University of California, San Francisco Logo

University of California, San Francisco | About UCSF | Search UCSF | UCSF Medical Center

Adjusting HIV prevalence for survey non-response using mortality rates; an application of the method using surveillance data from rural South Africa
Global Health Sciences Literature Digest
Published January 03, 2011
Journal Article

Nyirenda M, Zaba B, Barnighausen T, Hosegood V, Newell ML. Adjusting HIV prevalence for survey non-response using mortality rates; an application of the method using surveillance data from rural South Africa. PLoS One. 2010;5 (8):e12370.

In Context

Population-based estimates of HIV prevalence in sub-Saharan Africa have been primarily based on sentinel surveillance of pregnant women attending antenatal clinics. Increasingly, they have been based on population-based household surveys that include HIV testing. These household surveys typically have national coverage and provide data for women and men and in urban and rural areas. However, the main limitations are based on the level of non-response due to refusal or absence from the household, which can lead to potential bias in the estimates of HIV prevalence.(1, 2) A review of 20 population-based household-based surveys conducted since 2001 in 19 sub-Saharan countries demonstrated that the proportion of women who refused HIV testing was highest in South Africa (30.2%), ranging from 0.3 to 14.4% in the other countries. Refusal rates among men were also highest in South Africa (34.6%).(1) In a more recent national household survey in South Africa, refusal rates for HIV testing were highest (37%) in KwaZulu-Natal province where the current study was conducted.(3) Due to the high level of non-response in this area of South Africa, the authors sought to develop a simple model, using available mortality data, to obtain an adjusted estimate of HIV prevalence.


To evaluate a model which uses mortality rates by HIV status, obtained from longitudinal demographic surveillance data, to estimate HIV prevalence among the population who refused to participate in HIV surveillance. This estimated prevalence among those who refused testing was then used to adjust the overall prevalence in the population in the same rural area of South Africa.


Population-based household survey in South Africa.


The longitudinal Africa Centre Demographic Information system (ACDIS) conducts a bi-annual household surveillance and annual individual surveillance in a rural district of KwaZulu-Natal Province. These systems conduct routine recording of deaths and the population at risk. In the individual surveillance, HIV testing is conducted. Data from the ACDIS were used to derive mortality rates for persons who tested negative and positive during HIV surveillance in 2005 (testers) and for those who refused to participate (non-testers) whose HIV status is unknown. Person years of exposure were estimated from the date of HIV test or date of visit for individuals who refused HIV testing and right censored at December 31, 2007, at death, outmigration, or household membership ending. Mortality rates were calculated by dividing deaths by person years of exposure for those who tested HIV negative, HIV positive, and HIV status unknown. The model uses these mortality rates by HIV status to infer the HIV prevalence among those who refused to participate in the HIV surveillance. The model assumes that mortality among those who HIV status is unknown is a weighted average of mortality rates among those whose HIV status is known. Non-parametric bootstrapping was used to estimate the 95% confidence intervals around the estimates.


Overall, 59% of the 21,305 individuals in the 2005 survey year did not participate in HIV testing. Overall mortality rates were significantly higher in the untested group (16.9 per 1000 person-years; 95%CI: 14.5,17.4) than in the tested group (11.6 per 1000 person years; 95%CI: 9.6,12.5) partly due to differences in the age and sex composition, but also due to actual mortality differences presumed to be the result of higher HIV prevalence in those untested. Adjusted HIV prevalence for females (15-49 years) was 31.6% (95%CI: 26.1,37.1) compared to observed prevalence of 25.2% (95%CI: 24.0,26.4). For males (15-49 years), adjusted HIV prevalence was 19.8% (95%CI: 14.8,24.8) compared to an observed prevalence of 13.2% (95%CI: 12.1,14.3). For both sexes combined, the adjusted HIV prevalence was 27.5% (95%CI: 23.6,31.3) and observed prevalence was 19.7% (95%CI 19.6,21.3). Prevalence was highest in the 25-34 years age group (50.4%;95%CI: 37.8,63.0), but the 95% CI was relatively wide. Overall, the observed prevalence based on those who consented to HIV testing in the 2005 survey underestimated the adjusted prevalence in the population as a whole by approximately 7 percentage points (37% relative difference).


These results suggest that the true HIV prevalence in the study populations is likely underestimated if only the observed data are considered. After adjusting for the untested population, there was a significant difference between the overall adjusted and observed prevalence of 7 percentage points, but little change in the pattern of age specific HIV prevalence rates.

Study Quality

Limitation of this model is the assumption that the population consenting to participate in HIV surveillance is not significantly different from those not consenting with respect to factors other than HIV that determine mortality. If non-response is high among persons with selected characteristics that pre-dispose them to lower or higher mortality from HIV, such as disease stage, or causes other than HIV, this will bias the adjustments. Due to high non-response rates, data on 29% of deaths with known HIV status was used to adjust the HIV prevalence in 59% of the population with unknown HIV status which may have biased the results. Small numbers of older persons and males limit the interpretation of adjusted data in these groups. In addition, this method cannot be used to adjust cross-sectional national HIV prevalence data where there is no prospective adult mortality data available.

Programmatic Implication

This approach uses a simple model to obtain an adjusted estimate of HIV prevalence in a population with a high non-response rate using HIV surveillance data and reliable mortality data. With this approach, only aggregate mortality and HIV prevalence data are needed. This is in contrast to more complex approaches which require individual-level analysis with detailed demographic data and complex computer simulations. This simple approach is applicable even in settings who have only one HIV sero-survey but longitudinal mortality follow-up of testers and non-testers. However, this method should be validated in other surveillance sites with lower refusal rates and reliable mortality data. Application of this method in populations with increasingly availability of ART will require further adjustments to validate this model, including potentially identifying people on ART as a separate group.


  1. García-Calleja JM, Gouws E, Ghys PD. National population based HIV prevalence surveys in sub-Saharan Africa: results and implications for HIV and AIDS estimates. Sex Transm Infect 2006 Jun; 82 Suppl 3:iii64-70.
  2. Boerma JT, Ghys, PD, Walker N. Estimates of HIV-1 prevalence from national population-based surveys as a new gold standard. Lancet 2003; 362(9399):1929-1931
  3. Shishana O, Rehle T, Simbayi LC, et al. South African national HIV prevalence, incidence, behavior, and communication survey 2008: A turning tide among teenagers? Cape Town: HSRC Press