A crash was defined as any event where a motorized vehicle, including motorcycles, was involved in a collision with another vehicle, road user, or other obstacle, and reported in either of the police, ambulance, and hospital ED datasets 13 15 . RTI was defined as any person incurring a physical injury as a result of a crash reported to any of the above datasets 13 15 .

Since 2004, the National Highway & Motorway Police (NHMP) has been enforcing traffic rules 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 km), beat 34 (63 to 114 km), beat 33 (115 to 162 km), and beat 32 (163 to 212 km). NHMP deploys on each beat four motor vehicles and four patrolling officers per eight-hour shift 19 .

For every crash, a standard accident analysis report is filed during the first 24 hours by the attending NHMP officer 20 . A copy of this report is kept in the NHMP regional office. Details on the 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, gender, outcome (dead; transported to hospital; and not transported to hospital), and, if transported, name of the hospital.

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 injured patients, and funds are raised by transporting other patients. In most cases, ambulances are only staffed by the driver. A clerk at the post can come with the driver if he thinks this is justified, for instance, crashes with multiple patients. The ambulance communicates with the emergency post through a wireless system or 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). When reaching the scene, attendants separate injured from dead patients. Those severely injured are transported to the nearest hospital; preference is given to the government hospital if available. All information on the intervention, including crash location, injured patients 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, gender, age, user type (pedestrian, motorcycle rider, or vehicle occupant), and outcome (died, including whether the person died at crash scene, during transport, or at hospital ED; injured and transported, including hospital taken to; injured and not transported) were extracted from these log books 21 .

The Road Traffic Injury Research & Prevention Centre (RTIRP) at the Jinnah Post Graduate Medical Centre (JPMC) has systematically collected standard forms since September 2006 24 , information on RTI patients presenting at the hospital 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 described elsewhere 24 25 .

This dataset includes information on the crash date, time, and location, and patient's name, age, gender, 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, range 1 to 75) 26 , and outcome (discharged, admitted/referred, or died) of patients were recorded during their stay in the hospital ED. Information on RTI patients involved in crashes on selected road section was extracted from this dataset.

All information was recorded in Excel^® spreadsheets. We compared percentages for crash and injury patient characteristics across the three datasets. For the hospital ED dataset, we described outcome for the following NISS categories: minor injury, scores ranging from 1 to 3; major injury, scores ranging from 4 to 8; and severe injury, scores above 9 26 . Same records present in two or more datasets were matched using crash date and time, name, age, and gender of RTI patients. For matched records, we identified differences in reported outcome. To estimate total fatalities, a person reported injured in police statistics, but dead in ambulance data was considered as dead. The number of unique deaths and injuries were then assessed after removing duplicates of records appearing in two or more datasets. Ascertainment rate for police, ambulance, and hospital 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 hospital ED dataset, thus violating one of the basic assumptions of the method 15 . The unique records and traffic counts from highway authority were used to compute overall traffic fatality and injury rates per vehicle kilometre in 2008 for this road section 17 . Considering that there would be missing information for variables used in linking datasets, we carried a secondary analysis considering situations where at least one of the variables could be matched.

All the police, ambulance, and hospital ED data used in this study were publicly accessible and analyses were conducted with approval from the respective institutions. Furthermore, this manuscript did not permit identification of any RTI patient.

In 2008, 143 RTIs were reported to the police, 531 to EAS, and 661 to hospital ED. Names were available for 67.1% (n = 96) of police and 78.0% (n = 414) of EAS reported injury patients (Table 1). Information on age was available for 74.1% (n = 106) of police and 67.6% (n = 359) of EAS reported injury patients. Few records in the hospital ED dataset were without names (n = 13) or age (n = 5). The most injured patients in the three datasets were aged 16-45 years: 61.5% (n = 88) in police, 55.0% (n = 292) in EAS, and 78.1% (n = 516) in hospital ED. Males accounted for a majority of injuries, 92.1% (n = 609) of injured patients in hospital ED.

The proportion of pedestrians in police reported crashes was 3.5% (n = 5), whereas this was 7.5% (n = 40) in the EAS and 12.7% (n = 83) in the hospital ED. The proportion of motorcycle riders in police reported crashes were 4.2% (n = 6), whereas this was 9.2% (n = 49) in EAS and 30.6% (n = 203) in hospital ED. Occupants of four wheeled vehicles accounted for a majority of injuries in the three datasets: 83.9% (n = 120) in police, 75.9% (n = 403) in EAS, and 49.5% (n = 327) in hospital ED. In the hospital ED, only 13.6% (n = 21) of the 154 injury patients riding motorcycles were wearing helmets. Similarly, only 3.2% (n = 3) out of 93 four-wheeled vehicle occupants were wearing a seat belt at time of crash.

In 2008, police reported 43 crashes, whereas 255 crashes were reported to EAS and 449 to hospital 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 hospital ED. No information on crash outcome was available in 13.3% of EAS reported crashes, and 6.7% of those reported to hospital ED. Over half (n = 80, 55.9%) of police-reported injured patients received hospital care; 50.0% (n = 40) of these patients were injured between km 16 and km 120 and treated in Karachi. RTIRP hospitals accounted for 17 of those treated in Karachi. Nearly one fifth (n = 27, 18.8%) of RTI patients reported in police records died, whereas this proportion was 10.4% (n = 55) for EAS and 9.1% (n = 60) for hospital ED reported patients (Table 1). One fourth (n = 36, 25.2%) of police reported injured patients were not transported to the hospital, whereas this was 9.0% (n = 48) for EAS reported patients.

Out of 661 patients presenting to hospital 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 8.8% (n = 58) patients. NISS were available for 92.8% (n = 614) of 661 hospital ED patients; 34.2% (n = 210) had minor, 28.0% (n = 172) had major, and 37.8% (n = 232) had severe RTI. All those who were reported to have died had a severe RTI. Of 206 admitted patients, 5.8% (n = 12) had minor injuries and 24.8% (n = 51) had major injuries.

A total of 108 patients were found in two or more datasets yielding 1 214 unique records from the three datasets (Figure 1). Of 143 police reported patients 20.3% (n = 29) were observed in other datasets; 19.5% (n = 28) in EAS and 9.8% (n = 14) in hospital ED. Of 531 EAS reported patients 20.2% (n = 107) were observed in other datasets; 5.3% (n = 28) in police and 17.3% (n = 92) in hospital ED. Of 661 hospital ED reported patients 14.1% (n = 93) were observed in other datasets; 2.1% (n = 14) in police and 13.9% (n = 92) in EAS.

Discrepancies were observed for outcome of injuries reported in police and ambulance records: 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 hospital ED records, and nine of 84 injured patients in EAS were reported dead in hospital ED records.

Based on matching, 119 patients died in 2008 on this interurban road section (Table 2); of these, police recorded 22.6% (n = 27), EAS 46.2% (n = 55), and hospital ED 50.4% (n = 60). Similarly, of 1 095 unique injured patients, police recorded 10.6% (n = 116), EAS 43.5% (n = 476), and hospital ED 54.9% (n = 601). Traffic fatality was 54 deaths and injuries were slightly over 500 per 109 vehicle kilometres travelled on this road section. Matching of nameless police and ambulance records, when any of the crash dates, time, age, and gender details was available, decreased the overall estimates by 4 deaths and 73 injuries. Corrected traffic fatality rate was 53 deaths and injuries 467 per 109 vehicle kilometres travelled on this road section.