Deep Learning-Based Prediction of Heart Failure Readmission Using Longitudinal Cardiac Device Data

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Heart failure (HF) is a major cause of hospitalizations and readmissions.
• Cardiac implantable electronic devices (CIEDs) provide continuous physiologic data.
• CIED data may enable early risk stratification after HF decompensation.

Purpose of the Study:

• To predict 60-day HF readmission risk.
• Utilize deep learning models for prediction.
• Train models on post-discharge CIED diagnostic parameters.

Main Methods:

• Retrospective analysis of patients with ICDs/CRT-Ds (2007-2021).
• Analyzed 5 daily CIED diagnostics over 30 days post-discharge.
• Trained and evaluated two neural network architectures.

Main Results:

• Included 5734 patients and 12,369 HF hospitalization events.
• Temporal CIED data differed significantly between readmitted and non-readmitted patients.
• 1D CNN achieved 0.77 sensitivity, 0.98 specificity, and 0.89 AUROC.

Conclusions:

• Deep learning models can predict 60-day HF readmission risk.
• Leverages post-discharge CIED diagnostic trajectories.
• Supports targeted early follow-up for high-risk HF patients.