Optimizing respondent-driven sampling to find undiagnosed HIV-infected people who inject drugs

We evaluated whether identification of undiagnosed HIV-infected people who inject drugs (PWID) via respondent-driven sampling (RDS) can be enhanced through a precision RDS (pRDS) approach.

First, using prior RDS data from PWID in India, we built a prediction algorithm for recruiting undiagnosed HIV-infected PWID. pRDS was tested in Morinda, Punjab where participants were randomly assigned to standard or pRDS. In the standard RDS approach, all participants received two recruitment coupons. For pRDS, the algorithm determined an individual's probability of recruiting an undiagnosed PWID, and individuals received either two (low probability) or five (high probability) coupons. Efficiency in identifying undiagnosed HIV-infected PWID for the RDS approaches was evaluated in two ways: the number needed to recruit (NNR) and identification rate/week.

Predictors of recruiting undiagnosed PWID included HIV/HCV infection, network size, syringe services utilization, and injection environment. 1631 PWID were recruited in Morinda. From the standard RDS approach, 615 were recruited, including 39 undiagnosed; from pRDS, 1012 were recruited, including 77 undiagnosed. In pRDS, those with higher predicted probability were more likely to recruit others with HIV/HCV co-infection, undiagnosed and viremic HIV, and who utilized services. pRDS had a significantly higher identification rate of undiagnosed PWID (1.5/week) compared with the standard (0.8/week). The NNR for pRDS (13.1) was not significantly lower than the standard approach (15.8).

pRDS identified twice as many undiagnosed and viremic PWID significantly faster than the standard approach. Leveraging RDS or similar network-based strategies should be considered alongside other strategies to ensure meeting UNAIDS targets.