Landmark proportional subdistribution hazards models for dynamic prediction of cumulative incidence functions

An individualized dynamic risk prediction model that incorporates all available information collected over the follow-up can be used to choose an optimal treatment strategy in realtime, although existing methods have not been designed to handle competing risks. In this study, we developed a landmark proportional subdistribution hazard (PSH) model and a comprehensive supermodel for dynamic risk prediction with competing risks. Simulations showed that our proposed models perform satisfactorily (assessed by the time-dependent relative difference, Brier score and area under the receiver operating characteristics curve) under PSH or non-PSH settings. The models were used to predict the probabilities of developing a distant metastasis among breast cancer patients where death was treated as a competing risk. Prediction can be estimated by using standard statistical packages.