Evaluating HIV Estimates from National Population-Based Surveys

Evaluating HIV Estimates from National Population-Based Surveys for Bias Resulting from Non-Response (English)

Reliable data on HIV prevalence are essential for assessing the scope of and effectively managing the response to the epidemic. Antenatal clinic-based surveillance is commonly used to monitor trends in HIV in developing countries that have generalized epidemics. Recently, HIV seroprevalence data have been also collected in national population-based surveys, such as the Demographic and Health Surveys (DHS) and AIDS Indicators Surveys (AIS). Such surveys enable direct estimation of population HIV prevalence. A major challenge for population-based surveys is bias resulting from non-response, both from refusal and absence. In this study, we evaluate national HIV prevalence estimates from DHS and AIS surveys for bias resulting from non-response in the surveys. Data are from 17 recent national DHS and AIS surveys with HIV testing – Burkina Faso, Cambodia, Cameroon, Cote d’Ivoire, the Dominican Republic, Ethiopia, Ghana, India, Kenya, Lesotho, Malawi, Mali, Rwanda, Tanzania, Uganda, Zambia, and Zimbabwe – conducted during 2001 and 2006. Blood samples were collected and tested for HIV using standard laboratory and quality-control procedures. In the first three surveys, in Mali, Zambia, and the Dominican Republic, HIV status could not be linked to the characteristics and behaviors of the survey respondents. For each of the other 14 countries with HIV serostatus data linked to individual characteristics and behaviors, we predict HIV prevalence among non-responding adults on the basis of multivariate statistical models of HIV for those who were interviewed and tested, using a common set of predictor variables. Predictions are made separately for two groups of non-respondents: not interviewed/not tested and interviewed/not tested. Adjusted HIV prevalence is calculated as a weighted average of observed prevalence in the interviewed/tested group and predicted prevalence in the two non-tested groups. Predictions are made separately for adult males and females. In the 14 countries with linked data, the HIV testing rate varied from a low of 63 percent among men in Malawi and Zimbabwe to a high of 97 percent among women in Rwanda. Non-response rate was higher among urban, more educated, and wealthier men and women but had no clear association with various risk and protective behavioral factors. Non-tested men had significantly higher predicted HIV prevalence than those tested in 8 of the 14 countries, and non-tested women had significantly higher predicted prevalence than those tested in 7 of the 14 countries. Although non-tested men and women tend to have higher predicted HIV prevalence than those tested, the overall effect of non-response bias on observed prevalence estimates was small and not significant in all countries. In the 14 countries, HIV prevalence estimates adjusted for non-response bias were on average only 3 percent and 2 percent higher than the observed, non-adjusted estimates for men and women, respectively. The study finds that non-response for HIV testing tends to have small, non-significant effects on national HIV seroprevalence estimates obtained from national household surveys. National population-based surveys are an important source of reliable data on HIV prevalence that can enhance surveillance-based estimates in generalized epidemics.

full text in pdf about 3 mb