Validation of Polar H10 chest strap and Fitbit Inspire 2 tracker for measuring continuous heart rate in cardiac patients: impact of artefact removal algorithm

Abstract

Abstract Funding Acknowledgements Type of funding sources: Public hospital(s). Main funding source(s): Antwerp University Hospital Background In the last decade, heart rate (HR) monitors have become popular consumer tools for physical activity registration and training guidance. HR tracking could assist in following up cardiac patients. However, a profound validation of such monitors in cardiac populations, with HR monitors worn for a long period of time and compared to the gold standard Holter monitoring, is lacking. Purpose This pilot study assessed the accuracy of 2 HR monitors (Polar H10 chest strap and Fitbit Inspire 2 tracker) compared to Holter in patients with atrial fibrillation (AF), heart failure (HF) and following cardiac rehabilitation (CR). Methods A total of 15 patients (5 AF, 5 HF and 5 CR) with a scheduled 24-hour Holter monitor were asked to wear both Polar H10 and Fitbit Inspire 2 simultaneously with their Holter. All three devices were perfectly aligned in time. To be comparable with Polar and Fitbit data, Holter data was represented as a moving average of 3 and 5 data points, respectively. Raw Polar and Fitbit data were processed by a first version of an in-house designed artefact removal algorithm that labelled extreme values as artefacts based on the expected HR values per patient and the steepness of HR increases/decreases, resulting in cleaned data. Accuracy was assessed in raw and cleaned data using Pearson correlation coefficients (r), mean absolute error (MAE), mean absolute percent error (MAPE, criterion: ≤10%) and Bland-Altman analysis (95% Limits of Agreement, LOA). Results Raw Polar H10 HR data showed a strong correlation (r=0.932), a MAE of 3.43 beats per minute (bpm) and a MAPE of 4.86%. The cleaned Polar H10 data showed a small further improvement. Raw Fitbit Inspire 2 HR data showed a weak correlation (r=0.678), a MAE of 8.68 bpm and a MAPE of 14.56%. The agreement in the cleaned Fitbit Inspire 2 data improved to a moderate correlation (r=0.758), a MAE of 7.12 bpm and a MAPE of 12.24%. The findings varied per patient group, with highest accuracies in the CR group and lowest in the AF group (based on MAE/MAPE and Polar correlation) and in the HF group (based on Fitbit correlation). Bland-Altman analysis indicated that Polar H10 over- and underestimated HR equally in all groups, but to a lesser extent in the cleaned data. In contrast, Fitbit Inspire 2 overestimated HR more frequently than underestimating it, both overall and for the groups separately, but again to a lesser extent in the cleaned data (Table 1 and Figure 1). Conclusions Compared to gold standard Holter monitoring, Polar H10 is very accurate at assessing HR over a 24h time window, while the accuracy of Fitbit Inspire 2 is weaker. The results differ between several cardiac patient groups. We were able to improve accuracy by detecting and correcting artefacts. We are further exploring technical possibilities, striving for an improved and automatic way to continuously monitor HR in cardiac patients, which may help to monitor and guide their physical activities.