Effects of electrocardiogram qrs detection algorithms in heart rate variability metrics

Accurate detection of the electrocardiogram QRS complexes is essential for computing Heart Rate Variability (HRV) metrics. While many QRS detection algorithms exist, their impact on the accuracy of HRV metrics remains underexplored. This study addresses this gap by assessing how QRS detection errors affect HRV analysis across different algorithms and recording setups. We evaluated eight widely used QRS detectors: pan_tompkins, two_average, swt, christov, hamilton, matched_filter, engzee, and wqrs, using ECG recordings from 25 healthy participants under rest, cognitive load, and physical activity conditions. Two acquisition setups were considered: “chest strap”, which offers the most consistent signal quality in this dataset, and “loose cables”, a configuration more susceptible to artifacts. We used the manually annotated R-peaks to calculate the ground-truth HRV metric values. We evaluated the relationship between detector performance and HRV errors for 23 metrics using the concordance correlation coefficient (CCC), which combines accuracy and precision to quantify agreement with ground-truth HRV values. Results showed significant variability in detector performance across algorithms and setups. No single QRS detection algorithm consistently outperformed the others. Across all metrics, CCC values ranged from 0.99 down to 0 or slightly negative depending on the detector and condition, with loose cable data showing up to 0.10–0.15 higher agreement than chest strap recordings for the most robust metrics. Based on aggregated CCC values across HRV metrics and conditions, the two_average detector achieved the highest overall agreement, whereas engzee showed the lowest performance. These findings highlight the critical role of QRS detector selection and signal acquisition conditions in HRV analysis. They underscore the need for context-specific benchmarking, particularly for wearable and ambulatory applications where signal quality can vary.