Improving Underestimation of HIV Prevalence in Surveys Using Time-Location Sampling.

We sought to find a method that improves HIV estimates obtained through time-location sampling (TLS) used to recruit most-at-risk populations (MARPs). The calibration on residuals (CARES) method attributes weights to TLS sampled individuals depending on the percentile to which their logistic regression residues belong. Using a real country database, provided by EMIS-2010, with 9591 men who have sex with men (MSM) and an HIV prevalence of 12.1%, we simulated three populations (termed "pseudo-populations") with different levels of HIV. From each pseudo-population, 1000 TLS samples were drawn, and the HIV prevalence estimated by the TLS method and by the CARES method were recorded and compared with the HIV prevalence of the 9591 men. Results showed that the CARES method improves estimates given by the TLS method by getting closer to the real HIV prevalence.