Machine learning in sudden cardiac death risk prediction: a systematic review.

Area of Science:

• Cardiology
• Artificial Intelligence
• Predictive Analytics

Background:

• Many patients receiving implantable cardioverter defibrillators (ICDs) for primary prevention do not require therapy.
• Sudden cardiac death (SCD) occurs in up to 50% of individuals deemed low risk by conventional criteria.
• Machine learning (ML) presents a novel approach for risk stratification in ICD assignment.

Purpose of the Study:

• To systematically review the application of machine learning (ML) models for predicting sudden cardiac death (SCD).
• To assess the performance and quality of existing ML models for SCD risk prediction.
• To identify gaps and future directions for ML in SCD risk stratification.

Main Methods:

• A systematic search was conducted across multiple databases (MEDLINE, Embase, etc.) for studies using ML to model SCD risk.
• Studies were screened, and eligible research underwent assessment for transparency, quality, and risk of bias (TRIPOD, PROBAST).
• Data from 11 included studies, with participant numbers ranging from 122 to 124,097, were analyzed.

Main Results:

• Eleven studies met inclusion criteria, utilizing diverse data sources (demographic, clinical, ECG, genetic) and 4-72 variables.
• The area under the receiver operator characteristic curve ranged from 0.71 to 0.96, indicating variable predictive performance.
• Machine learning models outperformed traditional regression models in five of six comparative studies.
• No studies adhered to reporting standards, and five were assessed as high risk of bias.

Conclusions:

• Machine learning for SCD prediction is under-applied and implemented with limitations.
• ML shows potential for incremental utility in predicting SCD compared to traditional models.
• Standardized reporting guidelines are crucial to enhance the quality of evidence in ML for SCD prediction.