AI Tool Predicts Optimal Chemotherapy for Advanced Pancreatic Cancer Patients

A new artificial intelligence-based tool can predict which of two available chemotherapy options for pancreatic cancer would be more effective for an individual patient, according to results published in the February 11, 2026, edition of the Journal of Clinical Oncology. The tool, built on the Computational Histology Artificial Intelligence (CHAI) digital pathology platform, addresses an unmet clinical need in selecting first-line chemotherapy for advanced pancreatic ductal adenocarcinoma (PDAC).

Currently, no conclusive data exists to show which of the two approved chemotherapy regimens for patients with advanced pancreatic cancer is more effective. Two first-line multi-agent chemotherapy regimens approved by the US Food and Drug Administration for the treatment of advanced or metastatic PDAC include FOLFIRINOX (5-fluorouracil, leucovorin, irinotecan, oxaliplatin) and gemcitabine plus nab-paclitaxel. Both approaches are used to treat patients with good performance status, with NALIRIFOX, which combines liposomal irinotecan, 5-fluorouracil, leucovorin, and oxaliplatin, also recently approved for metastatic disease.

The problem with the current approach is that putting an ill patient on a chemotherapy regimen that isn't working worsens their health rather than improving it. Biomarkers from blood or tissue can help predict treatment response and guide these decisions in other cancer types, but currently, no biomarkers exist for pancreatic cancer.

To develop the tool, investigators used CHAI to analyze images of microscope slides containing tumor tissue samples, which were stained to highlight minute details of the cells. Almost all patients have these samples taken when their tumors are biopsied. The team analyzed tissue characteristics in samples from 25,000 pancreatic cancer patients who had received one chemotherapy regimen or the other. The platform's AI capabilities enabled analysis of more than 30,000 features of the tissue samples. Investigators then matched tissue characteristics to treatment response to create the predictive tool.

In the multinational study, the CHAI platform extracted quantitative histomorphologic features from diagnostic biopsies. In a development cohort of 178 patients, features associated with differential outcomes measured by time to next treatment or death (TNTD) between fluoropyrimidine-based chemotherapy (F-chemo)–treated and gemcitabine-based chemotherapy (G-chemo)–treated patients were used to develop continuous biomarker scores that were dichotomized into G-pref (favoring G-chemo) or F-pref (favoring F-chemo) results (GvF biomarker).

Among 299 patients in the validation cohort, there were 126 G-pref patients and 173 F-pref patients. Among G-pref patients, 43 received G-chemo; among F-pref patients, 113 received F-chemo. Among G-pref patients, the G-chemo group had significantly better TNTD vs the F-chemo group (median = 9.6 vs 7.2 months, P = .038), with no significant benefit in overall survival observed (median = 14.3 vs 12.4 months, P = .52).

Among F-pref patients, the F-chemo group had significantly better TNTD (median = 8.6 vs 7.5 months, P = .035) and significantly better overall survival (median = 14.4 vs 11.7 months, P = .003) vs the G-chemo group. In propensity score–weighted analysis, the GvF biomarker predicted the treatment effect (biomarker-treatment interaction: TNTD = P < .001, overall survival = P = .005).

Unlike most biomarker tests, where an extra sample of tissue or blood is needed, this test requires only a scanned image of the patient's existing biopsy slide. The image is sent electronically and quickly receives a result with the treatment preference. The result indicates not just which treatment is preferred, but how much more effective it is likely to be.

Pancreatic ductal adenocarcinoma is the most common form of pancreatic cancer. It is a challenging disease, which often presents with non-specific symptoms at an advanced stage, negatively impacting patients' health-related quality of life. PDAC is estimated to make up more than 80% of all cases of pancreatic cancer, which remains a lethal disease with an aggressive tumor biology. Median survival is approximately 4 months with a 5-year survival of 13%.

Over the past several decades, the incidence of pancreatic cancer has markedly increased. The disease ranks as the fourth leading cause of cancer death in men and the third leading cause of cancer death in women. Current estimates predict that pancreatic cancer will become the second leading cause of cancer-related death by 2030.

Based on the current outcomes, the tool needs further validation in patients undergoing treatment before it is ready for clinical use, but with that validation, it could eventually be applied to other solid tumor types. It could even compare the potential benefit of different types of therapy, such as radiation therapy versus surgery.

The study was supported by the Pancreatic Cancer Action Network (PanCAN–Know Your Tumor), University Health Network, Toronto, and Valar Labs, Inc.