Many studies have indicated that addiction is not an "all-or-nothing" phenomenon (Edwards et al., 1981), and that recovery - in terms of abstinence and/or stabilization - constitutes a dynamic process, which includes different dimensions and stages (Prochaska and DiClemente, 1983; Cramer and Schippers, 1994; Driessen et al., 1999; Hser et al., 2001). In the present study population, stabilization or termination of illicit drug use can be expected to encompass corresponding improvements in the patient's physical and mental health condition and his level of psychosocial functioning. The primary effect measure of the study should, therefore, include each of these domains of functioning, and provide a clear operationalization of how these domains are interrelated. Two approaches have been considered. According to the first approach, social integration and rehabilitation is the most important - and possibly the most difficult to achieve - outcome of treatment in the study population. From this viewpoint, improvements in the area of social integration and rehabilitation are preceded by improvements in the patient's health condition and drug use status, and/or reductions in his illegal activities. Improved social integration is, therefore, considered as the primary outcome measure according to this approach, while health and illicit substance use are regarded as intermediate outcome variables. Consequently, since treatment response is likely to occur more frequently on these intermediate variables than on the primary outcome measure, the statistical power analysis should be based on social integration and rehabilitation. In the second approach, treatment outcome is considered to represent a multidimensional process as well, but no assumptions are made as to the hierarchy across the outcome domains. According to this approach, the primary outcome measure should be operationalized in terms of a dichotomous index, which refers to improvements in each of the relevant domains of functioning. As a consequence, the statistical power analysis in this case should be based on the dichotomous outcome measure.
In clinical studies in the field of psychiatry, patients are usually selected on the basis of a clinical diagnosis, whereas the effects of treatment are evaluated on the basis of a severity scale. When using a severity scale, treatment outcome is often operationalized in terms of mean improvements since the baseline assessment, or in terms of percentage treatment responders. Both approaches have advantages and disadvantages. An outcome measure in terms of mean improvements can be used relatively easily in statistical analyses, and is quite sensitive. A statistically significant difference can, therefore, be demonstrated with a relatively small number of subjects. The clinical relevance of pre-treatment to post-treatment changes on a scale, however, is often difficult to interpret, particularly if the results are compared with those in a placebo or active comparison group, or if various scales are used, which diverge with respect to the observed level or direction of change. When using a dichotomous outcome measure, on the other hand, the degree of improvement, considered to be clinically relevant, is defined before the start of a study. The clinical relevance of a difference in percentage of patients who meet this definition - for example 60% responders in the experimental condition and 35% in the control condition - is evident. However, when using a dichotomous outcome measure, more patients are generally needed to demonstrate statistical significance.
With regard to these approaches, the European College of Neuropsychopharmacology (1995) concluded in a consensus meeting, that " ( …. ) a statistically significant difference between drug and placebo treatment with respect to the percentage responders defined as a pre-established degree of reduction on a pivotal severity scale may provide the most objective information about clinical relevance yet available" (pp. 533). If carefully selected, the use of a composite measure, in which the relevant outcome domains are integrated, offers the advantage of clear clinical relevance. In addition, it is important to note, that all information can still be retrieved if a dichotomous measure is used.

Based on these considerations, treatment response was defined in the present study as a dichotomous, multi-domain outcome index. In general terms, patients were considered as responders if they showed at least 40% improvement at the month 12 outcome assessment, compared to the situation at baseline, in at least one of the areas in which they functioned poorly at the start of the study (i.e. on the basis of which they were included at baseline), while these improvements should not have gone at the expense of similar deterioration in functioning in any of the other outcome domains. The operational definition of response is given in Table 5.

In addition to these criteria, the person should not have been in a controlled environment (e.g. detention, hospital, residential addiction treatment) for more than seven days in the month prior to the outcome assessment, in order to qualify as responder.