Predicting Electrical Storm Using Episodes' Parameters from ICD Recorded Data.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Electrical storm (ES) is a critical condition for patients with implantable cardioverter defibrillators (ICDs).
• Patients experiencing previous episodes are at increased risk for ES.
• Predictive models for ES using prior ICD data are lacking.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the short-term risk of ES.
• To utilize anonymized ICD remote monitoring data for ES risk prediction.
• To identify key episode parameters for ES risk assessment.

Main Methods:

• Utilized episode ICD-summaries from 16,022 patients.
• Constructed and evaluated logistic regression and random forest models.
• Included parameters such as total sustained episodes, shocks delivered, and cycle length.

Main Results:

• Random forest model demonstrated superior performance over logistic regression (P < 0.01).
• Random forest achieved a test accuracy of 0.99 and an AUC of 0.93.
• Logistic regression achieved a test accuracy of 0.98 and an AUC of 0.90.
• The total number of previous sustained episodes was the most significant predictor in both models.

Conclusions:

• Machine learning, particularly random forest, effectively predicts short-term ES risk in ICD patients.
• Previous sustained episodes are a critical factor in ES risk prediction.
• This approach offers a novel method for proactive management of ES in ICD recipients.