Reproducibility of Respondent-Driven Sampling (RDS) in Repeat Surveys of Men Who have Sex with Men, Unguja, Zanzibar

Ahmed Khatib, Shaaban Haji, Maryam Khamis, Christen Said, Farhat Khalid, Mohammed Dahoma, Ameir Ali, Asha Othman, Susie Welty, Willi McFarland

To assess the reproducibility of respondent-driven sampling (RDS) in obtaining comparable samples across two survey rounds, we conducted integrated bio-behavioral surveillance surveys (IBBSS) using RDS in 2007 and 2011 among men who have sex with men (MSM) on Unguja island in Zanzibar. Differences in the two rounds were assessed by comparing RDS-adjusted population estimates, stratified estimates, and bottleneck plots. Participants in the 2011 survey round were younger (31.4 vs. 9.9% under 19 years old, p < 0.001), more likely to have tested for HIV in the last year (53.7 vs. 10.6%, p < 0.001), and less likely to have injected drugs in the last 3 months (1.0 vs. 23.2%, p < 0.001) compared to participants in the 2007 round. HIV prevalence was 12.3% in 2007 compared to 2.6% in 2011 (p < 0.001). The difference in HIV prevalence persisted after stratifying and adjusting for known differences in the two surveys rounds. Bottleneck plots suggest that recruitment chains were "trapped" in the social networks of MSM who injected drugs to a greater extent in 2007 than in 2011. We conclude that the two rounds of RDS sampled different subsets of the MSM population on Unguja, particularly with respect to inclusion of MSM within the social networks of people who inject drugs. Findings underscore the need to evaluate the reproducibility of RDS in repeated rounds of IBBSS and to develop new sampling methods for key populations at high risk for HIV in order to track the epidemic, develop evidence-based prevention and care programs, and assess their impact.