Long-Term Prognosis of Schizophrenia

Abstract

The question of determining prognostically relevant features for schizophrenia was approached with multivariate statistical methods applied to the data from the Bonn longitudinal study of 502 schizophrenic patients. In this study, personal interviews according to a clinical classification scheme allowed every patient to be ranked within each of three different outcome classes: psychopathological remission, occupational remission, and course type. Our multivariate analysis encompassed a total of 50 items pertinent to the time up to and including the first 6 months after the first psychotic manifestation. Despite the introduction of mathematical methods considerably more sophisticated than those employed in earlier studies, no satisfactory solution could be found to the problem of predicting end states of schizophrenia. Reliable predictions could be made only for 'extreme' end states (i.e. full remission versus (1) characteristic residues in the narrower sense; (2) total unemployment, or (3) surging or simple courses to mixed residues or to typical schizophrenic defect psychoses). Accordingly, sufficiently reliable assertions applied only to a minority of about 1/3 of patients, whereas for the majority of 2/3, no generalizable prognostication was possible (67-71% true-positive predictions on 36-63% of total population in extreme states). By contrast, our analyses have clearly uncovered a fundamental problem of investigations into the long-term prognosis of schizophrenia: the extreme dependence of results on the clinical definition of end states. The further the phenomenon 'end state' is qualitatively subdivided, the poorer and less reproducible is the mutual discrimination between intermediate states and the less reliable are allocations of patients to these intermediate states by means of multivariate classifiers. Furthermore, our analyses have also demonstrated the usefulness of multivariate, adaptive procedures for investigations into the structural properties of long-term courses, so that predictions might be considerably improved if more reliable definitions of schizophrenic end states are available.