Better methods for measuring HIV-related deaths

I remember hearing David Bourne talk while I was taking classes at UCT a few years ago. He was a forceful speaker, but the most powerful memory I have of the talk was his presentation of the data that later became a 2005 paper outlining how deaths from HIV were being undercounted, and adjusting for this using the clever approach of looking at trends in likely opportunistic infections (OIs) such as pneumonia and tuberculosis.

The idea was that increases in these causes of death – not generally things we would expect an increase in given South Africa’s level and growth direction of economic well-being – over and above some pre-serious-HIV-mortality timepoint (they use 1996, which makes sense given the selective vital statistics data available prior to fully-representative democracy in 1994) could probably be associated with HIV infection. They match the estimates found in this way with existing dynamic mathematical models of the epidemic, and suggest that death certificates had been underestimating HIV-related mortality by more than 60%. This was particularly important at the time due to the combative mood in the country over the magnitude of the HIV epidemic, with some suggesting that the problem was not as significant since few people were being recorded as dying of HIV or AIDS.

Now, there is a possibly even cleverer study in a recent edition of the Bulletin of the WHO. They note that the estimates made in 2005 relied on a lack of misclassification of causes of death in 1996 – the baseline year. To account for potential misclassifcation, they use worldwide data to estimate the relative risk of death for each age group for various categories of cause of death, compared to a reference age range of 65 to 80, where HIV infection is expected to have been irrelevant at any meaningful level. The authors then compare these global relative death risks (RDR) to South African RDRs in each cause of death category. Those which stand out as different are marked down as potential sources of misclassified deaths. The differences between the worldwide and South African RDRs were then used to estimate the number of misclassified deaths.

And the result: “this study suggests that during 1996–2006 as many as 94% of all HIV/AIDS deaths in the country were being misclassified, especially among young to middle-aged females and among males in the middle-aged and older groups.” This is even more than the 2005 study suggested. But doesn’t really seem to qualitatively change our impressions. Nevertheless, and despite various caveats the authors’ note, this was a nice attempt to further strengthen a methodology to evaluate the level of misclassification present in cause of death notification. And it reminded me of a good, and always thoughtful, man who is sadly no longer with us.