Room: Track 3
Purpose: To identify the parts of heart that correlate with radiation-induced cardiac symptoms after radiotherapy for stage III non-small cell lung cancer (NSCLC). To estimate the corresponding dose-response relations using two popular NTCP models and evaluate their goodness-of-fit.
Methods: 112 patients, who received dose-escalated radiotherapy to 70-90 Gy on six prospective trials were studied. The dose volume histograms of left and right ventricles (LV, RV), left and right atrium (LA, RA), left anterior descending artery (LAD), heart and pericardium were derived. Cardiac toxicity was divided into four categories: ischemia, arrhythmia, pericardial and combination of them. Baseline risk was assessed using the WHO/ISH score. The Lyman-Kutcher-Burman (LKB) and Relative Seriality (RS) NTCP models were used to fit the clinical data. The area under the receiver operating characteristic curve (AUC) and Odds Ratio (OR) were used to evaluate data fitting.
Results: In this cohort, 7 patients had ischemia (6.3%), 12 arrhythmia (10.7%), 9 pericardial symptoms (8.0%) and 25 (22.3%) had at least one of those cardiac symptoms at 26 months (median) post-RT. Ischemia was correlated with dose to heart, LV and LAD (AUC: 0.68, 0.76, 0.72). Arrhythmia was correlated with dose to heart, LA, pericardium, RA and RV (AUC: 0.69, 0.62, 0.69, 0.65, 0.70). Pericardial symptoms were correlated with dose to heart, LA, pericardium, RA and RV (AUC: 0.72, 0.72, 0.74, 0.74, 0.73). The combined cardiac symptoms were correlated with dose to heart and LV (AUC: 0.71, 0.71). The OR values ranged between 2.7 to 12.2 and were statistically significant in 12 of the 15 cases.
Conclusion: The parts of heart that show significant correlation with different cardiac symptoms were identified. The LKB and RS models showed equivalency in fitting the clinical data. The derived model parameter sets can be used as NTCP objective functions or constraints in treatment plan optimization.
Not Applicable / None Entered.
TH- Response Assessment: Modeling: other than machine learning