The epidemiology of HIV infection in Zambia

Abstract

Population surveys of health and fertility are an important source of information about demographic trends and their likely impact on the HIV/AIDS epidemic. In contrast to groups sampled at health facilities they can provide nationally and regionally representative estimates of a range of variables. Data on HIV-sero-status were collected in the 2001 Zambia Demographic and Health Survey (ZDHS) and made available in a separate data file in which HIV status was linked to a very limited set of demographic variables. We utilized this data set to examine associations between HIV prevalence, gender, age and geographical location. We applied the generalized geo-additive semi-parametric model as an alternative to the common linear model, in the context of analyzing the prevalence of HIV infection. This model enabled us to account for spatial auto-correlation, non-linear, location effects on the prevalence of HIV infection at the disaggregated provincial level (nine provinces) and assess temporal and geographical variation in the prevalence of HIV infection, while simultaneously controlling for important risk factors. Of the overall sample of 3950, 54% was female. The overall HIV-positivity rate was 565 (14.3%). The mean age at HIV diagnosis for male was 30.3 (SD=11.2) and 27.7 (SD=9.3) for female respectively. Lusaka and Copperbelt have the first and second highest prevalence of AIDS/HIV (marginal odds ratios of 3.24 and 2.88, respectively) but when the younger age of the urban population and the spatial auto-correlation was taken into account, Lusaka and Copperbelt were no longer among the areas with the highest prevalence. Non-linear effects of age at HIV diagnosis are also discussed and the importance of spatial residual effects and control of confounders on the prevalence of HIV infection. The study was conducted to assess the spatial pattern and the effect of confounding risk factors on AIDS/HIV prevalence and to develop a means of adjusting estimates of AIDS/HIV prevalence on the important risk factors. Controlling for important risk factors, such as geographical location (spatial auto-correlation), age structure of the population and gender, gave estimates of prevalence that are statistically robust. Researchers should be encouraged to use all available information in the data to account for important risk factors when reporting AIDS/HIV prevalence. Where this is not possible, correction factors should be applied, particularly where estimates of AIDS/HIV prevalence are pooled in systematic reviews. Our maps can be used for policy planning and management of AIDS/HIV in Zambia.

Keywords:

• HIV/AIDS
• Demographic Health Survey
• maps
• Zambia
• non-linear effects

Notes

1. Briefly, we replace the strictly linear predictor (1) with a logic link function with dynamic and spatial effects Pr( y _i =1∣η _ι ) = e ^ηι /(1 + e ^ηι ) with a geo-additive semi-parametric predictor µ_i=h(η _ι ):η _ι =f ₁ (x _i1 ) + … + f _p (x _ip ) + f _spat (s _i ) + w’ _iγ (2) where h is a known response function with a logit link function and, f ₁ , … ,f _p are non-linear smooth effects of the metrical covariates (age at HIV diagnosis) and f _spat (s _i ) is the effect of the spatial covariate s _i ∈{1, … ,S} labelling the region in Zambia. Covariates in w’ _I are usual categorical variables such as gender and urban-rural residence. Regression models with predictors as in (2) are sometimes referred to as geo-additive models. P-spline priors are assigned to the functions f ₁ , … ,f _p and for f _spat (s _i ) we used Markov random field prior (Fahrmeir and Lang, 2001; Kandala, 2006 & 2007). The analysis was carried out using BayesX version 0.9, software for Bayesian inference based on Markov Chain Monte Carlo simulation techniques.