Predicting survival in heart failure: a risk score based on machine-learning and change point algorithm.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Heart failure (HF) poses a significant mortality risk, particularly in East Asian populations.
• Accurate risk prediction is crucial for effective HF management.
• Existing prediction models may not fully capture the complexity of HF mortality.

Purpose of the Study:

• To develop and validate a machine learning (ML) based risk score model for predicting mortality in East Asian patients with heart failure (HF).
• To effectively select relevant clinical features and segment continuous variables for improved prediction accuracy.
• To compare the performance of the novel ML risk score against conventional prediction models.

Main Methods:

• Utilized data from 3683 East Asian patients in the Korean Acute Heart Failure (KorAHF) registry.
• Employed Grouped Lasso for feature selection and a novel change-point analysis-based algorithm for continuous variable segmentation.
• Developed an ML risk score reflecting nonlinear relationships between features and survival, assigning an integer score up to 100.

Main Results:

• Identified 15 highly significant independent clinical features using Grouped Lasso.
• The ML risk score ranged from 1 to 71, with a median of 36.
• Demonstrated superior 1-year mortality prediction (AUC 0.751 vs. 0.711) compared to the MAGGIC-HF score.
• Showed significant differences in 3-year survival across risk score quintiles (80% in 1st vs. 17% in 5th).

Conclusions:

• A novel ML risk score model, incorporating advanced feature selection and variable segmentation, effectively predicts mortality in East Asian HF patients.
• The developed ML model demonstrates superior performance compared to conventional prediction models.
• This approach offers a more accurate tool for risk stratification and personalized management of heart failure.