[The necessity of validating regression models--characterization of the most important methods].

In rehabilitation research regression models are often used in analysing the effects of several independent factors on important outcomes in rehabilitation. But the results of such models are rarely used in rehabilitation practice for diagnosis or prognosis of patient outcomes. The main reason for this is the fact that the applicability of such models for new patient data is often unknown. A good fit of the model with respect to the data used in modeling does not guarantee a well-fitting model in the future. Thus, it is necessary to examine the performance of the model for new patient data. This examination is called model validation. The main aspect in model validation is the investigation of the prediction error caused by a too optimistic estimation of the model parameters. This prediction error is due to the twofold use of the data set: for estimating the regression coefficients AND for examining the model fit. In this paper this error is discussed and the main methods for regression model validation are presented. Finally an example illustrates the effects of model validation for prognostic purposes in a logistic regression model estimating the risk of early retirement after in-patient rehabilitation.