WEPDC0102 - Poster Discussion Abstract
Measuring and accounting for outcome-correlated recruitment and geographic recruitment bias in a respondent-driven sample of people who inject drugs in Tijuana, Mexico
Presented by Abby Rudolph (United States).
A. Rudolph1, T. Gaines2, R. Lozada3, A. Vera2, K. Brouwer2
1Pacific Institute for Research and Evaluation, Calverton, United States, 2University of California San Diego, San Diego, United States, 3Pro-COMUSIDA, Tijuana Baja California, United States
sampling (RDS) is a recruitment/analytic approach for hidden populations, including
those at risk for HIV/AIDS, which aims to generate unbiased estimates of disease
prevalence. Its widespread use combined with its reliance on untested
assumptions creates a need for exploratory and diagnostic techniques for RDS
data. We will assess two potential sources of bias: geographic recruitment bias
and outcome-correlated recruitment.
surveys assessed demographics, drug/sex behaviors, residential coordinates, and
recruiter-recruit ties among 1,048 people who inject drugs (PWID) recruited
through RDS in Tijuana, Mexico.
Blood samples were obtained through venipuncture. Simulations assessed
geographic and network clustering of active syphilis (RPR titers>1:8).
Gender-specific predicted probabilities
were estimated using logistic regression with GEE (clustered on RDS
chain) and robust standard errors.
prevalence was 4.4% (men=3.5%, women=9.7%).
unadjusted prevalence of active syphilis was 5.7% among men and 16.6% among
women. RDS-weighted estimates were not significantly different. Syphilis
clustered spatially in the Zona Norte, a neighborhood known for drug and sex
markets. Network simulations revealed geographic recruitment bias; respondents living/working/injecting/buying
drugs in the Zona Norte preferentially recruited others living/working/injecting/buying
drugs there. Network simulations also identified non-random recruitment by
syphilis; those with syphilis were 3 times more likely to recruit or be
recruited by other syphilis-infected PWID (P< 0.0001).
Gender-specific prevalence estimates accounting for clustering among RDS chains
were highest among those
directly/indirectly connected to syphilis cases and who lived/worked/injected/bought
drugs in the Zona Norte (men:15.9%, women:25.6%) and
lowest among those without network or neighborhood-level exposures (men:3.0%,
women:6.1%). Prevalence among those with a network or neighborhood exposure
were ~3% for men and 6-13% for women. Significant interaction between network
and neighborhood covariates was observed only for men (P< 0.001).
Conclusions: Our findings underscore the importance of considering
spatial and network dependencies and recruitment patterns when estimating
disease prevalence in key populations. Understanding how measures may be biased
and the limitations on the representativeness of estimates has important
implications for accurately estimating the prevalence of HIV/STIs for disease
surveillance and epidemiologic research. Estimates which do not account for
these dependencies and recruitment biases could lead to inaccurate prevalence
estimates with artificially narrow confidence intervals.
Back to the Programme-at-a-Glance