Association between depth of IPRO-α response and overall survival (OS) in patients with advanced non–small cell lung cancer (aNSCLC) treated with first-line (1L) pembrolizumab monotherapy.

e20521 Background: Anticipating clinical benefit early in the treatment course remains a major challenge in oncology clinical research. While standardized measures per RECIST 1.1 are used in randomized clinical trials (RCT) to define treatment response and disease progression, they are not used in clinical care. IPRO (Imaging-based Prognostication)-α is a fully automated, prognostic AI model that generates a survival prediction from a thoracic CT scan. By generating IPRO-α scores at baseline (BL) and on-treatment visits, changes in a patient's prognosis can be quantified. We assessed whether the magnitude of % change from BL (CFB) IPRO-α, reflecting the depth of IPRO-α response, correlates with OS. Methods: IPRO-α was evaluated in an external real-world dataset of 85 aNSCLC patients treated with 1L pembrolizumab monotherapy at 17 cancer centers, who had a BL and week 8 (± 4 weeks) scan after 1L treatment initiation. We divided the cohort into 3 IPRO-α response groups based on % CFB: 1) no response (≤ 0%), 2) intermediate response (0-40%) and 3) high response ( > 40%), to evaluate the association between the magnitude of IPRO-α response and OS. After landmarking at the week 8 CT scan, Kaplan-Meier (KM) and Cox proportional hazard analyses were performed to evaluate the OS benefit for the intermediate and high response groups over the no response group. Results: The median age of the cohort was 68 (IQR 62-75), and 42.3% (n = 36) were males. The median OS of the cohort was 20.4 months (95% CI: 10.5-28.3 months). At week 8, 22 (25.9%) patients had intermediate IPRO-α response, while 8 patients (9.4%) had high response. The median OS of patients with no response was 10.5 months (95% CI: 9.1-23.7 months), compared to 27.0 months (95% CI: 13.3-60.6 months) with intermediate response, and 50.8 months (95% CI: 14.6-not reached) for high response. The HR for intermediate response relative to no response was 0.57 (0.32-1.00, p = 0.05) and for high response was 0.36 (0.15-0.90, p = 0.029). Conclusions: These findings suggest that early IPRO-α response is associated with OS benefit, where greater improvements in IPRO-α were associated with progressively longer OS. This supports the potential of IPRO-α response as an outcome measure and early endpoint in this patient population. Future work will focus on validation in RCTs and comparisons to RECIST-based surrogate endpoints.