shows the sample characteristics at baseline and nine months. Methamphetamine users were over two years younger than others, on average. Methamphetamine was disproportionately used by Whites, as expected. At baseline, those who did not use methamphetamine (a group composed of 485 users of other drugs and 64 nonusers of drugs) had more prior treatment episodes than did methamphetamine users. Methamphetamine users at baseline were more likely to have graduated from high school but less likely to have been employed prior to incarceration than others. There were very few people who reported having same-sex partners. A significant majority of MA users were from the most western of the research sites (Oregon) and the fewest were in the northeastern site (Connecticut).
There were no significant differences in marital/partner status by methamphetamine use. We note that more respondents reported “living with a spouse or partner” than reported either being “legally married” or “living as married.” It appears, therefore, that respondents who were cohabiting were reluctant to report that they were “living as married.” Based on the assumption of its greater accuracy, the “living with a spouse or partner” variable was used in subsequent analyses.
displays the bivariate odds ratios relating risk behaviors (methamphetamine use, injection drug use that also involved potential HIV risk, and sexual risk behaviors) at each of the two time points. There was a marked decrease in the prevalence of risk behaviors nine months after incarceration, as compared with their prevalence before prison. Despite this decrease, methamphetamine use was significantly related to injection risk at baseline and to all types of sexual risk behavior at both time points. However, because cell sizes were much smaller at nine months, confidence intervals were quite wide for the odds ratios relating methamphetamine use to unprotected sex with a casual partner and to unprotected sex while a partner was high. Indeed, because only 11 people engaged in injection risk behaviors at nine months, two of the odds ratios involving this variable could not be estimated precisely enough to be meaningful. Other risk behaviors were not interrelated with each other as clearly or consistently, except in the trivial case when one behavior was a subset of one another (e.g., those having unprotected sex and those having unprotected sex when at least one of the partners was high), and these odds ratios are not displayed. We also note the attrition of 164 cases between the two time points. A set of descriptive analyses (not shown) indicated that there were no significant differences in the baseline analysis variables between those who were lost to follow-up and those who completed the nine-month interview. An additional set of descriptive analyses (also not shown) showed no significant differences in the baseline analysis variables between non-MA users and MA users who were lost to follow-up.
Bivariate Associations Among Risk Behaviors at Baseline and Nine Months (Odds Ratio, (95% Confidence Interval))
also shows the interrelations among risk behaviors over time. Those who used methamphetamine at baseline were far more likely than others to be using methamphetamine at nine months. Baseline methamphetamine users also were more likely than others to engage in all other forms of risk behavior at nine months. This consistency in association over time was not observed among most other risk behaviors, except for those who had unprotected sex with casual partners at baseline.
shows the multivariate analyses of methamphetamine use and sexual risk behaviors at baseline. As previously stated, the number of people reporting same-sex partners was too few to be able to control for this factor in the multivariate analyses at either time point. Methamphetamine use (col. 1) was strongly related to geographic area and to ethnicity, as expected, and to age. Also, as hypothesized, methamphetamine use was related to all three sexual risk behaviors, controlling for other factors (cols. 2–4). Sexual risk behavior was more likely to occur at the Kentucky site than at the Oregon site. Those with a greater number of prior treatments were more likely than those with fewer prior treatments to have unprotected sex with a casual partner. Living with a spouse or partner predicted engaging in both unprotected sex and unprotected sex while a partner was high.
Logistic Regression Analyses of Methamphetamine Use and Risk Behavior at Baseline
displays the multivariate analyses of methamphetamine use and sexual risk behaviors at nine months. Methamphetamine use at baseline was not included in the model of nine-month methamphetamine use because of the prohibitively small number of clients (eight out of 48) who reported using methamphetamine at nine months but did not report using it at baseline. Including this variable produced an untenably high odds ratio and reduced estimated effects of other variables to nonsignificance.
Logistic Regression Analyses of Methamphetamine Use and Risk Behavior at Nine Months
As hypothesized, methamphetamine use was related to all sexual risk behaviors, after controlling for other factors, although as in , confidence intervals were very wide in the analyses of unprotected sex with a casual partner and while a partner was high. Baseline methamphetamine use was not related to sexual risk behaviors, suggesting that its relationship with the nine-month risk behaviors shown in was entirely accounted for by current methamphetamine use. The same thing was found for the relationship between baseline sexual risk behaviors and their corresponding nine-month sexual risk behaviors, except that unprotected sex with a casual partner at nine months was predicted by the same baseline behavior. The influence of geography on methamphetamine use persisted at nine months. Unlike at baseline, however, compared with Oregon, unprotected sex was less likely to occur at all other sites, and unprotected sex with a casual partner was less likely to occur in Colorado. Older people were less likely than younger people to engage in any of the sexual risk behaviors. Number of prior treatments was not related to unprotected sex with a casual partner, as it was at baseline. Living with a spouse or partner still was related to unprotected sex in general and while someone was high.
Sensitivity analyses using GLMMs produced a number of differences compared with the models in and . Several were changes in the significance level of findings that nonetheless stayed significant or became significant at a more rigorous level. We report only others in which a finding changed from significant to nonsignificant or the reverse. There were no substantive changes in the magnitudes of the estimated effects.
GLMM produced a significant effect of living with a spouse or partner on methamphetamine use. The effect of age on unprotected sex became nonsignificant, but older age became significantly associated with a lower risk of unprotected sex with a casual partner. The effect of number of prior treatments on unprotected sex with a casual partner became nonsignificant.
(Nine-Month Follow-up) Models
In the models of nine-month behavior, GLMM again indicated an insignificant effect of age upon unprotected sex. In addition, being Hispanic (vs. White) became significantly associated with less risk of unprotected sex while a partner was high.