Adjusting for non-participation bias at an HIV surveillance site in rural South Africa

McGovern, M.E. and Marra, G. and Radice, Rosalba and Canning, D. and Newell, M.L. and Bärnighausen, T. (2015) Adjusting for non-participation bias at an HIV surveillance site in rural South Africa. Journal of the International AIDS Society 18 , ISSN 1758-2652.

Abstract

- Introduction: HIV testing is a cornerstone of efforts to combat the HIV epidemic, and testing conducted as part of surveillance sites provides invaluable data on the spread of infection and the effectiveness of campaigns to reduce the transmission of HIV. However, participation rates can be low, and if respondents systematically select into not testing because they know or suspect they are HIV positive (and fear disclosure), prevalence estimates may be biased. We implement Heckman-type selection models which can be used to adjust for missing data which are not missing at random, and establish the extent of selection bias in a population-based study of a hyperendemic community in rural South Africa. - Methods: Using data from residents of the 2009 Africa Centre Surveillance cohort in KwaZulu-Natal, where 5,565 women (35%) and 2,567 men (27%) provided blood for an HIV test, we account for missing data using interviewer identity as a selection variable which predicts testing participation but not HIV status. Our approach involves using our selection variable to examine the HIV status of residents who would ordinarily refuse to test, except they were allocated a persuasive interviewer. Our copula model relaxes a key parametric assumption typically used to implement this method. - Results: Our selection model HIV prevalence estimate of 33% (95% CI 27-40) for women compares to 24% from respondents who participated in testing, or 27% using imputation analysis to predict missing data on HIV status. For men, we find 25% HIV prevalence (95% CI 15-35) using the selection model, compared to 16% among those who participated in testing and 18% with imputation. We provide new confidence intervals which correct for the fact that relationship between testing and HIV status is unknown and requires estimation. - Conclusions: We confirm the feasibility and value of adopting selection models to account for missing data in Demographic Surveillance Systems. This approach should be routinely used in HIV prevalence estimation to assess potential selection bias. Elements of survey design, such as interviewer identity, present the opportunity to adopt this approach. Where non-participation is high, confidence intervals are much wider than standard approaches to dealing with missing data suggest.

Metadata

Statistics

DownloadsShow export options

Export as XMLJSONCSV

Activity Overview

6 month trend

225Downloads

6 month trend

437Hits

Additional statistics are available via IRStats2.

Archive Staff Only (login required)

Edit/View Item