ColoLDB: a machine learning-based predictive model for colorectal cancer using routine laboratory parameters

Area of Science:

• Oncology
• Biomarkers
• Machine Learning in Healthcare

Background:

• Colorectal cancer (CRC) is a prevalent global health concern.
• Current screening methods like colonoscopy are invasive and can miss early-stage tumors.
• There is a need for simpler, more accessible early detection methods for CRC.

Purpose of the Study:

• To develop a non-invasive screening method for early detection of colorectal cancer (CRC).
• To identify key laboratory parameters indicative of CRC risk.
• To create a predictive model for assisting clinicians in CRC diagnosis.

Main Methods:

• Utilized hospitalization numbers for data identification and excluded invalid records.
• Collected diverse laboratory test data including tumor markers and biochemical parameters.
• Applied machine learning models (LightGBM, LR, RF, XGBoost) and SHAP for interpretation.

Main Results:

• Identified eight key laboratory parameters: SG, CA19-9, CEA, age, ALB, CYFRA21-1, HDL-C, and CA72-4.
• The Random Forest (RF) model achieved an AUC of 0.863, demonstrating high diagnostic performance.
• The developed ColoLDB model outperformed a diagnostic model using only CEA and CA19-9.

Conclusions:

• Eight laboratory indicators are associated with CRC risk.
• The RF-based ColoLDB model is an effective tool for predicting CRC occurrence.
• This novel approach enhances diagnostic efficiency and shows promise for CRC screening.