Eaton, J.W., Bacaër, N., Bershteyn, A., et al. The Lancet Global Health (October 2015), 3(10): e598–608, doi: 10.1016/S2214-109X(15)00080-7.
This study compared 10 mathematical model projections of HIV prevalence, HIV incidence, and antiretroviral therapy coverage for South Africa against data from a large household survey done in 2012, seeking to validate past model projections. The authors reported that five models projected that prevalence in adults aged 15–49 years in 2012 would change by ≤ 0.3 percentage points from prevalence in 2008. Three models projected declines of 0.7 to 1.3 percentage points; one projected an increase of 0.9 percentage points. However, the household survey estimated that adult prevalence increased from 16.9 percent in 2008 to 18.8 percent in 2012. The disparity between the 2012 survey estimate and those in the 10 models was mainly because eight of the models projected that prevalence would decline among men, whereas 2012 household survey data estimated that prevalence increased by 2.9 percent among men. The authors concluded that the models might have been overly optimistic, especially for mid-aged adults (age 25–49 years), among whom prevalence and incidence were consistently higher than anticipated. However, they urged program planners and implementers to continue to collect surveillance and trial data to validate and improve the information provided through mathematical models.