196 - Identifying Candidates for Postoperative Radiotherapy in Patients with Non-Small Lung Cancer: Multicenter Deep Learning Model Development and Prospective Validation

Z. Ma¹, and Z. Hui²; ¹Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, ²Department of VIP Medical Services, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China

Purpose/Objective(s): The value of postoperative radiotherapy (PORT) for patients with non-small cell lung cancer (NSCLC) is under great debate. This study aimed to evaluate the efficacy of a deep learning model based on key clinicopathological variables in predicting disease-free survival (DFS) and to identify which patients could benefit from PORT. Materials/

Methods: Patients across five independent medical centers with histologically proven pN2 NSCLC who underwent complete resection were enrolled and divided into training and testing datasets at an 8:2 ratio. Further external validation was performed on a randomized controlled trial cohort. A deep learning algorithm, DeepSurv, was trained on key clinicopathological variables, selected according to clinical expertise, including sex, age, smoking history, positive lymph node amounts, histology, pathologic tumor stage, lymphovascular invasion, and PORT. The models performance was assessed using the concordance index (C-index), and the clinical impact of model-recommended treatments was determined by comparing DFS in different subgroups of patients.
Results: The training, testing, and external validation datasets comprised 1792, 449, and 364 individuals. The DeepSurv model demonstrated a C-index of 0.77 (CI, 0.75-0.78), 0.77 (CI, 0.74-0.79), and 0.72 (CI, 0.68-0.75) across these datasets, respectively. DeepSurv effectively distinguished patients who could benefit from PORT, recommending PORT for 55% of the patients. In the subgroup advised to undergo PORT by DeepSurv, those who received PORT exhibited improved median DFS (20.1 months, CI, 11.8-38.1) compared to those who did not (12.2 months, CI, 9.6-18.4, P=0.03). Conversely, in the subgroup where PORT was not recommended by DeepSurv, patients undergoing PORT had median DFS (20.5 months, CI, 16.8-28.3) to those who did not receive PORT (27.1 months, CI, 15.3-Inf, P=0.22). Patients who followed DeepSurv treatment recommendations achieved significantly better DFS than those who did not (HR=0.73, CI, 0.56-0.96, P=0.02).
Conclusion: Based on key clinicopathological variables, the deep learning model could predict DFS and tailor PORT for patients with NSCLC.