Impact of normal lung volume choices on radiation pneumonitis risk prediction in locally advanced NSCLC radiotherapy

This study is to evaluate the impact of different normal lung volumes on radiation pneumonitis (RP) prediction in patients with locally advanced non-small-cell-lung-cancer (LA-NSCLC) receiving radiotherapy. Three dosimetric variables (V20, V5, and mean lung dose (MLD)) were calculated using treatment plans from 442 patients with LA-NSCLC enrolled in NRG Oncology RTOG 0617. Three lung volumes were defined based on the treatment plan: total lung excluding gross-tumor-target (TL-GTV), total lung excluding clinical-target-volume (TL-CTV), and total lung excluding planning-target-volume (TL-PTV). Binary classification was used for clinical endpoints: no-RP2 (N = 377) for patients with acute RP grades 0 or 1, and RP2 (N = 65) for patients with acute grades 2 or higher. The impact of lung volume definition on RP prediction was investigated with statistical analyses and two supervised machine learning (ML) models: logistic regression (LR) and random forest (RF). Balanced data was generated for comparison to prediction obtained with the imbalanced data. Areas under curve (AUCs) and the Shapley Additive eXplanations (SHAP) were used to examine ML performance and explain the contribution of each feature to the prediction, respectively. Statistical analyses revealed that V20 and MLD were associated with RP but no difference among different lung volume definitions. Mean AUC values from 10-fold cross validation using imbalanced and balanced datasets were < 0.6 except that RF from the latter dataset yielded AUC values > 0.7. SHAP values indicated that MLD and V20 were the most prominent features in RP prediction. In all assessments, no difference among various volume definitions was found. Different lung volumes showed insignificant impact on RP prediction. Low AUC values suggest limited effectiveness of these dosimetric variables in predicting RP risk and more powerful predictors are needed.
Comment: 13 page, 6 figures