Non-participation bias in unlinked anonymous HIV-prevalence surveys in England and Wales

Abstract

The objective was to assess the potential bias in unlinked anonymous HIV-seroprevalence surveys from objections to specimens being included. Objection rates in seroprevalence surveys were examined. Statistically large clusters of objections were considered to be the result of health care worker behaviour, and were disregarded. Underlying objection rates were estimated from remaining data and compared to seroprevalence. Overall objection rates approached or exceeded seroprevalence in many participating centres. However, underlying objection rates declined with time while prevalences were generally unchanging. Also, underlying rates correlated poorly with observed seroprevalences. Findings were therefore consistent with processes producing the clusters of objections and underlying objection rates independently of serostatus of individuals. Although national seroprevalence estimates produced by the surveys are reasonably free from objection bias, regional seroprevalence estimates outside London remain vulnerable to bias as a result of some centres returning data whose quality cannot be guaranteed.