Association Between Risk Factors and Major Cancers: Explainable Machine Learning Approach.

Area of Science:

• Oncology
• Data Science
• Medical Informatics

Background:

• Cancer remains a leading cause of death globally, with rising incidence rates, particularly among younger populations.
• Early screening and risk factor monitoring are crucial for managing and reducing cancer risk.
• Understanding cancer diagnosis risk profiles is essential for improving patient outcomes and health equity.

Purpose of the Study:

• To utilize explainable machine learning (ML) models to identify and analyze key risk factors for breast, colorectal, lung, and prostate cancers.
• To enhance the understanding of cancer diagnosis risk profiles by uncovering significant associations between risk factors and cancer types.
• To facilitate precise screening, early detection, and personalized prevention strategies for major cancers.

Main Methods:

• Deidentified electronic health record data from the Medical Information Mart for Intensive Care (MIMIC)-III database was used.
• Propensity score matching was employed to combine cancer patient records with non-cancer controls.
• Three advanced ML models—penalized logistic regression, random forest, and multilayer perceptron (MLP)—were utilized to rank risk factors, with feature importance analysis for random forest and MLP.

Main Results:

• The MLP model demonstrated superior prediction performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.78 for breast cancer, 0.76 for colorectal cancer, 0.84 for lung cancer, and 0.78 for prostate cancer.
• Prominent nontraditional risk factors, including hyperlipidemia, diabetes, depressive disorders, heart diseases, and anemia, showed significant associations across cancer types.
• Rank biased overlap analysis revealed a unique risk factor pattern for lung cancer compared to other cancer types.

Conclusions:

• Explainable ML models are effective in assessing nontraditional cancer risk factors and identifying unique risk profiles for different cancer types.
• The study provides a hypothesis-generating foundation for future research in cancer diagnosis risk analysis and management.
• Collaboration with clinical experts for external validation is recommended to refine model outputs and integrate findings into clinical practice for improved patient care and cancer prevention.