Typically, cardiac substructures are neither delineated nor analyzed during radiation treatment planning. Therefore, we developed a novel machine learning model to evaluate the impact of cardiac substructure dose for predicting radiation-induced pericardial effusion (PCE).