An indirect comparison of EGFR TKI combination therapies in advanced <i>EGFR</i> -mutated NSCLC: Comparing manual and AI-generated analyses.

Abstract

e20625 Background: The treatment landscape for stage IV EGFR-mutated non-small cell lung cancer (NSCLC) is rapidly evolving. MARIPOSA (amivantamab plus lazertinib) and FLAURA2 (osimertinib plus chemotherapy) have established combination strategies that improve survival outcomes compared with osimertinib monotherapy. However, the comparative efficacy remains unclear. Additionally, with the growing integration of artificial intelligence (AI) into clinical research, this study aimed to evaluate AI-generated analysis compared to traditional manual analysis. Methods: An indirect comparison analysis was performed using Bucher statistical method, which enables comparative effectiveness estimation between interventions evaluated in separate randomized trials through a shared common comparator. Efficacy and safety outcomes from the MARIPOSA trial were indirectly compared against those of the FLAURA2 trial. Outcomes of interests include PFS, OS, objective response rates (ORR), and adverse events (AEs). Additionally, an independent AI-driven analysis using large language model (LLM) (GPT-5.2) was generated using predefined clinical trial keywords and aggregate trial data. Furthermore, exploratory AI-generated analyses were conducted, including digitization and reconstruction of Kaplan–Meier curves to estimate time-to-event outcomes. Results: In the manual analysis, the hazard ratio (HR) of PFS for Amivantamab-Lazertinib versus Osimertinib + Chemotherapy (as reference) was 1.13 (95% CI 0.83 – 1.53; p = 0.44). The indirect OS HR was 0.97 (95% CI 0.72 – 1.32; p = 0.87). The odds ratio (OR) of ORR in the Amivantamab-Lazertinib arm compared to Osimertinib + Chemotherapy was 0.69 (95% CI 0.39 – 1.22; p = 0.20). Lastly, the risk of experiencing any grade ≥3 AE was similar between treatment arms (OR = 0.81; 95% CI 0.51 – 1.3; p = 0.39). The manual analysis did not find any statistically significant differences between the two treatment arms. The AI-driven analysis produced concordant results. AI generated reconstructed Kaplan–Meier analyses further demonstrated minimal differences in restricted mean survival time at 36 months ( &lt; 1 month) between the two combination regimens, with no statistically significant separation observed. Conclusions: Indirect comparison using both manual and AI LLM-generated approaches demonstrated no significant differences in efficacy or safety between amivantamab–lazertinib and osimertinib plus chemotherapy. Together, these results suggest that both combination strategies provide broadly comparable clinical benefit, supporting treatment selection based on toxicity tolerance and clinical preference. Comparison Value + 95% Confidence Interval P-value Indirect PFS (HR) 1.13 (0.83 – 1.53) 0.44 Indirect OS (HR) 0.97 (0.72 – 1.32) 0.87 Indirect ORR (OR) 0.69 (0.39 – 1.22) 0.20 Indirect grade ≥3 AE (OR) 0.81 (0.51 – 1.31) 0.39