Objective(s): Investigate the effect of different study protocols with respect to the disclosure of HIV test results on non-response bias in HIV prevalence estimates.
Design: Nine-month surveillance of hospital admissions in Addis Ababa in which patients were approached for an HIV test. Patients had the choice between three consent levels: testing and post-test counseling (levelA); testing without post-test counseling (level B); and total refusal (level C). For all patients, information was collected on basic sociodemographic background characteristics and admission diagnosis. That information is used to predict HIV status in those who refuse testing.
Methods: We first investigate the covariates of different levels of consent. We then quantify bias in HIV prevalence estimates due to refusal for testing via Heckman regression models that account for sample selection.
Results: We find that refusal positively correlates with the likelihood of infection and that non-response bias in HIV prevalence surveys depends on the study protocol: if disclosure of HIV status is implied in study participation, the bias is likely to be much larger than in a scenario where respondents can opt out of post-test counselling. We also find that consent for testing increased since the introduction of antiretroviral therapy in Ethiopia. Other covariates of refusal are age, gender, marital status, educational status and counsellor. Conclusions: Disclosure or non-disclosure of test results is an important consideration in studies that wish to minimize non-response bias in HIV prevalence surveys. The availability of ART is likely to reduce refusal rates.