英语论文网

he days employed, we would test the other employment measures. We first modeled days employed as a Poisson distribution, which yielded a poor model fit.27 We then modeled the data with negative binomial regression models with a log link function that provided a good model fit. After finding a statistically significant association between group and number of days a person was employed during a month, we also examined whether this association would also be significant if employment outcome was measured as any day of work within a month and as any full-time employment during amonth. In step 4, we examined the prospective association between abstinence and outcomes in 2 ways: (1) we modeled whether the proportion of days abstinent in the first 12 months was associated with the number of employed days per month in the second year, and (2) we used time-varying covariates to test whether abstinence in a preceding month was associated with employment days in a following month (i.e., a prospective association between 1 month and the next). We conducted all analyses with Stata 9.2 software.28 RESULTS Baseline characteristics of the sample have been described elsewhere.9 Briefly, the group did not differ on sociodemographic or addiction severity measures at baseline except that ICM clients were older (mean=37.0 years; SD=6.6 years) than usual care clients (mean=35.5 years; SD=8.1 years). Figure1depicts the proportion of individuals within each group who were abstinent. The mean abstinence rate across the months for the ICM group was significantly higher than that for the usual care group (odds ratio [OR]=2.11; 95% confidence interval [CI]=1.36, 3.29). Additionally, there was a significant interaction between group and time, indicating that the differences in abstinence rates among those in ICM were growing over the assessment period (OR=1.05; 95% CI=1.02, 1.09). Figure 2 depicts the average number of days worked, by group, with superimposed lines derived from the negative binomial models. The model-derived lines demonstrate that the model provides a good fit to the data and highlight that differences in employment days diverge over time. In the multivariate analyses, there was not a main effect for group but a statistically significant interaction between time and group (incidence rate ratio [IRR]=1.03; 95% CI=1.02, 1.04), indicating that the rate of improvement over time in employment for the ICM group was greater than for the usual care group. In post hoc analyses, we found that there were significant time group interactions for having any work days in a month (OR=1.03; 95% CI=1.01,1.04). There was a main effect of individuals in the ICM group having greater odds of being employed full-time (OR=1.68; 95% CI=1.12, 2.51) as well as a significant interaction between group and time that was indicative of a greater rate of increase in fulltime employment for the ICM group over time than that for the usual care group (OR=1.05; 95% CI=1.03, 1.07). For illustrative purposes, employment rates across 3 measures of employment for ICM and usual care as well as for the comparison sample are shown for the past 3 months of follow-up (Table1). Two observations emerged from these data. First, the substance-dependent groups had lower rates than did the comparison group of having any employment or full-time employment. Second, differences between the ICM and usual care groups grew across the months. The proportion of days abstinent in year 1 was signifi