An AI-powered computational histology biomarker was able to predict benefit from chemotherapy intensification in the first-line treatment of metastatic colorectal cancer (mCRC). The biomarker’s development and validation in prospective, randomized phase III trials were presented at the 2026 ASCO Annual Meeting (Abstract 3513).  

Presenting author Paolo Ciracì, MD, of the Unit of Medical Oncology 2 at Azienda Ospedaliera Universitaria Pisana and the Department of Translational Research and New Technologies in Medicine and Surgery at the University of Pisa in Italy, suggested that use of the biomarker could help personalize upfront treatment strategies while potentially sparing some patients unnecessary treatment-related adverse effects.  

Biomarker Development 

The researchers used the pan-cancer foundation model CHAI, a pan-cancer foundation model that captures and interprets histomorphologic features from hematoxylin-eosin (H&E)–stained whole-slide images, to develop a predictive biomarker designed to support first-line treatment decision-making in mCRC.  

The model was pretrained on pan-cancer H&E slides and subsequently fine-tuned for mCRC using 294 whole-slide images from the randomized MRC FOCUS trial. Patients in the trial received fluorouracil alone or in combination with irinotecan (FOLFIRI) or oxaliplatin (FOLFOX). After feature extraction and quantification, the investigators used an additional 169 whole-slide images to develop the biomarker from reweighted features.  

Key histomorphologic features used to develop the biomarker included tumor glandularity, tumor cell abnormality, tumor invasiveness, proliferative activity, and stromal maturity, among others.  

In the development cohort, the hazard ratio (HR) for progression-free survival with doublet chemotherapy vs fluorouracil monotherapy was 0.65 among biomarker-positive patients (P = .095) and 0.86 among patients with biomarker-negative tumors (P = .0490).  

Biomarker Validation 

The locked biomarker was then applied to two independent, pooled phase III cohorts. These included 159 whole-slide images from the TRIBE trial, which compared FOLFIRI plus bevacizumab with FOLFOXIRI plus bevacizumab, and 227 whole-slide images from the TRIBE-2 trial, which compared FOLFOXIRI plus bevacizumab with FOLFOX plus bevacizumab followed by FOLFIRI plus bevacizumab after disease progression. The primary endpoint for biomarker validation was progression-free survival.  

In the validation cohorts, 48% of patients were classified as biomarker-positive. Slightly more than half of biomarker-positive patients received doublet chemotherapy, whereas slightly more than half of biomarker-negative patients received triplet chemotherapy. No significant differences in baseline characteristics were observed between the biomarker-positive and -negative groups, which, according to Dr. Ciracì, “potentially suggest[s] that this biomarker is not truly related to well-known prognostic factors in this setting,” he said.  

For progression-free survival, the HR for triplet vs doublet chemotherapy was 0.51 (95% confidence interval [CI] = 0.37–0.70; P < .001) among biomarker-positive patients. The median progression-free survival was 14.7 months (95% CI = 11.9–17.5) with the triplet regimen and 8.6 months (95% CI = 7.4–10.2) with the doublet regimen. Among biomarker-negative patients, the HR for progression-free survival with triplet vs doublet chemotherapy was 1.30 (95% CI = 0.96–1.74; P = .090).  

For overall survival, the HR for triplet vs doublet chemotherapy was 0.51 (95% CI = 0.35–0.73; P < .001) among biomarker-positive patients and 1.35 (95% CI = 0.94–1.92; P = .100) among biomarker-negative patients. In the biomarker-positive group, median overall survival was 30.1 months (95% CI = 27.2 to not reached) with the triplet regimen compared with 19.9 months (95% CI = 16.6–24.9) with the doublet.  

The P value for interaction was less than .001 for both progression-free and overall survival. 

Subgroup Findings 

Subgroup analysis demonstrated that the biomarker could identify patients who derived benefit from first-line treatment intensification.  

The biomarker maintained its predictive value in an analysis of patients with right-sided tumors and/or RAS mutations (n = 294), a subgroup Dr. Ciracì described as being of particular clinical interest because treatment selection is less clearly defined than it is for patients with left-sided tumors and BRAF mutations. “So the room to choose whether to intensify or not the upfront chemotherapy backbone is in patients with the right-sided and/or RAS-mutated tumors,” he commented.  

In this subgroup, the median progression-free survival among biomarker-positive patients was 15.0 months (95% CI = 11.9–17.9) with the triplet backbone vs 7.8 months (95% CI = 6.0–10.0) with the doublet backbone (HR = 0.43; 95% CI = 0.31–0.62; P < .001). The HR among biomarker-negative patients was 0.82 (95% CI = 0.59–1.13; P = .220).  

Among biomarker-positive patients, median overall survival was 30.7 months (95% CI = 26.2 to not reached) with FOLFOXIRI plus bevacizumab compared with 18.3 months (95% CI = 15.6–24.7) with FOLFOX or FOLFIRI plus bevacizumab (HR = 0.43; 95% CI = 0.29–0.64; P < .001). Among biomarker-negative patients, the HR for overall survival was 0.78 (95% CI = 0.55–1.12; P = .186).  

Dr. Ciracì concluded by noting that the research team plans to further validate the biomarker across independent patient cohorts.  

DISCLOSURES: Dr. Ciracì had no relationships to disclose. For disclosures of the other study authors, visit asco.org.  

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.