Abstract
Participation bias (selection bias) may be a problem in studies that attempt to evaluate the effects of needle-exchange programs (NEPs). The present study looked at only those injection drug users (IDUs) who were randomly placed in the needle-exchange condition in a two-arm randomized clinical trial of needle exchange. Time to follow-up between the experimental NEP condition (n = 296; median = 261 days) and pharmacy sales condition (n = 304; median = 256 days) was not statistically different [chi(2) (1, N = 600) = 0.42, P =.52]. Within the NEP condition, a similar analysis comparing time to follow-up between IDUs who used the NEP (n = 65; median = 199 days) and those who refrained from using the NEP (n = 231; median = 286 days) was highly significant, chi(2) (1, n = 296) = 17.3, P=.0001. The final logistic regression model predicting use of the NEP was the log 10 transformation of the number of times injected heroin in the last 30 days (odds ratio [OR ]=4.9, confidence interval [CI ]2.8, 8.9), ever injected amphetamine in the last 30 days (OR =4.9, CI 1.09, 22.5), and ever shared injection equipment in the last 30 days (OR =2.9, CI 1.5, 5.5). Within the NEP condition, follow-up rates differed between those who used the NEP and those who did not use it. Of drug users randomly assigned to an NEP, the ones who actually used the NEP had higher levels of drug use. Predictors of who used the NEP were consistent with the studies of Hagen et al. This study demonstrates that selection bias occurs among IDUs who use NEPs.