Title: Modeling of Radiation Pneumonitis after Lung Stereotactic Body Radiotherapy: A Bayesian Network Approach

Abstract: Background and Purpose: Stereotactic body radiotherapy (SBRT) for lung cancer accompanies a non-negligible risk of radiation pneumonitis (RP). This study presents a Bayesian network (BN) model that connects biological, dosimetric, and clinical RP risk factors. Material and Methods: 43 non-small-cell lung cancer patients treated with SBRT with 5 fractions or less were studied. Candidate RP risk factors included dose-volume parameters, previously reported clinical RP factors, 6 protein biomarkers at baseline and 6 weeks post-treatment. A BN ensemble model was built from a subset of the variables in a training cohort (N=32), and further tested in an independent validation cohort (N=11). Results: Key factors identified in the BN ensemble for predicting RP risk were ipsilateral V5, lung volume receiving more than 105% of prescription, and decrease in angiotensin converting enzyme (ACE) from baseline to 6 weeks. External validation of the BN ensemble model yielded an area under the curve of 0.8. Conclusions: The BN model identified potential key players in SBRT-induced RP such as high dose spillage in lung and changes in ACE expression levels. Predictive potential of the model is promising due to its probabilistic characteristics.