When did HIV incidence peak in Harare, Zimbabwe? Back-calculation from mortality statistics

Abstract

HIV prevalence has recently begun to decline in Zimbabwe, a result of both high levels of AIDS mortality and a reduction in incident infections. An important component in understanding the dynamics in HIV prevalence is knowledge of past trends in incidence, such as when incidence peaked and at what level. However, empirical measurements of incidence over an extended time period are not available from Zimbabwe or elsewhere in sub-Saharan Africa. Using mortality data, we use a back-calculation technique to reconstruct historic trends in incidence. From AIDS mortality data, extracted from death registration in Harare, together with an estimate of survival post-infection, HIV incidence trends were reconstructed that would give rise to the observed patterns of AIDS mortality. Models were fitted assuming three parametric forms of the incidence curve and under nine different assumptions regarding combinations of trends in non-AIDS mortality and patterns of survival post-infection with HIV. HIV prevalence was forward-projected from the fitted incidence and mortality curves. Models that constrained the incidence pattern to a cubic spline function were flexible and produced well-fitting, realistic patterns of incidence. In models assuming constant levels of non-AIDS mortality, annual incidence peaked between 4 and 5% between 1988 and 1990. Under other assumptions the peak level ranged from 3 to 8% per annum. However, scenarios assuming increasing levels of non-AIDS mortality resulted in implausibly low estimates of peak prevalence (11%), whereas models with decreasing underlying crude mortality could be consistent with the prevalence and mortality data. HIV incidence is most likely to have peaked in Harare between 1988 and 1990, which may have preceded the peak elsewhere in Zimbabwe. This finding, considered alongside the timing and location of HIV prevention activities, will give insight into the decline of HIV prevalence in Zimbabwe.

Figure 1. Crude mortality rate from death registration in Harare, with three scenarios of stable, decreasing and increasing underlying non-AIDS mortality.

Y-axis is deaths per 1000 population per year.

Figure 2. Comparison of three parametric forms of the incidence curve: log-logistic (A) gamma (B) and cubic spline (C).

Incidence curves (blue line) and generated mortality curve (red line) was fitted to observed excess AIDS mortality (black points) and prevalence (orange shade). All assume median survival post infection is 11.3 years and stable non-AIDS mortality.

Figure 3. Time trends of adult HIV prevalence (A&B) projected from parameterised incidence curves (C&D).

Higher prevalence results assuming declining background mortality (orange lines) compared with stable (green lines) or increasing background mortality (blue lines). Darker shaded lines represent longer survival post-infection; lighter lines represent shorter survival. UNAIDS prevalence estimates are shown for comparison (symbol, ref). HIV incidence measured in Harare male factory workers (○[7]) and Harare post-natal women (□[27]) are plotted alongside modelled incidence in C&D. Prevalence and incidence estimates from back-calculation are for 15 to 59 year olds, empirical and national estimates generally refer to 15 to 49 year olds.