Isolated outlier heart rate spikes in athletes: a signal acknowledged, not missed Reply to: isolated outlier heart rate spikes in athletes: a potentially missed signal for short lived arrhythmias

Abstract

analysis, very brief arrhythmic episodes might escape detection by appearing only as isolated outliers below the HRmax Tan threshold. We share the concern for a detection gap, as we stated explicitly in the manuscript. Moreover, as confirmed by our validation case, isolated outlier events may represent short arrhythmias. 2 The question is whether this carries meaningful clinical consequences in this population. Current guidelines do not support clinical investigation of very brief ar-rhythmic episodes in asymptomatic endurance athletes with structurally normal hearts, and the significance of such events has not been established. 3 Using a '80% of HRmax Tan ' as detection threshold would not resolve this question. Moreover, the specificity of such findings would likely be low since endurance athletes routinely reach such heart rates (HR) during high-intensity training: for an athlete with a HRmax Tan of 200 beats per minute (bpm), 160 bpm is unremarkable. Zhang suggests an additional 'steep slope' criterion for such events, which indeed would improve specificity. Our study was not aimed at defining the best discriminating criteria for proper arrhythmia detection, mainly since we had no 'golden standard', i.e. simultaneous electrocardiogram (ECG). On the other hand, we believe that the clinically meaningful signal lies at the extreme: as we have shown, paroxysmal spikes above the physiological maximum are rare, and should be of concern (if artefacts are ruled out) since they might be suggestive of real arrhythmias and/or of underlying disease. 4 We disagree that excluding brief events could have artificially inflated the sensitivity of paroxysmal spikes for arrhythmia detection, since the reverse is true: including more detections (isolated outliers and/or lower HR thresholds) would increase the sensitivity (but lower the specificity). The 71.4% sensitivity of paroxysmal spikes is therefore a conservative lower bound. Increasing sensitivity is certainly possible, but would come at a cost for more work to eliminate false positives. Future work with combined datasets (continuous ECG and HR read-outs) will allow optimisation of detection criteria, as rightfully pointed out. We fully agree with the concern that the 0.27% incidence figure is dependent on the proprietary averaging algorithms across manufacturers, and even within manufacturers if they change their algorithms. It is certainly conceivable that the averaging algorithm shapes spike morphology as it may create artefactual time lags between ECG-confirmed onset and the recorded spike, as we have shown. However, as we mentioned, our gen-eralised linear mixed model (GLMM) found no significant effect between manufacturers, which indicates that our 'paroxysmal spike' definition to define unusual (and likely arrhythmic) HR events, is rather robust over different brands and devices. Nevertheless, we share Zhang's view that manufacturers could be more transparent about their averaging algorithms. This would allow future studies to characterize device-level differences more precisely and strengthen the interpretability of wearable-derived incidence figures. Moreover, it would certainly be advantageous to have broader representation of manufacturers in future prospective studies, as we pointed out in our manuscript already. Although having access to raw signals would certainly be helpful from a scientific viewpoint, we acknowledge that industry protects its proprietary algorithms.