Application of machine learning algorithms in an epidemiologic study of mortality

Area of Science:

• Epidemiology
• Biostatistics
• Health Informatics

Background:

• Epidemiologic studies are crucial for understanding mortality risk factors.
• Machine learning (ML) offers powerful tools for analyzing complex health data.
• Identifying mortality predictors is vital for public health interventions.

Purpose of the Study:

• To apply and evaluate various machine learning algorithms for identifying all-cause mortality risk factors.
• To assess the performance of different ML models in predicting mortality.
• To interpret ML model findings for actionable insights.

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) data (2009-2016) linked to mortality data through 2019.
• Applied logistic regression, random forests, k-Nearest Neighbors, MARS, SVM, XGBoost, and super learner ML algorithms.
• Evaluated model performance using AUC-ROC, sensitivity, and NPV, with SHapley Additive exPlanations for interpretation.

Main Results:

• Machine learning models achieved an AUC-ROC ranging from 0.80 to 0.87.
• The super learner model demonstrated the highest performance (AUC-ROC: 0.87, sensitivity: 0.86, NPV: 0.98).
• Identified key mortality risk factors: advanced age, increased waist circumference, male sex, and systolic blood pressure. Protective factors included marriage, higher income, and education.

Conclusions:

• Machine learning is a valuable tool for identifying mortality risk factors in epidemiologic research.
• These findings support the development of individualized, targeted interventions.
• ML-driven insights can enhance public health strategies for mortality reduction.