I was struck by the following graph that appeared in a comment by John Bongaarts, François Pelletier, and Patrick Gerland in the Lancet last week. The graph compares estimates of HIV mortality made by the WHO in the early 2000s via the Global Burden of Disease Project (GBD), updated estimates GBD estimates from the WHO, UNAIDS, and recent projections put out by the UN.


By now, I hope most people reading this blog have realized that due to a number of changes, most notably the collection of population based estimates of HIV prevalence have led to significant decreases in the projected levels of HIV mortality globally. That along with updated models explains the difference between earlier HIV mortality estimates – including the earlier GBD estimates – (consistent with the top line in the graph) and the three bottom lines, which generally now incorporate the new prevalence estimates. (Side note: I recently saw a paper from David Canning who has suggested that non-random selection effects in DHS and other population based estimates are likely underestimating true prevalence by a significant amount…more when that paper goes public).

But why do the updated GBD WHO estimates decline so much faster than the recent estimates by the UN? The difference can largely be explained by assumptions about the scale up and impact of HIV treatment programs. The authors of the comment argue that the WHO estimates “can only be achieved by massive scale-up of antiretroviral treatment to lead to near universal access worldwide in 2015-30”. Great progress has been made on this front, true, but how realistic of an assumption is this? In particular in this new economic climate?

Given how widely these estimates are used for global health policy, I am somewhat surprised how optimistic of an assumption the WHO is using – if anything it only weakens their advocacy argument for more funding. But this comment clearly demonstrates that even the best statistics in global health are based on some very powerful – and potentially incorrect – assumptions.

Share on Facebook
 

Leave a Reply



Analytics Plugin created by Web Hosting