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

TL;DR

This study develops a Bayesian network model to predict radiation pneumonitis risk after lung SBRT, integrating biological, dosimetric, and clinical factors, with promising validation results.

Contribution

It introduces a novel Bayesian network approach that combines multiple risk factors for better RP prediction in lung SBRT patients.

Findings

Key predictive factors include ipsilateral V5 and ACE level changes.

The model achieved an AUC of 0.8 in validation.

Biological and dosimetric data improve RP risk assessment.

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.