The model predicting entering professional AOD treatment subsequent to SAL services had good overall classification10, an increase in correct classification from 68.4% to 89.5%, and an increase in selectivity from 0% to 75%, with a minimal loss of sensitivity from 100% to 96.2%. This model explained 64.5% of the variance in entering AOD treatment and did not depart from an ideal model11. Unfortunately only 40 cases with complete data could be included in this analysis and two (5%) were removed as outliers, resulting in 38 cases in the final analysis.
Table 14 displays the results of the Logistic Regression analysis predicting entering AOD treatment, including beta values, Wald statistics, alpha values, and odds ratios. The significant predictors of entering AOD treatment were final SOCRATES taking steps scale scores from the drug form and needing inpatient treatment. The odds ratio of 1.5 for the taking steps scale means that for every one point increase in final SOCRATES drug form taking steps scores participants had a 50% increase in their likelihood of entering AOD treatment. The odds ratio of .03 for needing inpatient treatment means that participants who were designated, according to ASAM criteria, as needing such treatment were 97% less likely to enter AOD treatment at all (note that the regression coefficient bears a negative sign.).
TABLE 14: LOGISTIC REGRESSION RESULTS PREDICTING ENTERED AOD TREATMENT
Pretest Taking Steps (Alcohol)
Pretest Taking Steps (Drugs)
Final Taking Steps (Alcohol)
Final Taking Steps (Drugs)
Need Inpatient Treatment
Prior CPS Referrals
Criterion Variable = Started AOD
An additional analysis was conducted substituting the variable of needing inpatient treatment with a variable of needing any treatment at all according to ASAM criteria (0 = No, 1 = Yes). When this variable was applied it was found to be a constant for all 38 cases that had available data and were not model outliers. As a constant the new “any need for treatment” variable was excluded from the analysis, so no treatment needed variable was included. This second model had good overall classification12, an increase in correct classification from 68.4% to 81.6% and
an increase in selectivity from 0% to 67%, with a modest loss of sensitivity from 100% to 88.5%. This model explained 63.2% of the variance in entering AOD treat- ment and did not depart from an ideal model13.
10 Model = 23.4, df = 7, a = .001. Hosmer and Lemeshow Goodness of Fit = 8.07, df = 8, df = .43. 12 Model = 22.7, df = 6, df = .0009. 13Hosmer and Lemeshow Goodness of Fit = 6.6, df = 8, a = .59. 11
Idaho Pre-Treatment Program