Outlier detection and reliability of heart rate measures in sedentary middle-aged participants

Abstract

Background: While recommendations exist regarding the appropriate usage, filtering and analysis of heart rate variability (HRV) data, there are no recommendations regarding the selection of appropriate data segments or outlier detection. Aim: Determine whether subjectively selected ('most stable'), predefined ('last 5 minutes'), or'controlled breathing' 5-minute data segments produce the highest HRV reliability in a sedentary population. Method: Eight males (age: 56.3 ± 4.6; BMI: 26.4 ± 4; on medication: n=6) and 11 females (age: 55.5 ± 6.7; BMI: 26.8 ± 5.6; on medication: n=6; mean± standard deviation) attended 2 measurements 1 - 2 weeks apart. Supine resting HRV (Polar RS800CX) was assessed during spontaneous breathing (10 minutes) and controlled breathing (6 minutes). Visual inspection, the Median Absolute Deviation, and 2X Standard Deviation were considered for outlier detection. Results: The 'last 5 minutes' demonstrated the best relative reliability in heart rate (HR: 0.82 (0.58 - 0.92 intraclass correlation coefficient and 95% confidence limits)), root mean square successive difference (rMSSD: 0.82 (0.56 - 0.93)), high frequency (HF: 0.76 (0.43 - 0.91)) and low frequency (LF: 0.42 (-0.08 - 0.75). The standard deviation of NN intervals (SDNN: 0.57 (0.13 - 0.82)) was most relatively reliable in the 'most stable' segment. The 'controlled breathing' and 'most stable' data segments produced best absolute reliability for HR (5.0% (3.8 - 7.5%) coefficient of variation (95% Confidence Limits)) and LF (56.4% (37.8 - 109.3%)); and SDNN (28.4% (20.4 - 46.3%)) and HF (61 % (41.7 - 111.9%)) respectively. The 'last 5 minutes' produced the best absolute reliability for rMSSO (24.4% (17.5 - 40.2%)). Discussion: We found it surprising that the 'last 5 minutes' rather than the 'most stable' or 'controlled breathing' intervals provided the most reliable measurement for measurements reflecting parasympathetic activity (i.e. HR, rMSSD and HF). The results based on the classification categories used in this study indicate that the high frequency measurement in the frequency domain and the rMSSD measurement in the time domain were sufficiently reliable in this population. However, the typical error is still large, and results, (especially SDNN, LF and HF) should be interpreted with caution. Regarding outlier detection, the median absolute deviation method was the most sensitive methods while using the standard method of removing data points greater than 2X SD failed to detect clearly outlying data points. Conclusion: The rMSSO from the last five minutes of a 10 minute resting period during spontaneous breathing is more reliable than subjectively selecting the 'most stable' 5 minute portion, or using controlled breathing. Research in a larger cohort and with 2 or 3 measurements taken at closer intervals such as successive days is needed to confirm these findings.