study cases, and other factors, these seem unlikely to be a major factor operating in the Tacoma study.
The case-control study report (Hagan et al., in press) can be viewed as having design issues that affect confidence in the inferences that can be drawn. One is the selection of controls.
Controls for this study were injection drug users who were ascertained through entry into drug treatment or enrollment in the county's HIV testing program. Although cases in this study can be argued to be representative of all cases for hepatitis B (Pierce County is a CDC sentinel surveillance site), the selection of controls through the two stated mechanisms raises questions. The underlying issue is the extent to which new drug treatment admissions or enrollees for HIV testing are ''representative" of the overall uninfected population (including those not in treatment or uninterested in HIV testing). The descriptive epidemiologic literature on the characteristics of persons with a history of treatment versus no history of treatment suggests demographic differences; this literature hints at the possibility for bias when limiting ascertainment to treatment or testing programs (Alcabes, 1993). However, analyses of drug-use risk factors for infection for those in treatment compared with those out of treatment indicate that, despite differences in distribution of certain characteristics, patterns of drug-use risk behaviors and infection are similar for both groups. This argues that treatment samples are not biased as far as susceptibility to infections is concerned (Alcabes et al., 1993). Also, prospective studies of factors associated with entry into treatment note similar injection risk behaviors, but increased complications of drug abuse in those entering treatment (Schütz et al., 1994); this suggests that the effect of sampling treatment entrants for the purpose of studying viral infection is likely to be minor.
Nevertheless, the issue of potential of bias for controls selected among treatment entrants and injection drug users who arrive for HIV testing could be real if the needle exchange program makes active referrals into treatment and HIV testing. In the case of the Tacoma study, this apparently occurred, but the researchers made explicit the point that direct referrals were identified and excluded from analysis.
The issue remains whether the probability that controls who were picked from treatment and testing were somehow more likely to have used needle exchange than HBV-seronegative drug users not going into treatment or testing. If the proportion of drug users in a community that used needle exchange was small (e.g., 5 percent) and the proportion in this sample was high, then the discrepancy might be attributed to sampling. However, as Hagan and colleagues note, the proportion of the entire Tacoma community of injection drug users that has been estimated to be enrolled or covered by needle exchange is high (i.e., 50 to 70 percent; see Hagan et al., 1994:10-11)