Second-line chemotherapy after gemcitabine plus nab-paclitaxel in metastatic pancreatic cancer: comparative outcomes and AI-guided treatment selection

Abstract Background International guidelines recommend 5FU/LV, Nal-IRI + 5FU/LV, FOLFIRI, FOLFOX, or (m)FOLFIRINOX as second-line (2L) chemotherapy for patients with metastatic pancreatic ductal adenocarcinoma (mPDAC) after failure of gemcitabine+Nab-paclitaxel (GnP). However, a head-to-head comparison has not been performed. Patients and methods We conducted an observational cohort study of consecutive mPDAC patients treated with 2L chemotherapy after GnP failure at 41 Italian centers. Progression-free survival (PFS) and overall survival (OS) were compared using inverse probability of treatment weighting. Interpretable artificial intelligence methods were applied to optimize treatment allocation. A counterfactual Cox model was trained on baseline characteristics to estimate 12-month PFS under each regimen, and an Optimal Policy Tree (OPT) was derived to generate treatment recommendations, validated in a test set. Net-benefit curves evaluated clinical utility. Results Among 704 eligible patients, 56 (8.0%) received 5FU/LV, 153 (21.7%) FOLFIRI, 209 (29.7%) FOLFOX, 209 (29.7%) Nal-IRI + 5FU/LV, and 77 (10.9%) FOLFIRINOX. FOLFIRINOX was associated with the longest PFS and OS. Median PFS was comparable among doublets (3.5 months FOLFOX, 3.6 FOLFIRI, 3.3 Nal-IRI + 5FU/LV), though Nal-IRI + 5FU/LV showed a long-term benefit. The OPT recommended Nal-IRI + 5FU/LV for patients with head/body tumors, Eastern Cooperative Oncology Group performance status (PS) 0, or CA19.9 < 109 U/mL in those with PS > 0. Net-benefit analysis showed that the OPT consistently outperformed uniform treatment strategies, achieving a 2.5 percentage-point net benefit at a threshold probability of ∼9%. Conclusions FOLFIRINOX appears the most effective option for carefully selected, fit patients eligible for 2L chemotherapy after GnP failure. Interpretable artificial intelligence-derived treatment policies may provide superior net clinical benefit compared to uniform approaches and guide individualized therapy, warranting integration with upcoming targeted strategies such as RAS inhibitors.