Beatquency domain and machine learning improve prediction of cardiovascular death after acute coronary syndrome

Area of Science:

• Cardiovascular Physiology
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Frequency domain measures of heart rate variability (HRV) are linked to adverse outcomes post-myocardial infarction.
• Traditional frequency domain analysis (Hz) may be confounded by varying heart rates.
• Beatquency (frequency relative to heartbeats) offers an alternative perspective on cardiac phenomena.

Purpose of the Study:

• To compare the predictive performance of frequency and beatquency domains for patient risk stratification after acute coronary syndrome.
• To investigate the utility of machine learning in optimizing HRV-based risk prediction using both domains.

Main Methods:

• Evaluated pre-defined frequency and beatquency bands for the LF/HF ratio in a clinical trial dataset (N=2302).
• Compared predictive accuracy (Area Under the Curve - AUC) for cardiovascular death within one year.
• Employed machine learning to identify optimal frequency and beatquency bands for enhanced prediction.

Main Results:

• Beatquency domain analysis (AUC=0.730) significantly outperformed traditional frequency domain analysis (AUC=0.704) for predicting cardiovascular death.
• Machine learning further improved beatquency-based prediction to an AUC of 0.753.
• Machine learning did not significantly improve predictive performance for the traditional frequency domain.

Conclusions:

• Beatquency domain analysis represents a more robust method for HRV-based risk prediction compared to traditional frequency domain measures.
• Machine learning can effectively optimize beatquency bands for superior cardiovascular risk stratification.
• Beatquency and machine learning offer valuable advancements for physiological studies of HRV and clinical risk assessment.