Patterns of external validity

Background

Generalizability depends on the quality of the prediction model as developed for the development setting (internal validity), and on characteristics of the population where the model is applied (validity of regression coefficients and distribution of predictor values). The general framework of validity of predictions was discussed in Chap. 17 (see in particular Fig. 17.1). Here, we first consider a number of typical situations that we may encounter when a prediction model is applied in an external setting. Theoretical relationships are illustrated with a large sample simulation and findings in some case studies. Approximate power calculations are given for tests of invalidity of a prediction model.