Evaluating Outcome-Correlated Recruitment and Geographic Recruitment Bias in a Respondent-Driven Sample of People Who Inject Drugs in Tijuana, Mexico

Abstract

Respondent-driven sampling’s (RDS) widespread use and reliance on untested assumptions suggests a need for new exploratory/diagnostic tests. We assessed geographic recruitment bias and outcome-correlated recruitment among 1048 RDS-recruited people who inject drugs (Tijuana, Mexico). Surveys gathered demographics, drug/sex behaviors, activity locations, and recruiter-recruit pairs. Simulations assessed geographic and network clustering of active syphilis (RPR titers≥1:8). Gender-specific predicted probabilities were estimated using logistic regression with GEE and robust standard errors. Active syphilis prevalence was 7% (crude: men=5.7% and women=16.6%; RDS-adjusted: men=6.7% and women=7.6%). Syphilis clustered in the Zona Norte, a neighborhood known for drug and sex markets. Network simulations revealed geographic recruitment bias and non-random recruitment by syphilis status. Gender-specific prevalence estimates accounting for clustering were highest among those living/working/injecting/buying drugs in the Zona Norte and directly/indirectly connected to syphilis cases (men:15.9%, women:25.6%) and lowest among those with neither exposure (men:3.0%, women:6.1%). Future RDS analyses should assess/account for network and spatial dependencies.

The relationship between life stressors and drug and sexual behaviors among a population-based sample of young Black men who have sex with men in Chicago

“…Respondent Driven Sampling (RDS) has been widely used in public health research as a method to enable valid statistical inference in samples of “hard-to-reach populations (Jenness, Neaigus, Wendel, Gelpi-Acosta, & Hagan, 2014; Rudolph, Gaines, Lozada, Vera & Brouwer, 2014). Therefore, methods to generate valid estimates and standard errors from RDS studies have received much attention in the literature (Gile & Handcock, 2010; Goel & Salganik, 2009; Goel & Salganik, 2010; Salganik & Heckathorn, 2004).…”

Polydrug Use and HIV Risk Among People Who Inject Heroin in Tijuana, Mexico: A Latent Class Analysis

“…From the initial group, 24 seeds recruited eligible peers. Additional information on recruitment procedures, tolerance, sample convergence, recruitment chains, and sample characteristics has previously been reported (Brouwer et al, 2012; Rudolph et al, 2014; Strathdee et al, 2008). Participants completed a quantitative survey and underwent testing for HIV, TB, and syphilis at baseline and every six months thereafter for three years (N=1,056).…”

“…To account for this recruitment homophily on polydrug use class, Model 2 controlled for each participant’s recruiter’s polydrug use class. Additionally, to account for the increased similarity among individuals in the same recruitment chain compared with those in different chains, Model 2 used a generalized estimation equation approach (like the approach used by Rudolph et al 2014) which clustered on recruitment chain membership in STATA 10 (StataCorp 2007). …”

Social venue range and referral chain impact: Implications for the sampling of hidden communities

“…With RDS, the researcher keeps track of referral linkages among participants and adjusts for the influence of well-connected participants on the composition of the sample. RDS is widely utilized in public health studies, mostly in an attempt to generate samples of people for which official rosters or population-based samples are unfeasible, such as men who have sex with men (MSM) [ 7 , 8 , 9 , 10 , 11 ].…”