O International Epidemiologies] Association 1998

Printed In Great Britain

International Journal of Epidemiology 1998:27:866-570

Estimating deaths and injuries due to road traffic accidents in Karachi, Pakistan, through the capture-recapture method
Junaid A Razzak a ' b and Stephen P
Background Road traffic accidents (RTA) are an important yet preventable cause of death and disability in developing countries like Pakistan. Yet accurate epidemiological data on injuries in developing country injuries is often difficult to obtain. We applied the capture-recapture method to estimate the death and injury rates due to RTA in Karachi. Methods We applied the two-sample capture-recapture method using traffic police records as one source of capture and the logs of a non-government ambulance service as the second capture source for the same 10 months and 20 days for which 1994 data were available. We generated a conservative adjusted estimate of injuries and deaths by considering entries in the two sources as matched if they reported the same date, time, and place, and at least one of the other matching variables, of name, vehicle registration number, vehicle types or patient outcome. We then compared the estimated rates with the police rates. Results In 1994 police reported 544 deaths and 793 injuries due to RTA while ambulance records noted 343 deaths and 2048 injuries. The capture-recapture analysis estimated at least 972 (95% CI : 912-1031) deaths and 18 936 (95% CI : 15 50722 342) injuries attributable to RTA during the study period. Official sources counted only 56% of deaths and 4% of serious injuries. The estimated rates for the year 1994 were 185 injuries and 11.2 deaths per 100 000 population. Conclusion Road traffic injuries and deaths in Karachi are a much more substantial health problem than is evident from official statistics. Keywords Road traffic accidents, Karachi, Pakistan, capture-recapture method, official data, ambulance data Accepted 10 December 1997 Road traffic accidents (RTA) are a significant,1'2 but preventable,34 cause of death, disability and economic loss in developing countries. Among males of the economically active age group, motor vehicle injuries are the third most important cause of death in these countries.5 It is projected that RTA will be the second most common cause of disability-adjusted life years in developing countries in the year 2020.6 Despite this, there is little recognition of the health and economic burden of this problem.7 The World Health Organization (WHO) in its international conference on RTA noted the importance of adequate data on traffic injuries.8 Indeed, accurate estimates of the public health burden of RTA can establish the priority of
this public health problem, and provide a rational basis for policy decisions. Yet, accurate epidemiological data from many of the developing countries are difficult to find in the literature. 9 Injury data from developing countries is primarily based upon either hospital logs ' ' ' or is derived from police records. 12 " 16 It is believed that both these sources underestimate the total burden of the injuries. Hospitals tend to see more severe forms of injuries only.1 Furthermore, published hospital series are unlikely to reflect all those who seek health care following RTA. In Karachi, Pakistan, for example, it is difficult to find out the total number of hospitals, let alone the RTA related injuries and deaths seen by each of them. Rates based on figures from a few of these hospitals would be expected to underestimate even the severe forms of injuries in the community. Similarly, studies which simply compared the police data with hospital data have consistently found that police data underestimate RTA15"17 ranging from 60% underestimation in Al Ain City17 to 80% underestimation in Saudi Arabia. 18 Since none of these

Downloaded from http://ije.oxfordjournals.org/ by guest on February 14, 2012

a Department of Community Health Sciences, The Aga Khan University, Karachi, Pakistan, b Current address: Section of Emergency Medicine, Yale New Haven Hospital New Haven, CT, USA Reprint requests to: Junaid A Razzak, 437 Meloy Road, #B1. West Haven, CT 06512, USA.

866

CAPTURE-RECAPTURE FOR ESTIMATION OF RTA DEATH AND INJURY compared police records with the actual rates in the community it is likely that they have underestimated the underreporting by the police. While accurate death and injury rates would best be generated through community-based studies, such studies are expensive and therefore difficult especially for a developing country like Pakistan. The capture-recapture method has been used extensively in the biological sciences19 and medicine for estimating difficult to count populations. 20 ' 21 It has been used in many different settings. 22 " 24 The method is based on matching two independent samples to arrive at an estimate of the total. In this study, the capture-recapture method was applied using combined records from police sources, and a large ambulance service in the city to estimate more accurately the death and injury rates due to RTA in Karachi, Pakistan.

867

Methods
Karachi is the largest dry of Pakistan, and one of the seven largest dties in the developing world with a population of 9.9 million, 25 859 459 registered vehides 26 and 245 km of roads. 27 Police records are the main source of information regarding RTA in the dty. Some of the acddents are registered by the police at the time, as is required by offidal policy when there is loss or damage to life or property. However, most are reported to the police through the medicolegal officer based in five government run hospitals, where presumably most of the trauma patients go for treatment. Although it is a legal requirement to report all motor vehicle collisions resulting in injuries, it has been observed that only a few cases are reported formally. One reason is that people would rather pay the modest bribe to the police rather than get involved in a tedious and often frustrating legal process for which they receive no apparent benefit. 28 ' 29 All cases of offidally recorded acddents are aggregated at the traffic police headquarters and a daily report is made, giving details of the traffic acddents reported during the last 24 hours. Information on all the offidally reported RTA are sent daily to the Traffic Engineering Bureau, a subsidiary of local government. They then analyse each event with the objective of determining the cause of the RTA and make recommendations to decrease their inddence. A yearly report is then published after an exhaustive analysis of the available data. Another source of information on RTA related injuries and acddents in Karadii is the Edhi Ambulance Service (EAS). This is a private philanthropic institution that provides, among other services, ambulances to transport the sick. We selected EAS as it is by far the largest in the dty; operating 35 (77%) of the 45 ambulance dispatch centres in Karachi. Edhi Ambulance Service covers the entire metropolitan area as its dispatch centres, supported by 75 working ambulances and 150 drivers , . , , , . _, , , , are spread evenly throughout the aty. The other ambulance services have centres only in front of hospitals and none of them have more than two centres each or more than 10% of ' the total ambulances. These other ambulance services are used primarily by patients of the nearby hospital for transport to f i_i_ 1 ^ 1 1 r_ J home or to another hospital. Typically, just after an accident has occurred, the EAS control tower or the regional centre is informed by one of the witnesses of the acddent. The injured or dead are picked up by the ambulance and are transported to the

nearest hospital. Information regarding the name and age of the victim, the time of accident, the place of accident and vehide type and registration number is noted. The National Identity Card, which is carried by most city dwellers is used in cases where victims are unable to give their name and age. Once the patient has been transported to the hospital, the ambulance driver reports the information to the Edhi information bureau which records all such information in a log. We used a two sample capture-recapture method. The first capture was all deaths and injuries due to RTA reported by the traffic police in their daily report. The recapture was the EAS logs. Cases were matched for the year 1994 except for 42 days (25 March-5 May) for which the ambulance log was missing. Each injury case in the police data was matched with EAS data on seven variables: the name of the victim, date, time and place of the accident, the types and the registration number of the vehide involved, and the outcome (alive or dead). The name of the victim in the ambulance service log was in Urdu, the national language of Pakistan, while police reported it in English. The names were considered matched if the pronundation was similar in both languages. The street system in Karachi is poorly organized. Street names are hard to find and numerical addresses rarely available. Many major roads of the dty have more than one name. Places in Karachi are often identified by a landmark like a well known shop or a mosque or other building. If the study reviewers, who were knowledgeable about the dty, judged the place similar in both the capture and recapture, it was considered matched. Time' was considered matched if the recorded time from the two sources was within one hour of each other. Date, type of vehide and registration number of vehicle were considered as matched if they were exactly the same in both the capture and recapture, The degree of matching was then defined based on four different standards (A-D). Each of the standards required date and time to be the same, and place to be similar. Standard A, the strictest standard required the name of the victim, the date, time and place of the event, the type and registration number of vehides involved and the patient outcome (survived or died) to be matched. For standards B-D the criteria were progressively slackened, so that each subsequent standard required one less criteria for a match (Table 1). In some cases one or both data sets did not have the name of the victim and registration number of vehide. If one of the data sets had no information on one of the variables while the other one had it, the variable was considered unmatched on that variable,

Downloaded from http://ije.oxfordjournals.org/ by guest on February 14, 2012

l i b l e x The four

&11"1 standards used to define a match between ambulance and police records of road traffic accidents, Karachi, Pakistan 1994 Name of victim, registration n o . vehicles, outcome All same
three of th e a t

. . Standard Date
A B C

, Place : Same Same Similar

_ Time

. P.

Same

. .*W. .. . .. . ?Y e . sam .^ A11?.tw0.of * e at>ove s a m e Same Same Similar Any one of the above same .
Same Similai

Similar

868

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY the two groups identified 220 matches, yielding an estimate of 963 deaths/year (95% Cl: 912-1031). Using more restrictive standards for matching resulted in progressively larger estimates of injuries and deaths reaching as high as 47 140 injuries and 3116 deaths using the most restrictive matching standard, standard A (Table 2). The incidence of deaths due to RTA using the capturerecapture method was at least 963 deaths which is equivalent to 9.72 deaths/100 000 population, 11.3 deaths/10 000 vehicles and 3.9 deaths/km of road. Estimated rates for serious injuries due to RTA were 190 injuries/100 000 population, 218 injuries/ 10 000 vehicles, and 77 injuries/km of road. The case fatality rate was 5.1.

An estimate of the total number of RTA victims was made with the formula (cl + I)(c2 -1 Estimation of n = h = m +1 Where cl and c2 are the numbers of people in the first and second capture respectively, and m is the number of people identified in both captures (matches). The following formulae30 were used to calculate the variance and 95% confidence interval (Cl) for the estimate of n:
Variance =
95% Cl = n1.96VVariance

(c, + I)(c2 +

l)(cl-m)(c2-m)

(m + I)2 (m + 2)

Rates and Cl for the whole year were calculated by extrapolating the data of 10 months 20 days to the whole year to calculate the annual incidence. This was done by first dividing the estimated number with 323 (i.e. total number of days for which data was available) and then multiplying this per day estimate by 365. Rates were calculated using 9.9 million as the total Karachi's population26 and 859 459 registered vehicles in Karachi.27 The length of roads in Karachi was taken from the Sindh Bureau of Statistic report 1994. 28

Discussion
Even with the least restrictive (most sensitive) matching criteria, criteria D (requiring a matching date, time and place, with one of the following: name of victim, registration number, vehide or outcome matched) the serious injury rates from road traffic injuries in Karachi were 21 times higher than what is offidally reported by the police. Using more strict (more spedfic) criteria increased the difference even further. Since most of the data collection reporting process both by the police force and ambulance drivers was by word of mouth, there was a high likelihood of mismatch on recall intensive data like name of the victim, registration number and the type of vehides. Thus, the least restrictive matching standard, standard D, would likely be most appropriate. These rate estimates would thus be conservative, but would be quite unlikely to falsely overestimate rates. Even these most conservative estimates, however, show that the public health importance of road traffic injuries in Karachi is markedly understated by offidal statistics. The rates for RTA related injuries estimated by the capture-recapture technique are at least 190/100 000 population compared to 9.0 and 20.9 injuries/100 000 population using the Police or the EAS data separately. Offidal sources thus counted only 56% of deaths and 4% of serious injuries. Indeed, the total number of injuries for this one dty, Karachi, is greater than the offidal estimate of the road traffic injuries for the whole country, which has over ten times Karachi's population.23 The death rate of 9.7/100 000 population, on the other hand is 179% higher than the
Downloaded from http://ije.oxfordjournals.org/ by guest on February 14, 2012

Results
Records from traffic police identified 544 deaths and 793 injuries from RTA in Karachi during the 10 month 20 day study period, while the ambulance services data identified 343 deaths and 2048 injuries for the same time period. Most of the officially, as well as unofficially, reported victims were males (93% and 89% for deaths and 88% and 86% for injuries respectively). The most frequent victim in both sources were pedestrians (35% and 30%) and motorcyclists (19% and 20%). The date, time and place of the accident was recorded in all of the cases, while vehicles Involved were unknown in 1.4% of cases, name of victim was not mentioned in 10% of cases, and vehicle registration number was not known in 1.5% of cases. Using the least strict standard for matching, standard D (when the time, date and place matched along with any one of the other four variables i.e. name of the victim, registration number of vehide, type of vehicle and patient outcome), comparison of

Table 2 Estimation of deaths and injuries due to road traffic accidents in Karachi for 1994 using the capture-recapture method Numbers unmatched in police data
476 754 405 725 335 701 324 695

Match type
A B

Outcome Deaths Injuries Deaths Injuries Deaths Injuries Deaths Injuries

Number of matches
68 39 139 68 209 92 220 98

Number unmatched InEdhl data

Estimated number of total deaths /injuries 3116 47 140 1524 27 036 1013 19 983
963

Estimated rate per 100 000 persons per year 31.5

95% confidence Interval for rate 25.5-38 335-626 13.8-17.2 214-337 10.1-11.0 166-242 9.2-10.4 157-226

2009
204

15.4
273

1980
134

10.3
202 9.7 189

1956
123

1950

18 760

CAPTURE-RECAPTURE FOR ESTIMATION OF RTA DEATH AND INJURY 5.49/100 000 reported by police and 234% higher than the unofficial figure of 3.46 reported by EAS. Compared to other cities the death rates for Karachi are much higher than cities in developed countries. For example, while 11.3 people die per 10 000 vehicles registered in Karachi, 1.4 people die in Tokyo, and 2.8 people die in Greater Manchester (UK) for a similar number of vehicles. The rates are however, comparable to developing world cities like Bangkok, Thailand with rates of 10 deaths/10 000 vehicles. Four major assumptions 18 ' 32 must be satisfied for the capturerecapture methodology to produce reliable results. The first assumption is that all the members of the population have the same probability of being captured. All the accidents occurring anywhere in the city were accessible to the extensive networks of both the traffic police and the EAS. Both institutions have good wireless/radio communication systems and both respond to all the RTA which are brought to their notice. As there is no area of the city that was apparently uncovered by either of the systems, all serious RTA In the city have an apparently equal chance of either being taken care of by police or transported by EAS to the hospital or both. The second assumption is that the capture sources should be independent. This is a difficult assumption to fulfil in most studies which employ capture-recapture methodology.26 In our study too it will be the more serious injuries which will require ambulance transport and will go to a major government hospital casualty department where there is a medicolegal officer on duty, so there is some degree of positive dependence. Any positive dependence, however, will lead to an overestimate of the number of matches, and so an underestimate of the total number of injuries, thus calculating only the serious forms of injuries. This may be one of the reasons for a higher fatality rate of 5.13% as compared to 1.7% for other developing countries. Indeed, if we assume that the case fatality rate of Karachi is similar to other developing countries the number of serious injuries represented by the number of deaths noted in the analysis would be three times larger. The third assumption using capture-recapture methods is that the population should be dosed. This assumption ensures that there are no major changes in the total population between the time period between the two captures. In our study, since the capture and the recapture was taking place at the same time, there was no chance of any change in the population. The fourth assumption is that the capture history of all the cases should be accurate. We assume that because of basic legal requirements, police tend to record the date, time, place, vehicle involved and name of the victim much more accurately as opposed to the actual event and burden of mistake. The information on EAS logs are first hand information from the ambulance driver himself who is the actual witness at the site of accident, which in itself reduces the chances of inaccurate reporting. Moreover, to address the difference in the language of the names and inconsistency in the addresses we used loose criteria for matching. Importantly, for this study, RTA fatality was determined at the time of transport for EAS or the time of emergency room discharge for the police data. Both of these approaches underestimate fatality, and again emphasizes that the mortality estimates provided by this analysis are minimum estimates.

869

By using the capture-recapture method an estimate of the burden of ill health can be made especially in developing countries where there is no working system to accurately record the number of injuries and deaths. Pakistan, like other developing countries, is likely to have an increase in traffic which can lead to substantially increased deaths and disability unless efforts are made to understand the problem and steps taken now.

References
Smith GS, Barss PG. Intentional injuries In developing countries: the epidemiology of a neglected problem. Epidemiol Rev 1991; 13:228-66.
2

Murry CJL, Lopez AD. Mortality by cause for eight regions of the world: Global Burden of Disease Study. Lancet 1997;349t 1269-76. VTano DC. Cause and control of automotive trauma. Bull NY Acad Med 1988:64:376-421. Sosin DM, Sacks JJ, Holmgreen P. Head injury-associated deaths from motor cycle crashes. JAMA 1990;264:2395-99. Soderlund N, Zwi AB. Traffic related mortality in industrialized and less developed countries. BuU World Health Organ 1995;73:175-82.

Downloaded from http://ije.oxfordjournals.org/ by guest on February 14, 2012

'Murray CJL, Lopez AD. Alternative projections of mortality and disability by cause 1990-2020: Global Burden of Disease Study. Lancet 1997,349:1498-504.
7

Zwi AB, Forjouh S, Murugusamphillay S et al. Injuries in developing countries: policy response needed now. Trans R Soc Trop Med Hyg 1996;9O:593-95. Road Traffic Accidents in Developing Countries. Report of a WHO Meeting. Mexico City. 9-13 November 1981. Mohan D. Accidental death and disability in India: A stocktaking. Acad Anal Prev 1984; 16:279-88. Balogun JA, Abereoje OK. Pattern of road traffic accident cases in a Nigerian University teaching hospital between 1987 and 1990. J Trop Med Hyg 1992,95:23-29.

10

' Weddell JM, McDougall A. Road traffic injuries in Sharjah. Int J Epidemiol 1981:10:155-59. Asogwa SE. Road traffic accidents in Nigeria: A review and a reappraisal. Accid Anal Prev 1992;24:149-55. Oluwasanml AJ. Road accidents trends in Nigeria. Accid Anal Prev 1993:25:48 5-87. Jayasuriya R. Trends in the epidemiology of injuries due to road traffic accidents In Papua New Guinea. Asia Pac J Public Health 1991;5:41-48.

l2

13

14

"Bangdiwala SI, Anzola-Perez E. Traffic accidents as a serious health problem in selected developing countries of the Americas. Bull Pan Am Health Organ 1972; 1:38-47.
16

Mohan D. An analysis of road traffic fatalities in Delhi India. Accid Anal Prev 1985,17:33-45. Bener A, Achan NV, Sankaran-Kutty M. Road traffic injuries in Al-Ain City, United Arab Emirates. J Roy Soc Health 1992:112:273-76. Ofosu JB, Abouammouh AM, Bener A. A study of road traffic aeddents In Saudi Arabia. Accid Anal Prev 1988:20:95-101. Cook LM, Brower LP, Croze HJ. The accuracy of population estimation from multiple recapture data. J Animal Ecol 1967;36:494-95. Mastro TD, Kitayapom D, Weniger BG et al. Estimating the number of HTV-lnfected injection drug users In Bangkok: A capture-recapture method. Am J Public Health 1994,84:1094-99. Sutter RW, Cochi SL. Pertussis hospltalization and mortality In the United States, 1985-1988: evaluation of the completeness of national reporting. JAMA 1992;267:386-91. Hartnoll R, Lewis R, Mitcheson M, Bryer S. Estimating the prevalence of opioid dependence. Lancet 1985:1:203-05.

17

18

19

20

21

22

870

INTERNATIONAL J O U R N A L OF EPIDEMIOLOGY

23

van Haastrecht HJA, van den Hoek JAR, Bardoux C, LeentvaarKuyprs A, Coutinho RA. The course of the HTV epidemic among intravenous drug users in Amsterdam, the Netherlands. Am J Public Health 1991;81:59-62. Frischer M. Estimating prevalence of injecting drug use in Glasgow. Br J Addict 1992,20:997-1000. The Challenge of Urbanization: The World's Large Cities. United Nations. New York, 1995. Number of registered vehicles in Karachi in 1994. Traffic Engineering Bureau, Karachi Development Authority. Pakistan Statistical Yearbook 1995. Government of Pakistan, Islamabad. Staff Reporter. Transporters accuse traffic police of harassment. Daily Dawn, 9 September 1997.

29

Azizullah Sharif. Traffic cops expert at harassing motorists. Daily Dawn, 22 September 1997 McCarty DJ, Tull ES, Moy CS, Kwho CK, LaPorte RE. Ascertainment corrected rates: Application of capture recapture methods. Int J Epidemiol 1993:22:559-65. Jacobs GB, Sayer IA. Road Accidents in Developing CountriesUrban Problems and Remedial Measures. Transport and Road Research Laboratory Supplementary Report 1984, p.839. Wittes JT, Sidel VW. A generalization of the simple capture-recapture model with application to epidemiological research. J Chron Dis 1968:21:287-301. AM J, Naraynsingh V. Potential impact of advanced trauma life support (ATLS) program in a third world country. intSurg 1987;72:179-84.

30

24

31

25

26

32

27 28

33

Downloaded from http://ije.oxfordjournals.org/ by guest on February 14, 2012