Patient therapy outcome modeling in cancer organoids is improved by cancer-associated fibroblasts and organoid assembly convolution.

Patient-derived organoids (PDOs) are becoming established as preclinical models for predicting therapeutic responses in cancer, yet their clinical accuracy remains limited by the insufficient representation of the tumor microenvironment and a reliance on static viability readouts. Here, we utilized a living biobank of 30 histopathologically and genetically characterized PDOs, alongside a microenvironment-derived from pancreatic, colon, and gastric cancers, to systematically evaluate their ability to respond to standard-of-care or experimental therapies and model patient outcomes. We assessed the impact of incorporating tissue-matched cancer-associated fibroblasts (CAFs) on treatment responses, finding that their presence not only increased chemoresistance in viability assays but significantly improved patient outcome prediction. To further enhance this predictive accuracy, we developed the Organoid Convolution Assay (OCA), a live-cell imaging-based approach that quantitatively captures dynamics of cell migration, clustering, and assembly during organoid formation. Mathematical modeling of these parameters enabled the significant stratification of donor tumors by stage (T0-T2 vs. T3-T4) and the prediction of patient clinical outcomes. Together, our findings demonstrate that incorporating either tumor microenvironment components or dynamic organoid assembly metrics improves the clinical relevance of PDO-based models.