[Multicenter trials in the treatment of depression: some methodological issues]

In recent year, multicenter psychopharmacological trials have met an increasing interest since through them it has become easier, in a relatively short time, to collect large samples of patients. Methodological problems common to all psychopharmacological trials may be sharpened in this kind of investigation. Others, that are specifically related to multicenter studies, may also arise. Methodological difficulties addressed in this paper are those related to patients selection procedure, to strategies for data analysis and to issues about design or management of the study. Strictly speaking, the results of a such a trial could only be generalised to patients exhibiting the characteristics of the studied sample. Moreover, before pooling the data obtained at each center, preliminary analyses should demonstrate that centers do not differentiate themselves on the basis of patient characteristics. The new drug has to be compared to both a placebo and a reference compound. In this way, and through the replication of the expected difference between the placebo and the reference drug, the methodological validity of the trial could be ascertained. The treatment with a placebo is warranted by the facts that illness natural course and non specific treatment effects have to be taken into account. Satisfactory levels of inter-rater reliability have to be achieved and maintained through training sessions with the participation of all the centers involved in the trial. Statistical analyses generally focus on the intent to treat sample. However, due to the delayed activity of antidepressant drugs, efficacy analyses could be limited to patients treated at least 14 days. Multivariate analysis of variance is particularly well suited for data analysis but can be invalidated by an excess of dropouts. To overcome this problem, other techniques such as survival analysis and random regression models have been proposed. This latter procedure seems to be particularly promising for analysing data of multicenter trials since it can assess treatment difference while adjusting for inter-center variability.