Comparative analysis of heart rate variability indices from ballistocardiogram and electrocardiogram: a study on measurement agreement.

Abstract

Ballistocardiogram (BCG) captures minute vibrations generated by heart movements. These vibrations are converted into heart rate variability (HRV) indices, allowing their unobtrusive monitoring over extended periods, while reducing the burden on patients or subjects. In this study, to evaluate the agreement between the HRV indices, we compared the HRV indices estimated from the BCG device with those obtained from the gold standard electrocardiogram (ECG). Twenty-five healthy volunteers (mean age: 40.6 ± 12.8 years; 14 males and 11 females) rested in the supine position on a bed with a BCG device placed under a pillow while ECG electrodes were attached. BCG and ECG measurements were simultaneously recorded for 20 min. Five min of time-series data for the JJ and RR intervals obtained from BCG and ECG were converted into HRV indices. These indices included the time-domain measures (mean inter-beat intervals [IBIs], standard deviation of normal-to-normal intervals [SDNN], root mean square of successive differences [RMSSD], and percent of difference between adjacent normal RR intervals greater than 50 ms [pNN50]) and frequency-domain measures (normalized low-frequency [LF], high-frequency power [HF], and LF/HF ratio). Of the 25 individuals, data of 22 (mean age: 38.9 ± 12.3 years; 13 males and 9 females) were used to assess the agreement between the two methods, excluding 3 (1 male and 2 females) with frequent premature ventricular contractions observed on ECG. Correlations between measurements were examined using scatter plots and Pearson's product-moment correlation coefficients; in contrast, differences between measurements were evaluated using paired t-tests. The Bland-Altman analysis was used to assess the agreement. For the mean IBIs, the correlation coefficient was 0.999 (p < 0.001), and the limits of agreement ranged from - 8.35 to 11.70, with no evidence of fixed bias (p = 0.139) or proportional bias (p = 0.402), indicating excellent agreement. In contrast, the correlation coefficients for SDNN, RMSSD, and pNN50 were 0.931 (p < 0.001), 0.923 (p < 0.001), and 0.964 (p < 0.001), respectively, showing high correlations. However, a fixed bias was observed in RMSSD (p = 0.007) and pNN50 (p = 0.010), and a proportional bias in SDNN (p = 0.002). The correlation coefficients for LF, HF, and LF/HF ratio were approximately 0.7, indicating lower agreement owing to observed fixed and proportional biases. These results indicate that, while the degree of agreement varies among HRV indices, the JJ intervals measured from BCG can be used as a suitable alternative to the RR intervals from ECG.