Given the expectations, efforts and costs of the Combining Medications and Behavioral Interventions (COMBINE) study for alcohol dependence [1,2], which matches about 1400 patients to nine different combinations of pharmacological and behavioural interventions, the overall results are comparably meagre and disappointing, to those in Project MATCH : (i) all combinations of interventions work, but none is superior; (ii) the few significant effects disappear at the end of treatment and/or are not of clinical relevance; (iii) the hypotheses on the additive effects of combining two pharmacological interventions (acamprosate and naltrexone) or of combining behavioural and pharmacological interventions, reflecting the present state of scientific knowledge (or belief?), can be consigned to the sea. As pointed out by Bergmark , the authors of the COMBINE study did not really state and discuss the complete failure of their theoretical study concept (e.g. ), as none of the major hypotheses received the necessary support. This lack of concern is indeed astonishing, compared to the thorough discussion of consequences for future research by the Project MATCH group (e.g. [6,7]). Alternatively, to publish unexpected but scientifically correct results could be interpreted as an outstandingly brave act. However, this hypothesis does not fit with the lack of theoretical discussion of the results and their consequences, especially against the background of the carefully worded recommendations as far as the pharmaceutical agents are concerned. We agree with Bergmark  that, based on the study results as reported, the conclusion to recommend solely naltrexone for implementation in health care  is more than questionable.
And the lessons to be learned?
- • Study the mechanisms of change in formal interventions.
- • Following the publications of Morgenstern & Longabaugh , we know that cognitive–behavioural therapy (CBT) works, but not by the hypothesized mechanisms of action. The COMBINE study is outstanding for its lack of knowledge on the mechanisms of change responsible for onset, course and cessation of substance use disorders. Therefore future research should concentrate on mediators of change, for both behavioural and pharmacological interventions and possible interactions. It can be assumed that not one single process, but a complex interaction of mediating and interacting contributors, is responsible for any change of behaviour. Our models of change need to be adapted accordingly. Given the present state of knowledge, the classic black box randomized controlled trial will not help us to gain urgently needed information, e.g. on the allocation of individualized interventions to patient or disorder profiles [9,10], without an understanding of the moderators and mediators of change.
- • There is plenty of empirical support for the notion that change processes in substance use disorders are influenced by a wide range of factors, which go far beyond the changes induced by formal treatment. For instance, in spite of intensive efforts to standardize interventions in multi-site studies, we found site effects in both studies, COMBINE  and Project MATCH . Additional factors play a significant, possibly dominant role: (i) pre-intervention patient and disorder variables (preparedness, pressure and commitment for change); (ii) setting variables during treatment (site characteristics, therapist behaviour); and (iii) social environment (social support system, risk factors for relapse). Given these long-lasting influences on maintenance or change of problematic substance use behaviour, the rather brief and partly overlapping study interventions (e.g. medical management and CBT) and the ‘up to 12 hours’ research assessment for all groups , it is not surprising that the ability to detect treatment outcome variance between interventions and patients has been extremely limited until now.
COMBINE and Project MATCH were necessary milestones along the path to learn more about treatment outcomes and stability of major interventions. However, to overcome our impression that ‘everything works’, and to improve treatment outcomes generally and treatment allocation procedures specifically, we have to open the black box to understand the processes of change and the factors which stimulate or impede them , making use of adequate research designs .