(Monday, 7/15/2019) 4:30 PM - 5:30 PM
Room: Exhibit Hall | Forum 3
Purpose: Since cardiac substructures are not typically delineated or analyzed during radiation treatment planning, we evaluated the impact of incorporating cardiac substructure dose into toxicity modeling for radiation-induced pericardial effusion. We hypothesized that incorporating cardiac substructure dose would improve heart toxicity models.
Methods: One-hundred and forty-one stage III NSCLC patients, who received radiation therapy in a prospective clinical trial, were included in this analysis. The impact of DVH metrics (mean and max dose, V5–V70) for the total heart, left atrium, left ventricle, right atrium, and right ventricle on pericardial effusion toxicity (≥grade 2, CTCAE v4.0 grading) were examined. Elastic net logistic regression, using repeated cross-validation (n=100 iterations, for avoiding observation selection bias), was conducted with cardiac-based DVH metrics as covariates. Models were trained with 75% of the data (training set), while the remaining 25% were reserved for testing model performance (test set). To assess the impact of cardiac substructure dose on toxicity modeling, we compared the following models: (a) standard model that included total heart DVH metrics-only; (b) models trained with both total heart and substructure DVH metrics; and (c) models using substructure DVH metrics-only. Model performance was assessed using AUC, calibration slope and intercept, and accuracy on the test set of each iteration.
Results: Grade 2 PCE incidence was 49.6% (n=70). The model using total heart+all substructure dose had the highest performance (AUC=0.819±0.063; calibration slope, intercept=1.824±1.197, -0.400±0.586; accuracy=0.731±0.070). Moreover, the standard model (total heart dose-only: AUC=0.799±0.063; calibration slope, intercept=4.445±5.432, -1.708±2.723; accuracy=0.696±0.085) was outperformed by many models that used various cardiac substructure DVH metrics only. Left atrium mean dose and right ventricle max dose were the most recurring predictors of toxicity.
Conclusion: Inclusion of cardiac substructure dose markedly improves toxicity modeling for radiation-induced pericardial effusion. Incorporation of cardiac substructures in treatment planning may better predict radiation-induced pericardial effusion.
Funding Support, Disclosures, and Conflict of Interest: This research was supported in part by The University of Texas MD Anderson Cancer Center Institutional Research Grant (IRG) Program.
Keywords
Radiation Effects, Modeling, Dose Response
Taxonomy
TH- response assessment : Machine learning
Contact Email
