• Record: found
  • Abstract: found
  • Article: found
Is Open Access

Non-response bias in estimates of HIV prevalence due to the mobility of absentees in national population-based surveys: a study of nine national surveys

, ,

Sexually Transmitted Infections

BMJ Publishing Group

Read this article at

      There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.



      To measure the bias in national estimates of HIV prevalence in population-based surveys caused by mobility and refusal to test.


      Data from nine demographic and health surveys and AIDS indicator surveys were used. Non-responders were divided into three groups: (i) “refusals” who were interviewed but not tested; (ii) “refusals” who were present in the household but not interviewed or tested; and (iii) “absentees” who were absent from the household. Correction for HIV status was made for the non-responders using multiple imputation methods with logistic regression models based on a common set of household-level and individual-level sociodemographic and behavioural factors for those tested and stratified by mobility status.


      The non-response groups were corrected to have higher risks of HIV than those who participated in the HIV tests, although these were only detected to be statistically significant in some of the countries. In Lesotho, the corrected prevalence for the absent household members was significantly higher than for those who were present in the household. However, the adjusted prevalences differed by less than a percentage point from the prevalences observed among those who were tested, so the overall effects of non-response on national estimates of HIV prevalence are minimal.


      The results indicate that the mobility of absentees does not substantially bias estimates of HIV prevalence from population-based surveys. None the less, if levels of non-response are high or if non-responders differ greatly from those who participate in HIV testing with respect to HIV status, non-response could still bias national estimates of HIV prevalence.

      Related collections

      Most cited references 17

      • Record: found
      • Abstract: found
      • Article: not found

      HIV testing in national population-based surveys: experience from the Demographic and Health Surveys.

      To describe the methods used in the Demographic and Health Surveys (DHS) to collect nationally representative data on the prevalence of human immunodeficiency virus (HIV) and assess the value of such data to country HIV surveillance systems. During 2001-04, national samples of adult women and men in Burkina Faso, Cameroon, Dominican Republic, Ghana, Mali, Kenya, United Republic of Tanzania and Zambia were tested for HIV. Dried blood spot samples were collected for HIV testing, following internationally accepted ethical standards. The results for each country are presented by age, sex, and urban versus rural residence. To estimate the effects of non-response, HIV prevalence among non-responding males and females was predicted using multivariate statistical models for those who were tested, with a common set of predictor variables. Rates of HIV testing varied from 70% among Kenyan men to 92% among women in Burkina Faso and Cameroon. Despite large differences in HIV prevalence between the surveys (1-16%), fairly consistent patterns of HIV infection were observed by age, sex and urban versus rural residence, with considerably higher rates in urban areas and in women, especially at younger ages. Analysis of non-response bias indicates that although predicted HIV prevalence tended to be higher in non-tested males and females than in those tested, the overall effects of non-response on the observed national estimates of HIV prevalence are insignificant. Population-based surveys can provide reliable, direct estimates of national and regional HIV seroprevalence among men and women irrespective of pregnancy status. Survey data greatly enhance surveillance systems and the accuracy of national estimates in generalized epidemics.
        • Record: found
        • Abstract: not found
        • Article: not found

        Estimates of HIV-1 prevalence from national population-based surveys as a new gold standard.

          • Record: found
          • Abstract: found
          • Article: not found

          Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia.

          To establish population-based HIV survey data in selected populations, and to assess the validity of extrapolation from HIV sentinel surveillance amongst antenatal clinic attenders (ANC) to the general population. In a population survey, adults aged > or = 15 years were selected by stratified random cluster sampling (n = 4195). The survey was carried out in catchment populations of clinics used for national HIV surveillance. The methodology allows detailed comparisons of HIV infection patterns to be made in two areas (urban and rural). Whereas the sentinel surveillance used serum-based HIV testing, the population survey used saliva (93.5% consented to provide a saliva sample). Surveillance of ANC tended to underestimate the overall HIV prevalence of the general population, but differences were not statistically significant. In the urban area, the adjusted overall HIV prevalence rate of ANC (aged 15-39 years) was 24.4% [95% confidence interval (CI), 20.9-28.0] compared with 26.0% (95% CI, 23.4-28.6) in the general population. The respective rural estimates were 12.5% (95% CI, 9.3-15.6) versus 16.4% (95% CI, 12.1-20.6). Age-specific prevalence rates showed ANC to overestimate infection in teenagers (aged 15-19 years), whereas in the reverse direction of those aged > or = 30 years. Teenagers analysed by single year of age revealed both ANC and women in the general population with about the same steep increase in prevalence by age, but the former at consistently higher rates. Extrapolations might be biased substantially due to the higher pregnancy rates amongst uninfected individuals. ANC-based data might draw a rather distorted picture of current dynamics of the HIV epidemic. Even though representing an obvious oversimplification, extrapolations of overall prevalence rates may correlate with that of the general population.

            Author and article information

            Centre for Population Studies, London School of Hygiene and Tropical Medicine, London, UK
            Author notes
            Correspondence to: Milly Marston, Centre for Population Studies, London School of Hygiene and Tropical Medicine, London WC1B 3DP, UK; milly.marston@
            Sex Transm Infect
            Sexually Transmitted Infections
            BMJ Publishing Group (BMA House, Tavistock Square, London, WC1H 9JR )
            August 2008
            22 July 2008
            22 July 2008
            : 84
            : Suppl_1 , Improved data, methods and tools for the 2007 HIV and AIDS estimates and projections
            : i71-i77
            © Marston et al 2008

            This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.


            Sexual medicine


            Comment on this article