[Methodological problems of longitudinal studies on schizophrenia].

Abstract

Longitudinal studies are a key to understanding schizophrenia. They are the more informative, the longer the periods covered. Hence, good studies into the course of schizophrenia almost exclusively involve a lot of effort and cost. In practice, however, time-consuming methods and design variables must be avoided. The pitfalls this constraint produces are instructive of the difficulties longitudinal studies are faced with in striving for valid results. For reasons of research economy, requirements must be adjusted to study objectives. Studies into the short term course are less time-consuming, but because of the rapid changes in the illness course study intervals should be defined clearly and observed strictly. In long-term studies, too, one source of error lies in the highly varying lengths of illness of the patients studied. Even some of the classic long-term studies are marred by this error. The beginning of the follow-up period should be comparable across the study cohort and as close to illness onset as possible. To obtain generally valid results the probands must be representative of all the illness cases in the general population not only at the outset, but also all the later stages of the study. Besides the efforts to avoid attrition in the study cohort, ways must be found for correcting and estimating data for an acceptable proportion of drop-outs. In the analysis of course and outcome the indicators chosen must be apt to the traditional subtypes as well as to a theoretical symptom patterns and empirical symptom structures. In the context of typical design variables of longitudinal studies the assets and weaknesses of two retrospective and one prospective design will be discussed. Concerning the social course, importance of disease-independent factors, such as age, sex and level of social development at illness onset, as well as of control groups will be demonstrated. Predictor models will be discussed with reference to the direct and indirect influences involved. Examples of such analyses will be given.