An H&E-derived score added modest prognostic information in resected pancreatic cancer

Design

Multicohort prognostic model development and validation, including a prospective cohort

Population

644 training cases and six validation cohorts of resected pancreatic ductal adenocarcinoma

Primary outcome

Overall survival

Effect

C-index gain 0.015–0.047; prospective cohort HR 4.45 (95% CI 2.28 to 8.70)

TIGERS is a nine-feature score computed from routine H&E whole-slide images of resected pancreatic ductal adenocarcinoma, capturing tumour burden, gland and nuclear morphology, chromatin texture and tumour–stroma arrangement. It was trained on 644 cases and tested in internal, post-neoadjuvant, external, TCGA and prospectively accrued cohorts.

Adding the score to clinicopathological models raised the C-index in all six retrospective cohorts, by 0.015 to 0.047. In the prospective cohort of 155, mortality was 47.5% in the high-risk and 22.7% in the low-risk group (adjusted HR 4.45, 95% CI 2.28 to 8.70). But in one external cohort the fixed cut-off labelled 96% of patients high risk, showing it does not transfer reliably between laboratories.

This is a research tool. It suggests routine slides hold prognostic information beyond stage, but it is not ready to change reports or treatment.

• Continue staging and grading pancreatic cancer by standard criteria; do not report an AI risk score in routine practice yet.
• Note that scanner and staining differences can shift an AI threshold, as happened in one external cohort.
• Keep high-quality whole-slide images of resections where digital pathology is available, for future validation work.
• Expect prospective interventional evidence before such scores guide adjuvant treatment.

Why it matters

It shows routine slides may carry prognostic information beyond stage, and why fixed AI cut-offs break between laboratories.

Don't overread it

The C-index gains were small and the fixed cut-off failed in one external cohort.

The statistics, in plain English

The C-index measures how well a model ranks patients by survival; a gain of 0.015 to 0.047 is modest. A hazard ratio of 4.45 in the prospective cohort is large, but rests on 155 patients, so the interval is wide.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.