Abstract 3654: Comparative efficacy of FOLFIRINOX versus gemcitabine plus nab-paclitaxel in metastatic pancreatic cancer: computing IPTW estimator while limiting patient data exposure with federated learning across 3 institutions

Abstract Background: Two first-line treatments are commonly used for metastatic pancreatic cancer: FOLFIRINOX and gemcitabine plus nab-paclitaxel. However, a direct head-to-head comparison in a randomized clinical trial has not yet been conducted. Collecting enough data for this comparison requires collaboration among multiple partner institutions. Herein we use a privacy enhancing technology called federated learning combined with a novel algorithmic methodology: FedECA. This method aims to facilitate collaboration by enabling institutions to retain ownership of their data, sharing only aggregated information across a network. Methods: We deployed a federated network across three cancer centers: the Fédération Francophone de Cancérologie Digestive (FFCD) hosting data from two past clinical trials (PRODIGE 35 - NCT02352337, PRODIGE 37 - NCT02827201), the Institut d’Investigació Biomèdica de Girona (IDIBGI) and the Pancreatic Cancer Action Network (PanCAN) that hold data from clinical practice. Patients with metastatic pancreatic adenocarcinoma, a performance status eastern cooperative oncology group (ECOG) no larger than 2, and receiving one of the two treatments as first-line were included. To compare the efficacy of the two treatments, we applied FedECA, the federated version of Inverse Probability of Treatment Weighting (IPTW). The following covariates were used as confounders to estimate the propensity score: age, biological gender, liver metastasis and ECOG score. Missing covariates were imputed using the MissForest algorithms independently on each center. The hazard ratio computed with a propensity score-weighted Cox Model is the estimand of the treatment effect on overall survival. The bootstrap variance estimator is used as advised when using IPTW. The entire analysis was performed without data leaving the servers of the corresponding institutions. Results: 514 patients were included for the treatment effect estimation (215 FOLFIRINOX and 299 gemcitabine plus nab-paclitaxel). The estimated hazard ratio output by our method, FedECA, is 0.75 (95% CI=0.63-0.92, p=0.005) in favor of FOLFIRINOX over gemcitabine plus nab-placlitaxel. These results show the survival benefit of using FOLFIRINOX over gemcitabine plus nab-paclitaxel. The same analysis conducted in each center independently failed to demonstrate a significant effect likely due to the statistical tests being underpowered. Conclusion: In this observational study leveraging historical trial data and real-world data, we find that FOLFIRINOX is superior to gemcitabine plus nab-paclitaxel. FedECA is the first method to allow performing IPTW in a federated setting i.e. without pooling data on a central server, facilitating collaboration across institutions and countries while ensuring data privacy. Citation Format: Quentin Klopfenstein, Jean Ogier du Terrail, Honghao Li, Imke Mayer, Nicolas Loiseau, Mohammad Hallal, Michael Debouver, Thibault Camalon, Thibault Fouqueray, Jorge Arellano Castro, Zahia Yanes, Laëtitia Dahan, Julien Taïeb, Pierre Laurent-Puig, Jean-Baptiste Bachet, Shulin Zhao, Remy Nicolle, Jérôme Cros, Daniel Gonzalez, Robert Carreras-Torres, Adelaida Garcia Velasco, Kawther Abdilleh, Sudheer Doss, Félix Balazard, Mathieu Andreux. Comparative efficacy of FOLFIRINOX versus gemcitabine plus nab-paclitaxel in metastatic pancreatic cancer: computing IPTW estimator while limiting patient data exposure with federated learning across 3 institutions [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3654.