Article Text

PDF

Epidemiology poster session 4: Methodological aspects: RDS & recruitement
P1-S4.34 Can RDS be used to recruit unbiased samples from the same population with repeated sampling?
  1. J Risser,
  2. P Padgett,
  3. J Montealegre
  1. School of Public Health, University of Texas, Houston, USA

Abstract

Background Respondent Driven Sampling (RDS) is a modified snowball sampling method that allows for recruitment of a probability sample, even when we cannot enumerate the population of interest. The National HIV Behavioural Surveillance project, designed to assess trends in risky behaviours, used RDS to recruit injection drug users (IDU) and heterosexuals at high risk of HIV infection (HET). In order to examine changes in behaviours over time, the sampling method must recruit homogeneous populations, so that observed differences across cycles can be attributed to changes in behaviour, rather than to errors in recruitment. This analysis was designed to determine if crude and adjusted measures within and between the IDU and HET cycles, are homogeneous on basic demographic characteristics. Our goal to assess if the samples may be considered to be from the same target population.

Methods Data are from the Houston Texas site of NHBS-IDU 1 and 2 (2005, 2009), and HET 1 and 2 (2006, 2010). Adjusted population prevalence estimates are calculated using RDSAT, adjusting for design effect and sampling characteristics. Estimates were compared using the Mantel Haenszel test for heterogeneity.

Results Comparing IDU1 to IDU2, we found similar population estimates (p for heterogeneity >0.05) for age 40–49 years; Black race; had current health insurance; and currently homeless. The populations differed (p for heterogeneity <0.05) by the proportion that: graduated from high school; were arrested in the last year; and visited a doctor in the last year. Comparing HET1 and HET2, we found similar population estimates for the proportions that: had health insurance; visited a doctor in the last year; were arrested in the last year; graduated from high school; and were aged 30–49. The populations differed in the percent: Black; currently homeless; and aged 18–29.

Conclusion Using RDS to assess behaviour changes over time will be difficult if the study samples do not represent a fixed population. We looked at variables we thought would be stable over the two cycles and were surprised to find instability. With these differences, we cannot attribute the observed changes in risky behaviour solely to prevention efforts that occurred between the surveillance cycles. However, RDS remains an exceedingly easy and efficient sampling method and with appropriate multivariable analysis techniques can be used with repeated samples over time.

Statistics from Altmetric.com

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.