Institutional members access full text with Ovid®

Underestimation of HIV prevalence in surveys when some people already know their status, and ways to reduce the bias

Floyd, Siana; Molesworth, Annaa,b; Dube, Albertb; Crampin, Amelia C.a,b; Houben, Reina,b; Chihana, Menardb; Price, Alisona,b; Kayuni, Ndoliweb; Saul, Jacquelinea; French, Neila,b; Glynn, Judith R.a

doi: 10.1097/QAD.0b013e32835848ab
Epidemiology and Social

Objective: To quantify refusal bias due to prior HIV testing, and its effect on HIV prevalence estimates, in general-population surveys.

Design: Four annual, cross-sectional, house-to-house HIV serosurveys conducted during 2006–2010 within a demographic surveillance population of 33 000 in northern Malawi.

Methods: The effect of prior knowledge of HIV status on test acceptance in subsequent surveys was analysed. HIV prevalence was then estimated using ten adjustment methods, including age-standardization; multiple imputation of missing data; a conditional probability equations approach incorporating refusal bias; using longitudinal data on previous and subsequent HIV results; including self-reported HIV status; and including linked antiretroviral therapy clinic data.

Results: HIV test acceptance was 55–65% in each serosurvey. By 2009/2010 79% of men and 85% of women had tested at least once. Known HIV-positive individuals were more likely to be absent, and refuse interviewing and testing. Using longitudinal data, and adjusting for refusal bias, the best estimate of HIV prevalence was 7% in men and 9% in women in 2008/2009. Estimates using multiple imputations were 4.8 and 6.4%, respectively. Using the conditional probability approach gave good estimates using the refusal risk ratio of HIV-positive to HIV-negative individuals observed in this study, but not when using the only previously published estimate of this ratio, even though this was also from Malawi.

Conclusion: As the proportion of the population who know their HIV-status increases, survey-based prevalence estimates become increasingly biased. As an adjustment method for cross-sectional data remains elusive, sources of data with high coverage, such as antenatal clinics surveillance, remain important.

aLondon School of Hygiene and Tropical Medicine, Keppel Street, London, UK

bKaronga Prevention Study, Chilumba, Karonga district, Malawi.

Correspondence to Sian Floyd, London School of Hygiene and Tropical Medicine, Keppel Street, WC1E 7HT, London, UK. Tel: +44 2076127888; fax: +44 2076368739; e-mail:

Received 1 May, 2012

Revised 14 July, 2012

Accepted 18 July, 2012

Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's Website (

© 2013 Lippincott Williams & Wilkins, Inc.