Injury-related mortality in South Africa: a retrospective descriptive study of postmortem investigations

Abstract Objective To investigate injury-related mortality in South Africa using a nationally representative sample and compare the results with previous estimates. Methods We conducted a retrospective descriptive study of medico-legal postmortem investigation data from mortuaries using a multistage random sample, stratified by urban and non-urban areas and mortuary size. We calculated age-specific and age-standardized mortality rates for external causes of death. Findings Postmortem reports revealed 52 493 injury-related deaths in 2009 (95% confidence interval, CI: 46 930–58 057). Almost half (25 499) were intentionally inflicted. Age-standardized mortality rates per 100 000 population were as follows: all injuries: 109.0 (95% CI: 97.1–121.0); homicide 38.4 (95% CI: 33.8–43.0; suicide 13.4 (95% CI: 11.6–15.2) and road-traffic injury 36.1 (95% CI: 30.9–41.3). Using postmortem reports, we found more than three times as many deaths from homicide and road-traffic injury than had been recorded by vital registration for this period. The homicide rate was similar to the estimate for South Africa from a global analysis, but road-traffic and suicide rates were almost fourfold higher. Conclusion This is the first nationally representative sample of injury-related mortality in South Africa. It provides more accurate estimates and cause-specific profiles that are not available from other sources.


Introduction
In South Africa in the year 2000, injury-related mortality accounted for 12% of deaths and 16% of years of life lost. 1 This was primarily due to high mortality rates from road-traffic injury and homicide, which were approximately twice and eight times the global average, respectively. 1,2 A previous South African national study of the burden of injury-related mortality used triangulation and modelling techniques 2 to overcome deficiencies in vital registration data and national statistics, such as underreporting 3,4 and the urban bias of national injury mortality surveillance. 5,6 These surveillance data are no longer suitable for burden of disease modelling. They are not nationally representative, since they are only available for two of nine provinces, and use mortuary registers rather than postmortem reports. 7 For deaths of undetermined cause, mortuary registers fail to differentiate routinely between deaths from natural or external causes and, for external-cause deaths, between accidental and deliberate events.
Under the Inquests Act of 1959, 8 postmortem investigations are a statutory requirement for all deaths that are not clearly from natural causes. This is a potentially useful alternative source of data on injury-related mortality. Here, we use postmortem records to provide a more accurate cause-specific profile of injury-related mortality in South Africa for the year 2009. This enables comparison with data from several sources including official statistics, the national survey of female homicides 9 and global burden of disease estimates. The study was commissioned by the South African Medical Research Council as part of its second national burden of disease study. 10

Methods
We conducted a retrospective descriptive study, using routine data collected through postmortem investigations during 2009. Data were obtained from postmortem reports and ancillary documentation, including police reports and hospital records. A multistage stratified cluster sample was drawn for eight provinces, using mortuaries as the primary sampling unit. A sampling frame of 57 274 postmortem reports from 106 mortuaries was used to draw a representative sample stratified by metro and non-metro area and mortuary size (stratified as less than 500, 501-1500, and more than 1500 cases). Forty-five mortuaries were selected with an expected sample of 22 733 records. All records for the Western Cape were obtained from the Provincial Injury Mortality Surveillance System 11 to complete the national sample. We assessed whether each death was from natural, external or undetermined cause. Field workers recorded the date of death, circumstances of death and the apparent manner of death (homicide, suicide, transport-related, or other unintentional or undetermined intent) consistent with the 10th revision of the International Statistical Classification of Diseases and Related Health Problems, 2007 (ICD-10; Table 1).
We excluded deaths from natural causes, fetal deaths and deaths that occurred outside South Africa. To account for the selection probabilities of mortuaries within survey strata, we applied analysis weights. Cases with unknown-age were proportionally redistributed to all other ages using a scaling factor. Age-standardized mortality rates were calculated for manner of death by age, sex, race, metro and non-metro Objective To investigate injury-related mortality in South Africa using a nationally representative sample and compare the results with previous estimates. Methods We conducted a retrospective descriptive study of medico-legal postmortem investigation data from mortuaries using a multistage random sample, stratified by urban and non-urban areas and mortuary size. We calculated age-specific and age-standardized mortality rates for external causes of death. Findings Postmortem reports revealed 52 493 injury-related deaths in 2009 (95% confidence interval, CI: 46 930-58 057). Almost half (25 499) were intentionally inflicted. Age-standardized mortality rates per 100 000 population were as follows: all injuries: 109.0 (95% CI: 97.1-121.0); homicide 38.4 (95% CI: 33.8-43.0; suicide 13.4 (95% CI: 11.6-15.2) and road-traffic injury 36.1 (95% CI: 30.9-41.3). Using postmortem reports, we found more than three times as many deaths from homicide and road-traffic injury than had been recorded by vital registration for this period. The homicide rate was similar to the estimate for South Africa from a global analysis, but road-traffic and suicide rates were almost fourfold higher. Conclusion This is the first nationally representative sample of injury-related mortality in South Africa. It provides more accurate estimates and cause-specific profiles that are not available from other sources.
Injury-related mortality in South Africa Richard Matzopoulos et al.
area using alternate mid-year population estimates 12 and the World Health Organization's (WHO) world standard population. 13 We recruited field workers and tested them for their ability to extract data from records. Field workers used a mobile phone based questionnaire to collect demographic information from the postmortem report, including age, sex and race of the deceased. Postmortem and police reports categorize individuals by the races black, coloured, Asian and white, and we kept those categories when conducting the study. We also recorded whether each death was related to a legal intervention, occurred in custody or if there was evidence of sexual assault. The mortuary death register number and the death notification number were collected as identifiers for follow-up to resolve data capture errors. The data captured on the mobile phone questionnaire application (Mobenzi Researcher, Cape Town, South Africa) were submitted to a central web-based platform. The project manager and biostatistician conducted quality checks while data were collected and resolved any data quality problems with the national level coordinator. Interobserver reliability was tested by two fieldworkers collecting data independently from the same folder on the same day for 5% of the sample. Reliability was high for cause of death (Κ = 0.86; 95% confidence interval, CI: 0.84-0.88), age (Κ = 0.95; 95% CI: 0.93-0.98) and sex (Κ = 0.94; 95% CI: 0.92-0. 97). Further details are available from corresponding author.

Ethics
The South African Medical Research Council's Health Research Ethics Committee approved the study.

Results
A total of 22 583 cases were drawn from the eight provinces -more than 99% of the expected total of 22 733. The discrepancy arose from invalid entries that had been included in the sampling frame -such as deaths that occurred before 2009 -and a small number of cases not recorded in mortuary registers or lacking records. A further 9418 cases were appended from the Western Cape database, providing a total unweighted data set of 32 001 records. After the application of sampling weights it was estimated that a total of 66 693 (95% CI: 60 356-73 030) deaths were processed by the forensic pathology service in 2009, of which 52 493 (95% CI: 46 930-58 057), or 78.7%, were from external causes (Fig. 1) 95% CI: 58.9-75.8) was significantly higher than the female rate (11.3; 95% CI: 9.5-13.0).The male-to-female ratio of homicide (6 male deaths per female death) was higher than for any other apparent manner of death. This was due to the particularly high rate ratios for three major external causes of death that were attributed to homicide: sharp force injuries/stabbing, gunshot injuries and blunt-force injuries ( Table 2).
The suicide rate of 13.4 per 100 000 population (95% CI: 11.6-15.2) was approximately one third of the homicide rate. Males were again over represented and overall there were five male suicides for every female suicide.   The numbers for total deaths exceed the sum of deaths by sex due to 145 deaths in which sex could not be determined. (. . .continued)

Research
Injury-related mortality in South Africa Richard Matzopoulos et al.
Transport-related injuries accounted for more than one-third of all external causes of death 33.8% (17 742/52 493; 95% CI: 31.9-35.7) and the majority of deaths due to unintentional injury. Most of the transport-related deaths were due to road traffic injuries, which represented 32.6% (17 103/52 493) of all injury-related deaths (95% CI: 30.7-34.5) or 36.1 per 100 000 population (95% CI: 30.9-41.3). Pedestrian deaths accounted for 40.0% (5604/14 010) of the road traffic deaths in which the road user group was defined. Drivers and passengers accounted for 22.9% (3205/14 010) and 32.6% (4572/14 010) of deaths respectively ( Table 2). The other major causes of unintentional injuries included burns, drowning and falls. The male-tofemale mortality ratios were lower for accidental than for intentional injuries.
For a small subset of deaths from external causes (4.0%; 2099/52 493) it was not possible for the medical examiners to determine intent. This was most common among deaths arising from the ingestion of poisonous substances (including drugs), deaths from fires, burns and hot substances, and instances where decomposed bodies, bones or skeletons were found. Fig. 2 shows the distribution of male and female fatalities.
Gunshot injuries were a leading cause across several categories, accounting for 6428 deaths, equivalent to 17.6 firearm-related deaths per day (95% CI: 15.7-19.6). Of these, 5513 were homicides, 780 were suicides, 48 were unintentional and 87 were deaths of undetermined intent. Table 3 presents the metro and non-metro mortality rates for homicide, suicide and road traffic injuries by sex, age and race. For all injuries and for homicide, metro mortality rates were notably higher than for non-metro areas. This finding was consistent when the data were stratified by sex and age.
Male injury rates were consistently and significantly higher than female rates in both metro and non-metro areas, with the highest male-to-female mortality ratio presenting among metro homicides (7 male deaths for every female death) and non-metro suicides (4.6 male deaths for every female). For road traffic fatalities, there were 3.6 male deaths for every female.
The age pattern for homicide was similar across metro and non-metro areas, albeit with the metro rates being noticeably higher in all but the youngest age category. Homicide rates were highest among teenagers and young adults in the 15-29 years age group in metro areas, and the 30-44 years age group in non-metro areas. Within age strata, the differences between metro and non-metro areas were not significant except for teenagers and young adults (15-29 years age group) among whom metro homicide rates were significantly higher, subjecting them to twice the risk of their non-metro counterparts. Suicide rates by age followed a similar pattern to homicide rates. Fig. 3 shows the homicide, suicide and road-traffic injury rates by race for metro and non-metro areas. The metro and non-metro homicide and suicide patterns by race were inconsistent. People categorized as coloured experienced the highest homicide risk overall. Homicide rates among people categorized as black were highest in metro areas; conversely, among people categorized as white, the highest rates were found in non-metro areas.
Road traffic mortality rates were higher in non-metro areas for people categorized as black, children and the elderly.
Differences in metro/non-metro mortality risk affected the provincial mortality profiles depicted in Fig. 4. Ho-micide ranked highest for five provinces: Eastern Cape, Gauteng, KwaZulu-Natal, the Northern Cape and the Western Cape, of which all except the Northern Cape have large urban centres.

Discussion
This study provides a comprehensive profile of injury-related mortality in South Africa in the year 2009 and provides cause-specific rates that are not available from other sources. Vital registration data suggest that there has been an overall decrease in external causes of death from 1997 to 2009, 14-16 but the reasons are unclear. In those data, a high proportion of deaths were recorded as "other external causes" of accidental injury -63% (31 166/49 456) in 2009. 15 The ICD-10 convention to code injury deaths with limited information on intent as accidental (X59) is a common cause of information loss for injuryrelated mortality data internationally. 17 We found more than three times as many deaths from homicide and roadtraffic injury than were recorded by vital registration. 15 Underreporting was also apparent in other official statistics. We recorded significantly more homicides (13% higher) than the 16 834 recorded by the South African Police Service in   2009 and significantly more road deaths (24% higher) than the 13 802 recorded by the Road Traffic Management Corporation. 18,19 The estimated total number of injury-related deaths in our study did not differ significantly from the 49 456 deaths from external causes recorded in vital registration, 15 but better causespecific detail is required for modelling of burden of disease trends. Previously, the first South African burden of disease study 2 provided the only detailed national estimates by age and sex for major causes of injury. There has been an overall decrease in total injury-related deaths from an estimated 59 935 in the year 2000. 1,2 The homicide rate of 38.4 per 100 000 population still places South Africa among the most violent countries, but there has been a significant decrease since 2000 when the national homicide rate was estimated at 64.8 per 100 000 population. This is consistent with national police statistics 18 and retrospective national surveys of female homicides that also indicate a decrease. 9 According to police statistics, homicide decreased from 18 793 to 15 609 between 2004 and 2011 18 . Two nationally representative retrospective surveys measured a 38% decrease (3793 to 2363 deaths) in female homicide between 1999 and 2009. 9 The decrease in female homicide has been attributed partly to the effectiveness of the Firearms Control Act of 2000. 9 A recent analysis of homicide across five South African cities from 2001 to 2005 confirms a substantial year-on-year decrease in homicides involving firearms, coinciding with the implementation of the Firearms Control Act, alongside a more modest decrease in other means of homicide. 20 We estimate that homicides involving firearms in metro areas accounted for just 38.5% of homicides in 2009, compared to 44% in 2005. 20 This suggests that homicide involving firearms has declined more rapidly than homicide by other means, at least in urban areas. The exceptionally high homicide rate among males has been noted previously, 2 but the maleto-female ratio has increased since 2000 indicating that the decrease in male homicide has not kept pace with the greater decline among females. After taking into account the overall decrease, the pattern of homicides by age group was similar to that in the year 2000.
Suicide remains an important contributor to injury-related mortality, although our study does suggest a decrease in the female suicide rate from an estimated 6.1 per 100 000 population  in 2000 to 4.5 per 100 000 population in 2009. Analysis by age shows a slight decrease, compared to 2000 estimates, among adults younger than 44 years, which is offset by an increase among older adults. As has been shown previously, 21 higher suicide rates were associated with increasing socioeconomic status, which was consistent across metro and non-metro areas.
Traffic authorities reported a slight increase in deaths related to road traffic injuries. 22 In contrast, a recent global study reported a consistent rate for road traffic mortality in South Africa from 2000 to 2010. 23 We found that deaths from road-traffic injury have not decreased significantly since the year 2000. If the homicide rate continues to decline, deaths from road-traffic injury are on course to become the leading cause of injury-related mortality. Road traffic injuries are the largest contributor to injury-related mortality in three of South Africa's predominantly rural provinces: Limpopo, Mpumalanga and the North West Province. At a national level, the prominence of pedestrian fatalities is of particular concern as it suggests that the strategy to improve road safety has not met the needs of vulnerable road users. Pedestrian safety relies on reduced exposure to risk through improved safety infrastructure and the provision of alternative transport modes for vulnerable road users, as behavioural modifications to reduce the risk of crash involvement and severity in the event of a collision are not easily attained. 24 This study presents a national profile of injury-related mortality by race, which in South Africa provides a rough proxy for socioeconomic status. Contrary to the conventional discourse that violence is more concentrated in areas of poverty and deprivation, 25,26 our study reveals that peopled categorized as coloured (who are, on average, more affluent than black people), 27 have comparable rates of homicide to black people. This is due to the relatively high rates of homicide among people categorized as coloured in non-metro areas, especially in the Western and Northern Cape. These two provinces have high levels of violence and of alcohol-related harm. Metro homicide rates were higher among racial categories that are, on average, poorer. 20 Comparison of our findings with estimates 2 for the year 2000 highlights important changes in the profile of deaths from external causes. Closer inspection reveals potential inaccuracies from the modelling process and the triangulation of inherently limited data sources that may have affected the earlier estimates. A previous study recognized the urban bias of mortuary-based surveillance data 2 as well as the poor distinction between deaths of undetermined cause (i.e. whether from natural or external causes) and injury-related deaths of undetermined intent. Our data confirm the incompleteness and misclassification of vital registration data from the Department of Home Affairs, which codes as R99 (i.e. an ill-defined natural cause death) any death that is under investigation at the time of certification.
The homicide rate in our study was 5% higher than the 36.4 per 100 000 population in the global burden of disease study, 23 while mortality from road-traffic injury and suicide were approximately fourfold higher than the 8.9 and 3.6 per 100 000, respectively, in the global burden of disease study. We have demonstrated the feasibility and utility of using mortuary-based data to provide timely, accurate and representative injury-related mortality information to monitor major injury trends and to identify at-risk groups.