Pulse Rate Variability Is Not the Same as Heart Rate Variability: Implications for Sports Performance and Injury Prevention

Abstract

Background

A key metric utilized by coaches and athletes to track athlete performance is heart rate variability (HRV). HRV calculated via electrocardiogram (ECG) has been shown to track autonomic function. However, many wearable devices utilize photoplethysmography (PPG) to calculate pulse rate variability (PRV). Our study investigated the agreement between PRV and HRV in a beat-to-beat analysis in a sample of American football players. Data from 103 male, Division I collegiate American football athletes, collected over three seasons, were analyzed. Heart rate (HR), pulse rate (PR), and two time-domain indices for PRV/HRV were measured (rMSSD and SDNN). Agreement between PRV and HRV was assessed using Bland-Altman analysis (bias, limits of agreement, and confidence intervals) with supporting error metrics (MAE, RMSE, Pearson r, and Lin’s concordance correlation coefficient). To evaluate whether PRV detected autonomic deviations of the same magnitude to HRV on the same day, a sliding-window z-score threshold-crossing analysis (0.5–2.0 SD) quantified PRV detection rates and PRV delay.

Results

HR and PR were similar at (59.4 (10.3) bpm vs. 59.7 (10.3) bpm). In contrast, PPG-PRV values were lower than ECG-HRV for both rMSSD and SDNN (80.9 (23.1) ms vs. 103.9 (22.0) ms, 141.3 (41.7) ms vs. 167.9 (40.0) ms). Bland–Altman analysis showed negative bias for rMSSD and SDNN across PPG wavelengths, race, and obesity, while HR/PR agreement was high with small bias (0.24–0.44 bpm). In the deviation analysis, PRV detected fewer autonomic deterioration events on the same day as HRV, capturing 16.7% to 56.7% of HRV-identified rMSSD events and 16.8%-52.0% of SDNN events across thresholds, and exhibited an average delay of 1.8–5.5 days in detecting changes already identified by ECG-HRV.

Conclusions

Our findings refute PPG-PRV as an equivalent surrogate for ECG-HRV for tracking autonomic function over time. Specifically, the delay in response when using PPG-PRV to track athlete autonomic function will result in missed opportunities for coaches to prevent autonomic deterioration. Finally, the PRV calculated with PPG should not be called HRV as it confuses scientists and consumers.

Key Points

• Our study contradicts the perception that pulse rate variability, measured via light-based technology, is an acceptable surrogate for heart rate variability among collegiate, Division I American football athletes.

• Pulse rate variability misestimated heart rate variability and showed substantial delays in detecting changes in autonomic nervous system function.

• Wearable health devices utilized for athletic performance would ideally employ electrocardiographic technology for monitoring heart rate variability and autonomic function.

Introduction

The autonomic nervous system (ANS) controls physiological processes like cardiovascular activity, energy metabolism, skeletal muscle contraction, and adrenal levels [1]. Optimal function of the ANS is paramount for peak physical performance as it directly impacts an athlete’s power output, agility, strength, and speed [2–5]. Of concern for coaches and trainers, disturbances in ANS function can manifest from non-functional overreaching, overtraining, insufficient recovery periods, and hyperthermic exposures [4, 6, 7]. Unfortunately, by the time symptoms of ANS dysfunction become observable (e.g., precipitous reductions in speed, coordination, and strength) significant physiological damage has already occurred. Depending on the severity of ANS dysfunction, recovery periods may range from a few days to one year [8–10]. Subsequently, monitoring athletes’ ANS activity has gained significant traction to enhance performance and prevent injury by utilizing non-invasive wearable devices that measure heart rate variability (HRV) [11].

HRV is defined as the variation in time intervals between successive heartbeats and is most accurately measured via electrocardiography (ECG). HRV reflects rapid fluctuations in autonomic nervous system (ANS) activity. The electrical impulses that initiate the cardiac cycle are highly sensitive to the neurotransmitters released by the sympathetic (epinephrine, norepinephrine) and parasympathetic (acetylcholine) branches of the ANS [1]. The ECG measures the depolarization and repolarization events of electrical signals, capturing atrial and ventricular electrical activity. While SA node activation precedes the P wave, it is not directly visible on the ECG [12]. With this, HRV is quantified by analyzing variations in successive R-R intervals, thus serving as a strong measure of ANS activity and function. Despite the high precision of ECG in enabling HRV analysis to quantify sympathetic-parasympathetic interplay, many wearable monitors worn by athletes in sports utilize an indirect, alternative methodology: photoplethysmography (PPG). Of concern, PPG fundamentally differs from ECG, which has led to increasing skepticism [12] regarding the accuracy of its derivative, pulse rate variability (PRV), and the validity of PRV as a surrogate for HRV.

Unlike ECG, PPG utilizes light-based technology to detect changes in blood volume occurring in the peripheral microvasculature (e.g., bicep, wrist, finger) [13, 14]. Specifically, the light emitted from PPG sensors penetrates the skin and underlying tissues (e.g., adipose) reaching the capillary beds. The extent to which the intensity of the penetrating light is absorbed by the capillary beds is largely influenced by blood volume near the measurement site with larger blood volume absorbing a greater intensity of light [15]. Because blood volume changes follow each phase of the cardiac cycle, PPG-derived pulse rate (PR) tracks closely with heart rate (HR). Consequently, the strong relationship between PR and HR is mistakenly assumed present between PRV and HRV. However, several internal factors significantly affect changes in blood volume such as blood pressure, peripheral resistance and vascular compliance potentially leading to discrepancies between PRV and HRV [16, 17]. Further, blood vessels act as a natural filter, dampening the PPG signal. This mechanical low pass filter results in a smoothing effect on PPG signals causing PPG peaks to be more rounded [18] and less defined compared to the sharp, clear QRS intervals in ECG tracings. As a result, the PPG signal lacks the fine temporal resolution that ECG provides, which is crucial for accurately capturing the subtle variations in autonomic nervous system activity. Inherently, these factors remove the high frequency components from PPG signals which are critical to detect subtle changes in cardiac and autonomic activity. Thus, making PRV calculated using PPG less sensitive to these changes which are present in electrocardiographic signals and HRV.

Also, external factors like skin pigment, tattoos, and ambient temperature affect the amount of light penetrating the skin and intensity absorbed subsequently impacting PPG-derived PRV and weakening the correlation with HRV [19–21]. These significant, fundamental differences between PPG and ECG further explain the observed disagreement between PRV and HRV metrics. Despite this evidence, PRV is widely utilized in wearable devices labeled and considered a surrogate for HRV [22]. However, in the sports realm, utilizing a highly sensitive metric, like HRV, allows for coaches to track HRV patterns that predict performance- and/or injury-related outcomes like training adaptations, signs of abnormal trends in ANS activity, and preemptive onset of non-functional overreaching. Thus, coaches must utilize methodologies that reliably and accurately track HRV and ANS activity. Currently, however, no studies previously evaluated the surrogacy of PPG-PRV for ECG-HRV, using a single non-invasive device, for detecting changes in ANS activity consequent to exercise training regimens in collegiate athletes. The current study investigated the agreement between PPG-PRV and ECG-HRV and their predictive power in measuring changes in ANS activity throughout 27 weeks of exercise training for 3 consecutive seasons, in a sample of division I collegiate American football athletes. We hypothesized that PPG-PRV would exhibit poor agreement with ECG-HRV, significantly underestimating ECG-HRV.

Materials and Methods

Study Design

The current study analyzed data from a prospective cohort that examined ANS function of Division I collegiate American football athletes throughout 3, 27-week seasons (2022, 2023 and 2024) [2–4, 23]. Baseline HR, PR, HRV and PRV were measured for 7 min before each practice via an armband monitor equipped with ECG and PPG capabilities and measured during the weeks the athletes trained (n = 27 training weeks).

Subjects

Athletes from one Division I collegiate American football team located in southeast Florida, United States, were recruited to participate in the prospective cohort in 2022, 2023, and 2024. Minimal differences in participant demographic or anthropometric profiles were observed between the 3 seasons and therefore, pooled for analyses. One hundred and three healthy, male football athletes volunteered to participate in the ongoing study. On average, athletes were 22.2 (± 2.1) years old, weighed 106.6 (± 22.5) kg, and were 188.2 (± 6.4) cm tall. Athletes predominantly identified as non-Hispanic (NH) black (73.3%), followed by NH White (14.3%), Asian or Pacific Islander (2.9%) or Hispanic/Latino (11.4%). The anthropometric profile of the athletes showed an average body mass index of 29.9 ± 5.1 kg/m², 34.2% were classified as obese, with a body fat percentage of 14.0 ± 6.7% Tables 1 and 2. All athletes, prior to the study, were fully informed of its protocols, benefits and risks and voluntarily consented. Informed consent was then obtained from all subjects involved in the study. All study protocols followed the ethical principles defined in the Declaration of Helsinki and were approved by the University of Miami’s Institutional Review Board (IRB #20191223).

Table 1.

Demographic and anthropometric profiles for the study sample of D1 collegiate male american football athletes

Demographic profile	Mean (SD) or %
Age (yrs)	22.2 (2.1)
Race/ethnicity (%)
NH white	14.3
NH black	73.3
Asian/Pacific Islander	2.9
Ethnicity (% yes)
Hispanic or Latino	11.4
Anthropometric Profile
Height (cm)	188.2 (6.4)
Weight (kg)	106.6 (22.5)
BMI (kg/m2)	29.9 (5.1)
Obesity (%)	34.2
Body fat (%)	14.0 (6.7)

Obesity defined as a BMI ≥ 30 kg/m²

Yrs: years; NH: non-Hispanic; cm: centimeters; kg: kilograms; BMI: body mass index

Table 2.

Position- and training-related data in a sample of D1 collegiate male american football athletes

Player position	No. or %
Offense	57.1%
Offensive line	15.2%
Skill	41.9%
Defense	42.9%
Defensive line	14.3%
Linebackers	11.4%
Defensive backs	17.1%
Training
No. of athletes	103
No. of sessions	398
No. of seasons	3
Length of season (months)	7

D1: Division I; No. : number

Exercise Training Sessions

The American football athletes, in all 3 seasons, participated in the current study during their 27-week season consisting of two, 4-week summer camps, each separated by a week of rest, one 4-week preseason camp and a 13-weeks in-season (see Fig. 1). All athletes were exposed to the same training which varied in intensity and exercises performed within and between sessions. Training occurred during football practices, were prescribed by the coaching staff and included strength and power-focused resistive exercises, short-distance sprint intervals, aerobic training, and agility training. Importantly, the study protocols did not modify any aspect of the prescribed training regimen.

Fig. 1.

American Football 27-Week Training Schematic

ANS Function via Heart Rate and Pulse Rate Variability

Heart rate (HR), pulse rate (PR) and two, time-domain metrics for HRV and PRV were continuously measured using an armband monitor (Warfighter Monitor™, [WFM], Tiger Tech Solutions, Miami, FL) equipped with electrocardiographic and photoplethysmographic technology. The WFM was previously validated in several subpopulations [2, 24–27]. Athletes were fitted with a WFM on the upper left arm around the widest posterior aspect of the biceps muscle and secured with an elastic strap in the early morning (06:00 to 07:00) prior to the start of practice. For the data shown in the analysis, HR, PR, HRV and PRV, athletes were instructed to remain seated in an upright position, nearly motionless and breathing at their normal rate for 7 min [28].

Heart Rate and Heart Rate Variability

HR/HRV metrics were calculated using the changes in inter-beat RR intervals. Heart rate was calculated as the number of QRS complexes detected per minute and averaged over five minutes. RR intervals are defined as the time between R waves on consecutive QRS complexes and NN intervals were noise-free RR intervals. R peaks were detected utilizing a modified Pan-Tompkins algorithm [29, 30]. Noise-free RR intervals were validated using established signal quality indices (SQI) [31]. From this data, two separate time-domain HRV indices were derived including SDNN (standard deviation of the NN interval) and rMSSD (the root mean square of successive differences between NN intervals). These HRV time-domain indices are well known to reflect parasympathetic and sympathetic autonomic output [32–34]. We utilized an ECG/PPG sampling rate of 100 Hz which provides sufficient bandwidth to detect QRS peaks bandpass filtered between 8 and 15 Hz. Importantly, the WFM previously demonstrated strong correlations with a standard 2-lead chest ECG (R² = 0.95) for measuring the frequency and variations in R-R intervals [25, 27].

Pulse Rate and Pulse Rate Variability

PPG technology, housed in the WFM, was used to measure PR and PRV via blood volumetric changes. Using a derivative based algorithm, peaks in PPG generated pulse waves were detected and defined as the highest amplitude reached for each pulse wave recorded. PR was defined as the frequency of pulse wave peaks detected in a 60 s interval and averaged over 5 min. As in previous studies [15], pulse wave peaks in the current study were assumed equivalent to the R peak on QRS complex measured on an ECG. Thus, the methods for extracting noise-free pulse-to-pulse (PP) intervals and subsequent indices of PRV were identical with those utilized for HRV described above.

Signal Processing and Beat Detection for ECG and PPG

ECG was preprocessed with a bandpass filter (8–15 Hz) to support accurate R-peak detection and reduce baseline wander and high-frequency noise. R peaks were identified using a modified Pan–Tompkins algorithm [29, 30]. The ECG was sampled at 100 Hz to provide sufficient resolution for QRS peak detection [35, 36].

PR and PRV metrics were derived from variations in PPG pulse intervals. Pulse intervals were defined as the time between the largest amplitude in two consecutive pulse wave after filtering. PPG was preprocessed with a bandpass filter (0.25–5 Hz) to remove wander and high frequency noise. PPG peaks were identified by taking the derivative of the filtered signal assessing where the derivative crossed zero and then looking for the maximum PPG value closest to where the zero crossing was detected [30]. The PPG was also sampled at 100 Hz to provide sufficient resolution for pulse peak detection.

Signal Quality Control and Artifact Rejection

To reduce the impact of data artifacts, data quality was evaluated continuously and only high-quality segments were used for HR/PR and HRV/PRV calculations. Signal quality indices (SQI) were computed in 10 s windows and were based on ECG and PPG stability, baseline drift, and amplitude excursions [31]. Windows failing SQI thresholds were excluded. After bad sections were removed and the QRS and Pulse peak intervals were calculated, interval-level checks removed/found missed or extra detections using physiologic RR/PP bounds and comparisons of outlier intervals to local interval statistics. Only segments where both PPG and ECG intervals were both good were used and the same segments were used for both HRV and PRV calculations.

Statistical Analyses

For demographic (age, race/ethnicity) and anthropometric (height, weight, BMI, obesity classification, body fat %) variables, continuous outcomes are summarized as mean ± standard deviation and categorical outcomes as counts and proportions. Analyses included data from 103 athletes, 398 training sessions, and 12,726 individual datasets.

Agreement Between ECG-HRV and PPG-PRV

Because this dataset includes repeated observations per athlete across many sessions, method comparison between ECG-derived HRV and PPG-derived PRV was evaluated using agreement and error analyses. Agreement was quantified using Bland–Altman analysis [37] with complementary error metrics for rMSSD, SDNN, and HR/PR. For each paired observation , ECG-HRV and PPP-PRV values were denoted as . and . Differences were computed as , and the paired mean was computed as . Mean bias was reported as . The standard deviation of was also computed, and limits of agreement (LoA) were calculated as . Uncertainty in mean bias was summarized using a 95% confidence interval based on the t distribution. Uncertainty in each LoA endpoint was summarized using a 95% confidence interval.

In addition to Bland–Altman statistics, the following error metrics were reported for each outcome mean absolute error (MAE) and root mean squared error (RMSE). Association and concordance were summarized using Pearson’s correlation coefficient and Lin’s concordance correlation coefficient (CCC).

ECG/PPG Agreement for Subgroups (PPG Wavelength, Race, Obesity)

Agreement between ECG-derived HRV and PPG-derived PRV was performed for rMSSD, SDNN, and HR/PR separately for each PPG wavelength (red, green, and infrared). Agreement was also evaluated within subgroups defined by BMI and race/ethnicity, where BMI subgroups were defined as obese (BMI > 30 kg/m²) and non-obese (BMI < 30 kg/m²), and race/ethnicity was analyzed within the available categories (non-Hispanic Black, non-Hispanic White, and Hispanic). For each wavelength and subgroup stratum, the same Bland–Altman calculations and error metrics were computed using paired observations within that stratum. Summary metrics included bias (with 95% CI), limits of agreement (LoA; with 95% CI for each endpoint), mean absolute error (MAE), root mean squared error (RMSE), Pearson correlation coefficient (r), and Lin’s concordance correlation coefficient (CCC).

Threshold-Crossing and Delay Analysis for Deviation Detection

To quantify time delay in PRV detecting changes of the same magnitude to HRV, a threshold-crossing analysis was performed using standardized z-scores derived from a 21-day moving window. For each athlete and each HRV/PRV metric, a 21-day mean and standard deviation were computed, and daily z-scores were calculated. Deterioration events were defined when the ECG-HRV z-score fell below a threshold. Thresholds evaluated were 0.5, 1.0, 1.5, and 2.0 standard deviations below the rolling mean.

For each threshold level, the day on which ECG-HRV first crossed the deterioration threshold was treated as the HRV event day. The corresponding PRV crossing day was identified using the same threshold definition applied to the PRV z-score. Detection delay was defined as the number of days between the PRV crossing day and the HRV crossing day. A delay of 0 indicated PRV crossed on the same day as HRV, a delay of 1 indicated PRV crossed the following day, and a delay of N days indicated PRV crossed the threshold N days after HRV. This was capped at 10 days if PRV had not crossed by then. Delays were summarized across all events and athletes at each threshold level. All statistical analyses were performed in MATLAB, version 2021b (MathWorks, Natick, MA, USA) with an a priori statistical significance level set at α < 0.05.

Results

Average ANS activity, including HR/HRV from ECG and PR/PRV from PPG, is summarized in Table 3. Prior to training sessions, athletes exhibited similar mean HR and PR (59.4 (10.3) bpm vs. 59.7 (10.3) bpm), indicating close agreement for rate. In contrast, time-domain variability indices differed, with lower PPG-derived values relative to ECG-HRV for both rMSSD (80.9 (23.1) ms vs. 103.9 (22.0) ms) and SDNN (141.3 (41.7) ms vs. 167.9 (44.0) ms).

Table 3.

Weekly averages of autonomic nervous system activity measured prior to training sessions

Baseline (pre-training)	ECG	PPG
HR (bpm) and PR (bpm)	59.4 (10.3)	59.7 (10.3)
HRV and PRV Indices
rMSSD (ms)	103.9 (22.0)	80.9 (23.1)
SDNN (ms)	167.9 (40.0)	141.3 (41.7)

ECG: electrocardiography; PPG: photoplethysmography; HR: heart rate (derived from ECG); PR: pulse rate (derived from PPG); bpm: beats per minute; HRV: heart rate variability (derived via ECG); PRV: pulse rate variability (derived via PPG); ms: milliseconds

Agreement and error metrics for PRV relative to HRV are presented in Table 4. Across wavelengths and subgroups, PRV systematically underestimated HRV for rMSSD and SDNN, while HR/PR showed minimal bias and strong concordance. Further, using red PPG, rMSSD exhibited a mean bias of − 23.004 ms (CCC = 0.627), and SDNN exhibited a mean bias of − 26.683 ms (CCC = 0.788). In comparison, HR/PR agreement was high bias 0.244 bpm (CCC = 0.995). Similar patterns were observed for green and IR PPG. Subgroup analyses (race/ethnicity and BMI) showed the same directionality, with consistently small HR/PR bias and persistent negative bias for rMSSD and SDNN.

Table 4.

Agreement and error metrics (PRV–HRV)

Metric/category	Units	n	Bias (95% CI)	LoA (lower, upper)	LoA lower 95% CI	LoA upper 95% CI	MAE	RMSE	r	CCC
Red PPG–ECG
RMSSD	ms	12,726	-23.004 (-23.126, -22.883)	-36.680, -9.328	-36.888, -36.473	-9.536, -9.121	23.006	24.039	0.953	0.627
SDNN	ms	12,726	-26.683 (-26.895, -26.472)	-50.541, -2.825	-50.903, -50.180	-3.187, -2.464	26.822	29.328	0.956	0.788
HR/PR	BPM	12,726	0.244 (0.227, 0.261)	-1.704, 2.192	-1.734, -1.675	2.163, 2.222	0.817	1.023	0.995	0.995
Green PPG–ECG
RMSSD	ms	12,726	-31.614 (-31.733, -31.495)	-45.022, -18.206	-45.225, -44.819	-18.410, -18.003	31.614	32.346	0.955	0.481
SDNN	ms	12,726	-22.533 (-22.736, -22.329)	-45.473, 0.408	-45.821, -45.126	0.060, 0.755	22.794	25.391	0.960	0.832
HR/PR	BPM	12,726	0.444 (0.427, 0.460)	-1.448, 2.335	-1.476, -1.419	2.306, 2.364	0.850	1.062	0.996	0.995
IR PPG–ECG
RMSSD	ms	12,726	-28.720 (-28.845, -28.595)	-42.819, -14.621	-43.032, -42.605	-14.834, -14.407	28.720	29.607	0.951	0.525
SDNN	ms	12,726	-29.593 (-29.818, -29.368)	-54.974, -4.212	-55.358, -54.589	-4.596, -3.827	29.709	32.302	0.951	0.754
HR/PR	BPM	12,726	0.404 (0.387, 0.422)	-1.584, 2.392	-1.614, -1.554	2.362, 2.422	0.873	1.092	0.995	0.994
PPG–ECG (NH Black)
RMSSD	ms	9545	-25.253 (-25.401, -25.104)	-39.780, -10.726	-40.034, -39.525	-10.980, -10.472	25.255	26.318	0.946	0.574
SDNN	ms	9545	-29.081 (-29.336, -28.827)	-53.952, -4.211	-54.387, -53.516	-4.646, -3.776	29.199	31.729	0.951	0.752
HR/PR	BPM	9545	0.341 (0.321, 0.362)	-1.654, 2.336	-1.688, -1.619	2.301, 2.371	0.858	1.073	0.994	0.993
PPG–ECG (Hispanic)
RMSSD	ms	1400	-23.970 (-24.340, -23.600)	-37.815, -10.125	-38.448, -37.182	-10.758, -9.492	23.970	24.988	0.961	0.658
SDNN	ms	1400	-27.720 (-28.351, -27.088)	-51.318, -4.121	-52.397, -50.239	-5.200, -3.042	27.775	30.220	0.964	0.807
HR/PR	BPM	1400	0.366 (0.314, 0.418)	-1.587, 2.320	-1.677, -1.498	2.231, 2.409	0.846	1.062	0.997	0.997
PPG–ECG (NH White)
RMSSD	ms	11,326	-21.521 (-21.641, -21.402)	-34.239, -8.804	-34.443, -34.035	-9.008, -8.600	21.523	22.478	0.960	0.659
SDNN	ms	11,326	-27.345 (-27.558, -27.132)	-50.016, -4.674	-50.380, -49.652	-5.038, -4.310	27.428	29.691	0.961	0.786
HR/PR	BPM	11,326	0.063 (0.048, 0.078)	-1.564, 1.690	-1.590, -1.538	1.664, 1.716	0.664	0.832	0.997	0.997
PPG–ECG (BMI > 30)
RMSSD	ms	4734	-25.385 (-25.595, -25.175)	-39.821, -10.949	-40.179, -39.462	-11.308, -10.591	25.387	26.432	0.943	0.554
SDNN	ms	4734	-28.885 (-29.247, -28.522)	-53.805, -3.964	-54.424, -53.186	-4.584, -3.345	28.999	31.559	0.947	0.740
HR/PR	BPM	4734	0.349 (0.320, 0.379)	-1.659, 2.357	-1.708, -1.609	2.308, 2.407	0.866	1.082	0.994	0.993
PPG–ECG (BMI < 30)
RMSSD	ms	7992	-23.554 (-23.704, -23.403)	-37.023, -10.085	-37.280, -36.765	-10.342, -9.827	23.555	24.536	0.957	0.630
SDNN	ms	7992	-27.277 (-27.539, -27.016)	-50.630, -3.925	-51.076, -50.183	-4.372, -3.479	27.385	29.766	0.960	0.793
HR/PR	BPM	7992	0.341 (0.320, 0.362)	-1.558, 2.239	-1.594, -1.521	2.203, 2.276	0.820	1.027	0.996	0.996

Figure 2 presents the number of autonomic deterioration events defined by threshold crossings for 7-, 14-, 21-, and 28-day sliding window z score analysis, using z-score thresholds from 0.5 to 2.0 SD below the rolling mean. Across windows, PRV detected fewer same day deterioration events (HRV/PRV values below z-score thresholds) than HRV for both rMSSD and SDNN. For rMSSD, PRV captured 23.0% to 48.9% of the HRV identified events across thresholds and window sizes. For SDNN, PRV captured 16.7% to 56.7% of HRV identified events across thresholds and window sizes. These findings indicate that, although PRV tracks overall trends, it frequently does not detect autonomic deterioration events on the same day as ECG-HRV.

Fig. 2.

Number of autonomic deterioration events across z score thresholds. For each athlete, daily z scores for rMSSD and SDNN were computed using a rolling window mean and standard deviation, with window sizes ranging from 7 to 28 days. A deterioration event was defined as an HRV value below a Z-score threshold. Same day detections were counted when PRV crossed the same threshold on the same day as HRV. PRV thresholds were calculated from PRV data using the same rolling window approach, not from HRV. HRV is derived from electrocardiography and PRV is derived from photoplethysmography. SD: standard deviation. HRV: heart rate variability. PRV: pulse rate variability. ANS: autonomic nervous system

The delay in PRV detecting the same magnitude of change relative to ECG-HRV for a 21-day window is shown in Fig. 3. Across thresholds, PRV detection lagged HRV by 1.8 to 5.5 days for both rMSSD and SDNN. Importantly, the largest reductions in ANS activity were associated with the longest PRV detection delays. This pattern was consistent across rMSSD and SDNN, supporting that PRV is slower and less sensitive than ECG-HRV for identifying short-term deteriorations when using threshold based deviation detection.

Fig. 3.

Average delay for PRV to detect the same change in rMSSD and SDNN relative to HRV using a 21 day rolling window. For each HRV deterioration event, day 0 was the day the HRV z score first crossed the selected threshold. PRV delay was calculated as the number of days until PRV crossed its corresponding threshold using the same z score definition, where 0 indicates the same day, 1 indicates the following day, and so on. Delays were then averaged across all events for each threshold and metric. To reduce the effect of long lag outliers on the mean, delays were capped at 7 days. SD: standard deviation. HRV: heart rate variability measured via electrocardiography. PRV: pulse rate variability measured via photoplethysmography

Figure 4 provides an illustrative example of HRV and PRV time series with z score thresholds. In this trace, HRV crosses one or more deterioration thresholds on three occasions while PRV does not cross the corresponding thresholds. Although the two series are strongly correlated (r = 0.95, p < < 0.01), this example highlights that correlation does not capture agreement in threshold based detection of autonomic deterioration.

Fig. 4.

Example HRV (upper) and PRV (lower) time series with z score thresholds shown as standard deviation bands. In this example, HRV crosses one or more deterioration thresholds on three occasions while PRV does not cross the corresponding thresholds. Although the two signals are strongly correlated in this trace visually and statistically (r: 0.95, p < < 0.01), correlation reflects overall association and does not indicate agreement in ANS deterioration detection

.

Discussion

The current study investigated the agreement between PPG-PRV and ECG-HRV and their ability to detect ANS activity in a sample of division I collegiate American football athletes. The major findings of this study were (1) significant differences were observed between PPG-PRV and ECG-HRV, such that PPG-PRV underestimated both rMSSD and SDNN compared to ECG-HRV regardless of race, PPG wavelength, and obesity, (2) PPG-PRV reported a substantially lower proportion of athletes eliciting specific thresholds of ANS degradation (0.5, 1.0, 1.5, 2.0 SD) compared to ECG-HRV and (3) PPG-PRV exhibited a significant time delay in detecting changes in autonomic function compared to ECG-HRV at each threshold of ANS deterioration. Collectively, our findings demonstrate that PPG-PRV is not ECG-HRV in an athletic population and as such, PPG-PRV should not be accepted as an appropriate surrogate for HRV.

Our study novelly observed that PRV measured using multiple PPG wavelengths significantly deviated from HRV measured via ECG in athletes. PPG-PRV consistently underestimated ECG-HRV, showing lower PPG-PRV values in athletes prior to training sessions. These marked differences present challenges in quantifying an athlete’s physiological tolerance, adaptations and recovery to acute and cumulative exercise training sessions. At the elite level, the magnitude of the perturbations in ANS responses to training must be accurately measured as it dictates recovery time and successive training sessions that require extremely precise specificity, loading schematics and strategic periodization [38–40]. Interestingly, while the findings of the current study align with others also documenting significant disagreement between PPG-PRV and ECG-HRV, some studies still concluded PPG-PRV provides a “satisfactory” level of agreement and serves as a “reasonable” surrogate for ECG-HRV [12, 22]. These conclusions highlight a pervasive misunderstanding regarding the importance of detecting the subtle variations in ANS activity, which are only identifiable using ECG-HRV. By the time PPG-PRV detects perturbations in ANS activity, sport performance may have already slowed or even regressed [41]. Consequently, coaches are unable to intervene preemptively, either before the onset or during the early stages of autonomic dysfunction. Unfortunately, the nuanced difference between PPG-PRV and ECG-HRV remains uncommunicated resulting in sports coaches, device manufacturers, athletes, scientists, and consumers mistakenly assuming wearable monitoring devices using PPG-PRV accurately track athletes’ ECG-HRV.

Uniquely, this study uncovered a time delay in PPG-PRV detecting changes in ANS activity. For every threshold in change of ANS activity evaluated (0.5, 1.0, 1.5, and 2.0 SD), PPG-PRV detected only a fraction of the autonomic events as compared to ECG-HRV. Specifically, PPG-PRV detected the same magnitude of change in ANS activity in only 16.7% to 56.7% of the events detected using ECG-HRV, with the accuracy worsening amid increases in magnitude. Changes in HRV, whether positive, negative, large or small, provide information on the athletes’ ANS function in response to training and/or competitions, possibly necessitating modifications to training programs and/or recovery time [42, 43]. As shown, PPG-PRV failed to accurately detect changes in HRV in a large proportion of athletes, meaning sport coaches are less likely to evaluate the impact of training or competition on ANS activity, possibly leading to misinformed decision-making. Importantly, our study found a significant time delay in PPG-PRV detecting the same change in ANS activity as ECG-HRV. Further, this time delay widened when identifying larger changes in ANS activity. Suggesting PRV lacks sufficient predictive strength to serve as a reliable surrogate for time-resolved or event-based HRV monitoring.

Clinical Implications

Time is critical for preventing injuries like non-functional overreaching and overtraining, where early signs and symptoms are not easily observable [9]. Previous studies show significant acute and cumulative changes in HRV following exercise training [2, 3, 23, 44, 45]. In our previous works, HRV remained depressed 24 h post training, higher baseline HR and slow HR recovery following acute bouts of exercise, indicating insufficient recovery. Those findings demonstrated immediate (within 24 h) manifestation and detection of alterations in ANS activity and function. The current study highlights the insufficient sensitivity of PPG-PRV to capture such changes as indicated by the large time delay in detecting similar changes as ECG-HRV. Consequently, delayed detection of altered ANS activity, specifically depression, precludes coaches from intervening earlier to mitigate disturbances like prescribing longer recovery and/or modifying subsequent training sessions.

Given the fundamental differences between PPG and ECG, the observed significant disagreement and misestimation of PPG-PRV are expected. The consistently strong agreement between PPG pulse rate and ECG heart rate may have contributed to the common assumption that PPG-derived PRV is interchangeable with ECG-derived HRV [46, 47]. However, the physiological information in HRV is carried by small beat to beat changes in the RR interval, which reflect autonomic modulation. In sport settings, these small changes are often the signal of interest, as they can relate to training adaptation or excessive fatigue [48]. Obtaining this information depends on the ability of the measurement method to resolve small beat to beat timing changes, which can differ between ECG and PPG. Because PPG reflects peripheral blood volume changes, it is influenced by vascular and hemodynamic factors such as blood pressure and arterial compliance, which can alter pulse waveform shape and timing and may attenuate short term variability in pulse intervals [16, 17, 21]. ECG-derived QRS complex shows definitive peaks and valleys, clearly delineating each phase of the cardiac cycle, most importantly the R peak [49]. The PPG-derived pulse wave, however, shows smooth and rounded peaks [18]. These waveform characteristics make peak timing less precise and reduce sensitivity to beat-to-beat variability, contributing to differences between PRV and HRV.

Strengths and Limitations

Our study has several strengths that warrant attention. First, this study uniquely evaluated ECG-HRV and PPG-PRV in athletes, whose sport performance heavily depends on an optimally functioning ANS. Previous studies in athletes mostly (1) utilized devices equipped only with PPG technology, (2) evaluated HRV (or PRV) in the acute, post-exercise period and (3) included a small number of datapoints. Here, with the use of ECG and PPG, we determined that the most widely used technology (PPG/PRV) in the sports realm, does not accurately track HRV. Further, the large number of athletes (n = 103) and datasets (n = 12,726) used strengthens the validity of the study findings. Lastly, the poor agreement and significant differences observed between PPG-PRV and ECG-HRV were independent of PPG wavelength (red, green and infrared), ethnicity, or obesity. Our study also has limitations including the inclusion of only male athletes who played one collegiate sport, American football, and PPG was only measured on the upper arm which potentially restricts the generalizability of the current study.

Conclusion

Our study demonstrated significant limitations in the ability of PPG to act as a surrogate for ECG in HRV measurements in athletes. This was observed in a sample of Division 1 male, collegiate American football athletes. Here, PPG-derived PRV significantly underestimated HRV. Further, PPG-PRV only detected a small fraction of athletes exhibiting significant ANS changes identified via ECG-HRV (between 16.7% to 56.7% depending on the magnitude of change). Lastly, PPG-PRV demonstrated a significant time delay in identifying athletes eliciting ANS changes of the same magnitude detected by ECG-HRV. PRV lagged HRV by 1.8 to 5.5 days to detect the same magnitude of change. These findings show that PPG-PRV is not a direct surrogate for ECG-HRV and suggest using PPG-PRV may increase the risk of coaches missing a critical window for preventing overtraining and declines in sport performance attributed to undetected, ongoing perturbations in ANS activity. Our findings suggest that coaches should utilize ECG-based monitoring devices to accurately track HRV. While PPG-PRV poorly estimates HRV, PPG technology can be utilized to estimate, in some circumstances, other physiological factors important to health including oxygen saturation. Further studies should consider evaluating female athletes, and different types of sports. Lastly, inter-beat interval metrics calculated utilizing PPG should be exclusively marketed and labeled as PRV and not HRV.

Author Contributions

All authors listed on the current manuscript fully met the ICJME criteria for authorship: Conceptualization: SHW, HLW, MJW, ABK and KBD; methodology: HLW, MJW, EJR, JG, KB and LAF; resources: SHW, HLW, EJR, LAF, ABK; data curation: HLW, EJR, and MJW; data visualization: HLW and MJW; project administration: HLW, MJW, JG, KB, EJR and SHW; supervision: HLW and MJW; formal analyses: HLW, EJR, LAF, and MJW; original writing: SMM; and editing/reviewing: MJW, HLW, ABK, KBD, MM, EJR and LAF.

Funding

This study received no funding.

Data Availability

Upon request to the corresponding author, data and materials may be made available.

Declarations

Ethics Approval and Consent to Participate

All athletes, prior to the study, were fully informed of its protocols, benefits and risks and voluntarily consented. Informed consent was then obtained from all subjects involved in the study. All study protocols followed the ethical principles defined in the Declaration of Helsinki and were approved by the University of Miami’s Institutional Review Board (IRB #20191223).

Competing Interests

The following authors are paid employees of Tiger Tech Solutions, Inc, the company that developed the Warfighter Monitor™: HLW, MJW, SMM, and SHW. The remaining authors have nothing to disclose. All data was analyzed and prepared in a double-blind fashion. Each data set was anonymized and given a random ID for each athlete and date. PR, HR, PRV, and HRV were calculated for each unique ID which corresponded to a date and athlete. The resulting ID and metrics were then combined independently before analysis took place.

Transparency Statement

All authors affirm this manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as originally planned have been explained.