Reliability of five-minute vs. one-hour heart rate variability metrics in individuals with spinal cord injury

Introduction

Heart rate variability (HRV) is the physiological phenomenon of variation in the time interval between consecutive heartbeats which is generated by the SA node in the right atrium of the heart (Shaffer & Ginsberg, 2017). Variations in heart rate are analyzed by different methods such as time domain, frequency domain or combined time-frequency measures (Shaffer & Ginsberg, 2017). HRV is usually measured with short (approximately 5 min) and long (24-h) durations (Malik et al., 1996; Shaffer & Ginsberg, 2017). Interestingly, short term and long term HRV reflect different underlying physiological processes. Long term HRV is attributed to changes in circadian rhythm, core body temperature, the renin-angiotensin system, and the sleep cycle (Malik et al., 1996; Shaffer, Meehan & Zerr, 2020). Short term HRV is thought to be generated by four sources: (1) interactions between sympathetic and parasympathetic autonomic regulation, (2) respiration, (3) baroreceptor reflex, that regulates blood pressure, and (4) rhythmic adjustment in blood vessel diameter (Shaffer & Ginsberg, 2017). Thus, long term HRV is not interchangeable with short term HRV (Shaffer, Meehan & Zerr, 2020). Long term HRV has been found to be a powerful and independent predictor of an adverse event such as morbidity or death in multiple patient populations (Fang, Wu & Tsai, 2020; Hohnloser et al., 1997; Nunan, Sandercock & Brodie, 2010; Patel et al., 2017). However, some studies demonstrated that short term or even ultra-short-term HRV recording (<5 min) might also be used to predict adverse events (Fei et al., 1996; Karp et al., 2009; Kautzner et al., 1998; La Rovere et al., 2003).

To employ HRV in clinical practice, apart from its validity, the HRV metrics should be reliable and therefore applicable across individual patients. For reliability analysis, it is necessary to estimate outcome variability within individuals and in different populations. It was found that clinical populations showed poorer reliability than healthy subjects (La Fountaine et al., 2010; Lord et al., 2001). Additionally, reliability was poorer during interventions such as tilt or pharmacological stimulation compared to rest (Sandercock, Bromley & Brodie, 2005). Sandercock, Bromley & Brodie (2005) stated that describing HRV in general as a reliable measurement technique is an oversimplification, because the results of reliability studies are heterogenous and depend on a number of factors, thus studies in different clinical populations are required to assess HRV reliability.

The spinal cord is a compact bundle of neural structures that facilitate the transmission of both motor and sensory information between the brain and the rest of the body. Consequently, any damage to the spinal cord can disrupt the conduction of signals associated with sensory, motor, and autonomic functions across the site of the injury (Rupp et al., 2021). Spinal cord injury (SCI) can be categorized into two primary groups, determined by the level of the lesion: tetraplegia refers to the impairment or loss of motor and/or sensory function in the cervical segments of the spinal cord, while paraplegia refers to the impairment or loss of motor and/or sensory function in the thoracic, lumbar or sacral segments (Rupp et al., 2021). Individuals with SCI also suffer from consequences of autonomic dysfunction, including issues such as sexual dysfunction as well as bowel and bladder problems (Hou & Rabchevsky, 2014). Regarding the autonomic control of the heart, the parasympathetic system is undamaged in individuals with SCI, since its regulation is from the vagal nuclei in the brain stem (Takahashi et al., 2007). On the other hand, sympathetic control of the heart relies on bulbospinal input via the cervical region and spinal sympathetic preganglionic neurons located in the upper half (T1–T6) of the thoracic segments. As a result, those with injuries at or above T6 have compromised sympathetic control of the heart (Fossey et al., 2022; Malmqvist et al., 2015; Rodrigues et al., 2016). Research indicates that individuals with SCI exhibit lesion-dependent impairment in resting cardiovascular function. Among these individuals, those with the highest injuries (tetraplegia) demonstrate the greatest degree of cardiovascular dysfunction compared to individuals with high thoracic (T1–T6) and low thoracic (below T6) paraplegia (West, Mills & Krassioukov, 2012). Notably, cardiovascular disease is the leading cause of death after SCI (Sabre et al., 2013; Savic et al., 2017). Additionally, individuals with SCI, particularly individuals with tetraplegia, have higher risk of cardiovascular events and deaths when compared to their able-bodied counterparts (Chamberlain et al., 2019; Cragg et al., 2013; Jia et al., 2023). Apart from an altered metabolism and decrease in physical activity, autonomic dysfunction is one of the proposed contributing factors (Raguindin et al., 2021). In comparison with healthy counterparts, participants with SCI had lower HRV values for both time and frequency domain metrics (Rodrigues et al., 2016; Serra-Ano et al., 2015).

Our previous study in 45 individuals with SCI and without any restriction on activity showed that when the duration of HRV recording was shorter (i.e. 1, 3 or 6 h), the reliability was lower than long-duration recording (24 h) (Ruangsuphaphichat et al., 2023). However, 24-h recording may not be convenient in clinical practice and it has not been integrated in general medical care (Shaffer, Meehan & Zerr, 2020). Additionally, the low reliability observed for 1-h recording in our previous study may also have resulted from placing no restriction on activity. We thus aimed to investigate test-retest reliability of short term HRV measurement, namely 5-min and 1-h durations, when participants rest in the supine position. The 5-min duration was chosen because it is a standard and generally used method that requires a short time to perform (Malik et al., 1996; Sassi et al., 2015; Shaffer & Ginsberg, 2017). The 1-h duration was used in order to be able to establish the test-retest reliability of the ultra-low-frequency (ULF) outcome since the shortest time period for ULF to be valid is 1 h (Tan & Jiang, 2013).

Materials and Methods

Results

Overall reliability of HRV in SCI

There were no significant differences (p > 0.05) in any pairs of HRV values for both durations (5 min and 1 h) in the overall analysis (i.e., without consideration of lesion level). For the 5-min duration overall, HF had moderate to good relative reliability (ICC = 0.50), but all other outcomes had poor reliability (ICC 0.06–0.36). For the 1-h duration, overall relative reliability was moderate to good for SDNN, RMSSD, HF, ULF and TP (ICC 0.40–0.71), while it was poor for LF and VLF (ICCs of 0.13 and 0.28, respectively). For all outcomes, relative reliability was higher (higher ICC) for the 1-h than for the 5-min duration.

In terms of absolute reliability, the CVs for the 1-h duration were better than for the 5-min duration. In general, time domain metrics showed better reliability than frequency domain metrics for both durations. The CVs for the 5-min HRV outcomes were 47.2% (SDNN) and 67.3% (RMSSD) for the time domain metrics and in the range of 86.6% to 158.0% for the frequency domain metrics. For the 1-h HRV outcomes, CVs were 28.8% (SDNN) and 51.4% (RMSSD) for the time domain metrics and in the range of 47.5% to 103.2% for the frequency domain metrics. The lowest CV and narrowest 95% LoA were found for SDNN. Similar to the CVs, the limits of agreement were narrower for the 1-h duration compared to the 5-min duration for all outcomes (Table 2, Fig 1).

Table 2:

Test-retest reliability; 5-min and 1-h recordings; n = 37.

	Day 1, Median (p25, p75)	Day 2, Median (p25, p75)	p-value	ICC (95% CI)	CV	95% LoA
5-min
RR interval (ms)	840.0 (138.0)	815.0 (111.0)
HR (bpm)	73.4 (12.6)	75.0 (10.5)
SDNN (ms), (n = 36)	35.9 (22.2, 49.5)	38.1 (28.3, 46.5)	0.36	0.27 [−0.05 to 0.54]	47.2	±0.95 $\overline{\mathrm{X}}$
RMSSD (ms)	19.4 (12.1, 26.1)	21.0 (14.7, 33.2)	0.44	0.36 [0.04–0.61]	67.3	±1.22 $\overline{\mathrm{X}}$
HF (ms2), (n = 27)	151.7 (31.9, 417.0)	224.5 (88.3, 542.8)	0.36	0.50 [0.15–0.74]	158.0	±1.71 $\overline{\mathrm{X}}$
LF (ms2), (n = 33)	259.1 (122.2, 665.9)	298.8 (172.0, 556.0)	0.90	0.06 [−0.27 to 0.39]	151.6	±1.68 $\overline{\mathrm{X}}$
TP (ms2), (n = 25)	922.2 (493.2, 2,304.3)	1,434.1 (761.0, 2,011.5)	0.34	0.35 [−0.04 to 0.65]	86.6	±1.33 $\overline{\mathrm{X}}$
1-h
RR interval (ms)	854.0 (134.0)	843.0 (109.0)
HR (bpm)	72.1 (12.1)	72.4 (9.9)
SDNN (ms)	57.5 (47.8, 71.1)	61.2 (48.4, 79.4)	0.14	0.61 [0.37–0.78]	28.8	±0.65 $\overline{\mathrm{X}}$
RMSSD (ms)	22.9 (14.0, 29.7)	25.2 (16.7, 33.6)	0.46	0.46 [0.15–0.68]	51.4	±1.03 $\overline{\mathrm{X}}$
HF (ms2), (n = 28)	270.0 (153.1, 867.5)	432.7 (166.2, 646.2)	0.35	0.66 [0.40–0.83]	75.9	±1.30 $\overline{\mathrm{X}}$
LF (ms2), (n = 33)	467.1 (182.5, 859.1)	615.0 (261.2, 880.6)	0.55	0.13 [−0.21 to 0.45]	97.4	±1.45 $\overline{\mathrm{X}}$
VLF (ms2), (n = 36)	976.4 (479.5, 1,880.1)	1,234.3 (728.9, 2,053.7)	0.31	0.28 [−0.04 to 0.55]	67.5	±1.18 $\overline{\mathrm{X}}$
ULF (ms2)	1,297.0 (705.1, 2,620.8)	1,593.2 (875.3, 3,810.2)	0.22	0.40 [0.10–0.64]	103.2	±1.47 $\overline{\mathrm{X}}$
TP (ms2), (n = 27)	3,198.8 (1,819.6, 6,140.9)	3,622.6 (1,687.0, 7,245.0)	0.70	0.71 [0.46–0.86]	47.5	±0.97 $\overline{\mathrm{X}}$

DOI: 10.7717/peerj.16564/table-2

Note:

CI, confidence interval; CV, coefficient of variation; HF, high frequency power; ICC, intraclass correlation coefficient; LF, low frequency power; LoA, limits of agreement; p25, 25^th percentile; p75, 75^th percentile; RMSSD, root mean square of successive differences between normal heartbeats; SDNN, standard deviation of all normal-to-normal R-R intervals; TP, total power; ULF, ultra-low frequency power; VLF, very low frequency power; $\overline{\mathrm{X}}$ data mean.

Discussion

We aimed to investigate test-retest reliability of short term HRV measurement, namely 5-min and 1-h durations, when participants rest in the supine position. In general, participants showed a lower degree of HRV for 5-min and 1-h metrics compared to healthy subjects. For 5-min HRV, the medians of HF, LF and TP were 151.7, 259.1 and 922.2 ms², respectively (Table 2). The corresponding mean HRV values in healthy subjects were reported elsewhere to be 975, 1,170 and 3,466 ms² (Malik et al., 1996). Regarding 1-h HRV outcomes, mean SDNN and RMSSD were 90.4 and 47.8 ms in healthy subjects (Evrengul et al., 2006) and the medians of SDNN and RMSSD were 57.5 and 22.9 ms in our population (Table 2). The finding of lower HRV values in participants with SCI compared to healthy participants is consistent with multiple previous studies (La Fountaine et al., 2010; Rodrigues et al., 2016; Serra-Ano et al., 2015; Thayer et al., 2016).

SDNN showed lowest CV and narrowest 95% LoA. However, the overall CV of SDNN between test and retest measures of the 5-min HRV was large (47.2%) compared to previous reports in healthy participants (8.0%) (Sinnreich et al., 1998). Additionally, populations with SCI had lower relative reliability (ICC of 0.27 and 0.36 for SDNN and RMSSD, respectively) compared to healthy participants measured for the same duration and in the same position (ICC of 0.82 and 0.76 for SDNN and RMSSD) (Pinna et al., 2007).

Our study agrees with previous studies which reported that frequency domain measures (HF and LF) demonstrated larger variation and higher CV compared to time domain measures (Nunan, Sandercock & Brodie, 2010; Pinna et al., 2007; Ponikowski et al., 1996; Salo et al., 1999). The relative reliability of frequency domain measures was lower than in healthy participants. For example, for 5-min HRV reliability, our ICCs for HF and LF were 0.50 and 0.06 while the respective values in healthy participants were 0.63 and 0.79 (Pinna et al., 2007). Additionally, for the 1-h duration in a supine position, the ICCs of HF and LF were 0.66 and 0.13 in our study, and 0.81 and 0.86 in normal subjects (da Cruz et al., 2019). Our study shows much lower reliability of 1-h HRV in frequency domain measures in participants with tetraplegia (ICCs of HF and LF were 0.54 and 0.07) compared to a previous study: La Fountaine et al. (2010) found that the ICCs for HF and LF were 0.66 and 0.44 in individuals with tetraplegia and the ICCs were 0.90 and 0.74 in healthy individuals. To consider the least reliable HRV parameters among LF and HF, the results were inconsistent among studies: some studies found lower relative reliability of LF and some found the opposite (Freed et al., 1994; Nunan, Sandercock & Brodie, 2010; Sinnreich et al., 1998). Our study showed that LF had lowest relative reliability in participants with SCI based on 1-h HRV metrics.

Regarding absolute reliability, the trend of higher variability in the frequency domain was found in both healthy participants and in populations with diseases such as chronic heart failure. Nunan, Sandercock & Brodie (2010) showed that the overall CVs for SDNN and RMSSD were 32% and 37% while the CVs in LF and HF were 56% and 116% for the short term recording (5-min) of HRV in healthy adults. Ponikowski et al. (1996) demonstrated that the CVs of HF, LF, VLF and TP were 66.4%, 81.5%, 56.0% and 45.9% while the CV of SDNN was 25.4% in 5-min recording data in patients with chronic heart failure. Lord et al. (2001) showed that the CV for LF was 45% in healthy participants and 76% in cardiac transplant recipients. Our study demonstrated the CV of SDNN and RMSSD of 47.2% and 67.3% while the CV of HF, LF, VLF and TP were 158.0, 151.6, 114.8 and 86.6%. LF, HF, VLF and ULF seem to have lower absolute reliability based on CV and LoA compared to TP.

Poor reliability of short term (5-min) HRV has been identified in our study. This is in line with other studies which were done in patients with chronic heart failure, type-2 diabetes and cardiac transplant recipients (Lord et al., 2001; Ponikowski et al., 1996; Sacre et al., 2012). One possible explanation is the low initial HRV values reported in clinical populations, therefore, slight changes in the test-retest value can result in a high CV. Lower absolute values among participants may also lead to lower ICC (Sandercock, Bromley & Brodie, 2005). Regarding 5-min and 1-h reliability of HRV, we found that the supine position did not provide better reliability compared to unrestricted activity in participants with SCI (Ruangsuphaphichat et al., 2023). This has to be interpreted with caution. Firstly, the results not showing improvement in reliability may be partially explained by differences in group proportions: the present study had a higher proportion of participants with tetraplegia (37.8%) compared to the previous study (22.2%) (Ruangsuphaphichat et al., 2023). We also had data from participants with complete tetraplegia while the former study did not. We found that individuals with tetraplegia showed poorer HRV reliability compared to those with paraplegia. This may be explained by the alteration in cardiogenic autonomic control in participants with higher lesion level, which results in higher inter-day variation of HRV (Bauman et al., 2012). Secondly, the influence of different body positions might contribute to the observed differences in reliability results. In a previous study, a reliability analysis of HRV in 37 non-SCI young men in different positions demonstrated noteworthy differences. The ICCs of HRV in the supine position were found to be lower than in the standing position. Specifically, the ICCs were as follows: log of SDNN (ICC of 0.60 vs 0.88 (supine vs. standing)), log of RMSSD (ICC of 0.74 vs. 0.86) and log of HF (ICC of 0.81 vs. 0.84). However, no such trend was observed for LF (ICC of 0.63 vs. 0.32) (da Cruz et al., 2019). This might imply that distinct body postures influence the reliability of HRV, particularly in the context of individuals with SCI. Given these findings, there emerges a potential need for further investigation into the impact of different postures on HRV metrics and their reliability within the SCI population.

Our participants had various medications, and some medications have been shown to alter HRV, e.g., gabapentin, antihypertensive medication or tricyclic antidepressants (Ermis et al., 2010; Miyabara et al., 2017; Pavithran et al., 2010; van Zyl, Hasegawa & Nagata, 2008). However, since the medications were taken at regular times, these should not have directly affected test-retest differences. Our study had too few participants to conduct a robust comparison of HRV data in individuals with respect to whether there was a complete or incomplete lesion, which may lead to additional differences in reliability of HRV. Multiple invalid data sets due to poor signal recording has to be improved for clinical application, e.g., by careful monitoring or regularly checking the data recording during the test period. Our results were limited to the supine position, and the single centre study may limit data generalization. It may be possible that different positions, or measurements with paced breathing, may lead to different reliability outcomes (Nunan, Sandercock & Brodie, 2010; Pinna et al., 2007; Sinnreich et al., 1998; Tonello et al., 2015). Additionally, the inclusion of age- and sex-matched able-bodied controls might have provided more comparable data in the same measurement setting and allowed more insightful interpretation. For future studies, emphasis on examining reliability in different body positions or during paced breathing sessions might offer avenues to improve the reliability of HRV measurements within this population. By including a larger number of participants, it would enable the analysis of the potential impact of variables such as the severity and completeness of the lesion.

Conclusions

The supine position did not provide better reliability compared to unrestricted activity in participants with SCI. HRV analysis using a 5-min duration is of limited value due to poor reliability in individuals with SCI. For the 1-h analysis duration, SDNN, RMSSD, HF and TP have moderate to excellent reliability overall and when grouped based on lesion level; ULF has moderate to good reliability overall and for paraplegia; VLF has moderate to good reliability for paraplegia only; but LF has poor reliability overall and in both groups. Interpretation of the reliability of HRV varies according to lesion level: it is recommended to take lesion level into account when interpreting reliability measures.

Supplemental Information