A framework for the creation of prediction models for serious adverse events

In oncology, the risk of Serious Adverse Events (SAEs) is an important factor in the decision for a certain treatment, because SAEs can be life threatening and/or they can reduce the quality of life to a large extent. Prediction models can be made, based on data available from clinical trials. Building such prediction models requires great effort and expertise. On the one hand, data mining expertise is required to mold the data, build and validate the models. On the other hand, clinical insight is required in order to select interesting features and judge the applicability of the model. The greater the involvement of the clinical expert in the generation of a model, the larger the chance that the model will be adopted for further investigation or use in clinical practice. This paper presents a framework which provides a graphical interface which helps a clinical expert to be largely involved in the generation of a (crude) first set of prediction models without great effort or data mining expertise. This allows unpromising paths of research to be easily discarded while promising models can be further teased out by a data mining expert.