Identification and Validation of an Explainable Predictive Model For Heart Failure in Patients With Hypertension.

Area of Science:

• Cardiology
• Biomedical Informatics
• Machine Learning

Background:

• Heart failure (HF) affects over 60 million globally, with hypertension as a key risk factor.
• Early HF risk prediction in hypertensive individuals is clinically significant.
• This study focused on developing an interpretable machine learning (ML) model for HF prediction.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for predicting heart failure (HF) risk.
• To identify key predictors of HF in patients with hypertension.
• To leverage machine learning for improved clinical decision-making in HF prevention.

Main Methods:

• Utilized data from the Systolic Blood Pressure Intervention Trial (SPRINT).
• Applied random under-sampling to balance a dataset of 324 participants (1:1 ratio).
• Evaluated seven ML algorithms (SVM, Adaboost, NB, LR, GBM, RF, MLP), using AUC, DCA, and calibration curves; employed SHAP for interpretability.

Main Results:

• 162 patients (1.8%) developed incident HF over a median follow-up of 3.88 years.
• Gradient Boosting Machine (GBM) exhibited the best performance among the tested models.
• The final GBM model achieved 0.731 accuracy, 0.770 precision, and an AUC of 0.763 (0.676-0.840), using 14 LASSO-selected features.

Conclusions:

• The Gradient Boosting Machine (GBM)-based explainable prediction model effectively predicts HF risk in hypertensive patients.
• The model's interpretability using SHAP enhances clinical trust and utility.
• This approach offers a promising tool for proactive HF management in high-risk populations.