Investigators Find New Way to Forecast Survival for Patients With Colon Cancer

Published on: 

The overall survival after adjusting for tumor stage showed tumor adipose feature was independently prognostic as both a binary feature and as a semiquantitative categorical feature.

Using machine learning techniques, investigators have identified a way to forecast survival for patients with later stage colon cancer.1

A team, led by Vincenzo L’Imperio, MD, Department of Medicine and Surgery, Pathology, University of Milano-Bicocca, IRCCS (Scientific Institute for Research, Hospitalization and Healthcare) Fondazione San Gerardo dei Tintori, evaluated whether pathologist scoring of a histopathologic feature previously identified by machine learning is linked to survival among patients with colon cancer.

“Identifying new prognostic features in colon cancer has the potential to refine histopathologic review and inform patient care,” the authors wrote. “Although prognostic artificial intelligence systems have recently demonstrated significant risk stratification for several cancer types, studies have not yet shown that the machine learning–derived features associated with these prognostic artificial intelligence systems are both interpretable and usable by pathologists.”

In the prognostic study, the investigators used deidentified, archived colorectal cancer cases between 2013-2015 from the University of Milano-Bicocca.

The team used all available histologic slides from 258 consecutive colon adenocarcinoma cases and sought main outcomes of the prognostic value of tumor adipose feature for overall survival and disease-specific survival, measured by univariable and multivariable regression analyses.They also evaluated interpathologist agreement in TAF scoring. The median age of the patient population was 67 years and 53% (n = 138) were men.

In addition, 119 patients had stage II colon cancer and 139 patients had stage III colon cancer.

The results show tumor adipose feature was found in 120 total cases, 63 of which were widespread, 31 were multifocal, and 26 were unifocal.

The overall survival after adjusting for tumor stage showed tumor adipose feature was independently prognostic as both a binary feature (presence vs absence: hazard ratio [HR] for presence of TAF, 1.55; 95% confidence interval [CI], 1.07-2.25]; P = .02) and as a semiquantitative categorical feature (HR for widespread TAF, 1.87; 95% CI, 1.23-2.85; P = .004).

The interpathologist agreement for widespread tumor adipose feature compared to lower categories like absent, unifocal, or multifocal was 90%, which corresponds to a K metric at the threshold of 0.69 (95% CI, 0.58-0.80).

“In this prognostic study, pathologists were able to learn and reproducibly score for TAF, providing significant risk stratification on this independent data set,” the authors wrote. “Although additional work is warranted to understand the biological significance of this feature and to establish broadly reproducible TAF scoring, this work represents the first validation to date of human expert learning from machine learning in pathology. Specifically, this validation demonstrates that a computationally identified histologic feature can represent a human-identifiable, prognostic feature with the potential for integration into pathology practice.”


L’Imperio V, Wulczyn E, Plass M, et al. Pathologist Validation of a Machine Learning–Derived Feature for Colon Cancer Risk Stratification. JAMA Netw Open. 2023;6(3):e2254891. doi:10.1001/jamanetworkopen.2022.54891