Machine Learning Risk Prediction for Incident Heart Failure in Patients With Atrial Fibrillation.

Area of Science:

• Cardiology
• Medical Informatics
• Predictive Analytics

Background:

• Atrial fibrillation (AF) significantly elevates the risk of developing heart failure (HF).
• Current risk stratification and prevention strategies for incident HF in AF patients are underexplored.
• There is a critical need for improved methods to identify high-risk individuals within the AF population.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting heart failure (HF) hospitalization in patients with atrial fibrillation (AF).
• To identify key predictors of HF hospitalization in patients with AF.
• To enable better risk stratification for HF prevention in AF patients.

Main Methods:

• Utilized the Fushimi AF Registry, a community-based prospective survey, dividing data into derivation (n=2,383) and validation (n=2,011) cohorts.
• Constructed an ML model using the derivation cohort to predict HF hospitalization incidence.
• Validated the model's predictive ability using the independent validation cohort.

Main Results:

• Heart failure hospitalization occurred in 14% of patients over a median follow-up of 4.4 years.
• A random forest ML model using 7 variables (including age, creatinine clearance, LV ejection fraction, and LV asynergy) achieved an AUC of 0.75, outperforming the Framingham HF risk model (AUC=0.67).
• The ML model demonstrated significant ability to stratify HF hospitalization risk (log-rank P < 0.001).

Conclusions:

• The developed ML model effectively predicts HF hospitalization in AF patients.
• Key predictors identified by the model offer insights into HF pathogenesis in this cohort.
• This risk stratification tool provides opportunities for targeted HF prevention strategies in patients with AF.