Automated Batch Processing of Diurnal Cardiac Activity: Comparison of Fully Automated Batch- to Gold-Standard Manual Processing

Abstract

The analysis of long-term variation patterns in heart rate (HR) and heart rate variability (HRV) provides insights into autonomic nervous system function beyond short-term recordings taken under resting or experimental conditions. Yet, traditional processing pipelines often require time- and labor-intensive visual inspection of electrocardiography (ECG) data and manual artifact removal. This study evaluated the performance of 3 code-based fully automated batch-processing pipelines—NeuroKit2, RHRV, and Systole—against the manual gold standard utilizing Kubios for both (diurnal) HR and HRV estimates derived from raw 48-h ECG recordings. Results illustrate that while automated pipelines yield HR estimates in good agreement to the gold standard (r = 0.91-0.99; α = 0.90-0.99), HRV estimates exhibit greater deviations (r = 0.66-0.87; α = 0.76-0.90). Cosinor analyses of diurnal HR patterns indicate strong consistency between Kubios and NeuroKit2 (r = 0.94-0.99; α = 0.97-0.99), but weaker correlations with RHRV and Systole (r = 0.58-0.87; α = 0.63-0.93). HRV cosinor parameters showed even larger discrepancies, with parameter-dependent correlations ranging from r = 0.41 to 0.86 and Cronbach’s alphas from α = 0.59 to 0.91. Findings suggest that automated batch processing of ECG data for analyzing diurnal variation patterns in HR and HRV produces results that show moderate to good agreement with the gold standard including visual inspection and manual processing. However, caution is warranted, as existing toolboxes and pipelines may lead to different results.