demographic characteristics, medical and drug treatment history, and recent drug-use and sexual behavior of respondents. Serologic screening for HIV-1 antibodies used venous specimens. Specimens repeatedly reactive in whole-virus lysate enzyme-linked immunosorbent assays (ELISA) were confirmed by Western blot.

The sample of 641 seronegative injection drug users was primarily male (76 percent); about two-thirds were age 40 or younger; somewhat more than half were African American, 26 percent white, and 18 percent Hispanic; and almost half reported injecting twice a day or more at recruitment. As of September 1992—4 years after the program began—the status of the cohort was 83 (13 percent) observed to have seroconverted, 42 (7 percent) had died and were seronegative at last serologic evaluation, 208 (32 percent) were seronegative at last evaluation and were lost to follow-up, and 308 (48 percent) were seronegative at the end of the study period. The cohort had second measures on 88 percent (513 of 641) of the baseline sample, and over-time follow-up response rates were 85, 84, 82, 78, and 68 percent, respectively, of the base sample in each of the successive waves.

Figure 6.1 shows the rate of HIV seroconversions per 100 person years (py). A decline in seroincidence was observed, from 8.4 per hundred py at the first follow-up wave to approximately 2.0 per hundred py after 1990.

Table 6.1 presents the percentages of the cohort reporting that they had engaged in HIV-related risk behavior since their last interview. The denominator for each wave is the number observed at each wave at risk of seroconverting (i.e., prior seroconverters are eliminated). The gender-specific prevalence of sexual risk behavior is nearly identical at each wave.

Survival analysis (i.e., Cox proportional hazard survival model) of HIV seroconversion incorporated the two behavioral risk variables, injecting risk and sexual risk, that varied over interviewing waves, as well as gender, age, race, and site/neighborhood. The only variable found to be associated with HIV seroconversion was the time-dependent behavioral risk variable "injecting risk" (RR = 9.8, p < .001). The time-dependent sexual risk variable and all demographic variables were not found to be associated with seroconverting.

Risky injection behavior was reported by 100 percent of the study cohort at baseline (1988) and only 14 percent at wave six (1992). (Neither risk behavior nor HIV status was an initial recruitment factor.) The observed decline in seroincidence in this cohort appears to have resulted from changes in drug-use risk behaviors attributable, at least in part, to the intervention.

This study design obviously cannot completely rule out alternative explanations for the observed decline in seroconversion in the studied cohort. One such alternative explanation could be that the findings were the result of other prevention efforts, especially given the increased amount of media attention directed at HIV risk reduction and other community efforts occurring