CHAI Biomarker for Chemotherapy Choice in Advanced Pancreatic Cancer
A platform powered by computational histology artificial intelligence (CHAI) could be used to predict whether gemcitabine- or fluoropyrimidine-based chemotherapy would be preferred as first-line treatment in advanced pancreatic ductal adenocarcinoma. A report on the development and validation of the CHAI-backed biomarker was published by Hendifar et al in the Journal of Clinical Oncology.
These findings were also presented in a poster at the 2026 ASCO Gastrointestinal Cancers Symposium (Abstract 764).
Model Methods
In the multinational study, deep learning models, which were trained on about 25,000 pan-cancer slides, processed digitized H&E-stained slides to learn to segment tissue and cell data. The models could identify 30,000 different histomorphologic features from diagnostic biopsies, and the CHAI platform extracted related features with weighting to find a relevant signature for treatment choice.
In a development cohort of 178 patients from Cedars-Sinai Medical Center and the University of Pittsburgh Medical Center, features associated with differential outcomes measured by time to next treatment or death (TNTD) between patients treated with fluoropyrimidine-based chemotherapy and those treated with gemcitabine-based chemotherapy were both used to develop continuous biomarker scores that were dichotomized into those that favored gemcitabine or fluoropyrimidine chemotherapy (GvF biomarker). The biomarker represented the threshold for the greatest difference in the hazard ratio for patients receiving gemcitabine-based chemotherapy vs those receiving fluoropyrimidine-based chemotherapy.
Data used in the development cohort was not used in the independent validation cohort, which consisted of patients from the Know Your Tumor molecular profiling initiative from the Pancreatic Cancer Action Network, and the prospective Canadian sequencing COMPASS trial.
The main outcomes of interest were TNTD and overall survival (OS) according to receipt of gemcitabine- or fluoropyrimidine-based chemotherapy among patients who were more sensitive to gemcitabine or fluoropyrimidine.
Key Findings
Among 299 patients in the validation cohort, there were 126 patients favoring gemcitabine and 173 patients favoring fluoropyrimidine. Among patients preferring gemcitabine, 43 received gemcitabine-based chemotherapy; among patients preferring fluoropyrimidine, 113 received fluoropyrimidine-based chemotherapy.
Among gemcitabine-sensitive patients, the group treated with gemcitabine-based chemotherapy had significantly better TNTD vs the group treated with fluoropyrimidine-based chemotherapy (median = 9.6 vs 7.2 months; P = .038), with no significant benefit in OS observed (median = 14.3 vs 12.4 months; P = .52).
Among patients who preferred fluoropyrimidine, the group treated with fluoropyrimidine-based chemotherapy had significantly better TNTD (median = 8.6 vs 7.5 months; P = .035) and significantly better OS (median = 14.4 vs 11.7 months; P = .003) vs the group treated with gemcitabine-based chemotherapy.
In propensity score–weighted analysis, the GvF biomarker predicted the treatment effect (biomarker-treatment interaction: P for TNTD < .001; P for OS = .005).
Basal and classical RNA subtypes were associated with both TNTD and OS in the validation cohort (hazard ratio = 0.48; 95% confidence interval = 0.34–0.70; P for OS < .0001), but the subtypes were not associated with treatment effect (P for TNTD = .3; P for OS = .8). Status of the GvF biomarker was not correlated with RNA subtyping. When a model adjusted for RNA subtype, the interaction with treatment was still significant (P < .05).
The investigators concluded: “The histomorphology-based GvF biomarker predicted differential treatment benefit of first-line GvF. This biomarker can guide optimal treatment selection for first-line therapy in advanced pancreatic ductal adenocarcinoma.”
Viswesh Krishna, BS, of Valar Labs, Inc., Palo Alto, California, is the corresponding author for the Journal of Clinical Oncology article.
DISCLOSURES: The study was supported by the Pancreatic Cancer Action Network (PanCAN–Know Your Tumor), University Health Network, Toronto, and Valar Labs, Inc. For full disclosures of the study authors, visit ascopubs.org.
AI in Practice: Have you developed or used an AI biomarker in cancer prognostication or diagnostics? Tell ASCO AI in Oncology about your experiences with these tools.
ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.
Performance of a convolutional neural network in determining differentiation levels of cutaneous squamous cell carcinomas was on par with that of experienced dermatologists, according to the results of a recent study published in JAAD International.
“This type of cancer, which is a result of mutations of the most common cell type in the top layer of the skin, is strongly linked to accumulated [ultraviolet] radiation over time. It develops in sun-exposed areas, often on skin already showing signs of sun damage, with rough scaly patches, uneven pigmentation, and decreased elasticity,” stated lead researcher Sam Polesie, MD, PhD, Associate Professor of Dermatology and Venereology at the University of Gothenburg and Practicing Dermatologist at Sahlgrenska University Hospital, both in Gothenburg, Sweden.