Predicting alcoholism treatment outcome

The detrimental effects of alcoholism on society have stimulated the growth of addiction treatment centers. These programs are characterized by low completion rates. This fact has promoted a great deal of research aimed toward predicting treatment completion. If those "at risk" for dropping out of programs can be identified, they can be singled out for special consideration which could result in their success with treatment. Alternatively, if it can be determined that clients with certain characteristics have a high probability of completing treatment at specific centers, then patient characteristics can be "matched" with the program shown to offer such people the best opportunity for treatment completion. The majority of studies in this area have used MMPI scales and/or combinations of demographic variables for prediction. In general, these studies have not been very successful or have failed to replicate. Some reasons for this are small sample sizes, a limited number of variables used in prediction, and lack of cross validation. The present research addresses these problems by using large numbers of subjects and predictor variables. Cross validation was performed on an independent sample. Phase One subjects were drawn from archival records; a sample of three hundred and seventy subjects was obtained; two hundred were treatment completers and one hundred seventy non completers. Variables included in the analysis were; age, sex, race, education. marital status, nnuummbbeerr of dependents, employment status, previous treatments, weeks sober prior to treatment, place of residence, prescription medication, referring agent, self reported reasons for referral, and the three validity and ten standard clinical scales of the MMPI, Through discriminant analysis, an overall successful classification rate of 65.4% was obtained. Treatment completers were classified correctly 74.0% and non completers 55.3%. The cross validation sample was obtained and variables collected in the same manner as in phase one. Data from one hundred treatment completers and eighty non completers was collected. The discriminant function from phase one derived an overall successful classification rate of 56.1%. Treatment completers were classified correctly 69.0% and non completers 40.0%. Results highlight a dramatic failure to predict treatment dropouts. However, treatment completers could be predicted. The relevance of this finding for treatment matching was discussed. It was concluded that, due to the heterogeneity of alcoholic samples, personality measures such as the MMPI should only be used to describe population characteristics at specific treatment centers; generalization should not be expected. It was hypothesized that, by looking for specific predictors at each treatment center instead of searching for global predictors, treatment matching is feasible, and may be very helpful in reducing dropout rates