Deep-learning model reading histology slides identifies which pancreatic cancer patients benefit from a given adjuvant chemotherapy regimen

Deep learning applied to routine histology slides from resected pancreatic cancer produced a biomarker (PANCprAId) predicting which patients derive greater benefit from gemcitabine versus mFOLFIRINOX. The model was developed in 231 patients who underwent pancreatectomy and received adjuvant chemotherapy, then examined in the randomised PRODIGE-24/CCTG PA6 trial (n=313), where the identified subgroups differed in disease-free survival. The signal draws on both epithelial and stromal tumour features. This is a development-and-validation study rather than a practice-changing trial: the biomarker is not available in routine diagnostics and requires prospective confirmation.