Methodological problems in studies of depressive disorder: utility of the discontinuation design.