Machine learning-based model for worsening heart failure risk in Chinese chronic heart failure patients

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Chronic heart failure (CHF) poses a significant health burden, with predicting worsening heart failure (WHF) crucial for timely intervention.
• Existing risk prediction models may not fully capture the complexity of WHF development.

Purpose of the Study:

• To develop and validate an optimal machine learning (ML) model for predicting WHF in patients with CHF.
• To identify critical clinical and quality of life predictors for WHF.
• To create a user-friendly clinical risk assessment tool based on the validated model.

Main Methods:

• A nested case-control study involving 200 CHF patients.
• Collected 65 variables including demographics, clinical data, and quality of life scores.
• Employed LASSO regression, univariate analysis, and multiple ML models for variable selection.
• Compared nine ML algorithms, including random forest (RF), for predictive performance.
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.
• Developed an online risk prediction tool based on the best-performing model.

Main Results:

• Sixty participants (30%) developed WHF within a 3-month follow-up period.
• Key predictors identified: N-terminal pro-brain natriuretic peptide (NT-proBNP), creatinine (Cr), uric acid (UA), hemoglobin (Hb), and emotional quality of life score.
• The RF model demonstrated the highest predictive accuracy (AUC: 0.842) on the test set.
• SHAP analysis confirmed NT-proBNP, UA, and Cr as highly significant predictors.

Conclusions:

• NT-proBNP, Cr, UA, Hb, and emotional scores are critical indicators for WHF in CHF patients.
• The random forest model offers superior predictive accuracy for WHF compared to other assessed ML algorithms.
• An online risk prediction tool utilizing the RF model can aid in early and personalized WHF risk assessment.