Individualized Triplet Chemotherapy Decision-Making in Metastatic Colorectal Cancer: A Machine-Learning-Driven Study

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Identifying optimal candidates for triplet chemotherapy in metastatic colorectal cancer (mCRC) is challenging.
• Triplet regimens like FOLFOXIRI/FOLFIRINOX offer potential benefits but carry increased toxicity.
• Predictive biomarkers are needed to guide treatment decisions in mCRC.

Purpose of the Study:

• To develop a machine learning-based predictive biomarker for identifying mCRC patients who benefit from first-line triplet chemotherapy.
• To determine clinical variables that predict individual treatment effect (ITE) for triplet therapy.

Main Methods:

• Retrospective analysis of clinical and biochemical data from 136 de novo mCRC patients.
• Machine learning models (T-Learner) used for ITE estimation and prediction of progression-free survival (PFS) ≥ 270 days.
• SHAP analysis employed to identify key predictive features for model reduction.

Main Results:

• The full model achieved a high predictive performance (ROC AUC 0.919).
• Top predictive variables included primary tumor site, ferritin, CA19-9, CRP, uric acid, TSH, triglycerides, total protein, LDL, and platelet count.
• A reduced model using these 10 features retained strong predictive power (AUC 0.869).

Conclusions:

• A machine learning model shows promise for selecting mCRC patients for triplet chemotherapy.
• The identified biomarkers can aid in optimizing treatment strategies and managing toxicity.
• Prospective validation in larger cohorts is crucial for clinical implementation.