Estimation of the completeness of road traffic mortality data in Zambia using a three source capture recapture method

Highlights

• •A lack of reliable data on road traffic accidents in many low- and medium-income countries has led to an underestimation of the severity of the problem.
• •This study estimates the completeness of road traffic crash fatality data in Zambia.
• •Data on road traffic fatalities from the police, hospitals, and the civil registration and vital statistics databases was collected and analyzed using a three-source capture-recapture technique.
• •The capture-recapture technique estimated the completeness of police, hospital, and CRVS databases to be 19%, 12% and 14% respectively. A combination of the three data sets was found to increase completeness to 37%.
• •No single database contains complete data to provide a comprehensive picture of the burden of road traffic injury and we recommend that a combination of sources be used for the country’s official reporting.

Abstract

Background

Accurate and reliable data are essential for tracking progress and evaluating the effectiveness of road safety intervention measures. However, in many low- and medium-income countries, good quality data on road traffic crashes are often difficult to obtain. This situation has led to an underestimation of the severity of the problem and distortions in trends when the reporting changes over time. This study estimates the completeness of road traffic crash fatality data in Zambia.

Methods

Data from the police, hospitals, and the civil registration and vital statistics (CRVS) databases was collected for the period 1st January to 31st December 2020 and analyzed using a three-source capture-recapture technique.

Results

A total of 666 unique records on mortalities as a result of road traffic crashes were collected from the three data sources during the period under review. The capture-recapture technique estimated the completeness of police, hospital, and CRVS databases to be 19%, 11% and 14% respectively. The combination of the three data sets was found to increase completeness to 37%. Based on this completion rate, we estimate that the actual number of people who died as a result of road traffic crashes in Lusaka Province in the year 2020 was approximately 1,786 (95% CI [1,448–2,274]). This corresponds to an estimated mortality rate of around 53 deaths per 100,000 population.

Conclusions

There is no single database contains complete data to provide a comprehensive picture of Lusaka province and by extension the country’s road traffic injury burden. This study has shown how capture and recapture method can address this problem. It shows the need for the continuous review of the data collection processes and procedures in order to identify gaps and bottlenecks, improve efficiency, and increase the quality and completeness of road traffic data on injuries and fatalities. Based on the findings of this study, it is recommended that the city of Lusaka province and Zambia as a whole utilize more than one database for official reporting of road traffic fatalities to increase completeness.

1. Introduction

Globally, road traffic crashes (RTCs) are among the leading causes of death, injury, and disability. The 2018 Global Status Report on Road Safety, produced by the World Health Organization, estimates that 1.35 million lives are lost on our roads every year and over 50 million more are injured (WHO, 2018). Sub-Saharan African has the highest traffic fatalities rates in the world with 26.6 deaths per 100,000 population (Kazeem, 2019).

The collection and use of accurate and reliable data on road traffic fatalities is critical in the fight to reduce road traffic injuries and fatalities. Public health efforts to combat the problem of RTC in most African countries are hindered by the lack of reliable data to inform an effective response (Adeloye, 2012). Good road traffic data can also help build support among policy makers and other partners by; 1) describing the nature and gravity of the problem; 2) monitoring and evaluating the effectiveness of intervention measures and; 3) providing information on the social and economic costs of crashes and the value of life (WHO, 2010). Good traffic crash data is also required to track progress toward the UN’s SDG Target 3.6, which calls for reducing the number of global deaths and injuries from road traffic accidents by half, and Target 11.2, which calls for increased road safety and access to transportation systems, as well as the expansion of public transportation (Nneka and Namita, 2021).

Road traffic crash data collection and reporting systems in many countries on the African continent are inadequate and underdeveloped (Thomas et al., 2017). Under-reporting, aggregate reporting, inadequate specificity, inconsistency, inaccessibility, and data release delays are just a few of the major challenges with RTC data across the continent (Chang et al., 2020). There are huge discrepancies between the official road traffic fatalities as reported by many African countries and the estimated number of deaths as modelled by the WHO (WHO, 2018). Without complete and reliable information on when, where, and how crashes occur on roads, mitigation measures cannot be appropriately designed and implemented in areas where they are most needed to avoid future deaths. Monitoring data related to road traffic crashes necessitates a multi-sectoral approach involving collaborative efforts and participation from various stakeholders.

Various strategies have been suggested to improve the reporting of RTC fatality data and obtain better estimates of the burden of RTC. Among them are the use of mathematical models (Fisa et al., 2019, WHO, 2018), linking and merging of various sources (Short and Caulfield, 2016), and the capture-recapture method (Abegaz et al., 2014, Sango et al., 2012). Capture-recapture is a methodological approach that provides estimates of the outcome of an event based on cases commonly captured by different data sources (Tercero and Andersson, 2004). Originally, the method was developed to provide reliable estimates for wildlife populations (Abegaz et al., 2014, Ramos et al., 2020, Rongsheng et al., 2005). However, in recent years, it has been commonly utilized in various epidemiological studies (Laure and Stéphane, 2008, Rongsheng et al., 2005); and to estimate the prevalence of a variety of diseases, including HIV, stroke, tuberous sclerosis, cancer, diabetes, meningococcal disease, and inflammatory bowel disease (Morrison and Stone, 2000, Sekikawa et al., 1999, Tilling et al., 2001, Héraud-Bousquet et al., 2012).

The analysis of road traffic mortality data using the capture-recapture technique in low- and middle-income countries has provided a means to overcome the limitations of incomplete data sources. The methods have been used in Malawi (Samuel et al., 2012) and Ethiopia (Abegaz et al., 2014) to determine the prevalence of under-reporting using police and hospital records. These studies have helped to determine what future efforts are required to improve the reporting of road traffic deaths at the national level.

Official road traffic crash data in Zambia comes solely from Zambia Police Records (RTSA, 2020, RTSA, 2021). In 2020, the country recorded a total of 28,484 road traffic crashes, resulting in 1,690 deaths. However, WHO estimates that traffic fatalities in Zambia claim around 3,654 (World Health Organization, 2020) lives each year, which is much higher than the official national reported figure (Amend & FIA Foundation, 2019). An estimated 50% of fatalities are not captured in the official reporting of road traffic crashes in Zambia (Amend & FIA Foundation, 2019).

Due to inadequate capacity for response and follow-up on wounded victims and differing traffic fatality definitions, police data tends to underreport road traffic injuries and fatalities (Chand et al., 2021). Other factors which lead to underreporting highlighted in literature include; lack of reporting to relevant departments (Wu et al., 2016, Ahmed et al., 2017); lack of adherence to operational definitions used to classify road traffic injuries and deaths (WHO, 2010); unqualified or inexperienced staff, outdated, unfinished reporting, and incomplete statements (Odonkor et al., 2020) and; performance based administrative policies which may lead traffic police and road safety agencies to report fewer deaths and injuries to avoid criticism or attract reward (Alcorn, 2011, Ahmed et al., 2017). Many countries are using hospital and CRVS data to supplement the data collected by the police (Abdulhafedh, 2017). Other important sources of road traffic crash data include; the emergency services, local authorities, and insurance companies (PIARC).

Lusaka Province is Zambia’s most populous and densely inhabited province, with a projected population of 3.48 million people accounting for 18.94 percent of the country’s total population (Knoema). Lusaka City, Zambia’s capital, is located in Lusaka province, is the most urbanized city in the country and one of the fastest growing cities in Sub-Saharan Africa. Lusaka Province was chosen as the study area as it is home to almost half of the country’s motor vehicle population and has the highest rate of road traffic collisions (RTSA, 2021) and based on discussions with the national experts from the country regarding the population, the number of crashes and the number of deaths.

The main sources of road traffic fatality data in Zamba include the police, hospitals and CRVS registers compiled by the Department of National Registration, Passports and Citizenship (DNRPC). Officer from the Zambia Police Service (ZP) collect data from the scene of the crash and also from hospitals. Health facilities collect data on fatalities from various paper-based registers and from the health management information systems. By law, all deaths must be registered with the Registrar General’s Office within 14 days of the date of death (GRZ, 1964). This can be done by a relative of the deceased or a person present at the time of death. The registration process involves filling out a Death Registration Form and submitting it to the Registrar General’s Office, which is responsible for maintaining the CRVS system. The CRVS system records information about the deceased person, including their name, age, and cause of death, and this information is used to inform policy and programmatic responses to road traffic fatalities.

The objective of this study is to address this problem by estimating the completeness of road traffic crash fatality data by using a three-source capture-recapture method in Lusaka province. The study aims to share knowledge on our experience in utilizing this method for the first time ever in Zambia so that other researchers and road safety experts can undertake their own studies or to learn from the findings of this article. In addition, the article urges decision-makers to assess the extent of the problem of underreporting of traffic deaths using approaches other than the negative binomial regression employed in World Health Organization models (Iaych, 2011). This study is a follow-up to a study by Sichembe et al. (Sichembe et al., 2019) undertaken to investigate the epidemiology of RTCs in rural Zambia using hospital data, which recommended the use of capture-recapture on multiple data sources to determine the true burden of road trauma in the country.

2. Materials and methods

2.1. Data sources

The three sources of data for the study included police, hospital, and CRVS records. Road Traffic Fatality data was collected from records/books in all the police stations and posts with traffic sections in Lusaka Province. Surveillance was also conducted at the casualty and the mortuary sections of the 10 level one hospitals in Lusaka province. These hospitals provide comprehensive trauma care and treatment for RTC victims. Records from the Department of National Registration, Passports and Citizenship in the CRVS death register attributed to causes of death related to road traffic crashes was also collected and analyzed.

Only crashes that resulted in at least one fatality were enrolled in the study. A road traffic fatality in Zambia is defined as “a death occurring within 30 days of a road traffic crash, which is reported to the police and recorded in the police database” (RTSA, 2021). Structured checklists were prepared and used for the data collection process. Variables collected included the name of the victim, gender, age, time and place of the incident. Victims who share two or more key identifier variables in different data sources were considered a match. The study captured all crashes in police, CRVS, and hospital records for a period of one year from 1st January to 31st December 2020.

2.2. Data analysis

Capture-recapture is a methodological approach that provides estimates of the size of a population when it is not practical to count every individual (Tercero and Andersson, 2004). The technique was originally developed to provide reliable estimates for populations of wildlife species (Abegaz et al., 2014, Ramos et al., 2020, Rongsheng et al., 2005). In recent years, capture-recapture has been widely used in various epidemiological studies to estimate the prevalence of a variety of diseases, including HIV, stroke, tuberous sclerosis, cancer, diabetes, meningococcal disease, and inflammatory bowel disease (Morrison and Stone, 2000, Sekikawa et al., 1999, Tilling et al., 2001, Héraud-Bousquet et al., 2012). Various studies have also employed the technique to estimate the incidence of road traffic crash injuries and fatalities (Morrison and Stone, 2000, Samuel et al., 2012, Abegaz et al., 2014).

The analysis of data from the capture-recapture experiment consists of determining the best-fitting model and estimating the population size based on that model. The basic information required to perform a three-source capture-recapture analysis includes the number of cases identified by each source and their interactions. This information is captured in a set of capture histories of individually identified records. The capture histories are arranged and presented in a 2³ contingency table with 3 variables. We used a “1” to represent “identified” and “0” to represent “not identified”. The data is presented in Table 1 below.

Table 1.

Capture history contingency table.

Source 1 (Police)	Source 2 (Hospital)	Source 3 (CRVS)	Frequency
1	1	1	11
0	1	1	20
1	0	1	73
0	0	1	152
1	1	0	14
0	1	0	127
1	0	0	172
0	0	0	?

Eight log-linear models were generated and fitted to the data in the capture histories table (Table 1 above). The logarithm of the number of cases was used as the outcome variable for each model. The analysis was conducted using the “Rcapture” software package in R programming language (Rivest and Baillargeon, 2019). The model with the smallest Bayesian Information Criterion (BIC) was selected as the final model. The final model will provide an estimate of the number of deaths which are missed out of the three databases. Linear extrapolation will be used to estimate the number of road traffic deaths for the entire country.

3. Results

A total of 795 records were collected from at least one of the three data sources as traffic mortalities that occurred during the period 1st January to 31st December 2020. Hospital, CRVS and Police data identified 205, 256 and 334 fatalities, respectively. All three data sets had missing data (unknown or unidentified), which included victims’ name (N = 97), age (N = 145) and gender (N = 4). Most victims in all three data sources were males and aged between 20 and 30 years. Table 2 below gives a summary of the data collected by age group and gender and Table 3 presents the number of records collected from each source.

Table 2.

RTF by age group and by gender.

	CRVS		Hospital		Police		Total
	Count	%	Count	%	Count	%	Count	%
By Age Group
0–19	32	12.51%	21	9.30%	37	10.87%	90	11.65%
20–29	74	29.22%	47	25.56%	77	21.56%	198	25.97%
30–39	64	24.99%	39	17.71%	79	23.73%	182	24.01%
40–49	47	18.38%	20	7.39%	33	9.60%	100	14.19%
50–59	17	6.53%	6	1.39%	28	8.30%	51	6.88%
60–69	9	3.36%	5	1.86%	7	2.03%	21	2.74%
Above 70	4	1.62%	1	0.46%	3	1.11%	8	1.34%
Unknown Age	9	3.39%	66	36.32%	70	22.81%	145	13.22%
Grand Total	256	100%	205	100%	334	100%	795	100%
By Gender
Female	50	19.46%	56	29.28%	63	17.10%	169	19.19%
Male	206	80.54%	147	70.63%	268	82.15%	621	80.51%
Unknown Gender	0	0.00%	2	0.09%	3	0.75%	5	0.30%
Grand Total	256	100%	205	100%	334	100%	795	100%

Table 3.

Number and percentage of records collected from each database.

Data source	Number of records	%
Police Data	334	42%
CRVS Data	256	32%
Hospital Data	205	26%
Total (excluding duplicates)	666	84%
Total Number of records	795	100%

3.1. Comparison of data sources

A comparison of the three sources reveals that 11 records were common in all three databases. The matching process also identifies 236 records unique to the police database, 160 records unique to the hospital database, and 152 records were unique to the CRVS database. Fig. 1 below shows a Venn diagram which depicts the number of common and unique records in the databases from the three sources.

Fig. 1.

Venn diagram depicting the results of matching the three RTF databases.

3.2. Capture-recapture analysis

For the purposes of the capture-recapture analysis, 97 records that were unknown and unidentified were removed from the dataset as there was no means of matching them across the three databases. The unknown records were later added to the final estimates. The results of the log-linear modeling are presented in Table 3 below. The third model (PHC PC – with interactions between police and CRVS data) had the smallest BIC values and thus was selected to be the best-fitting model. According to these results, we estimate that around 1,689 (95% confidence interval [CI], 1,351 to 2,177) (Table 4) road traffic mortalities occurred in Lusaka during the period under review. To get the correct estimated figure, we added the number of unknowns or unidentified to the estimated figure, 1,689, received from Table 3 above. This means the estimated figure will be 1,786 with a 95% CI(1,448 – 2,274).

Table 4.

Results of the regression models.

Model No	Interactions	Poisson, multinomial estimation, and model fit, 95% profile likelihood confidence interval:
		$\widehat{\mathit{N}}$	stderr	deviance	df	AIC	BIC	infCL	supCL
1	PHC	1157.5	75.3	31.256	3	79.501	96.877	1023.9	1321.6
2	PHC PH	1026.5	67.2	18.735	2	68.98	90.699	908.5	1174.4
3	PHC PC	1689.4	205.1	2.661	2	52.906	74.626	1351	2176.6
4	PHC HC	1093.8	76.5	28.591	2	78.836	100.555	959.9	1263.1
5	PHC PH PC	1534.2	247	2.025	1	54.27	80.333	1156.9	2183.2
6	PHC PH HC	927.1	60.9	11.05	1	63.295	89.358	822.1	1063.8
7	PHC PC HC	2129.3	456.9	0.113	1	52.358	78.421	1465.5	3422.7
8	PHC PH PC HC	2355.8	887.8	0	0	54.245	84.652	1207.1	5218.1

Key:

PH – Police and Hospital.

PC – Police and CRVS.

HC – CRVS and Hospital.

PHC – Police, Hospital and CRVS.

The completeness of police and CRVS data on road traffic mortality is shown in Table 5 below. The completeness of police and CRVS data combined (excluding duplicates) is also calculated against the estimated number of road traffic mortalities calculated in Lusaka Province. The table shows that the estimated completeness of police data is around 19% with a 95% CI of 15% to 26%. Hospital data and CRVS data was estimated to be 12% and 15% complete, respectively. The results reveal that a combination of the three data sources resulted in an improved completeness rate of 37%.

Table 5.

Completeness of police, hospital and CRVS data of road traffic fatalities from capture recapture analysis.

Database	Count	Completeness	(95% CI, %)
Police Data	334	19%	(15, 23)
CRVS Data	256	14%	(11, 18)
Hospital Data	205	11%	(9, 14)
Total (excluding duplicates)	666	37%	(29, 46)

Based on the estimates of completeness calculated by the capture-recapture analysis above and the country’s population projections for the year 2020 (Central Statistical Office, 2013), we calculated the incidence of road traffic fatalities in Lusaka Province and in Zambia. The rates per 100,000 population are presented in Table 6 below.

Table 6.

Estimated number of road traffic deaths per 100,000 population.

	Projected population in 2020	Reported number of fatalities in 2020	Estimated number of fatalities	Estimated Number of deaths per 100,000 population
Lusaka	3,360,183	445	1,786	53.2
Zambia	17,885,422	1,690	6,783	37.9

4. Discussion

According to the 2020 Annual Road Transport and Safety Status Report (RTSA, 2021), Lusaka province reported a total of 15,871 road traffic crashes resulting in 445 deaths. These figures represent 56% of all the crashes recorded in the country and 26% of all the deaths. Capture-recapture analysis estimated that a total of 1,786 deaths occurred in Lusaka province as a result of RTCs. Based on this estimate, we calculate the number of road traffic deaths for Zambia by using a linear extrapolation. The number of road traffic deaths to the entire country, gives that around 6,783 (95% CI [5,499–8,636] lives were lost on our roads in the year 2020. This corresponds to an estimated mortality rate of around 53.2 deaths per 100,000 population for Lusaka and around 37.9 deaths per 100,000 population for the entire country.

The estimated number of yearly RTC fatalities from the capture-recapture technique employed in this study is much higher than the 3,654 deaths for Zambia estimated using statistical modelling techniques in the 2019 global health estimates (GHE) (World Health Organization, 2020). Low levels of completeness are however consistent with findings from other low- and middle-income countries. A two-source capture-recapture technique was used to estimate the number of road traffic deaths in Lilongwe (Samuel et al., 2012), Addis Ababa (Abegaz et al., 2014) and Bamako (Sango et al., 2012) yielding similar results.

High levels of under-reporting of road traffic crash fatality data can be explained by the countries’ reporting from only one source, reliance on paper-based systems, inconsistent and varying case definitions and inaccurate classification and coding of RTC deaths. The low completeness levels of road traffic crash fatality data makes it difficult to assess the effectiveness of RTC interventions and to analyse trends to be able to measure the countries’ progress towards achievement of both national and global road safety targets.

The authors note that the estimations of completeness presented in this study should be viewed in light of some limitations. Hospitals do not collect data on the location of crashes, so it is likely that some records found in the hospital databases were from RTCs that occurred outside of Lusaka province then subsequently rushed or transferred to a health facility in Lusaka. In addition, a significant number of entries lacked information, such as name, age, and gender, which affected the process of matching records from one database to another. Another potential bias to consider when interpreting the study results is the possibility of higher levels of under-reporting in less developed provinces compared to Lusaka province, which may be attributed to the linear extrapolation technique used to estimate the number of crashes reported in the entire country. Further, the data were gathered during the COVID-19 period, which may have led to distortions in the number and patterns of RTCs compared to prior years.

5. Conclusion

This study estimated the level of under-reporting of road traffic crash fatality data in the province of Lusaka and extends this findings to the whole of Zambia. Data on road traffic crash fatalities from three databases which included, the Zambia police service, hospitals and the Department of National Registration, Passports and Citizenship was collected and compared to estimate the level of completeness of police data using a matching algorithm and a three-source capture-recapture technique.

Comparing and matching the three sources revealed that no single database provides enough information to give a complete picture of the burden of road traffic injuries in the country. The study found that the actual number of traffic deaths is far greater than the official data indicates and that there is need to strengthen the country’s RTC data systems in order to accurately address the problem. A thorough review of data collection processes and procedures is required to identify gaps and bottlenecks, improve efficiency, and improve the quality and completeness of road traffic data on injuries and fatalities.

The study recommends that the country utilize more than one database for official reporting of road traffic fatalities. A multi-sectorial team should be put together and tasked with the responsibility of compiling and analyzing road traffic crash data from multiple-department data sources. This data linkage is valuable for the validation of data quality against each source and provides comprehensive information concerning road traffic injuries and deaths.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.