Development and Validation of a Computational Histology Artificial Intelligence–Powered Predictive Biomarker for Selection of Chemotherapy in Advanced Pancreatic Cancer

Researchers developed and validated the GvF biomarker, which reads standard diagnostic biopsies using the CHAI computational histology platform to indicate whether first-line treatment of advanced pancreatic cancer is likely to work better with a fluoropyrimidine-based or a gemcitabine-based regimen. The development and validation cohorts together comprised 477 patients. In validation, patients classified as fluoropyrimidine-preferring and treated accordingly had a longer time to next treatment (8.6 vs 7.5 months) and longer overall survival (14.4 vs 11.7 months); in the gemcitabine-preferring group, matched treatment lengthened time to next treatment (9.6 vs 7.2 months) but no significant overall survival difference was shown. Pancreatic cancer has very few predictive markers, so the direction matters, but this is a development-and-validation study on retrospective data rather than a trial randomising patients to treatment by biomarker.