[Predictive studies in psychopharmacology (author's transl)]

Once the efficacy of a drug has been proven (by controlled trials), it may be interesting to define "responder" and "non-responder" profiles. This can be done with a non-comparative prediction trial in which initial (pre-treatment) characteristics of subjects with good or poor result are compared, or with a randomized trial in which one attempts to define initial features of favorable or unfavorable cases, which could lead later to a better choice for a future given patient between these therapies (and only these ones) given in similar conditions. Predictor variables can be: demographic, physiologic, nosologic, symptomatic. Statistical methods can be: simple comparison of responses between sub-classes according to one initial feature, or multivariate techniques particularly discriminant analysis (when results are expressed in a dichotomic way), multiple regression (when responses are graded, as with an improvement score). Validation of results of such trials can be done by checking their predictive value on subjects independant of those from which prediction algorithms have been set up.