ABSTRACT
The standard method for building risk prediction models based on right-censored time-to-event data is Cox regression. The general aim of including variables in the prediction model is to increase the prediction performance of the model. This can only be achieved with predictor variables that are associated with the outcome of interest. Before including a categorical variable in the risk prediction model it is useful to count the number of patients in the dataset across the specified categories. In the uncensored binary outcome case, the null model predicts the outcome risk simply estimated by the number of events at the prediction time horizon divided by the number of subjects in the dataset. In the case with right-censored time-to-event outcome when there are no competing risks, we use the Kaplan-Meier estimator to obtain the benchmark model. In the case with competing risks, the null model is obtained with the Aalen-Johansen estimator.
