GBD-2013 substantially underestimates road deaths in Iran

Blogger Katie Leach-Kemon writes on IHME’s website that there have been large improvements in road safety in Iran. She points to GBD-2013 estimates that show a steadily declining trend since 1990. However, as I’ve argued before, GBD-2013 results for Iran are not correct.

The Figure below shows various estimates of road deaths in Iran:


This figure compares GBD-2013 estimates of road deaths in Iran with official government statistics, GBD-2010 estimates, and our estimates (Bhalla et al. 2009). Note that GBD-2010 and GBD-2013 are two different runs of the GBD models; each run of the GBD model produces a full time history of estimates since 1990.

  • The green line shows official government statistics. These are based on deaths reported by the national forensic medicine organization and traffic police. Official government statistics in low- and middle-income countries tend to underreport road traffic deaths. Therefore, the true estimate of road deaths is expected to be higher than the official statistics.
  •  We (Bhalla et al. 2009, in purple) estimated road deaths in Iran in 2005 using the Iranian vital registration data. We estimated that the true death toll was 11% more than reported by national forensic medicine. Vital registration data are a good source for such estimates because a relatively high proportion of deaths in Iran get registered and the coding of causes of death is of reasonably high quality.
  •  GBD-2010 estimates (in blue) also relied on Iranian vital registration data. The GBD-2010 estimates for the year 2010 are broadly consistent with deaths reported by forensic medicine. The mean estimate from GBD-2010 is 18% higher than reported by forensic medicine and the GBD-2010 uncertainty range encompasses the death count reported by forensic medicine. (Note that I have only plotted estimates for 2010 from GBD-2010 because I only found GBD-2010 estimates for one year (2010) in IHME’s data archive.)
  •  Importantly, GBD-2013 estimates (in red) of road deaths in Iran are substantially lower than official government statistics. Notably, GBD-2010 estimates and GBD-2013 estimates for the year 2010 are significantly different even though both estimates rely on national vital registration data. GBD-2013 estimates are less than half of GBD-2010 estimates. In fact, the uncertainty bounds of GBD-2013 and GBD-2010 estimates do not overlap. Such a large difference between GBD-2010 and GBD-2013 is very unusual especially for a country like Iran which has high-quality death registration data.

Since my previous post on the issue, I’ve had a lengthy email exchange with Professor Ali Mokdad at IHME about the GBD-2013 estimates for Iran. He says that we shouldn’t compare results from GBD-2010 and GBD-2013. I disagree. I think that such comparisons are useful because (1) they help identify mistakes (like the one in GBD 2013 that I have highlighted here), and (2) these comparisons help understand the true uncertainty in estimates produced by GBD. Regardless, the fact that the GBD-2013 is dramatically lower than the national official statistics shows that GBD-2013 is substantially underestimating the road death toll in Iran.

I understand that IHME is expected to release GBD-2015 soon; I hope they’ll address this problem.


  • Bahadorimonfared, A., Soori, H., Mehrabi, Y., Delpisheh, A., Esmaili, A., Salehi, M., & Bakhtiyari, M. (2013). Trends of Fatal Road Traffic Injuries in Iran (2004–2011). PLoS ONE, 8(5), e65198–5.
  • Bhalla, K., Naghavi, M., Shahraz, S., Bartels, D., & Murray, C. J. L. (2009). Building national estimates of the burden of road traffic injuries in developing countries from all available data sources: Iran. Injury Prevention : Journal of the International Society for Child and Adolescent Injury Prevention, 15(3), 150–156.

WHO’s Global Status Report underestimates global pedestrian death toll

In a recent blog post, I pointed out that WHO’s latest Global Status Report on Road Safety, underestimates pedestrian deaths in South East Asia because they use statistics from India that are known to be incorrect. In this blog post, I’ll present evidence that the Global Status Report also underestimates pedestrian deaths in China. The overall implication is that the global pedestrian death toll is substantially underreported by WHO.

Let me explain: WHO’s Global Status Report estimates that there are 261,367 road traffic deaths annually in China, of which 26% are estimated to be pedestrians.

Notably, the estimate of total road traffic deaths is not based on official government statistics from China. Instead, it is based on WHO’s analysis of data from the China’s vital registration system and the national disease surveillance system. WHO’s modeled estimate of 261,367 deaths is more than four times the official government statistics of 58,539 deaths for the year 2013 — i.e. the Chinese government’s official statistics only report about one in five road traffic deaths in the country.

Despite the very severe issue of underreporting of road traffic deaths in China, the Global Status Report uses the proportion of pedestrian deaths (26%) reported in the official government statistics as their estimate of pedestrian deaths in China. This is very strange because we do not expect underreporting to be the same for all types of victims. Deaths among certain groups, such as the poor (which pedestrians tend to be), are more likely to be underreported.  A more reasonable thing to do would have been for WHO to estimate pedestrian deaths using the same data sources that they used to estimate total road deaths (i.e. death registration data). I do not know why they did not do this.

A recent IHME/GBD paper in the Lancet used national death registration data to estimate causes of death in China. They report that in 2013, 53% of all road traffic deaths in China were pedestrians — more than twice the proportion reported by the WHO Global Status Report.

In a previous blog post, I had explained that pedestrians likely comprise about 37% of road deaths in India rather than the 9% estimated by WHO. In this blog post, I’ve presented evidence that pedestrians likely comprise about 53% of road deaths in China rather than the 26% estimated by WHO. Because India and China together comprise a substantial proportion of the global population, an error in their estimates has a substantial effect on the global road death toll. WHO estimates that 22% of global road deaths are pedestrians. However, if we correct the estimates for India and China as discussed above, the corrected estimate at the global level would attribute 33% of the total global road deaths to pedestrians, which is a very large increase from what WHO reports.

Of course, this only accounts for the undercounting of pedestrians in China and India. We should expect pedestrians to be underrepresented in other countries as well.

The fundamental issue here is that WHO has chosen to use the proportional distribution of road deaths reported in official government statistics without corrections. They should have modeled these proportions just like what they did to estimate the national total road traffic death estimates.

2015 WHO Global Status Report: Committees are not a source of reliable statistics!

WHO has recently released their 2015 Global Status Report on Road Safety (2015 WHO GSRRS). This is the third in a series of reports produced by WHO every three years that provide an assessment of global progress towards road safety. Over coming days, I hope to share my thoughts on the report as I work my way through it. (Click here to see all posts on the subject)

Like the two previous status reports, the 2015 WHO GSRRS provides one page of road safety data for each country. From my standpoint, these country-pages contain three types of information:

  1. Statistics that I expect are mostly reliable (e.g. the country’s population, GDP per capita, number of registered vehicles of different types) because in most cases this information arises from fairly well understood and often reliable data collection processes.
  2. Statistics that need to be interpreted carefully in the light of how these data are generated. This includes, for instance, the official statistics of road traffic deaths in each country, which is often from traffic police (who may or may not have the capacity to collect such data) or from vital registers (which may be incomplete, poorly coded, or the reported statistics may involve statistical corrections that may not be transparent).
  3. Statistics that I simply ignore because I think they are likely too unreliable to be of any value. These are the data generated by asking country experts questions that I don’t think could possibly be answered by experts unless they have access to empirical data (which I often know doesn’t exist).
Generating information for (3) is likely a substantial part of the work undertaken by WHO to produce these status reports, and a substantive part of the new insights these reports seek to add. I say this because information about (1) and (2) are often readily available by searching the web or from existing cross-national databases.
WHO collects such consensus data by appointing a national data coordinator, who collects the answers provided by a committee of local experts, and organizes a meeting to get consensus on the final responses. An example of such a questionnaire is available online here.  Here is a typical question, “In your opinion, how effective is enforcement of speed limits in your country (please rate enforcement on a scale of 0 to 10, where 0 is not effective at all and 10 is highly effective)”. I can imagine that this question could be answered through a carefully designed study but I wouldn’t trust any expert who hasn’t conducted such a study to say anything remotely reliable.
Of course, WHO doesn’t just collect such data, it also tries to assess them and draw conclusions. I wanted to understand to what extent this can be done and decided to take a closer look at this statement from their new global status report:

“This assessment found that only 27 countries (15% of participating countries) rate their enforcement of speed laws as “good” (8 or above on a scale of 0 to 10), suggesting that without ongoing and visible enforcement of speed limit legislation, the potential impact of speed legislation to save lives globally remains vastly unattained.” (Pg 22, 2015 WHO GSRRS)

Population wide speed enforcement that covers the entire road network is notoriously difficult to do, and I wondered which 27 countries could have succeeded. Sweden? France? Netherlands? Probably.

Well, I extracted the information from the status report and here is how all countries rank on speed enforcement:

This table shows the level of speed enforcement in all countries on a scale of 0 (worst) to 10 (best) as reported in the 2015 WHO GSRRS.

The two countries that report having the best speed enforcement, Turkmenistan and United Arab Emirates, seem unlikely candidates for the top spot. Both have death rates substantially higher than OECD countries. Revealingly, both countries also gave themselves a perfect score on all other types of enforcement (drink driving law, helmet law, and seatbelt law). Perhaps the countries have recently implemented road safety programs and their local expert panel was optimistic enough to consider them the best in the world!
There is a lot written about speed control in France (see here, for example), so I’m not surprised to see its score of 9. However, it’s impossible that Nicaragua, Oman and Togo deserve to be listed alongside France. Neither is it possible that countries like Kazakhstan, Tajikistan, and Zambia should be in the 8+ score that WHO considers “good” in the text quoted above. These countries have death rates higher than the global average and many times higher than countries in Western Europe that score much lower on speed enforcement (e.g. Sweden and Netherlands).  While most countries worldwide struggle with speed enforcement, Nicaragua would appear to have pulled off a miracle because in the 2009 WHO GSRRS, it had only scored a 3 on speed enforcement.
I don’t think WHO should generate such statistics through expert consultation. In the best case, it generates silly data as I’ve highlighted here. In the much worse case, publishing such tables under the WHO banner gives bad data false legitimacy. If you think that nobody would really uses this data for serious statistical analysis, you would be wrong. It’s increasingly common to extract data from the GSRRS and use them in cross-national statistical analysis. Here are a few examples.
And, there are other similar papers. I’ve listed the name of the journal here to highlight that these studies are being published in well regarded academic journals, not in obscure places where articles may not be properly peer reviewed before publication. For instance, Accident Analysis and Prevention, is arguably one of the best academic journals for injury research where some of the best researchers in the road safety field publish. Unfortunately, once WHO has published such data, many researchers, peer reviewers, and journal editors assume that the data are meaningful.

2015 WHO Global Status Report: Overstating GDP losses due to traffic crashes?

WHO has recently released their 2015 Global Status Report on Road Safety (2015 WHO GSRRS). This is the third in a series of reports produced by WHO every three years that provide an assessment of global progress towards road safety. Over coming days, I hope to share my thoughts on the report as I work my way through it. (Click here to see all posts on the subject)

I think it’s time for all of us who work in road safety to speak clearly and truthfully about the costs of traffic crashes globally. Although I’ve been prompted by the 2015 WHO GSRRS to write this blog post, the issue is much more pervasive across the road safety field. Policy reports (such as the 2015 WHO GSRRS and numerous others) are likely overstating the impact of crashes on the economy. I understand that this may be a good faith attempt at trying to desperately bring more attention to road safety. However, if we lie about such statistics now we will stifle our ability to do serious work in the future.

Consider the following quotes from the 2015 WHO GSRRS:

  1. Road traffic injuries … cost governments approximately 3% of GDP.” Page VII, Foreword written by the Director General of WHO, Margaret Chan.
  2. … low- and middle- income countries lose approximately 3% of GDP as a result of road traffic crashes”, Page IX, Executive Summary.
  3. Road traffic deaths and injuries in low- and middle- income countries are estimated to cause economic losses of up to 5% of GDP. ” Sidebar on Pg. XI.
  4. Pages 77 to 270 provide a one-page summary of crash statistics for each country. This includes each country’s “Estimated GDP lost due to road traffic crashes:” (As an aside, the numbers reported are difficult to justify. For instance, Jamaica reports a GDP loss of 0.2% and Bulgaria 2.0%, even though Jamaica has a higher road traffic death rate than Bulgaria.)

The claim that low- and middle- income countries lose 3% of GDP due to road traffic injuries is clearly intended to mean that a country that has a GDP of US$100 billion, would have had a GDP of US$103 billion if it didn’t have any road traffic injuries. However, that is a serious misrepresentation of the research from which such estimates are derived. There are lots of research studies that have estimated the economic burden of road traffic injuries but there isn’t a single study published yet that aims to measure the macroeconomic impact of road traffic injuries on a country’s economy (and its GDP loss). Let me explain the difference.

If not GDP loss, what do injury costing studies actually measure?

It is important to understand what studies of the economic burden of injuries aim to measure. Broadly speaking, most health costing studies start by estimating the value of a statistical life that is lost (or, when considering non-fatal disabilities, the value of a statistical life-year lost). If we have a way to put a monetary value on a unit of health loss, and we already know the total amount of health loss in a society due to injuries, then we can simply multiply the two to get the total monetary value of the health lost due to injuries. This is the main logic of injury costing studies. Note that this has little to do with GDP, which is the net monetary value of all goods and services produced by the society.

The difference becomes clearer once we understand how economists estimate the monetary value of a unit of health loss? There are two ways that are commonly used:

1. Conducting studies to reveal how much people are willing to pay to avoid health risks (i.e. the Willingness-to-pay approach): For instance, a survey may be conducted that asks people how much they would be willing to pay to reduce their risk of injuries by a small amount. From this, the economic value that people place on a death or disability can be estimated. Such methods are controversial and economists have debated extensively about how to measure what people are willing to pay to avoid illness and injury. Nevertheless, everybody agrees that these willingness-to-pay estimates are not intended to measure the effect of injuries on a society’s economic production — i.e. these studies do not measure the impact of injuries on GDP.  Much of what people are willing to pay to avoid a death likely has to do with the emotional pain that they suffer when they lose a loved one, which isn’t related with how that person affected the economy. GDP loss is the amount of market production that did not happen because of traffic crashes. Pain and grief matter little from the perspective of the economy.

2. Estimating the tangible monetary impacts of ill health (called the Human Capital approach or Cost-of-Illness approach): This method aims to estimate the direct and indirect costs incurred due to road traffic injuries. Examples of direct costs include medical treatment (ambulance and hospital costs) and funeral costs. Examples of indirect costs include the income that is lost due to the loss of wages when a person cannot return to work because of disability or death. There are many unresolved measurement questions with this method as well (e.g. how do we handle non-market labor such as housework? ). However, again, one thing is clear — the method does not measure the macroeconomic impact of injuries on the economy and does not measure GDP loss. To see why, consider the following:
— Labor losses: These losses are usually a large value, typically accounting for 75-80% of the total costs estimated using this method. The reason these values are large is because they are computed by assuming that when a person dies the total of all their future earnings are lost. However, this is a dramatic overestimate of the loss borne by the economy. In most countries, there is a pool of unemployed people who will usually fill the opening created by a person who can no longer work. Thus, as far as the economy (and the GDP) is concerned, the main loss involved could be only the cost of retraining new hires. The fact that it is a different person who is earning these wages now doesn’t affect the GDP.
— Medical costs: The cost of providing medical care for injuries are the main direct costs in the Human Capital method, typically accounting for 15-20% of the total costs using this method. However, these costs (and all other direct costs) are actually part of the formal economy — they contribute to the GDP.  As far as the economy is concerned, doctors and hospitals produce value, and this is included in the GDP as a positive contribution. One person’s medical payments go towards another’s salary.

So, how should we measure the true impact of traffic injuries on GDP?
If we must measure GDP loss, we need other tools. For instance, one approach would be to develop a macroeconomic model that relates the incidence of injuries with the the loss of labor supply (due to disability and death of workers) and the loss of capital (such as due to health care expenditures that may otherwise have gone towards savings or investments) with economic production. WHO has a tool for this purpose called Projecting the Economic Costs of Ill-Health (EPIC) that has been applied previously to assess the global macroeconomic impact of non-communicable diseases. However, to my knowledge, this tool or a similar methodology has never been applied to road traffic injuries.

The problems that I’m highlighting here are not new. In fact, most of these issues are discussed at length in the 2009 WHO Guide to Identifying the Economic Consequences of Disease and Injury, which was produced by WHO after consultation meetings with the best health economists from across the globe.

In closing I should add that although claims about GDP losses are made commonly in policy reports and presentations, they are rarely made in the formal research literature or in policy reports that are led by economists. To give a few notable examples:

  • The landmark 2006 study by Corso, Finkelstein, Miller and colleagues estimated the costs of injuries in the US. Their paper presented medical costs and productivity losses. They did not mention impact of injuries on the GDP of the US.
  • New Zealand’s costs of injury report estimates costs using willingness-to-pay methods but they never present this as impact on the national GDP.
  • Bishai and Bachani’s chapter titled Injury Costing Frameworks (paywall) reviews methods for estimating the economic burden of injuries but then explains that to measure impact on the economy, we would need to develop general equilibrium methods for macroeconomic analysis of the effects of road traffic injuries.

There are numerous research publications where GDP is explicitly mentioned but this is done to enable cross-country comparisons or to put a large dollar figure into perspective.  Thus:

  • The 2008 IRAP report by McMahon and Dahdah provided a simple algebraic equation to estimate the net economic burden of injuries in settings where detailed data about injury costs are not readily available. Their equation uses GDP as an index for estimating the value of (statistical) deaths and injuries in any country based on empirical evidence that suggests a relationship between national income and willingness-to-pay estimates. However, they do not claim that their equation provides an estimate of the amount of GDP that is lost.
  • A 2011 paper by Lim et al. that reported the economic burden of injuries in South Korea included the following sentence after their cost estimate, “as a yardstick, this represents 2.7% and 4.9% of Korea’s GDP”. Note that they are using the GDP to provide perspective to their economic burden estimate — they are not claiming that they have measured  the amount of GDP lost in South Korea to injuries.
  • A 2011 paper by Perez Nunez et al. (paywall) reports the economic burden of injuries in a state in Mexico and then adds “This figure equals 1.3% of the GDP of the state.” Again, like Lim et al., they are only using GDP as a yardstick.
  • In my own work (such as this report on the cost of road injuries in Latin America), I estimate the net economic burden of road traffic crashes in four countries in Latin America and then present these figures as “equivalent to X% of GDP” to allow a comparison across countries.

To summarize, we do not know how much road traffic injuries hurt the economy nor do we know how much GDP is lost. I suspect that the macroeconomic impact would likely be much smaller than the 3 to 5% figure provided in the 2015 WHO-GSRRS but we won’t really know until somebody does the correct analysis. Personally, I believe that we do not need such analysis to justify our safety investments because the immense loss of health due to traffic injuries that has been documented extensively is already reason enough.

2015 WHO Global Status Report: Attributing too much importance to trauma care?

WHO has recently released their 2015 Global Status Report on Road Safety (2015 WHO GSRRS). This is the third in a series of reports produced by WHO every three years that provide an assessment of global progress towards road safety. Over coming days, I hope to share my thoughts on the report as I work my way through it. (Click here to see all posts on the subject)

Here is a little mistake in the 2015 WHO GSRRS that symbolizes much of what is wrong with global road safety work:

The report has the following section:

Lack of emergency care creates injury outcome disparities
The gross disparities in injury outcomes between high-income countries and low- and middle-income countries relate directly to the level of care received immediately post-crash, and later in a health- care facility. Some estimate that if trauma care systems for seriously injured patients in low- and middle income countries could be brought up to the levels of high- performing countries, an estimated half a million lives could be saved each year (12).” (Page 12, 2015 WHO GSRRS)

Half a million lives annually!! That’s almost half of the road traffic deaths that happen in low- and middle- income countries.

Finding this statistic hard to believe, I looked up Reference 12 cited in the text above, which is available online here. The study does not estimate the lives that would be saved by improving trauma care systems. Instead, their goal is to “ estimate the number of lives that could be saved if injury mortality rates in low- and middle-income countries could be reduced to rates in high-income countries” — i.e. it estimates the net effect of not just trauma care but all injury prevention activities. It includes the effects of designing cars to be safer, investing in road safety infrastructure, strong government regulation of how people are allowed to behave on roads, and decades of work in designing, implementing, and sustaining national safety programs in high-income countries.

I didn’t see other quantitative estimates of lives that could be saved by any of the other interventions in the 2015 WHO GSRRS. The text cited above is the only explicit estimate of lives saved and it attributes all the gains made by high-income countries to better trauma care.

It’s only a tiny mistake but I think it is revealing that it would slip by the many eyes that would have seen this report before it was finalized. Global road safety efforts are dominated by people with medical or public health backgrounds. As a result, we have a tendency to think that medical interventions and behavior change interventions are far more important for road safety than is likely the case. Trauma systems are obviously important but they are only one part of a Safe System.

WHO’s 2015 Global Status Report: Underestimating need for pedestrian safety?

WHO has recently released their 2015 Global Status Report on Road Safety (2015 WHO GSRRS). This is the third in a series of reports produced by WHO every three years that provide an assessment of global progress towards road safety. Over coming days, I hope to share my thoughts on the report as I work my way through it. (Click here to see all posts on the subject)

I cringed when I read the following section in the 2015 WHO GSRRS, which claims that the proportion of pedestrian deaths in South East Asia is relatively low and that this is partly due to safety measures:

… the likelihood of dying on the road as a motorcyclist, cyclist or pedestrian varies by region: the African Region has the highest proportion of pedestrian and cyclist deaths at 43% of all road traffic deaths, while these rates are relatively low in the South-East Asia Region (see Figure 7). This partly reflects the level of safety measures in place to protect different road users and the predominant forms of mobility in the different regions – for example, walking and cycling are important forms of mobility in the African Region, while in the South-East Asia Region and the Western Pacific Region, motorcycles are frequently used as the family vehicle.” (Pg 8, 2015 GSRRS)


Well, here is the Figure 7 that is cited in the text from the report above:


This is Figure 7 in the 2015 WHO Global Status Report on Road Safety, which shows the distribution of road traffic deaths (by type of road user) in world regions. (Source: 2015 WHO GSRRS, Pg 8)

Indeed, the proportion of road traffic deaths that are pedestrians in South-East Asia (13%) shown in this Figure is surprisingly low. Conventional belief is that pedestrians comprise a large proportion of road traffic deaths in low- and middle-income countries, including in South-East Asia. So, what is going on here?

I’m nearly certain that I have an explanation. India accounts for over two-thirds of the population of the WHO’s South East Asia Region and thus India’s statistics have a large influence on the regions statistics. Road safety researchers familiar with India know that the country’s official road safety statistics do not correctly report the types of road users killed. Pg 147 of the 2015 WHO GSRRS, which provides official statistics from India, reports that only 9% of road traffic deaths in India are pedestrians. While it should be obvious to road safety researchers that this number couldn’t possibly be correct, here are a few points that illustrate the issue:

  • India’s official road safety statistics are generated from police reports by the National Crime Records Bureau. The data in their 2014 annual report (available here) shows that many cities report no pedestrian deaths. For instance, the city of Chennai reported 1046 road traffic deaths in 2014 but none of these were pedestrians. That’s none! Zero!
  • Pedestrians comprise a  much higher proportions of road traffic deaths in countries that neighbor India according to the 2015 WHO GSRRS (Bangladesh: 32%, China: 26%, Sri Lanka: 29%).
  • India’s national sample registration system, which ascertained the causes of 122,000 deaths from all causes between 2001 and 2003 in a nationally representative sample of households in India reported that 37% of road traffic deaths were pedestrians. (Hsiao et al. 2013; available here) This is likely the most reliable source we have for this statistic.
  • There have been lots of studies published from across India that report the descriptive epidemiology of road traffic injuries from data collected from other sources (such as hospital and ER surveillance, mortuary records, household surveys, etc). These studies invariably find that pedestrians comprise much higher proportions of road traffic deaths than the 9% figure reported in official government statistics and the 2015 WHO GSRRS.

The problem with official government statistics in India is discussed routinely at scientific meetings that are attended and/or organized by WHO. Therefore, I find it surprising that WHO has continued to present these statistics from India uncritically. Although this is mildly bothersome, it is data from a single country. What I find more problematic is that WHO would allow the data from India to pollute the regional statistics for South East Asia, and suggest (as they do in the text cited above) that pedestrians may be safer in the region partly due to safety measures.

GBD estimates for South Africa compared with newly published data

The current issue of the Bulletin of the World Health Organization carries an article by Matzopoulos et al. that reports estimates of deaths due to various types of injuries in South Africa. They report that “… 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.

Reliable measurement of injury deaths in Sub-Saharan Africa are rare but this is an unusually good study. The authors investigated the medico-legal information available for a nationally representative sample of postmortem reports from 2009. Trained field workers visited 45 mortuaries; inspected the forensic pathology reports and accompanying police reports and hospital records for 32,000 deaths; and recorded data on the demographics of victim and causes of death.

Here is how the results of Matzopoulos et al. compare with GBD-2010:

                                GBD-2010               Matzopoulos et al.
Road deaths            4,479                       17,103
Suicide                     1,786                         6,471
Homicide                18,281                      19,028
All Injuries               44,438                      52,493

As Matzopoulos et al. report, road deaths and suicides in South Africa are four times those reported by GBD.

The fact that GBD-2010 estimates for road deaths in South Africa were problematic is something that I have discussed before. In our Burden of Road Injuries in Sub-Saharan Africa Report, we presented GBD-2010 estimates but noted that “estimates of road injury deaths in Southern SSA are likely too low. For instance, in South Africa, which has 71% of the regional population, the road injury death rate based on national traffic police statistics is 2.3 times the GBD estimate.” In 2010, national traffic police reported 14,804 road traffic deaths, much higher than GBD but only 13% lower than the estimate by Matzopoulos et al.

Why the GBD-2010 estimates of road deaths in South Africa are so low is a mystery. GBD uses national death registration data from South Africa as a primary source of data for estimating road traffic deaths.  ICD-coded cause-of-death tabulations for South Africa are available online from the WHO Mortality Database. However, even a crude calculation using these tabulations provides results that are much closer to those reported by Matzopoulos et al. than the GBD-2010 estimates.

Consider the following simple-minded calculation: The South African death registration data for 2010 reports the following death counts using ICD-10 cause codes:

  • 5,646   deaths coded to V00-V89, which roughly corresponds to “road deaths”;
  • 14,207 deaths coded to V90-X58, which corresponds to “accidental deaths other than road deaths”; and
  • 15,576 deaths coded to X59, which corresponds to “unspecified accidental deaths”

First, note that even before any adjustments, there are 26% more deaths coded to V00-V89 than the GBD estimate of 4,479 deaths. Second, a fairly large number of deaths are coded to the dump code X59. Prior to GBD-2010, the usual practice in GBD was to prorate these deaths to specified causes. This is usually done with data disaggregated into age- sex- groups but working with the aggregate totals is likely not a bad approximation. If we were to prorate X59 deaths in this manner, we would assign 28% of the 15,576 X59 deaths to road injuries. Doing this would bring the road death toll up to 10,076 deaths, which is much closer to the road deaths reported by Matzopoulos et al. This is prior to the distribution of deaths coded to various other dump codes (including “unspecified type of injury”, and the broadest “unspecified causes of death”), and prior to adjustment for incomplete death registration.  It is conceivable that doing these additional adjustments would bring the road death toll to close to the value reported by Matzopoulos et al.

GBD-2010 now uses new methods for handling deaths coded to unspecified causes. These methods are substantially more complex and I have found them impossible to understand using what is documented by the study. My previous blog posts and paper in IJE demonstrate that these methods perform poorly.