Machine learning-based risk factor analysis and prediction model construction for mortality in chronic heart failure.

Area of Science:

• Cardiology
• Biostatistics
• Public Health

Background:

• Chronic heart failure (CHF) poses a significant global mortality burden.
• Traditional risk prediction tools for CHF lack accuracy and comprehensiveness.
• Machine learning (ML) and health ecology frameworks offer potential for improved risk prediction.

Purpose of the Study:

• To develop an ML-based model for predicting CHF mortality.
• To analyze CHF mortality risk factors using a health ecology framework.
• To integrate health ecology theory with ML for systematic risk factor identification.

Main Methods:

• Utilized data from 489 CHF patients with a 10-year mortality follow-up.
• Applied a five-layer health ecology framework to select 58 variables.
• Employed SMOTE-ENN for data imbalance and XGBoost for mortality prediction, validated via 10-fold cross-validation.

Main Results:

• Identified 24 key mortality risk factors across individual traits, behaviors, and living conditions.
• The SMOTE-ENN and XGBoost model achieved 81.58% accuracy and an AUC of 0.83.
• Key predictors included age, BMI, medication use, blood pressure, metabolic markers, and environmental factors.

Conclusions:

• Successfully categorized CHF mortality risk factors by integrating health ecology and ML.
• The developed model shows high accuracy but requires further optimization for clinical application.
• This approach enhances understanding of multi-dimensional risk factors in CHF mortality.