Prescription data and demographics: An explainable machine learning exploration of colorectal cancer risk factors...

Area of Science:

• Computational biology
• Oncology
• Health informatics

Background:

• Colorectal cancer remains a significant global health challenge despite advances in treatment and prevention.
• Predictive models are crucial for early detection and personalized risk management.

Purpose of the Study:

• To evaluate machine learning models for predicting colorectal cancer risk.
• To utilize demographic and prescribed drug data for risk prediction.
• To enhance model interpretability using explainable AI techniques.

Main Methods:

• Developed and assessed five machine learning algorithms: Logistic Regression, XGBoost, Random Forests, kNN, and Voting Classifier.
• Evaluated predictive performance across multiple time horizons (3, 6, 12, 36 months).
• Employed explainable AI for feature contribution analysis (age, sex, social status, medications).

Main Results:

• The Voting Classifier demonstrated high precision (>0.99) in identifying at-risk patients.
• Recall was moderate (~0.6), indicating room for improvement in comprehensive detection.
• Model performance was consistent across different prediction timeframes.

Conclusions:

• Machine learning effectively identifies individuals at elevated risk for colorectal cancer.
• Early intervention and personalized strategies are facilitated by these predictive models.
• Further research is necessary before widespread clinical implementation.