A Novel Classification of Intrahepatic Cholangiocarcinoma Phenotypes Using Machine Learning Techniques: An...

Area of Science:

• Hepatobiliary Surgery
• Oncology
• Machine Learning in Medicine

Background:

• Intrahepatic cholangiocarcinoma (ICC) patients often have poor prognoses.
• Significant heterogeneity exists in ICC presentation and outcomes.
• Identifying distinct patient subgroups is crucial for personalized treatment.

Purpose of the Study:

• To identify distinct clusters of ICC patients based on preoperative characteristics.
• To determine if these clusters correlate with different patient outcomes.
• To develop a tool for prospective patient classification.

Main Methods:

• Utilized a multi-institutional database of 826 ICC patients undergoing curative-intent resection (2000-2017).
• Performed cluster analysis on preoperative variables (tumor size, CA 19-9, NLR).
• Developed a classification tree for prospective patient assignment into identified clusters.

Main Results:

• Identified three distinct ICC presentation clusters: common (58.9%), proliferative (34.9%), and inflammatory (6.2%).
• Clusters differed significantly in tumor size, CA 19-9, and neutrophil-to-lymphocyte ratio (NLR).
• Median overall survival (OS) decreased progressively across clusters (60.4, 27.2, 13.3 months; p<0.001).
• Classification tree showed excellent agreement (κ=0.93) for cluster assignment.

Conclusions:

• Machine learning effectively identified three prognostic ICC clusters using only preoperative data.
• Preoperative patient heterogeneity can be characterized using machine learning.
• This approach can assist physicians in preoperative selection and risk stratification for ICC patients.