3640 - AI-Derived Thymic Health and Locoregional Recurrence Among Patients with Unresectable Non-Small Cell Lung Cancer Undergoing Definitive Chemoradiation
Presenter(s)
L. L. Thompson1,2, V. Prudente1,2, S. Bernatz1,2, J. Yoon3, N. A. Saeed2, A. T. Gregg3, S. B. Shah3, P. M. Amin3, A. Warrington4, F. K. Keane5, R. H. Mak1,2, and H. Aerts1,4; 1Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA, 2Department of Radiation Oncology, Brigham and Women's Hospital and Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, 3Harvard Medical School, Boston, MA, 4Department of Radiation Oncology, Mass General Brigham/Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, 5Department of Radiation Oncology, Massachusetts General Hospital, Boston, MA
Purpose/Objective(s): Thymic function may influence T-cell repertoire diversity and the related risk of locoregional recurrence (LRR), but its role as a biomarker in unresectable locally-advanced NSCLC (LA-NSCLC) remains underexplored. We examined relationships between a validated artificial intelligence (AI)-based measure of thymic health, thymic radiation dose, and locoregional recurrence (LRR) among patients with LA-NSCLC undergoing chemoradiation.
Materials/Methods: We reviewed the records of LA-NSCLC patients who completed chemoradiation at our institution 11/01/98-01/31/14, abstracting clinical characteristics, treatment details, and oncologic outcomes. We used a validated deep-learning system to determine thymic health scores from pre-treatment computed tomography scans (score range: 0-1; higher scores=better function). We assessed associations between thymic health, thymic dose, and LRR using a Fine-Gray subdistribution hazard model with consideration of death as a competing risk. The model was adjusted for age, sex, histology, stage, functional status, prescribed radiation dose, and significant covariates (univariate p <0.10).
Results: Overall, 402 patients (median age 67 years, 52.0% female, 93.1% stage III, 38.6% adenocarcinoma) completed chemoradiation. Median thymic health was 0.17 (IQR 1.64 x 10-5-0.30). Median mean thymic dose was 50.6 Gy (IQR 39.8-59.8 Gy). At a median follow-up of 16.1 months, 44.6% of patients had LRR (154/180 in-field, 26/180 out-of-field). Cumulative incidence of LRR adjusted for competing risk of death was 22.1% at 1 year, 33.0% at 2 years, and 39.2% at 5 years. In the adjusted Fine-Gray model, higher thymic health was associated with reduced risk of LRR (SHR=0.49, 95% CI:0.25-0.96, p=0.040), but mean thymic dose was not (SHR=0.99, 95% CI:0.98-1.00).
Conclusion: AI-based thymic health metrics may have utility as clinical biomarkers for LRR.