Evaluation of respondent-driven sampling in seven studies of people who use drugs from rural populations: Findings from the rural opioid initiative

Georgiy V. Bobashev; William A. Zule; Abby E Rudolph; Robin M Nance; Georgiy V. Bobashev; Daniel Brook; Wajiha Z Akhtar; Ryan R Cook; Hannah L. Cooper; Peter D Friedmann; Simon D W Frost; Vivian F. Go; Wiley D Jenkins; Philip T Korthuis; William C Miller; Mai T Pho; Stephanie A Ruderman; David W Seal; Thomas J Stopka; Ryan P Westergaard; April M Young; William A. Zule; Judith I Tsui; Heidi M Crane; Bridget M Whitney; Joseph A C Delaney

Evaluation of respondent-driven sampling in seven studies of people who use drugs from rural populations

Findings from the rural opioid initiative

Rudolph, A. E., Nance, R. M., Bobashev, G., Brook, D., Akhtar, W., Cook, R., Cooper, H. L., Friedmann, P. D., Frost, S. D. W., Go, V. F., Jenkins, W. D., Korthuis, P. T., Miller, W. C., Pho, M. T., Ruderman, S. A., Seal, D. W., Stopka, T. J., Westergaard, R. P., Young, A. M., ... Delaney, J. A. C. (2024). Evaluation of respondent-driven sampling in seven studies of people who use drugs from rural populations: Findings from the rural opioid initiative. BMC Medical Research Methodology, 24(1), 94. Article 94. https://doi.org/10.1186/s12874-024-02206-5

Copy citation

Abstract

BACKGROUND: Accurate prevalence estimates of drug use and its harms are important to characterize burden and develop interventions to reduce negative health outcomes and disparities. Lack of a sampling frame for marginalized/stigmatized populations, including persons who use drugs (PWUD) in rural settings, makes this challenging. Respondent-driven sampling (RDS) is frequently used to recruit PWUD. However, the validity of RDS-generated population-level prevalence estimates relies on assumptions that should be evaluated.

METHODS: RDS was used to recruit PWUD across seven Rural Opioid Initiative studies between 2018-2020. To evaluate RDS assumptions, we computed recruitment homophily and design effects, generated convergence and bottleneck plots, and tested for recruitment and degree differences. We compared sample proportions with three RDS-adjusted estimators (two variations of RDS-I and RDS-II) for five variables of interest (past 30-day use of heroin, fentanyl, and methamphetamine; past 6-month homelessness; and being positive for hepatitis C virus (HCV) antibody) using linear regression with robust confidence intervals. We compared regression estimates for the associations between HCV positive antibody status and (a) heroin use, (b) fentanyl use, and (c) age using RDS-1 and RDS-II probability weights and no weights using logistic and modified Poisson regression and random-effects meta-analyses.

RESULTS: Among 2,842 PWUD, median age was 34 years and 43% were female. Most participants (54%) reported opioids as their drug of choice, however regional differences were present (e.g., methamphetamine range: 4-52%). Many recruitment chains were not long enough to achieve sample equilibrium. Recruitment homophily was present for some variables. Differences with respect to recruitment and degree varied across studies. Prevalence estimates varied only slightly with different RDS weighting approaches, most confidence intervals overlapped. Variations in measures of association varied little based on weighting approach.

CONCLUSIONS: RDS was a useful recruitment tool for PWUD in rural settings. However, several violations of key RDS assumptions were observed which slightly impacts estimation of proportion although not associations.