Toward practical screening of mortality risk: Insights from interpretable machine learning in NHANES

Area of Science:

• Public Health
• Biostatistics
• Cardiology

Background:

• Community-based mortality risk screening faces challenges in identifying robust and practical predictors.
• Limited consensus exists on effective factors for population-level risk assessment.
• Interpretable machine learning offers a novel approach to identify efficient mortality predictors.

Purpose of the Study:

• To identify efficient predictors of all-cause and cardiovascular mortality using interpretable machine learning.
• To evaluate the prognostic implications of leading predictors in a nationally representative cohort.
• To develop a parsimonious model for practical community screening.

Main Methods:

• Analysis of 9957 adults (≥40 years) from NHANES 1999-2004 with mortality follow-up.
• Evaluation of 134 demographic, lifestyle, and biomarker variables using multiple machine learning algorithms.
• Assessment of model interpretability with Shapley Additive Explanations (SHAP) and prognostic implications via Kaplan-Meier analyses.

Main Results:

• Age, troponin T (TNT), and N-terminal pro-B type natriuretic peptide (NT-proBNP) were consistently the most influential predictors of mortality.
• Elevated TNT and NT-proBNP levels were associated with significantly poorer survival outcomes.
• A five-variable model (age, TNT, NT-proBNP, physical activity, gender) demonstrated good discrimination (AUC=0.841) and calibration.

Conclusions:

• A concise set of five predictors (age, gender, physical activity, TNT, NT-proBNP) enables efficient mortality risk stratification.
• These findings support the potential utility of these predictors in practical community-based screening programs.
• The study highlights the value of interpretable machine learning in identifying key health risk factors.