Artificial intelligence-powered real-time model for predicting survival in advanced <i>EGFR</i> -mutant NSCLC.

Abstract

8636 Background: Novel targeted therapies have led to improved survival in EGFR -mutant NSCLC. However, survival outcomes and cancer progression vary. Accurate and practical prediction of progression, survival, and T790M status in advanced EGFR -mutant NSCLC is crucial for optimizing patient outcomes and personalizing treatment strategies. This study developed and validated an AI model for predicting survival, and the T790M mutation, integrating clinical, pathological, laboratory, and radiologic data. Methods: The model was developed and internally validated using data from Samsung Medical Center (SMC) collected baseline data (at the time of starting EGFR-TKI) and longitudinal laboratory data (during the EGFR-TKI treatment and follow-up) of patients with EGFR (deletion 19 or L858R) mutant NSCLC who received EGFR-TKI between 2008 and 2023. The primary outcome was the prediction of progression-free survival (PFS) event within 3, 6, and 12 months from each monitoring point during EGFR-TKI treatment. Secondary outcomes included predicting overall survival (OS) within 3, 6, and 12 months from each monitoring points and the detection of the T790M mutation. Results: A total of 3,095 patients participated in the study, with a median follow-up period of 41.5 months. At the time of data lock, 2,713 (87.7%) patients had experienced disease progression, 311 (10.0%) patients continued on first-line EGFR TKI treatment, and 71 (2.3%) patients were lost to follow-up. Among the patients who progressed on first-line EGFR-TKI, 1,083 (39.9%) patients acquired the T790M mutation, and 815 patients received third-generation EGFR-TKI as second-line treatment. Of the 1,630 patients without T790M or with an unknown T790M status, 174 received third-generation EGFR-TKI (117 for leptomeningeal seeding and 57 in a clinical trial), 865 were treated with cytotoxic chemotherapy or other therapies, and 591 were lost to follow-up. A total of 2,985 patients were included in the AI model. Median PFS in total population was 24.0 months (95% CI, 22.5-25.0). and medial overall OS was 50.7 months (95% CI, 48.7-52.9). The training set consisted of 1,910 patients, the validation set had 478 patients, and the test set included 597 patients. The AUC for predicting PFS events at 3, 6, and 12 months from the monitoring point was 0.780, 0.755, and 0.698, respectively. The AUC for predicting OS events at 3, 6, and 12 months from the monitoring point was 0.924, 0.886, and 0.812, respectively. The AUC for predicting T790M detection from the monitoring point at 3, 6, and 12 months was 0.768, 0.737, and 0.666, respectively. Conclusions: This study demonstrates a real-time AI-powered model to predict survival outcomes and T790M mutation status in advanced EGFR-mutant NSCLC during EGFR-TKI treatment. The model’s ability to accurately forecast PFS, OS, and T790M acquisition offers valuable insights for personalized treatment strategies.