PURPOSE: Follow-up investigations in locally advanced stage non-small cell lung cancer (NSCLC) patients treated with radiochemotherapy (RCHT) regularly focus around lung toxicity. However, Cardiac Adverse Events (CAE) can occur much earlier in patients than originally anticipated with serious repercussions for patient quality-of-life and survival.Therefore, here we investigated spatial dependencies of dose within the heart and their correlation with toxicity, with dosimetric parameters of sub-regions of the heart at the focus of this analysis.Additionally, we aimed to explore the connection between cardiac toxicity and pulmonary toxicity.

METHODS: Patient treatment plans with dosimetric data for the lungs and the heart, as well as toxicity data for 502 NSCLC patients treated with either passively scattered proton therapy (PSPT), intensity modulated radiation therapy (IMRT), three-dimensional conformal radiation therapy (3DCRT) or volumetric arc therapy (VMAT) with or without chemotherapy was retrospectively retrieved from prospective clinical studies of three international centers. Cardiac toxicity data was not available for all patients. Data was randomly split into a training set (336) and validation set (166). Statistical analyses were performed using binomial logistic regression.

RESULTS: In univariate modeling, the Mean Lung Dose (MLD) significantly predicted CAE grade ≥ 3 in the training-set (p MLD = 0.02, AUC train = 0.69), which was confirmed in validation (AUC val, = 0.77). No suitable candidates for the construction of multivariate models could be identified. Parameters of the heart and its subregions did not significantly predict CAE grade ≥ 3 in the investigated cohorts. No parameters were found to significantly predict CAE grade ≥ 2 or RP. Finally, no spatial dependency was found in the investigated toxicity data.

CONCLUSION: The pulmonary dosimetric parameter MLD successfully predicted CAE grade ≥ 3 in a cohort treated with either photons or protons. Cardiac dosimetric parameters as well as spatial parameters did not perform similarly. No parameters were found to significantly predict RP in the investigated cohorts.