dans les PRE montrent clairement que ces facteurs sont impliqués dans un quart des accidents et que les interventions mise en œuvre sur les routes pourraient réduire les accidents de 20 % [49].

12

L'objectif de cette thèse était de contribuer à une meilleure connaissance du fardeau des traumatismes routiers et de leurs déterminants spécifiques dans les PRBM. Pour répondre à ces objectifs, cinq études descriptives et analytiques ont été réalisées. Le manque de données sur les traumatismes de la circulation dans les PRBM africains lors de la revue de la littérature réalisée en 2007 par le responsable de l’équipe de recherche dans laquelle cette thèse a été menée [50], nous a conduit à commencer notre travail en décrivant ce problème de santé publique pour certaines situations spécifiques. Peu d'études ayant été publiées sur la sécurité routière interurbaine à l'époque, nous avons commencé par évaluer la charge d’accidents de la circulation sur l’axe Yaoundé-Douala, Cameroun (étude I). Les résultats de cette étude nous ont permis de mieux apprécier le processus de déclaration d'accidents de la circulation dans ce pays. Nous avons répété une étude similaire au Pakistan, mais cette fois nous avons pu recueillir des données provenant de sources multiples, y compris la police, les ambulances et les urgences (étude II). La comparaison de ces rapports a été utile pour évaluer les divergences avec les données de la police, souvent la seule source de rapport sur l'accident comme en témoigne l’étude faite au Cameroun. La littérature publiée à partir des PRBM et les résultats de l'étude du Cameroun ont toujours montré une plus faible contribution des facteurs situationnels dans les accidents que les informations rapportées dans les études de PRE [49]. Malgré cela, nous avons observé que certains facteurs situationnels ont été fréquemment observés sur les sites d’accidents. Il est apparu intéressant d'évaluer les facteurs situationnels liés à des sites d’accidents corporels sur l’axe Yaoundé-Douala par une étude cas-témoins, une méthode jamais utilisée auparavant pour évaluer ces contributions (étude III). De même, la contribution significative d’accidents sur les zones en travaux nous a conduits à évaluer le risque de mortalité routière sur ces zones par rapport aux autres zones. Nous avons évalué ce risque à partir d’une étude de cohorte historique (étude IV). Les deux études ci-dessus ont montré que les circonstances des accidents pourraient être mieux expliquées en évaluant les interactions entres les facteurs liés au conducteur et aux situations. Cela nous a suggéré de développer et de tester une nouvelle méthode d'évaluation des interactions entre la perception de la dangerosité et les facteurs situationnels sur des tronçons accidentogènes chez des conducteurs volontaires pakistanais (étude V). Ces études réalisées à partir d’approches originales devraient permettre de mieux apprécier le fardeau des traumatismes routiers et d’identifier les facteurs situationnels qui pourraient être modifiés pour diminuer le risque d’accidents sur les routes interurbaines dans les PRBM.

13

1. Introduction Road Traffic Injuries (RTIs) are a major yet highly neglected public health problem in Low- and Middle-Income Countries (LMICs) [1]. A recent road safety survey conducted simultaneously in 178 countries demonstrated that more than 90% of 1.2 million estimated yearly road fatalities occur in these countries [2]. Further, RTIs are the major cause of Disability-Adjusted Life Year (DALY) losses in LMICs, because many children and men in their productive ages suffer these injuries [3]. RTIs could cost a LMIC up to 1-1.5% of its Gross Domestic Product (GDP) [4]. It is expected that road fatalities would increase by 80% from 1990 to 2020 in LMICs, unless appropriate measures are implemented [5]. Yet road transport is a crucial determinant of development in LMICs [6]. Nearly 90% of public and goods in LMICs are transported through the urban and interurban road network [6, 7]. Traffic safety on these roads thus becomes a strategic part of the broader development process [8]. Even in developed countries, these roads contribute dramatically to traffic fatalities and severe injuries [9]. In Pakistan, for instance more than 27% of the fatal Road Traffic Crashes (RTCs) occur on interurban roads although these roads represent less than 5% of the entire network [10]. Potential benefits of implementing traffic safety measures on such roads are potentially enormous, as shown by studies in developed countries [11]. Research plays a pivotal role in implementing traffic interventions [12]. The relative safer travel conditions in High-Income Countries (HIC) owe much to the traffic safety research that had been carried out in the 1960s and 1970s [13]. For instance in Sweden, it has been demonstrated that urban speed management research has contributed substantially in reducing traffic-related mortality and morbidity, with a good cost-benefit ratio [14]. Unfortunately, injury prevention and control research still remains rudimentary in LMICs [15]. The World Bank reported that interventions with proven effectiveness exist but their implementation in LMICs is impeded by the lack of research to document and understand specific local injury problems [16]. If interventions are not adapted to the local situation, they may not yield the same success evidenced in HICs [17]. The lack of research in LMICs is exemplified by our literature review of the RTI burden and risk factors associated with interurban roads (Appendix 1): Comparable traffic mortality indicators are rarely assessed and reported; some studies mention some specificity, such as the over-involvement of pedestrians and commercial vehicle occupants in interurban road crashes, but the actual road-user distribution involved in such crashes is not known; most epidemiological studies focussed on risky road behaviours and road situational factors (physical characteristics and environmental circumstances) were almost never investigated [18-48]. Previous research in HICs clearly shows that such factors are involved in one fourth of crashes, and that implementing road interventions could reduce crashes by 20% [49]. The goal of this thesis was to contribute to a better knowledge of the RTC burden and specific determinants in LMICs. This was done through descriptive studies assessing the burden of RTIs using comparable indicators (studies I & II), analytical studies of situational factors involved in RTCs (studies III & IV), and applying a novel method to assess interactions between situational factors with driver-related characteristics on interurban roads in LMICs (Study V).

14

2. Background 2.1 Road injury burden in LMICs A recent World Health Organization (WHO) report showed that road mortality is twice as high in LMICs than in HIC (20 vs. 10 per 100 000 inhabitants) [2]. This trend was clearly visible when selecting the world most populated nations according to their income groups (Figure 1). Traffic fatality is 4 to 6 times higher in LMICs like Pakistan, Nigeria, and Russia than in HICs like United Kingdom and France. Among different world regions, LMICs situated in East Mediterranean and African region had the highest traffic mortality rates (32.2 per 100 000) as compared to other regions with similar economic situations [2, 5]. Furthermore, for every traffic crash death, many more are injured, with temporary or permanent disability [51]. For instance in India, for every reported death, 25 more people are hospitalized due to traffic injuries [52].

Figure 1. Traffic fatality per 100 000 inhabitants in various countries [2]

High-income country

Middle-income country

Low-income country

0 5 10 15 20 25 30 35

Nigeria

Pakistan

Russia

Mexico

Brasil

India

China

Indonesia

Vietnam

United States

Bangladesh

France

Germany

United Kingdom

Japan

Co

un

try

Fatality per 100 000 inhabitants

2.2 Road safety is a key to development in LMICs RTCs are one of the three leading causes of death worldwide for persons aged 15-45 years [53]. They account for 2.7% of DALY losses worldwide, and 3.7% in middle-income countries [53]. Almost half of the traffic fatalities are recorded among vulnerable road users (Figure 2) such as pedestrians, bicyclists, and motorcyclists [2]. The impact of RTIs on the social fabric within LMICs is not straightforward [54]. Death of a man from a low- or middle-income group in their productive age significantly reduces the income of his household and leads to direct and indirect economic losses to the country [3]. Traffic safety is thus more than a health problem, and its improvement in LMICs may have significant consequences on overall national development [8].

15

Figure 2. Traffic fatalities according to road user groups in different countries

0

20

40

60

80

100

United

Sta

tes

Mex

ico

Russia

Franc

e

Germ

any

United

King

dom

Japa

n

Bangla

desh

China

India

Brazil

Indo

nesia

Country

%

Occupants of four wheeled motorized vehicles Vulnerable road users Others

2.3 Risk factors Crash prevention is a priority to reduce RTI burden [55]. Gordon and Gibson were the first scientists who classified injuries as a public health problem by clearly defining it in terms of interaction between host (road user), agent (vehicle), and environmental factors (1949) [56]. William Haddon, Jr., further conceptualized that the involvement of these factors within the phases of influence such as pre-crash, crash, and post crash [56]. Haddon’s matrix provided a means to identify risk factors and preventive measures [57]. For instance for speeding, which is one of the foremost factors involved in highway crashes, this matrix identifies different preventive measures related to road-user, vehicle and environmental factors; these measures, implemented before, during, or after the crash can contribute to the control of RTIs (Table 1).

Table 1. The Haddon Matrix applied to highway crashes involving speed (examples of intervention)

Phases Possible outcome Factors Road user Vehicle and

equipment Environment

Pre crash Crash prevention Traffic enforcement to reduce speeding

Regular vehicle controls to assess braking

Speed calming measures or improved road design

Crash Injury prevention Increased restraint use

Airbag installation Impact-absorbing barriers

Post crash Life preservation Training possible bystanders in first aid skills

Foldable wind screen for emergency exit

Pre-hospital and trauma care system

16

2.4 Multiple factors involved in traffic crashes RTCs are usually consequences of multiple factors [49]. Two studies, conducted independently in the United States and Great Britain, showed that although road user-related factors were identified in up to 94% of RTCs, other factors were also involved in a third of them (Figure 3) [55]. In most cases, such crashes resulted when both road user- and road situational factors (indicated in red) were involved. These situational factors can be fixed road environmental characteristics, such as road geometry, or transient environmental circumstances, such as weather, light, or traffic conditions [55]. Current evidence also suggests that interventions on road environment-related factors can prevent driver-related errors and violations, the foremost cause of traffic crashes reported elsewhere [58]. In this regard, the two known approaches are to make roads “self-explaining” and “forgiving” [3]. For instance, speeding is clearly facilitated by plain road profile and installation of speed-calming measures in such situations decreases crash likelihood by indirectly influencing drivers to reduce their speeds (self-explaining roads) [11]. Further, installation of impact absorbing barriers would mitigate the severity of crash if the crash occurs at all (forgiving roads) [14].

Figure 3. Contribution of risk factors in road traf fic crashes (adapted from [55])

Great Britain United States of America

2 1

1

65

2

424

Road andenvironment

28%

Vehicle8%

Road User94%

Road andenvironment

34%

Vehicle12%

Road user93%

3 1

327

2

6

67

2.5 Implications of interactions for highway safety Highways are the backbone of the economy in all countries. In absence of effective railways and motorways, mixed dual- and single-roads are the major link for almost all transportation of consumables from farms to markets [6]. These roads are over-involved in crash fatalities. For instance in the United States of America (USA), 54% of traffic fatalities occur on such type of road sections [59], similarly, interurban road sections in France account for one third of road crashes but two thirds of road fatalities (Table 2) [9]. Undoubtedly, severity of traffic crashes on these roads is higher than those on urban roads. The key determinant of the high crash fatality is travel speed, itself allowed on these road types [19, 24]. Relationship between

17

road situational factors, high speed, and crash locations has also been demonstrated on rural roads in HICs [60].

Table 2. Traffic fatalities according to road network type in France, 2004 Road network Injury crashes Deaths Crash severity* Urban roads 57 825 1 451 2.5 Rural roads 27 565 3 781 13.7 Motorways 8 182 584 7.1 National roads 5 436 951 17.5 District roads 13 947 2 246 16.1 * Fatality per crash × 100

Traffic prevention on these road sections implies preventing risky road behaviours [61]. Road design, surface, markings, furniture, and traffic management play an important role in reducing crash likelihood by reducing the inappropriate road user behaviours on these road sections [60]. Previous research in developed countries has clearly demonstrated that engineering measures were highly cost effective in reducing injury crashes compared to those targeting only road behaviours or vehicle factors [49]. A British study showed that complete upgrading of national highways to motorways reduced crashes by 76% and traffic fatalities by 81% [62]. Similarly, installation of wired guardrail reduced the likelihood of head-on crashes on undivided rural road sections in Sweden [14]. However, development of these measures requires rigorous research methods to assess their appropriateness to local traffic conditions and demands [14, 16]. 2.6 Factors limiting implementation of engineering measures in LMICs Highway traffic safety has not received appropriate attention in LMICs, both in terms of estimating the injury burden and assessing risk factors [11]. Some of the few studies conducted in such settings suggest that these road sections are important concentrations of traffic crashes in LMICs, probably due to over-involvement of vulnerable road users [17, 19, 63]. Although there is evidence of an increasing injury burden, adaptation and implementation of proven engineering interventions in LMICs is impeded by major knowledge gaps [50]: Firstly, reporting of injuries and availability of RTC and RTI data remains in general the most important difficulty in LMICs. A study in Pakistan showed that police statistics accounts for only 56% of traffic fatalities and 4% of severe injuries in urban settings [64]. Similar results were observed in Iran where the official data source for traffic fatalities was compared with health facility data [65]. As most LMICs do not have vital registration data, WHO recently estimated traffic fatalities in those countries while including information on traffic exposure, risk, preventive, and mitigating factors in their model [2]. The results showed that in most highly populated LMICs, official statistics included only half or less of the actual traffic fatalities occurred in those countries (Figure 4). Thus, without proper estimates, it becomes very difficult to advocate for preventive measures in these countries [50].

18

Figure 4. Percentage difference of crash fatalities between official reported and

estimated figures (adapted from WHO, 2009 [2])

0.1

1

10

100

Ethiop

ia

Nigeria

Pakist

an

Bangla

desh

Indo

nesia

China

Egypt

India

Turke

y

Country

Un

der

rep

ort

ing

%*

* (Estimated deaths – reported deaths) / reported deaths Almost all of the countries use secondary datasets from police, health, or transport departments to assess traffic fatalities and injuries. It becomes equally difficult to obtain disaggregated data on highways, unless appropriate steps, as practiced in the France [9] and the USA [55], are taken for data collection. Further, delays in publication of data, inherent to their collection, pose problems for measuring the impact of interventions [6]. Among secondary datasets, police statistics are used in more than 50% of the countries, particularly LMICs [2]. The problem with police data is that their main focus is to determine human responsibility for a given crash, and not to assess all contributory factors [7]. Previous studies in LMICs have shown that the situational factors for which cost-effective engineering measures exist are less likely to be identified in crash reports (Figure 5) [7]. Similarly hospital data, that report more fatalities than police data [66], do not provide, in most cases, information on crash circumstances, location, or the use of safety devices during the crash [43]. This again leads to knowledge gaps regarding which injury control measures would be suitable for the affected population.

19

Figure 5. Causes of road crashes as determined by the police in developing countries

(adapted from Wootton and Jacobs 1996 [7])

0

20

40

60

80

100

Cypru

s (19

82)

Botsw

ana

(198

2)

Pakist

an (1

984)

Zimba

bwe

(197

9)

Mala

ysia

(198

5)

Philipp

ines (

1984

)

Ethiop

ia (1

982)

India

(198

0)

Afgha

nista

n (1

984)

Iran

(198

4)

Country (Year)

%

Road user Vehicle defects Road factors Other Another limitation to choose and implement interventions is the unavailability of exposition measures such as vehicle-kilometres (vehicle-km) driven on highways in LMICs [6]. Without these measures, it is difficult to compare traffic safety experience of highways in LMICs to those in HICs. Exposition measures allow to document injury experience for specific vehicles suspected to be over-involved in crashes [11]. Moreover, even if policy makers would implement preventive interventions, outcome measures in terms of lives saved or injuries prevented could not be adjusted for change in exposition in the post-intervention phase [17]. Last but not least, the approach of traffic safety is different in LMICs and in HICs, and needs specific research. For instance, although vulnerable road users account for a majority of traffic fatalities in LMICs [11], traffic separation interventions such as construction of access controlled roads for intercity travel would divide the local population in two zones [17]. As the people from LMICs are mostly less motorized, those crossing such road sections will expose themselves to high-speed vehicles and injury risk. Further, the costs of most interventions from HICs are often enormous, thus make their implementation difficult in LMICs [17, 62]. Injury prevention interventions need to be adapted to the local situation in LMICs [16]. Little research has been carried out to assess the local factors and potential benefit of related interventions in LMICs. For instance, out of 236 studies evaluating traffic safety interventions, only six were from the LMICs [67]. Injury research funding, even in the US, does not correspond to the burden, and LMICs are far behind in placing injuries on their research agenda [16, 68]. Thus research should focus on determining the crash burden of factors involved as well as the development of appropriate interventions in LMICs [50].

20

2.7 Interurban road safety research gaps in LMICs To our knowledge, no review on the epidemiology of interurban RTC in LMICs was available. As part of a systematic review of original studies published in Medline® with interurban road settings (Appendix 1), we only found 31 articles from these countries. Most of them used police or health data to assess crash burden. We did not find any study comparing differences in injury reporting between these types of data. Only one study reported injury burden per 100 million vehicle-km. Situational factors other than the light or weather conditions were almost never documented in these studies. None of them assessed the interactions between driver- and road-related crash factors. Only few studies assessed high-risk crash site identification methods on inter-urban road sections in LMICs. Finally, few road interventions were assessed in these countries, mostly by non comparative methods.

3. Objectives To contribute to filling the research gaps on RTCs and RTIs in LMICs, the objectives of this thesis were:

1. To assess the road crash and injury burden on selected interurban roads in LMICs–Studies I & II;

2. To describe road user groups and situational factors involved in interurban road

crashes–Studies I & II; 3. To assess the association of situational factors with injury crashes on selected roads of

LMICs–Studies III & IV; 4. To assess the road hazard perception of the high-risk crash sites in voluntary drivers–

Study V. The thesis presents two descriptive studies (I & II) and three analytical studies (III-V) in line with these objectives. In study I, the crash burden, number of persons who died or were injured per vehicle-km was assessed on Yaoundé-Douala road section in Cameroon. Further, associated crash factors and types were described using police reports. In study II, we compared RTIs per vehicle-km reported to police, ambulance, and hospital on Karachi-Hala road section during a one-year period. In study III, we assessed situational factors associated with injury crash sites using case-control methods on the Yaoundé-Douala road section. In study IV, we compared the incidence density rates estimated from events (crash, fatality, and severe injury) and vehicle-km between highway work zones and other traffic zones on Karachi-Hala road section. In study V, we assessed the hazard perception of high-risk crash sites and those not involved in crashes from the above two road sections by showing their videos to voluntary Pakistani drivers. Furthermore, situational factors associated with drivers’ hazard perception were assessed using multivariate models. Study I and III have been published whereas manuscript of study IV is under review and those for study II and V are in preparation.

21

4. Descriptive studies The paucity of traffic injury data in African LMICs, as discussed in the literature review published in 2007 by the head of the research team in which this thesis was conducted [50], made interesting for us to start our work by describing this public health problem for some specific setting. Since few studies were published on interurban traffic safety at that time, we started by assessing traffic crash burden on Yaoundé-Douala road section, Cameroon (Study I) . Results from this study permitted us to better appreciate the process of traffic injury reporting in this country. We repeated a similar study in Pakistan, but this time we were able to collect RTI data from multiple sources including police, ambulance, and hospitals (Study II) . Comparison of injury reporting was useful to assess discrepancies in police data, often the only source of crash reporting as evidenced in Cameroon. 4.1 Study I: Traffic crash and injury burden on Yaoundé-Douala road section, Cameroon Located in Central and West Africa on the Gulf of Guinea, Cameroon covers an area of 475 440 square kilometres and has a population of nearly 19 million inhabitants. Cameroon has one of the ten highest per capita GDPs—about 2 300 US$ (2008)—in sub-Saharan Africa. The life expectancy at birth is 54 years [69]. The total road network is 50 000 km long, with only 10% paved [69]. Its interurban road network is highly strategic in the region as it is expected that its economy will grow substantially in coming years, with increasing oil imports from Chad and industrialization [70]. Cameroon is in pre-motorization stage with 16.8 registered vehicles per 1 000 persons [2]. An increasing trend in crash fatalities has been observed in Cameroon since 1970s [3]. Traffic fatalities increased from under 400 in 1972 to over 1 150 in 2005 [2]. Over half of those who died are passengers and drivers of four-wheeled vehicles, whereas pedestrians, cyclists, and motorcyclists account for the rest of deaths [2]. A recent estimate suggested that actual traffic fatalities could be around five thousand per year in Cameroon [2]. Speed limit, drunk-driving, seat-belt, and helmet laws are poorly enforced [71].The formal pre-hospital care system is still in development and RTCs are the principal reason for such interventions in urban settings [71]. Traffic safety on interurban road network is increasingly becoming the matter of public concern in Cameroon. Crashes that drew extensive public attention were mostly those involving public transportation vehicles with several fatalities [72]. When our study was started, no comparative fatality and injury estimates were available for these road sections. Further, no description was available in the literature of crash types, causes, and road users involved. Several reasons could explain unavailability of this information: firstly, not all concerned police reports were forwarded to the department in charge of traffic injury surveillance and, secondly, no detailed data were neither collected nor analyzed at the national level, that could provide a better picture of traffic safety on interurban roads [72].

22

Objectives The objectives of this first study were:

1. To assess crash burden on an interurban road section in Cameroon.

2. To describe crash types, causes, and situational factors.

3. To describe outcome of traffic injuries, according to road user-related factors.

This study was published as: Sobngwi-Tambekou J, Bhatti J, Kounga G, Salmi L-R, Lagarde E. Road traffic crashes on the Yaoundé–Douala road section, Cameroon. Accident Analysis and Prevention 2010;42(2):422-6 (Appendix 2). The data was collected under supervision of Dr. Sobngwi-Tambekou and Dr. Lagarde. I contributed to the analysis and manuscript writing. Methods Setting and study design The study settings were the Yaoundé-Douala road section. This 243-km long, mostly two-lane, un-separated road section serves as a major link between two most populous cities of the country, Yaoundé and Douala. To assess road burden and describe factors involved in highway crashes, traffic and crash data from road authorities and highway police stations was retrospectively collected. Traffic counts Traffic count surveys were conducted by the Ministry of Public Works, Cameroon, on five locations during two seven-day periods in May and November 2005 [73]. Daytime traffic counts were recorded from 6:00 am to 10:00 pm, whereas night-time counts were recorded from 10:00 pm to 6:00 am. All passing vehicles were counted and classified. Results showed that the daily traffic count varied between 2 269 and 3 553 vehicles on Yaoundé-Douala road section; mid-city sections had the lowest traffic counts, and those close to the two main cities had the highest traffic counts. Personal vehicles accounted for 55%, public transportation for 21%, and trucks for 24% of the traffic [73]. Police reports In June and July 2007, all 13 police stations within and outside Yaoundé and Douala were visited to collect police traffic crash reports that had been filed between January 2004 and May 2007. In principle, a police report is established for all injury crashes and for a number of non-injury crashes involving several counterparts, for which property loss and civil responsibilities are at stake. All police reports retrieved from the police station archives were scanned. The State Defence Secretariat Bureau in Yaoundé, in charge of centralizing police reports in the country, was also visited to seek reports not found in the police stations. As a number of police reports are lost or destroyed, we evaluated the completeness of data collection by comparing the availability of police reports with crash events reported in the main police station registers. In all police stations, such a register is continuously updated with one line for all events related to police interventions. This includes RTCs, which are listed and identified by a report number and the corresponding police report number. Because the exact crash site is not specified in the registers and some crashes could have occurred on

23

secondary roads, the register could not be used to assess the crash rate on the Yaoundé-Douala section. However, the register made it possible to estimate the completeness of police reports retrieved and digitized. This exercise was only possible in the eight largest police stations and was not conducted in the remaining five smaller ones. All reports from all 13 police stations were however included in this study. Data extraction Scanned reports were coded, using a grid adapted from the standardized French data surveillance system [9], and were recorded using EPIINFO 2000 [74]. Crash details included hour, day, month, and year of crash, number of vehicles and road users involved, situational factors such as light and weather conditions, horizontal and vertical road profile, road and shoulder surface conditions, and whether situated in urban zones, at intersection, or near a school. Involved vehicles were categorized as trucks, personal, utilitarian, and passenger vehicles. Involved road users were classified by vehicle type or as pedestrians, bicyclists, and motorcycle riders. Outcome of these road-users were graded by the police as: not injured, slightly injured, seriously injured (when needing hospital admission), and died. Road user-related variables such as age, sex, profession, means of transport to hospital, and road user category (driver, passenger and pedestrian) were recorded when available. For motorized vehicle users, information on the use of helmet or seat belt and Driving While Intoxicated (DWI) was recorded. Similarly, for pedestrian, information on crossing facilities was recorded. In addition, all reports were independently read by two investigators (Emmanuel Lagarde and Jöelle Sobngwi-Tambekou), and coded using a standardized grid for crash type, and one or two possible causes. No inconsistencies were found for crash type, and when two different causes were identified by the two investigators, both were recorded. On some occasions, the crash corresponded to several types (for example, a two-wheel motorized vehicle versus a pedestrian). Such crash type was therefore coded as the type involving the user of highest vulnerability. The vulnerability decreasing order was: one or more pedestrians involved; one or more two-wheel motorized vehicles; vehicles travelling in opposite directions; single vehicle running off the road; vehicles travelling in the same direction; one or more still or manoeuvring vehicle; crash at an intersection). Analysis Crash and injury risks per 100 million kilometres travelled were computed, with numerators being numbers of persons or events in the study period, and denominators estimated number of kilometres travelled on the Yaoundé-Douala road section during the study period [73]. These rates were corrected taking into account our estimates of police report completeness, as compared with the events listed in main police registers. To assess the respective involvement of personal, utility, passenger vehicles and trucks, we divided the number of vehicles of each type involved in non-fatal and fatal crashes by kilometres travelled by same vehicle type. Crash and injury risks for specific vehicle and road user group were compared using rate ratios and their 95% confidence intervals. Injury severity for different road user groups was assessed by fatality ratio, fatalities per total injured. Percentage increases in traffic fatalities and injuries were computed for the year 2006 compared to 2004. Furthermore, proportions of crash causes and situational factors were computed according to crash types. Crash fatality ratios and fatality per crash were compared for crash types and causes. Lastly, the outcome of injuries was compared according to road user characteristics. Data were analyzed using SAS version 9.1.3 [75].

24

Results Crash burden A total of 935 RTCs police reports corresponding to our study period were retrievable from 13 police stations. These RTCs could be classified into fatal (N=228), injury (N=428), and materiel-damage only (N=279) crashes. A total of 3 039 persons were involved in these crashes; 12.3% died, and 49.2% were injured. Among those who died (N=374), 74.3% died immediately as a result of crash impact, 5.6% died after impact on the crash scene, 2.7% died during transportation, and 17.4% died in the hospital. Among the injured (N=1 494), 48.6% were declared as seriously injured in the police report. When these figures were compared for their completeness to a total of 1 400 RTCs reported in registers of eight police stations, we estimated that these results accounted for 65.4% of all crashes, 62.7% of injury crashes and 76.1% of fatal crashes. With an estimated 655 465 vehicle-km travelled daily, the corrected mortality and morbidity estimates were approximately 73 deaths and 240 injuries per 100 million vehicle-km travelled (Table 3). Occupants of personal and passenger vehicles were about twice as often involved in an injury or fatal crashes than trucks’ occupants. Similarly, the injury and fatality risk was four times higher for occupants of personal and passenger vehicles than for trucks’ occupants. Injury severity was higher for vulnerable road users: pedestrians (0.43), cyclist (0.36), and motorcyclist (0.25), as compared to other vehicles’ occupants (≤ 0.20) (Figure 6).

Table 3 Burden of road traffic crashes and injuries, according to vehicle type on Yaoundé-Douala road section (2004-2007)

Non fatal Fatal

N Rate* RR 95% CI N Rate* RR 95% CI

Crash risk by vehicle

- Truck 86 45 1 53 27 1

- Personal vehicle 340 102 2.29 1.81-2.90 142 43 1.55 1.13-2.13

- Utilitarian vehicle 85 70 1.56 1.16-2.11 41 34 1.22 0.81-1.84

- Passenger vehicle 159 93 2.10 1.61-2.72 84 49 1.80 1.27-2.53

Total** 428 69 228 44

Injury risk by road user

- Passenger

- Truck 108 56 1 19 10 1

- Personal vehicle 557 167 2.99 2.43-3.67 133 40 4.06 2.51-6.56

- Utilitarian vehicle 171 140 2.50 1.97-3.18 22 18 1.83 0.99-3.38

- Passenger vehicle 392 230 4.11 3.32-5.09 85 50 5.07 3.08-8.34

- Pedestrian 77 † 57 †

- Cyclist 7 † 4 †

- Two-wheeled motorized vehicle 110 † 36 †

- Other or unspecified 72 † 18 †

Total‡ 1494 240 374 73

* Per 100 million vehicle-km † Denominator not available ‡ Corrected for under-reporting as per police registers

25

Figure 6. Traffic injury outcome, according to road user group on Yaoundé-

Douala road section (2004-2007)

0.00

0.20

0.40

0.60

0.80

1.00

Pedes

trian

Cyclis

t*

Two-whe

eled v

ehicle*

Perso

nal v

ehicl

e*

Utilitar

ian ve

hicle*

Truck

*

Passe

nger

vehic

le*

Others/

unsp

ecified

*

Road user

Fat

alit

y p

er t

ota

l in

jure

d

FatalityInjury

* Driver or occupant

Time trends Over 2004 to 2006, traffic injuries and fatalities increased (Figure 7). For instance, as compared to 2004 (N=105), a 20% increase in road fatalities was observed in 2006 (N=126). Similarly, the injuries increased from 492 in 2004 to 687 in 2006, a 45% increase. A peak of RTI occurrence was observed on Friday, Saturday, Sunday, accounting for 55.1% of road fatalities and 54.4% of RTIs (Appendix 3).

26

Figure 7. Monthly trend of traffic fatalities and injuries on the Yaoundé-Douala

road section (Jan 2004 to May 2007)

0

10

20

30

40

50

60

70

80

90

100

J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M

2004 2005 2006 2007

Time period

N

Fatality

Injuries

Crash type patterns Out of 935 RTCs, the major crash types were: 1) collisions in the same direction (19.3%); 2) single vehicles running off the road (19.2%); 3) collisions of vehicles travelling in opposite directions (16.7%); and 4) RTCs involving pedestrians (15.0%) (Table 4). Crash fatality was high for those involving collision of vehicles in opposite direction (0.9) and pedestrian crashes (0.5) (Appendix 3). In most crashes, road user-related factors, such as hazardous overtaking (28.6%), excessive speed (19.5%), inattention/distraction (14.5%), loss of control (12.8%), and hazardous manoeuvre (12.2%), were identified. Vehicle-related factors, such as tyre problems and other mechanical causes, were identified in 18.0% of crashes. Environmental factors were identified in 4.0% of the crashes. Among 99 RTCs where no cause could be identified, 44 involved pedestrians left without assistance by a run-away vehicle. Weather conditions were normal for most crashes (81.4%). A significant proportion of crashes occurred in built-up areas (41.6%), at intersections (19.3%), and on sections with flat-road profile (70.7%). Same-direction crashes were frequent with hazardous overtaking (47.5%) and loss of control (22.7%). Running-off-the-road crashes were frequently observed with tyre (35.0%) and mechanical problems (52.8%), and when shoulder conditions were irregular (21.3%). Similarly, opposite direction were frequent with hazardous overtaking (56.4%), and pedestrian crashes were more frequent than average with inattention and distraction (29.8%) and in built-up areas (54.3%). Two-wheeled motorized vehicle crashes were frequent with hazardous overtaking (39.5%), in built-up areas (76.0%), at intersections (29.0%), and with straight road profile 75.3%). Crashes at intersections were frequent with excessive speed (33.9%), inattention and distraction (62.9%), hazardous manoeuvre (35.5%), built-up areas (77.1%), and straight road profile (79.0%).

27

Table 4. Crash types, causes, and situational factors on the Yaoundé-Douala road section (2004-2007) All (N=935) Same

direction Running off

the road Opposite direction

Pedestrian Other still or manoeuvring

vehicle

Two-wheeled

motorized vehicle

Intersection

N % (N=181) (N=180) (N=156) (N=141) (N=84) (N=81) (N=62) % % % % % % % Causes Hazardous overtaking 268 28.6 47.5 18.3 56.4 5.7 14.3 39.5 8.1 Excessive speed 182 19.5 18.8 18.3 18.6 20.6 20.2 12.4 33.9 Inattention, distraction 136 14.5 9.4 1.7 1.3 29.8 13.1 21.0 62.9 Loss of control 120 12.8 22.7 11.1 16.0 4.3 13.1 11.1 3.2 Hazardous manoeuvre 114 12.2 13.3 6.1 7.1 3.6 22.6 22.2 35.5 Unsafe parking 74 7.9 8.3 2.8 9.0 2.8 35.7 3.7 3.2 Tyre puncture/burst/loss 98 10.5 3.8 35.0 7.1 4.3 1.2 3.7 0.0 Other mechanical cause 71 7.8 8.3 52.8 12.8 7.1 10.7 4.9 1.6 Environmental factors 37 4.0 5.0 7.2 1.9 6.4 2.4 0.0 0.0 Light conditions Light 552 59.2 58.0 70.4 53.9 57.1 49.4 60.5 54.0 Twilight 92 9.9 8.3 10.1 9.0 9.3 16.9 8.6 12.0 Night 288 30.9 33.7 19.6 37.2 33.6 33.7 27.4 34.0 Weather conditions Normal 749 81.4 75.6 87.1 75.0 90.4 72.3 93.8 75.8 Rain 131 14.2 18.9 9.6 19.7 8.1 16.9 5.0 17.7 Other 40 4.6 5.6 3.4 5.3 1.5 10.8 1.3 6.5 Built-up area 381 41.6 34.4 18.8 27.5 54.3 46.3 76.0 77.1 Intersection 168 19.3 16.5 6.1 7.6 24.0 19.7 29.1 72.1 Flat road profile 656 70.7 68.5 65.9 68.6 72.1 77.4 76.3 72.1 Straight road profile 570 61.3 61.3 42.7 53.3 68.6 73.8 75.3 79.0 Irregular surface condition 7 0.8 1.1 1.1 0.7 0.0 0.0 0.0 3.3 Irregular shoulder condition 79 11.1 7.4 21.3 14.8 8.6 9.1 3.1 2.1 Nearby school 16 1.7 0.6 1.7 0.0 5.1 0.0 3.7 1.6

28

Injury outcome patterns Most of injured road users were aged 15-59 years (93.1%) (Table 5). Those aged 0-14 years and 60 years or more accounted for 15% of road fatalities. Females accounted for 15.7% of those involved, but 25.1% of those injured and 26.2% of those who died. Pedestrians accounted for 5.4% of road users involved in crashes, but 18.4% of those who died. Twenty five pedestrians died while crossing, whereas twelve of them died on the roadside. Nearly four out of five injured persons were transported by private means to the hospital. Ambulances were used to transport 4.0% of those injured and 8.4% of those died. Police forces evacuated 2.0% of those injured and 6.0% of those who died. Table 5. Traffic injury outcome, according to road-user characteristics, on the Yaoundé-

Douala road section (2004-2007) Total Injured Died N % N % N % Age (y) - 0-14 52 2.4 35 3.8 15 6.5 - 15-29 563 25.8 284 30.8 68 29.6 - 30-44 1 019 46.8 399 43.3 71 30.9

- 45-59 485 22.3 175 19.0 58 25.2 - ≥ 60 59 2.7 29 3.1 18 7.8 Sex - Male 2 322 84.3 952 74.9 262 73.8

- Female 431 15.7 319 25.1 93 26.2 Profession - Professional drivers 1 049 52.7 230 29.0 46 35.9

- Office managers/employees 442 22.2 219 27.6 24 18.8 - Self employed 193 9.7 112 14.1 17 13.3 - Manual 127 6.4 97 12.2 15 11.7

- Students 74 3.7 53 6.7 15 11.7 - Other 107 5.4 82 10.3 11 8.6 Road user

- Driver 1 673 55.1 436 29.2 98 26.2 - Vehicle occupants 1 200 39.5 968 64.8 208 55.6 - Pedestrian 163 5.4 89 6.0 68 18.2

Pre-hospital transport - Not transported 1 297 43.5 116 7.9 63 17.1 - Private 1 539 51.7 1 260 85.7 252 68.5

- Ambulance 90 3.0 59 4.0 31 8.4 - Police force 52 1.7 29 2.0 22 6.0

Seat-belt wearing was reported for only three persons out of over three thousand vehicle occupants. Blood-alcohol concentration of responsible drivers was not systematically performed and no record was available. Apparent DWI was identified in only two drivers. In about 9.3% of the drivers no driving permit was available. Twenty pedestrian died at sites where no zebra crossing was available.

29

4.2 Study II: Differences in police, ambulance, and emergency department reporting of traffic injuries on Karachi-Hala road section, Pakistan Pakistan located at the junction of Middle-East, South-East Asia, China, and Central Asian States, is the sixth most populous nation of the world [69]. Approximately 1.4 million RTCs occurred in Pakistan during 1999, resulting in over 7 000 fatalities [76]. A recent WHO report showed that actual traffic fatalities could be 4 to 10 times higher than official statistic [2]. Similarly, two independent population-based surveys estimated incidence of traffic injuries around 15 to 17 per 1 000 persons per year [77, 78]; RTIs contribute significantly to the work load in hospitals [79]. The direct cost of RTIs to Pakistani economy is over 1 billon US$ [80]. Interurban road sections are the back-bone of Pakistani economy. Its strategic interurban road network of approximately 8 000 km plays a significant role in transport, as it carries more than 80% of inland passenger and freight traffic [76, 81]. Although these road sections account for 4% of the entire network, a high proportion of traffic fatalities (27%) occur on these road sections [63]. Similarly, previous research in Pakistan has shown that injury severity was higher for crashes in rural areas [82]. However, no distinction has been made whether this refers to interurban or other rural roads. Higher speeds and inappropriate geometrical design can explain this high fatality ratio, but no comparison indicators were available for these road sections [60]. The catchment area for traffic injuries is difficult to define on interurban road sections in LMICs, and police records remain to date the most reliable source of evaluating interurban road safety [7]. The use of police statistics alone can lead to underestimation of road burden in LMICs, as previously illustrated by linking police and ambulance datasets in Karachi, Pakistan [64]. No notable research has been carried out to compare the differences in injury reporting while linking different datasets with interurban road settings in LMICs [2, 3]. Objectives The objectives of this study were:

1. To assess differences for crash and injury reporting in police, ambulance, and emergency department (ED) datasets on an interurban road section in Pakistan. 2. To estimate variations of traffic fatality and injury per vehicle-km travelled when linking these datasets.

The manuscript of this study is currently prepared for submission: Bhatti JA, Razzak JA, Lagarde E, Salmi L.-R. Differences in police, ambulance, and emergency department reporting of traffic injuries on Karachi-Hala road, Pakistan (Appendix 4). I have been involved in all steps of this study, from conception, to data collection, analysis, interpretation of results, and manuscript writing.

30

Methods The study setting was 196-km long Karachi-Hala road section (km 16 to km 212 from Karachi city centre). This is a four-lane highway, two lanes in each direction. The lanes are separated by a ground surface, but there are no physical barriers. It has one of the highest traffic counts in the province, with over 24,000 vehicles per day [83]. This high traffic count can be explained by the economic activity in Karachi, the most populous city of Pakistan, which accounts for 70% of the government’s revenue through trade and industry [84]. In this retrospective study, information on traffic injuries reported to Police, ambulance service, and ED during 2008 (Jan to Dec) was collected and compared. Police data Since 2004, the National Highway & Motorway Police (NHMP) ensures traffic enforcement on this 196-km long road section. Administratively, this section is considered as Sector I of South-Zone of NHMP and is divided further in four 46- to 51-km-long beats: beat 35 (km 16 to 62), beat 34 (63 to 114), beat 33 (115 to 162), and beat 32 (163 to 212). NHMP deploys four motor vehicles and four patrolling officers in an eight-hour shift on these beats [10]. For every crash, a standard accident analysis report is filed by the attending NHMP officer [85]. A copy of this report is kept in the NHMP regional office. Similarly, details on crash and those involved are recorded on a separate accident register. From these reports and registers, information was principally extracted on time, date, location of crash, and whether it was fatal, involved injury, or was without injury. Similarly, we extracted information on name, age, sex, outcome (dead; severe injury, defined as transported to hospital; and mild injury, defined as not transported to hospital), and hospital brought to (when available) of those involved in crashes. Ambulance data Ambulance records were obtained from Edhi Ambulance Service (EAS) logbooks. EAS is the largest private philanthropic ambulance service in the world [86]. EAS has been providing ambulance service to injury patients on this road section for the last 15-20 years. For this purpose, EAS has established six ambulance posts, mostly near main towns, to provide care to traffic injury patients. Location of these posts are: 1) Sohrab Goth (12 km from Karachi centre), 2) Karachi toll plaza (km 28), 3) Edhi centre (km 56), 4) Nooriabad (km 94), 5) Hala Naka (km 160), and 6) Hala (km 212). This service is freely available to injury patients, and funds are raised by transporting other patients. Ambulance staff consists of, in most cases, only the driver. A clerk at the post can accompany the driver if he thinks this justified, for instance in case of crash with multiple patients. Ambulance communicates with emergency post through wireless system or by cell phone. RTI patients or bystanders can contact the services using the free emergency access number 115, which connects them to the main city centre [86]. Information is then transmitted by wireless or cell phone to nearby posts, which finally dispatches the ambulance(s). After reaching the scene, attendants separate injured and dead patients. Those severely injured are transported to the nearest hospital; preference is given to the government hospital if available. All information on the RTI intervention including crash location, RTI patient identity and outcome is then transmitted by wireless or telephone to the regional centre, which records the information in a central log book. We obtained these log books from the regional centre at Karachi. Crash details such as date, time, location, and whether it was fatal or involved injury were extracted from these books. Similarly, road user details such as name, sex, age, user type (pedestrian, motorcycle rider, or vehicle occupant), and outcome (died, including when the

31

person died at crash scene, during transport, or at ED; injured and transported, including hospital taken to; injured and not transported) were extracted from these books. Hospital records The Road Traffic Injury Research & Prevention Centre (RTIRP) at the Jinnah Post Graduate Medical Centre (JPMC) has been working since September 2006 [87]. This centre systematically collects, on standard Performa sheets, information on RTI patients presenting at the ED of the five largest teaching hospitals in Karachi: 1) JPMC, 2) Abbasi Shaheed Hospital, 3) Civil Hospital Karachi, 4) Liaqat National Hospital, and 5) The Aga Khan University Hospital. This dataset includes information on the crash date, time, and location as well as patient’s name, age, sex, road user group. Further information on whether the patient was wearing helmet or seat belt was available. The New Injury Severity Score (NISS) [88] and outcome (discharged, admitted/referred, or died) of patients were recorded during their stay in the ED. Information on RTI patients involved in crashes on selected road section was extracted from this dataset. Analysis All information was recorded on Excel® spreadsheets. We computed percentages for crash and injury patient characteristics for three datasets and compared them with each other. For the ED dataset, the distribution of NISS according to the outcome was plotted. Records from the three datasets were then matched for crash date, name, age, and sex of RTI patients involved. For matched records, we identified changes in outcome. Total deaths and injuries were then assessed while removing the records appearing in two or more datasets. Ascertainment rates for police, ambulance, and ED records as compared to these total fatalities and injuries were computed. Unique record and traffic counts from National Highway Authority (NHA) were used to compute overall traffic fatality and injury rates per vehicle-km for 2008 [83]. Results Crash outcome In 2008, police reported 43 crashes, whereas 255 crashes were reported to EAS and 449 were reported to ED. One out of two police reported crashes (N=19, 44.4%) was fatal, whereas this proportion was 14.5% (N=37) for those reported to EAS and 10.4% (N=47) for those reported to ED. No information on crash outcome was available in 13.3% of EAS reported crashes and 6.7% of those reported to ED. Injury outcome A total of 143 RTIs were reported to police, 531 to EAS, and 661 to ED. Monthly trends indicated higher proportions of RTIs in June and July 2008 (Figure 8). Over half of police-reported injury patients received hospital care (N=80, 55.9%) (Table 6). Half of these patients (N=40), injured between km 16 and km 120 were treated in Karachi; RTIRP hospitals treated 17 of them. Nearly one fifth of RTI patients reported in police records died (N=27, 18.8%), whereas this proportion was 10.4% for EAS and 9.1% for ED reported patients. One fourth of police-reported injury patients (N=25.2%) were not transported to the hospital, whereas this was 9.0% for EAS-reported patients (N=48). Out of 661 patients presenting to ED, 47.7% (N=315) arrived by private means, whereas 43.0% (N=284) arrived in ambulances. Police transported only four of these patients, and no information was available on the remaining 58

32

patients. In the ED, those with NISS scores from 4 to 8 had a higher likelihood of hospital admission (81.0% vs. 19.0%, P<0.001) than those with NISS scores from 1 to 3 (Figure 9). Those who were reported to have died had NISS scores 10 or more. Figure 8. Month-wise police, ambulance, and emergency department reporting of traffic

injuries on Karachi-Hala road section (2008).

0

20

40

60

80

100

120

140

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Month

N

Police

Prehospital

Emergency Department

33

Figure 9. Outcome of traffic injuries reported to emergency department according to

New Injury Severity Score (NISS) on Karachi-Hala road section (2008)

0

20

40

60

80

100

120

1 2 3 4 5 6 8 9 10 11 12 13 14 16 17 18 19 21 22 25 26 27 29 33 34 35 38 43 50

NISS

N

Died

Admitted

Discharged

Patient demographics Names were available for 67.1% of police- and 78% of EAS-reported injury patients (Table 6). Information on age was available for 74.1% of police- and 67.6% of EAS-reported injury patients. Few records in ED dataset were without names (N=12) and age (N=5). Most injury patients in the three datasets were aged 16-45 years: 61.5% in police, 55.0% in EAS, and 78.1% in ED. Similarly, men accounted for a majority of injuries, up to 92.1% of injury patients in ED. Road user group The proportion of pedestrians in police-reported crashes was 3.5% (N=5), whereas this was 7.5% in the EAS and 12.7% in the ED. Similarly, the proportion of motorcycle riders in police reported crashes were 4.2% whereas this was 9.2% in EAS and 30.6% in ED. Occupants of four-wheeled vehicles accounted for a majority of injuries in the three datasets: 83.9% in police, 75.9% in EAS and 49.5% in ED. In the ED, only 15.7% (N=21) of the 154 injury patients riding motorcycles were wearing helmets. Similarly, only three out of ninety-three four-wheeled vehicle occupants were wearing a seat belt at time of crash.

34

Table 6. Traffic injuries reported to police, ambulance, and emergency department on

Karachi-Hala road section (2008).

Police Ambulance Emergency department

N % N % N % Road traffic crash - Fatal 19 44.1 37 14.5 47 10.4 - Not Fatal 24 55.8 184 72.2 372 82.9 - Unknown 0 0.0 34 13.3 30 6.7 Road traffic injury - Deaths 27 18.8 55 10.4 60 9.1 - Transported to hospital 80 55.9 428 80.6 601 90.9 - Not transported to hospital 36 25.2 48 9.0 NA Name of patient available - Yes 96 67.1 414 78.0 648 98.0 - No 47 32.9 117 22.0 13 2.0 Age (y) - 0-15 14 9.8 34 6.4 62 9.4 - 16-45 88 61.5 292 55.0 516 78.1 - >45 4 2.8 33 6.2 78 11.8 - Unknown 37 25.9 172 32.4 5 0.7 Sex - Male 93 65.0 364 68.5 609 92.1 - Female 12 8.4 78 14.7 52 7.9 - Unknown 38 26.6 89 16.8 0 0.0 Road user group - Pedestrian 5 3.5 40 7.5 83 12.7 - Motorcycle riders 6 4.2 49 9.2 203 30.6 - Four-wheeled vehicles’ occupants 120 83.9 403 75.9 327 49.5 - Others 0 0.0 1 0.2 4 0.6 - Unknown 12 8.4 38 7.2 44 6.6

NA – Not applicable

Matched records A total of 108 patients were found in two or more datasets (Figure 10), including 13 who were found in all datasets, 28 who were found in police and EAS datasets, and 14 who were found in both police and ED datasets; 93 records were common between ambulance and ED datasets. Some discrepancies were observed for outcome of injuries reported in police and ambulance records (Table 7): four of the 17 injured in police dataset were reported dead in EAS records. Similarly, one of eight injured in police records was reported dead in ED records, and nine of 84 injured patients in EAS were reported dead in ED records. Ascertainment of road fatalities and injuries Based on matching, 119 unique patients were reported to have died in 2008 on this interurban road section (Table 8). Police recorded 22.6%, EAS 46.2% and ED 50.4% of them. Similarly, a total of 1 095 injuries were reported injured in three datasets after identifying matched

35

records. Police accounted for 10.6%, EAS 43.5%, and ED 54.9%. Traffic mortality rate was 54 deaths per 109 vehicle-km and RTI rate was slightly over 500 injuries per 109 vehicle-km on this road section. Matching of nameless police and ambulance records, when any of the crash dates, time, age, and sex details was available, decreased the overall estimates by 4 deaths and 73 injuries. Corrected traffic fatality was 53 deaths and injuries were 467 per 109 vehicle-km travelled on this road section.

Figure 10. Traffic injuries reported to police, ambulance service, and emergency departments on Karachi-Hala road section in 2008 (N=1 214)

PoliceN=143

AmbulanceN=531

HospitalN=661

114

1315 1

424 56879

Table 7. Differences in outcome of traffic injury among police, ambulance, and emergency department for same patient on Karachi-Hala road section, 2008 (N=108)

Ambulance Hospital Injured Died Discharged Admitted Died N (%) N (%) N (%) N (%) N (%) Police Injured 13 46.4 4 14.3 6 42.9 1 7.1 1 7.1 Died 0 0.0 11 39.3 0 0.0 0 0.0 6 42.9 Ambulance Injured 49 53.3 26 28.3 9 9.7 Died 0 0.0 0 0.0 8 8.7

Table 8. Ascertainment of police, ambulance, and emergency department records for traffic fatalities and injuries on Karachi-Hala road section (N=1 214)

Outcome Police Ambulance Hospital Total Rate* N %† N %† N %† N %

Deaths 27 22.6 55 46.2 60 50.4 119 9.8 54.4 Injuries 116 10.6 476 43.5 601 54.9 1 095 91.2 500.4

* per 109 km travelled. † Ascertainment rate; record numbers (N) divided by total (N) for the given outcome.

36

5. Analytical Studies Previous literature from LMICs, and results from the Cameroon study consistently showed a lower contribution of situational factors in crashes than those reported in studies from HICs [49]. Nevertheless, we observed that some situational factors were frequently observed at crash sites. We thought it interesting to assess situational factors associated with injury crash sites on Yaoundé-Douala road section by case-control study, a method never used previously to assess such contributions (Study III) . Similarly, a high crash fatality associated with road section undergoing maintenance, made interesting for us to assess such burden and compare it with that of normal traffic using historical cohort study methods (Study IV). Both of the above studies showed that crash circumstances could be better explained when driver- and situation-related factors are considered simultaneously. This suggested us to develop and test a novel method to assess interactions between hazard perception and situational factors at high-risk crash sites in voluntary Pakistani drivers (Study V). 5.1 Study III: Situational factors at traffic crash sites: a case-control study on Yaoundé-Douala road section, Cameroon Situational factors play an important role in determining risk and severity of an RTC [60]. Improving road design for instance, can decrease crash risk by increasing road network ability to compensate for driving errors [62]. Similarly, installation of side impact barriers at curves or removal of solid objects reduces crash severity. It is estimated from crash data of HICs that implementing road interventions can reduce 20% of the preventable RTCs [49]. Installation of road interventions in LMICs is often not effectively advocated, due to underreporting of involved situational factors at crash sites [50]. As discussed earlier, police reports are the most common RTC surveillance mechanism in LMICs, but they tend to focus only on road user-related crash factors (Figure 5, page 19). In countries like Botswana, India, and Zimbabwe, road factors were reported in only one percent of crashes [7]. Our results from Cameroon showed that road situational factors were identified in less than 5% of interurban road crashes. A similar proportion was observed for interurban crashes in Pakistan [89]. These proportions were certainly less than the expected involvement of such factors in crashes as shown in the USA and Great Britain [55]. Moreover, previous traffic safety research in LMICs focused more on transient factors such as crash time or adverse weather conditions [90]. It was observed that, when situational factors were reported as a crash cause, adverse weather or reduced visibility were identified in over half of these crashes [89]. In LMICs, police are not trained to report road factors, information essential to advocate and implement local as well as area-wide road interventions [7, 91]. We found only one cross-sectional study from Brazil, showing that road surface conditions were significantly associated with injury crashes on interurban roads [21]. Involvement and contribution of modifiable situational factors other than weather conditions is rarely investigated in LMICs, particularly using case-control methods [90, 92].

37

Objectives The objectives of this study were:

1. To assess situational factors associated with injury crashes sites on an interurban road section in Cameroon.

2. To assess attributable risk proportion for such crash factors.

This study has been published as: Bhatti JA, Sobngwi-Tambekou J, Lagarde E, Salmi LR. Situational factors associated with road traffic crashes: A case-control study on the Yaounde-Douala road section, Cameroon. International Journal of Injury Control and Safety Promotion 2010 Mar 30:1-8 (Appendix 5). I contributed principally in data collection from videos, analysis, interpretation of results, and manuscript writing. Methods Design and setting This case-control study was conducted on the Yaoundé-Douala road section. This 243-km, undivided, mostly two-lane road connects the two most populated cities (over one million inhabitants each); Yaoundé at 750 m above sea level and Douala, the port city. This road section passes through several smaller towns (<20 000 inhabitants), except for Edéa (Figure 11). This explains the low traffic counts of 2 269 vehicles per day on central road sections, and around 3 553 vehicles per day near cities [73]. The road section from Yaoundé to Edéa passes mostly through hilly regions, while from Edéa to Douala it passes mostly through plain regions. Case and control sites Cases were defined as sites involved in an injury traffic crash that occurred on this road section during the period from Jan 2004 to May 2007. Police reports collected for the road burden assessment (see previous study) were used to select these sites. All sites were localized during subsequent visits conducted with help of one of the police officers who had attended the crash. A site where several crashes occurred during study period was considered as a single site. For each case site, a control site was selected by drawing a random number of km towards Douala: a random integer between 1 and 10 was read from a random number table and added to the kilometre location of the case site. This strategy was chosen to account for the large observed variations in average daily traffic counts in various road sub-sections. Indeed, a random sampling of cases over the whole road under study would have led to sub-section imbalances between the number of cases and controls. Measures The same set of situational variables was collected from case and control sites in two steps: (i) At the time of the site visits using a standardized coding sheet adapted from the French accident analysis reports coding sheet [9]. This step included information on built-up or rural area, horizontal and vertical profiles, road width, surface regularity, verge slope, depth at 10 m from the verge, location and type of nearby obstacles (within a road distance of 50 m in each direction), horizontal marking and presence of an intersection. This data was collected

38

under supervision of Dr. Sobngwi-Tambekou and Dr. Lagarde. (ii) A second set of variables was generated from the examination of the video recording of the road section, taken during daylight hours. Case and control sites were identified in the video using their Global Positioning System (GPS) coordinates as recorded both in the video and on the standardized coding sheets. Data recorded from video watching included: number of lanes, side markings, presence of road drain, type of intersection (four-legged, three-legged, and access), intersection control (sign, signals), intersection treatment (speed calming, improving visibility by angling the entering side road), speed control measures (speed sign, hazard sign, road sign with rumble strips), the maximum in both directions of the length of straight section (length of closest straight section if the site is on a curve), important structures (housing and other non-shop buildings, shops, toll plaza). I conducted this second step of data extraction from video recording, blindly to the case or control status of the sites. Analysis Situational variables, with the exception of intersection type, important structures at roadside, and speed control measures, were grouped into two categories. The maximum in both directions of the length of straight section at or near the crash and control sites was measured to account for the potential for vehicles to gather speed. For intersections and sections before curves only, the variable indicated whether the maximum in both directions of the straight section was above vs. below or equal to this value of 85th percentile of this maximum length of straight section. Associations between injury crash site (outcome variable) and situational factors (independent variables) were determined using logistic regression models. Because controls for cases occurring in the section closer (the last 10 km) to Douala could not be selected, the data set was not analyzed as an individually matched case-control study. To control for variations in traffic density throughout the road section, the road was divided into 10 sub-sections of 25 km in length from Yaoundé to Douala and the corresponding nominal variable was entered as a road section random effect in the models. We fitted three multivariate models including all variables significantly associated (p<0.05) with crash risk, including significant interactions, using a stepwise backward selection strategy. Model 1 included only variables coded during site visit, model 2 included only variables coded from video recordings, and model 3 included all variables. Attributable risks were estimated with a method that remains valid when confounding exists [93]. For a given characteristic associated with crash risk, the attributable fraction is pd*(RR-1)/RR, where pd is the proportion of cases (crash sites) exposed to the characteristic and RR the adjusted relative risk of this risk factor. Crash being an infrequent event, we estimated adjusted RR by the adjusted OR (aOR) estimated by model 3. All analyses were performed on SAS version 9.2 [75]. Results Overall, 656 injury crashes occurred at 554 case sites, for which 554 control sites were randomly selected. Eleven of them were located beyond the road section and were therefore excluded, leaving 543 control sites. For logistical reasons, the last 10-km section of the road section (near Douala) was not recorded, leaving 474 case sites and 509 control sites with data extracted from video. No significant spatial autocorrelation for selection was observed for

39

case and control sites, as estimated from Moran’s I statistic (P ≥ 0.12). High crash site density was observed near Yaoundé and on the Edéa-Douala road section (Figure 11).

Figure 11. Injury crash site density along 25-km stretches of Yaoundé-Douala road section

1.3-1.4

1.5-2.5

2.5-6.4

Douala

YaoundéEdéa

Pouma Boumnyebel

Mbankomo

Site/kmInhabitants per city (,000)

>200

>20-200

<2025 km

N

Situational factors such as flat and straight road profile, null verge depth, built-up areas, irregular road and shoulder conditions, nearby roadside obstacles, lane width > 8 m, and intersections were significantly more frequent at case sites than controls (Table 9). Video scrutiny showed that shops and schools were more frequent at case sites than controls. No associations were observed between lengthy straight sections (>872 m) and case sites. In the model including only variables from site visits, crash sites were more likely to be located at intersections than control sites. Similarly, the crash sites were more likely to be located in built-up areas as compared to control sites. Similarly, likelihood of crashes on flat road sections increased when the road was wider than 8 m as compared to narrow road sections. Moreover, injury crash likelihood increased in presence of nearby obstacles when road surface was irregular than regular. In the model including only variables from video, crash sites were more likely near shops than control sites. In the model including both types of variables (model 3), crash sites were more likely to be located on road sections with flat road profiles, irregular surface conditions, near (< 4 m) solid obstacles, except for crash barriers, and at three- and four-legged intersections (Table 9). Furthermore, crash likelihood increased in built-up areas when verge depth was null as compared to verge depth more than 0 m. Attributable crash risk proportions were 21.9% for flat road profiles, 11.3% for irregular surface conditions, 5.5% for nearby (< 4 m) obstacles, 4.6% for three-legged intersections, and 3.4% for four-legged intersections. Attributable crash risk proportion was 16.1% for built-up areas as estimated from model 1.

40

Table 9. Situational variables at case and control sites on Yaoundé-Douala road section, Cameroon Proportion Proportion P Model 1 Model 2 Model3 Attributable among controls among cases aOR* [95% CI] aOR* [95% CI] aOR* [95% CI] risk (%) Variables from site visit N=543 N=554 Flat vs. hill road profile 53.0 64.1 <0.01 1.52 1.15 - 2.01 21.9 0 m vs. > 0 m verge depth (10 m from verge) 39.8 46.2 0.03 Straight vs. curved road section 37.2 44.2 0.02 Built-up vs. rural road section 35.0 48.9 <0.01 Irregular vs. regular surface conditions 30.6 37.9 0.01 1.43 1.04 - 1.99 11.3 Continuous road-marking vs. intermittent 28.1 28.9 0.79 Verge slope > 70° vs. ≤ 70° 23.0 23.3 0.91 Lane width > 8 vs. ≤ 8 m 10.1 17.2 <0.01 Irregular vs. regular road shoulder 6.5 12.3 <0.01 Roadside obstacle at <4 m vs. > 4 m § 4.2 11.0 <0.01 1.99 1.09 - 3.63 5.5 Intersection vs. no intersection 3.5 11.2 <0.01 2.30 1.29 - 4.08 Built-up vs. rural road section - if verge depth 0 m 2.65 2.12 - 3.32 2.33 1.97 - 2.77 - if verge depth > 0 m 1.49 1.02 - 2.19 1.22 0.79 - 1.87 16.1† Lane width > 8 vs. ≤ 8 m - if flat road profile 4.19 2.84 - 6.19 - if hill road profile 0.88 0.50 - 1.58 Roadside obstacle at < 4 m vs. > 4 m ‡ - if regular road surface conditions 1.46 0.80 - 2.66 - if irregular road surface conditions 18.54 6.61 -51.93

Variables extracted from video recordings N=509 N=474 2 vs. > 2 lanes 91.9 90.1 0.46 Speed control 0.02 - Speed sign 15.8 17.5 - Hazard sign 14.8 20.7 - Road sign with rumble strips or speed blocks 2.2 1.9 Length of straight road section (>85 vs. < 85 percentile) § 8.2 8.8 0.73 Side-road markings 5.1 6.1 0.15 Intersection control vs. none (signs, signals, blocks) 3.7 8.9 0.01 Presence of a road drain 2.9 2.8 0.77 Intersection treatment║ 2.6 4.2 0.18 Important structures at roadside <0.01 - Shops 11.8 19.0 1.76 1.06 - 2.09 - Houses and other non-shop buildings 8.5 6.8 0.66 0.37 - 1.17 - Check points 2.6 3.0 1.36 0.55 - 3.39 - Bridge/train track 1.4 1.9 0.48 0.10 - 2.42 - School 1.2 2.5 1.81 0.63 - 5.19 Intersection <0.01 - Access to shops or homes 11.4 16.9 1.62 1.00 - 2.64 1.40 0.90 - 2.17 - Three-legged 2.6 6.8 3.43 1.63 - 7.21 3.11 1.15 - 8.39 4.6 - Four-legged 1.3 4.9 3.74 1.39 -10.14 3.23 1.51 - 6.92 3.4 Model 1 Association between road traffic crash sites and situational variables collected during the site visit. Model 2 Association between road traffic crash sites and situational variables collected from the video. Model 3 Association between road traffic crash sites and situational variables collected from both the methods * aOR- adjusted odds ratio, from logistic regression with road section random effect

† Estimated from model 1, using proportion of case sites in built-up areas ‡ Except for crash barrier and in a 100-m stretch § Maximum of both directions before intersections and curves ║ Speed calming, improving visibility by angling the entering side road

41

5.2 Study IV: Burden and factors associated with highway work zone crashes, Karachi-Hala road section, Pakistan With aging of highways, road authorities spend a considerable proportion of their budget on their preservation [94]. For instance, federal and state highway departments of transportation in the USA invest 10 to 15% of their annual budget on the maintenance of highways, amounting to tens of billions of US$ each year [95, 96]. Similarly, highway maintenance becomes an essential component of development fund spending in LMICs [97, 98]. These construction zones, often named as Highway Work Zones (HWZs), are present on road networks in all countries [99, 100]. In HWZs, road lanes, which are normally available to accommodate the actual traffic flow, are closed, shifted, or encroached upon for construction purposes [99]. Although it is often better to provide detours to the commuters, this remains impractical in highway settings [99, 101]. Thus, traffic flows need to be restricted giving rise to safety challenges [100]. For instance, 63% of fatal crashes and one-third of injury crashes took place on HWZs of the two-lane highways in Kansas [100]. In the US, the estimated cost of HWZ crashes between 1995 and 1997 was 6.2 billion US$, with an average cost of 3,687 US$ per crash [102]. In HICs, traffic safety issues related to HWZs have been studied in detail, and appropriate traffic control interventions are put in place before the construction begins [103]. In Pakistan, the interurban network of over 8 000 km suffers extensively from wear and tear mostly due to overloading, heavy traffic, and delayed maintenance [81]. The pavement-condition survey conducted in 2001 showed that 50% of the National interurban road network is in need of major rehabilitation [81]. The maintenance demands have been consistently increased from 10 billion Pakistani rupees (PKR) in 1991 to over 30 billion PKR in 2005, yet slightly over 10 billion PKR were spent in 2005 for highway maintenance [81]. It is highly likely that a significant proportion of the current road network undergoes maintenance, but no data is available to estimate the traffic safety or exposition related to such conditions. The case of Pakistan is certainly not different from most LMICs, as almost no research has been carried out to assess this problem in these countries [3, 90]. Objectives The objectives of this study were:

1. To assess burden of HWZ crashes on an interurban road section in Pakistan.

2. To assess factors associated with such crashes in Pakistan.

This study is under second revision in Injury Prevention: Bhatti JA, Razzak JA, Lagarde E, Salmi L.-R. Burden and factors associated with highway work-zone crashes, Karachi-Hala road section, Pakistan. Injury Prevention (Appendix 6). I have been involved in all steps of this study, from conception, to data collection, analysis, interpretation of results, and manuscript writing.

42

Methods Study design and setting Study design was based on a historical cohort study [104, 105]. Highway police crash reports and traffic statistics for a three-year period (2006-2008) were used to assess the injury risks in HWZs. The study setting was a 196-km long four-lane, separated, section of the Karachi-Hala road in the province of Sindh, Pakistan (Figure 12). This is one of the most heavily used road sections, with traffic counts ranging from 16,356 to 24,707 vehicles per day [83]. The NHA Pakistan manages maintenance and upgrading operations on this road section. The NHMP have been enforcing traffic rules on this road section since 2004.

Figure 12. Karachi-Hala Road Section, province of Sindh, Pakistan

Arabian Sea

IndiaBaluchistan

India

China

Afghanistan

Arabian sea

Selected road

Other roadBoundries

Cities

Traffic data Annual average daily traffic survey data were collected from NHA headquarters. These surveys are conducted each year to assess traffic counts on different road sections under Federal administration. Locations near toll plazas are selected to assess 24-hour counts by NHA personnel. We extracted information on traffic counts observed between Karachi-Hyderabad (146 km) and Hyderabad-Hala (50 km) sub-sections. Variables included in traffic surveys were number, type (trucks, buses transporting ≥ 20 passengers, mini-truck, minibus or coasters transporting < 20 passengers, cars or jeeps, and motorcycles), and direction of vehicles (North-bound or South-bound) [83]. In Pakistan, during maintenance works on separated interurban roads, two or more lanes in a given direction are completely blocked and traffic, on most occasions, is diverted to oppositely directed lanes (Figure 13). Police and highway authorities facilitate private contracting agencies in traffic control during construction periods. Details of HWZ start and end dates and km locations are recorded in their memos. We collected this data from NHA and NHMP regional offices.

43

Figure 13. Examples of normal traffic zone (A) and work zone (B) on interurban road

section in the province of Sindh, Pakistan

A

B

Crash data After a crash, a NHMP patrolling officer files the details of crash on a standard four-page accident analysis report [85]. A copy of this report is kept in the regional office, whereas the original is sent to the NHA headquarters. Moreover, each crash is recorded on a separate accident register in the regional office [85]. All police crash reports and registers from Jan 06

44

to Dec 08 were retrieved and photocopied from regional NHMP offices with permission of the officer in charge. Variables coded from accident registers included date, time, number and type of involved vehicles, number of persons injured or who died in a reported crash, and whether crash occurred during maintenance works. Variables coded from crash reports included date, time, location, direction of lane (North-bound or South-bound), light, weather, horizontal and vertical road profile, road surface and shoulder condition, ongoing maintenance, and crash cause [106].Type of crash was defined as single vehicle, same direction, opposite direction, sidewise, pedestrian. When more than one type was identified, crashes were coded as crash of most vulnerable involved road user; the vulnerability decreasing order was: pedestrian; opposite direction; sidewise or at intersection; single vehicle; same direction [106]. Details on number, injury severity, and type of road user involved (pedestrian, riders of two-wheelers, or occupants of cars/jeeps, minibuses, buses, or trucks) were coded separately. Severity was defined as ‘severe’ when involved person was transported to hospital and ‘fatal’ when involved road user died at crash scene or at hospital in first 24 hours of the event [85]. Analysis Information on crashes from registers and reports were linked to make a single file based on crash location (km) and crash date, available for all crashes. Crash, fatality, and severe injury per 109 vehicle-km travelled for vehicle type and direction were computed using traffic counts survey [83]. Proportion of crashes, fatalities, and injuries on HWZ were computed and compared to other zones. Information on HWZ dates was only available for the Hyderabad-Hala sub-section. Due to this limitation, crash, fatality and injury rates for work and normal traffic zone were computed only for the 50-km-long sub-section. Crash, fatality, and severe injury risks according to road directions, vehicle types and traffic conditions were compared using rate ratios with 95% confidence intervals, rate differences, and attributable risk proportions where appropriate [104, 107]. Associations of factors with HWZs crashes were estimated from a multiple logistic regression model, including all variables weakly associated (P<0.2) with HWZs with backward selection strategy [108]. Results Overall injury burden A total of 180 crashes were identified from police registers. Overall, 612 road users were injured in these crashes; 14.8% (N=91) died, and 55.3% (N=339) were severely injured. The crash fatality rate on this highway, excluding HWZ crashes, was 13 per 109 vehicle-km (Table 10). The crash rate was significantly higher in North-bound as compared to South-bound direction (rate ratio (RR) = 1.81; 95% confidence interval (95%CI) = 1.30, 2.52). Compared to trucks, the crash rate was lower for passenger cars (RR = 0.57; 95% CI = 0.42, 0.78), but the fatality rate was twice as high for passenger cars compared to trucks (RR = 1.93; 95%CI = 1.04, 3.61). Similarly, fatality risk was significantly higher for occupants of buses (RR = 3.32; 95%CI = 1.52, 7.22) and minivans (RR = 4.75; 95%CI = 1.84, 12.24), as compared to trucks’ occupants.

45

Table 10. Road crash fatality and injury risk per 109 vehicle-km on the Karachi-Hala

road section, Pakistan (2006-08)

Vehicle-km Crash Fatality Severe injury 109 N Rate* N Rate* N Rate* All (except work zone) 4.84 153 31.6 63 13.0 287 59.3 -North-bound direction 2.40 98 40.8 38 15.8 191 79.6 -South-bound direction 2.44 55 22.5 25 10.2 96 39.3 Vehicle† -Motorcycle 0.19 10 52.6 8 42.1 10 52.6 -Car/jeep 1.90 60 31.6 29 15.3 107 56.3 -Mini-van (<20 passengers) 0.16 11 68.8 6 37.5 33 206.2 -Mini-truck 0.27 16 59.3 7 25.9 21 77.8 -Buses 0.42 39 92.9 11 26.2 105 250.0 -Trucks 1.90 105 55.3 15 7.9 50 26.3 * Per 109 vehicle-km travelled † At least one of the involved vehicles in the crash was from the category. Work-zone crash and injury burden Fifteen percent (N=27) of crashes occurred in HWZs, accounting for 30.8% (N=28) of all fatalities and 15.3% (N=52) of those severely injured. On the 50 km-long sub-section for which work-zone traffic exposition was available, 17.6% of the 0.89 billion vehicle-km travelled over three years were during maintenance works. On average, HWZs were 5.7-km-long (SD = 4.3). Two work zones were 10 and 14-km-long and lasted more than 300 days. Crash (32.5 vs. 31.6 per 109 km travelled) and fatality risk (16.3 vs. 13.0 per 109 km travelled) at normal traffic zone were similar to that of the total road section, whereas severe injury rate (32.5 vs. 59.3 per 109 km travelled) was higher to that of the total road section (Tables 10 & 11). Significantly higher crash (RR=2.35), fatality (RR=4.70), and severe injury risks (RR=1.92) were observed on HWZs than on other zones (P≤0.004) on this sub-section (Table 11). Table 11. Highway work zone crash fatality and injury risk per 109 vehicle-km on 50-km

long sub-section on Karachi-Hala road, Pakistan (2006-08) Vehicle-km Crash Fatality Severe injury 109 N Rate* N Rate* N Rate* Work zone 0.157 12 76.4 12 76.4 27 172.0 Normal traffic 0.738 24 32.5 12 16.3 66 89.4 Rate ratio (RR) 2.35 4.70 1.92 95% Confidence Interval 1.17 ; 4.70 2.11 ; 10.46 1.23 ; 3.01 Rate difference 43.9 60.2 82.5 Attributable proportion (%) 57.4 78.7 48.0

* Per 109 vehicle-km Factors associated with HWZ crashes on the 196-km-long section Complete reports were available for 93.3% (N=168) of all traffic crashes, 96.8% of fatal crashes (N=63), and 98.9% of severe injury crashes (N=88). As compared to crashes between vehicles moving in the same direction, crashes between oppositely moving vehicles and those involving pedestrians were more likely on HWZ than on non HWZ (Table 12). Similar associations (P≤0.01) were observed for both 146- and 50-km-long sub-sections, where pedestrian and opposite direction crashes accounted for most (≥73.2%) HWZ crashes. Similarly, wet surface crashes were significantly more likely to occur on HWZ than on

46

normal zones. However, this association was not observed when these two sub-sections were analyzed separately. Moreover, as compared to other traffic crashes, hazardous overtaking was the most common cause of HWZ crashes (55.6% vs. 7.1%). Table 12. Factors associated with work-zone crashes on the 196-km-long Karachi-Hala

road section, Pakistan (2006-08) Work-zone crashes Other crashes P Adjusted 95%CI* N=27 N=141 odds ratio N (%) N (%) Severity 0.003 - Mild or no injury 0 0.0 20 14.2 - Severe injury 10 37.0 77 54.6 - Death 17 63.0 44 31.2

Crash type <0.001

- Same direction 4 14.8 67 47.5 1 - Opposite/sidewise 17 63.0 25 17.7 10.65 3.22 - 35.25

- Pedestrian 5 18.5 13 9.2 6.03 1.39 - 26.20 - Single vehicle 1 3.7 36 25.5 0.25 0.02 - 2.94 Light 0.29 - Daylight 11 40.7 73 51.8

- Night 16 59.3 68 48.2 Surroundings 0.41

- Built-up 13 48.2 80 56.7

- Rural 14 51.8 61 43.3

Horizontal profile 0.19 - Straight 21 77.8 123 87.2 - Curve 6 22.2 18 12.8

Vertical profile 0.09 - Plain 26 96.3 119 84.4

- Slope 1 3.7 22 15.6 Road surface 0.47 - Regular 24 88.9 131 92.9 - Irregular 3 11.1 10 7.1

Shoulder surface 0.72 - Regular 25 92.6 133 94.3 - Irregular/absent 2 7.4 8 5.7 Intersection 0.60

- Yes 1 3.7 9 6.4

- No 26 96.3 132 93.6

Surface condition 0.08 - Dry 23 85.2 134 95.0 1 - Wet 4 14.8 7 5.0 7.26 4.15 - 48.8

Cause† ‡

- Overtaking 15 55.6 10 7.1 - Sudden entry 5 18.5 7 5.0 - Bad weather 4 14.8 9 6.4

- Drowsiness 3 11.1 29 20.6 - Speeding 1 3.7 17 12.1

- Tyre-burst 0 0.0 20 14.2

* 95% confidence interval † A crash may be designated with two causes ‡ Not included in multivariate analysis.

47

5.3 Study V: Road hazard perception at high-risk crash sites in voluntary Pakistani drivers Traffic crashes are unevenly distributed along interurban road networks [3]. They occur in clusters at single sites , often called high-risk sites or zones, along particular sections of the road [109]. They can be defined as sites having a higher expected number of crashes than other similar sites [110]. Theoretically, inadaptability of driving behaviour to local road and traffic hazards leads to crashes at these sites [109]. It has been documented that design improvement at these sites can result in significant decreases in crash risk [11]. However, this remains an expensive option and not all high-risk sites can be improved in a timely manner [111]. Hazard perception is the driver’s ability to identify potential hazardous situations and taking necessary actions to avoid them [112, 113]. Driver-related factors such as age, sex, familiarity with road, driving experience, attitudes, and self-assessment of skills can influence this ability [114-119]. Road elements such as sharp bends, decreased widths, and presence of lane markings can increase hazard perception [120, 121]. Previous research has shown that augmenting hazard perception by driver training or by implementing appropriate road furniture could significantly reduce the likelihood of RTCs [122, 123]. Much work on high-risk crash sites focused on identifying these sites using statistical methods so that safety work could be prioritized [124]. Interactions between driver- and site-related factors has not been investigated in detail, particularly for the interurban road settings in LMICs [49]. To our knowledge, the hypothesis that high-risk crash sites might not be perceived as dangerous by drivers has not been tested as yet. Insight into how such sites are perceived by drivers could be useful in developing and implementing less expensive interventions, particularly in LMICs [116]. Objectives The objectives of this study were:

1. To compare hazard perceptions of high-risk crash sites to those not involved in crashes in voluntary drivers.

2. To assess driver- and site-related factors associated with hazard perception level.

3. To assess whether associations between factors and hazard perception were different

at high-risk crash sites and at sites not involved in crashes. The manuscript of this study is currently prepared for submission to Health Psychology: Bhatti JA, Razzak JA, Lagarde E, Sobngwi-Tambekou J, Alioum A, Salmi L.-R. Hazard perception at high- and low-risk crash sites (Appendix 7). I have been principally involved in study conception, analysis, interpretation of results, and manuscript writing. Site and video data from Cameroon was collected under supervision of Dr. Sobngwi-Tambekou and Dr. Lagarde. Site, video, and participant data from Pakistan was collected under my direct supervision.

48

Methods Study design and settings The study settings were interurban road sections situated in Cameroon and Pakistan: 1/ Karachi-Hala road section in Pakistan (196-km-long mostly four-lane separated road), and 2/ Yaoundé-Douala road section in Cameroon (243-km-long mostly two-lane non-separated road). A matched strategy was used to select sites. ‘High-risk sites’ were those involved in three or more RTCs in a precedent three-year period, whereas ‘low-risk sites’ were those not involved in a RTC, during the same period. For each high-risk site, a low-risk site was randomly selected on the same road section. Hazard perception was assessed by showing videos of these sites to voluntary Pakistani drivers. Ethical approval of the study was obtained from the Aga Khan University Ethics Research Committee in May 2009 (Reference ERC/2009/1179). Site selection In Pakistan, NHMP regional office was visited in the month of June 2009. Crash reports and registers for the three-year period from Jan 1, 2006 to Dec 12, 2008 were retrieved and photocopied. High-risk sites with given kilometre location were then identified with GPS coordinates with help of a police officer. Similarly, traffic police offices in Cameroon were visited and such sites were subsequently identified in June 2007. The two road sections were filmed from a four-wheeled sedan car moving within the authorized speed limit (July 2009 in Pakistan and July 2007 in Cameroon). All high- and low-risk sites were then identified by linking GPS coordinates to the videos. For each high-risk site, a low-risk site was randomly selected out of all sites which were not involved in crashes on the same road section. Video sets To measure hazard perception, video of sites were cut so that each video showed a 500 metre-long-road section during 30 seconds, including the last 100 m corresponding to the high- or low-risk site (Figure 14). Further, a yellow indicator blinked five times to help drivers identify the site for which they had to emit a judgement on hazard perception right after video projection. We determined sample size to be 26 pairs of sites, assuming that 95% of the high-risk sites would be identified as dangerous and 80% of the low-risk sites as not dangerous with a precision of 7.5 [125]. Participant selection Participants were Pakistani nationals residing in Karachi, aged 18 years or more, with a valid driving permit, who had driven a motorized vehicle on the Karachi-Hala road section in the previous seven days. Random sampling was not possible because of heavy-traffic and higher speeds conditions on this road section [30]. Thus, a convenience, but representative, sampling method was used to recruit 100 drivers. For this, we determined the drivers’ sex and vehicular distribution by observing traffic from a pilot study (N=5 496). It was observed that cars accounted for 39.1%, heavy trucks for 36.5%, minibuses and mini-trucks for 7.8%, buses for 9.6%, and motorcycles for 6.3% of the vehicles entering Karachi (Appendix 8). Distribution of cars and heavy vehicles was similar to that recorded by highway authority [83]. Almost all drivers were men (99.9%). Based on these findings, personal vehicle male drivers were invited from a roadside gas station at start of the highway near Karachi, and commercial vehicle drivers were invited from transport company offices at six different locations in Karachi. Data collection Face-to-face interviews with drivers were conducted in Urdu language. Questions were developed from an English language questionnaire, using back translation, independent linguistic verification, and testing on five drivers. Interviews were either conducted at the Aga

49

Khan University (AKU) Campus or at the company offices in separate rooms. Driver-related variables included socio-demographic variables (age, sex, marital status, education, and employment), whether driving permit was issued without practical test, frequency of reported risky driving behaviours (sleepy driving, mobile phone use while driving, seat-belt use, traffic tickets, DWI during last three months), and involvement in RTC during last one year. Using 17-inch video screens, five test videos (three from Pakistan and two from Cameroon) were shown to drivers before presenting selected sites. The order of sites was randomly drawn for each participant. To avoid confusion from right- and left-hand driving practiced in Cameroon and Pakistan, site videos from Cameroon followed those from Pakistan. For each video shown, drivers were asked to report their perception of site and traffic, on a four-level scale; 1) Certainly safe; 2) Probably safe; 3) Probably dangerous; 4) Certainly dangerous (Figure 14). Further, they were asked to record their preferred speed (in km/h) for each site. Each site was characterized by the main investigator, using definitions used in our previous study conducted in Cameroon [126]. Site-related variables assessed were built-up or rural area, horizontal and vertical road profile, road width, surface regularity, verge slope, depth at 10 m from the verge, location and type of nearby obstacles (within a road distance of 50 m in each direction), horizontal marking, vertical road signs, and presence of an intersection or a U-turn [126]. Traffic-related variables assessed were traffic moving in same or opposite direction, visible pedestrian, motorcyclist, or heavy vehicle, rain or wet surface, manoeuvring vehicle (crossing or overtaking), and number of lanes [127].

50

Figure 14. Picture extracted of a high-risk site video and related questions, from the

Karachi-Hala road section

54

51

Analysis Proportions of site- and driver-related characteristics were computed. Discordance (D) of appreciation for a matched high- and low-risk site was defined as “minor” when difference of hazard perception level was one on the Likert scale and “major” when level difference was more than one. Positive sign (D+) was used to show that hazard perception was higher for the high-risk site than its matched low-risk site, and negative sign (D-) to show that hazard perception was lower for the high-risk site than its matched low-risk site. Wilcoxon test was used to assess whether these discordances were significantly higher or lower for high-risk site than low-risk site. Similarly, differences in reported speeds for matched high- and low-risk site pairs were compared using a paired t test. Correlations between reported speeds for high- and low-risk site pairs were assessed by intra-class correlation coefficient (ICC). Associations of driver-, site-, and traffic-factors with road hazard perception level1 were assessed using logistic regression with a backward selection strategy including significant (P<0.05) variables and interactions with risk status (high- or low-risk site) [108]. To assess whether these associations remained significant while adjusting for site- and driver-related factors, two other models were constructed including the site and participant identification as random effects: Model 1:

iiiikiiiiki TVriskSVriskDVTVSVcountryriskRHPLogit **)( 76543210, ββββββββ +++++++= Model 2 with site as random effect:

iiiiikiiiiki bTVriskSVriskDVTVSVcountryriskRHPLogit 076543210, **)( ++++++++= ββββββββ

Model 3 with driver as random effect:

kiiiikiiiiki bTVriskSVriskDVTVSVcountryriskRHPLogit 176543210, **)( ++++++++= ββββββββ Where i=1,2,3…52 videos; k=1,2,3…100 drivers; RHP, road hazard perception, ‘0’ for safe and ‘1’ for dangerous; risk, ‘1’ for high-risk site and ‘0’ for low-risk site; country, ‘0’ for Pakistan and ‘1’ for Cameroon; SV, site-related variables; TV, traffic-related variable; DV, driver-related variable; b0, random effect for site; b1, random effect for driver. The association of multiple-category variables (age and vehicle driven) with hazard perception was assessed using log-likelihood test for model 2 and 3 [108]. Results Sites Out of 131 crash sites identified in Pakistan, 16 were involved in three or more crashes. Similarly, out of 474 crash sites identified in Cameroon, 18 were involved in three or more crashes. We randomly selected 10 high-risk sites. In Pakistan, most high-risk sites were with straight road profile (87%) whereas this proportion was 20% in Cameroon (Table 13). Road surface conditions were irregular on most high-risk sites in Pakistan (75% vs. 65%) and Cameroon (90% vs. 40%) than low-risk sites. Almost half of high-risk sites in Pakistan (62%) and Cameroon (50%) were with flat road profile. A vertical road sign was visible at 38% of high-risk sites in Pakistan and at 40% such sites in Cameroon. Fewer high-risk sites were 1 For these analyses, road hazard perception categories were regrouped as ‘safe’ when driver reported his perception to be ‘certainly safe’ or ‘probably safe’ and ‘dangerous’ when he reported his perception to be ‘probably dangerous’ or ‘certainly dangerous.’

52

located in built-up area in Pakistan (31%) than in Cameroon (50%). Similarly, one third of high-risk sites were at intersection in Pakistan (31%) and Cameroon (40%). In Pakistan, 19% of the high-risk sites were situated at a U-turn and on 13%, maintenance works was ongoing.

Table 13. Characteristics of high- and low-risk sites on Yaoundé-Douala and Karachi-Hala road sections

Pakistan Cameroon High risk (%) Low risk (%) High risk (%) Low risk (%) Characteristics (N=16) (N=16) (N=10) (N=10) Site factors Straight 87 87 20 50

Irregular road shoulder 81 81 100 90

Irregular surface conditions 75 63 90 40

Lane width ≤ 8 75 100 80 90

Flat 62 81 50 60

Visible road side obstacle 50 38 100 90

Vertical road sign 38 19 40 10

Built-up road section 31 56 50 10 Intersection 31 31 40 0

U-turn 19 19 0 0

Diversion 13 6 0 0

Continuous road-markings 13 0 50 10

Traffic factors

Visible heavy vehicle 88 69 40 40

Same direction moving vehicle 81 88 60 60

Manoeuvring vehicle (overtake, crossing) 69 56 30 10

Opposite direction moving vehicle 25 19 40 50

Visible pedestrian 19 31 50 20

Visible motorcyclists 13 19 10 0

Rain 0 0 20 20

Wet surface 0 0 70 50

Participants Out of 100 participants, 44 were interviewed at the AKU. Most participants were aged between 26-45 years and one fifth had received no education (Table 14). While all drivers lived in Karachi, 46 of them had a residence in other regions of Pakistan as well. Seventy-four drivers reported either not wearing a seat-belt at all or wearing it occasionally, and 92 of them reported a cell phone use while driving. No significant association was found for any of the driver-related factor or interview location with recent crash history.

53

Table 14. Characteristics of Pakistani drivers included in sample (N=100).

Proportion (%) (N=100)

Recent crash (%) (N=20)

Age (y) -18-25 9 5 -26-35 38 30 -36-45 28 45 - > 45 25 20 Vehicle driven -Truck 33 30 -Motorcar 43 45 -Bus 11 0 -Mini-bus 6 15 -Mini-truck 5 5 -Motorcycle 2 5 Permanent domicile -Karachi 54 50 -Sindh 9 20 -Punjab 29 20 -NWFP/Baluchistan 8 10 Education (y) - None 22 35 - 1-5 22 10 - 6-10 37 40 - >10 19 15 Profession -Driver 85 85 -Other 15 15 Married 88 80 Familiar with road 83 90 Licensed after test 45 50 Seat-belt use -None 27 20 -Occasional 47 55 -Frequent 26 25 Sleepy driving 29 40 Phone dialling 84 80 Phone receiving 92 85 Traffic Ticket 49 50 Drunk driving 2 10

Hazard perception of crash and non crash sites In twelve site pairs, five from Pakistan and seven from Cameroon, road hazard perception was significantly higher for high-risk than low-risk sites and reported preferred speeds were significantly lower for high-risk than low-risk sites (Table 15). Correlations of pair-wise reported speeds were moderate to high (0.51≥ICC≤0.95). The highest negative speed differences (> 25 km/h) were observed for pairs 4 (toll plaza at high-risk site), 16 (built-up area with markets and traffic at high-risk site), and 24 (a curve, with rain, oil tanker, and parked vehicle on high-risk site) (Appendix 8). Most high-risk sites where site hazard perception was not different or lower than at low-risk sites were straight (N=10) and flat (N=8).

54

Table 15. Differences in hazard perception, and reported preferred speeds for high- and

low-risk site pairs on Yaoundé-Douala and Karachi-Hala road sections Road hazard perception Traffic hazard perception Difference in preferred speed D+ D++ D- D-- P D+ D++ D- D-- P Mean P ICC 95% CI Pakistan

1 16 37 4 5 <0.001 1 1 11 27 <0.001 -4.50 0.021 0.81 0.72, 0.87 2 8 14 6 15 0.838 9 24 3 11 0.001 -5.55 0.002 0.84 0.76, 0.89 3 6 10 7 28 0.002 4 28 8 9 0.016 3.50 0.051 0.81 0.71, 0.87 4 10 82 0 0 <0.001 7 60 1 7 <0.001 -43.15 <0.001 0.60 0.40, 0.73 5 1 5 1 3 0.566 6 6 1 1 0.015 -5.00 <0.001 0.95 0.93, 0.97 6 7 8 12 37 <0.001 0 4 5 17 0.005 12.75 <0.001 0.83 0.75, 0.89 7 11 26 8 13 0.024 3 2 11 26 <0.001 1.85 0.240 0.85 0.78, 0.90 8 18 30 2 7 <0.001 2 4 11 45 <0.001 14.35 <0.001 0.51 0.27, 0.67 9 6 9 12 23 0.010 2 4 11 22 <0.001 7.15 <0.001 0.90 0.84, 0.93

10 11 17 9 21 0.616 6 11 8 23 0.075 2.10 0.238 0.82 0.73, 0.88 11 16 24 8 7 <0.001 16 25 7 6 <0.001 -6.45 <0.001 0.82 0.73, 0.88 12 3 3 13 37 <0.001 2 2 16 22 <0.001 16.50 <0.001 0.83 0.75, 0.89 13 0 4 9 49 <0.001 2 6 2 8 0.545 10.25 <0.001 0.84 0.76, 0.89 14 9 13 14 27 0.010 10 32 3 7 <0.001 -4.85 0.007 0.85 0.78, 0.90 15 12 51 1 3 <0.001 7 48 3 4 <0.001 -15.45 <0.001 0.88 0.82, 0.92 16 10 68 0 1 <0.001 13 75 0 1 <0.001 -25.01 <0.001 0.86 0.80, 0.91

Cameroon 17 12 14 3 3 <0.001 6 50 0 4 <0.001 -5.55 0.001 0.80 0.70, 0.86 18 14 17 7 9 0.114 0 1 6 72 <0.001 14.10 <0.001 0.88 0.82, 0.92 19 11 5 1 1 <0.001 4 8 0 12 0.685 -8.65 <0.001 0.90 0.85, 0.93 20 11 11 11 10 0.844 1 2 5 67 <0.001 11.10 <0.001 0.88 0.82, 0.92 21 15 20 2 1 <0.001 9 65 0 0 <0.001 -12.00 <0.001 0.86 0.80, 0.91 22 12 20 2 4 <0.001 4 16 6 15 0.904 -3.95 0.007 0.88 0.82, 0.92 23 4 0 10 23 <0.001 4 1 8 37 <0.001 14.75 <0.001 0.89 0.84, 0.93 24 14 49 5 1 <0.001 6 70 2 0 <0.001 -25.09 <0.001 0.84 0.76, 0.89 25 15 20 4 4 <0.001 9 77 0 0 <0.001 -17.10 <0.001 0.77 0.66, 0.85 26 13 23 10 15 0.068 1 85 0 0 <0.001 -19.55 <0.001 0.84 0.76, 0.89

D+ - Minor discordance: High-risk site was ranked one level more hazardous than low-risk site D++ - Major discordance: High-risk site was ranked two levels more hazardous than low-risk site D- - Minor discordance: Low-risk site was ranked one level more hazardous than high-risk site D-- - Major discordance: Low-risk site was ranked two levels more hazardous than high-risk site ICC-Intra class correlation coefficient CI- confidence interval

Factors associated with hazard perception Driver age and vehicle were significantly associated with site hazard perception. Compared to middle-aged drivers, significantly high hazard perception was reported by drivers aged 18-25 years and those aged 26-35 years (Table 16). Similarly, compared to heavy truck drivers, those driving cars or mini-trucks reported significantly lower hazard perception. Vehicle driven remained significantly associated with hazard perception in models 2 and 3. The association of age with hazard perception was not significant in model 3 (0.05 ≥ P ≤ 0.10). Hazard perception of Cameroonian sites was significantly higher than Pakistani sites. Hazard perception of at sites with irregular surface conditions, at intersections, and with ongoing maintenance works was significantly higher than those without them. Hazard perception of sites with straight and flat road profile was significantly lower than those with curve and slope road profile. Hazard perception of high-risk sites in built-up areas and having road width narrower than 8 m was significantly lower than low-risk sites with same features. Hazard perception of high-risk sites with visible hazard sign or a U-turn was significantly higher than low-risk sites with same features. Hazard perception of sites videos with manoeuvring or oppositely moving vehicles was higher than site videos without them. Hazard perception of site videos with heavy vehicle or motorcycles was significantly lower than site videos without them. Hazard perception for high risk site videos with rain was significantly lower than low risk site videos with same conditions.

55

Table 16. Factors associated with hazard perception of high- and low-risk sites on

Yaoundé-Douala and Karachi-Hala road sections

Model 1 Model 2 Model 3 OR 95% CI OR 95% CI OR 95% CI Driver Age (y) - 18-25 1.93 1.50-2.48 1.95 1.71-2.22 2.09 1.53-2.82 - 26-35 1.21 1.03-1.42 1.20 1.12-1.30 1.23 1.02-1.49 - 36-45 1 1 1 - > 45 1.00 0.84-1.19 1.00 0.91-1.08 1.00 0.81-1.23 Vehicle driven - Truck* 1 1 - Motorcar 0.71 0.62-0.82 0.70 0.65-0.75 0.69 0.58-0.81 - Mini-bus 0.85 0.64-1.13 0.85 0.73-0.98 0.83 0.61-1.16 - Mini-truck 0.68 0.50-0.92 0.67 0.58-0.78 0.66 0.46-0.94 - Bus 1.38 1.10-1.74 1.39 1.24-1.55 1.40 1.08-1.86 Site Cameroon vs. Pakistan 6.53 4.91-8.68 6.88 5.26-9.02 7.61 6.55-8.85 Straight vs. curve 0.72 0.58-0.88 0.69 0.56-0.86 0.69 0.62-0.76 Irregular vs. regular road surface 4.78 3.74-6.09 4.61 3.71-5.75 5.36 4.71-6.11 Flat vs. hill 0.57 0.48-0.69 0.54 0.45-0.65 0.54 0.50-0.60 Intersection vs. none 1.46 1.13-1.90 1.40 1.08-1.82 1.49 1.30-1.73 Work zone 24.34 15.00-39.51 23.33 14.43-37.71 31.82 24.53-41.26 Lane width ≤ 8 m vs. > 8 m - High-risk 0.51 0.43-0.61 0.50 0.37-0.68 0.48 0.42-0.56 - Low-risk 18.48 10.39-32.85 22.64 12.42-41.26 23.57 17.29-32.14 Built-up area vs. rural - High-risk 0.58 0.51-0.68 0.67 0.49-0.90 0.57 0.46-0.69 - Low-risk 2.04 1.51-2.74 2.29 1.69-3.09 2.18 1.86-2.56 Vertical road sign - High-risk 2.75 2.38-3.16 2.64 2.02-3.44 2.94 2.48-3.50 - Low-risk 0.50 0.34-0.72 0.50 0.36-0.70 0.47 0.39-0.58 U-turn - High-risk 8.00 6.36-10.22 7.69 5.09-11.62 9.58 7.50-12.24 - Low-risk 0.62 0.39-0.97 0.65 0.43-0.99 0.60 0.47-0.76 Traffic Heavy traffic 0.37 0.29-0.47 0.41 0.33-0.51 0.34 0.30-0.39 Manoeuvring vehicle 1.91 1.50-2.42 1.82 1.45-2.32 2.01 1.77-2.29 Oppositely coming traffic 2.05 1.56-2.69 2.15 1.64-2.82 2.18 1.88-2.53 Motorcycle 0.53 0.38-0.75 0.52 0.37-0.74 0.51 0.43-0.61 Rain - High-risk 0.19 0.15-0.24 0.17 0.11-0.27 0.16 0.13-0.20 - Low-risk 0.47 0.29-0.78 0.48 0.29-0.79 0.45 0.35-0.58

See Appendix 8 for more details on univariate analyses OR – Odds ratio 95% CI – 95% confidence interval Model 1 – Without random effect Model 2 – Site as random effect Model 3 – Driver as random effect * Motorcyclists were not analyzed separately, as no differences were observed.

56

6. Discussion 6.1 Originality of studies Research methods described here could contribute to the documentation of gaps, impeding development and implementation of specific interventions in LMICs [16]. Availability of traffic exposition from two countries showed an increasingly high road disease burden compared to similar roads in HICs, even with police underreporting documented earlier [9, 64]. Broader implementation of such simple methods, which is not the case at present, could be useful to identify high-risk road networks in LMICs [128]. Nevertheless, a complementary study, that we conducted, comparing the concordance between fatality indicators computed with difference expositions showed that traffic exposition is not reliably collected in LMICs [129]. Vehicle counts on both road sections were based on surveys and were an approximate estimate of traffic exposition. As road burden and the impact of interventions to reduce this burden could not be accurately analyzed without adjusting for traffic exposition [128], reliability and validity of methods employed to collect this information on interurban roads in LMICs needs to be further assessed [130]. Furthermore, we documented involvement of situational factors in RTCs using case control and cohort study methods. Such methods were almost never used for assessing specific road risk factors and their attributable risk proportions in LMICs [3, 92]. Similarly, hazard perception study methods were based on methods for assessing the performance of diagnostic tests, to show the accuracy of drivers in differentiating high-risk sites from the low-risk ones [125]. This method assessed odds of poor hazard perception of high-risk sites with given characteristics compared to low-risk sites with same characteristics [108]. Such a method has two benefits; firstly, it identified high-risk sites with low hazard perception and secondly, it identified those site factors which are perceived less hazardous at high-risk sites than low-risk ones. This information could be useful in developing and implementing specific interventions for such settings. These methods can be applicable to understand the interactions between site- and driver-related factors in HICs and LMICs [49]. 6.2 Comparison with published literature Previous research showed that traffic fatality per population was two to three times higher in LMICs than in HICs [2]. However, comparable traffic safety indicator, fatality per vehicle km, is rarely available for interurban roads in LMICs. Our results consistently showed that interurban traffic fatality and injury per km travelled was dramatically higher on interurban roads in Cameroon and Pakistan than similar roads in HICs [9]. Traffic fatality was 73 per 100 million vehicle-km on the Yaoundé-Douala road section, a rate 35 times higher than for a similar road in the US (≤ 2 per 100 million vehicle-km). Similarly, traffic fatality was 53 per 109 km vehicle-km on the Karachi-Hala road section, a rate 13 times higher than a similar road in France (≤ 4 death per 109 km travelled). Further, an alarming increase in road fatality was observed from 2004 to 2006 in Cameroon. These results clearly point out the need to assess comparable and sensitive traffic fatality indicators in LMICs, so that interurban road burden can be measured reliably over time [105]. This may help to advocate appropriate resources for road interventions in these countries [50]. Both our descriptive studies consistently showed that vehicle occupants accounted for most traffic injuries on these road sections. These results were not surprising, as these vehicles account for half of the traffic on these roads [83]. However, when adjusted for km travelled, fatality and severe injury risk was significantly higher for occupants of light vehicles (cars) and buses as compared to trucks’ occupants. Several factors could explain this high mortality in such road user groups. Speeding was identified as one of the major cause of crashes on

57

these road sections. Our complementary speed measurements (Appendix 2) and other studies in the African region clearly showed that the likelihood for over-speeding is higher for smaller and passenger vehicles [24, 106, 122]. Moreover, a high injury severity can be explained by failure to use seat belts in LMICs [131]. Previous research has shown that poor law enforcement and old vehicles are two important factors limiting the use of seat belts in these countries [131, 132]. In both settings, seat belt use among the injured was not reported. Further, we observed in our pilot study that only one-third of the car drivers wore seat belts in Pakistan. Seat-belt wearing is not mandatory for all vehicle occupants in Pakistan [2]. Most commercial buses do not have seat belts for their driver and passengers [10]. With higher speeds allowed on interurban road sections, injury severity for such occupants can be higher in case of a crash [3]. Thus, there is a need to improve traffic law and their enforcement on interurban roads in LMICs [2]. Our hazard perception study did provide some useful insights with respect to drivers of smaller vehicles. Results showed that drivers of these vehicles were not fully aware of road hazards, as compared to drivers of heavy vehicles. Furthermore, a low seat-belt use and a high cell phone use reflected the low hazard perception observed in those driving such vehicles [133]. This points out the need to improve traffic enforcement as well as the development of hazard perception interventions for drivers of such smaller yet powerful vehicles, for which traffic fatality risks are higher [134]. Emergency care to the injured remains an important Government priority in LMICs [135]. Previous research has clearly shown that essential airway and life-saving equipment were often not available in the hospitals around major road network in LMICs [136]. In this thesis, our focus was mostly to assess the pre-event situational factors involved in crashes. However, we noted that transport of the victims and availability of pre-hospital care is rudimentary on both road sections. In Cameroon, most patients came to hospitals by private means. Similarly, despite the availability ambulance service in Pakistan, still an important proportion of those who attended ED came by their own means. In fact, there was no system of triaging patients according to their severity at site, and decision to shift the patient are either taken bystanders, the police, or unqualified ambulance personals [137]. Moreover, there was no organized trauma care system with designated trauma centres on both road sections, although they are near major cities in their countries. Improving access to skilled care can decrease this high road burden outside urban areas [82]. These results point out the need to assess current trauma care systems along interurban road sections in LMICs, and make policies to use them efficiently to decrease this road burden [138]. In Pakistan, police traffic injury records were compared to ambulance and hospital records. To our knowledge, this type of record linking between different datasets for interurban road injuries has not been done before in a LMIC [3]. This permitted to measure under-reporting in police records for this road sections. We estimated that police identified only one out of five traffic fatalities and one of ten severe injuries on this road section. The traffic fatality underreporting was high on this interurban road compared to one of two fatalities reported by police, estimated in Karachi city [64]. Further, even when ambulance and hospital records appeared to have recorded higher traffic injury numbers, both accounted for only half of traffic fatalities and injuries separately. Most HICs have their specific police based road traffic injury surveillance system with a fair coverage of injury crashes on interurban road [9, 55]. Without such surveillance system, accurate assessment of road burden will remain a problem in LMICs. Moreover, we observed that the United Nations definition for road fatality was not implemented in police data [3]. Clearly, police did not follow injury patients up to 30 days.

58

Although adjustments are possible, overall injury severity of interurban crashes can be higher than urban crashes [82]. Previous research has shown that an important proportion of road fatalities occurred on these road sections in Pakistan [63]. These results suggested that caution should be taken while interpreting traffic injury outcome for interurban road crashes and validity of adjustment models for estimating such traffic fatalities should be assessed in LMICs [139]. We noted differential reporting of involved road user groups in police data as compared to health data [140]. It was observed that overall proportion of vulnerable road users was half in police data compared to health services data. These findings were similar to those observed in a HIC [141]. These results indicated that police reporting needs improvement to avoid such bias, so that interventions to protect vulnerable road users in interurban road settings in LMICs can be effectively advocated [11]. Traffic situation in LMICs is different from HICs because of different traffic mix, poor road conditions, development of built-up areas around interurban roads, and older vehicles [11]. Involvement of situational factors in crashes on two road sections was less than 5%, when police records were used to assess crash factors [55]. However, confronting police reported situational factors with identified crash types showed specific patterns. Crashes involving pedestrians and motorcycles for instance, were higher in built-up areas, at intersections, and near schools as compared to the overall crashes. Moreover, excessive speed was frequently observed in crashes at intersections. Similarly, running-off-the-road crashes were more frequent on road sections with curve and flat profiles. Certainly, most police reports in both countries focused on identifying road user error and violations [7]. Information on situational factors is rarely interpreted in LMICs. The standardizing of information using a grid can be useful to assess the involvement of situational factors in crashes in these countries. Moreover, it was observed that there was no system for identification of hazardous sites on neither road sections. Safety audits, which are supposed to be part of road projects, as recommended by the World Bank and other donor agencies, are often not properly implemented in LMICs [17]. There is a strong need to establish close coordination between enforcement and road agencies to improve safety data collection and its utilization in LMICs [3, 128, 142]. Such measures may help to envisage site-specific as well as area-wide road interventions in these countries [91]. Previous research pointed out that built-up areas are over-represented in interurban road crashes in LMICs [19]. Likely explanations could be the presence of vulnerable road users, excessive speed, and inadequate indications along these road sections [17, 19, 25]. In Cameroon, it was observed that 29% of injury crashes that occurred in built-up areas, involved one or more pedestrians, versus only 13% in rural crashes. Similarly, 18% of injury crashes that occurred in built-up areas involved one or more two-wheeled motor vehicles, versus only 4% in rural crashes. The attributable risk for built-area was high (16.1%), but the intervention consequences of such a result are not straightforward, as those areas cover a large part of the road section, 35% as estimated from control sites. These results suggest that area-wide traffic calming measures in the most populated and plain areas should, however, be considered, in particular to protect the most vulnerable road users. Crash sites were also associated with intersections, which can become a hub of commercial activity with a lot of pedestrian movements and traffic conflicts on intercity road sections in LMICs [17]. It should be noted that two-wheeled motor vehicle crashes were overrepresented at intersections (21% versus 9%), but this was not the case for pedestrians (19% versus 21%). These results indicate that service lanes and crossing facilities in built-up environment might therefore reduce the risk of crashes on this road [11, 17].

59

The hazard perception study showed that sites in built-up areas and intersections had higher hazard perception than those in rural areas. However, the hazard perception of high-risk sites within built-up areas was significantly less than sites not involved in crashes situated in such areas. Speed-reducing measures such as hazard signs, traffic lights, pavement markings, and lane-width modifications were extremely rare in built-up areas on both road sections [143]. This implies that improved hazard perception with speed-calming interventions can be useful in reducing crashes at known high-risk sites [11, 60, 143]. In both settings, most crashes occurred on road sections with flat and straight road profile. The case-control study indicated that a flat road profile was associated with injury crashes in Cameroon. Here again, no particular crash types were represented at sites with a flat road profile. Increased likelihood of crashes on flat and wide road indicated that these results could be explained by speed and hazardous overtaking [19, 39, 92]. Similarly, relative flat profile could explain a high crash rate on the Edéa-Douala road, but this hypothesis could not be tested in absence of population density and speed measures for this sub-section. The hazard perception study showed some interesting results regarding the above findings. High-risk sites with flat and straight road profile were not perceived as hazardous by drivers. Drivers choose significantly higher speeds for such road sections, compared to sites with slope and curved road profile [24]. Traffic speed monitoring thus in such context becomes the cornerstone for crash prevention [144]. Although NHMP implemented a strict speed enforcement measures on the Pakistani highway in the daylight settings, these measures were not consistently implemented in Cameroon at present. Nevertheless, available information points out the need to improve speed enforcement in such road situations [3]. In Cameroon and Pakistan, a significant proportion of crashes occurred as a result of loss of control by drivers and vehicles crashed off the road. Indeed, overall speed reduction is an important preventive intervention, but we have observed that in Cameroon the nearby obstacles (< 4 m) were associated with injury crash sites. These obstacles not only increase the crash risk by blocking the drivers’ view, but also increase the vulnerability of occupants when a crash occurs. Removal of these obstacles should be considered if feasible, otherwise measures such as crash barriers should be prioritized to reduce the risk of injuries [60]. This is all the more interesting as the proportion of such sites is relatively low, 4.2% as estimated from control sites. The hazardous overtaking appeared to be an important cause of severe crashes with non-separated traffic conditions. Almost half of the traffic in both countries is composed of trucks or other slow-moving vehicles [24]. Increased traffic volume could lead to this hazardous overtaking by smaller and faster vehicles, thus increasing traffic conflict and crash risk in such conditions. Therefore, it becomes imperative to separate the traffic by safety barriers in the high-risk areas where such crashes are observed in excess numbers [14]. These interventions have reduced the crash burden and were shown to be cost-effective in Sweden, but their feasibility in LMICs remains to be assessed [14]. Highways in LMICs undergo extensive wear and tear for several reasons. Firstly, overloading of vehicles is a frequent traffic violation, as observed in Pakistan [81]. Drivers after paying a traffic ticket continue to travel on the road network, thus damaging roads on its way. Secondly, delays in reconstruction and maintenance are frequent because LMICs often have financial difficulties [80]. Our results showed that road surface irregularities were associated with injury-crash sites. It is possible that these factors increased the likelihood of crashes due to loss of control, break failures, forced lane change, and abrupt breaking [92, 145]. Further

60

observational studies on how the degree of surface irregularity impacts vehicle speed and direction could be useful to assess crash risk with similar road conditions [11]. Mechanical problems, in particular tyre bursts, were identified as important causes of fatal crashes in both settings. Tyre bursts were previously identified as an important vehicle factor involved in crashes in LMICs [145]. In most LMICs, car owners prefer using used tyres due to financial constraints [50]. Previous research has clearly shown that these countries do not have an effective vehicle inspection system [19, 92]. Irregular road surface conditions on interurban roads may facilitate frequent tyre burst in LMICs [11]. The relative contribution of tyre problems in crashes indicated that realistic inspection system on interurban road sections in LMICs, keeping in mind that car owner might not be able to afford the corresponding maintenance costs [50]. Road surface improvement may help reduce crashes attributed to tyre problems in these countries. Traffic demands on the road sector and consistently augmenting in LMICs. Highway traffic is expected to triple in Pakistan over a period from 2005 to 2015 [81]. The reconstruction and upgrading of highways will certainly increase in these countries. The study on HWZ safety in Pakistan, for the first time showed a significantly high traffic crash and fatality burden due to such conditions. Guidelines for work zone management exist but so far no mechanism for HWZ design, performance, and enforcement evaluation has been defined or implemented in Pakistan [146]. Indeed, there is a need to improve institutional capacity, as well as inspection mechanisms, so that road agencies could be accountable for ensuring HWZ safety [81, 99, 146]. Further, it was observed that lengthy work zones lasted for over 10 months. Thus, efforts are required to reduce HWZ duration to impact crash risks [146]. Moreover, one of two HWZ crashes occurred between opposite-direction vehicles; the likely explanation was the high volume un-separated traffic conditions and hazardous overtaking. This points out the need to carefully plan and execute the safe flow of traffic during maintenance works [147]. Enforcing harsher penalties for overtaking, providing alternate lanes, and traffic separation may be some useful measures to decrease hazardous situations leading to HWZs crashes in Pakistan [146]. Our study showed that pedestrians were significantly involved in HWZ crashes. Sudden entry onto the highway was reported as the major cause of such crashes. Similarly, wet surfaces increased the risk of HWZ crashes. Such involvements, although less important, were found in HWZ crashes elsewhere [100]. Human judgement error is indeed one of the principal factor identified in HWZ crashes [100, 101]. These results indicated that provision of advance warning area, clear zones to enhance visibility, road markings, and hazard signage in the work area can be useful interventions to reduce HWZ crashes in Pakistan [94, 146]. Adverse weather conditions were involved in a significant proportion of interurban road crashes in Cameroon and Pakistan. We observed that there were no specific speed-restrictions imposed for such conditions on both road sections. Hazard perception for rainy conditions was lower for high-risk crash sites compared to low-risk sites. These results point out the need for measures that would help drivers better identify hazards in such conditions at high-risk sites. Specific measures, such as all-weather pavement markings, reduced speed limits, and surface improvements may help reduce such crashes in LMICs [99]. The hazard perception study showed that drivers were able to identify only half of high-risk sites. A reciprocal relationship of road and traffic hazard perception with reported speed was observed. The finding that hazard perception of high-risk sites was significantly different from that of low-risk sites for some of the site factors may have implications for road safety. Hazard perception of high-risk crash sites, for instance with hazard signs, was significantly

61

higher than that of low-risk sites. It is likely that drivers perceived a high-risk crash site more hazardous when road furniture and hazard signage was available on such site [11, 17]. Interurban road maintenance has not received adequate attention in LMICs as shown by few road signs observed in Cameroon and Pakistan. These results suggested that development and implementation of interventions improving hazard perception may be useful in reducing the crash risk on interurban road sections in LMICs [14]. Further, we observed that hazard perception was higher for Cameroonian sites as compared to Pakistani sites. This could be explained by traffic conditions, separated in Pakistan compared to non-separated in Cameroon. A higher hazard perception of sites with work zones in Pakistan, where traffic was not separated showed a similar trend for higher hazard perception [148, 149]. Further, mountainous terrain, unfamiliarity with the road section, and right-hand driving could augment hazard perception for Cameroonian site videos [150, 151]. These results showed that hazard perception measures developed elsewhere should be validated cross culturally and video methods can be useful to conduct such studies.

7. Limitations and perspectives In line with the increasing crash burden in LMIC, the United Nations has declared the next decade as the “Decade of Action for Road Safety 2011-2020” [3]. Such actions should be monitored with valid and reliable indicators. The police data that we used in Cameroon might have two limitations for estimating traffic fatality. Firstly, all police reports were not collected and secondly, reported injured patients were followed at most for eight days. This leads to underestimation of our rates, which could not be corrected in the analyses [3]. Although we were able to record all police reported injuries on the Pakistani road section, we observed that police recording of traffic fatalities and injuries was very low when compared to ambulance or ED data [64]. The record linkage helped to estimate overall injury rates but failed to identify high-risk crash sites or other information that could be useful for implementing preventive interventions [152]. Furthermore, information on safety equipment use by vehicle occupants and motorcycle rider was almost never reported in police data. All this showed that there is a need to improve the existing interurban road injury surveillance system so that we could measure the impact of safety interventions in the future. The police should be provided with adequate resources and training to collect information as is the case in HICs [55]. Preliminary results from this thesis were communicated to the NHMP and a complementary study is currently undergoing to evaluate the impact of strict seat-belt law implementation at Karachi-Hala road section. We hope to work with them in the future on my return to Pakistan, so that we can plan other studies to better understand which factors impede safety equipment use and are responsible for the high fatality among passenger occupants. Associations of situational factors with injury crash sites did not imply a causal relationship. As road design and furniture improvements are planned on Yaoundé-Douala road section, we would like to reassess the road burden on the same road section in coming years. Few studies have reported the effectiveness of road interventions in LMICs [90], and such data collection in Cameroon and possibly on other interurban roads in Pakistan will help measure the impact of interventions, particularly with respect to specific situational factors. In the studies described here, most injury patients came to the hospital on their own, on both road sections. In Pakistan, slightly higher proportion of the patients was transported by the ambulance service. In connection with the field work in Pakistan, a study was conducted to assess the quality of pre-hospital care available to RTI patients on interurban roads in the province of Sindh. Trauma care quality has rarely been assessed in LMICs [2, 3]. Our evaluation of trauma care in a Moroccan region identified several opportunities for improvement in the current healthcare system [153]. Results from such a study can be useful

62

to understand the deficiencies in the trauma care system as well as propose recommendations to the stake holders of road safety. Finally, results from the hazard perception study indicated a significant interaction between driver- and site-related factors at high-risk sites. Up to now, methods to prioritize high-risk site improvement concentrated on crash data, whereas driver perception of site was never studied in this regard. These results encourage pilot testing of projects for integrating these methods in prioritizing crash sites for improvement. Further, because of limited resources at the time of study initiation, we were not able to construct a portable simulator to measure the driver reactions such as video speed selection or deviation of steering wheel upon perceiving a hazard while video projection. Such a method could further validate our hypothesis and help in understanding more the above interactions. I would like to continue working on this topic and consider conducting such program in a LMIC to help in development and implementation of cost-effective preventive measures in such settings.

8. Conclusion The studies conducted here clearly demonstrate the need to reduce the crash burden on interurban road sections in LMICs. Several lessons can be learnt from these studies: Firstly, a reliable and accurate injury surveillance system is required to assess the highway injury burden in LMICs. Police reporting has been efficiently used in HICs to assess road burden and there is no reason that this cannot be achieved in LMICs, with improved resources and mechanism. Further, efforts are required to improve vulnerable road users reporting in crashes. Exposition measures should be used to compute comparable and sensitive traffic safety indicators in these countries. Secondly, there is a need to improve coordination between road and police authorities to properly use the information on crash site location and situational factors for prevention purposes by the stake holders. Our results indicate that traffic calming and speed enforcement interventions in built-up areas and on flat sections of the road should be prioritized. Moreover, road surface maintenance and the removal of nearby roadside obstacles are likely to prevent many serious crashes. Thirdly, HWZ activity is expected to increase in Pakistan and other LMICs. The results suggest that a monitoring system is needed to examine the HWZ safety interventions by agencies involved in maintenance works. Moreover, efforts are required to reduce the duration of HWZs. These results orient toward prevention interventions such as harsher punishment for traffic violations such as overtaking, traffic separation, advanced warning area, hazard signage, and safe passages for pedestrians at HWZs in Pakistan. Feasibility and effectiveness of their implementation, however remains to be evaluated. Fourthly, the difference in hazard perception of high-risk sites as compared to low-risk sites with same site factors suggested the need to make roads “self explaining” in LMICs. Hazard perception interventions are mostly of low cost and easy to implement compared to design changes. These results point out that implementing such interventions may reduce traffic speed at high-risk sites [154]. Moreover, the study methods can be useful for the road authorities to assess the impact of these interventions on drivers’ hazard perception before their implementation [155]. Lastly, the focus of this thesis was to document contribution of situational factors in interurban road crashes. This should not hinder implementing interventions to counter other interurban road risk factors such as mechanical failure, tyre problems, speeding, DWI, and low seatbelt and helmet use.

63

References [1] Mohan D. Road traffic injuries--a neglected pandemic. Bull World Health Organ. 2003;81(9):684-5. [2] World Health Organization. Global status report on road safety. Geneva: World Health Organization; 2009. [3] Peden M, Scurfiled R, Sleet D, Mohan D, Hyder A, Jarawan E. World report on road traffic injury prevention. Geneva: World Health Organization; 2004. [4] Jacobs G, Aeron-Thomas A, Astrop A. Estimating global road fatalities. Crowthorne (UK): Transport Research Laboratory; 2000. Report No.: TRL Report No. 445. [5] Kopits E, Cropper M. Traffic fatalities and economic growth. Accid Anal Prev. 2005 Jan;37(1):169-78. [6] International Road Federation. World Road Statistics 2008 (Data 2001 to 2006). Geneva: International Road Federation; 2008. [7] Wootton J, Jacobs GD. Safe roads: A dream or a reality. Crowthorne (UK): Transport Research Laboratory; 1996. Report No.: PA 3184/96. [8] Khayesi M, Peden M. Road safety in Africa. BMJ. 2005 Oct 1;331(7519):710-1. [9] Observatoire National Interministériel de Sécurité Routière. La sécurité routière en France: Bilan de l'année 2004. Paris: La documentation Française; 2005. [10] Nishtar S, Mohamud KB, Razzak J, Ghaffar A, Ahmed A, Khan SA, et al. Injury prevention and control: National Action Plan for NCD Prevention, Control and Health Promotion in Pakistan. J Pak Med Assoc. 2004 Dec;54(12 Suppl 3):S57-68. [11] Ross A, Baguely C, Hills B, McDonald M, Silcock D. Towards safer roads in developing countries—a guide for planners and engineers. Crowthorne (UK): Transport Research Laboratory; 1991. [12] Thomson H, Jepson R, Hurley F, Douglas M. Assessing the unintended health impacts of road transport policies and interventions: translating research evidence for use in policy and practice. BMC Public Health. 2008;8:339. [13] Forjuoh SN, Li G. A review of successful transport and home injury interventions to guide developing countries. Soc Sci Med. 1996 Dec;43(11):1551-60. [14] Elvik R, Kolbenstvedt M, Elvebakk B, Hervik A, Braein L. Costs and benefits to Sweden of Swedish road safety research. Accid Anal Prev. 2009 May;41(3):387-92. [15] Borse NN, Hyder AA. Call for more research on injury from the developing world: results of a bibliometric analysis. Indian J Med Res. 2009 Mar;129(3):321-6. [16] Jamison DT, Mosley WH, Measham AB, Bobadilla JL, eds. Disease Control Priorities in Developing Countries. Washington, DC: Oxford University Press; 1993. [17] Mohan D. Road safety in less-motorized environments: future concerns. Int J Epidemiol. 2002 Jun;31(3):527-32. [18] Abbas KA. Traffic safety assessment and development of predictive models for accidents on rural roads in Egypt. Accid Anal Prev. 2004 Mar;36(2):149-63. [19] Afukaar FK. Speed control in developing countries: issues, challenges and opportunities in reducing road traffic injuries. Inj Control Saf Promot. 2003 Mar-Jun;10(1-2):77-81. [20] Akgungor AP, Yildiz O. Sensitivity analysis of an accident prediction model by the fractional factorial method. Accid Anal Prev. 2007 Jan;39(1):63-8. [21] Almeida LV, Pignatti MG, Espinosa MM. [Main factors associated with motor vehicle accidents on Federal Highway 163, Mato Grosso State, Brazil, 2004]. Cad Saude Publica. 2009 Feb;25(2):303-12. [22] Bagaria V, Bagaria S. A geographic information system to study trauma epidemiology in India. J Trauma Manag Outcomes. 2007;1(1):3. [23] Bishai D, Asiimwe B, Abbas S, Hyder AA, Bazeyo W. Cost-effectiveness of traffic enforcement: case study from Uganda. Inj Prev. 2008 Aug;14(4):223-7.

64

[24] Damsere-Derry J, Afukaar FK, Donkor P, Mock C. Assessment of vehicle speeds on different categories of roadways in Ghana. Int J Inj Contr Saf Promot. 2008 Jun;15(2):83-91. [25] Derry JD, Afukaar FK, Donkor P, Mock C. Study of vehicle speeds on a major highway in Ghana: implication for monitoring and control. Traffic Inj Prev. 2007 Jun;8(2):142-6. [26] Duric P, Miladinov-Mikov M. [Epidemiological characteristics of road traffic injuries in AP Vojvodina]. Med Pregl. 2009 Jan-Feb;62(1-2):17-22. [27] Ehikhamenor EE, Obianwuho HO. Digital screening for Blood Alcohol Concentration (BAC) in a Southern Nigeria City. Traffic Inj Prev. 2006 Mar;7(1):70-5. [28] Erdogan S, Yilmaz I, Baybura T, Gullu M. Geographical information systems aided traffic accident analysis system case study: city of Afyonkarahisar. Accid Anal Prev. 2008 Jan;40(1):174-81. [29] Guanche Garcell H, Suarez Enriquez T, Gutierrez Garcia F, Martinez Quesada C, Mendoza Perez R. [Drunk driving in professional drivers in the Via Blanca highway in Cuba]. Gac Sanit. 2006 Sep-Oct;20(5):407-9. [30] Hijar M, Carrillo C, Flores M, Anaya R, Lopez V. Risk factors in highway traffic accidents: a case control study. Accid Anal Prev. 2000 Sep;32(5):703-9. [31] Hijar M, Flores M, Lopez MV, Rosovsky H. Alcohol intake and severity of injuries on highways in Mexico: a comparative analysis. Addiction. 1998 Oct;93(10):1543-51. [32] Hijar-Medina MC, Carrillo-Ordaz CE, Flores-Aldana ME, Anaya R, Lopez-Lopez MV. [Risk factors for injuries caused by traffic accidents and the impact of an intervention on the road]. Rev Saude Publica. 1999 Oct;33(5):505-12. [33] Hijar-Medina MC, Flores-Aldana ME, Lopez-Lopez MV. [Safety belt use and severity of injuries in traffic accidents]. Salud Publica Mex. 1996 Mar-Apr;38(2):118-27. [34] Liu GF, Han S, Liang DH, Wang FZ, Shi XZ, Yu J, et al. Driver sleepiness and risk of car crashes in Shenyang, a Chinese northeastern city: population-based case-control study. Biomed Environ Sci. 2003 Sep;16(3):219-26. [35] Loo BP. Validating crash locations for quantitative spatial analysis: a GIS-based approach. Accid Anal Prev. 2006 Sep;38(5):879-86. [36] Majdzadeh R, Khalagi K, Naraghi K, Motevalian A, Eshraghian MR. Determinants of traffic injuries in drivers and motorcyclists involved in an accident. Accid Anal Prev. 2008 Jan;40(1):17-23. [37] Mutto M, Kobusingye OC, Lett RR. The effect of an overpass on pedestrian injuries on a major highway in Kampala - Uganda. Afr Health Sci. 2002 Dec;2(3):89-93. [38] Nascimento EC, Nascimento E, Silva Jde P. [Alcohol and amphetamines use among long-distance truck drivers.]. Rev Saude Publica. 2007 Apr;41(2):290-3. [39] Odero W, Khayesi M, Heda PM. Road traffic injuries in Kenya: magnitude, causes and status of intervention. Inj Control Saf Promot. 2003 Mar-Jun;10(1-2):53-61. [40] Poli de Figueiredo LF, Rasslan S, Bruscagin V, Cruz R, Rocha e Silva M. Increases in fines and driver licence withdrawal have effectively reduced immediate deaths from trauma on Brazilian roads: first-year report on the new traffic code. Injury. 2001 Mar;32(2):91-4. [41] Rey de Castro J, Gallo J, Loureiro H. [Tiredness and sleepiness in bus drivers and road accidents in Peru: a quantitative study]. Rev Panam Salud Publica. 2004 Jul;16(1):11-8. [42] Routley V, Ozanne-Smith J, Li D, Yu M, Wang J, Zhang J, et al. China belting up or down? Seat belt wearing trends in Nanjing and Zhoushan. Accid Anal Prev. 2008 Nov;40(6):1850-8. [43] Saidi HS, Kahoro P. Experience with road traffic accident victims at The Nairobi Hospital. East Afr Med J. 2001 Aug;78(8):441-4. [44] Souza JC, Paiva T, Reimao R. Sleep habits, sleepiness and accidents among truck drivers. Arq Neuropsiquiatr. 2005 Dec;63(4):925-30. [45] van Schoor O, van Niekerk JL, Grobbelaar B. Mechanical failures as a contributing cause to motor vehicle accidents South Africa. Accid Anal Prev. 2001 Nov;33(6):713-21.

65

[46] Viegas CA, de Oliveira HW. Prevalence of risk factors for obstructive sleep apnea syndrome in interstate bus drivers. J Bras Pneumol. 2006 Mar-Apr;32(2):144-9. [47] Yildirim RC. Caffeine consumption in drivers of heavy vehicles in Turkey. Public Health. 2003 Sep;117(5):329-32. [48] Zhou JH, Zhao XC, Jiang ZQ, Zhu SY, Zhou JC. Investigation on road traffic safety of the mountain areas in southwest China based on Wulong County, Chongqing Municipality. Traffic Inj Prev. 2005 Jun;6(2):193-6. [49] Sabey BE, Taylor H. The Known Risks We Run: The Highway. In: Schwing SC, Alber WA, eds. Societal Risk Assessment: How safe is safe enough. New York: Plenum Press; 1980. [50] Lagarde E. Road traffic injury is an escalating burden in Africa and deserves proportionate research efforts. PLoS Med. 2007 Jun;4(6):e170. [51] Mock C, Cherian MN. The global burden of musculoskeletal injuries: challenges and solutions. Clin Orthop Relat Res. 2008 Oct;466(10):2306-16. [52] Gururaj G. Road traffic deaths, injuries and disabilities in India: current scenario. Natl Med J India. 2008 Jan-Feb;21(1):14-20. [53] Mathers C, Boerma T, Ma Fat D. The global burden of disease: 2004 update. Geneva: World Health Organization; 2008. [54] Nantulya VM, Reich MR. Equity dimensions of road traffic injuries in low- and middle-income countries. Inj Control Saf Promot. 2003 Mar-Jun;10(1-2):13-20. [55] Evans L. Traffic Safety. Bloomfield Hills, MI: Science Serving Society; 2004. [56] Runyan CW. Introduction: back to the future--revisiting Haddon's conceptualization of injury epidemiology and prevention. Epidemiol Rev. 2003;25:60-4. [57] Runyan CW. Using the Haddon matrix: introducing the third dimension. Inj Prev. 1998 Dec;4(4):302-7. [58] NSW Roads and Traffic Authority. Why speeding? Regret is such a short distance. Sydney: NSW Roads and Traffic Authority; 1996. [59] Subramanian R. Geospatial Analysis of Rural Motor Vehicle Traffic Fatalities. Washington, DC: National Highway Traffic Safety Administration; 2009. [60] Oxley J, Corben B, Koppel S, Fildes B, Jacques N, Symmons M, et al. Cost-effective infrastructure measures on rural roads. Clayton, Victoria: Monash University Accident Research Centre; 2004. [61] Petridou E, Moustaki M. Human factors in the causation of road traffic crashes. Eur J Epidemiol. 2000;16(9):819-26. [62] Crane EO, Augustine A, Tait GR. The M77 Highway: saving lives and money. Injury. 2008 Sep;39(9):1071-4. [63] National Transport Research Centre. Traffic Counter measures in Pakistan. Islamabad: National Transport Research Centre, Ministry of Communication; 1985. Report No.: NTRC publication 85. [64] Razzak JA, Luby SP. Estimating deaths and injuries due to road traffic accidents in Karachi, Pakistan, through the capture-recapture method. Int J Epidemiol. 1998 Oct;27(5):866-70. [65] Zavareh DK, Mohammadi R, Laflamme L, Naghavi M, Zarei A, Haglund BJ. Estimating road traffic mortality more accurately: use of the capture-recapture method in the West Azarbaijan province of Iran. Int J Inj Contr Saf Promot. 2008 Mar;15(1):9-17. [66] Andrade SM, Prado de Mello-Jorge MH. [Traffic accidents in a city in Southern Brazil: an evaluation of coverage and quality of data]. Cad Saude Publica. 2001 Nov-Dec;17(6):1449-56. [67] Perel P, Ker K, Ivers R, Blackhall K. Road safety in low- and middle-income countries: a neglected research area. Inj Prev. 2007 Aug;13(4):227. [68] Perel P, McGuire M, Eapen K, Ferraro A. Research on preventing road traffic injuries in developing countries is needed. BMJ. 2004 Apr 10;328(7444):895.

66

[69] Central Intelligence Agency. The World Factbook. Langley, VA: Directorate of Intelligence; 2009. [70] Regional Department Centre (ORCE). Country Strategy Paper 2010-2014, Cameroon. Tunis-Belvedère: African Development Fund, African Development Bank; 2009. [71] Ze Minkande J, Simo Moyo J, Afane Ela A, Penlap Temdie E, Nnomoko Bilounga E, Beyiha G, et al. Bilan des trois premières années d'activités du SAMU de Yaoundé (Cameroun). Med Trop (Mars). 2009;69(6):577-80. [72] Jane and Justice. Les accidents de la circulation au Cameroun de 1982 à 2001. Yaoundé: Jane and Justice; 2003. [73] Ministère des Travaux Publiques. Comptages routiers 2005 : Synthèse des campagnes de mai et novembre. Yaoundé: Direction des investissements et de l’entretien routier, Cellule de la programmation, Ministère des Travaux Publiques; 2006. [74] Epi Info [Computer program]. Version 3.5.1. Atlanta, GA: Centers for Disease Control and Prevention; 2008. [75] SAS [Computer program]. Version 9.1.3. Cary, NC: SAS Institute Inc.; 2004. [76] National Transport Research Centre. Manual of road safety improvement by the use of low cost engineering countermeasures. Islamabad: National Transport Research Centre, National Highway Authority, and Finnroad OY; 1999. [77] Ghaffar A, Hyder AA, Masud TI. The burden of road traffic injuries in developing countries: the 1st national injury survey of Pakistan. Public Health. 2004 Apr;118(3):211-7. [78] Fatmi Z, Hadden WC, Razzak JA, Qureshi HI, Hyder AA, Pappas G. Incidence, patterns and severity of reported unintentional injuries in Pakistan for persons five years and older: results of the National Health Survey of Pakistan 1990-94. BMC Public Health. 2007;7:152. [79] Raja IA, Vohra AH, Ahmed M. Neurotrauma in Pakistan. World J Surg. 2001 Sep;25(9):1230-7. [80] Ahmed A. Road Safety in Pakistan. Islamabad: National Road Safety Secretariat; 2007. [81] Japan International Cooperation Agency, National Transport Research Centre, Government of Pakistan. Pakistan Transport Plan Study in the Islamic Republic Of Pakistan. Islamabad: tripartite collaboration of the Japan International Cooperation Agency, National Transport Research Centre, and Ministry of Communications, Government of Pakistan; 2007. [82] Shah SG, Khoumbati K, Soomro B. The pattern of deaths in road traffic crashes in Sindh, Pakistan. Int J Inj Contr Saf Promot. 2007 Dec;14(4):231-9. [83] National Highway Authority. Traffic survey. Islamabad: National Highway Authority; 2008. [84] Government of Pakistan. Pakistan Economic Survey 2008-09. Islamabad: Economic Advisor's Wing, Finance Division, Government of Pakistan; 2009. [85] Khoso A. Analysis of National Highways & Motorway Police injury surveillance system with respect to WHO injury surveillance guidelines. Stockholm: Karolinska Institutet; 2007. [86] Razzak JA, Luby SP, Laflamme L, Chotani H. Injuries among children in Karachi, Pakistan--what, where and how. Public Health. 2004 Mar;118(2):114-20. [87] Road Injury Research and Prevention Centre. Report of road injury surveillance project of the road injury research and prevention centre of JPMC, Karachi. Karachi: Jinnah Post Graduate Medical Centre (JPMC); 2008. [88] Stevenson M, Segui-Gomez M, Lescohier I, Di Scala C, McDonald-Smith G. An overview of the injury severity score and the new injury severity score. Inj Prev. 2001 Mar;7(1):10-3. [89] National Highway and Motorway Police. Accident Analysis Report N5 (Year 2006). Islamabad: National Highway and Motorway Police; 2007.

67

[90] Odero W, Garner P, Zwi A. Road traffic injuries in developing countries: a comprehensive review of epidemiological studies. Trop Med Int Health. 1997 May;2(5):445-60. [91] Morency P, Cloutier MS. From targeted "black spots" to area-wide pedestrian safety. Inj Prev. 2006 Dec;12(6):360-4. [92] Nordberg E. Injuries as a public health problem in sub-Saharan Africa: epidemiology and prospects for control. East Afr Med J. 2000 Dec;77(12 Suppl):S1-43. [93] Rockhill B, Newman B, Weinberg C. Use and misuse of population attributable fractions. Am J Public Health. 1998 Jan;88(1):15-9. [94] Li Y, Bai Y. Development of crash-severity-index models for the measurement of work zone risk levels. Accid Anal Prev. 2008 Sep;40(5):1724-31. [95] Department of Transportation. Highway Statistics 2007: Funding For Highways and Disposition of Highway-User Revenues. Washington, DC: Department of Transportation; 2007. [96] Department of Transportation. FY 2008 Budget in brief. Washington, DC: Department of Transportation; 2008. [97] Republic of Ghana, European Community. Country Strategy Paper and Indicative Programme for the period 2002–2007 Accra: European Union delegation in Ghana; 2002. Report No.: DEV/054/2002-EN. [98] Masood H, Seetharam KC, Utami D, Nakazaki T, Ahmad K, Beg B. Report and recommendation of the President to the board of directors on proposed loans to the Islamic Republic of Pakistan for the North-West Frontier Province road development sector and subregional connectivity project. Manila: Asian Development Bank; 2004. Report No.: RRP: PAK 36052. [99] American Association of State Highway and Transportation Officials. A policy on geometric design of highways and streets. Washington, DC: American Association of State Highway and Transportation Officials; 2004. [100] Li Y, Bai Y. Comparison of characteristics between fatal and injury accidents in highway work zones. Safety Sci. 2008;46(4):646-60. [101] Li Y, Bai Y. Highway work zone factors and their impact on crash severity. J Transp Eng. 2009;135(10):694-701. [102] Mohan SB, Gautam P. Cost of highway work zone injuries. Practical Periodical on Structural Design and Construction. 2002;7(ASCE (2)):68-73. [103] Department of Transportation. Manual of Uniform Traffic Control Devices for Streets and Highways Washington, DC: United States Department of Transportation, Federal Highway Administration; 1988. [104] Salmi LR, Battista RN. Epidemiologic assessment of hazardous roadway locations. Epidemiology. 1990 Jul;1(4):311-4. [105] Lassarre S. Some Statistical Models for Road Risk Analysis. In: Tiwari G, Mohan D, Muhlrad N, eds. The Way Forward: Transportation Planning and Road Safety. New Delhi: MacMillan India Ltd.; 2005. [106] Sobngwi-Tambekou J, Bhatti J, Kounga G, Salmi L-R, Lagarde E. Road traffic crashes on the Yaoundé–Douala road section, Cameroon. Accid Anal Prev. 2010;42(2):422-6. [107] Rothman KJ, Greenland S, eds. Modern Epidemiology. Second ed. Philadelphia, PA: Lippincott Williams & Wilkins; 1998. [108] Hosmer DW, Lemeshow S. Applied Logistic Regression. Second ed. Danvers, MA: John Wiley & Sons, Inc.; 2000. [109] Geurts K, Wets G. Black Spot Analysis Methods: Literature Review. Diepenbeek: Steunpunt Verkeersveiligheid Bij Stijgende Mobiliteit.; 2003. Report No.: RA-2003-07. [110] Elvik R. A survey of operational definitions of hazardous road locations in some European countries. Accid Anal Prev. 2008 Nov;40(6):1830-5.

68

[111] Geurts K, Wets G, Brijs T, Vanhoof K, Karlis D. Ranking and selecting dangerous crash locations: correcting for the number of passengers and Bayesian ranking plots. J Safety Res. 2006;37(1):83-91. [112] Benda HV, Hoyos CG. Estimating hazards in traffic situations. Accid Anal Prev. 1983;15(1):1-9. [113] SWOV. SWOV fact sheet: hazard perception. Leidschendam: SWOV Institute for Road Safety Research; 2008. [114] DeJoy DM. The optimism bias and traffic accident risk perception. Accid Anal Prev. 1989 Aug;21(4):333-40. [115] Finn P, Bragg BW. Perception of the risk of an accident by young and older drivers. Accid Anal Prev. 1986 Aug;18(4):289-98. [116] Harre N. Risk evaluation, driving, and adolescents: a typology. Dev Rev. 2000;20:206-26. [117] Mayhew DR, Simpson HM, Pak A. Changes in collision rates among novice drivers during the first months of driving. Accid Anal Prev. 2003;35(683-91). [118] McKenna FP, Stanier RA, Lewis C. Factors underlying illusory self-assessment of driving skill in males and females. Accid Anal Prev. 1991;23:45-52. [119] Trankle U, Gelau C, Metker T. Risk perception and age-specific accidents of young drivers. Accid Anal Prev. 1990 Apr;22(2):119-25. [120] Goldenbeld C, van Schagen I. The credibility of speed limits on 80 km/h rural roads: The effects of road and person(ality) characteristics. Accid Anal Prev. 2007 Nov;39(6):1121-30. [121] Kanellaidis G, Zervas A, Karagioules V. Drivers' risk perception of road design elements. Transp Hum Factors. 2000;2(1):39-48. [122] Deery HA. Hazard and risk perception among young novice drivers. J Safety Res. 1999;30(4):225-36. [123] Rundmo T, Iversen H. Risk perception and driving behaviour among adolescents in two Norwegian counties before and after a traffic safety campaign. Safety Sci. 2004;42(1):1-21. [124] Montella A. A comparative analysis of hotspot identification methods. Accid Anal Prev. 2010;42(2):571-81. [125] Flahault A, Cadilhac M, Thomas G. Sample size calculation should be performed for design accuracy in diagnostic test studies. J Clin Epidemiol. 2005 Aug;58(8):859-62. [126] Bhatti JA, Sobngwi-Tambekou J, Lagarde E, Salmi LR. Situational factors associated with road traffic crashes: A case-control study on the Yaoundé-Douala road section, Cameroon. Int J Inj Contr Saf Promot. 2010 Mar 30:1-8. [127] Sümer N, Ünal AB, Birdal A, Çinar P, Çevikoglu S. Comparison of self-reported and computer-based hazard perception skills among novice and experienced drivers. In: Fourth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, 2007, 9-12 July, Stevenson, WA. Washington, DC: Transport Research Board; 2007. [128] Tiwari G, Mohan D, Muhlrad N, eds. The Way Forward: Transportation Planning and Road Safety. Delhi: McMillan India Ltd; 2005. [129] Bhatti JA, Salmi LR, Lagarde E, Razzak JA. Concordance between road mortality indicators in high-income and low- and middle-income countries. Traffic Inj Prev. Apr;11(2):173-7. [130] Sauerzapf V, Jones AP, Haynes R. The problems in determining international road mortality. Accid Anal Prev. 2010;42(2):492-9. [131] Iribhogbe PE, Osime CO. Compliance with seat belt use in Benin City, Nigeria. Prehosp Disaster Med. 2008 Jan-Feb;23(1):16-9. [132] Sangowawa AO, Alagh BT, Ekanem SE, Ebong IP, Faseru B, Adekunle BJ, et al. An observational study of seatbelt use among vehicle occupants in Nigeria. Inj Prev. 2010 Apr;16(2):85-9.

69

[133] Perneger T, Smith GS. The driver's role in fatal two-car crashes: a paired "case-control" study. Am J Epidemiol. 1991 Nov 15;134(10):1138-45. [134] Bener A, Ghaffar A, Azab A, Sankaran-Kutty M, Toth F, Lovasz G. The impact of four-wheel drives on traffic disability and deaths compared to passenger cars. J Coll Physicians Surg Pak. 2006 Apr;16(4):257-60. [135] Jooma R, Khan SJ, Razzak JA. The Trauma Centre: what it should mean. J Pak Med Assoc. 2008 Oct;58(10):530-1. [136] Quansah R. Availability of emergency medical services along major highways. Ghana Med J. 2001;35:8-10. [137] Razzak JA, Cone DC, Rehmani R. Emergency medical services and cultural determinants of an emergency in Karachi, Pakistan. Prehosp Emerg Care. 2001 Jul-Sep;5(3):312-6. [138] Razzak JA, Kellermann AL. Emergency medical care in developing countries: is it worthwhile? Bull World Health Organ. 2002;80(11):900-5. [139] Bhalla K, Ezzati M, Mahal A, Salomon J, Reich M. A risk-based method for modeling traffic fatalities. Risk Anal. 2007 Feb;27(1):125-36. [140] Jarvis SN, Lowe PJ, Avery A, Levene S, Cormack RM. Children are not goldfish--mark/recapture techniques and their application to injury data. Inj Prev. 2000 Mar;6(1):46-50. [141] Dhillon PK, Lightstone AS, Peek-Asa C, Kraus JF. Assessment of hospital and police ascertainment of automobile versus childhood pedestrian and bicyclist collisions. Accid Anal Prev. 2001 Jul;33(4):529-37. [142] Khorasani-Zavareh D, Khankeh HR, Mohammadi R, Laflamme L, Bikmoradi A, Haglund BJ. Post-crash management of road traffic injury victims in Iran. Stakeholders' views on current barriers and potential facilitators. BMC Emerg Med. 2009;9:8. [143] Retting RA, Ferguson SA, McCartt AT. A review of evidence-based traffic engineering measures designed to reduce pedestrian-motor vehicle crashes. Am J Public Health. 2003 Sep;93(9):1456-63. [144] Retting RA, Kyrychenko SY, McCartt AT. Evaluation of automated speed enforcement on Loop 101 freeway in Scottsdale, Arizona. Accid Anal Prev. 2008 Jul;40(4):1506-12. [145] Nzegwu MA, Nzegwu CO. Review of causes of road traffic accidents in Benin City, Nigeria: a 1-year study, August 2003-July 2004. Emerg Med Australas. 2007 Feb;19(1):77-8. [146] Ahmed A. National Road Safety Plan 2007-2012. Islamabad: National Road Safety Secretariat, Ministry of Communications; 2007. [147] American Association of State Highway and Transportation Officials. Summary report on work zone crashes. Standing committee on highway traffic safety. Washington, DC: American Association of State Highway and Transportation Officials; 1987. [148] Lewis-Evans B, Charlton SG. Explicit and implicit processes in behavioural adaptation to road width. Accid Anal Prev. 2006 May;38(3):610-7. [149] Morgan JF, Duley AR, Hancock PA. Driver responses to differing urban work zone configurations. Accid Anal Prev. 2010 May;42(3):978-85. [150] McKenna FP, Horswill MS, Alexander JL. Does anticipation training affect drivers' risk taking? J Exp Psychol Appl. 2006 Mar;12(1):1-10. [151] Sagberg F, Bjornskau T. Hazard perception and driving experience among novice drivers. Accid Anal Prev. 2006 Mar;38(2):407-14. [152] Razzak JA, Laflamme L. Limitations of secondary data sets for road traffic injury epidemiology: a study from Karachi, Pakistan. Prehosp Emerg Care. 2005 Jul-Sep;9(3):355-60. [153] Tachfouti N, Bhatti JA, Nejjari C, Kanjaa N, Salmi LR. Emergency trauma care for severe injuries in a moroccan region: conformance to French and World Health Organization standards. J Healthc Qual. 2010 Jun 30:1-9.

70

[154] Rosenbloom T, Shahar A, Elharar A, Danino O. Risk perception of driving as a function of advanced training aimed at recognizing and handling risks in demanding driving situations. Accid Anal Prev. 2008 Mar;40(2):697-703. [155] Milleville-Pennel I, Jean-Michel H, Elise J [Sic: inversion nom/prénoms par PubMed]. The use of hazard road signs to improve the perception of severe bends. Accid Anal Prev. 2007 Jul;39(4):721-30. [156] Mohan D, Tiwari G, Meleckidzedeck K, Fredrick MN. Road traffic injury prevention manual. Geneva: World Health Organization and Indian Institute of Technology; 2006. [157] Welander G, Svanstrom L, Eckman R. Safety Promotion: An Introduction. Stockholm: Karolinska Institutet, Department of Public Health Sciences, Division of Social Medicine; 2000. [158] Widome MD. Remembering as we look ahead: the three E's and firearm injuries. Pediatrics. 1991 Aug;88(2):379-83. [159] Salmi LR. Lecture critique et communication médicale scientifique : Comment lire, présenter, rédiger et publier une étude clinique ou épidémiologique. 2ème édition. Paris: Elsevier; 2002.

71

Publications (peer-reviewed) Related to thesis

Articles published 1. Sobngwi-Tambekou J, Bhatti J, Kounga G, Salmi L.-R, Lagarde E. Road traffic

crashes on the Yaoundé-Douala road section, Cameroon. Accid Anal Prev. 2010 Mar;42(2):422-6.

2. Bhatti JA, Sobngwi-Tambekou J, Lagarde E, Salmi L.-R. Situational factors associated with road traffic crashes: a case-control study on the Yaoundé-Douala road section, Cameroon. Int J Inj Contr Saf Promot. 2010. [Epub Mar 30:1-8]

Manuscript under review 3. Bhatti JA, Razzak JA, Lagarde E, Salmi L.-R. Burden and factors associated with

highway work-zone crashes, Karachi-Hala road section, Pakistan. Inj Prev 2010.

Manuscripts in preparation 4. Bhatti JA. Razzak JA, Lagarde E, Salmi L.-R. Difference in police, hospital, and

prehospital reporting of road traffic injuries to on an interurban road, Pakistan. 5. Bhatti JA, Razzak JA, Lagarde E, Sobngwi-Tambekou J, Alioum A, Salmi L.-R.

Hazard perception at high- and low-risk crash sites.

Abstracts published 6. Bhatti JA, Sobngwi-Tambekou J, Salmi L.-R, Lagarde, E. Road traffic injuries and

associated road-factors on Yaoundé-Douala road-section, Cameroon [Abstract]. Int J Emerg Med 2008;1:237. [Oral presentation in 12th National Health Sciences Research Symposium, Karachi, 26-28 August, 2008].

7. Bhatti JA, Salmi L.-R. Perception de la dangerosité des tronçons accidentogènes chez des conducteurs volontaires [Résumé]. Rev Epidemiol Sante Publique 2010 [sous presse]. [Présentation orale acceptée au IVème congrès d’épidémiologie ADELF-EPITER, Marseille, 15-17 septembre 2010].

8. Bhatti JA, Razzak JA, Lagarde E, Salmi L.-R. Burden and factors associated with work-zone crashes on an interurban highway in Pakistan [Abstract]. Inj Prev 2010. [In press]. [Poster presentation accepted in Safety 2010 World Conference, London, 22-24 September, 2010]

9. Bhatti JA, Razzak JA, Lagarde E, Salmi L.-R. Road hazard perception of high risk sites in voluntary Pakistani drivers [Abstract]. Inj Prev 2010 [In press]. [Poster presentation accepted in Safety 2010 World Conference, London, 22-24 September, 2010].

Other articles related to traffic injuries

Articles published 1. Bhatti JA, Salmi L.-R, Lagarde E, Razzak JA. Concordance between road-mortality

indicators in high-income and low- and middle-income countries. Traffic Inj Prev 2010;11(2): 173-7.

2. Bhatti JA, Constant A, Salmi L.-R, Chiron M, Lafont S, Zins M, Lagarde E. Impact of retirement on risky driving behavior and attitudes towards road safety among a large cohort of French drivers (the GAZEL cohort). Scand J Work Environ Health. 2008 Aug;34(4):307-15.

3. Bhatti JA. Prioritization of road traffic injury prevention fund spending in developing countries [Letter]. Inj Prev. 2010;16(3): 214-5.

72

4. Farooq U, Bhatti JA, Siddiq M, Majeed M, Malik N, Razzak JA, Khan MM. Road traffic injuries in Rawalpindi city, Pakistan. East Mediterr Health J 2010. [In press]

Abstracts published 5. Bhatti JA, Salmi L.-R. Faiblesse des capacités de mesurer l'impact d’une politique de

sécurité routière dans les pays en développement (PED) : Le cas de la Région de l’Est Méditerranéen (EMRO) [Résumé]. Rev Epidemiol Sante Publique 2010 [sous presse]. [Présentation orale acceptée au IVème congrès d’épidémiologie ADELF-EPITER, Marseille, 15-17 septembre 2010].

6. Bhatti JA, Salmi L.-R. Retirement transition: modification of driving behaviours and attitudes toward road safety [Abstract]. Int J Emerg Med 2008;1:235. [Oral presentation in 12th National Health Sciences Research Symposium, Karachi, 26-28 August, 2008].

Contributed to Report: 7. Eastern Mediterranean status report on road safety: Call for action. Cairo; World

Health Organization regional office for Eastern Mediterranean: 2010. ISBN 978 92 9021 701 5.

Related to other injuries

Articles published 1. Bhatti JA, Razzak JA. Railway associated injuries in Pakistan. Int J Inj Contr Saf

Promot. 2010 Mar; 17(1):41-44. 2. Butt BA, Bhatti JA, Manzoor MS, Malik KS, Shafi MS. Experience of makeshift

spinal cord injury rehabilitation center established after the 2005 earthquake in Pakistan [Letter]. Disaster Med Public Health Prep 2010;11(1):8-9.

3. Farooq U, Majeed M, Bhatti JA, Khan JS, Razzak JA, Khan MM. Differences in reporting of violence and deliberate self harm related injuries to health and police authorities, Rawalpindi, Pakistan. PLoS One 2010 Feb 23;5(2): e9373.

4. Tachfouti N, Bhatti JA, Nejjari C, Kanjaa N, Salmi L.-R. Emergency trauma care for severe injuries in a Moroccan region: Conformance to French and World Health Organization standards. J Healthc Qual. 2010 [Epub Jun 30]

Manuscript under review 5. Bhatti JA, Shahid M, Mehmood A, Razzak JA, Akbar S, Akhtar U. Epidemiological

patterns of suicide-terrorism in civilian Pakistani population. Int J Inj Contr Saf Promot 2010.

Abstract published 6. Bhatti JA, Shahid M, Mehmood A, Razzak JA, Akbar S, Akhtar U. Suicide terrorism:

what, where, when, and how in Pakistan. J Pak Med Assoc 2010;60(1 Supp);S23-24.

73

Appendices Index of appendices Appendix 1: Literature review on interurban road injury burden in LMICs ........................... 74 Appendix 2- Published article - Study I ................................................................................... 81 Appendix 3: Study I supplementary results ............................................................................. 86 Appendix 4: Manuscript in preparation – Study II .................................................................. 88 Appendix 5: Article published – Study III ............................................................................. 102 Appendix 6: Article under review – Study IV ....................................................................... 110 Appendix 7: Manuscript in preparation – Study V ................................................................ 123 Appendix 8: Study V supplementary results.......................................................................... 137

74

Appendix 1: Literature review on interurban road in jury burden in LMICs Methods Several approaches can be used to review transport safety research such as safety promotion model , Haddon’s matrix, and the three ‘E’ (Enforcement, Engineering, and Education) [57, 156-158]. For this systematic review, the public health approach was used. Considering, this interrelationship of risk factor, crash outcome, trauma care, and their reporting, we reviewed studies if their objective corresponded with one or more of the following research interests:

1. Assessing traffic injury burden (outcome) with respect to age, sex, and road user type. 2. Assessing factors associated with outcome. 3. Assessing trauma care of crash and injury event and their outcome. 4. Assessing reporting mechanism and bias for assessing traffic burden outcomes, risk

factors, and trauma care. 5. Assessing the impact of traffic safety interventions.

The focus was studies with interurban road settings to document research needs. Original research published in peer-reviewed journals indexed on MEDLINE for a period from Jan 1995 to Dec 2009 was included; reviews, editorials, commentaries, and letters to the editor were excluded. Study title and abstract were searched using the following Medical Subject Headings (MeSH): “traffic accident” and “developing countries” or individual names of LMICs given in “Global Status Report on Road Safety.” Then articles concerning interurban roads were selected. Articles written in English, French, Portuguese, Spanish, and Romanian Language were included if they satisfied the following criteria [159]: objective(s) clearly stated; inclusion criteria given and adequate; ethical standards observed; reliable and valid principal measures. Results A total of 3 663 abstracts were retrieved. After removing double entries and case reports, we were left with 1 960 study abstracts (Figure 15). After careful search of studies conducted at highway or interurban road settings, 37 abstracts corresponded to our criteria. Full texts of 32 studies were available; however, we did not include a study reporting the results of traffic management intervention using simulation methods. Road crash and injury burden A Kenyan study showed that almost 60% of police-reported injury crashes occurred on interurban roads [39]. Injuries including fatalities per crash were higher for interurban roads (1.8) than urban roads (1.4). Interurban traffic crash, fatality, and injury per vehicle-km were reported in one Egyptian study only. Average traffic fatality per km was 6 per 100 million vehicles-km (range 3 to 11) in the precedent three-year period [18]. Injury risk was 35 per 100 million vehicle-km (range 29 to 79). Fatality per crash ratio was on average 0.30 (range 0.17-0.43) on Egyptian interurban roads [18]. A Chinese study, where fatalities and daily traffic measures were available, showed that fatality could vary from 31.1 to 72.8 per 100 million vehicle-km travelled on mountainous, un-separated interurban roads [48]. Road users involved in these crashes were mostly aged 15-45 years-old and men [22, 31]. Although some studies showed that pedestrians [37] and passengers of public transport vehicles [19] were over-involved in highway crashes, road user distribution was almost never reported.

75

Figure 15. Literature available for assessing research needs on interurban traffic safety

(1995-2009)

Original articlesN=1 960

Related to highwayN=37

Full text retrievedN=32

Qualified for reviewN=31

BurdenN=5*

RoadBehaviours

N=9*

CrashfactorsN=18*

CareEvaluation

N=0

* A research may have been counted for one or more categories

InterventionsN=6*

Quality of RTC reporting Out of 18 studies on crash factors and outcome, nine used police-related sources [18, 19, 21, 23, 26, 35, 37, 39, 48]. Among them, crash reports were analyzed in only four studies [19, 21, 35, 37]. In three studies, data from traffic agencies was used to assess crash factors, including two where detailed reports were available [28, 40, 45]. Four studies used ambulance-related crash reports which included information on injured patients [30-33]. Emergency department (ED) statistics for assessing RTI burden were used in three studies [22, 40, 43]. Only one study used multiple crash data sources but did not assess discrepancies between them [40]. Quality of documentation was assessed for ED data, only type of data that showed that seat-belt use was reported in 27.3%, crash factors in 18.8%, alcohol use in 18.2%, and extrication methods in 3.1% of crash cases (including highways) reporting to ED [43]. Crash types were mostly defined using police or traffic agency specific definitions, international definitions such as the Microcomputer Accident Analysis Package or External injury codes were available in only five studies [19, 30-33]. Further, traffic fatality in all crash burden studies was not defined according to the WHO definition requiring RTI patient follow-up to 30 days. Similarly, in most studies, injury severity was defined as ‘mild’ mostly when treated at crash scene and ‘severe’ when requiring a hospital visit or admission [30-33]. Only one study defined injuries by using the New Injury Severity Score (NISS) [36]. We found four studies related to identification and reporting of high-risk crash sites on interurban roads in LMICs [22, 28, 35, 39]. A Kenyan study reported that a total of 145 high-risk sites with five or more crashes (N=1 261) were identified on rural road networks in one year, but no details were given regarding whether they were defined using Global Positioning System (GPS) coordinates or other length measures [39]. A Turkish study showed that information on traffic crashes is mostly recorded in textual form and is not geo-referenced, which could pose problems in their identification and improvement afterwards [28]. The authors used GPS coordinates of sites and assessed the usability of this system from a LMIC point of view. Similarly, the Indian study examined the usefulness of GPS coordinates to identify specific injury events on a highway [22]. Using hospital data and onsite investigation,

76

they identified a cluster of long bone fracture as a result of motorcycle crashes. In depth analysis revealed that the presence of speed bump without light resulted in these crashes during low-light conditions. Further, a study from Hong Kong, a relatively resourceful setting, showed that in 12% of police reports, GPS coordinates were not recorded correctly, limiting their utilization by road agencies [35]. Crash factors Road user factors Contribution of road user, vehicle, and road situational factors in interurban crashes was given in only one study [18]. This study showed that these were involved in 65.2% (range 58.6 to 73.5) of interurban road crashes. These included loss of control (30.0%), over-speeding (12.4%), misjudging traffic gap (11.9%), sudden slowing (7.9%), and careless overtaking (6.1%) [18]. Almost all nine studies which analyzed factors associated with crashes and injuries, focussed on road user-related factors. Only three of these used case-control designs whereas rest of them assessed factors associated with crash or injury severity using single source data and cross-sectional designs (Table 17). For instance, a study showed that frontal and pedestrian collisions were more significantly associated with injury crashes than rear end collisions [21]. Similarly, several studies confirmed that DWI was significantly associated with injury crashes on interurban roads [30, 31]. Moreover, not wearing a seat-belt or a helmet was shown to be significantly associated with injury crashes in two different settings [31, 33, 36]. An interstate bus driver survey showed that Body Mass Index (BMI) ≥ 30 km/m² was significantly associated with drowsiness while driving (50.0% vs. 30.1% in those with BMI < 30 kg/m²; P <0.05), and involvement in a sleep related crash (13.0% vs. 6.5%; P <0.05) than [46]. A Chinese population-based case-control study showed that crashes were more likely in drivers reporting shift or night work than in those driving during daylight or without shifts [34]. Vehicle factors Vehicle factors were reportedly involved in 22.5% (range 15.5-32.8) of interurban road crashes in a LMIC [18]. Common circumstances were tire burst and over turning of vehicle [18]. Similar involvements were shown in a South-African study where brake and tire problems were common vehicle factors involved in crashes [45]. Furthermore, the Kenyan study suggested that inappropriate inner vehicle design could increase injury severity of passenger vehicle users [19]. A survey of vehicles travelling on South-African highways showed that 29% of vehicles had potential mechanical defects such as overloading (10%), brake problems (8%), and worn-off tyres (5%). Most passenger vehicles’ tyres (52-56%) were underinflated and those vehicles had travelled over 300 000 kilometres. Road environmental factors Road environment-related factors were identified in 6.2% (1.2 to 8.5) of interurban road crashes in a LMIC [18]. Among them, notable causes were U-turns2 (2.7%) and bad weather conditions (2.9%) [18]. Urban areas, in particular intersections, along the highways are frequent focal points of RTCs [19, 26, 37]. A study from Turkey examining the sensitivity of the crash prediction model showed that average daily traffic is one of the principal parameters explaining number of traffic crashes on interurban roads [20]. A cross-sectional analytical study showed that injury crashes were significantly higher for substandard pavement conditions than for unpaved road sections [21].

2 U-turns were considered as situational risk factors in this study. Typically, they represent interactions between situational (absence of hazard signs, pavement marking, road barriers, and traffic enforcement) and risk-taking behaviours.

77

Behaviours Speeding Prevalence of risky road behaviours on interurban studies was assessed by few studies. For instance, a study reported that speed alone could account for over 50% of the RTCs reported in Ghana [19]. In two different studies, over-speeding was measured as a function of vehicle and road type on interurban road sections [24, 25]. Both studies showed that 90% of the vehicles travelling through built-up areas on such roads exceeded posted speed limits. Even on rural road sections, nearly half of the sampled vehicles had exceeded posted speed limits. Mean speeds varied with road types, with higher speeds noted for national highways than on regional and inter-regional highways. Further, the highest vehicle speeds were associated with private cars and large buses [25]. Driving while intoxicated (DWI) A Nigerian study showed that 44.6% of the drivers involved in traffic violations on highways had Blood Alcohol Concentration (BAC) higher than 0.05% [27]. A Brazilian study showed that alcohol consumption could be higher in some risk groups such as truckers, where weekly incidence of DWI could be as high as 91% [38]. Nearly half consumed it at the fuel station. Similarly, the incidence of DWI in Cuban commercial drivers on highways was 8.2% (N=66/832; 95% CI=5.9, 10.4), with 20% of them having a BAC ≥ 0.05% [29]. Sleepy driving Long working hours was an important cause of sleepiness in Brazil [38]. A Peruvian highway bus driver survey showed that up to 80% of them drove continuously for more than five hours [41]. Sleep deprivation was measured in two different studies from Latin America, suggesting that nearly one-fourth of commercial drivers slept less than 5 hours in the preceding 24-hour period [41, 44]. Use of stimulant drugs In relation to sleepiness in commercial drivers, several studies assessed stimulant use in these types of samples. A Peruvian study reported that 14% of them used coffee, 4% smoked, 4% chewed coca, and 2% took alcohol mixed with coca leaves [41]. Higher caffeine use (95%) was reported elsewhere [44]. A Turkish study showed that 75% of such drivers used a medicinal drug with caffeine and paracetamol while driving, mostly for headache and fatigue. Its use resulted in sedation for 30 to 60 min (78.5%), stumbling (21.5%), and loss in visual acuity (6.5%) [47]. Two Brazilian studies reported that stimulants such as amphetamines were used by commercial truck drivers to cope with sleepiness [38, 44]. According to one [44], its use was 66%, while the other reported it to be around 11%. These substances were available on the highway and most of them used it for night driving [38]. Pre-hospital and essential trauma care We found four studies conducted on the Mexico-Cuernavaca highway that indicated that there was some pre-hospital care system on those road sections [30-32]. A brief description was available, but structure, process, and outcome of care were not described. An urban hospital-based study showed that 15.9% of all injured patients received in ED had been involved in interurban road crashes [43]. A majority of these patients were transported by private means. None of the study reported commonly known evaluation parameters such as response time or length of stay for those injured in interurban road crashes. Evidence on interventions Several prevention and control measures were evaluated, almost all in non-controlled studies with a before and after study design (Table 18). A study showed that installation of rumble strips on a busy junction on a highway decreased crashes by 35% and fatalities by 55% [19]; a 51-percent decrease in pedestrian collisions was recorded. This intervention was four times

78

more cost-effective than redesigning a junction or installing a pedestrian bridge. Similarly, a Ugandan study observed the influence of overhead pedestrian bridge installation on pedestrian crashes on the Kampala-Jinja highway [37]. No significant differences were observed in pedestrian crash count between the two periods. Females and children used the bridge more than men and adults. A convenient sample of 123 pedestrians reported that 52.0% of them used it. A majority of those not using it reported extra walking distance and time and high stairs as causes of their non-use. Traffic enforcement with increased fines and a rigid penalty scoring on Brazilian highways system decreased fatalities by 24.7%, hospital admissions 33.2%, and RTCs by 21.3%. Stricter penalties almost halved the tickets issued to drivers [40]. Similar results were observed on four sub-urban road sections in Uganda, when more resources were provided to the highway police. Adequately equipped traffic police facilitated a drop in road fatalities by 27.3% (95% confidence interval (95% CI)= 6.0; 48.9].[23]. Considering average age of person dying due to crash and life expectancy, traffic enforcement could save up to USD 27 (95% CI=15-118] per disability adjusted life year saved year at 3% discount rate. Traffic enforcement appeared to be one of the most cost-effective behavioural interventions and overall system improvement approach. Two separate studies showed that road environmental improvements lead to significant reductions in injury crash rates. For instance, multi-sector intervention involving removal of hard metal bars and installation of road furniture, luminous traffic signs, and security fence decreased injuries from 2.1 to 1.4 per 10 000 vehicles adjusted on seat-belt use on a Mexican highway [32]. Similarly, security fence installation on a mountainous Chinese highway decreased overall traffic crashes by 43.8% and severe crashes by 70.8% [48]. Overall design improvement on the same highway led to a decrease in crashes by 37.2% and traffic fatalities by 47.1% as compared to the other highways adjusted on traffic counts.

79

Table 17. Analytical studies of traffic crash and injury risk on interurban road sections in developing countries. Author (year) Objective Study design Setting Data Outcome Results Almeida, et al (2009)

Factors associated with injury crashes

Cross-sectional Federal highway 163, Brazil

Highway police data for year 2004 Injury crash in which at least one person has been injured or died

Associations were assessed for three different set of independent variables: Surface condition: Substandard pavement OR=1.89; 95%CI=1.32-2.70 (Reference: unpaved) Crash types: Frontal (OR=14.14; 95%CI=8.96-22.32) and pedestrian (OR=35.95; 95%CI=8.10-159.92) collisions (reference: rear collisions) Contributory factors: Highway maintenance problems (OR=4.35; 95%CI=1.94-9.75) and disobeying traffic signs (OR=5.69; 95%CI=2.01-16.12) (reference: not keeping safe distance)

Hijar, et al (1996) Association of non-use of seat belt with injury severity

Cross-sectional Mexico–Cuernavaca highway, Mexico

Seven-month ambulance data (1994) Crash requiring hospital treatment

Not wearing a seat-belt (OR=2.94; 95% CI=1.13-7.66) was significantly associated with injury crash

Hijar, et al (1998) Association of alcohol intake with injury crashes

Cross-sectional As above As above Crash requiring hospital treatment

Alcohol intake (OR=6.09; 95%CI=1.55-24.00) was significantly associated with injury crashes

Hijar, et al (2000) Factors associated with highway crashes

Case-control As above Three-month ambulance data (1996: Cases, involved in crash & transported by ambulance; controls, not involved in a crash, recruited from two points.

Crash Age <25 years (OR=3.18; 95%CI=1.53-6.57), alcohol intake (OR=5.02; 95%CI=1.71-14.72), Mexico-Cuernavaca direction (OR=2.68; 95%CI=1.67-4.33), weekday (OR=2.69; 95%CI=1.66-4.38), daylight (OR=4.16; 95%CI=2.31-7.48), and adverse weather (OR=5.56; 95%CI=3.57-8.66) were significantly associated with injury crashes.

Liu, et al (2003) Driver sleepiness and risk of crash

Case-control Highways in Huanggu district, China

Nine-month traffic & police data (2001-02): Cases, drivers involved in a traffic crash; controls, randomly selected with police officers from 28 road locations

Police reported crash

Daytime sleepiness measured by Epworth Sleepiness Scale (ESS) > 10 (OR=2.07; 95%CI=1.30-3.29) and night or shift work (OR=2.14; 95%CI=1.50-3.05) were significantly associated with police reported crash. Associations remained significant even after removing those with alcohol intake determined by breath-analyzers applied to both case and control groups.

Majdzadeh, et al (2008)

Factors associated with injuries in drivers & motorcyclists

Case-control Qazvin-Loshan Road, Iran

Four-month traffic & police data (2005): Cases, injured drivers & motorcyclists involved in a crash; controls, uninjured drivers & motorcyclists on the same road involved in a crash.

Mild traffic injury (NISS ≤ 15) and moderate traffic injury (NISS > 15)

Female sex (OR=7.78; 95%CI=2.77-21.85), safety equipment use (OR=0.44; 95%CI=0.23-0.84), and motorcycle involvement (OR=5.06; 95%CI=1.42-18.02) were significantly associated with mild injury crashes. Female sex (OR=4.78; 95%CI=1.36-16.80) and adverse weather condition (OR=4.32, 95%CI=1.13-16.50) were significantly associated with moderate injury crashes.

Rey de Castro, et al (2004)

Association of fatigue and drowsiness with reported crashes in bus drivers

Cross-sectional Northern Pan American highway, Peru

Bus drivers operating at bus-stand on km 10 of the highway. Interviews were conducted with 238 out of 400 drivers working on the bus stand.

Crash or near-crash

Forty five percent drivers reported having a crash or near crash in a precedent period. Driving hours per day was significantly higher in drivers who reported a crash or near crash than drivers not reporting such event (7.9 vs. 6.9 hours, P=0.003). Similarly, 35% drivers reported driving while sleepy, which was significantly higher in those reporting a crash or near crash than drivers not reporting such event (P<0.001); proportions were not given.

Souza, et al (2005) Association of sleepiness with crashes

Cross-sectional Federal highways, Brazil

Interviews conducted with 260 truck drivers

Crash Thirteen percent (N=27) drivers were involved in a crash in precedent five years. Mean age (34.1 vs. 38.8; P=0.03) was lower in drivers involved in crashes than those not involved. Excessive daytime sleepiness measured by ESS was significantly higher in drivers involved in crash than those not involved in crash (8.2 vs. 6.3; P=0.02).

Viegas et al (2005) Prevalence of risk factors for sleep apnoea syndrome in commercial bus drivers

Cross-sectional Federal highways, Brazil

Interviews conducted with 262 male interstate bus drivers

Crash Fifty percent drivers had a Body Mass Index (BMI) ranging from 25.0 to 29.9 kg/m² and 17% had BMI ≥ 30 kg/m². Forty eight percent reported drowsiness while driving and 42% of them were involved in crashes. BMI ≥ 30kg/m² was significantly associated with drowsiness while driving (50.0% vs. 30.1% in those with BMI < 30 kg/m²; P <0.05) and involvement in a sleep related crash (13.0% vs. 6.5%; P <0.05).

OR: adjusted odds ratio 95% CI: 95% confidence interval NISS: New Injury Severity Score

80

Table 18. Traffic injury intervention studies on interurban road sections in developing countries. Author (year) Intervention Study design Setting Data Outcome Results Afukaar (2003) Rumble strips on an urban

intersection Before and after study without control

Accra-Kumasi highway, Ghana

Police data Per year 1. Crash (N) 2. Crash type 3. Fatality (N)

Road crashes decreased by 55% and road fatalities decreased by 35%. Head-on, side-swipes, and hitting fixed object crashes decreased by 100%, whereas pedestrian involving crashes decreased by 51%.

Hijar , et al (1999) Road furniture: luminous signs, traffic lights, rumble strips, and replacement of lateral metal bars by security fence

Before-after without control group

Mexico–Cuernavaca highway, Mexico

Ambulance (pre-hospital care data) for two period 1994 & 1996

Per 10 000 vehicles travelled 1. Crash 2. Injury

No significant differences were observed for crashes between the two periods. Injury per 10 000 vehicles significantly reduced from 2.1 to 1.4 over the same period. Results were adjusted on age, speed, use of seat belt, alcohol intake, and external causes of injuries.

Bishai, et al (2008) Traffic police enforcement with vehicles and speed camera

Before and after study without control

Four highways entering Kampala, Uganda

Police data

Per year 1. Citation (N) Per month 2. Fatality (N)

Number of citations increased from USD 72 000 in before to 327 311 in after period. A 17% drop in road fatality per month was recorded.

Poli de Figueiredo, et al (2001)

1. Increased fines 2. Rigid penalty scoring

Before and after study without control. Intervention applied in 1998

Interstate highways, Brazil

Traffic agency data & level I trauma care data for 1998 compared to 1997

1. Crashes 2. Deaths 3. Injuries 4. Citations 5. Emergency admissions

All the outcome variables decreased in the after period as compared to before period. Crashes decreased by 21.3% (327 640 to 257 688); road deaths decreased by 24.7% saving 5 962 lives; traffic citations decreased by 49.5% (601 977 to 304 785). Similarly, emergency room admissions decreased by 33.2% (787 to 526).

Mutto, et al (2002) Overhead pedestrian bridge in a locality situated on a highway

Before and after study without control

Kampala-Jinja highway

Police data 1. Crash 2. Crash severity

Total crashes augmented from 35 to 71. The difference was +74.5% (51 vs. 13) for minor, +21.4% (17 vs. 14) serious, and -75.0% (2 vs. 8) for fatal crash.

Zhou, et al (2005) 1. Guardrail installation 2. Design improvements

Two comparisons; 1. Before after study (guardrail) 2. Control sections (design improvements)

Three sections of National highway 319 Wulong (intervention), Pengshui & Qianjiang counties

Traffic police 1. Crashes 2. Crash severity 3. Deaths

As compared to before period, traffic crashes decreased by 43.6% (84 vs. 149) and severe crashes decreased by 70.8% (7 vs. 24) in the after period. No changes in overall fatalities were observed; as compared to similar road sections, a decrease in crashes (≥ -37.2%) and traffic deaths (≥ -47.1%) was observed in the same period (1997-2001).

81

Appendix 2- Published article - Study I

82

83

84

85

86

Appendix 3: Study I supplementary results

Figure 16. Weekly pattern of traffic fatalities and injuries on Yaoundé-Douala road section (2004-2007)

0

50

100

150

200

250

300

350

400

Monday Tuesday Wednesday Thursday Friday Saturday Sunday

Day

N

Fatalities

Injuries

Figure 17. Hourly pattern of traffic crashes and fatalities on Yaoundé-Douala road section (2004-2007)

0

10

20

30

40

50

60

70

80

90

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Hour

N

Crashes

Fatality

87

Table 19. Traffic fatalities and injuries according to crash types and causes on Yaoundé-Douala road section (2004-3007)

Crashes Non-fatal

injuries Injury per

crash Fatalities Fatality

per crash N N N Crash types Vehicles travelling in the same direction 181 211 1.2 31 0.2 Single vehicle running off the road 180 399 2.2 69 0.4 Vehicles travelling in opposite directions 156 313 2.0 136 0.9 One or more pedestrians 141 150 1.1 71 0.5 One or more still or manoeuvring vehicles 84 122 1.5 11 0.1 One or more two-wheeled motor vehicles 81 99 1.2 29 0.4 Intersection 62 104 1.2 3 0.1 Others 50 96 1.9 24 0.5 Crash causes* Human factors Hazardous overtaking 268 454 0.6 87 0.3 Excessive speed 182 355 2.0 74 0.5 Inattention, distraction 136 177 1.3 27 0.2 Loss of control 120 194 1.6 42 0.4 Hazardous manoeuvre 114 124 1.1 18 0.2 Unsafe parking 74 126 1.7 14 0.2 Other human factors 50 83 1.1 40 0.8 Any human factors 700 1093 1.6 224 0.3 Mechanical failures Tyre puncture, burst or loss 98 266 2.7 66 0.7 Other mechanical failures 71 153 2.2 38 0.5 Any mechanical failures 168 418 2.5 103 0.6 Environmental factors 37 49 1.3 21 0.6 Unknown causes 91 84 0.9 56 0.6 Total 935 1494 1.6 374 0.4

* A crash may appear more than one time

88

Appendix 4: Manuscript in preparation – Study II TITLE DIFFERENCES IN POLICE, AMBULANCE, AND EMERGENCY DEPARTMENT REPORTING OF TRAFFIC INJURIES ON KARACHI-HALA ROAD, PAKISTAN AUTHORS 1. Junaid Ahmad BHATTI a, 2. Junaid Abdul RAZZAK b 3. Emmanuel LAGARDE a 4. L.-Rachid SALMI a, c, d

AFFILIATIONS a. Equipe Avenir « Prévention et Prise en Charge des Traumatismes », Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897), Bordeaux, France. b. Department of Emergency Medicine, The Aga Khan University, Karachi, Pakistan. c. Institut de Santé Publique, d’Epidémiologie et de Développement (ISPED), Université Victor Segalen Bordeaux 2, Bordeaux, France. d. Service d’information médicale, Centre Hospitalier Universitaire de Bordeaux, Bordeaux, France. CORRESPONDING AUTHOR Junaid A. BHATTI Equipe Avenir « Prévention et Prise en Charge des Traumatismes » Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897) 146 rue Léo Saignat 33076 Bordeaux cedex France Tel: + 33 (5) 57 57 45 50 Fax: + 33 (5) 57 57 45 28 Email: [email protected] COUNTS Abstract: 213 words Manuscript: 2 782 words References: 36 Tables: 3 Figure: 1

89

ABSTRACT Background: Previous research in developing countries assessed discrepancies in police Road Traffic Injury (RTI) reporting for urban settings only. The objective of this study was to assess differences in RTI reporting among police, ambulance, and Emergency Department (ED) datasets on an interurban road section in Pakistan. Methods: The study setting was 196-km long Karachi-Hala road section. RTIs reported to the police, Edhi Ambulance Service (EAS), and five EDs in Karachi during 2008 (Jan to Dec) were compared. Further, records from these data were matched to assess their ascertainment. Results: A total of 143 RTIs were reported to the police, 531 to EAS, and 661 to ED. Fatality per hundred traffic injuries was twice as high in police records (N=27, 18.8%) than in ambulance (N=55, 10.4%) and hospital records (N=60, 9.1%). On the contrary, pedestrian and motorcyclist involvement per hundred traffic injuries was twice as low in police records (N=11, 7.7%) than in ambulance (N=89, 16.7%) and hospital records (N=286, 43.3%). Based on matching, police recorded 22.6%, EAS 46.2%, and ED 50.4% of the 119 reportedly died patients. Similarly, police data accounted for 10.6%, EAS 43.5%, and ED 54.9% of the 1 095 reportedly injured patients. Conclusion: Police reporting, particularly of non-fatal RTIs and those involving vulnerable road users, should be improved in Pakistan. KEYWORDS: Highway; injury severity; surveillance; traffic accident. ABBREVIATIONS: EAS, Edhi Ambulance Service; ED, Emergency Department; HIC, High-Income Country; LMIC, Low- and Middle-Income Country; NHMP, National Highway & Motorway Police; NISS, New Injury Severity Scores; RTC, road traffic crash; RTI, road traffic injury; US, United States of America; WHO, World Health Organization; $, Dollar.

90

BACKGROUND Pakistan, located at the junction of Middle-East, South-East, and Central Asia , is the sixth most populous nation of the world [1]. According to transport authorities, approximately 1.4 million Road Traffic Crashes (RTCs) occurred in Pakistan in 1999, resulting in over 7 000 fatalities [2]. Two independent population-based surveys estimated incidence of Road Traffic Injuries (RTIs) around 15 to 17 per 1 000 persons per year [3, 4]. These injuries contribute significantly to the workload in hospitals, leading to direct costs of over 1 billon US$ to the Pakistani economy [5, 6]. Interurban road sections are the backbone of Pakistani economy. Its strategic road network of approximately 8 000 km plays a significant role in transport, as it carries more than 80% of inland passenger and freight traffic [2, 7]. Although these road sections account for 4% of the entire network, they account for a high proportion of traffic fatalities (27%) [8]. Previous research in Pakistan has shown that injury severity was higher for crashes in rural areas, but no distinction was made between interurban or other rural roads [9]. Higher speeds, presence of vulnerable road users, and complex road traffic conditions can explain this high fatality ratio, but no comparison indicators were available for these road sections [10]. Police records remain, to date, the most used source for evaluating interurban traffic safety, because of geographical distances and complexity of trauma care in such settings [9, 11]. The use of these statistics only, however, can lead to underestimation of RTI burden in Low- and Middle-Income Countries (LMICs) like Pakistan [12]. A recent World Health Organization (WHO) report showed that actual traffic fatalities could be 4 to 10 times higher than the official statistics in Pakistan [13]. A previous study in Karachi city showed that police records accounted for only 56% of traffic fatalities and 4% of such severe injuries [14]. No notable research has been carried out to compare the differences in injury reporting by linking different datasets for interurban road settings in Pakistan [12, 13]. The World Bank reported that interventions with proven effectiveness exist but their implementations are impeded by the lack of documenting the specific disease burden in LMICs [15]. The objective of this study were to assess differences in crash and injury reporting between police, ambulance, and Emergency Department (ED) datasets on an interurban road section in Pakistan. Further, these datasets were linked to assess variations in traffic fatality and injury per vehicle-kilometres (vehicle-km) travelled on the road section. METHODS The study setting was the 196-km-long Karachi-Hala road section (km 16 to km 212 from Karachi city centre). This is a four-lane highway, two lanes in each direction [16]. The lanes are separated by a ground surface, but there are no physical barriers. Traffic counts range 16 356 to 24 707 vehicles per day on this section [17]. This high traffic count can be explained by the economic activity in Karachi, the most populous city of Pakistan, which accounts for 70% of the government’s revenue through trade and industry [18]. In this retrospective study, information on traffic injuries reported to highway police, ambulance service, and ED during 2008 (Jan to Dec) was collected and compared. Police data Since 2004, the National Highway & Motorway Police (NHMP) ensures traffic enforcement on this road section. Administratively, this section is considered as Sector I of South-Zone of NHMP and is divided further in four 46- to 51-km-long beats: beat 35 (km 16 to 62), beat 34

91

(63 to 114), beat 33 (115 to 162 km), and beat 32 (163 to 212 km). NHMP deploys four motor vehicles and four patrolling officers in an eight-hour shift on these beats [19]. For every crash, a standard accident analysis report is filed by the attending NHMP officer during the first 24 hours [20]. A copy of this report is kept in the NHMP regional office. Similarly, details on crash and those involved are recorded on a separate accident register. From these reports and registers, information was extracted on time, date, location of crash, and whether it was fatal, involved injury, or was without injury. We also extracted information on name, age, sex, outcome (dead; severe injury, defined as transported to hospital; and mild injury, defined as not transported to hospital), and hospital brought to of those involved in crashes. Ambulance data Ambulance records were obtained from Edhi Ambulance Service (EAS) logbooks. EAS is the largest private philanthropic ambulance service in the world [21]. Since 1973, the EAS has been progressively increasing its ambulance posts from main Pakistani cities to the important highways in Pakistan [22, 23]. For transporting injured patients, EAS has established six ambulance posts, mostly near main towns on Karachi-Hala road section: 1/ Sohrab Goth (12 km from Karachi centre), 2/ Karachi toll plaza (km 28), 3/ Edhi centre (km 56), 4/ Nooriabad (km 94), 5/ Hala Naka (km 160), and 6/ Hala city (km 212). This service is freely available to injury patients, and funds are raised by transporting other patients. Ambulance staff consists of, in most of cases, only the driver. A clerk at the post can accompany the driver if he thinks this justified, for instance in case of crash with multiple patients. Ambulance communicates with emergency post through a wireless system or by cell phone. RTI patients or bystanders can contact EAS using the free emergency access number 115, which connects them to the main city centre [21]. Information is then transmitted by wireless or cell phone to nearby posts, which finally dispatches the ambulance(s). After reaching the scene, injured and dead patients are separated. Those severely injured are transported to the nearest hospital; preference is given to the government hospital if available. All information on the RTI intervention, including crash location, RTI patient identity and outcome, is then transmitted by wireless or telephone to the regional centre, which records the information in a central log book. We photocopied these log books from the regional centre at Karachi. Crash details such as date, time, location, and whether it was fatal or involved injury were extracted from these books. Similarly, road user details such as name, sex, age, user type (pedestrian, motorcycle rider, or vehicle occupant), and outcome (died, including when the person died at crash scene, during transport, or at ED; injured and transported, including hospital taken to; injured and not transported) were extracted from these books [21]. Hospital records The Road Traffic Injury Research & Prevention Centre (RTIRP) at the Jinnah Post Graduate Medical Centre (JPMC) has been working since September 2006 [24]. This centre systematically collects, on standard Proforma sheets, information on RTI patients presenting at the ED of the five largest teaching hospitals in Karachi: 1/ JPMC, 2/ Abbasi Shaheed Hospital, 3/ Civil Hospital Karachi, 4/ Liaqat National Hospital, and 5/ The Aga Khan University Hospital. Details on their data collection methods are available elsewhere [24, 25]. This dataset includes information on the crash date, time, and location as well as patient’s name, age, sex, road user type (pedestrian, motorcycle rider, or vehicle occupant). Further information on whether the patient was wearing a helmet or seat belt was available. The New Injury Severity Scores (NISS) [26] and outcome (discharged, admitted/referred, or died) of

92

patients were recorded during their stay in the ED. Information on RTI patients involved in crashes on selected road section was extracted from this dataset. Analysis All information was recorded on Excel® spreadsheets. We compared percentages for crash and injury patient characteristics for three datasets. For the ED dataset, we compared outcome for following NISS categories: 1 to 3; 4 to 8; and ≥ 9. Records from the three datasets were then matched for crash date, name, age, and sex of RTI patients involved. For matched records, we identified differences in reported outcome. A person reported injured in police statistics, but dead in ambulance data was considered as dead. Number of unique deaths and injuries were then assessed by removing records appearing in two or more datasets. Ascertainment rate for police, ambulance, and ED records as compared to these total fatalities and injuries were computed [27]. Capture-recapture methods were not used to estimate road burden because RTIs away from Karachi might not have the same probability of being captured in the ED dataset. These unique records and traffic counts from NHMP were used to compute overall traffic fatality and injury rates per vehicle-km in 2008 for this road section [28]. RESULTS Crash and injury outcome In 2008, police reported 43 crashes, whereas 255 crashes were reported to EAS and 449 to ED. One out of two police reported crashes (N=19, 44.4%) was fatal, whereas this proportion was 14.5% (N=37) for those reported to EAS, and 10.4% (N=47) for ED. No information on crash outcome was available in 13.3% of EAS reported crashes, and 6.7% of those reported to ED. A total of 143 RTIs were reported to police, 531 to EAS, and 661 to ED. Monthly trends indicated higher proportions of RTIs in June and July 2008. Over half of police-reported injury patients received hospital care (N=80, 55.9%). Half of these patients (N=40), injured between km 16 and km 120 were treated in Karachi; RTIRP hospitals treated 17 of them. Nearly one fifth of RTI patients reported in police records died (N=27, 18.8%), whereas this proportion was 10.4% for EAS- and 9.1% for ED-reported patients (Table 1). One fourth of police-reported injury patients (N=25.2%) were not transported to the hospital, whereas this was 9.0% for EAS-reported patients (N=48). Out of 661 patients presenting to ED, 47.7% (N=315) arrived by private means, whereas 43.0% (N=284) arrived in ambulances. Police transported only four of these patients, and no information was available on the remaining 58 patients. In the ED, those with a NISS from 4 to 8 had a higher likelihood of hospital admission than those with NISS from 1 to 3 (81.0% vs. 19.0%, P<0.001). Those who were reported to have died had a NISS of 10 or more. Patient characteristics Names were available for 67.1% of police- and 78.0% of EAS-reported injury patients (Table 1). Information on age was available for 74.1% of police- and 67.6% of EAS-reported injury patients. Few records in the ED dataset were without names (N=12) or age (N=5). Most injury patients in the three datasets were aged 16-45 years: 61.5% in police, 55.0% in EAS, and 78.1% in ED. Men accounted for a majority of injuries, up to 92.1% of injury patients in ED. The proportion of pedestrians in police-reported crashes was 3.5% (N=5), whereas this was 7.5% in the EAS and 12.7% in the ED. The proportion of motorcycle riders in police-reported crashes were 4.2%, whereas this was 9.2% in EAS and 30.6% in ED. Occupants of four-

93

wheeled vehicles accounted for a majority of injuries in the three datasets: 83.9% in police, 75.9% in EAS, and 49.5% in ED. In the ED, only 15.7% (N=21) of the 154 injury patients riding motorcycles were wearing helmets. Similarly, only three out of 93 four-wheeled vehicle occupants were wearing a seat belt at time of crash. Concordance between databases Matching yielded 1 214 unique records from the three datasets (Figure 1). A total of 108 patients were found in two or more datasets, including 13 who were found in all datasets, 28 who were found in police and EAS datasets, and 14 who were found in both police and ED datasets; 93 records were common between ambulance and ED datasets. Discrepancies were observed for outcome of injuries reported in police and ambulance records (Table 2): four out of the 17 injured in police dataset were reported dead in EAS records. Similarly, one of eight injured in police records was reported dead in ED records, and nine of 84 injured patients in EAS were reported dead in ED records. Ascertainment of road fatalities and injuries Based on matching, 119 unique patients were reported to have died in 2008 on this interurban road section (Table 3). Police recorded 22.6%, EAS 46.2%, and ED 50.4% of them. Similarly, a total of 1 095 patients were reported injured in three datasets. Police accounted for 10.6%, EAS 43.5%, and ED 54.9%. Traffic fatality was 54 deaths and injuries were slightly over 500 per 109 vehicle-km travelled on this road section. Matching of nameless police and ambulance records, when any of the crash dates, time, age, and sex details was available, decreased the overall estimates by 4 deaths and 73 injuries. Corrected traffic fatality was 53 deaths and injuries were 467 per 109 vehicle-km travelled on this road section. DISCUSSION This study showed crash and injury numbers reported by police were several times less than ambulance and ED data on this road section in a one-year period [27]. Fatality per hundred traffic injuries was twice as higher in police records than in ambulance and hospital records. On the contrary, pedestrian and motorcyclist involvement per hundred traffic injuries was twice as less in police records than in ambulance and hospital records. Compared to overall estimated RTIs, police reported one in five traffic fatalities and one in ten severe injuries on this road section. Underreporting in police traffic crash data cannot be denied, particularly in LMICs. This study showed that this could be particularly high for road sections outside the cities. Police accounted for only one of five traffic fatalities, compared to one out of two in Karachi city [14]. This type of documenting disparity could jeopardize the resource allocation for traffic safety interventions in these settings [29]. Police reporting has been more reliable and complete in high-income countries (HICs) and this improvement should be an important priority in LMICs like Pakistan, to better estimate and monitor traffic safety programs in these countries [12, 30]. The proportion of traffic fatalities was higher in police than in both EAS and ED records. In Pakistan and many other LMICs, police performance is judged by few event numbers [31]. Since RTCs are part of these statistics, higher traffic injury numbers could reflect poor enforcement. It is possible that police records did not include non-fatal traffic injuries because of such reasons [14, 19]. These results showed that this differential reporting needs to be considered seriously. Documentation might be improved by implementing performance evaluation based on number of crashes in which the police intervened for public safety [31].

94

This might motivate police officers to report RTIs, to better identify the high-risk groups and crash sites [13]. Furthermore, police reported fewer pedestrian and motorcyclist involvement per hundred traffic injuries. There could be several explanations: Firstly, it is likely that these injuries took place near built-up areas, so patients were transported by bystanders or ambulances directly to hospital, without police intervention [13, 32]. Secondly, it is possible that such road users belonged to lower socioeconomic status and did not want to be involved in cumbersome and expensive legal procedures, and settled their issues without police [19]. Nevertheless, efforts are required to improve documentation of such road users to better design and implement effective crash prevention policies [33]. Limitations of secondary datasets such as ambulance or ED for RTC prevention have been considered previously in Pakistan [34]. Availability of NISS was exceptional in this study, because of the existing RTI surveillance system [24]. It was observed that both EAS and ED recorded the approximate location of traffic crash (town, motel…), whereas police data included the km location of the sites. Linking of these datasets permitted to show a high crash and injury burden, but failed to identify high-risk crash sites. Moreover, seat-belt and helmet use was not reported in a majority of ED patients, and not recorded at all in police data. This shows the need to improve police reporting of crash factors, information that could help in developing policies adapted to local settings [11]. Finally, this study may have some limitation regarding RTI estimates because names were not available for one of three police and one of five ambulance records [34]. Some of these police and ambulance records could be matched with only one common parameter, thus RTIs could be slightly overestimated in this study. Nevertheless, corrected fatality and injury rates were higher than a similar road in an HIC [35]. Moreover, fatality numbers could be even higher, because patients were not followed for over 30 days, as in the WHO definition [13]. Furthermore, half of the police-reported patients were injured away from Karachi and were transported to hospitals outside Karachi [32]. This shows that the ascertainment of police records could be much lower than reported in this study. In conclusion, interurban traffic crash burden appears to be several times higher in Pakistan than other HICs [35]. Police RTI documentation, particularly of non-fatal injuries and those involving vulnerable road users, should be improved in Pakistan [12, 14, 34]. Revising police performance evaluation, to account for number of traffic crashes in which the police intervened, might motivate officers to report RTIs [13, 36]. Furthermore, a linked and comprehensive database would be useful to monitor and implement traffic safety interventions in Pakistan [14]. ACKNOWLEDGEMENTS We are especially thankful to Dr. Aftab Ahmed PATHAN, Deputy Inspector General of Police, Mr. Irshad SODHAR, Senior Patrolling Officer, and Mr. Naeemullah SHIEKH, Senior Patrolling Officer, National Highway and Motorway Police south sector III office, Pakistan for their support in data collection. We are also thankful to Pr. Rasheed JOOMA (JPMC) and Mr. Ameer HUSSAIN (JPMC) for providing us the ED data. Special thanks to Mr. Faisal EDHI for providing us the EAS log books. ETHICAL APPROVAL

95

All the police, ambulance, and ED data used in this study was publicly accessible and data analysis was conducted with approval from their respective institutions. Furthermore, this manuscript does not permit identification of any RTI patient. AUTHORS’ CONTRIBUTION This study is the part of PhD thesis work of JB supervised by LRS who contributed equally to study conception, design, analysis, and manuscript writing. JAR and EL provided technical help in all of the above work. COMPETING INTERESTS The authors declare that they have no competing interests. FUNDING First author is the PhD candidate at Université Victor Segalen Bordeaux 2. This position is funded by Higher Education Commission of Pakistan. Institut National de la Santé et de la Recherche Médicale Unité 897, France, funded the logistics for data collection. Funding bodies had no input in study design, analysis and interpretation of results. REFERENCES 1. Central Intelligence Agency: The World Factbook. Langley, VA: Directorate of

Intelligence; 2009. 2. National Transport Research Centre: Manual of road safety improvement by the

use of low cost engineering countermeasures. Islamabad: National Transport Research Centre, National Highway Authority, and Finnroad OY; 1999.

3. Ghaffar A, Hyder AA, Masud TI: The burden of road traffic injuries in developing countries: the 1st national injury survey of Pakistan. Public health 2004, 118(3):211-217.

4. Fatmi Z, Hadden WC, Razzak JA, Qureshi HI, Hyder AA, Pappas G: Incidence, patterns and severity of reported unintentional injuries in Pakistan for persons five years and older: results of the National Health Survey of Pakistan 1990-94. BMC public health 2007, 7:152.

5. Ahmed A: Road Safety in Pakistan. Islamabad: National Road Safety Secreteriat; 2007.

6. Raja IA, Vohra AH, Ahmed M: Neurotrauma in Pakistan. World journal of surgery 2001, 25(9):1230-1237.

7. Government of Pakistan: Pakistan Transport Plan Study in the Islamic Republic Of Pakistan. Islamabad: tripartite collaboration of the Japan International Cooperation Agency (JICA); National Transport Research Centre (NTRC), and Ministry of Communications, Government of Pakistan; 2007.

8. National Transport Research Centre: Traffic Counter measures in Pakistan. Islamabad: National Transport Research Centre, Ministry of Communication; 1985.

9. Shah SG, Khoumbati K, Soomro B: The pattern of deaths in road traffic crashes in Sindh, Pakistan. International journal of injury control and safety promotion 2007, 14(4):231-239.

10. Oxley J, Corben B, Koppel S, Fildes B, Jacques N, Symmons M, Johnston I: Cost-effective infrastructure measures on rural roads. Clayton, Victoria: Monash University Accident Research Centre; 2004.

11. Wootton J, Jacobs GD: Safe roads: A dream or a reality. Crowthorne: Transport Research Laboratory; 1996.

96

12. Peden M, Scurfiled R, Sleet D, Mohan D, Hyder A, Jarawan E: World report on road traffic injury prevention . Geneva: World Health Organization; 2004.

13. WHO: Global status report on road safety. Geneva: World Health Organization (WHO); 2009.

14. Razzak JA, Luby SP: Estimating deaths and injuries due to road traffic accidents in Karachi, Pakistan, through the capture-recapture method. International journal of epidemiology 1998, 27(5):866-870.

15. Jamison DT, Mosley WH, Measham AB, Bobadilla JL (Eds.): Disease Control Priorities in Developing Countries. Washington, DC: Oxford University Press; 1993.

16. JICA, NTRC, Government of Pakistan: Pakistan Transport Plan Study in the Islamic Republic Of Pakistan. Islamabad: tripartite collaboration of the Japan International Cooperation Agency (JICA); National Transport Research Centre (NTRC), and Ministry of Communications, Government of Pakistan; 2007.

17. National Highway Authority (NHA): Traffic survey. Islamabad: NHA; 2008. 18. Government of Pakistan: Pakistan Economic Survey 2008-09. Islamabad: Economic

Advisor's Wing, Finance Division, Government of Pakistan; 2009. 19. Nishtar S, Mohamud KB, Razzak J, Ghaffar A, Ahmed A, Khan SA, Mirza YA:

Injury prevention and control: National Action Plan for NCD Prevention, Control and Health Promotion in Pakistan. Jpma 2004, 54(12 Suppl 3):S57-68.

20. Khoso A: Analysis of National Highways & Motorway Police Injury Surveillance System with respect to WHO Injury Surveillance Guidelines. Kerolinska Institutet, Department of Public Health Sciences; 2007.

21. Razzak JA, Luby SP, Laflamme L, Chotani H: Injuries among children in Karachi, Pakistan--what, where and how. Public health 2004, 118(2):114-120.

22. Wikipedia: Edhi Foundation. . San Francisco, CA: Wikimedia Foundation Inc.; 2010. [Available at URL: http://en.wikipedia.org/wiki/Edhi_Foundation] [Cited July 9, 2010].

23. Raftery KA: Emergency medicine in southern Pakistan. Annals of emergency medicine 1996, 27(1):79-83.

24. Road Injury Research and Prevention Centre: Report of road injury surveillance project of the road injury research and prevention centre of JPMC, Karachi. Karachi: Jinnah Post Graduate Medical Centre (JPMC); 2008.

25. Ahmed A: National Road Safety Plan 2007-2012. Islamabad: National Road Safety Secretariat, Ministry of Communications; 2007.

26. Stevenson M, Segui-Gomez M, Lescohier I, Di Scala C, McDonald-Smith G: An overview of the injury severity score and the new injury severity score. Inj Prev 2001, 7(1):10-13.

27. Amoros E, Martin JL, Laumon B: Estimating non-fatal road casualties in a large French county, using the capture-recapture method. Accident; analysis and prevention 2007, 39(3):483-490.

28. NHA: Traffic survey. Islamabad: National Highway Authority (NHA); 2008. 29. Lagarde E: Road traffic injury is an escalating burden in Africa and deserves

proportionate research efforts. PLoS medicine 2007, 4(6):e170. 30. Evans L: Traffic Safety. Bloomfield Hills, MI: Science Serving Society; 2004. 31. Suddle MS: Police-Executive Relationship in Pakistan In Police Reform in South

Asia: Sharing of Experiences. New Dehli: Commonwealth Human Rights Initiative (CHRI); 2007.

32. Razzak JA, Cone DC, Rehmani R: Emergency medical services and cultural determinants of an emergency in Karachi, Pakistan. Prehosp Emerg Care 2001, 5(3):312-316.

97

33. Mohan D: Road safety in less-motorized environments: future concerns. International journal of epidemiology 2002, 31(3):527-532.

34. Razzak JA, Laflamme L: Limitations of secondary data sets for road traffic injury epidemiology: a study from Karachi, Pakistan. Prehosp Emerg Care 2005, 9(3):355-360.

35. Observatoire National Interministériel de Sécurité Routière: La sécurité routière en France: Bilan de l'année 2004. Paris: La documentation Française; 2005.

36. MacKay JM, Macpherson AK, Pike I, Vincenten J, McClure R: Action indicators for injury prevention . Inj Prev, 16(3):204-207.

98

Table 1. Traffic injuries reported to police, ambulance, and emergency department on Karachi-Hala road section (2008).

Police Ambulance Emergency department

N % N % N % Road traffic crash - Fatal 19 44.1 37 14.5 47 10.4 - Not fatal 24 55.8 184 72.2 372 82.9 - Unknown 0 0.0 34 13.3 30 6.7 Road traffic injury - Deaths 27 18.8 55 10.4 60 9.1 - Transported to hospital 80 55.9 428 80.6 601 90.9 - Not transported to hospital 36 25.2 48 9.0 NA Name of patient available - Yes 96 67.1 414 78.0 648 98.0 - No 47 32.9 117 22.0 13 2.0 Age (y) - 0-15 14 9.8 34 6.4 62 9.4 - 16-45 88 61.5 292 55.0 516 78.1 - >45 4 2.8 33 6.2 78 11.8 - Unknown 37 25.9 172 32.4 5 0.7 Sex - Male 93 65.0 364 68.5 609 92.1 - Female 12 8.4 78 14.7 52 7.9 - Unknown 38 26.6 89 16.8 0 0.0 Road user group - Pedestrian 5 3.5 40 7.5 83 12.7 - Motorcycle riders 6 4.2 49 9.2 203 30.6 - Four-wheeled vehicles’ occupants 120 83.9 403 75.9 327 49.5 - Others 0 0.0 1 0.2 4 0.6 - Unknown 12 8.4 38 7.2 44 6.6

NA – Not applicable

99

Table 2. Differences in outcome of traffic injury for matched patients (N=108) identified in police, ambulance, and emergency department records for Karachi-Hala road section, 2008

Ambulance Hospital Injured Died Discharged Admitted Died N (%) N (%) N (%) N (%) N (%) Police Injured 13 46.4 4 14.3 6 42.9 1 7.1 1 7.1 Died 0 0.0 11 39.3 0 0.0 0 0.0 6 42.9 Ambulance Injured 49 53.3 26 28.3 9 9.7 Died 0 0.0 0 0.0 8 8.7

100

Table 3. Ascertainment of police, ambulance, and emergency department records for traffic

fatalities and injuries on Karachi-Hala road section (N=1 214)

Outcome Police Ambulance Hospital Total Rate† N %* N %* N %* N % Deaths 27 22.6 55 46.2 60 50.4 119 9.8 54.4 Injuries 116 10.6 476 43.5 601 54.9 1 095 91.2 500.4 * Ascertainment rate; numbers of record divided by total for the given outcome. † per 109 km travelled.

101

Figure 1. Unique records of traffic injury patients reported to police, ambulance service, and emergency departments on Karachi-Hala road section in 2008 (N=1 214)

PoliceN=143

AmbulanceN=531

HospitalN=661

114

1315 1

424 56879

102

Appendix 5: Article published – Study III

103

104

105

106

107

108

109

110

Appendix 6: Article under review – Study IV Note: Revisions are underlined. TITLE Burden and factors associated with highway work-zone crashes, Karachi-Hala road section, Pakistan AUTHORS 1. Junaid Ahmad BHATTI a, 2. Junaid Abdul RAZZAKb 3. Emmanuel LAGARDE a 4. L.-Rachid SALMI a,c,d AFFILIATIONS a. Equipe Avenir « Prévention et Prise en Charge des Traumatismes », Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897), Bordeaux, France. b. Department of Emergency Medicine, The Aga Khan University, Karachi, Pakistan. c. Institut de Santé Publique, d’Epidémiologie et de Développement (ISPED), Université Victor Segalen Bordeaux 2, Bordeaux, France. d. Service d’information médicale, Centre Hospitalier Universitaire de Bordeaux, Bordeaux, France. CORRESPONDING AUTHOR Junaid A. BHATTI Equipe Avenir « Prévention et Prise en Charge des Traumatismes » Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897) 146 rue Léo Saignât 33076 Bordeaux cedex France Tel: +(33) 5 57 57 45 50 Fax: +(33) 5 57 57 45 28 Email: [email protected] COUNTS Abstract: 178 words Manuscript: 2 317 words References: 28 Tables: 3 Figure: 1

111

ABSTRACT Objective: To assess the burden and factors associated with highway work-zone (HWZ) crashes. Design: Historical cohort. Setting: Karachi-Hala road-section, Pakistan (196 km). Data: Police reported crashes and traffic statistics from Jan 06 to Dec 08. Analysis: Crash and fatality risk between HWZ and other zones for a 50-km-long section were compared. Crash locations were described for a further 146-km-long section on which factors associated with HWZ crashes were assessed. Results: HWZs accounted for 15.0% of traffic crashes (N=180) and 30.8% of road fatalities (N=91) on the 196-km-long section. Rates were higher in HWZ compared to other zones for crash (rate ratio (RR) = 2.35, 95% confidence interval (95%CI) = 1.17-4.70) and fatality (RR = 4.70, 95%CI = 2.11-10.46). Opposite-direction (adjusted odds ratio (aOR)=10.65, 95% CI=3.22-35.25) and traffic crashes involving pedestrians (aOR=6.03, 95%CI=1.39-26.20) and on wet surfaces (aOR=7.26, 95%CI=4.15-48.89) were significantly associated with HWZs. Conclusion: These results orient toward prevention measures such as strict traffic enforcement, traffic separation, improving pedestrians’ conspicuity, and hazard signage at HWZs in Pakistan. Feasibility and effectiveness of these measures remains to be evaluated. Keywords: Developing country; road/traffic accidents; severity; trauma. Abbreviations: HWZ, highway work zones; km, kilometre; LMIC, low and middle-income countries; NHA, National Highway Authority; NHMP, National Highway & Motorway Police; PKR, Pakistani rupee; USA, United States of America.

112

1. INTRODUCTION With the aging of highways, road authorities spend a considerable proportion of their budget on maintenance.1-3 For instance, federal and state highway departments of transportation in the United States of America (USA) invest 10-15% of their annual budget on the maintenance of highways, amounting to tens of billions of US$ each year.4 5 Similarly, Low- and Middle-Income Countries (LMICs) spend a considerable portion of their development assistance on highway maintenance projects.6 7 Construction zones, often named Highway Work Zones (HWZ), are present on all road networks anywhere in the world.1-7 HWZs are different from work zones in urban areas, because provision of detours is often impractical.1 2 Thus regular traffic flows need to be restricted, leading to safety challenges.8 Previous research in developed countries has demonstrated an increased crash and fatality risk in HWZ as compared to other parts of the transport network.9 For instance, in two-lane highways of Kansas, 63% of fatal crashes and one-third of injury crashes took place on HWZs.10 The estimated cost of HWZ crashes between 1995 and 1997 in the USA was 6.2 billion US$ with an average cost of 3,687 US$ per crash.11 Nevertheless, the safety problem related to HWZ has received some attention in high-income countries (HIC) and appropriate traffic control interventions are implemented during the construction periods.2 Pakistan has a strategic national highway network of over 8 000 km. Over 90% of the inland traffic passes through these road sections and, consequently, these highways undergo extensive wear and tear due to overloading, heavy traffic, and delayed maintenance.12-14 Previous research has shown that 27% of road fatalities occurred on these roads although that they accounted for only 4% of the network.14 A survey conducted in 2000 showed that 50% of the national highway network was in need of major pavement reconstruction.12 15 The maintenance demand had consistently increased from 10 billion Pakistani rupees (PKR) in 1991 to over 30 billion PKR in 2005, yet only around 10 billion PKR were available in 2005 for highway maintenance.12 To date, no study has ever estimated the road crash burden due to such traffic conditions in Pakistan.13 16 The objective of this study was to assess the burden and factors associated with HWZ crashes on an interurban highway in Pakistan. 2. METHODS 2.1 Study design and setting As we compared the incidence density rates, estimated from events (crash, fatality, or injury) and person-time exposure (km travelled) measures, between the HWZ and normal traffic zones, the study design was similar to an historical cohort study.17 18 The study setting was a 196-km-long four-lane, separated, non-access controlled Karachi-Hala road section in the province of Sindh, Pakistan. Traffic counts ranged from 16,356 vehicles per day on Hyderabad–Hala sub section (50 km) to 24,707 vehicles per day on Karachi-Hyderabad sub section (146 km).19 The National Highway Authority (NHA) manages the overall maintenance and upgrade of this road section, mostly by private contractors.12 The National Highway and Motorway Police (NHMP) have been enforcing traffic rules on this road section since 2004. 2.2 Traffic data Annual average daily traffic survey data were collected from NHA headquarters. These surveys are conducted each year to assess traffic counts on different road sections under the Federal administration.19 Locations near toll plazas are selected to assess 24-hour counts by the NHA personals. We extracted information on traffic counts observed between Karachi-Hyderabad (146 km) and Hyderabad-Hala (50 km) road sections. Variables coded from traffic

113

surveys included number, type (trucks, buses transporting ≥ 20 passengers, mini-truck, minibus or coasters transporting < 20 passengers, cars or jeeps, and motorcycles), and direction of vehicle (North-bound or South-bound).19 In Pakistan, during maintenance works on separated highways, two or more lanes in a given direction are completely blocked and traffic is diverted, most of the times, to the opposite directed lanes (figure 1A & 1B). The police and highway authorities facilitate traffic during the construction period. Detail of HWZ commencement and completion dates and km locations of maintenance works are recorded in their memos. We collected this data from NHA and NHMP regional offices, but these records were available only for the 50-km-long Hyderabad-Hala sub section. 2.3 Crash data After a crash, NHMP patrolling officer files the details of crash on a standard four-page accident analysis report.20 A copy of this report is kept in the regional office, whereas the original is sent to the NHA headquarters. Moreover, the crash is recorded on a separate accident register in each regional office.20 All police crash reports and registers for the period from Jan 06 to Dec 08 were retrieved and photocopied from regional NHMP offices with the permission of the officer in charge. Variables coded from accident registers included date, time, number and type of involved vehicles, number of persons injured or who died in a reported crash, and whether the crash occurred during maintenance works. Variables coded from crash reports included date, time, location, direction of lane (North-bound or South-bound), light, weather, horizontal and vertical road profile, road surface and shoulder condition, ongoing maintenance, and cause and type of crash.21 Type of crash was defined as single vehicle, same direction, opposite direction, sidewise, pedestrian. When more than one type was identified, crashes were coded as crash of the most vulnerable involved road user; the vulnerability decreasing order was: pedestrian; opposite directions; sidewise or at intersection; single vehicle; same direction.21 Information on number, injury severity, and type of road user involved (pedestrian, riders of two-wheelers, or occupants of cars/jeeps, minibuses, buses, or trucks) were coded separately. Severity was defined as ‘severe’ when the involved person was transported to the hospital and ‘fatal’ when the involved road user died at the crash scene or at hospital within the first 24 hours following the event.20 2.4 Analysis Information on crashes from registers and reports were linked to make a single file based on crash location (km) and crash date, available for all crashes. Crashes, fatality, and severe injury per 109 vehicle-km travelled for vehicle type and direction were computed using traffic counts survey. Due to limited data on traffic exposition of work zones, these rates for work and normal traffic zone were computed using information on work zone dates and average daily traffic for the 50-km-long sub-section. Crash, fatality, and severe injury risks according to road directions, vehicle types and traffic conditions were compared using rate ratios with 95% confidence intervals, rate differences, and attributable risk proportions where appropriate.17 22 Associations of factors with HWZs crashes were estimated from a multiple logistic regression model, including all variables weakly associated (P<0.2) with HWZs with backward selection strategy.23

3. RESULTS 3.1 Overall crash and injury burden

114

A total of 180 crashes were identified from the police registers. Overall 612 road users were injured in these crashes; 14.8% (N=91) died, and 55.3% (N=339) were severely injured. The road fatality rate on this highway, excluding HWZ crashes, was 13.0 per 109 vehicle-km (table 1). The crash rate was significantly higher in North-bound as compared to South-bound direction (rate ratio (RR) = 1.81, 95% confidence interval (95%CI) = 1.30, 2.52). Compared to trucks, the crash rate was lower for passenger cars (RR = 0.57, 95% CI = 0.42, 0.78), but the fatality rate was twice as high for passenger cars compared to trucks (RR = 1.93, 95%CI = 1.04, 3.61). Similarly, fatality rate was significantly higher for occupants of buses (RR = 3.32; 95%CI = 1.52, 7.22) and minivans (RR = 4.75; 95%CI = 1.84, 12.24), as compared to trucks’ occupants. 3.2 Work zone related crash and injury burden Fifteen percent (N=27) of the traffic crashes occurred in HWZs, accounting for 30.8% (N=28) of all fatalities and 15.3% (N=52) of those severely injured on the 196-km-long road section. During the three-year period, 0.89 billion vehicle-km travelled on the 50-km-long sub-section for which HWZ dates were available. HWZ accounted for 17.6% of the vehicle km travelled on this sub section. On average, HWZs were 5.7-km-long (SD = 4.3). Two work zones were 10 and 14-km-long and lasted more than 300 days. The crash (32.5 vs. 31.6 per 109 km travelled) and the fatality (16.3 vs. 13.0 per 109 km travelled) rates observed on normal traffic zone of this sub-section were similar to that for the whole road section whereas severe injury rate (89.4 vs. 59.3 per 109 km travelled) was higher as compared to the whole road section. Significantly higher crash (RR = 2.35), fatality (RR = 4.70), and severe injury risks (RR = 1.92) were observed on HWZs compared to other zones (P≤0.004) on this sub section (table 2). 3.3 Factors associated with HWZ crashes Complete reports were available for 93.3% (N=168) of all traffic crashes, 96.8% of fatal crashes (N=63), and 98.9% of severe injury crashes (N=88). Crashes between vehicles moving in opposite direction and those involving pedestrians were more likely on HWZ than on other sections (table 3). Similar associations were observed for both sub-sections separately, where pedestrian and opposite direction crashes accounted for most (≥73.2%, P≤0.01) HWZ crashes. Similarly, wet surface crashes were significantly more likely to occur on HWZ than on non HWZ. However, this association was not observed when the two sub-sections were analyzed separately. Hazardous overtaking was the major cause of crash in HWZs (55.6%) whereas sudden entry on the road was identified as crash cause in all of the pedestrian involving HWZ crashes (N=5, 18.5%). 4. DISCUSSION These results showed that HWZs lead to increased road crash and fatality risk on this national highway in Pakistan. Overall, HWZs accounted for one third of fatalities and the crash fatality risk was four times higher in HWZs as compared to normal traffic. In HWZ, one out of two crashes occurred between opposite-direction vehicles; the likely explanation of was the high volume un-separated traffic conditions and hazardous overtaking. The highway traffic is expected to triple from 2005 to 2025 in Pakistan.12 The Government has envisaged meeting these demands by upgrading and improving the current highway network, exposure to HWZs is thus expected to increase.12 Although guidelines for work zone management exist in Pakistan, so far no mechanism for HWZ design, performance, and enforcement evaluation has been defined or implemented.15 These results suggested the need to improve institutional capacity as well as inspection mechanisms so that road agencies

115

should be accountable for ensuring HWZ safety.2 12 15 Further, reducing HWZ duration could be useful in decreasing the resulting crash risk.15 Almost half of the traffic on the highway was composed of heavy trucks which have an overall low speed.12 As the space to accommodate the traffic volume is reduced in HWZ, it is likely that in absence of harsher penalties and barriers, smaller and faster vehicles performed hazardous overtaking. Our results consistently showed that most of the crashes occurred as a result of traffic conflict between the oppositely moving traffic in the work zones (figure 1B).21 This points out the need to carefully plan and regulate the traffic flow during maintenance works. Enforcing harsher penalties for overtaking, providing alternate lanes, and traffic separation might be some of the useful measures to decrease hazardous situations leading to HWZs crashes in Pakistan.15 This study showed that pedestrians, probably including workers, were significantly involved in HWZ crashes. Sudden entry onto the highway was reported as the major cause of such crashes. Similarly, wet surface increased the risk of HWZ crashes. Such involvements, although less important, were found in HWZ crashes elsewhere.10 Human judgement error is indeed one of the principal factor identified in HWZ crashes.1 10 These results indicated that prevention measures such as advance warning area, clear zones to enhance visibility, road markings, hazard signage in the work area, and conspicuity equipment for workers could be useful to reduce such crashes in Pakistan.3 15 Finally, the overall traffic fatality risk on this highway is several times higher as compared to a limited access road in the HICs.24 Factors such as drowsiness, speeding, hazardous overtaking, and poor vehicle condition were highly involved in these crashes.25 26 These results were not surprising as the traffic conditions are quite different from HICs which had more crash prevention and control measures on their roads.21 Interestingly, the fatality risk for the occupants of cars, minibuses, and buses were two times or more high than for those travelling in trucks. The higher fatality risk associated with car and bus occupants as compared to truck occupants could be due to the high number of passengers and the non use of seat belts by both drivers and passengers, substandard vehicles, and higher traffic speeds.26 This study demonstrates the need to investigate and control high vulnerability of car occupants in LMICs. This study may have some limitations. Firstly, we included only police reported crashes, which were shown to report only 56% of road fatalities and 4% of severe injuries in Pakistan.27 Thus given results could under estimate the crash risk which we had assumed to be same for both the HWZ and normal traffic zones. Further, traffic measuring was based on 24-hour surveys. This could lead to both underestimation and overestimation of the crash risk.17 Nevertheless, these had been used previously to compare the crash risk for vehicles and road types in LMICs.21 Finally, little information was available on involved drivers to account for those factors in adjusted analyses.1 Pakistan, like many other LMICs, is passing through economic transition.12 With the expected increase in HWZ activity in the future, several lessons could be learnt from this study. Firstly, a high road injury burden in HWZ indicated that a monitoring system is needed to examine the HWZ safety measures by the agencies involved in maintenance works.15 Secondly, more efforts are required to reduce the duration of HWZs.2 Finally, these results orient toward prevention measures such as harsher punishment for traffic violations such as overtaking, traffic separation, advanced warning area, hazard signage, and improving pedestrians’

116

conspicuity at HWZs in Pakistan.2 8 15 Feasibility and effectiveness of their implementation, however remains to be evaluated. ACKNOWLEDGEMENTS We are especially thankful to Dr. Aftab Ahmed PATHAN, Deputy Inspector General of Police, Mr. Irshad SODHAR, Senior Patrolling Officer, and Mr. Naeemullah SHIEKH, Senior Patrolling Officer, National Highway and Motorway Police south sector III office, Pakistan for their support in data collection. Authors also acknowledge Engr. Ali Bin Usman SHAH, Road Safety Expert at National Highway Authority for providing traffic survey reports. Finally, we would like to thank the editor and reviewers for their suggestions to improve the content of this manuscript. FUNDING First author is the PhD candidate at Université Victor Segalen Bordeaux 2. This position is funded by Higher Education Commission of Pakistan. Institut National de la Santé et de la Recherche Médicale Unité 897, France, funded the logistics for data collection. Funding bodies had no input in study design, analysis and interpretation of results. AUTHORS CONTRIBUTION This study is the part of PhD thesis work of JB supervised by LRS. JAR and EL provided technical help in study conception, design, analysis, and manuscript writing. COMPETING INTERESTS No competing interests were identified for any of the authors. LICENSE STATEMENT: The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd and its licencees, to permit this article (if accepted) to be published in IP and any other BMJ Group products and to exploit all subsidiary rights, as set out in our licence (http://ip.bmjjournals.com//ifora/licence.pdf). Important points What is already known on the subject?

• Highway work zones (HWZs) lead to increased crash and fatality risk. • Risks and factors associated with such zones were rarely studied in developing countries.

What this study adds: • Crash fatality risk was four times as high on HWZs as compared to other zones in Pakistan, a

low-income country. • Traffic separation, harsher penalties for hazardous overtaking, and appropriate hazard signage

at HWZs might reduce the risk of such crashes. Policy implications:

• Exposition to HWZ will tremendously increase in coming years in developing countries like Pakistan. Implementation of safety interventions at HWZs may significantly reduce road disease burden.

REFERENCES 1. Li Y, Bai Y. Highway work zone factors and their impact on crash severity. J Transp Eng

2009;135:694-701.

117

2. American Association of State Highway and Transportation Officials (AASHTO) . A policy on geometric design of highways and streets. Washington, DC: AASHTO, 2004.

3. Li Y , Bai Y. Development of crash-severity-index models for the measurement of work zone risk levels. Accid Anal Prev 2008;40:1724-31.

4. Department of Transportation (DOT). FY 2008 Budget in brief. Washington, DC: DOT, 2008.

5. Department of Transportation (DOT). Highway Statistics 2007: Funding For Highways and Disposition of Highway-User Revenues, All Units of Government, 2007. Washington, DC: DOT, 2007.

6. Republic of Ghana, European Community. Country Strategy Paper and Indicative Programme for the period 2002–2007. Accra: European Union delegation in Ghana, 2002.

7. Masood H, Seetharam KC, Utami D, et al. Report and recommendation of the President to the board of directors on proposed loans to the Islamic Republic of Pakistan for the North-West Frontier Province road development sector and subregional connectivity project. Manilla: Asian Development Bank, 2004.

8. American Association of State Highway and Transportation Officials (AASHTO) . Summary report on work zone crashes. Standing committee on highway traffic safety. Washington, DC: AASHTO, 1987.

9. Wang J, Hughes WE, Council FM, at al. Investigation of highway work zone crashes: what we know and what we do not know. J Transp Res Rec 1996;(1529):54-64.

10. Li Y , Bai Y. Comparison of characterstics between fatal and injury accidents in highway work zones. Safety Sci 2008;46:646-60.

11. Mohan SB, Gautam P. Cost of highway work zone injuries. Practical Periodical on Structural Design and Construction 2002;7:68-73.

12. JICA, NTRC, Government of Pakistan. Pakistan Transport Plan Study in the Islamic Republic of Pakistan. Islamabad: tripartite collaboration of the Japan International Cooperation Agency (JICA); National Transport Research Centre (NTRC), and Ministry of Communications, Government of Pakistan, 2007.

13. Nishtar S, Mohamud KB, Razzak J, et al. Injury prevention and control: National Action Plan for NCD Prevention, Control and Health Promotion in Pakistan. J Pak Med Assoc 2004;54:S57-68.

14. National Transport Research Centre. Traffic Counter measures in Pakistan: NTRC publication 85. Islamabad: National Transport Research Centre, Ministry of Communication, 1985.

15. Ahmed A. National Road Safety Plan 2007-2012. Islamabad: National Road Safety Secretariat, Ministry of Communications, 2007.

16. Hyder AA , Ghaffar AA, Sugerman DE, et al. Health and road transport in Pakistan. Public Health 2006;120:132-41.

17. Salmi LR, Battista RN. Epidemiologic assessment of hazardous roadway locations. Epidemiology 1990;1:311-4.

18. Lassarre S. Some Statistical Models for Road Risk Analysis. In: Tiwari G, Mohan D, Muhlrad N, editors. The Way Forward: Transportation Planning and Road Safety. New Delhi: MacMillan India Ltd., 2005.

19. National Highway Authority (NHA). Traffic survey. Islamabad: NHA, 2008. 20. Khoso A. Analysis of National Highways & Motorway Police Injury Surveillance System

with respect to WHO Injury Surveillance Guidelines. Stockholm; Kerolinska Institutet, 2007.

118

21. Sobngwi-Tambekou J, Bhatti J, Kounga G, et al. Road traffic crashes on the Yaoundé–Douala road section, Cameroon. Accid Anal Prev 2010;42:422-6.

22. Rothman KJ, Greenland S, editors. Modern Epidemiology. Second ed. Philadelphia, PA: Lippincott Williams & Wilkins, 1998.

23. Hosmer DW, Lemeshow S. Applied Logistic Regression. Second ed. Danvers, MA: John Wiley & Sons, Inc., 2000.

24. Observatoire National Interministériel de Sécurité Routière. La sécurité routière en France: Bilan de l'année 2004. Paris: La documentation Française, 2005.

25. Ross A, Baguely C, Hills B, et al. Towards safer roads in developing countries. Crowthorne: Transport Research Laboratory, 1994.

26. Tiwari G , Mohan D, Muhlrad N, editors. The Way Forward: Transportation Planning and Road Safety. Delhi: McMillan India Ltd, 2005.

27. Peden M, Scurfiled R, Sleet D, et al (editors). World report on road traffic injury prevention. Geneva: World Health Organization, 2004.

119

Table 1. Road crash fatality and injury risk per 109 vehicle-km on the Karachi-Hala


Vehicle-km Crash Fatality Severe injury 109 N Rate* N Rate* N Rate* All (except work zone) 4.84 153 31.61 63 13.02 287 59.30 -North-bound direction 2.40 98 40.83 38 15.83 191 79.58 -South-bound direction 2.44 55 22.54 25 10.25 96 39.34 Vehicle† -Motorcycle 0.19 10 52.63 8 42.11 10 52.63 -Car/jeep 1.90 60 31.58 29 15.26 107 56.32 -Mini-van (<20 passengers) 0.16 11 68.75 6 37.50 33 206.25 -Mini-truck 0.27 16 59.26 7 25.93 21 77.78 -Buses 0.42 39 92.86 11 26.19 105 250.00 -Trucks 1.90 105 55.26 15 7.89 50 26.32 * Per 109 vehicle-km travelled † At least one of the involved vehicles in the crash was from the category below.

120

Table 2. Highway work zone crash fatality and injury risk per 109 vehicle-km on Karachi-Hala road sub-section, Pakistan (2006-08)

Vehicle-km Crash Fatality Severe injury 109 N Rate† N Rate† N Rate† Work zone 0.157 12 76.43 12 76.43 27 171.97 Normal traffic 0.738 24 32.52 12 16.26 66 89.43 Rate ratio (RR) 2.35 4.70 1.92 95% Confidence Interval for RR 1.17; 4.70 2.11; 10.46 1.23; 3.01 Rate difference 43.91 60.17 82.54 Attributable proportion (%) 57.45 78.73 48.00

† Per 109 vehicle-km

121

Table 3. Factors associated with work-zone crashes on 196-km-long Karachi-Hala


Work-zone crashes Other crashes P Adjusted 95%CI† N=27 N=141 Odds ratio N (%) N (%) Severity 0.003 - Mild or no injury 0 0.0 20 14.2 - Severe injury 10 37.0 77 54.6 - Fatal 17 63.0 44 31.2

Crash type

- Same direction 4 14.8 67 47.5 1 - Opposite/sidewise 17 63.0 25 17.7 <0.001 10.65 [3.22; 35.25]

- Pedestrian 5 18.5 13 9.2 6.03 [1.39; 26.20] - Single vehicle 1 3.7 36 25.5 0.25 [0.02; 2.94] Light 0.29 - Daylight 11 40.7 73 51.8

- Night 16 59.3 68 48.2 Surroundings 0.41

- Built-up 13 48.2 80 56.7

- Rural 14 51.8 61 43.3

Horizontal profile 0.19 - Straight 21 77.8 123 87.2 - Curve 6 22.2 18 12.8

Vertical profile 0.09 - Plain 26 96.3 119 84.4

- Slope 1 3.7 22 15.6 Road surface 0.47 - Regular 24 88.9 131 92.9 - Irregular 3 11.1 10 7.1

Shoulder surface 0.72 - Regular 25 92.6 133 94.3 - Irregular/absent 2 7.4 8 5.7 Intersection 0.60

- Yes 1 3.7 9 6.4

- No 26 96.3 132 93.6

Surface condition* 0.08 - Dry 23 85.2 134 95.0 1 - Wet 4 14.8 7 5.0 7.26 [4.15; 48.89]

Cause** ‡

- Overtaking 15 55.6 10 7.1 - Sudden entry 5 18.5 7 5.0 - Bad weather 4 14.8 9 6.4

- Drowsiness 3 11.1 29 20.6 - Speeding 1 3.7 17 12.1

- Tyre-burst 0 0.0 20 14.2

† 95% confidence interval ‡ Not included in multivariate analysis. * Variable was kept in multivariate model based on the significant difference between log-likelihood of reduced and the full model (P<0.05). ** Only important causes are cited here; a crash may be designated with two causes.

122

123

Appendix 7: Manuscript in preparation – Study V TITLE HAZARD PERCEPTION AT HIGH- AND LOW-RISK ROAD SITES AUTHORS 1/ Junaid Ahmad BHATTI, MSca,b,* 2/ Junaid Abdul RAZZAK, MD, PhDc 3/ Emmanuel LAGARDE, PhDa,b 4/ Jöelle SOBNGWI-TAMBEKOU, MDa 5/ Ahmadou ALIOUM, PhDa 6/ L.-Rachid SALMI, MD, PhDa,b,d,* AFFILIATIONS a. Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897), Bordeaux, France b. Institut de Santé Publique, d’Epidémiologie de Développement (ISPED), Université Victor Segalen Bordeaux 2, Bordeaux, France. c. Department of Emergency Medicine, The Aga Khan University, Karachi, Pakistan. d. Service d’information médicale, Centre Hospitalier Universitaire de Bordeaux, Bordeaux, France. * Contributed equally CORRESPONDING AUTHOR Junaid A. BHATTI Equipe Avenir « Prévention et Prise en Charge des Traumatismes » Institut National de la Santé et de la Recherche Médicale Unité 897 (INSERM U897) 146 rue Léo Saignat 33076 Bordeaux cedex France Tel: +(33) 5 57 57 45 50 Fax: +(33) 5 57 57 45 28 Email: [email protected] COUNTS Abstract: 244 Manuscript: 3 165 References: 44 Tables: 4 Figures: 0 RUNNING TITLE Hazard perception of crash sites

124

ABSTRACT Objectives: Interurban roads contribute significantly to traffic fatalities in developing countries. In this study, hazard perception was compared for sites frequently involved in Road Traffic Crashes (RTCs) to those not involved in RTCs on the same road sections. Design: Study settings were Karachi-Hala road in Pakistan and Yaoundé-Douala road section in Cameroon. Videos of 26 high-risk sites (involved in ≥ 3 crashes in 3 years) and 26 low-risk sites (no crash reported), matched for the road section, were shown to 100 voluntary Pakistani drivers. Main outcome measures: Variations in perceived site hazardousness (Likert scale) and preferred speed for each site pair were assessed. Factors associated with hazard perception level of high-risk sites were assessed using logistic regression analyses. Results: The drivers reported a higher hazard perception and a lower preferred speed for high-risk sites than for their matched low-risk sites in only half of pairs (N=12, P≤0.02). High-risk sites situated in built-up areas (adjusted odds ratio [OR] =0.58, 95% confidence interval [95%CI] =0.51-0.68) and with lane width ≤ 8 m (OR=0.51, 95%CI=0.43-0.61) were perceived less hazardous than low-risk sites with same road situation. Further, high-risk sites with vertical road signs (OR=2.75, 95%CI=2.38-3.16) and U-turns (OR=8.00, 95%CI=6.36-10.22) were perceived more hazardous than low-risk sites with same situation. Conclusion: These methods identified factors influencing the hazard perception of high-risk sites on two road sections in developing countries. They might be useful in prioritizing high-risk sites for improvement as well as implementing low-cost interventions in such settings. Keywords: Black spots; developing countries; epidemiology; risk factors; road traffic accidents.

125

INTRODUCTION Road traffic safety is an important health problem worldwide, resulting in an

estimated 1.2 million deaths and another 50 million injuries each year (World Health Organization, 2009). Over 85% of these Road Traffic Crashes (RTCs) related deaths occur in Low- and Middle-Income Countries (LMICs) (Peden et al., 2004). Highways, which represent 5 to 10% of national road networks, lead to significant injury burden in these countries. For instance in Pakistan, 27% of all police-reported fatal RTCs occur on National highway-5 (N5) (National Transport Research Centre, 1985). Similarly, on Yaoundé-Douala road-section in Cameroon, approximately 73 deaths occurred per 100 million kilometer travelled, a rate 35 times higher than on similar type of roads in Europe (Sobngwi-Tambekou, Bhatti, Kounga, Salmi, & Lagarde, 2010).

RTCs are unevenly distributed along the network (Peden et al., 2004). They occur in

clusters at single sites, often called high-risk sites, along particular sections of the road (Geurts & Wets, 2003). They can be defined as sites having a higher expected number of crashes than other similar sites (Elvik, 2008). Theoretically, inadaptability of driving behavior to the local road and traffic hazards leads to crashes at these sites (Geurts & Wets, 2003). It has been documented that design improvement at these sites can result in significant decreases in crash risk (Ross, Bagunley, Hills, McDonald, & Silcock, 1991). However, this remains an expensive option, and not all high-risk sites can be improved in a timely manner (Geurts, Wets, Brijs, Vanhoof, & Karlis, 2006).

Hazard perception is the ability to identify potential hazardous situations and taking

necessary actions to avoid them (Benda & Hoyos, 1983; SWOV, 2008). Driver-related factors such as age, sex, familiarity with road, driving experience, attitudes, and self-assessment of skills could influence this ability (DeJoy, 1989; Finn & Bragg, 1986; Harre, 2000; Mayhew, Simpson, & Pak, 2003; McKenna, Stanier, & Lewis, 1991; Trankle, Gelau, & Metker, 1990). Road elements such as sharp bends, decreased widths, and presence of lane markings could increase hazard perception (Goldenbeld & van Schagen, 2007; Kanellaidis, Zervas, & Karagioules, 2000). Previous research has shown that augmenting hazard perception by driver training or by implementing appropriate road furniture could significantly reduce the likelihood of RTCs (Deery, 1999; Rundmo & Iversen, 2004).

Much work on high-risk crash sites stressed that identifying these sites using statistical

methods so that safety work could be prioritized (Montella, 2010). Interactions between driver- and site-related factors had not been investigated in detail to prioritize such sites, particularly on interurban roads in LMICs (Sabey & Taylor, 1980). To our knowledge, the hypothesis that high-risk crash sites might not be perceived as dangerous by some drivers has not been tested (Geurts & Wets, 2003). Insight into how high-risk crash sites are perceived by drivers could be useful in developing and implementing less expensive interventions, particularly in LMICs (Harre, 2000). The objective of this study was to compare hazard perceptions for sites involved in RTCs to those not involved in RTCs in voluntary drivers. Further, we assessed driver- and road- related factors associated with hazard perception level.

METHODS

Study design and settings The study settings were interurban road sections situated in Cameroon and Pakistan:

1/ Karachi-Hala road section in Pakistan (196-km-long mostly four lane separated road), and 2/ Yaoundé-Douala road section in Cameroon (243-km-long mostly two-lane non-separated

126

road). A matched strategy was used to select sites. ‘High-risk sites’ were those involved in three or more RTCs in a precedent three-year period, whereas ‘low-risk sites’ were those not involved in a RTC, during the same period. For each high-risk site, a low-risk site was randomly selected on the same road section. Hazard perception was assessed by showing videos of these sites to voluntary Pakistani drivers. Ethical approval of the study design was obtained from the Aga Khan University (AKU) Ethics Research Committee in May 2009 (Reference ERC/2009/1179).

Site selection

In Pakistan, National Highway & Motorway Police (NHMP) regional office was visited in June 2009. Crash reports and registers for the three-year period from Jan 1, 2006 to Dec 12, 2008 were retrieved and photocopied. High-risk sites with given kilometer location were then identified with Global Positioning System (GPS) coordinates with help of a police officer. Similarly, traffic police offices in Cameroon were visited and such sites were subsequently identified in June 2007 (Bhatti, Sobngwi-Tambekou, Lagarde, & Salmi). The two road sections were filmed from a four-wheeled sedan car moving within the authorized speed limit (July 2009 in Pakistan and July 2007 in Cameroon). All high- and low-risk sites were then identified by linking GPS coordinates to the videos. For each high-risk site, a low-risk site was randomly selected out of all sites on the same road section which were not involved in crashes.

Video sets

To measure hazard perception, video of sites were cut so that each video showed a 500-meter-long road section during 30 seconds, including the last 100 m corresponding to the high- or low-risk site. Further, a yellow indicator blinked five times to help drivers identify the site for which they had to emit a judgment on hazard perception during video projection. We determined sample size to be 26 pairs of sites, assuming that 95% of the high-risk sites would be identified as dangerous and 80% of the low-risk sites as not dangerous with a precision of 7.5 (Flahault, Cadilhac, & Thomas, 2005).

Participant selection

Participants were Pakistani nationals residing in Karachi, aged 18 years or more, with a valid driving permit, who had driven a motorized vehicle on the Karachi-Hala road section in the previous seven days. Random sampling was not possible because of heavy-traffic and higher speed conditions on this road section (Hijar, Carrillo, Flores, Anaya, & Lopez, 2000). Thus, a convenience, but representative, sampling method was used to recruit 100 drivers. For this, we determined the drivers’ sex and vehicular distribution by observing traffic from a pilot study (N=5 496). It was observed that cars accounted for 39.1%, heavy trucks for 36.5%, minibuses and mini-trucks for 7.8%, buses for 9.6%, and motorcycles for 6.3% of the vehicles entering Karachi. Distribution of cars and heavy vehicles was similar to that recorded by highway authority (NHA, 2008). Almost all drivers were men (99.9%). Based on these findings, personal vehicle male drivers were invited from a roadside gas station at start of the highway near Karachi, and commercial vehicle drivers were invited from transport company offices at six different locations in Karachi.

Data collection

Face-to-face interviews with drivers were conducted in Urdu language. These were developed from an English language questionnaire using back translation, independent linguistic verification, and testing on five drivers. Interviews were either conducted at the Aga Khan University (AKU) Campus or at the company offices in separate rooms. Driver-related

127

variables included socio-demographic variables (age, sex, marital status, education, and employment), whether driving permit was issued without practical test, frequency of reported risky driving behaviors (sleepy driving, cell phone use while driving, seat-belt use, traffic tickets, driving while intoxicated during previous three months), and involvement in RTC during previous year.

Using 17-inch video screens, five test videos (three from Pakistan and two from

Cameroon) were shown to drivers before presenting selected sites. The order of sites was randomly drawn for each participant. To avoid confusion from right- and left-hand driving practiced in Cameroon and Pakistan, site videos from Cameroon followed those from Pakistan. For each video shown, drivers were asked to report their perception of site and traffic, on a four-level scale; 1/ Certainly safe, 2/ Probably safe, 3/ Probably dangerous, 4/ Certainly dangerous. Further, they were asked to record their preferred speed (in km/h) for each site.

Each site was characterized by the main investigator, using definitions used in our

previous study conducted in Cameroon (Bhatti et al.). Site-related variables assessed were built-up or rural area, horizontal and vertical road profile, road width, surface regularity, verge slope, depth at 10 m from the verge, location and type of nearby obstacles (within a road distance of 50 m in each direction), horizontal marking, vertical road signs, and presence of an intersection or a U-turn. Traffic-related variables assessed were traffic moving in same or opposite direction, visible pedestrian, motorcyclist, or heavy vehicle, rain or wet surface, maneuvering vehicle (crossing or overtaking), and number of lanes (Sümer, Ünal, Birdal, Çinar, & Çevikoglu, 2007).

Analysis

Proportions of site- and driver-related characteristics were computed. Discordance (D) of appreciation for a matched high- and low-risk site pair was defined as “minor” when difference of hazard perception level was one on the Likert scale and “major” when the level difference was more than one. Positive sign (D+) was used to show that hazard perception level was higher for the high-risk site than its matched low-risk site, and negative sign (D-) to show that hazard perception level was lower for high-risk site than its matched low-risk site. Wilcoxon test was used to assess whether these discordances were significantly higher or lower for high-risk site than low-risk site. Similarly, differences in reported speeds for matched high- and low-risk site pairs were compared using a paired t test. Correlations between reported speeds for high- and low-risk site pairs were assessed by intra-class correlation coefficient (ICC).

Associations of driver-, site-, and traffic-factors with road hazard perception level were assessed using logistic regression (model 1) with a backward selection strategy including significant (P<0.05) variables and interactions with risk status (high- or low-risk site). For these analyses, site hazard perception levels were regrouped as ‘safe’ when driver reported his perception to be certainly or probably safe and ‘dangerous’ when he reported his perception to be probably or certainly dangerous. To assess whether these associations remained significant while adjusting for site- and driver-related factors, two other models were constructed including site (model 2) and participant identification (model 3) as random effects. The association of multiple category variables (age and vehicle driven) with hazard perception was assessed using log-likelihood test for model 2 and 3 (Hosmer & Lemeshow, 2000).

RESULTS

128

Sites

Out of 131 crash sites identified in Pakistan, 16 were involved in three or more crashes. Similarly, out of 474 crash sites identified in Cameroon, 18 were involved in three or more crashes. We randomly selected 10 high-risk sites. In Pakistan, most high-risk sites were with straight road profile (87%), whereas this proportion was 20% in Cameroon (Table 1). Road surface conditions were irregular on most high-risk sites in Pakistan (75% vs. 65%) and Cameroon (90% vs. 40%) than low-risk sites. Half of high-risk sites in Pakistan (62%) and Cameroon (50%) were with flat road profile. A vertical road sign was visible at 38% of high-risk sites in Pakistan and at 40% in Cameroon. Fewer high-risk sites were located in built-up area in Pakistan (31%) than in Cameroon (50%). Similarly, one third of high-risk sites were at intersection in Pakistan (31%) and Cameroon (40%). In Pakistan, 19% of the high-risk sites were situated at a U-turn and on 13%, maintenance works was ongoing.

Participants

Out of 100 participants, 44 were interviewed at the AKU. Most participants were aged between 26-45 years and one fifth had received no education (Table 2). While all drivers lived in Karachi, 46 of them had a residence in other regions of Pakistan as well. Seventy four drivers reported either not wearing a seat-belt at all or wearing it occasionally, and 92 of them reported using a cell phone while driving. No significant association was found for any of the driver-related factor or interview location with recent crash history.

Hazard perception of high- and low-risk sites

In twelve site pairs, five from Pakistan and seven from Cameroon, site hazard perception level was significantly higher and reported speeds were significantly lower for high-risk than for low-risk sites (Table 3). Correlations of pair-wise reported speeds were moderate to high (0.51≥ICC≤0.95). The highest negative speed differences (> 25 km/h) were observed for pair 4 (toll plaza at high-risk site), 16 (built-up area with markets and traffic at high-risk site), and 24 (a curve, with rain, oil tanker, and parked vehicle on high-risk site). Most high-risk sites where site hazard perception was not different or lower than low-risk sites were straight (N=10) and plain (N=8).

Factors associated with hazard perception

Compared to middle-aged drivers, significantly high hazard perception was reported by drivers aged 18-25 years and those aged 26-35 years (Table 4). Similarly, compared to heavy trucks, those driving cars or mini-trucks reported significantly lower hazard perception. Vehicle driven remained significantly associated with hazard perception in models 2 and 3. The association of age with hazard perception was not significant in model 3 (P > 0.05).

Hazard perception of Cameroonian sites was higher than Pakistani sites. Similarly

hazard perception at sites with irregular surface conditions, at intersections, and with ongoing maintenance works was significantly higher than those without them. Hazard perception of sites with straight and flat road profile was significantly lower than those with curve and slope road profile. Hazard perception of high-risk sites in built-up areas and having road width ≤ 8 m was significantly lower than low-risk sites with same features. Hazard perception of high-risk sites with visible hazard sign or a U-turn was significantly higher than low-risk sites with same features. Hazard perception of sites videos with maneuvering or oppositely moving vehicles was higher than site videos without them. Hazard perception of site videos with heavy vehicle or motorcycles was significantly lower than site videos without them. Hazard

129

perception for high risk site videos with rain was significantly lower than low risk site videos with same conditions.

DISCUSSION

This study showed that drivers were able to discriminate only half of high-risk sites from their matched low-risk sites. Further analysis showed that certain driver-, road-, and traffic-related characteristics were associated with a low hazard perception. For instance, participants who drove cars and mini-trucks had overall low hazard perception as compared to those driving trucks. Similarly, hazard perception of sites with flat and straight road profile was significantly lower than those without these profiles. Furthermore, high-risk crash sites situated in built-up area, with lane width ≤ 8 m, and during rainy conditions were perceived less hazardous than low-risk sites with same features.

The study methods were inspired from diagnostic test studies, to assess the accuracy

of drivers in differentiating high-risk sites from the low-risk ones (Flahault et al., 2005), and to analyze factors associated with low hazard perception. This study, for instance, showed that hazard perception of high-risk crash sites generally remained low, particularly for those sites which were straight and flat. Confronting these findings with the fact, previously shown on Yaoundé-Douala road section, that crashes were significantly higher for a road section with flat profile (Bhatti et al.); this study suggested that drivers preferred higher traffic speeds at flat road sites (Afukaar, 2003; Damsere-Derry, Afukaar, Donkor, & Mock, 2008). Such information could be extremely useful for safety experts, and these methods might facilitate prioritizing and providing insights into possible interventions at high-risk sites (Bishai, Asiimwe, Abbas, Hyder, & Bazeyo, 2008).

Furthermore, these methods assessed the odds of poor hazard perception of high-risk

compared to low-risk sites with same road features. For instance, the high-risk sites situated in built-up areas were not perceived hazardous compared to low-risk sites with similar road characteristics. Indeed, the incidence of RTCs on interurban road sections is higher in LMICs than in developed countries (Mohan, 2002). Ribbon development, improper crossing facilities, higher traffic mix, and absence of service lanes in LMICs could explain this crash risk in built-up areas (Ross et al., 1991). Our results suggested that implementing interventions which increase hazard perception might reduce crash risk on such sites (Bhatti et al.).

Moreover, narrow lane widths increased road hazard perception and reduced traffic

speed except for high-risk sites. It is likely that drivers were unable to perceive the hazardousness of such sites, because of little road furniture and hazard signage regarding speed adjustments, a condition frequent in LMICs (Mohan, 2002; Ross et al., 1991). This was consistent with the observation that high-risk sites with hazard signs resulted in higher hazard perception levels. This clearly indicated that proper installation and maintenance of such signs could have long-term road safety implications in LMICs (Milleville-Pennel, Hoc, & Jolly, 2007).

Previous studies have showed that adverse weather conditions significantly increased

interurban RTC risk in LMIC (Hijar et al., 2000; Majdzadeh, Khalagi, Naraghi, Motevalian, & Eshraghian, 2008). Our results showed that the ability to detect hazardousness of high-risk sites could be compromised during such weather conditions. Enforcing low traffic speeds and installation of real-time speed indications during such conditions could reduce crash risk on these roads (Konstantopoulos, Chapman, & Crundall).

130

Similarly, it was shown previously that drivers of lighter and more powerful vehicles were over-involved in crashes (Bener et al., 2006). Traffic composition on highways even in LMICs is significantly different from the cities, and passengers of such vehicles account for a majority of traffic injuries (Sobngwi-Tambekou et al., 2010). Furthermore, a low seat-belt and a high cell phone use reflected the low hazard perception observed in those driving such vehicles (Perneger & Smith, 1991). These results indicated that improving hazard awareness by enforcement and road measures might be useful in reducing crash risk in relatively vulnerable groups (Rosenbloom, Shahar, Elharar, & Danino, 2008).

Lastly, the high hazard perception of Cameroonian sites compared to Pakistani sites

could be explained by several factors. Firstly, the traffic was separated in Pakistan compared to non-separated in Cameroon. A higher hazard perception was observed for maintenance zones in Pakistan, where traffic was not separated (Lewis-Evans & Charlton, 2006; Morgan, Duley, & Hancock). Further, mountainous terrain, unfamiliarity with the road section, and right-hand drive could augment the hazard perception for Cameroonian sites among Pakistani drivers (McKenna, Horswill, & Alexander, 2006; Sagberg & Bjornskau, 2006). A reproduction of this study with Cameroonian drivers might help to assess the impact of familiarity on hazard perception (McKenna et al., 1991).

The study may have several limitations. Firstly, police underreporting of crashes

might have resulted in a small sample size to select high-risk sites (World Health Organization, 2009). Secondly, selection of voluntary drivers could lead to a prudent driver sample (Delhomme, 1991). Thirdly, drivers may respond higher hazard perception for sites in Pakistan known to them as high-risk ones (Lewis-Evans & Charlton, 2006). These biases might lead to high hazard perception of high-risk sites.

In conclusion, the study methods provided an opportunity to identify high-risk sites

with poor hazard perception features (Milleville-Pennel et al., 2007). These results showed that implementing cost effective interventions such as hazard signs at those sites could reduce traffic speeds (Rosenbloom et al., 2008). These methods could be useful to assess expected impact of such interventions at those sites (Milleville-Pennel et al., 2007). Furthermore, these might help in developing and implementing specific interventions adapted to local settings in HICs and LMICs (Jamison, Mosley, Measham, & Bobadilla, 1993). Feasibility of these methods however, remains to be assessed.

ACKNOWLEDGEMENTS Authors are especially thankful to Dr. Aftab Ahmed PATHAN (NHMP), Mr. Irshad SODHAR (NHMP), Mr. Naeemullah SHIEKH (NHMP), Mr. Javed SHAH (AKU), Dr. Sanaullah BASHIR (AKU), and Dr. Kiran EJAZ (AKU) for their support in data collection. Authors would like to thank all the drivers who participated in the hazard perception study and the owners of transport agencies who provided us the space to conduct the interviews. FUNDING First author is the PhD candidate at Université Victor Segalen Bordeaux 2. This position is funded by Higher Education Commission of Pakistan. Institut National de la Santé et de la Recherche Médicale Unité 897, France, funded the logistics for data collection. Funding bodies had no input in study design, analysis and interpretation of results. REFERENCES

131

Afukaar, F. K. (2003). Speed control in developing countries: issues, challenges and opportunities in reducing road traffic injuries. Inj Control Saf Promot, 10(1-2), 77-81. Benda, H. V., & Hoyos, C. G. (1983). Estimating hazards in traffic situations. Accid Anal Prev, 15(1), 1-9. Bener, A., Ghaffar, A., Azab, A., Sankaran-Kutty, M., Toth, F., & Lovasz, G. (2006). The impact of four-wheel drives on traffic disability and deaths compared to passenger cars. J Coll Physicians Surg Pak, 16(4), 257-260. Bhatti, J. A., Sobngwi-Tambekou, J., Lagarde, E., & Salmi, L. R. Situational factors associated with road traffic crashes: A case-control study on the Yaounde-Douala road section, Cameroon. Int J Inj Contr Saf Promot, 1-8. Bishai, D., Asiimwe, B., Abbas, S., Hyder, A. A., & Bazeyo, W. (2008). Cost-effectiveness of traffic enforcement: case study from Uganda. Inj Prev, 14(4), 223-227. Damsere-Derry, J., Afukaar, F. K., Donkor, P., & Mock, C. (2008). Assessment of vehicle speeds on different categories of roadways in Ghana. Int J Inj Contr Saf Promot, 15(2), 83-91. Deery, H. A. (1999). Hazard and risk perception among young novice drivers. J Safety Res, 30(4), 225-236. DeJoy, D. M. (1989). The optimism bias and traffic accident risk perception. Accid Anal Prev, 21(4), 333-340. Delhomme, P. (1991). Comparing one's driving with others': assessment of abilities and frequency of offences. Evidence for a superior conformity of self-bias? Accid Anal Prev, 23(6), 493-508. Elvik, R. (2008). A survey of operational definitions of hazardous road locations in some European countries. Accid Anal Prev, 40(6), 1830-1835. Finn, P., & Bragg, B. W. (1986). Perception of the risk of an accident by young and older drivers. Accid Anal Prev, 18(4), 289-298. Flahault, A., Cadilhac, M., & Thomas, G. (2005). Sample size calculation should be performed for design accuracy in diagnostic test studies. J Clin Epidemiol, 58(8), 859-862. Geurts, K., & Wets, G. (2003). Black Spot Analysis Methods: Literature Review (No. RA-2003-07). Diepenbeek: Steunpunt Verkeersveiligheid Bij Stijgende Mobiliteit. Geurts, K., Wets, G., Brijs, T., Vanhoof, K., & Karlis, D. (2006). Ranking and selecting dangerous crash locations: correcting for the number of passengers and Bayesian ranking plots. J Safety Res, 37(1), 83-91. Goldenbeld, C., & van Schagen, I. (2007). The credibility of speed limits on 80 km/h rural roads: The effects of road and person(ality) characteristics. Accid Anal Prev, 39(6), 1121-1130. Harre, N. (2000). Risk evaluation, driving, and adolescents: a typology. Dev Rev, 20, 206-226. Hijar, M., Carrillo, C., Flores, M., Anaya, R., & Lopez, V. (2000). Risk factors in highway traffic accidents: a case control study. Accid Anal Prev, 32(5), 703-709. Hosmer, D. W., & Lemeshow, S. (2000). Applied Logistic Regression (Second ed.). Danvers, MA: John Wiley & Sons, Inc. Jamison, D. T., Mosley, W. H., Measham, A. B., & Bobadilla, J. L. (Eds.). (1993). Disease Control Priorities in Developing Countries. Washington, DC: Oxford University Press. Kanellaidis, G., Zervas, A., & Karagioules, V. (2000). Drivers' Risk Perception of Road Design Elements. Transportation Human Factors, 2(1), 39-48. Konstantopoulos, P., Chapman, P., & Crundall, D. Driver's visual attention as a function of driving experience and visibility. Using a driving simulator to explore drivers' eye movements in day, night and rain driving. Accid Anal Prev, 42(3), 827-834. Lewis-Evans, B., & Charlton, S. G. (2006). Explicit and implicit processes in behavioural adaptation to road width. Accid Anal Prev, 38(3), 610-617.

132

Majdzadeh, R., Khalagi, K., Naraghi, K., Motevalian, A., & Eshraghian, M. R. (2008). Determinants of traffic injuries in drivers and motorcyclists involved in an accident. Accid Anal Prev, 40(1), 17-23. Mayhew, D. R., Simpson, H. M., & Pak, A. (2003). Changes in collision rates among novice drivers during the first months of driving. Accid Anal Prev, 35(683-91). McKenna, F. P., Horswill, M. S., & Alexander, J. L. (2006). Does anticipation training affect drivers' risk taking? J Exp Psychol Appl, 12(1), 1-10. McKenna, F. P., Stanier, R. A., & Lewis, C. (1991). Factors underlying illusory self-assessment of driving skill in males and females. Accid Anal Prev, 23, 45-52. Milleville-Pennel, I., Hoc, J.-M., & Jolly, E. (2007). The use of hazard road signs to improve the perception of severe bends. Accid Anal Prev, 39(4), 721-730. Mohan, D. (2002). Road safety in less-motorized environments: future concerns. Int J Epidemiol, 31(3), 527-532. Montella, A. (2010). A comparative analysis of hotspot identification methods. Accid Anal Prev, 42(2), 571-581. Morgan, J. F., Duley, A. R., & Hancock, P. A. Driver responses to differing urban work zone configurations. Accid Anal Prev, 42(3), 978-985. National Transport Research Centre. (1985). Traffic Counter measures in Pakistan: NTRC publication 85. Islamabad: National Transport Research Centre, Ministry of Communication. NHA. (2008). Traffic survey. Islamabad: National Highway Authority (NHA). Peden, M., Scurfiled, R., Sleet, D., Mohan, D., Hyder, A., & Jarawan, E. (2004). World report on road traffic injury prevention. Geneva: World Health Organization. Perneger, T., & Smith, G. S. (1991). The driver's role in fatal two-car crashes: a paired "case-control" study. Am J Epidemiol, 134(10), 1138-1145. Rosenbloom, T., Shahar, A., Elharar, A., & Danino, O. (2008). Risk perception of driving as a function of advanced training aimed at recognizing and handling risks in demanding driving situations. Accid Anal Prev, 40(2), 697-703. Ross, A., Bagunley, C., Hills, B., McDonald, M., & Silcock, D. (1991). Towards safer roads in developing countries—a guide for planners and engineers. Crowthorne, UK: Transport Research Laboratory. Rundmo, T., & Iversen, H. (2004). Risk perception and driving behaviour among adolescents in two Norwegian counties before and after a traffic safety campaign. Safety Sci, 42(1), 1-21. Sabey, B. E., & Taylor, H. (1980). The Known Risks We Run: The Highway. In S. C. Schwing & W. A. Alber (Eds.), Societal Risk Assessment: How safe is safe enough. New York: Plenum Press. Sagberg, F., & Bjornskau, T. (2006). Hazard perception and driving experience among novice drivers. Accid Anal Prev, 38(2), 407-414. Sobngwi-Tambekou, J., Bhatti, J., Kounga, G., Salmi, L.-R., & Lagarde, E. (2010). Road traffic crashes on the Yaoundé–Douala road section, Cameroon Accid Anal Prev, 42(2), 422-426. Sümer, N., Ünal, A. B., Birdal, A., Çinar, P., & Çevikoglu, S. (2007). Comparison of self-reported and computer-based Hazard perception skills among novice and experienced drivers. Paper presented at the Fourth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, Stevenson, Washington. SWOV. (2008). SWOV fact sheet: hazard perception. Leidschendam: SWOV Institute for Road Safety Research. Trankle, U., Gelau, C., & Metker, T. (1990). Risk perception and age-specific accidents of young drivers. Accid Anal Prev, 22(2), 119-125. World Health Organization. (2009). Global Status Report on Road Safety. Geneva: WHO.

133

Table 1. Characteristics of high- and low-risk sites on Yaoundé-Douala and Karachi-Hala road sections

Pakistan Cameroon High-risk (%) Low-risk (%) High-risk (%) Low-risk (%) N=16 N=16 N=10 N=10

Site factors

Straight 87 87 20 50

Irregular road shoulder 81 81 100 90

Irregular surface conditions 75 63 90 40

Lane width ≤ 8 75 100 80 90

Flat 62 81 50 60

Visible road side obstacle 50 38 100 90

Vertical road sign 38 19 40 10

Built-up road section 31 56 50 10

Intersection 31 31 40 0

U-turn 19 19 0 0

Diversion 13 6 0 0

Continuous road-markings 13 0 50 10

Traffic factors

Visible heavy vehicle 88 69 40 40

Same direction moving vehicle 81 88 60 60

Maneuvering vehicle (overtake, crossing) 69 56 30 10

Opposite direction moving vehicle 25 19 40 50

Visible pedestrian 19 31 50 20

Visible motorcyclists 13 19 10 0

Rain 0 0 20 20

Wet surface 0 0 70 50

134

Table 2. Characteristics of Pakistani drivers included in sample (N=100).

Proportion (%) Recent crash (%) N=100 N=20 Age (years) -18-25 9 5 - 26-35 38 30 - 36-45 28 45 - > 45 25 20 Vehicle driven -Truck 33 30 -Motorcar 43 45 -Bus 11 0 -Mini-bus 6 15 -Mini-truck 5 5 -Motorcycle 2 5 Permanent domicile -Karachi 54 50 -Sindh 9 20 -Punjab 29 20 -NWFP/Baluchistan 8 10 Education (years) - None 22 35 - 1-5 22 10 - 6-10 37 40 - >10 19 15 Profession -Driver 85 85 -Other 15 15 Married 88 80 Familiar with road 83 90 Licensed after test 45 50 Seat-belt use -None 27 20 -Occasional 47 55 -Frequent 26 25 Sleepy driving 29 40 Phone dialing 84 80 Phone receiving 92 85 Traffic Ticket 49 50 Drunk driving 2 10

135

Table 3. Differences in hazard perception and preferred speeds for high- and low-risk site

pairs on Yaoundé-Douala and Karachi-Hala road sections

Site hazard perception Traffic hazard perception Difference in traffic speed Site pair D+ D++ D- D- - P D+ D++ D- D- - P Mean P ICC 95% CI Pakistan 1 16 37 4 5 <0.001 1 1 11 27 <0.001 -4.50 0.021 0.81 0.72, 0.87 2 8 14 6 15 0.838 9 24 3 11 0.001 -5.55 0.002 0.84 0.76, 0.89 3 6 10 7 28 0.002 4 28 8 9 0.016 3.50 0.051 0.81 0.71, 0.87 4 10 82 0 0 <0.001 7 60 1 7 <0.001 -43.15 <0.001 0.60 0.40, 0.73 5 1 5 1 3 0.566 6 6 1 1 0.015 -5.00 <0.001 0.95 0.93, 0.97 6 7 8 12 37 <0.001 0 4 5 17 0.005 12.75 <0.001 0.83 0.75, 0.89 7 11 26 8 13 0.024 3 2 11 26 <0.001 1.85 0.240 0.85 0.78, 0.90 8 18 30 2 7 <0.001 2 4 11 45 <0.001 14.35 <0.001 0.51 0.27, 0.67 9 6 9 12 23 0.010 2 4 11 22 <0.001 7.15 <0.001 0.90 0.84, 0.93 10 11 17 9 21 0.616 6 11 8 23 0.075 2.10 0.238 0.82 0.73, 0.88 11 16 24 8 7 <0.001 16 25 7 6 <0.001 -6.45 <0.001 0.82 0.73, 0.88 12 3 3 13 37 <0.001 2 2 16 22 <0.001 16.50 <0.001 0.83 0.75, 0.89 13 0 4 9 49 <0.001 2 6 2 8 0.545 10.25 <0.001 0.84 0.76, 0.89 14 9 13 14 27 0.010 10 32 3 7 <0.001 -4.85 0.007 0.85 0.78, 0.90 15 12 51 1 3 <0.001 7 48 3 4 <0.001 -15.45 <0.001 0.88 0.82, 0.92 16 10 68 0 1 <0.001 13 75 0 1 <0.001 -25.01 <0.001 0.86 0.80, 0.91 Cameroon 17 12 14 3 3 <0.001 6 50 0 4 <0.001 -5.55 0.001 0.80 0.70, 0.86 18 14 17 7 9 0.114 0 1 6 72 <0.001 14.10 <0.001 0.88 0.82, 0.92 19 11 5 1 1 <0.001 4 8 0 12 0.685 -8.65 <0.001 0.90 0.85, 0.93 20 11 11 11 10 0.844 1 2 5 67 <0.001 11.10 <0.001 0.88 0.82, 0.92 21 15 20 2 1 <0.001 9 65 0 0 <0.001 -12.00 <0.001 0.86 0.80, 0.91 22 12 20 2 4 <0.001 4 16 6 15 0.904 -3.95 0.007 0.88 0.82, 0.92 23 4 0 10 23 <0.001 4 1 8 37 <0.001 14.75 <0.001 0.89 0.84, 0.93 24 14 49 5 1 <0.001 6 70 2 0 <0.001 -25.09 <0.001 0.84 0.76, 0.89 25 15 20 4 4 <0.001 9 77 0 0 <0.001 -17.10 <0.001 0.77 0.66, 0.85 26 13 23 10 15 0.068 1 85 0 0 <0.001 -19.55 <0.001 0.84 0.76, 0.89

D+ - Minor discordance: High-risk site was ranked one level more hazardous than low-risk site D++ - Major discordance: High-risk site was ranked two levels more hazardous than low-risk site D- - Minor discordance: Low-risk site was ranked one level more hazardous than high-risk site D- - - Major discordance: Low-risk site was ranked two levels more hazardous than high-risk site ICC-Intra class correlation coefficient CI- confidence interval

136

Table 4. Factors associated with hazard perception at high- and low-risk sites on Yaoundé-

Douala and Karachi-Hala road sections

Model 1 Model 2 Model 3 OR 95% CI OR 95% CI OR 95% CI Driver Age (years) -18-25 1.93 1.50-2.48 1.95 1.71-2.22 2.09 1.53-2.82 - 26-35 1.21 1.03-1.42 1.20 1.12-1.30 1.23 1.02-1.49 - 36-45 1 1 1 - > 45 1.00 0.84-1.19 1.00 0.91-1.08 1.00 0.81-1.23 Vehicle driven - Truck* 1 1 - Motorcar 0.71 0.62-0.82 0.70 0.65-0.75 0.69 0.58-0.81 - Mini-bus 0.85 0.64-1.13 0.85 0.73-0.98 0.83 0.61-1.16 - Mini-truck 0.68 0.50-0.92 0.67 0.58-0.78 0.66 0.46-0.94 - Bus 1.38 1.10-1.74 1.39 1.24-1.55 1.40 1.08-1.86 Site Cameroon vs. Pakistan 6.53 4.91-8.68 6.88 5.26-9.02 7.61 6.55-8.85 Straight vs. curve 0.72 0.58-0.88 0.69 0.56-0.86 0.69 0.62-0.76 Irregular vs. regular road surface 4.78 3.74-6.09 4.61 3.71-5.75 5.36 4.71-6.11 Flat vs. hill 0.57 0.48-0.69 0.54 0.45-0.65 0.54 0.50-0.60 Intersection vs. none 1.46 1.13-1.90 1.40 1.08-1.82 1.49 1.30-1.73 Work zone 24.34 15.00-39.51 23.33 14.43-37.71 31.82 24.53-41.26 Lane width ≤ 8 m vs. > 8 m - High-risk 0.51 0.43-0.61 0.50 0.37-0.68 0.48 0.42-0.56 - Low-risk 18.48 10.39-32.85 22.64 12.42-41.26 23.57 17.29-32.14 Built-up area vs. rural - High-risk 0.58 0.51-0.68 0.67 0.49-0.90 0.57 0.46-0.69 - Low-risk 2.04 1.51-2.74 2.29 1.69-3.09 2.18 1.86-2.56 Vertical road sign - High-risk 2.75 2.38-3.16 2.64 2.02-3.44 2.94 2.48-3.50 - Low-risk 0.50 0.34-0.72 0.50 0.36-0.70 0.47 0.39-0.58 U-turn - High-risk 8.00 6.36-10.22 7.69 5.09-11.62 9.58 7.50-12.24 - Low-risk 0.62 0.39-0.97 0.65 0.43-0.99 0.60 0.47-0.76 Traffic Heavy traffic 0.37 0.29-0.47 0.41 0.33-0.51 0.34 0.30-0.39 Maneuvering vehicle 1.91 1.50-2.42 1.82 1.45-2.32 2.01 1.77-2.29 Oppositely coming traffic 2.05 1.56-2.69 2.15 1.64-2.82 2.18 1.88-2.53 Motorcycle 0.53 0.38-0.75 0.52 0.37-0.74 0.51 0.43-0.61 Rain - High-risk 0.19 0.15-0.24 0.17 0.11-0.27 0.16 0.13-0.20 - Low-risk 0.47 0.29-0.78 0.48 0.29-0.79 0.45 0.35-0.58

OR – Odds ratio 95% CI – 95% confidence interval Model 1 – Without random effect Model 2 – Site as random effect Model 3 – Driver as random effect * Motorcyclists were not analyzed separately as no differences were observed.

137

Appendix 8: Study V supplementary results Table 20. Driver age, sex, and vehicles driven on Karachi-Hala road section (July 2009)

N (%) Seat belt /

helmet use P

Vehicle driven <0.001 - Motorcycle 341 6.2 35.5 - Car/jeep 2151 39.1 29.8 - Minibus/van 217 3.9 18.9 - Mini-truck 212 3.9 8.0 - Bus 529 9.6 7.9 - Truck/trailer 2006 36.5 3.3 - Other 40 0.8 - Sex - - Male 5483 99.9 16.8 - Female 7 0.1 71.4 Age (y)* <0.001 18-25 411 7.8 24.3 26-35 1850 35.3 22.1 36-45 2053 39.2 12.0 46-55 802 15.3 11.1 ≥56 123 2.3 8.1

* Two drivers were less than 18 years based on observation

© Bhatti, PPCT, INSERM U897 138

Table 21. Situational factors at high- and low- risk site pairs on Yaoundé-Douala and Karachi-Hala road sections Pair Site hazards Traffic hazards

Flat Curve Irregular surface

Irregular shoulder

Built-up Roadside obstacles

Intersection Same direction

Opposite direction

Pedestrian movement

Motorcyclist Speeding/ Overtaking

Rain Wet surface

Pakistan 1 D01 D10 D10 C11 D01 C11 D01 C11 C00 D01 C00 C00 C00 C00 2 D01 C00 C11 C11 D01 C00 D01 C11 D01 D01 D01 C11 C00 C00 3 C11 C00 D10 D10 D10 D10 D10 C11 D01 D10 C00 C00 C00 C00 4 C00 C00 D10 D10 C00 D10 D10 C11 D10 D10 D10 C11 C00 C00 5 C11 C00 D10 C11 C00 C00 C00 D10 C00 C00 C00 D10 C00 C00 6 C11 D01 C11 C11 C11 C11 D01 D01 D01 C00 C00 D01 C00 C00 7 D10 D10 C11 C11 D10 D10 D10 D01 C00 C00 C00 C11 C00 C00 8 D01 C00 D10 C11 D01 D01 D01 C11 C00 C00 C00 C11 C00 C00 9 D01 D01 D01 C11 D01 D01 C00 C11 C00 D01 C00 C11 C00 C00

10 D10 C00 D01 D01 D01 C00 C00 C11 D10 D01 C00 D01 C00 C00 11 D01 C00 C11 C11 D01 D01 C00 C11 C00 D01 D01 C11 C00 C00 12 C11 C00 D01 D01 C00 C00 C00 D01 C00 C00 C00 D01 C00 C00 13 C11 C00 C11 C11 D01 C00 C00 C11 C00 C00 C00 D10 C00 C00 14 C11 C00 C11 C11 D01 C11 D01 C11 C00 C00 D01 D10 C00 C00 15 C11 C00 D01 D01 D10 D10 D10 C11 D10 C00 C00 D10 C00 C00 16 C11 C00 D10 D10 D10 D10 D10 D10 D10 D10 D10 D10 C00 C00

Cameroon 17 C11 D10 D10 C11 D10 C11 D10 C00 C00 D10 C00 C00 C00 C11 18 C11 D10 C11 C11 D10 C11 C00 D01 D01 C00 C00 C00 D01 C11 19 D01 C11 D10 C11 D10 C11 C00 C11 C11 C11 C00 C00 C00 C00 20 C11 D10 C11 C11 C00 D10 C00 C00 D01 C00 C00 C00 D01 C11 21 D01 C11 D10 C11 D10 C11 D10 D10 C00 D10 C00 C00 C00 D10 22 C00 C11 C11 C11 D01 C11 C00 C11 D01 C00 C00 D01 C00 D10 23 C00 C11 D01 C11 C00 C11 D10 D01 C00 C11 C00 C00 C00 D01 24 D01 D10 D10 C11 C00 C11 C00 C11 D10 C00 C00 D10 D10 D10 25 D10 D01 D10 C11 D10 C11 D10 D10 D10 D10 D10 D10 C00 D10 26 D10 C00 D10 D10 C00 C11 C00 C11 C11 C00 C00 D10 D10 D01

Total - D10 4 6 12 4 9 6 9 4 6 6 3 8 2 4 - D01 8 3 5 3 9 3 5 5 6 4 3 4 2 2

C11 Concordant high- and low-risk site pair: hazard present C00 Concordant high- and low-risk site pair: hazard absent D10 Discordant high- and low-risk site pair: hazard at crash-site only D01 Discordant high- and low-risk site pair: hazard at non crash site only


Table 22. Driver-related factors associated with hazard perception of sites on Yaoundé-Douala and Karachi-Hala road sections

Sample Perception P* Safe Hazardous N=2 697 N=2 503 N % % Driver characteristics Age (y) <0.001 -18-25 9 7.2 10.9 - 26-35 38 37.9 38.1 - 36-45 28 29.1 26.9 - > 45 25 25.8 24.1 Education (y) 0.06 - None 22 22.8 21.1 - 1-5 22 21.8 22.3 - 6-10 37 35.6 38.6 - >10 19 19.8 18.1 Profession 0.02 - Driver 85 83.9 86.1 - Other 15 16.1 13.9 Marital status 0.88 - Married 88 88.1 87.9 - Single 12 11.9 12.1 Permanent domicile 0.001 - Karachi 54 56.5 51.3 - Sindh 9 8.9 9.1 - Punjab 29 26.9 31.3 - NWFP/Baluchistan 8 7.7 8.3 Vehicle driven <0.001 - Truck 33 30.6 35.6 - Motorcar 43 46.4 39.3 - Bus 11 9.6 12.5 - Mini-bus 6 6.2 5.8 - Mini-truck 5 5.3 4.7 - Motorcycle 2 1.9 2.1 Licensed after test 0.04 - Yes 45 43.7 46.4 - No 55 56.3 53.6 Seat-belt use 0.02 - None 27 26.0 28.1 - Occasional 47 46.5 47.6 - Frequent 26 27.5 24.3 Sleepy driving 0.80 - Yes 29 28.9 29.2 - No 71 71.1 70.8 Phone dialling 0.03 - Yes 84 83.0 85.1 - No 16 17.0 14.9 Phone receiving 0.98 - Yes 92 92.0 92.0 - No 8 8.0 8.0 Traffic Ticket 0.003 - Yes 49 47.0 51.1 - No 51 53.0 48.9 Drunk driving 0.85 - Yes 98 98.0 98.0 - No 2 2.0 2.0 Recent crash history 0.05 - Yes 20 21.0 18.9 - No 80 81.0 81.1

* Variables with P<0.05 were initially included in model 1 with backward selection strategy


Table 23. Situational factors associated with hazard perception of sites on Yaoundé-Douala and Karachi-Hala road sections (Drivers =100)

Sites Perception P* Sites Perception P* Safe Hazardous Safe Hazardous N=2 697 N=2 503 N=2 697 N=2 503 N % % % N % % % Site situation Site situation Vertical road profile <0.001 High risk crash site <0.001 - Flat 34 65.4 28.9 40.8 - Yes 26 50.0 44.6 55.8 - Hill 18 34.6 71.1 59.2 - No 26 50.0 55.4 44.2 Road surface <0.001 Country <0.001 - Irregular 35 67.3 60.3 74.9 - Pakistan 32 61.5 73.9 48.2 - Regular 17 32.7 39.7 25.1 - Cameroon 20 38.5 26.1 51.7 Road shoulder <0.001 - Irregular 45 86.5 82.8 90.5 Traffic situation - Regular road shoulder 7 13.4 17.2 9.5 Manoeuvring (overtake, crossing) vs. no <0.001 Surface markings <0.001 - Yes 24 46.2 49.4 42.7 - Continuous 8 15.4 10.6 20.5 - No 28 53.8 50.6 57.3 - Discontinuous 44 84.6 89.4 79.5 Same direction moving vehicle 0.45 Lane width <0.001 - Yes 39 75.0 74.5 75.5 - ≤ 8 m 45 86.5 90.3 82.5 - No 13 25.0 25.5 24.5 - > 8 m 7 13.5 9.7 17.5 Opposite direction moving vehicle <0.001 Horizontal road profile <0.001 - Yes 16 30.8 20.5 41.8 - Straight 35 67.3 78.9 54.8 - No 36 69.2 79.5 58.2 - Curve 17 32.7 21.1 45.2 Visible heavy vehicle vs. without <0.001 Built-up area <0.001 - Yes 33 63.5 59.5 32.9 - Yes 20 38.5 33.3 44.0 - No 19 36.5 40.5 67.1 - No 32 61.5 66.7 56.0 Motorcyclists <0.001 Visible road side obstacle <0.001 - Yes 6 11.5 91.9 84.7 - Yes 33 63.5 53.2 74.5 - No 46 88.5 8.1 15.3 - No 19 36.5 46.8 25.5 Pedestrian <0.001 Intersection vs. none <0.001 - Yes 15 28.9 18.9 39.6 - Yes 14 26.9 21.1 33.2 - No 37 71.1 81.1 60.4 - No 38 73.1 78.9 66.8 Rain <0.001 Vertical road sign <0.001 - Yes 4 7.7 6.6 8.9 - Yes 14 26.9 21.8 32.4 - No 48 92.3 93.4 91.1 - No 38 73.1 78.2 67.6 Wet surface <0.001 U-turn vs. none <0.001 - Yes 12 23.1 13.7 33.2 - Yes 6 11.5 9.5 13.7 - No 40 76.9 86.3 66.8 - No 46 88.5 90.5 86.3

141

Résumé Introduction : La sécurité routière sur le réseau interurbain est un problème majeur de santé publique dans les Pays à Revenu Bas et Moyen (PRBM) mais peu d'attention y a été consacrée. Les objectifs de cette thèse étaient d’évaluer le fardeau des traumatismes en relation avec le trafic interurbain, la déclaration des usagers blessés dans des bases de données différentes, d’analyser l’association entre les facteurs situationnels (caractéristiques physiques et circonstances environnementales) et les sites des accidents et la perception de la dangerosité des tronçons accidentogènes dans les PRBM. Méthodes et résultats : Pour répondre à ces objectifs, cinq études spécifiques ont été réalisées dans deux PRBM, le Cameroun et le Pakistan. L’étude I a évalué le nombre de tués par véhicules-km parcourus et les facteurs qui leur étaient associés, en utilisant les rapports de police entre 2004 et 2007 sur l’axe Yaoundé-Douala, Cameroun. Le taux de mortalité était de 73 par 100 millions véhicules km parcourus, un taux 35 fois plus élevé que sur un même type de route en pays à revenu élevé. La mortalité était plus élevée pour les accidents impliquant des usagers vulnérables, les véhicules roulant en sens opposé et ceux dus à une défaillance mécanique, y compris un éclatement de pneu. L’étude II a évalué les différences de déclaration d’accidents faites par les services de police, d’ambulance et des urgences en 2008 sur l’axe Karachi-Hala, Pakistan. La mortalité était de 53 par 109 véhicules-km parcourus ; le taux de mortalité était 13 fois plus élevé sur cet axe par rapport à un même type de route en France. La police a déclaré un mort sur cinq et un blessé grave sur dix. Les usagers de la route vulnérables, y compris les piétons et deux-roues ont été deux fois moins déclarés par la police que par les services d'ambulance ou des urgences. L’étude III a étudié les facteurs situationnels associés aux sites des accidents sur l’axe Yaoundé-Douala par une approche de type cas-témoins. Les facteurs tels que le profil routier plat (rapport de cotes [RC] ajusté =1,52 ; intervalle de confiance à 95 % [IC95 %]=1,15-2,04), les surfaces irrégulières (RC=1,43 ; IC95 %=1,04-1,99), les obstacles à proximité (RC=1,99 ; IC95 %=1,09-3,63) et les intersections à trois (RC=3,11 ; IC95 %=1,15-8,39) ou à quatre directions (RC=3,23 ; IC95 %=1,51-6,92) étaient significativement associés à des sites d’accidents corporels. De plus, la probabilité des accidents augmentait dans des zones urbaines situées dans des régions de plaine (RC=2,23 ; IC95 %=1,97-2,77). L’étude IV a étudié le fardeau des traumatismes dus aux accidents ainsi que les facteurs associés dans des zones en travaux sur l’axe Karachi-Hala en utilisant les méthodes de cohorte historique. Un tiers de la mortalité routière était survenu dans des zones en travaux et le risque de mortalité était quatre fois plus élevé dans ces zones que dans les autres zones. Un accident sur deux a eu lieu entre des véhicules roulant en sens opposé dans ces zones. L’étude V a étudié la perception de la dangerosité des tronçons accidentogènes (au moins 3 accidents sur 3 ans) et non accidentogènes (aucun accident déclaré) sur les deux axes des précédentes études, en montrant leurs vidéos à des conducteurs volontaires pakistanais. Les conducteurs n’ont perçu comme dangereux que la moitié des tronçons accidentogènes. La perception de la dangerosité des tronçons plats et droits était plus faible par rapport aux tronçons en courbes et avec une pente. La perception de la dangerosité en zone urbaine d’un tronçon accidentogène était significativement moins élevée (RC=0,58 ; IC95 %=0,51-0,68) que celle d’un tronçon non accidentogène ayant la même caractéristique (RC=2,04 ; IC95 %=1,51-2,74). La perception de la dangerosité d’un tronçon accidentogène avec panneau de signalisation était significativement plus élevée (RC=2,75 ; IC95 %=2,38-3,16) par rapport à des tronçons non accidentogènes ayant la même caractéristique (RC=0,50 ; IC95 %=0,34-0,72). Conclusion : Cette thèse montre combien des méthodes épidémiologiques simples, mais novatrices, peuvent être utiles pour évaluer le fardeau des traumatismes par accidents et leurs facteurs de risques dans les PRBM. Ces pays sont confrontés à un énorme fardeau de morbidité routière qui est souvent sous-déclarée dans les données de la police. Un système de surveillance fiable et valide est nécessaire dans les PRBM. De plus, la politique de prévention pourrait être améliorée par une meilleure communication d’information entre les autorités routières et policières concernant les facteurs situationnels. De la même façon, les mesures de sécurité dans les zones en travaux devraient être contrôlées par un système dédié. Enfin, la sécurité routière sur les routes interurbaines dans les PRBM pourrait être améliorée en rendant les routes plus « informant », en particulier avec l’application de mesures peu couteuses telles que les panneaux de signalisations sur les tronçons accidentogènes. Mots Clés: Accidents de la circulation; pays en développement ; trauma; usagers vulnérables.

Abstract Background: Interurban traffic safety is a major public health problem, but has received little attention in Low- and Middle-Income Countries (LMICs). The objectives of this thesis were to assess the burden of injury related to interurban traffic, and reporting of these injuries in different datasets, to analyze situational factors (physical characteristics and environmental circumstances) associated with crash sites, and road hazard perception of high-risk crash sites in LMICs. Methods and results: These objectives were assessed in five specific studies conducted in two LMICs, Cameroon and Pakistan. In study I, traffic fatality per vehicle-km and associated crash factors were assessed using police reports for years 2004 to 2007, on the two-lane Yaoundé-Douala road section in Cameroon. Traffic fatality was 73 per 100 million vehicle-km, a rate 35 times higher than a similar road in a high-income country. Fatality was higher for crashes involving vulnerable road users, crashes between oppositely-moving vehicles, and those due to mechanical failure including tyre burst. In study II , traffic injury reporting to police, ambulance, and Emergency Department (ED) in 2008 was assessed, on the four-lane Karachi-Hala road section in Pakistan. Crash fatality was over 53 per 109 vehicle-km, a rate 13 times higher than a similar road in France. Police reported only one out of five fatalities and one out of ten severe injuries. Vulnerable road users were two times less reported in police data than ambulance or ED data. In study III , situational factors associated with injury crash sites were assessed on the Yaoundé-Douala road section, using case-control methods. Factors such as flat road profiles (adjusted Odds Ratios [OR]=1.52; 95% Confidence Interval [95%CI]=1.15-2.01), irregular surface conditions (OR=1.43; 95%CI=1.04-1.99), nearby road obstacles (OR=1.99; 95%CI=1.09-3.63), and three- (OR=3.11; 95%CI=1.15-8.39) or four-legged (OR=3.23; 95%CI= 1.51-6.92) intersections were significantly associated with injury crash sites. Furthermore, the likelihood of crash increased with built-up areas situated in plain regions (OR=2.33; 95%CI=1.97-2.77). In study IV, traffic injury burden and factors associated with Highway Work Zones (HWZs) crashes were assessed on the Karachi-Hala road section, using historical cohort methods. HWZs accounted for one third of traffic fatalities, and fatality per vehicle-km was four times higher in HWZs than other zones. One out of two HWZ crashes occurred between oppositely moving vehicles. In study V, hazard perception of high-risk (with ≥ 3 crashes in 3 years) and low-risk sites (no crash reported) from the two above road sections was assessed by showing videos to voluntary Pakistani drivers. Drivers were able to identify only half of the high-risk sites as hazardous. Sites with a flat and straight road profile had a lower hazard perception compared to those with curved and slope road profile. High-risk sites situated in built-up areas were perceived less hazardous (OR = 0.58; 95%CI=0.51-0.68) compared to low-risk sites (OR = 2.04; 95%CI=1.51-2.74) with same road situation. Further, high-risk sites with vertical road signs were more likely to be perceived hazardous (OR = 2.75; 95%CI=2.38-3.16) than low-risk sites (OR = 0.50; 95%CI=0.34-0.72) with such signs. Conclusion: This thesis illustrates how innovative yet simple epidemiological methods can be useful in assessing the injury burden and specific risk factors in LMICs. These countries face a high burden of interurban road injuries, mostly under-reported in police data. A reliable and accurate injury surveillance system is needed in these countries. Moreover, prevention policy can be improved by better information transfer between road and police authorities regarding situational factors. Similarly, a monitoring system is required to examine the HWZ safety interventions in these countries. Lastly, interurban road safety can be improved by making roads self-explaining, especially by implementing low-cost interventions such as vertical signs at high-risk sites. Keywords: Developing country; highway safety; injury; prevention; vulnerable road users.

Documents

DOCTORAT DE L’UNIVERSITÉ BORDEAUX 2 · Université Victor Segalen Bordeaux 2 Année 2010 Thèse n°1 724 THÈSE pour le DOCTORAT DE L’UNIVERSITÉ BORDEAUX 2 Mention : Science,