Main Session
Sep
29
PQA 07 - Head and Neck Cancer, Lung Cancer/Thoracic Malignancies, and Nursing and Supportive Care
3612 - Descriptive and Machine Learning Analysis of Cardiovascular Adverse Events after Definitive Chemoradiation for Lung Cancer In an Appalachian Population: A Retrospective Study
Presenter(s)
Joseph Schmidlen, MD - West Virginia University School of Medicine, Morgantown, WV
J. A. Schmidlen1, J. C. Knoth2, T. Nguyen3, A. N. Joseph1, M. R. Trotta4, R. A. C. Siochi2, R. R. Raylman3, C. Bianco5, J. Ryckman2, D. A. Clump II2, M. F. Hanna3, P. Pifer2, and V. Salama2; 1West Virginia University School of Medicine, Morgantown, WV, 2Department of Radiation Oncology, West Virginia University School of Medicine, WVU Cancer Institute, Morgantown, WV, 3Department of Radiology, West Virginia University, School of Medicine, Morgantown, WV, 4Department of Radiation Oncology, University of Alabama at Birmingham, Birmingham, AL, 5Department of Cardiology and Cardiac Surgery, West Virginia University, School of Medicine, Morgantown, WV
Purpose/Objective(s):
Cardiovascular adverse events (CVAEs) after chemoradiotherapy (CRT) for lung cancer are a major concern in Appalachia due to high rates of smoking and pre-treatment cardiovascular disease (CVD). The objectives of this study were to characterize the incidence of CVAEs in this population and evaluate machine learning (ML) models for post-CRT CV risk stratification. We hypothesized that this population has a high burden of post-CRT CVAEs and that incorporating patient and dosimetric data into ML-models could identify high-risk patients.Materials/Methods:
An IRB-approved retrospective study of Appalachian patients with primary lung cancer treated with definitive CRT (2013—2025) was conducted. Patients receiving SBRT or re-irradiation were excluded. Baseline clinical variables, including demographics, smoking status, and pre-existing CVD were identified and compiled along with new post-CRT CVAEs. Heart dosimetric parameters were also recorded. CVAEs were dichotomized (CVAE vs No CVAE). ML-models [Random Forest (RF), Gradient Boosting (GBM), Support Vector Machine (SVM), Logistic Regression (LR)] were trained using 5-fold cross validation. Performance was evaluated using ROC-AUC, sensitivity, and specificity. Feature importance was evaluated using permutation. Wilcoxon Test and Chi-Squared Test were used for descriptive comparisons.Results:
Eighty-six patients (mean age 66, 47% M vs 53% F) were included. 82 (95%) patients were current/former smokers and 64 (75%) had baseline CVD. By diagnosis, 80% (n=69) were NSCLC and 20% (n=17) were LS-SCLC. 51 (59%) patients experienced a CVAE. Most frequent events were NSTEMI (n=15, 29.4%), pericardial disease (n=15, 29.4%), and arrhythmia (n=8, 15.7%). Baseline clinical variables did not differ significantly between outcome groups. Mean heart dose (MHD) was higher in the CVAE group (13.4 vs 9.4 Gy, p=0.27). Heart V20, V30, V40, and V50 were all numerically higher in the CVAE group without significant differences. GBM achieved the highest AUC (0.55, 95% CI 0.44-0.69) and sensitivity (75%). RF showed the highest sensitivity (80%, 95% CI 69%-90%). Important features for model prediction and discrimination were age at diagnosis, heart V20, V40, V50, MHD, and M-stage.Conclusion:
A high incidence of CVAEs was observed after CRT in this Appalachian cohort. Although ML models using only demographics, baseline CVD status, and whole-heart dosimetry had limited discrimination, tree-based approaches outperformed traditional regression and demonstrated high sensitivity for identifying patients who developed CVAEs. These models highlighted key predictors such as age at diagnosis and heart dose metrics. Incorporating cardiac substructure dosimetry, imaging biomarkers, and circulating biomarkers in a larger cohort may improve CVAE prediction after CRT.