ABSTRACT

After developing a prognostic model like the Cox model the question is often asked how the validity of the model can be assessed. Unfortunately, it is not easy to define validity in a concise and precise way. There is even ambiguity about the meaning of the term “model.” For an interesting discussion, see Altman & Royston (2000). Following van Houwelingen (2000), a statistical model is defined as a rule to compute (survival) probabilities given the observation of the covariates included in the model. With this definition of a model a statistical definition of validity can be based on the concept of calibration (“validation by calibration”).