Guidelines for Rigor and Reproducibility of Heart Rate Variability within Human Cardiovascular Research

Abstract

Heart rate variability (HRV) is a widely utilized approach for investigating cardiovascular health and human performance. Methodological approaches, analyses, and interpretations of HRV can vary extensively, and scientific debates on HRV reliability and utility remain. These debates have been amplified in recent years by the rapid growth and ubiquity of wearable devices, and their incorporation into scientific research studies. The present guidelines paper takes an evidence-based approach to propose more consistent and rigorous measurement and interpretation of HRV within human cardiovascular research and application. First, HRV has some utility as a cardiovascular risk stratification tool, but is not appropriate to employ as a specific marker of cardiac sympathetic outflow or sympathovagal balance. While time-domain and spectral analysis can more accurately reflect respiratory modulation of cardiac vagal (parasympathetic) activity, caution is necessary to avoid overinterpretation, particularly with respect to the concept of “vagal tone”. Numerous experimental, demographic, and environmental factors influence HRV assessment, interpretation, and reliability. Specifically, it is critical to account for non-modifiable (age, sex, race/ethnicity, etc.) and modifiable (i.e., exercise, physical fitness, etc.) factors within participant samples. Furthermore, technical approaches such as the HRV input signal (e.g., electrocardiography vs. photoplethysmography), length of recording, location of recordings (i.e., laboratory vs. field), respiratory rate and depth, and analytical approaches (i.e., time vs. frequency domain) can all impact rigor, reliability, and study interpretations. Finally, with respect to the rapid advancements of HRV assessment using wearable technology, investigators should interpret and contextualize findings within the limitations outlined in this guideline review.

INTRODUCTION

Heart rate variability (HRV) is a common physiological assessment in human cardiovascular and performance research. It is most accurately quantified using R-R intervals from electrocardiography (ECG), but the advancement of wearable technologies employing photoplethysmography (PPG) to assess continuous pulse rate has contributed to a dramatic rise in HRV research in recent years (Figure 1). To fully appreciate the experimental complexities and caveats of HRV requires some historical context. Briefly, several scientists and physicians have developed methods and use for HRV: a comprehensive review by Billman [1] highlights the ancient Greek physician and scientist Herophilos (335–280 BC), who is credited with the first estimates of heart rate via pulse measurements, Galen of Pergamon (131–200 AD), who wrote the initial reports of acute exercise on pulse rate, and John Floyer (1649–1734), who invented the “Physician Pulse Watch” which was used to assess pulse rate while simultaneously examining respiration under various conditions. The Reverend Stephen Hales (1677–1761) extended the work of Floyer by observing fluctuations in pulse rate – which represented the beginnings of HRV research – and reported that such fluctuations varied during the respiratory cycle [1, 2]. In 1847, Carl Ludwig used his invention, the smoked drum kymograph, to record fluctuations in pulse rate that increased during inspiration and decreased during expiration – documenting for the first time what we now know as respiratory sinus arrhythmia (RSA) [1, 3], or more recently termed respiratory heart rate variability (respHRV). However, it was not until the establishment of ECG by Willem Einthoven using galvanometers [1, 4], and the use of ambulatory ECG by Norman Holter [5], that HRV research became more practical and readily accessible to extensive networks of researchers in the early-to-mid 1900s.

Figure 1.

Pubmed search citations on the term “heart rate variability” over the past 100 years (date range: 1925 – 2025, data accessed December 7, 2025).

During the 1970’s, several laboratories began to embrace the digital/computing revolution and apply novel signal processing approaches to HRV, which led to the establishment of time and frequency domain analyses [1, 6–9]. In 1981, Akselrod and colleagues [10] published seminal work in Science that delineated the critical influence of parasympathetic (PNS) and sympathetic (SNS) autonomic nervous system activity on resultant HRV metrics through the utilization of pharmacological manipulation in dogs. Throughout the 1980’s and early 1990’s, the utilization of HRV expanded dramatically – but not without inconsistent methodology, misconceptions, and field debates. For example, one contentious topic was the misleading use of frequency-domain HRV metrics to interrogate general PNS and SNS function, not just to the heart. Such debates culminated in several attempts to add clarity, including the 1996 European Task Force Report [11], but even this attempt was promptly challenged [12]. Specifically, Eckberg [12] presented the underlying mathematical assumptions and limitations of power spectral analysis, and documented the lack of evidence for “sympathovagal balance” based upon some key pharmacological blockade studies. He also posited that while HRV has promise for “risk stratification” purposes, it should not be used to interrogate the independent activity of the PNS and SNS.

Physiological Basis for Heart Rate Variability

A healthy heart does not beat with metronomic precision (Figure 2). The intervals between heart beats exhibit variability that largely reflect the actions of the autonomic nervous system on the sinoatrial (SA) node of the heart, which receives both sympathetic and parasympathetic innervation. Cardiac sympathetic neurons increase heart rate (i.e. exert a positive chronotropic effect) and the force of contraction (i.e. a positive inotropic effect), while cardiac parasympathetic neurons – which travel in the vagus nerve – decrease heart rate and force of contraction (negative chronotropic and inotropic effects). However, differences in mechanisms underlying the chronotropic influences of PNS versus SNS branches help demonstrate why HRV is predominantly influenced by cardiac vagal activity. [13–15]. Reduced adenylyl cyclase activity and cAMP-production lead to downstream inhibition of the hyperpolarization-activated “pacemaker” current and reduced influx of ions such as calcium. The combination of potassium efflux and inhibition of adenylyl cyclase reduces the rate of spontaneous depolarization within pacemaker cells and reduces heart rate (i.e., increased heart period) [16]. Activation of adenylyl cyclase by ACh further opposes sympathetically--mediated β-adrenergic activation of adenylyl cyclase and cAMP production [13]. Importantly, ACh is rapidly degraded due to the effects of cholinesterase activity within the synaptic cleft, and prior studies have demonstrated that the rapid effects of vagal efferent activity on inter-beat cardiac chronotropic activity is heavily contingent on the swift degradation of ACh [17, 18]. Upon cardiac vagal withdrawal, the post-synaptic influences of previously released ACh are rapidly ameliorated. Conversely, removal and degradation of catecholamines at the level of the heart are primarily dependent on neuronal reuptake [19], leading to a prolonged post-synaptic effect which cannot underly the acute differences in heart period between consecutive beats. Therefore, the parasympathetic nervous system is uniquely equipped to carry out its influence on cardiac rhythmicity on a beat-by beat basis through 1) rapid axonal conduction; 2) direct and indirect influences on the rate of depolarization within cardiac pacemaker cells; and 3) rapid degradation and cessation of cardiac influence due to the presence of acetylcholinesterase within the synaptic cleft. Through these mechanisms, vagal innervation at the level of the SA node evokes beat-to-beat alterations in spontaneous rhythmicity of cardiac pacemaker cells, resulting in what we measure as HRV. These physiological observations offer a firm foundational basis for the assumption that HRV represents primarily vagal activity, rather than sympathetic influences.

Figure 2.

Hypothetical electrocardiograms illustrating how the root mean square of successive differences (RMSSD) of heart rate variability (HRV) can be widely different (65 ms in top panel vs. 130 ms in bottom panel), but still result in the same average heart rate of ~76 beats per min (bpm).

This current Guidelines paper aims to build upon the basic premise posited by Eckberg [12] that HRV has an important and expanding role in health, performance, and disease risk – particularly its potential for risk stratification purposes – but that its use for assessing cardiac sympathetic activity and/or sympathovagal balance should be avoided. We will present evidence from the past four decades that support this consensus stance, including several pivotal studies published within the American Journal of Physiology – Heart and Circulatory Physiology [20–24]. We accomplish our objective via ten focused questions related to key methodological and interpretive issues. We will then conclude with practical guidelines aimed at improving rigor and reproducibility of HRV within human cardiovascular research.

Question 1. What does HRV purport to measure, and what does HRV actually measure?

In the past decade HRV has become a buzzword in the fitness and wellness domain. Even popular media outlets have documented the overuse of HRV for health and fitness monitoring in recent years [25]. One of the many problems with the overuse of indices related to HRV is that the term itself is misleading. First, the name suggests that HRV is a measure of the variation in heart rate (i.e., beats/min), whereas it is instead a measure of the variability of the interval between adjacent R-waves (the R-R interval; RRI, or the heart period). Figure 2 depicts typical RRIs and illustrates how similar heart rates can have widely varying RRIs and accompanying assessments of HRV. This differentiation becomes important when one considers discrepancies in linearity between vagal activation, heart rate, and heart period [26–28] which are reciprocals of each other [29, 30]. Specifically, vagal activation exhibits a more linear association with heart period as opposed to heart rate [26, 27], and this association is less susceptible to influences of simultaneous sympathetic activation within the normal operating range [27]. These findings support the utility of heart period over heart rate when assessing parasympathetic regulation. As such, prior calls for the utilization of the term “heart period variability” have been put forward [31]. Nonetheless, the use of the phrase HRV is far more common across the fields of both physiology and psychology, and indeed across the exercise and fitness communities as well.

Traditionally, high-resolution ECG monitoring of RRI and HRV have been used to quantify various time and frequency domain indices (for detailed overview, see [11, 28]). Detailed descriptions and recommendations for use of each time and frequency domain HRV metric were summarized by the 1996 Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology [11]. These include time domain assessments of root mean squared of successive differences (RMSSD), standard deviation of normal-to-normal RRI (SDNN), standard deviation of the time between NN intervals (SDANN), and the percentage of RRIs that differ by greater than 50ms (pNN50), alongside both low (LF; 0.04–0.15Hz) and high (HF; 0.15–0.4Hz) frequency domain assessments [11]. Table 1 summarizes common time and frequency domain HRV metrics, as well as some less commonly employed nonlinear HRV metrics. Since the publication of the Task Force recommendations 30 years ago [11], significant technological advances have resulted in commercially available wearable devices, which advertise continuous heart rate and HRV monitoring [32]. Most of these devices rely on PPG, which detects beat to beat changes in blood flow, often measured over the wrist. In contrast to an ECG signal, PPG signals are light-derived and obtained by illuminating the skin; changes in microvascular blood volume are subsequently estimated by how absorption, scattering, and reflectance of light changes during each heartbeat back to a light sensor embedded in the device [33–35]. As such, many consumer devices are not monitoring HRV, but rather pulse rate variability. Due to the noninvasive nature of currently available wearable technology, monitoring of pulse rate variability for longer periods of time is readily attainable and accessible in large-scale populations. However, the PPG signal is subject to multiple limitations due to movement artifacts, environmental influences on vascular flow (e.g., temperature), and the absence of R-waves for detection and hence great temporal jitter [28]. Furthermore, skin pigmentation may influence accuracy of heart rate monitoring using PPG due to increased light absorption by melanin [36, 37], resulting in discrepancies in HRV monitoring across different populations. Therefore, caution is warranted when interpreting findings from such devices (see Question 10: Do most of today’s wearables provide rigorous and reliable HRV metrics?). While the technical language used to describe differing modes of HRV assessment may seem trivial, specificity in denoting what input signal is being used to quantify HRV (i.e., ECG, PPG, etc.) is needed to appropriately interpret the reliability of the findings, which has important implications for cardiac autonomic regulation. As discussed in Question 10, even if the quantification is validated, the ability to extend metrics about heart period variability to have meaning in other contexts (e.g., fatigue, sleep quality, exercise prescription) is currently limited and should be viewed with caution.

Table 1.

Common time and frequency domain measures of heart rate variability (HRV), as well as some less commonly reported nonlinear HRV metrics.

Variable	Units	Description	Analysis Type
SDNN	ms	Standard deviation of all normal-to-normal (NN) intervals.	Time Domain
SDANN	ms	Standard deviation of the average NN intervals.	Time Domain
RMSSD	ms	Root mean square of successive R-R interval differences	Time Domain
SDNN index	ms	Mean of the standard deviations of all NN intervals.	Time Domain
SDSD	ms	Standard deviation of differences between adjacent NN intervals.	Time Domain
NN50 count	count	Number of pairs of adjacent NN intervals differing by more than 50 ms.	Time Domain
pNN50	%	Percentage of adjacent NN interval pairs differing by more than 50 ms (NN50 count divided by the total number of NN intervals).	Time Domain
Total power	ms2	Variance of NN intervals over the 5-min recording segment; frequency range of ≤ 0.40 Hz	Freq. Domain
VLF	ms2	Power in the very low frequency band of 0.003 – 0.03 Hz	Freq. Domain
LF	ms2	Power in the low frequency band of 0.04 – 0.15 Hz	Freq. Domain
LF norm	n.u.	Low-frequency power expressed in normalized units	Freq. Domain
HF	ms2	Power in the high frequency band of 0.15 – 0.40 Hz	Freq. Domain
HF norm	n.u.	High-frequency power expressed in normalized units	Freq. Domain
LF/HF	ratio	Ratio of low- to high-frequency power	Freq. Domain
SampEn	unitless	Non-linear quantification of regularity of the HRV time series, or the differential entropy rate.	Other
DFA a1	unitless	Non-linear quantification of the fractal dynamics of the HRV time series, describing the short-term (4–11 beats) scaling of the signal’s fluctuations	Other
DFA a2	unitless	Non-linear quantification of the fractal dynamics of the HRV time series, describing the long-term (11+ beats) scaling of the signal’s fluctuations	Other
Heart Rate Asymmetry	%	Analyses of the differential contributions of HR accelerations and decelerations to HRV features	Other

Question 2. What are the key methodological considerations for enhancing HRV reliability and utility?

The reliability of HRV depends on careful consideration of several methodological factors including: (1) signal acquisition and preprocessing, (2) recording conditions and confounding/contextual variables, and (3) data processing and analytical approaches. Each of these are briefly discussed below, with a specific focus on laboratory-based assessments of HRV in humans. The interested reader is also pointed to other recent excellent reviews [28, 31, 33, 38].

Signal Acquisition and Preprocessing.

In the laboratory setting, the highest fidelity (and most common) recordings of cardiac electrical activity are obtained by an ECG using commercially available hardware and software. For most applications, surface electrode placements are chosen to yield a prominent R-wave relative to other waves in the ECG (most often Lead II). A modified Lead II electrode placement (i.e., active leads on a lower left rib and right collarbone, with reference leads on a lower right rib) is particularly advantageous because it minimizes movement artifact that can occur with placement on limbs [28]. This montage is also advantageous as high-pass filtering of the ECG signal provides an estimate of inspiratory activity if breathing is not being recorded directly from a respiratory transducer around the chest and/or abdomen or via nasal airflow or spirometry. It should be noted that deriving periodicity from waveform features other than the R-wave (e.g., P-wave onset) is not recommended for controlled, in-laboratory testing. Regardless, poor surface electrode contact, faulty conduction by lead wires, extraneous magnetic or electrical noise (e.g., 60 Hz), excessive participant movement, and/or respiratory EMG interference are all potential sources of artifact that can significantly impact the quality of the input signal. Care should be taken to minimize or eliminate these potential experimental confounders to the extent possible, with filtering of the ECG signal also reducing movement or sweat-induced fluctuations in baseline.

Most clinical and research ECG-based monitoring systems sample the analog ECG sample at 250–1,000 Hz (see Table 2), convert it to digital form, and store it via a computer-based beat-to-beat data acquisition system. Analytic software then utilizes algorithms to recognize ECG waveform characteristics. The sampling rate must be properly selected. Lower sampling rates can produce a “jitter” in the estimation of the R-wave fiducial point that causes error in accurately detecting R-waves, and also may introduce error in the HRV spectrum that increases with frequency, thus affecting more high-frequency spectral components [11, 39]. To reduce these potential sources of error, the optimal sampling frequency range is 250–500 Hz, though some recommendations suggest a sampling frequency of 1,000 Hz when assessing intra-ECG information, such as signal averaging of either P waves or the QRS [11, 40–42]. There appears to be limited additional time domain value at higher sampling frequencies, thus for many human HRV studies a sampling rate of 250 Hz is employed [28]. However, for frequency domain analyses, consistent with the Nyquist-Shannon sampling theorem, the sampling rate must be at least twice the highest frequency component of the target frequency in the signal.

Table 2.

Recommendations for standardizing HRV data collection in controlled laboratory settings versus real-world (ambulatory/wearable) settings.

Domain	In-Lab (Controlled) Recommendations	Real-World / Field (Ambulatory & Wearable) Recommendations
Pre-session diet, caffeine, alcohol & exercise	• Ideally, test the morning after overnight fast. Standardize and document a pre-session window (e.g., ≥3 h fasted, ≥6 h caffeine-free, ≥12 h alcohol-free, ≥12–24 h without vigorous exercise). • Apply the same window to every participant and visit; record and report any deviations.	• Strict standardization is possible with planning; consider “data collection bursts” where conditions parallel in-laboratory controls (i.e., fasted mornings) • Require participants to log food, caffeine, alcohol, and exercise (timestamped diary or app). • If utilizing longer durations with confounders, use logs to flag and contextualize — or exclude — confounded recording windows.
Sleep & time-of-day	• Schedule sessions at a consistent clock time across participants and especially repeat visits. • Capture prior-night sleep duration/quality (questionnaire or actigraphy) and avoid testing immediately upon waking unless that is the variable of interest.	• Leverage continuous recording to capture planned quiet rest periods. • With proper controls, nocturnal/sleep has been shown to offer stable and reproducible HRV assessment. • Shift work or irregular sleep schedules have not been adequately studied for reliability, and should be avoided.
Body position & posture	• Maintain a single posture (e.g., supine or seated) for the full session; standardize the setup (chair, table height) across participants. • Allow ≥10 min of quiet rest in that posture immediately before recording begins.	• Focus analysis to windows of relative inactivity (ideally stable morning recording and/or sleep) and report the activity context alongside HRV values.
Environmental conditions	• Standardize and report room temperature, humidity, lighting, and ambient noise; minimize conversation; no phone/device use by participants. • Keep conditions consistent across all sessions for a given participant and across participants.	• Record ambient context where possible (location & preceding activity tags). • Document variability conditions encountered (e.g., travel = different location and time zones) and acknowledge as a source of variability.
Equipment & input signal	• Use ECG with a modified Lead II placement; sample at ≥250 Hz (250–500 Hz preferred); confirm signal quality before and during recording. • Calibrate equipment on a regular, documented schedule.	• Report device make/model and input signal (e.g., PPG); favor devices independently validated against ECG under conditions resembling intended use. • Document firmware/algorithm versions where available and disclose known signal-quality limitations of the chosen device.
Recording duration & epoch selection	• Use ≥ 5-min (ideally ≥ 10-minute) steady-state recordings for short-term HRV; keep epoch length identical across conditions and participants. • Allow a brief stabilization period (ideally ≥ 10-minute) before the analysis window begins (same posture, minimal talking).	• Favor longer continuous recordings (e.g., multi-night) to offset noise, and pre-specify epoch-selection/exclusion rules (e.g., quiet wake, stable sleep). • Average multiple windows where possible
Respiration	• Record respiration directly (chest/abdominal transducer or nasal airflow); favor verified spontaneous breathing within the normal range over forced paced-breathing protocols unless respiration is the primary variable of interest.	• Direct respiration monitoring is challenging in the field and home studies, but not impossible. • Isolate assessment to relatively stable breathing periods (i.e., quiet rest, sleep, etc.).
Behavioral state during recording	• Instruct participants not to speak, move, or use devices during the recording. • Standardize any tasks/stressors and their order across participants.	• Use accelerometer/activity-tracker data to characterize and report behavioral state at the time of each recording (resting, walking, exercising, etc.). • Avoid pooling HRV across heterogeneous activity states.
Participant-level factors to document	• Record and, when appropriate, schedule around menstrual cycle phase/hormonal status, medication timing, and hydration; align repeat visits to similar menstrual phases. • Capture and document sociodemographic factors such as biological sex, gender, race, ethnicity, other health metrics (e.g., blood pressure, lipids), cardiorespiratory fitness, social determinants of health, and subjective assessments of stress, illness, etc. • Capture health behaviors such as physical activity, and dietary patterns, and sleep.	• Capture self-report data (symptoms, medications, stress, illness, menstrual phase) via logs or ecological momentary assessment to aid interpretation of free-living variability. • Capture and document sociodemographic factors such as biological sex, gender, race, ethnicity, other health metrics (e.g., blood pressure, lipids), cardiorespiratory fitness, social determinants of health, and subjective assessments of stress, illness, etc. • Capture health behaviors such as physical activity, and dietary patterns, and sleep.
Documentation & reporting	• Report all standardization procedures, deviations, and exclusions in the methods; provide sufficient detail for replication (electrode placement, software/version, filtering parameters, etc.). • Pre-register clinical trial (e.g., clinicaltrials.gov) when appropriate.	• Report device specifications, recording duration and compliance, data-cleaning/exclusion criteria, and the proportion of usable data. • Explicitly frame findings as derived from heterogeneous, lower-fidelity data and interpret accordingly. • Pre-register analytical plan (e.g., cos.io/initiatives/prereg) when appropriate.

This table provides actionable recommendations for study design and reporting. It is intended as a practical checklist rather than an exhaustive review. Investigators should adapt these recommendations to their specific research question while documenting any departures. Definitions: ECG, electrocardiography; HRV, heart rate variability; PPG, photoplethysmography.

As previously noted, PPG sensor-based recordings, via smartwatches, fitness bands, and other portable devices, are becoming increasingly common. Notable advantages of this approach include increased ease, convenience, and versatility of ambulatory and field-based data collection. However, these advantages must be appropriately weighed against various shortcomings, including a lower signal-to-noise ratio, poorer temporal resolution, more movement and motion artifact, differing and varying specifications across devices, proprietary algorithms that companies are reluctant or unwilling to share, and often less accuracy detecting waveform features [43–45]. Interested readers are directed to the following review articles for a more detailed discussion of PPG recordings [33, 38], as well as Question 10: Do most of today’s wearables provide rigorous and reliable HRV metrics?, for a detailed and comprehensive description of ambulatory and field-based assessments of HRV.

Recording Conditions and Confounding/Contextual Variables.

For laboratory studies, environmental control is crucial (e.g., laboratory temperature and humidity, ambient noise, lighting, time of day, etc.). Care should be taken to ensure that the recording environment is consistent for all participants, and these characteristics should be described in detail when disseminating results (Table 2).

The duration of the recording will be dictated by the nature of each investigation. Because HRV tends to increase with the length of the analyzed epoch, due to a greater likelihood of nonstationary and loss of signal stability, the length of recording time should be carefully considered [46]. In addition, the recording duration must allow enough cardiac cycles of the oscillation of interest to be included to calculate reliable estimations. Consideration of the experimental protocol and conditions for the recording, as well as the Nyquist-Shannon sampling theorem, are critical [11, 28]; however, for measurements of HRV, a minimum 5-minute recording is recommended. While time domain methods (SDNN and RMSSD) are generally considered more ideal for the analysis of longer-term recordings [11], they can also be used to investigate short-term recordings, as shown by the evidence that time domain analyses of recording epochs as short as 10–30-sec aggregated across multiple segments within a laboratory session closely approximate results obtained from longer recording durations [11, 47]. Nevertheless, for consistency and standardization, a minimum 5-minute recording is favored.

Modulatory variables, including respiration, speaking, and posture, can have a significant impact on measures of HRV [48, 49]. Respiratory rhythm has a profound impact on HRV: during inspiration, there is a relative inhibition of cardiac parasympathetic (vagal) activity, whereas the opposite occurs during expiration [48]. Respiratory influences on cardiac vagal activity have classically been referred to as RSA. While the term arrythmia in RSA implies pathophysiology, the respiratory variation in heart rate is not arrhythmic and certainly not pathological. A recent international consensus statement has argued that RSA should be redefined as respiratory heart rate variability, or simply respHRV [50]. We agree with this recommendation as an alternative term, as it appropriately emphasizes the respiratory component of HRV. As such, we refer to respiratory modulation of HRV as respHRV throughout the remainder of the guidelines.

The effects of breathing on cardiac autonomic function are observed at the peak of the high-frequency band of HRV (0.15–0.4 Hz). Within the typical range of resting breathing frequencies (~7–24 breaths/minute in adults), respHRV is inversely related to respiratory rate and directly related to tidal volume. In other words, slower respiratory rates or deep inspiratory volumes will lead to an increase in respHRV This relation is evident for both spontaneous and paced breathing. As such, both aspects of respiratory behavior should be considered, and use of a respiratory transducer during recordings of heart rate is recommended to monitor rate and depth of breathing (Table 2) [51, 52]. To minimize any confounding influence of respiration on indices and interpretations of HRV, recommendations have sometimes included the consideration of paced breathing, verifying that any experimental manipulations do not appreciably alter respiration, and/or statistically adjusting for changes in respiration to account for the respHRV [28]. If experimentally pacing respiration to external cues, investigators should be aware that voluntarily controlling breathing may be effortful for the participant, which itself can directly influence cardiac autonomic control (see Question 9. Do breathing rates and tidal volume need to be fixed for HRV?).

Speaking, which naturally disrupts the inspiratory-expiratory respiratory rhythm, also profoundly influences measures of HRV, including respHRV, beyond changes in breathing rate and tidal volume alone [53, 54]. As such, participants should be instructed not to speak during recordings at rest (Table 2). Given that some commonly applied laboratory stressors require vocalization (e.g., mental arithmetic, Trier Social Stress Test, etc.), with the attendant changes in respiratory rate, depth (and increases in intrathoracic pressure in some conditions), direct comparisons of HRV metrics and respHRV estimates between speaking and non-speaking periods should be avoided. Finally, postural changes impact cardiac autonomic function; for this reason, posture should be kept constant across experimental conditions and/or treatments, and between participants for cross sectional analyses [55].

Data Processing and Analytical Approaches.

In addition to high-quality signal acquisition, the ability to accurately detect the point in each cardiac cycle from which measures are made, during analysis, is critical for reliable HRV assessment. The consensus recommendation is to use automated detection software in concert with investigator visual inspection and confirmation [11, 28]. Artifacts due to measurement error, movement, and other non-physiological sources described below, as well as those due to experimenter- or algorithm-induced processing errors, can lead to spurious or missed heartbeats, both of which can impact the derivation of HRV metrics.

Because the inspection and correction of spurious and/or missed heartbeats can be tedious when done manually, especially with longer recordings or for many participants, most researchers and clinicians have adopted automated pre- and post-processing strategies to assist. These range from simple smoothing and filtering of the digital signal to peak detection algorithms that can detect potentially spurious beats based on the extent of variability in the duration of nearby cardiac cycles [56, 57]. Newer analytical techniques, such as independent components or wavelet-transform-based approaches, can further improve the accuracy of R-wave detection [58, 59]. Any automated approach should always be followed by careful visual inspection by a trained research team member, as automatically detected outliers are not always due to artifact. Indeed, abnormal or ectopic beats, arrhythmic events, missing data, and noise effects can also alter the R-R interval and thus estimates of HRV. To reduce this error, interpolation based on preceding/successive beats or on its autocorrelation function are commonly employed [11]. In these cases, visual inspection is necessary to justify corrections, and investigators should report both the relative number and relative duration of R-R intervals that were omitted or corrected.

When beat-to-beat fluctuations in RRIs are the metric of interest, analyses should be limited to data segments that are free of abnormal or ectopic beats. However, omitting these RRIs can lead to selection bias and should therefore be explicitly acknowledged, along with reporting removal of data [28]. When estimating HRV, handling of ectopic beats requires careful consideration [60]. If the ectopic beat is ventricular (e.g., premature ventricular contraction, PVC), the effect of the ectopic beat can be handled by assigning it the midpoint of the two beats immediately adjacent, which will minimally impact HRV. In some individuals, even if free of disease, the number of abnormal beats can be quite large (e.g., >10% of beats), with this likelihood increasing in older and less healthy participants [61]. In these situations, limiting analyses to data segments free of abnormal beats, while necessary, must be balanced against the potential influence of selection bias on estimates of HRV; any such bias would only be further exaggerated with an increasing number of abnormal beats [11, 28]. Moreover, it is important to acknowledge that after a PVC, the following beat will have increased contractility both from Starling effects and increased calcium release from the compensatory pause, which increases stroke volume and baroreflex simulation of the next beat. If there are a lot of PVCs, interpolation does not capture the regulation of the cardiac period and should be avoided. Finally, cases in which ectopic beats exceed 20–30% of all heartbeats are not considered suitable for HRV analyses [62]. Regardless of approach, all data that have been removed due to ectopic beats should be reported in methodologies for full transparency.

Question 3: Is HRV reliable/reproducible in traditional test retest?

The utility of HRV for cardiovascular risk stratification (see Question 6) and for monitoring changes in response to intervention (see below) is contingent upon its reliability. When examining reliability, both absolute (i.e., variation of a measure within an individual across multiple measurements) and relative (i.e., consistency of one’s rank within a larger group across multiple measurements) metrics need to be considered [63]. Numerous studies suggest that HRV exhibits moderate to excellent relative test-retest reliability (i.e., intraclass correlation coefficients (ICC)) when tested across days to weeks [64–70]. In contrast, absolute reliability, often measured using the coefficient of variation, standard error of measurement, and/or 95% limits of agreement, is generally poorer and indicates that HRV metrics display substantial within participant variability [64, 68, 69, 71]. This section aims to build upon prior reviews [28, 72] by highlighting studies which have assessed HRV reliability in response to various external and technical factors that are crucial to proper use of HRV within human cardiovascular research.

Within commonly employed short term HRV assessment (i.e., ranging from several seconds to several minutes), increased recording time is associated with improved absolute and relative reliability [64, 66, 73]. Longer times facilitate greater validity of comparisons with HRV values obtained using the recommended minimums [11] of 5-minute assessment periods [47, 64, 74]. For instance, Schroeder and colleagues [66] reported that the ICC values for SDNN and RMSSD recorded in 10-sec epochs 1–2 weeks apart ranged from 0.41–0.57, indicating poor to moderate reliability, but increased to 0.70–0.91 when using 2- and 6-minuterecordings. Improvements were also observed in frequency domain HRV metrics with the 2- to 6- minute recording times [66]. When multiple ultra-short term HRV recordings (i.e., multiple 10-sec recordings) were averaged, reliability improved [66]. Similar findings were described by Burma et al. [64], who reported improvements in the coefficient of variation (i.e., absolute reliability) and ICC (i.e., relative reliability) of HRV upon retesting two weeks apart when recording durations were increased from 30-seconds up to 5-minutes. However, when assessing longer durations utilizing techniques such as 24-hour ECG Holter monitoring, it becomes similarly important to account for the likely influence of factors such as daily activity patterns [75]. Under such circumstances, researchers should consider assessing HRV during times of stable activity such as during quiet rest periods or stable sleep free of apneas and other sleep disorders (and with respiratory data), where there is little confounding due to effects of movement or other behavioral factors (Table 2) [76].

While a longer recording duration appears to improve test-retest reliability, a prolonged time between initial and follow-up testing sessions may have the opposite effect [71, 73, 77]. For example, a test-retest assessment of 10-minute resting HRV across three time points separated by ~15 days (i.e., time point 1 vs. 2) and ~210 days (i.e., time point 2 vs. 3) found that ICCs across time points ranged from 0.23–0.56 for time domain, and 0.29–0.75 for frequency domain measures [77]. Alternatively, quantifying time domain HRV metrics using 24-hour ECG recordings resulted in improved ICCs ranging from 0.57–0.75 [77]. Subsequent studies similarly reported poor-to-moderate test-retest reliability in a longitudinal assessment of time (ICCs: 0.35–0.64) and frequency (ICCs: 0.48–0.60) HRV in 103 participants tested on 5 separate testing sessions 2–4 months apart [71]. Collectively, these studies demonstrate the variability in test-retest reliability during short- and long-term recordings.

Various studies suggest that time-domain assessments of HRV are more reliable than frequency-domain metrics in test-retest paradigms [64, 70, 78]. Schumann and colleagues [70] reported moderate-to-good ICCs amongst time-domain HRV metrics (ICC from 0.75–0.83), while frequency-domain metrics fell below this range (ICC from 0.54–0.74) when participants’ ECG data was collected for 15 minutes across two sessions separated by a week. Similarly, the absolute reliability of time domain metrics (with the exception of pNN50) was better (CVs = 19–21%) relative to frequency domain measures (CVs = 31–37%) [70]. Poorer absolute reliability in traditional frequency domain compared to time domain HRV measures was similarly observed by Burma et al. [64] who reported CVs between 4–21% in time-domain, and 16–40% in absolute frequency-domain measures following two, 5-minute recording sessions separated by two weeks. Of the frequency-domain metrics, some studies suggest the high-frequency component to be more reliable than the low frequency component [64, 69, 70, 76, 79–81], although this may be influenced by factors previously discussed such as recording duration and respiratory rate. Indeed, increasing recording duration from 2- to 6-minute [66] and 30-sec to 5-minute [64] significantly improved relative and absolute reliability of frequency domain HRV metrics, and led to a negligible difference in reliability between LF and HF components with greater recording time. Although test-retest reliability may be similar between frequency domain metrics under optimal recording conditions, the underlying physiological mechanisms appear to differ, and may be more ambiguous particularly in the case of LF and LF/HF measures.

Regarding the influence of overall health status, pre-existing chronic conditions such as heart failure, arrythmias, and spinal cord injury influence the reliability of HRV. Specifically, individuals with heart failure exhibit high coefficients of variation in both time (25–139%) and frequency (45–111%) domain HRV metrics, signifying poor absolute reliability [82]. In myocardial infarction patients, absolute reliability was low, while relative reliability exceeded an ICC > 0.7 [83]. Individuals with spinal cord injury appear to maintain relatively stable HRV over time (moderate to good ICC), particularly with increased recording durations [73, 84]. Lastly, HRV is significantly associated with and impacted by changes in behavioral states and environmental influences. Numerous studies have investigated the influence of factors such as breathing patterns (i.e., spontaneous versus controlled breathing) and posture (i.e., supine vs. sitting vs. standing) on HRV reliability, but the findings are largely ambiguous. For instance, while some studies report slight improvements in HRV inter-day reliability during controlled breathing (often at 15 breaths/minute) [68, 77], others report negligible differences between breathing conditions [67, 70, 85, 86]. Similarly, while HRV reliability may be reduced during postural challenge [68, 72, 77], others report marginal and non-systematic differences in HRV reliability compared between supine, seated, and standing positions [71, 86]. While acute changes in respiration and posture influence resting HRV, a fair deal of ambiguity remains surrounding the influence of these factors on reliability across time within an individual. [72]

Additional uncertainty presents itself in studies when behavioral alterations such as mental stress exposure [87, 88], acute exercise [89], and sleep [76, 90, 91] are performed. Though each of these behavioral parameters influence HRV acutely, far fewer reliability studies have been performed under differing environmental and behavioral constraints. During light exercise, HRV metrics appear to exhibit adequate absolute and relative reliability when quantified across epochs of ~3–6 minutes [92–94]. For instance, during light cycling and walking, the average ICCs of traditional time and frequency domain HRV metrics were ~0.79–0.92 and ~0.81–0.91, respectively [92–94]. Others have also reported moderate-to-good relative reliability of HRV during six-minutes of steady-state moderate intensity cycling (ICC: 0.58–0.85), but reduced reliability during high-intensity cycling (ICC: 0.38–0.76) [95]. Moreover, HRV monitored during the final 30 seconds of maximal exercise has demonstrated exceedingly low reliability [96]. When used to quantify post-exercise recovery, most HRV metrics exhibit moderate to good reliability across studies [92, 93, 96, 97]. In response to mental stress, some studies demonstrate greater relative reliability in both time and frequency domain HRV measures (ICC: 0.82–0.85) compared to rest (ICC: 0.70–0.73) [67], while others demonstrate the opposite [98]. In low activity states, such as sleep, most HRV metrics are stable and reliable with relative reliability exceeding an ICC > 0.80 [76, 79, 99]. However, reliability appears reduced during afternoon naps [80] and is poor when using LF and LF/HF frequency domain measures [76, 79, 80]. It should be noted that prior studies assessing HRV reliability during sleep are isolated to healthy populations. As such, nocturnal HRV reliability may be altered in populations with frequent nocturnal sleep and cardiovascular disruption such as obstructive sleep apnea [100].

Discrepancies in HRV reliability across behavioral states may be partly influenced by shorter recording duration [80, 96] and violation of signal stationarity (i.e., increased movement or sweating during exercise), which may bias the HRV assessment [101]. Under circumstances where signal stationarity is violated, as might be expected during exercise, for example, it may be preferable to utilize nonlinear HRV indices. Prior research has indicated moderate relative (ICC 0.50–0.60), but excellent absolute (CV 4.7–8.9%), reliability amongst measures of heart rate complexity, while other nonlinear HRV indices similarly exhibited excellent relative (ICC 0.74–0.82) and absolute (CV 14–21%) reliability at rest [70]. Additionally, reliability of nonlinear HRV metrics may differ based on the measure employed [102]. Nevertheless, some studies suggest that short term nonlinear HRV and heart rate complexity indices demonstrate acceptable relative and absolute reliability during low-intensity walking prior to (ICCs: 0.73–0.92; CV: 10–22%) and after (ICCs: 0.70–0.91; CV: 13–28%) maximal exercise [92], supporting the utility of nonlinear HRV indices for use during changes in behavioral states.

In sum, many studies support the relative reliability of HRV at rest and potentially during specific behavioral states (i.e., steady state exercise and sleep). The relative reliability of HRV supports its utilization as a potential risk stratification tool across individuals, as this indicates that one’s relative rank against a larger group of individuals is consistent over time. Conversely, absolute reliability is often lower when quantifying HRV. This inconsistency suggests caution should be considered when interpreting changes in HRV over time within an individual in response to a given intervention, perturbation, or treatment (i.e., longitudinal study designs). Similarly, careful consideration should be taken when interpreting HRV during different behavioral states (i.e., controlled breathing, postural changes, exercise, sleep, stress) due to ambiguity surrounding the reliability of HRV measures across time. In addition to constraints around the reliability of HRV measures, assumptions about the underlying physiological mechanisms of HRV may not hold true in specific behavioral and environmental situations. While resting HRV metrics, particularly in the time domain, represent primarily vagal activity (i.e., variation in vagal neural outflow) of heart rhythm, HRV responsiveness to behavioral states that may involve some level of cardiac sympathetic activation (i.e., psychological stress, exercise) undermines the ability to infer specific contributions of vagal withdrawal to the observed reductions in HRV. As such, caution is warranted when interpolating the relative influence of vagal activity or withdrawal during dynamic laboratory tasks. Adherence to the technical recommendations previously outlined in excellent methodological reviews [11, 28] as well as experimental standardization both within and between participants are critical. Additionally, uniform reporting of experimental and environmental factors which may impact HRV reliability such as recording duration, body posture, respiration, time between testing, etc. are necessary to ensure the accuracy and reliability of future HRV assessment in human cardiovascular research studies. Finally, future research should not only be attentive to experimental techniques that strengthen HRV reliability, but should also avoid redundancy through a priori selection of which HRV measures will be used based upon experimental design and goals [103].

Question 4: Are time-domain or frequency domain HRV measures appropriate indices of cardiac parasympathetic activity?

The parasympathetic nervous system is uniquely poised to modulate cardiac rhythmicity on a beat-to-beat basis due to high axonal conduction velocity, ACh-mediated influences on ion channel permeability and activation within cardiac pacemaker cells, and rapid ACh degradation leading to abrupt cessation of muscarinic receptor activation [13, 104]. However, perhaps the best evidence in support of a predominant influence of cardiac parasympathetic modulation on time and frequency domain HRV measures come from blockade studies performed in animals and humans. In early work, glycopyrrolate-induced cholinergic blockade completely abolished HF HRV, but only partially reduced HRV in the LF spectrum in conscious, anaesthetized dogs [10]. Only following both anticholinergic and β-blockade was the LF-HRV power spectrum abolished [10]. Subsequent work in humans demonstrated that atropine (a muscarinic receptor blocker) administration effectively abolished HF HRV (~0.25 Hz) in both supine and standing positions, and this effect was not impacted by further β-blockade using propranolol [105]. In contrast, atropine failed to completely eliminate low frequency HRV fluctuations in the supine position, and only when combined with propranolol administration was LF-HRV abolished in the standing position [105]. These findings are bolstered by subsequent studies demonstrating abolishment of HF-HRV alongside time-domain HRV measures (RMSSD and pNN50) following pharmacological blockade of parasympathetic activity [106, 107]. Conversely, β-blockade has been observed to reduce (albeit not completely abolish) LF-HRV in some studies [108]. Only following parasympathetic and sympathetic blockade is LF-HRV completely abolished [108]. Together these findings demonstrate that HF-HRV and time domain HRV measures are influenced by cardiac vagal activity, whereas LF-HRV is likely influenced by both parasympathetic and sympathetic innervation.

Further support of the predominant influence of cardiac vagal activity on HF-HRV comes from its close interrelation with respiratory patterns. There is consensus that the HF component of HRV reflects rapid changes in heart rate, (i.e. those mediated by the vagus nerve) and respHRV. Along with the reduction in intrathoracic pressure during inhalation, the inspiratory increase in heart rate, via respHRV, serves to increase venous return to the right side of the heart, delivering deoxygenated blood to the lungs. Simultaneously, oxygenated blood is delivered to the body from the left side of the heart. Experimental studies in animals indicate that cardiac vagal activity is highest during the post-inspiratory phase (immediately after the peak of inspiration) and lowest in late expiration, and that much of this respiratory modulation is governed by the Kolliker-Fuse nucleus in the pons [109] as well as by the preBötzinger nucleus in the medulla [110].

Microelectrode recordings from the human cervical vagus nerve have documented cardiac and respiratory modulation of multi-unit vagal activity [111, 112]. Single-unit recordings have identified myelinated vagal axons that behave as cardioinhibitory neurons associated with a marked respiratory modulation of firing and peak activity during the post-inspiratory phase [112, 113]. However, some respiratory synchronization of vagal activity may serve as a compensatory response to baroreflex activation. Respiratory fluctuations in HR and stroke volume are inversely related under normal conditions [114]. Following atropine blockade in humans, HRV within the respiratory frequency range is abolished, yet respiratory fluctuations in stroke volume were maintained and blood pressure variability elevated [114]. These findings infer that HRV may serve to buffer respiratory fluctuations in stroke volume (SV) to maintain stable cardiac output and blood pressure.

Many studies simply record ECG or PPG without recording respiration; we contend that all laboratory studies in which HRV is reported should also record respiration, which can easily be done with respiratory belt transducers wrapped around the chest and/or abdomen. Only then can the HF component of HRV be appropriately characterized as respiratory. Nevertheless, we can conclude that the HF component of HRV, and the time-based HRV metrics RMSSD and pNN50, do indeed reflect modulation of cardiac vagal drive.

Question 5: Are LF or LF/HF ratio appropriate indices of cardiac sympathetic or sympathovagal balance?

As noted above, autonomic outflow to the SA node of the heart mediates short-term fluctuations in heart rate, and frequency domain (or power spectral) analyses of HR signals are very commonly used to quantify cardiac autonomic regulation. Throughout the literature, there is widespread use of LF and HF to infer changes specific to a particular branch of the autonomic nervous system, or their ratio to infer changes in cardiovascular autonomic balance. While HF power as an index of PNS variability is considered robustly reliable and supported by pharmacological blockade studies (detailed above in Q3 and Q4), LF is not selective to sympathetic outflow.

Several studies have collectively provided considerable evidence that LF power is non-specific to cardiac autonomic activity. First, β-adrenergic receptor antagonists (e.g., beta-blockade) do not consistently affect LF power [115–118]. In fact, they generally cause increases in HF power. Second, SA node parasympathectomy reduces LF power in awake dogs [23], and LF power is further reduced by beta-adrenergic blockade. Third, LF power is not directly related to cardiac norepinephrine spillover [119]. Instead, LF power is influenced by neural reflexes, adrenergic receptor sensitivity and postsynaptic signal transduction, and electrochemical coupling demonstrated via various clinical examples of cardiac sympathetic denervation that included pure autonomic failure, cardiac transplantation, and dopamine beta-hydroxylase deficiency. Fourth, LF power is unrelated to extracardiac measurements of SNS activity. Saul et al. [24] demonstrated that at rest and during phenylephrine infusion, there was no relation between absolute and relative LF power with directly measured post-ganglionic efferent muscle SNA or plasma norepinephrine (Figure 3). However, during decreases in arterial pressure induced by nitroprusside, there were weak relations (r = 0.34–0.35) between relative LF power, muscle sympathetic outflow and plasma norepinephrine. The authors concluded that these findings likely reflect that LF fluctuations of heart rate are under both sympathetic and parasympathetic control, which has also been demonstrated experimentally [23]. These findings have been extended and substantiated by multiple groups since [21, 22, 120–124], with several of these studies providing additional evidence that changes in LF power induced by various experimental manipulations reflect modulation of cardiac autonomic outflow (e.g., via the baroreflexes) versus direct changes in cardiac sympathetic tone, as is often assumed [121, 122]. In sum, the literature indicates that LF power is jointly mediated by the SNS and PNS, related to factors in addition to cardiac SNA, and is therefore not purely reflective of, nor a valid indicator of, cardiac sympathetic regulation. Likewise, it is important to acknowledge that MSNA and plasma norepinephrine are not surrogates for cardiac SNS activity.

Figure 3.

Absolute (Panels A and C) and fractional (Panels B and D) low-frequency (LF) heart rate variability (HRV) compared with muscle sympathetic nerve activity and plasma norepinephrine at each infusion dose of nitroprusside and phenylephrine. During increases in sympathetic activity induced by nitroprusside, fractional LF (but not absolute LF) was modestly correlated to MSNA and plasma norepinephrine (r=~0.34–0.35, P<0.05). In contrast, decreases in sympathetic activity induced by phenylephrine were not significantly correlated to either absolute or fractional LF HRV. Reproduced with permission from Saul et al. [24].

The above studies also challenge the use of the LF:HF ratio as reflecting changes in “sympathovagal balance,” where an increase in the LF:HF ratio is suggested to reflect increasing ‘sympathetic dominance’ and a decrease reflective of an increase in ‘parasympathetic dominance’. This interpretation is dependent upon cardiac sympathetic activity being the primary determinant of the LF peak. As noted above, changes in LF power are not purely reflective of changes in cardiac sympathetic tone, as changes in parasympathetic outflow are also reflected in LF power [23]. Moreover, the interpretation of LF:HF in this manner inherently assumes that changes in cardiac sympathetic and parasympathetic drive are purely reciprocal, such that a decrease in one corresponds to an equivalent increase in the other. However, there are many examples where this assumed reciprocal relation is not valid, even in response to provocation. For example, following acute exercise there is rapid parasympathetic reactivation even while sympathetic tone remains elevated [125]. Fundamental work across multiple studies by Eckberg and colleagues [12, 126–128] described parallel changes in sympathetic and parasympathetic activity to chemoreceptor stimulation and cold water immersion of the face (e.g., the diving reflex), further demonstrating that the interaction of the sympathetic and parasympathetic nervous systems are complex and non-linear, and that simple reciprocal changes do not always occur. Finally, and as discussed by Eckberg in his critical review on sympathovagal balance [12], there are purely mathematical issues arising from the calculation of LF:HF using normalized power such that a change in LF must be reflected by a change in HF, and thus LF:HF increases exponentially as a function of normalized LF power. As such, the assumption that LF:HF reflects sympathovagal balance is inappropriate theoretically, in addition to the fact that LF does not purely reflect cardiac sympathetic activity. In conclusion, the use of LF:HF ratio as a quantitative probe for the balance between sympathetic and vagal cardiac outflow is not valid and should not be interpreted as such in human cardiovascular research.

Question 6: Is HRV appropriate for risk stratification?

Decades of research suggest that HRV can be used in stratifying risk for various chronic diseases. Overall, HRV for risk stratification has greatly improved our understanding of the pathophysiology of a variety of diseases, particularly those related to ventricular arrhythmias and cardiac conditions, but also with other disease states. For example, one of the first clinical studies supporting HRV as a risk stratification tool was in 1963 and demonstrated its association with fetal hypoxia, where acute losses of beat-to-beat variability represented fetal distress [129]. The use of HRV has since expanded to assess mortality risk in critically ill patients in the intensive care unit [130], progression of cardiovascular and neurovascular conditions including myocardial infarction [131], heart failure [132], stroke [133] and acute pancreatitis [134]. As noted throughout this review, various HRV metrics are used as proxy measures to assess autonomic function and these metrics may be differentially interpreted. With the noted lack of reliability and validity related to the use of LF power, VLF power and LH/HF measures, this section will focus on studies that used metrics with strong intrapartcipant reliability and validity (i.e., time domain metrics as well as HF power) for risk stratification.

There is extensive literature to date examining associations between HRV and cardiovascular disease (CVD) [135]. Hypertension, the most prominent CVD risk factor [136], has been associated with lower HRV in several studies. For example, Huikuri et al. [137] demonstrated that SDRR and SDNN was lower in adults with hypertension compared with individuals with normal blood pressure. This finding was corroborated in other studies demonstrating lower SDNN [138] and HF power [139] in adults with hypertension. In a cohort of over 19,000 adults, RMSSD was inversely associated with cardiovascular risk scores [140]. Given that 80% of the sample were males, however, the generalizability of these findings to females remains limited [140].

Along with cross-sectional findings implicating an association between low HRV and CVD, numerous studies have similarly reported a prospective association between resting HRV and incident cardiovascular risk. The Atherosclerosis Risk in Communities (ARIC) study cohort has provided multiple findings which characterize the utility of resting HRV as a prognostic variable associated with the development of hypertension, sudden cardiac death, coronary heart disease, and all-cause mortality [138, 141–143]. Results from the 3-year ARIC study follow-up demonstrate that the adjusted odds ratio (OR) for the development of hypertension in previously normotensive midlife adults increases with reductions in HF-HRV, with those in the lowest quartile exhibiting a 2.44 OR compared to the highest tertile [138]. Similar results were observed in the 9-year follow-up, although the associations were more modest compared with the 3-year follow-up. Specifically, those in the lowest HRV quartile demonstrating a hazard ratio (HR) of 1.24–1.44 relative to those in the highest HRV quartile [143]. Although direct comparison of the 3-year and 9-year findings warrants caution given the use of different statistical approaches (OR vs. HR, respectively), the directionally consistent association between low HRV and incident hypertension across both follow-up periods strengthens confidence in the relation. Notably, baseline HRV was also lower in individuals with hypertension, the authors report that the rate of change in HRV across the 9-year follow-up did not differ between groups [143], supporting the utility of HRV as a risk stratification assessment for cardiovascular risk. Building upon this early work from the ARIC cohort in a large, diverse cohort encompassing 192,110 person-years of follow-up, higher HRV during midlife was associated with a modestly (4–8%) lower lifetime CVD incidence in both males and females, with a stronger association observed in females [135]. In addition, in a dose-response meta-regression that included nearly 22,000 individuals without known CVD, every 1% increase in SDNN was associated with a 1% lower risk of fatal or non-fatal CVD [144]. This meta-analysis [144] also reported that lower time-domain HRV was associated with higher relative risk of CVD, as depicted in Figure 4. Collectively, these studies demonstrate that prior to advanced CVD, time domain HRV metrics and HF power may be altered as it relates to hypertension risk, and may have some associative utility.

Figure 4.

Meta-analysis comparing cardiovascular disease risk in low verses high heart rate variability as assessed in the time domain (standard deviation of N-N intervals). Reproduced with permission from Hillebrand et al. [144].

Regarding clinical outcome associations and risk stratification, HRV has demonstrated utility, particularly when assessed following cardiac arrest and mortality after a myocardial infarction. For example, 24-hour SDANN was over 2-foldlower in a small sample of patients with coronary artery disease who died from sudden cardiac arrest due to ventricular fibrillation during ambulatory monitoring compared with apparently healthy control participants [145]. Similarly, patients with sustained ventricular tachycardia exhibit lower SDNN compared to patients without repetitive ventricular arrhythmias. One study utilizing 24-hr recordings demonstrated that SDNN was lowest in the early morning after awakening [146], corresponding to the time period of which sudden cardiac death incidence is greatest [147]. Further, in the 2–4 weeks after a myocardial infarction, HRV time domain metrics are strongly associated with subsequent mortality [148–151]. Collectively, time domain HRV metrics are useful indicators of cardiac arrest and as prognostic indicators of mortality risk following major cardiac events, such as a myocardial infarction.

In patients with heart failure, several studies have demonstrated that reduced HRV is associated with more advanced symptoms, poorer outcomes, and increased mortality. The majority of studies that use HRV to assess poor prognosis or death in patients with heart failure found SDNN [152–154] or SDANN [152] to be the strongest correlates. Further, a clinical study with 105 patients with heart failure (17 females) reported that lower SDNN and SDANN correlated with a lower ejection fraction and that SDNN, SDANN and RMSSD were inversely correlated with functional heart failure classification [132], suggesting that patients with more severe heart failure symptoms have lower HRV. In an ischemic heart disease population, SDRR was associated with greater risk of ischemic events (sudden cardiac death and ventricular tachycardia), and was superior to left ventricular ejection fraction as a risk measure [155, 156]. In contrast, HRV metrics were not associated with event risk in dilated cardiomyopathy, but 24-hr blood pressure variability demonstrated enhanced risk stratification for sudden cardiac death [157]. Further, HRV may also be helpful for assessing prognosis after treatment in patients with heart failure. Landolina et al. [158] studied the effect of resynchronization therapy on HRV in patients with advanced heart failure, and found that SDANN ≤65ms at the first week and ≤76ms at the fourth week were associated with cardiac transplant or mortality. Thus, the collective literature indicates that HRV time domain metrics are potentially useful in identifying which patients with heart failure are more at-risk for additional morbidity and mortality, and thus who may benefit from the most intensive treatment and monitoring.

In summary, HRV is widely used for risk stratification related to poor secondary outcomes and mortality in cardiac conditions. Although only briefly touched on in this section, reduced HRV is also associated with cardiovascular risk factors that include hypertension [138, 143], diabetes [159] and stroke [133] when compared with apparently healthy individuals. The positive predictive value of HRV is modest, ranging from 14 to 40% [160], while the negative predictive value is higher, ranging from 77 to 98% [160]. The combination of HRV with other risk stratification variables, including left ventricular ejection fraction, non-sustained ventricular tachycardia, premature ventricular beats >10/h, late potentials, baroreflex sensitivity, ECG changes, etc., may increase the overall prognostic accuracy, at least for patients with cardiac conditions [156, 160–163]. It remains uncertain why HRV may be lower in individuals at risk for incident CVD. Low HRV may represent insufficient cardiac vagal modulation of heart period, thus inappropriate autonomic regulation of cardiovascular function. However, the causes of low HRV in individuals at risk for CVD are difficult to disentangle due to the compounding influence of numerous lifestyle and environmental factors (i.e., low physical activity, alcohol use, poor sleep, excessive stress) that are similarly associated with both reduced HRV and elevated CVD risk. Despite significant research related to HRV as a risk stratification tool, there is a current lack of consensus regarding the most clinically meaningful HRV metric in clinical populations. Evidence suggests that SDNN is a robust risk measure; however, there is no consensus about this HRV parameter in clinical use [164]. Further, while several studies demonstrate a reduction in many or most HRV metrics in the context of various conditions, the utility of HRV for risk stratification is complicated by substantial differences in recording methods and durations, as well as a current lack of population-specific normative data that would be necessary to inform risk stratification. Thus, while HRV holds promise as a risk stratification tool, and possibly as a useful prognostic metric, it will take significant work to move toward implementation within clinical practice.

Question 7: Do age, sex, and race/ethnicity impact HRV?

Lower HRV is associated with an increased risk of CVD, morbidity and mortality in older adults. With normal aging, HRV decreases across all metrics (time-domain and frequency-domain measures) [165, 166], reflecting reduced cardiac vagal activity and support of blood pressure [167–169]. For example, studies demonstrate a reduced parasympathetic support of blood pressure in older males and females following autonomic blockade [167–169], suggesting a physiological alteration in cardiac vagal activity with age. This effect of aging on HRV accelerates in young adulthood, but decreases after middle age [166, 170]. the rate of age-associated reductions in HRV occurs independently of health and lifestyle factors such as physical activity [171, 172] or the presence of cardiometabolic disease [173]. It is worth noting that both low physical activity levels and cardiometabolic disease are also both associated with lower absolute levels of HRV [171–173]. Therefore, sedentary individuals, or individuals with poorer cardiometabolic health, may experience greater cardiovascular risk at an earlier age than physically active, healthy individuals. HRV studies considering biological sex and age will be helpful in improving HRV reproducibility and replicability, and caution is needed to interpret changes in HRV longitudinal studies, especially when studies involve older participants.

A significant body of research indicates that sex systematically influences HRV. Multiple studies [170, 174–178] and meta-analyses [179, 180] have confirmed fundamental differences in cardiac autonomic control between healthy, young, biological males and females. Specifically, females at rest tend to present with higher HR and shorter RRI compared to males [180]. Despite having a faster heart rate, females exhibit greater HF power than males [170, 175, 180], suggestive of greater cardiac parasympathetic modulation. In contrast to the higher HF power, other studies observe lower overall HRV in females compared to males, as captured by SDNN [174, 180]. In addition to these sex differences in basal heart rate and HRV, several studies have demonstrated sex differences in autonomic responses to challenges such as changes in posture, the Valsalva maneuver and slow deep breathing [181]. In response to laboratory-based social stress, females present with a lower HRV [179] and a blunted reduction in HF [177] compared to males. These sex differences in HRV are more pronounced in younger and middle-aged adults, and tend to diminish with advancing age [170].

Differences in HRV have also been reported across different racial and ethnic groups, but the clinical meaning is still under investigation. In the US, there is a large body of literature documenting cardiovascular health disparities across different racial and ethnic groups, particularly in Black Americans [182, 183]. Social determinants of health (i.e., the conditions in which people are born and raised, live, work, and age) account for a large proportion of these disparities [184, 185]. It is important to note that although increased HRV is generally thought to be cardioprotective, Black adults exhibit a pattern labelled “cardiovascular conundrum” [186–188], where greater HRV is observed despite higher incidence of CVD [189]. A 2015 meta-analysis reported that Black Americans, in general, exhibit higher resting HRV than White Americans, even after controlling for age and health status [188]. A study in a large Brazilian cohort also reported HRV is greatest in Black, followed by lighter skinned (i.e., brown), relative to White participants, but only reported frequency domain measures [189]. In US samples, several studies have reported higher HRV in Black males compared with White males [190–192]. In contrast to the studies in males, findings in females suggest that HRV is not different among Black and White females [192, 193]. Therefore, it appears racial differences in HRV may be more pronounced in males than females. Given some inconsistencies, and relatively few studies in other underrepresented adults (i.e., Hispanic, Asian, etc.) [194–197], there remains a clear need for more controlled HRV data across racial and ethnic groups. More diverse cohorts will establish reliable normative data, provide insight to potential differences in physiology amongst different racial and ethnic backgrounds and be representative of the general population. As such, when conducting HRV studies, biological sex and race/ethnicity should be documented and reported [198, 199]. Further, social determinants of health impact cardiovascular risk, and thus should be assessed and incorporated into the study analysis understand impact on HRV [200].

Question 8: Does fitness level and/or basal HR impact HRV?

Physical activity and cardiorespiratory fitness (CRF) are well established correlates of cardiovascular and all-cause mortality [201–205]. The American Heart Association even recommends CRF as a clinical vital sign [206], but whether CRF impacts HRV may be less clear. Physical activity is voluntary movement produced by skeletal muscles that results in energy expenditure [207, 208], where exercise is planned, structured, and repetitive movement with the objective of improving or maintaining fitness [207, 208]. In contrast, CRF is a complex trait, influenced by a combination of heredity, exercise training, and environmental factors [209–212]. Additionally, CRF partially reflects how effectively the heart, lungs, and blood vessels deliver oxygen to the muscle and the subsequent capacity to utilize this oxygen during maximal exercise [207, 208]. For the purposes of this discussion, CRF is typically assessed using graded exercise testing, with indirect calorimetry, to determine peak oxygen consumption (e.g., VO₂peak or VO_2Max) and operationalized fitness [213–216].

With regular endurance exercise training, the cardiovascular system undergoes structural and functional (e.g., increased plasma volume) adaptations that enhance CRF [217–219]. Notably, the left ventricular remodeling and enlargement lead to greater stroke volume which contributes to greater maximal cardiac output [217]. For example, prospective studies have demonstrated that exercise training leads to increased left ventricular mass [220–223]. Cross sectional studies in elite athletes demonstrate remarkable cardiac size and structural characteristics [224–228], which are thought to reflect a combination of inherited traits and cumulative exposure to exceptional training volume and intensity. It is also important to note the nuance that not all forms of exercise (e.g. endurance compared with resistance exercise) produce equivalent cardiac remodeling [218, 220, 229]. Nonetheless, these adaptations likely underly the lower resting HR observed with higher CRF [230]. While higher CRF being associated with lower HR and greater HRV is also thought to reflect parasympathetic activity [231–233], studies using autonomic blockade demonstrate that trained individuals exhibit a lower intrinsic HR [234, 235]. These findings indicate that training-related bradycardia is not solely autonomically mediated but also reflects intrinsic cardiac adaptations.

Multiple studies have reported that CRF is associated with lower resting HR [215, 230, 236, 237]. The lower HR lengthens R–R intervals and contributes to higher resting HRV, such as greater RMSSD [238–240] and HF power [240–242], although this is not always the case [215, 231, 237, 243]. Examples of studies indicating higher CRF is associated with time-domain HRV indices include cross-sectional work in children [238] and adults [239], indicating resting HR and HRV indices (e.g., pNN50, RMSSD, and SDNN) were correlated with CRF. In the sample of children, only mean HR remained independently associated with $\dot{\mathrm{V}}{\mathrm{O}}_{2\text{peak}}$ and was the strongest predictor after statistical adjustment [238]. However, in the adult sample, SDNN remained associated with $\dot{\mathrm{V}}{\mathrm{O}}_{2\text{peak}}$ after statistical adjustment for clinical measurements such as HbA1c, which was relevant as the sample included apparently healthy adults and individuals with type 2 diabetes [239]. In another sample of adults with type 2 diabetes, $\dot{\mathrm{V}}{\mathrm{O}}_{2\text{peak}}$ correlated positively with RMSSD, HF, and pNN50%[240]. However, the analyses did not adjust for resting HR, age, sex, or comorbidities, so it is unclear whether HRV–CRF associations were independent of HR [240]. Regarding studies that focused more on spectral analyses, a cross-sectional comparison of 15 young adult athletes and 15 nonathletes reported that athletes exhibited higher HF power compared with nonathletes [241]. Given the study design, the groups likely differed in factors beyond VO_2max (e.g., exercise volume, genetics, sleep, etc.), however, these variables were not considered in this study. To this point, in another cross-sectional study that split young adult participants into tertiles of physical activity, higher objectively measured physical activity, especially more bouts of vigorous physical activity, were associated with higher HF power, as well as RMSSD and SDNN [242]. In adjusted regression models, only vigorous physical activity and RMSSD remained associated.

Together, these cross-sectional findings suggest that individuals with higher CRF generally exhibit lower resting HR and more favorable time- and frequency-domain HRV profiles. Importantly, prospective endurance-training studies extend this evidence by demonstrating that structured aerobic exercise increases HRV in previously sedentary individuals. In healthy adults, 3 months of moderate-volume endurance training lowered resting HR and increased multiple time- and frequency-domain HRV indices, with no further enhancement after 3 months (i.e., 6-, 9-, and 12 months) [244]. Similarly, in sedentary older adults, HRV improved progressively across 12 months of endurance training, with some measures reaching values similar to those observed in much younger individuals [245]. These prospective exercise studies support the idea that increases in CRF are accompanied by favorable adaptations in autonomic cardiac modulation.

Challenges in standardizing and interpreting HRV data make HR, alone, an attractive, accessible indicator of autonomic cardiac control. Indeed, there are data indicating that when all HRV indices are considered together, resting HR emerges as having the strongest association with CRF (likely reflecting contributions from intrinsic HR) [215, 238] along with physical activity levels [246]. Consistent with these findings, resting HR is also associated with cardiovascular and all-cause mortality [247–250]. The stronger association between CRF and HR compared with CRF and HRV, is likely influenced by numerous factors. There is evidence that higher resting HR with aging can be partially attenuated with exercise training, and to a greater extent than HRV [246]. Thus, careful consideration of these influences is essential when interpreting associations of CRF with HR and HRV from literature.

Question 9: Do breathing rates and tidal volume need to be fixed for HRV?

As detailed in Question 2, there is indisputable evidence that breathing rates and tidal volume influence HRV. This effect has been clear since the foundational work establishing respHRV [1]. Still, controversy remains as to whether investigators should experimentally fix or control respiratory rates and/or tidal volumes when assessing HRV in the laboratory. At rest, most healthy adults have breathing frequencies that range from 10–15 breaths per minute. The 0.15–0.4 Hz frequency range also corresponds to the HF domain of HRV, which is often used as an index of respHRV. It is well established that slower and deeper respiration impacts respHRV, with maximum effects occurring at ~0.1 Hz (or six breaths per minute). Such breathing frequency elicits the highest amplitude of respHRV and is also associated with the most efficient gas exchange [28, 48, 251, 252]. The relation between breathing frequency and respHRV is a driving factor in HRV biofeedback research, where slow breathing techniques are employed to activate the parasympathetic nervous system and reduce stress [251, 253].

While controlled, slow breathing may be applicable to some experimental designs aimed at HRV biofeedback and/or enhanced relaxation [48, 251], such slow breathing frequencies are not generalizable to most autonomic states at rest. Instead, most laboratory autonomic function studies aim for participants to naturally breathe in frequencies that align within the HF domain of HRV (0.15–0.4 Hz). While some studies suggest stronger HRV reliability with paced breathing [68, 254], others report no difference in reliability [77, 83, 86, 255]. Additionally, while slow breathing is often reported to mitigate stress, others report tangible effects of paced breathing on mental activity, stress and emotional valence, and altered HRV [256–259], suggesting that paced breathing has the potential to induce some level of mental stress due to a forced and/or abnormal breathing frequency.

The key to a robust study design, particularly for longitudinal or repeatable studies, is to ensure similar breathing frequencies across sessions – and to ensure the study sample is within the broader range of 0.15–0.4 Hz. Thus, respiration should be measured via chest/abdominal transducer or nasal airflow and reported (Table 2). We acknowledge that only spirometry can measure tidal volume (when airflow has been integrated and calibrated by a known volume), which can influence HRV [260–262], but the experimental modifications necessary to accurately record tidal volume are not realistic in most HRV studies, and similar to paced breathing could potentially cause altered mental stress that does not represent a ‘normal’ baseline. Finally, there may be circumstances where breathing frequencies should be considered as covariates in some studies and analyses, including when there are large ranges of breathing frequencies within a sample.

Question 10: Do most of today’s wearables provide rigorous and reliable HRV metrics?

So-called “smart” wearable devices can contribute to promoting positive health behaviors. For example, these devices prompt reminders, estimating steps taken or calories burned and information about sleep patterns and quantity. Overall, quantitative data regarding the accuracy and effectiveness in these areas are mixed [263–265]. Many popular smart wearable devices also measure HRV, and market HRV as an indicator of an individual’s health status or recovery status after acute exercise [32], or for planning exercise training [266]. However, the evidence base for these recommendations is lacking.

The scientific basis for the use of HR analyses in wearable devices comes from studies performed in clinical or laboratory settings, using equipment such as an ECG, a chest strap that measures HR, and/or a Holter monitor. With the development of smart devices, such as the Apple Watch, WHOOP, Fitbit, the Oura ring, and similar products, consumers can assess their HRV daily, without requiring a clinical or lab setting. These devices commonly track HR with ECG or PPG [33]. Some devices, such as the Apple Watch [267, 268] and Samsung Galaxy Watch [268] can utilize a user-initiated 30-second single-lead ECG for cardiac rhythm detection. Specifically, wearers can obtain a single-lead ECG by placing a finger on the watch’s crown or electrode, which completes an electrical circuit across the upper body and allows the device to record a ~30-second tracing. The extent to which the “user-friendly” modifications of the original laboratory measurements may degrade accuracy is unclear and varies based on device and the outcome measure [269].

The reliability of wearable-device derived HR and HRV has been compared to simultaneous ECG assessment. Most studies that support the comparability of PPG-derived HRV metrics to gold-standard ECG assessment have been performed under resting or sleeping conditions [269–273]. The combined findings from these research studies reach a consensus that wearable devices offer a reliable estimate of nocturnal HR [269–273]. However, the findings regarding the reliability of other PPG-derived variables are more ambiguous. For instance, relative to nocturnal ECG, the Oura Ring has demonstrated acceptable HRV comparability when assessing time-domain metrics such as the RMSSD across the entire sleeping period [272, 273]. Similar findings were reported when using the WHOOP wristband, which demonstrated low bias in HR during slow wave sleep compared to ECG (bias: <0.5%), and acceptable RMSSD (bias: 8–13%) particularly following natural log transform (bias: 2–3%) [271]. However, the correlation and agreement between PPG and ECG-derived variables is substantially reduced when assessing frequency-domain HRV measures and SDNN, and when data are assessed across a shorter (5 minute) time frame during the nocturnal period [272]. Similarly, the reliability of PPG-derived HRV under free-living conditions and behaviors is even less reliable [274]. Together these findings suggest that HRV derived from PPG devices is most reliable when assessed under controlled, resting laboratory conditions. However, in studies where this is not feasible, such as large-scale, at-home studies where wearables are often monitored, restriction of HRV analyses to low activity periods may confer the most reliable findings.

As has been described in the current review, many factors such as age, sex, fitness, race/ethnicity, psychosocial factors, and medications can affect HRV. Thus, tracking a person’s HRV over time and in response to various stressors or environmental factors is thought to provide additional utility compared to an individual “spot measurement” to a set of normal values, which have not been developed in a comprehensive, consistent way. However, this notion remains quite speculative as there is a lack of data that device based HRV measures outperform self-report measures (e.g., subjective feelings of readiness) or submaximal exercising HR in the context of performance [275, 276].

Despite having utility as a health promoting tool, HRV-based metrics from smart devices are quite limited in their ability to represent activity of the autonomic nervous system [12]. Notably, most wearables only report time domain measures such as RMSSD or SDNN or a normalized score based on these measures. Thus, while wearables can be valuable lifestyle tools, they are neither valid measures of cardiac/vagal physiology nor direct proxies of sympathetic outflow [277, 278]. Taken together, these limitations reflect a modern version of the McNamara fallacy or Goodhart’s law: focusing on what is easily measured by wearables rather than what is meaningful. That HRV can be quantified continuously does not guarantee that it provides actionable insight into autonomic function, performance, or readiness. Without evidence that device-based HRV metrics outperform simple self-report or basic physiological indices, relying on them as proxies of recovery or training status risks mistaking measurement for understanding.

Conclusion and Summary

HRV represents a commonly employed tool to investigate physiological regulation in humans across numerous disciplines including psychology and physiology. Since its conception and adoption in cardiovascular research, widespread use has led to discrepancies in experimental methodology, and debates related to the use of HRV as a robust and reliable physiological assessment of autonomic nervous system regulation. While the utility of HRV as it pertains to human health and cardiovascular physiology is appreciated, the present guideline is intended to help facilitate a more consistent use and interpretation of HRV in cardiovascular research and applications moving forward. A list of questions alongside a summary of answers discussed throughout the manuscript can be found in Table 3.

TABLE 3.

Summary and recommendations for enhancing heart rate variability (HRV) rigor and reproducibility within human cardiovascular research.

Question	Summary	Recommendation
What does HRV purport to measure, and what does it actually measure?	HRV technically reflects variability in heart period (R-R intervals), not variability of heart rate. Many wearables estimate HRV using photoplethysmography (PPG), which is not as precise as ECG-based HRV.	Although the term HRV is firmly established in the field, consider the term “heart period variability” within methodological description. Use ECG over photoplethysmography in controlled, laboratory settings. Be cautious when interpreting wearable-derived metrics, and always specify the input signal.
What are the key methodological considerations for enhancing HRV reliability?	ECG provides high-fidelity HRV data, especially with appropriate electrode placement and minimum sampling rates of 250–500 Hz. PPG offers convenience but has limitations. Environmental and procedural consistency, including respiration monitoring, is crucial for reliable HRV data.	For ECG-based HRV, use modified Lead II ECG placement to reduce artifact, and sample at ≥250 Hz. For PPG-based HRV, ensure proper sensor placement and appropriate finger warmth when performed under controlled conditions. Ensure consistent lab conditions and monitor respiration frequency. Consider statistical adjustments if respiratory rates vary widely.
Is HRV reliable/reproducible in traditional test-retest?	Longer recording durations improve both absolute and relative reliability of HRV metrics. Time domain metrics are generally more reliable than frequency domain metrics. Reliability varies across behavioral states and is influenced by signal quality and chronic conditions. Nonlinear HRV indices may offer better reliability in dynamic states.	Use ≥5-min recordings for improved reliability. Prioritize time domain metrics within longitudinal studies. Consider nonlinear HRV indices during exercise or when stationarity is violated. Account for behavioral and environmental influences. Be cautious when interpreting HRV in clinical populations.
Are time domain and frequency domain HRV measures appropriate indices of cardiac parasympathetic activity?	HRV reflects autonomic modulation of the heart, with time-domain and HF spectral metrics primarily reflective of modulation of cardiac vagal outflow. Cardiac sympathetic activity is not well captured by traditional HRV metrics. Respiration plays a key role in vagal modulation and should be recorded alongside HRV.	Recognize that RMSSD, pNN50, and HF reflect modulation of vagal tone. Record respiration in all HRV studies to properly interpret HF components. Adopt the term “respiratory HRV” (respHRV), rather than “respiratory sinus arrythmia” (RSA), to emphasize physiological relevance. Avoid interpreting HRV metrics as indicators of sympathetic drive.
Are LF or LF/HF ratio appropriate indices of cardiac sympathetic or sympathovagal balance?	LF power is not a specific marker of cardiac sympathetic activity and is influenced by both sympathetic and parasympathetic outflow, baroreflexes, and other factors. The LF:HF ratio does not validly represent sympathovagal balance.	Avoid interpreting LF power as a direct measure of cardiac sympathetic tone. Do not use LF:HF ratio to infer sympathovagal balance. Acknowledge the complex, non-reciprocal interactions between autonomic branches and the mathematical limitations of normalized spectral ratios.
Is HRV appropriate for risk stratification?	Time domain HRV metrics and HF power are useful for stratifying risk and outcomes in several cardiovascular conditions, including hypertension, myocardial infarction, and heart failure. However, predictive accuracy is modest and limited by methodological inconsistencies and lack of normative data.	Prioritize the use of SDNN, RMSSD, and HF power for risk stratification in cardiac populations. Combine HRV with other clinical biomarkers to improve predictive accuracy. Standardize recording methods and develop population-specific normative data to enhance clinical utility of HRV. Avoid reliance on LF and LF:HF metrics for risk prediction.
Do age, sex, and race/ethnicity impact HRV?	HRV differs by sex, age, and race. Females tend to have higher HF power, but lower SDNN compared to males. HRV declines with age, independent of lifestyle, though physical activity improves absolute HRV. Racial differences exist, particularly in males, but clinical implications remain unclear.	Account for sex, age, and race in HRV interpretation. Use caution when comparing HRV across diverse populations. Promote the inclusion of underrepresented groups to help establish normative data across race, ethnicity, and other factors. Consider hormonal and psychosocial factors when analyzing sex and gender-related HRV differences.
Does fitness level and/or basal HR impact HRV?	Higher cardiorespiratory fitness (CRF) is generally associated with lower resting HR and higher HRV, particularly RMSSD and HF power. However, associations vary across studies and may be confounded by age, sex, and health status. HR may be a more consistent correlate of CRF than HRV.	Consider both HR and HRV when evaluating autonomic function in relation to fitness. Use RMSSD and HF power as primary HRV metrics when assessing CRF. Adjust for confounding variables in analyses. Recognize that vigorous physical activity may have stronger associations with HRV than moderate activity.
Do breathing rates and tidal volume need to be fixed for HRV?	Breathing rate and tidal volume significantly influence HRV, particularly HF power and RSA. While slow breathing enhances RSA, it may not reflect typical resting autonomic states and can introduce mental stress in some instances.	Monitor and document breathing frequency. Acknowledge that forced, paced breathing may elicit stress in some individuals, and unintentionally impact HRV. Ensure breathing rates fall within the HF domain (0.15–0.4 Hz) for consistency. Consider respiratory rate as a covariate in analyses when variability is high across participants.
Do most of today’s wearables provide rigorous and reliable HRV metrics?	Wearables offer accessible HRV tracking and may help promote positive health behaviors, but wearable HRV metrics are limited in physiological validity. Most devices report time-domain metrics and do not accurately reflect cardiac autonomic nervous system activity and/or balance.	Use wearables for lifestyle tracking, not for precise autonomic assessment. Interpret HRV data from wearables cautiously. Prioritize longitudinal tracking over spot comparisons, but recognize large day-to-day variability. Avoid using wearable-derived HRV to infer sympathetic activity or “sympathovagal balance.”

CRF, cardiorespiratory fitness; ECG, electrocardiography; HF, high frequency; HRV, heart rate variability; LF, low frequency; pNN50, percentage of RRIs that differ by greater than 50ms; PPG, photoplethysmography; respHRV, respiratory HRV; RMSSD, root square mean of successive R-R interval differences; RSA, respiratory sinus arrythmia; SDNN, standard deviation of the time between NN intervals

The purpose of the present paper is to inform readers and researchers about the limitations in methodologies, interpretability, and physiological relevance of specific HRV metrics as autonomic assessment tools within cardiovascular research. We organized this through discussion of 10 key questions informed by inconsistencies and/or misconceptions. Researchers and clinicians hoping to utilize HRV measures within their studies should be aware of and recognize the numerous factors that influence HRV to aid in experimental design. Influences on HRV range from non-modifiable characteristics and processes (i.e., age, sex, circadian rhythms), disease states (i.e., hypertension, diabetes, congestive heart failure, etc.), environmental factors (i.e., pollution, temperature), behavioral factors (i.e., stress, physical activity, alcohol, caffeine, and drug use), and methodological factors (i.e., respiratory patterns, body posture, recording duration, data collection modality). The feasibility of controlling every factor with known influences on HRV is challenging and heavily contingent upon the experimental design being used and research question being asked Even under highly controlled, laboratory settings, it is difficult to fully remove external influences such as ambulation prior to laboratory arrival, daily living psychological stress, seasonal variations, etc. However, when practical, we recommend that researchers adhere to commonly reported experimental constraints in an autonomic laboratory environment [278] including 1) no caffeine, alcohol intake, or smoking for a minimum of 12 hours prior to testing; 2) no intense physical activity the day of testing and at least 12 hours prior to testing; 3) no active speech and movement during the recording period to maintain of stable breathing and posture; 4) and no meal intake for at least 3 hours prior to arrival to the laboratory environment (Table 2). Most importantly, it is critical that researchers using HRV measures in a laboratory setting ensure consistent study parameters within and between participants and explicitly describe in their methods section what standards are in place, while contextualizing potential limitations in physiological relevance.

It is important to recognize that controlling factors that likely impact HRV in the home setting is a major challenge and not feasible under most ambulatory study circumstances. Recording HRV only during low activity or sleep periods during the active recording period might improve wearable reliability [270–273]. The intent of the present guidelines is not to be overly prescriptive, but instead to ensure that HRV collected from wearable devices will be viewed through a critical, objective lens in light of the numerous limitations underlying its physiological relevance and implications. As such, we recommend that researchers are forthcoming in recognizing the limitations of using HRV, while paying particular attention to the variables they choose to report and their subsequent interpretations. It is preferable to collect and present HRV alongside other measures of autonomic and cardiovascular regulation such as high-fidelity microneurographic recording of muscle sympathetic nerve activity, or more commonly accessible measures such as beat-to-beat finger plethysmography, to contextualize HRV findings rather than treating them as a standalone outcome measure. Indeed, certain journals such as Clinical Autonomic Research, specifically state that papers using HRV as the only or primary outcome assessment have very low priority for the journal. While the current group of authors cannot speak for the editorial board, we collectively believe a similar prioritization should be considered for the American Journal of Physiology – Heart and Circulatory Physiology.

In summary, while HRV may offer utility as a risk stratification tool, its application as a specific marker of sympathetic or sympathetic-vagal balance is not appropriate. While there is stronger evidence that time-domain and HF assessments can more accurately reflect modulation of cardiac vagal activity, authors and investigators are cautioned against overinterpretation of HRV and its implications as a specific physiological measure of the autonomic nervous system. As numerous experimental, demographic, and environmental factors have been observed to significantly influence HRV measures and their reliability, careful consideration is necessary when designing experiments that utilize HRV as an outcome variable. These considerations should include accounting for non-modifiable (age, sex, race/ethnicity, etc.) and modifiable (i.e., exercise, fitness levels, etc.) factors within the participant population, as well as technical approach considerations such as the HRV input signal (i.e., ECG, PPG, etc.), data acquisition, laboratory/field conditions, and analytical approaches for HRV quantification (i.e., time vs. frequency domain). Finally, with respect to the rapid advancements of continuous HRV assessment using wearable technology, investigators should interpret the findings within the context of the limitations outlined in this review, with particular caution of any interpretations regarding autonomic activity and/or balance.

Key Abbreviations:

CVD

Cardiovascular disease

CRF:

Cardiorespiratory fitness

ECG:

Electrocardiogram

FFT:

Fast Fourier Transform

HR

Heart rate

HRV

Heart rate variability

ICC

Intraclass correlation coefficient

pNN50

% of successive normal-to-normal intervals that differ by > 50 ms

PNS

Parasympathetic nervous system

PPG

Photoplethysmography

RespHRV

Respiratory HRV, a preferred term for RSA

RRI

R-R interval

RMSSD

Root mean square of successive R-R interval differences

RSA

Respiratory Sinus Arrhythmia

SA node

Sinoatrial node

SDNN

Standard deviation of normal-to-normal intervals

SNS

Sympathetic nervous system

Data Availability

Not Applicable