ABSTRACT
A major use for statistical models is in prediction. Prognostic models predict the expected outcome of a patient, like the remaining survival time after the diagnosis of a cancer, or the probability that a certain outcome occurs, like survival probabilities. Predictions are made using the values of a set of prognostic variables. The prognostic variables can be a mixture of continuous variables, like age or blood pressure, binary variables, indicating the presence or absence of a certain factor, and categoric variables, with more than two categories. Developing and evaluating a prognostic model involves several steps1.The first step is the construction of the model, the model building. Here, decisions are made on relevant prognostic variables and the form of the relationship between the variables and the outcome.There is a trade-off between including all prognostic information available in the model and the danger of overfitted models, models with many estimated regression coefficients that fit very well on the current data set but perform poorly on new data.
