AI-Powered Histology Predicts the Best Adjuvant Chemotherapy for Resected Pancreatic Cancer (PANCprAId): JCO | August 2026

Introduction:

Following curative surgery for pancreatic ductal adenocarcinoma (PDAC), modified FOLFIRINOX (mFOLFIRINOX) has become the preferred adjuvant chemotherapy for fit patients, while gemcitabine (GEM) remains an option for those unable to tolerate intensive treatment. However, treatment selection is currently driven primarily by patient fitness rather than tumour biology. This study developed and validated PANCprAId, an artificial intelligence (AI)-based biomarker derived from routine histopathology slides to predict which patients are more likely to benefit from GEM or mFOLFIRINOX.

Why was this study needed?

Adjuvant chemotherapy selection in PDAC is largely based on performance status rather than tumor characteristics.

Some patients may derive greater benefit from gemcitabine despite being eligible for mFOLFIRINOX.

Reliable predictive biomarkers to personalize adjuvant chemotherapy are lacking.

Whole-slide digital pathology contains rich biological information that remains underutilized.

AI-based pathology has the potential to guide precision oncology using routinely available histology slides.

Results:

Researchers developed PANCprAId, a deep learning biomarker trained on routine histopathology images from resected pancreatic cancers.

The biomarker was externally validated using patients from the randomized PRODIGE-24/CCTG PA6 trial.

PANCprAId successfully identified patients with differential benefit from gemcitabine versus mFOLFIRINOX, predicting both disease-free survival and cancer-specific survival.

The AI model generated treatment-specific histology scores, enabling individualized comparisons of the expected benefit from each chemotherapy regimen.

Predicted sensitivity to gemcitabine and mFOLFIRINOX was associated with distinct epithelial and stromal morphological features, suggesting that routine histology captures biologically meaningful treatment-response signatures.

The study demonstrates that standard pathology slides can provide predictive information without requiring additional molecular or genomic testing.

Clinical Impact:

This study represents an important advance in AI-driven precision oncology for pancreatic cancer. Instead of relying solely on patient fitness, oncologists may soon be able to use routine digital pathology to select the adjuvant chemotherapy most likely to benefit an individual patient. Because whole-slide imaging is already part of routine pathology practice, PANCprAId could be integrated into existing clinical workflows with minimal additional cost. Prospective validation in future clinical trials will be essential before routine implementation.

Bottom Line:

PANCprAId is an AI-based histology biomarker that predicts whether patients with resected pancreatic cancer are more likely to benefit from gemcitabine or mFOLFIRINOX. This study highlights the potential of routine digital pathology to personalize adjuvant chemotherapy and marks an important step toward precision treatment in pancreatic cancer.