AbstractIntroductionHIV prevalence estimates is a key indicator to inform the coverage and effectiveness of HIV prevention measures. Many countries including Kenya transitioned from sentinel surveillance to the use of routine antenatal care data to estimate the burden of HIV. Countries in Sub Saharan Africa reported several challenges of this transition, including low uptake of HIV testing and sub national / site-level differences in HIV prevalence estimates.MethodsWe examine routine data from Kwale County, Kenya, for the period January 2015 to December 2019 and predict HIV prevalence among women attending antenatal care (ANC) at 100% HIV status ascertainment. We estimate the bias in HIV prevalence estimates as a result of imperfect uptake of HIV testing and make recommendations to improve the utility of ANC routine data for HIV surveillance. We used a generalized estimating equation with binomial distribution to model the observed HIV prevalence as explained by HIV status ascertainment and region (Sub County). We then used marginal standardization to predict the HIV prevalence at 100% HIV status ascertainment.ResultsHIV testing at ANC was at 91.3%, slightly above the global target of 90%. If there was 100% HIV status ascertainment at ANC, the HIV prevalence would be 2.7% (95% CI 2.3-3.2). This was 0.3% lower than the observed prevalence. Similar trends were observed with yearly predictions except for 2018 where the HIV prevalence was underestimated with an absolute bias of -0.2%. This implies missed opportunities for identifying new HIV infections in the year 2018.ConclusionsImperfect HIV status ascertainment at ANC overestimates HIV prevalence among women attending ANC in Kwale County. However, the use of ANC routine data may underestimate the true population prevalence. There is need to address both community level and health facility level barriers to the uptake of ANC services.Key questionsWhat is already known?▪HIV surveillance estimates from antenatal clinics (ANC) can serve as a useful proxy for HIV prevalence trends in the general female population.▪Kenya has conducted multiple studies which have shown that national HIV prevalence estimates from sentinel surveillance and those from routine program data to be similar.▪However, these studies have also revealed ongoing challenges to the suitability of using routine data as compared to sentinel surveillance including sub optimal uptake of HIV testing and sub national/ site-level differences in HIV prevalence estimates.What are the new findings?▪HIV positive pregnant women are more likely to be tested at ANC as compared to HIV negative women, leading to higher HIV prevalence estimates among women attending ANC.▪Health facility level HIV prevalence estimates are lower than that of the general population.What do the new findings imply?▪HIV positive women are underrepresented in antenatal clinics.▪In Kwale County (and similar contexts), use of routine ANC data is still not a reliable method to estimate HIV prevalence, both at facility and community level.
National HIV prevalence estimates for sub-Saharan Africa: controlling selection bias with Heckman-type selection models
