Survival prediction with Bayesian Networks in more than 6000 non-small cell lung cancer patients

A model that predicts survival in lung cancer as a function of treatment choices would be valuable for decision support. In this study we built data flow tasks and a data warehouse to collect from clinical databases a large non-small cell lung cancer dataset from MAASTRO (N=1781) and from Princess Margaret Hospital (PMH, N=4591). We learned Bayesian Network (BN) models for survival prediction from the MAASTRO data and evaluated the models in the PMH dataset. The BN model based on stage and radiotherapy dose had a high predictive accuracy (AUC 0.917). The model correctly showed that radical radiotherapy (>60Gy) is beneficial for non-small cell lung cancer patients and that this benefit is disease stage dependent.