Machine-Learning Algorithms Using Systemic Inflammatory Markers to Predict the Oncologic Outcomes of Colorectal...

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Colorectal cancer (CRC) survival prediction remains a challenge.
• Serum inflammatory markers are increasingly recognized for their prognostic potential.

Purpose of the Study:

• To evaluate the clinical significance of machine learning (ML) algorithms for predicting survival in colorectal cancer (CRC) patients.
• To develop and validate ML-based prediction scores using serum inflammatory markers.

Main Methods:

• Developed four prediction scores (DFS score-1 to -4) using random forest algorithms on 15 inflammatory marker compositions.
• Trained models on the Yonsei cohort (n=803) and validated on the Ulsan cohort (n=138) of stages I-III CRC patients.
• Utilized Cox proportional hazards model and Harrell's concordance index (C-index) to assess prognostic performance.

Main Results:

• DFS score-4 emerged as an independent prognostic factor in both training and test sets (HR 8.98 and 2.55, respectively).
• DFS score-4 achieved a higher C-index (0.727) than the lymphocyte-to-C-reactive protein ratio (LCR) (0.659) in the test set.
• DFS score-3 (C-index 0.725) demonstrated comparable predictive ability to DFS score-4.

Conclusions:

• ML-based approaches demonstrate significant prognostic utility for predicting disease-free survival (DFS) in CRC.
• These algorithms can enhance the clinical application of inflammatory markers for patient management in colorectal cancer.