Individual training prescribed by heart rate variability, heart rate and well-being scores in experienced cyclists

Abstract

Optimizing the training of endurance athletes involves the nuanced balance between overload and recovery. Monitoring recovery effectively requires integrating multiple variables. This study evaluates the efficacy of training protocols guided by vagally-mediated heart rate variability (vmHRV), resting heart rate (RHR), and subjective well-being (WB) scores in improving cycling performance. It also explores the relationships between physiological and subjective measures. Twenty-eight experienced male cyclists were divided into three groups: vmHRV-only (Group 1), vmHRV + WB (Group 2), and vmHRV + WB + RHR (Group 3). Over 40 days, participants recorded daily vmHRV, RHR, and WB scores and followed customised training protocols. Pre- and post-intervention cycling tests assessed maximal power (Pmax), 1-min, 5-min, 20-min, and functional threshold power (FTP™). Daily data analysis included correlation and autocorrelation function (ACF) assessments to evaluate trends and individual variability. Across all groups, significant performance improvements were observed for 1-min, 5-min, 20-min, FTP™, and FTP™/kg. Group 3 showed the greatest improvements, particularly in 5-min and 20-min efforts (310.5 ± 60 to 337.9 ± 71 watts, and 260.9 ± 55 to 284.5 ± 64 watts, respectively). ACF revealed stress as having the highest day-to-day consistency among subjective measures. Individual correlations revealed diverse strengths of the relationships between physiological and subjective markers. Combining vmHRV, RHR, and WB offers a more nuanced assessment of athlete readiness and enhances training outcomes compared to vmHRV-only guidance. The study underscores the value of integrating physiological and subjective measures for personalising training protocols and highlights future directions for improving monitoring systems with advanced analytics.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-13540-z.

Introduction

How athletes respond to training and competition is complex¹. Placing enough stress on the athlete primes the body for improvement, but such stress must be followed by periods of recovery to allow fitness adaptations to occur and to, importantly, avoid overtraining. Unfortunately, the boundary between successful training and overtraining is unclear, partly due to individual variability in the training response, since an appropriate load for one athlete may cause overtraining in another². To address these challenges and monitor training and performance readiness, a myriad of psychological and physiological markers has been proposed. Of these, heart rate variability (HRV), resting heart rate (RHR), and well-being scores (WB) are commonly used in practice to guide endurance training.

HRV is defined as the variation in the time interval between consecutive heartbeats, and is used as a non-invasive marker of autonomic nervous system (ANS) regulation of cardiac function. It reflects the dynamic interplay between the parasympathetic and the sympathetic branches of the ANS, which promote recovery and stress responses, respectively^3,4. A key physiological mechanism influencing HRV is baroreflex sensitivity, a mechanism through which blood pressure changes are buffered via autonomic control. This reflex operates via negative feedback loops, modulating heart rate in response to fluctuations in blood pressure. Higher baroreflex sensitivity is associated with greater parasympathetic tone and improved cardiovascular and autonomic health⁵. Specifically, vagally-mediated HRV (vmHRV) reflects the activity of the parasympathetic nervous system, which is largely regulated by the vagus nerve, the tenth cranial nerve and main afferent pathway of the parasympathetic system^6,7. Since the parasympathetic system promotes homeostasis, recovery, and energy conservation, vmHRV provides insight into the body’s capacity to adapt to external and internal demands, such as those imposed by training^8,9.

HRV is measured by time- or frequency-domain methods, with the time-domain Root Mean Square of Successive Differences (RMSSD) being the most frequently used parameter to estimate vmHRV^3,4. In athletes, reduced resting RMSSD has been associated with fatigue, overtraining and reduced performance^10–13. Additionally, several studies have successfully used vmHRV indices to prescribe endurance training in different sports, by individualizing the timing of high-intensity sessions^14–19. For instance, Kiviniemi et al. found that HRV-guided training significantly improved maximal running velocity¹⁶.

Another key physiological marker in monitoring training is resting heart rate (RHR), a well-established indicator of an athlete’s physical status²⁰. Elevated RHR has traditionally been associated with accumulated fatigue or illness and the need for recovery^21,22. Interestingly, research highlights the interplay between RHR and HRV, emphasizing that HRV is inherently influenced by average heart rate due to physiological and mathematical factors. Normalizing HRV by average HR is essential to reduce mathematical bias and improve interpretation, particularly in populations with varying RHR^23,24. Building on this foundation, the combination of RHR and HRV has been used to distinguish overtraining from recovery states in high-training load athletes, with promising potential for capturing subtle changes in fitness and fatigue^12,21. This combination shows potential for improving training precision, although further research is needed^25,26.

Beyond physiological markers, subjective variables also provide valuable insights into athlete’s states. Psychological factors such as impaired mood state, fatigue, insomnia or irritability are highly sensitive and early indicators of overtraining, often appearing before a drop in performance^27–30. They have also shown to provide information about the intensity of output that can be expected when training^31–33. WB scores, derived from self-reported ratings of quality of sleep, fatigue, stress, and muscle soreness, provide a particularly efficient means of monitoring overtraining and recovery^34–37.

Despite the value of both psychological and physiological tools in monitoring and predicting overtraining or readiness, neither provides the desired insights when used in isolation³⁸. Psychological stressors, for example, can impact performance even in sports that require minimal physical exertion, like golf³⁹, and alterations in mood can occur in the absence of changes in performance¹. Similarly, physiological markers like RHR cannot inform of all aspects of fatigue or performance²¹. Consequently, research increasingly supports integrating both approaches, because subjective parameters can reinforce and contextualize HRV data^13,21,34,40. As Bourdon et al.³¹ mentions, the combined approach balances athlete perception and quantifiable practice. Given the complexities of athlete performance, combining markers could improve the precision of training recommendations^12,41.

Moreover, individual differences play a critical role in the integration and interpretation of combined approaches. Research indicates that under similar training loads or stress conditions, athletes may exhibit different HRV responses due to variations in autonomic regulation and recovery capacity^19,21. Likewise, subjective variables can be influenced by psychological resilience or external stressors, leading to varied perceptions^27,39. These differences underscore the need for individualized monitoring strategies, as population-level trends may obscure significant personal variability, potentially reducing the precision of training guidance¹⁸.

The present study aims to compare interventions that combine vmHRV (RMSSD), RHR, and WB scores (based on fatigue, DOMS, stress, and sleep quality) to guide the intensity of training sessions for endurance athletes. The primary aim is to determine the extent to which combination approaches yield greater performance improvements compared to using each metric in isolation. Additionally, the study seeks to explore the shared and unique information contributed by these measures by assessing correlations and temporal patterns between the physiological and subjective markers.

Method

Participants

Participants were recruited through social media networks (Twitter and Instagram) and an advertisement published in HRV4Training’s monthly newsletter. A total of 119 individuals initially expressed interest in participating in the study. Participation was voluntary and anonymous, and each cyclist was informed that they could withdraw at any time. Written informed consent was obtained from each participant and ethical approval was granted by the local ethics committee from the Universitat Autònoma de Barcelona (protocol CEEAH-5745).

Final sample and dropout description

Of the 119 individuals who initially enrolled, 3 were excluded during screening due to pre-existing medical conditions (asthma, diabetes, or cardiovascular pathologies), resulting in 116 eligible participants. These participants were randomly allocated into one of three training groups before the start of the intervention. Over the course of the study, 88 participants did not complete the protocol. Specifically, 23 withdrew due to illness or physical injury, including infections, musculoskeletal injuries, or bike accidents. Another 66 participants voluntarily discontinued participation, with some of the reasons provided including loss of interest, competing training commitments, or work scheduling conflicts. Two participants were excluded due to excessive time between baseline testing and program start, which compromised the study protocol. The final sample therefore consisted of 28 participants who completed the study, with 8 participants in Groups 1 and 3, and 12 in Group 2. The uneven distribution across groups reflects the observed attrition that occurred during the study. A summary of anthropometric and training characteristics for the final sample is provided in Table 1.

Table 1.

Anthropometric and training characteristics of participants.

	Group 1 n = 8	Group 2 n = 12	Group 3 n = 8
Age	39.6 (10.0)	47.7 (15.7)	49.5 (10.3)
Age range	29–55	27–69	37–65
Height	175.4 (9.3)	182.1 (8.3)	181.3 (5.9)
Weight	71.2 (8.6)	76.8 (9.3)	83.1 (10.6)
BMI	23.8 (5.4)	23.8 (1.)	25.3 (2.7)
Cycling experience	18.3 (7.0)	18.8 (19.4)	20.5 (13.8)
Current hours/week training	9.4 (4.1)	11.3 (5.7)	10.0 (4.1)
Current days/week training	4.3 (1.8)	5.3 (1.6)	5.1 (1.3)

Group 1: vmHRV-based training group. Group 2: vmHRV and WB-based training group. Group 3: vmHRV, WB and RHR-based training group. BMI: Body Mass Index. Height is expressed in centimetres, weight in kilograms, cycling experience in years and current training in hours and days per week; and age range as years old. Data are expressed as mean and standard deviation (SD).

Instruments

Athlete Burnout Questionnaire (ABQ)

The ABQ was developed by Raedeke and Smith⁴² and is a self-reported inventory that addresses classic symptoms of burnout, including reduced sense of athletic accomplishment, devaluation of sports participation and emotional/physical exhaustion. In this study, both the English and the Spanish⁴³ versions were used. The questionnaire consists of 15 items and uses a Likert scale from 1 (“almost never”) to 5 (“almost always”). The higher the scores in all items (except 1, 11 and 15), the higher the burnout. Following a total sum of items, higher scores on the ABQ indicate that athletes are high in burnout. Scores over 70 are interpreted as burnout, while those below 50 are interpreted as no risk of burnout.

Oura ring, Whoop strap, smartphone with HRV4Training or EliteHRV

The vmHRV parameter of RMSSD was obtained using either an Oura ring (Gen3, Oulu, Finland), a Whoop strap (Whoop 4.0, Boston, USA) or the phone apps HRV4Training⁴⁴ or EliteHRV⁴⁵ linked to a cardiac chestband. The tools used have been previously validated^46–49.

Well-being (WB) questionnaire

The WB questionnaire consisted of four questions regarding perceived sleep quality, fatigue, muscle soreness (DOMS) and stress. This questionnaire is custom-made, based on the recommendations of³⁵ for monitoring well-being in athletes. Each question was scored from 1 to 7 (with 1 and 7 representing lowest and highest ratings, respectively). A daily WB score was determined by summing sleep quality and subtracting fatigue, DOMS and stress. The maximal and minimum WB scores were 10 and -20 arbitrary units, respectively.

Power meter

A power meter was used to collect data pertaining to pre- and post-intervention performance, as well as daily training sessions. Each cyclist used their own power meter, such as Assioma Duo (Favero Electronics SRL, Italy), or similar. Data were uploaded to each participant’s TrainingPeaks account⁵⁰. For the tests, the variables obtained included maximal power (Pmax), 1-min maximal power (1 min), 5-min maximal power (5 min), 20-min maximal power (20 min) and Functional Threshold Power™ (FTP™), which is the highest power that the rider can sustain for 1 h⁵¹. Power was measured in watts. For the training sessions, the power meter was used to measure the Intensity Factor™ (IF™), which reflects the relative intensity of a session in relation to the rider’s FTP™⁵¹. IF™ was used to define the intensity of each training session (see Supplementary Material A).

Procedure

Inclusion and exclusion criteria

Eligibility was assessed using an initial screening questionnaire that included questions about daily habits (e.g., smoking, medication use, pathologies) and cycling experience (e.g., years of cycling, training days per week, hours training per week). Participants also completed the ABQ. All candidates reported being injury-free, not taking medication, and not experiencing symptoms of burnout at the time of enrolment³.

General procedure

Individuals eligible for the study then signed a written consent and proceeded to perform the initial power tests on the bike. These tests were performed on their own bicycles using their own power meters. Participant weights were measured on the day of the tests. The participants then proceeded with a 40-day study period: 9 days to establish the baseline of the program followed by 31 days of intervention. Every morning upon waking up, participants recorded vmHRV (RMSSD) and their perceived WB. Participants using an Oura Ring or a Whoop Strap obtained the vmHRV values automatically from the device’s app. Those using the HRV4Training or EliteHRV platforms measured vmHRV for 3 min using the cardiac chest band, first thing in the morning, in a seated position, before any other movement, and in a fasted state. Participants registered their vmHRV and WB data into the study program and during the 31-day intervention period, they followed the program’s training recommendations, which advised them to train “High” or “Low” intensity, or to “Rest” (see Supplementary Material A). The procedure is illustrated in Fig. 1. The app used to record the data was AppSheet⁵², an application that enables the creation of mobile applications from other sources.

Fig. 1.

Illustration of the general procedure. During the 9-day baseline period, participants trained freely. During the 31 days of intervention, they followed individualized recommendations.

Participants were randomly allocated to one of three groups: vmHRV-guided group (Group 1), vmHRV-WB-guided group (Group 2) and vmHRV-WB-RHR group (Group 3). Group 1 was advised to train low, high or rest based on their daily vmHRV, whereas Group 2 had an advice based on vmHRV and WB, and Group 3 based on vmHRV, WB and RHR (see Supplementary Material A). In all cases, training sessions were registered and uploaded to TrainingPeaks and training intensity was measured using IF™. Participants exercised at the time of day that was most convenient for them. At the end of the 31 days of intervention, participants performed another set of power tests on the bike and registered their weight on the day of the tests. Participants were instructed to follow the program as closely as possible. If participants did not follow the program for 10 days or more in total (across the 31 days of intervention), they were excluded from the study.

Pre and post power cycling tests

The first day of testing consisted of a 15-min warm-up, followed by 2 × 1 min of riding at maximal power with 7-min recovery in between, followed by three maximum-effort sprints of about 15 s separated by7-min recovery, and a cool-down. The riders were instructed to do the tests standing and in a climb. The second day of testing consisted of a 15-min warm-up, 3 × 1-min intervals at a cadence over 100 rpm with 1 min recovery, 5 min of easy riding, 5 min at maximum intensity, 10 min of easy riding, 20 min at maximum intensity, and cool-down^53–55. The riders were instructed to do the tests in a climb with a constant grade⁵⁶. Participants were familiar with these testing procedures, as such protocols are routinely incorporated into training programs and have demonstrated high reliability in previous research⁵⁷. Participants were free to choose whether to perform the tests indoors or outdoors but were instructed to maintain consistency by using the same setting for the post-test as they had used for the pre-intervention test.

Statistical analysis

Descriptive statistics are reported as mean ± standard deviation (SD) for each measure and subgroup, unless otherwise stated. A 3 × 2 multivariate analysis of variance (MANOVA) was performed to analyse the differences between pre- and post-intervention for each cycling power parameter, comparing the results between the three intervention groups. The non-parametric Wilcoxon test was used to detect changes between the initial and final power tests, for each group separately. Mean data for each participant were used to calculate the percentage change between the initial and final tests (Post–Pre), then averaged for each group. A one-way analysis of variance compared the percentage changes across the three groups. Omnibus tests that achieved statistical significance were followed by Bonferroni post-hoc tests to compare pairs of groups. Effect sizes for one-way analyses of variance were reported using partial-eta squared (η_p²), with values of 0.01, 0.06, and > 0.14 indicating small, medium, and large effects, respectively⁵⁸. For post-hoc pairwise comparisons, effect sizes were reported as Cohen’s d, with values of 0.2, 0.5, and 0.8 indicating small, moderate, and large effects, respectively⁵⁹.

Correlations between daily vmHRV, RHR and WB parameters were calculated using Spearman’s rank correlation test, because the data did not meet the assumptions of normality, as revealed by a Shapiro–Wilk test. Daily vmHRV, RHR and WB parameters were also analysed as a time series using Autocorrelation Function (ACF) analysis to assess temporal consistency, which is the persistence of patterns or dependencies within a time series. Daily data were computed for 24 participants, instead of the total sample of n = 28, because some participants in Groups 1 and 2 did not provide complete RHR data. All statistical analyses were performed using IBM SPSS Statistics Package for Mac OS (version 28.0; SPSS Inc., Chicago, IL, USA). The threshold for statistical significance was set at p < 0.05. The raw data supporting the findings of this study are available on the Open Science Framework (OSF) repository.

Results

The results are subdivided into two sections. First, pre- and post-performance tests were used to evaluate changes in physical performance before and after the intervention. Second, daily data of vmHRV (RMSSD), RHR, and WB scores were examined, providing insights into their trends and interplay during the study.

Training and cycling test performance

Over the 31-day training period, athletes from Groups 1, 2 and 3 trained an average of 11 days of high intensity, 12, 14 and 12 of low intensity and 8, 6 and 8 of rest days, respectively. A MANOVA indicated no significant differences in the pre-intervention power data amongst groups for any of the tests, showing that the cyclists were at a comparable level when starting the intervention. Wilcoxon tests were then applied to detect changes between the initial and final power tests, for each group separately. The results are shown in Table 2. For Group 1, no significant difference between Pre and Post results for any of the tests. For Group 2, there were significant differences in 1 min and 5 min. For Group 3, significant differences were found for 5 min, 20 min, FTP™ and FTP™/kg. When considering all groups together, the most significance was found for 1 min, 5 min, 20 min, FTP™ and FTP™/kg. In all cases of significance, post-intervention values were higher than in the initial tests.

Table 2.

Results of tests Pmax, 1 min, 5 min, 20 min, FTP™ and FTP™/kg, pre- and post-intervention for Group 1 (n = 8), Group 2 (n = 12), Group 3 (n = 8), and the Total average from the three groups.

	Pre	Post	p value	ηp2
Pmax
Group 1	961 ± 150.86	979.13 ± 160.43	ns
Group 2	1016.67 ± 233.77	1014.92 ± 216.39	ns
Group 3	954.5 ± 242.32	1008.38 ± 175.62	ns
Total	983 ± 210.42	1002.82 ± 184.35	ns	0.025
1 min
Group 1	480.38 ± 77.53	474.00 ± 103.84	ns
Group 2	509.50 ± 105.12	557.75 ± 114.60	0.003a
Group 3	475.88 ± 89.52	487.50 ± 110.59	ns
Total	491.57 ± 91.60	513.75 ± 113.36	0.027a	0.180
5 min
Group 1	312.88 ± 58.84	311.38 ± 64.20	ns
Group 2	334.58 ± 52.57	352.25 ± 66.05	0.003a
Group 3	310.50 ± 60.34	337.88 ± 70.96	0.012a
Total	321.50 ± 55.69	336.46 ± 66.70	0.001a	0.357
20 min
Group 1	259.63 ± 44.13	262.75 ± 46.58	ns
Group 2	284.17 ± 43.01	293.75 ± 51.46	ns
Group 3	260.88 ± 54.72	284.50 ± 64.39	0.012a
Total	270.50 ± 46.70	282.25 ± 53.75	0.002a	0.315
FTP™
Group 1	244.75 ± 39.96	252.13 ± 47.08	ns
Group 2	270.50 ± 44.93	277.25 ± 50.93	ns
Group 3	248.75 ± 52.63	271.00 ± 62.65	0.018a
Total	256.93 ± 45.83	268.29 ± 52.58	0.000a	0.406
FTP™/kg
Group 1	3.48 ± 0.70	3.57 ± 0.79	ns
Group 2	3.65 ± 0.76	3.67 ± 0.81	ns
Group 3	3.00 ± 0.63	3.26 ± 0.71	0.018a
Total	3.41 ± 0.73	3.53 ± 0.77	0.020a	0.198

ns: non-significance; ^aPost > Pre; η_p² : partial-eta squared.

Figure 2 presents the percentage difference between the initial and final tests (Post–Pre) for each group. A one-way analysis of variance indicated significant differences in the results of 1 min (F(25,2) = 4.499, p = 0.021, η_p² = 0.265 [large effect]) and 5 min (F(25,2) = 5.082, p = 0.014, η_p² = 0.289 [large effect]) between groups and tendency to significance for FTP™ (F(25,2) = 2.541, p = 0.099, η_p² = 0.169 [large effect]). Bonferroni test showed statistically significant differences between groups 1 and 2 for 1 min (p = 0.025, d = 9.72 [large effect]) and between groups 1 and 3 for 5 min (p = 0.012, d = 6.47 [large effect]). Individual pre- and post-test values, along with percentage changes for each participant, are provided in Supplementary Material B.

Fig. 2.

Percentage change (Post–Pre) and standard error of the performance tests. ‡: significant difference between pairs of groups (Bonferroni, p < 0.05).

Daily data analyses

Individual correlation from daily data

Correlation coefficients of vmHRV (RMSSD) with RHR, and with WB variables for each athlete are shown in Table 3. The correlations were calculated individually for each participant by averaging the values across the 40 days of the study. RMSSD and RHR present significant negative correlation in 82% of the cases. Significant correlations between RMSSD and subjective variables were observed in a limited number of participants: WB in 4 individuals, fatigue in 6, DOMS in 5, sleep quality in 4, and stress in 1.

Table 3.

Spearman correlation coefficients between RMSSD and RHR, sleep quality, DOMS, fatigue, stress and WB, calculated individually for each athlete using their average daily data across the study period.

	ID	RHR	Sleep Quality	DOMS	Fatigue	Stress	WB
RMSSD	2	− 0.65**	0.22	− 0.44*	− 0.49*	− 0.32*	0.47*
	3	− 0.76**	0.40*	− 0.16	− 0.20	− 0.10	0.26
	4	0.46*	0.21	0.06	− 0.04	− 0.04	0.14
	7	− 0.41*	0.03	0.16	0.15	0.09	− 0.13
	8	− 0.90**	0.24	− 0.21	− 0.42*	− 0.01	0.37*
	9	− 0.48*	0.24	− 0.34*	− 0.36*	− 0.31	0.43*
	10	− 0.61**	0.02	0.12	0.01	− 0.23	0.05
	12	− 0.47*	0.12	− 0.06	− 0.05	0.13	0.01
	13	− 0.58**	0.30	− 0.04	0.00	− 0.28	0.22
	16	− 0.70**	0.08	− 0.22	− 0.26	− 0.13	0.24
	17	− 0.56**	− 0.05	− 0.22	− 0.13	− 0.12	0.17
	18	− 0.86**	0.41*	0.17	− 0.03	0.06	0.17
	19	− 0.18	0.09	− 0.07	0.02	− 0.02	0.05
	20	− 0.63**	− 0.04	0.02	0.08	0.17	− 0.06
	22	− 0.11	− 0.15	0.03	− 0.35*	− 0.15	− 0.03
	28	0.01	0.22	− 0.22	− 0.18	− 0.21	0.27
	29	− 0.81**	0.15	− 0.14	− 0.26	− 0.17	0.21
	31	− 0.76**	− 0.02	0.11	− 0.24	− 0.15	0.06
	32	− 0.08	0.03	0.32*	0.34*	0.29	− 0.27
	33	− 0.68**	0.36*	0.38*	0.03	0.13	− 0.02
	35	− 0.47*	− 0.02	− 0.28	− 0.31	− 0.20	0.30
	36	− 0.54**	0.34*	− 0.34*	− 0.32*	− 0.11	0.40*

The analysis included 22 participants, with data points for each variable collected over 40 days (n = 40). Spearman significance: *p < 0.05 and **p < 0.001. Participants 23, 30, and 36 were excluded as their stress values remained constant across all 40 days of the study, preventing meaningful correlation calculations.

Autocorrelations from daily data

Autocorrelation Function (ACF) analysis was used to detect patterns and temporal consistency for DOMS, fatigue, stress, sleep quality, WB, vmHRV (RMSSD), and RHR, recorded daily, across all groups. The ACF measures the correlation between a variable’s values on one day and its values at prior time points (lags). For example, Lag 1 represents the correlation between the values of a variable on one day and the preceding day, Lag 2 represents the correlation with two days prior, and so on. The heatmap in Fig. 3 visualizes these correlations, with darker colours indicating stronger autocorrelations for a given variable and lag. Notably, a high autocorrelation in stress at “Lag Day 1” suggests that stress levels on one day are closely related to those on the previous day, reflecting day-to-day consistency. Sleep quality and RMSSD also exhibit moderate autocorrelations at specific lags (e.g., Lags 1 and 5), suggesting recurring trends over time. In contrast, WB and RHR display consistently low autocorrelation values, indicating more variable or sporadic patterns over the study period.

Fig. 3.

Heatmap of ACF ratings by daily lag and variable. The heatmap displays ACF ratings across multiple lags (days 1 to 7) for each variable. Each lag represents the correlation of a variable’s values with its values from previous days. Rating 0–10: Darker colours indicate stronger autocorrelations. The analysis included 24 participants, with data points for each variable collected over 40 days (n = 40).

Discussion

The goal of the present study was to assess the effectiveness of training protocols guided by vagally-mediated heart rate variability (vmHRV, referring to RMSSD), well-being (WB), and resting heart rate (RHR) on cycling performance. The intervention was divided into three groups depending on the variables that guided training: vmHRV-only (Group 1), vmHRV + WB (Group 2), and vmHRV + WB + RHR (Group 3). Across all groups, improvements were observed in several performance metrics, with Group 3 exhibiting the most consistent performance gains. Daily data were also analysed to explore individual correlations and temporal patterns between WB and physiological markers. Significant correlations were observed between vmHRV and RHR, while correlations between vmHRV and the subjective-report variables were less consistent. ACF revealed strong day-to-day trends for stress, moderate for vmHRV and sleep quality, and sporadic for WB and RHR.

Changes in cycling performance

General improvements in performance

Participants showed significant improvements in 1 min, 5 min, 20 min, FTP™, and FTP™/kg (p < 0.05) when considering all groups collectively. These findings align with previous research, where cyclists who trained based on vmHRV showed enhanced ventilatory thresholds and performance in a 40-min time trial, compared to block-periodization training^15,60. Similar results have been reported in endurance runners, for whom vmHRV-based training improved maximal running velocity¹⁹, and countermovement jump height¹⁷.

Comparison between groups: the role of well-being in performance gains

Comparison between groups revealed that Groups 2 and 3 showed larger improvements in 1 min and 5 min, respectively, compared to Group 1 (p < 0.05). These results may underscore the importance of integrating WB scores into training recommendations. In Groups 2 and 3, high-intensity sessions were performed only when both vmHRV and WB scores were within or higher than their baseline, whereas a drop in those values triggered adjustments to “Low” or “Rest” days (see Supplementary Material A). This protocol aligns with evidence suggesting that pre-training wellness correlates with external training output, as WB appears to indicate the quality of the training output that might be produced on the day^33,35. In runners, a reduction in WB had likely negative to very likely negative impact on the ability of players to fulfil high-intensity efforts⁶¹, suggesting that higher WB scores reflect increased readiness and predisposition for exertion.

Importantly, the combination of vmHRV and subjective variables may enhance training monitoring by contextualizing vmHRV trends. This is because increases in vmHRV can indicate both optimal training adaptations and the onset of fatigue or overreaching, so adding WB markers help contextualize these changes, allowing for better interpretation⁹. As Buchheit ²¹ emphasized, measures of HR cannot fully inform on all aspects of wellness and fatigue, highlighting the importance of combining vmHRV with psychometric assessment for a more holistic view of athletes’ readiness to train.

The WB factors evaluated in the study -fatigue, stress, DOMS and sleep quality- are all known to influence performance. Perceived stress, for example, impacts the ability to perform high efforts⁶², by reducing concentration and vigour⁶³, while fatigue questionnaires has been nominated as the most sensitive tool to variations in training load and performance⁶⁴. DOMS, caused by eccentric muscle activity⁶⁵, is associated with a reduction in performance, as it impacts cycling economy, glycogen repletion and the intensity of subsequent training sessions^65,66. Finally, sleep quality, asides from being essential for health and emotional regulation, cognition and quality of life, is directly linked to athletic performance^67,68, with poor sleep quality predicting worse changes of winning in elite athletes^32,69, as well as impaired speed, endurance, attention, and memory⁶⁷. Despite this, many athletes fail to sleep enough for their activity levels and also fail to recognize the impact of sleep deprivation on performance^67,68, reinforcing the importance of tracking and integrating sleep metrics into training plans. Overall, the findings support the value of self-reported WB assessments in optimizing training schedules and improving athletic performance.

Comparison within groups: the role of RHR in performance gains

Within-group analyses revealed that Group 3 presented the greatest improvements from pre- to post-intervention, particularly in 5 min, 20 min, FTP™ and FTP™/kg. This suggests that combining RHR with vmHRV and WB scores provided valuable additional information for guiding training. Two key mechanisms may explain these findings. First, the inclusion of vmHRV and RHR in Group 3’s training recommendations addressed potential distortions caused by HRV’s dependence on average heart rate (HR). HRV, typically measured via R-R intervals, is influenced by average HR due to physiological and mathematical relationships. Normalizing HRV with respect to R-R intervals (the inverse of HR) mitigates this bias, allowing HRV to better reflect autonomic changes²⁴. Second, combining HRV with RHR helped mitigate HRV saturation, a phenomenon observed at high training loads or fitness levels. HRV saturation occurs when vagal tone is already elevated, reducing HRV’s sensitivity as a marker of parasympathetic activity, since it plateaus even as HR continues to decrease¹². This can complicate the differentiation between recovery and fatigue. For instance, a high HRV and low RHR typically indicate recovery, whereas stable HRV with elevated RHR may signal fatigue or overtraining²¹. The addition of RHR to the guide-training recommendations may help address this limitation by providing context for HRV changes²¹. Additionally, the use of rolling averages further enhances the utility of HRV and RHR in guiding training. RHR exhibits lower day-to-day variation than HRV, making it a practical measure for assessing training adaptation in real-time, while HRV is more sensitive to longer-term fatigue and fitness trends⁷⁰. Notably, averaging HRV and RHR over a week provided stronger correlations with fitness improvements, such as 10-km running performance, than daily measures alone⁷⁰. Our study adopted a similar approach, using weekly averages to capture a more reliable picture of training adaptations.

Overall, while vmHRV alone is useful in detecting nonfunctional overreaching, combining it with RHR may provide a more comprehensive view of an athlete’s condition^12,71. In this context, the greater performance improvements of Group 3 could stem from the combined use of RHR and vmHRV. However, more research is needed to refine these tools, as limitations in using HRV ratios have been noted⁹.

Lack of changes in performance

Interestingly, no significant improvements were found in Pmax across or within groups, and while Group 3 showed a tendency for improvements in 20 min, FTP™, and FTP™/kg compared to Groups 1 and 2, these differences were not statistically significant. A possible explanation is the reliance of these metrics on neuromuscular effort^72,73 and cycling economy^74,75, respectively, both of which are influenced by targeted strength training. For instance, enhancing Pmax typically required targeted strength training to increase muscle cross-sectional area and fibre composition⁷⁶, which endurance alone may not achieve effectively^74,77. Since participants in this study did not incorporate strength training, this may have contributed to limited improvements observed in both short-term power (Pmax) and longer endurance efforts (20 min, FTP™). Additionally, the intervention period of four weeks may have been insufficient to detect changes in peak power, as studies such as⁷⁸, have observed increases in peak power only after an eight-week training period. Another possible explanation is the high baseline fitness of the participants. Given the homogeneity in pre-training levels across groups, the observed differences in response to training programs are unlikely to result from initial disparities in pre-intervention levels⁷⁹. Experienced athletes often display limited improvements in endurance performance due to already optimized physiological characteristics, such as high blood volume or red blood cell counts⁸⁰, while less trained individuals tend to show greater responses to high-intensity interval and volume training⁸¹. Thus, the participants’ advanced training status may have constrained the potential for further performance enhancements in Pmax, 20 min and FTP™ efforts.

Additionally, while analysing within-group differences, Group 1 showed no significant improvement in performance. This finding is surprising, as vmHRV-only guided protocols have previously led to better performance^14–19. Asides from supporting the role of RHR and WB scores as markers to refine guided-training, discussed above, another explanation possible explanation for the lack of significant improvements in Group 1 could lie in methodological differences in how vmHRV is measured. For instance, while other studies such as¹⁶ used an orthostatic test to assess vmHRV, our study measured RMSSD in a supine or seated position. Since body position can significantly influence HRV readings, this may account for differences in findings^9,21,82. Furthermore, whereas other studies have analysed high-frequency (HF) power, this study used RMSSD due its lower sensitivity to breathing rate⁸³ and its reliability in short-term recordings^84,85. Additionally, although a logarithmic transformation of RMSSD (lnRMSSD) was not applied in the present study, this approach may help normalize the distribution and stabilize variance of RMSSD, reducing the influence of extreme values and enhancing comparability across individuals^13,86. Future research could explore different methodologies for HRV-guided training, including alternative parameters such as lnRMSSD and variations in recording positions.

Analysis of daily data: tailoring training recommendations

ACFs are used to examine the degree of correlation between a variable and its previous values over time, identifying trends of fluctuations. This is particularly useful tool for monitoring training and recovery, where both physiological (e.g., vmHRV) and psychological (e.g., stress, fatigue) markers can vary due to external and internal factors. In this study, ACF analysis revealed that stress had the highest autocorrelation at lag-1, indicating greater day-to-day consistency than the other studies variables. This temporal stability suggests that, in this sample, stress reflected a more persistent internal state. This consistency does not necessarily imply usefulness for guiding training, since the marker could remain the same regardless of training load (e.g., a daily rating of 5 within a 7-Likert scale). Still, stress may remain a relevant variable, as previous findings suggest that perceived stress can negatively affect performance and recovery and can be an early sign of overreaching⁸⁷. In contrast, DOMS, fatigue, sleep quality and WB showed more erratic ACF values. While some consistency emerged at short lags, such as around day 2, the irregularity of these variables highlights their susceptibility to daily fluctuations. For instance,⁶⁵ emphasized the need to carefully manage recovery when DOMS is elevated, recommending training cessation until it subsides. The instability of DOMS observed in this study underscores its potential limitations as a predictor of recovery and its critical role in overtraining prevention. Moreover, although physiological measures like HRV are often expected to show more predictable linear or quadratic patterns⁸⁸, the findings here showed only moderate correlations on certain days, such as days 1 and 5. This suggests that HRV might reflect longer-term adaptations rather than daily variability. Collectively, the results emphasize the importance of individualized monitoring, as athletes’ recovery responses can vary based on their unique physiological and psychological profiles. Moreover, the findings underscore the need to consider not only the day-to-day stability of each variable, but also its sensitivity to training load, when determining its usefulness for guiding training decisions.

Individual correlation analyses further revealed considerable variability in physiological and subjective responses. As expected, vmHRV presented a strong negative correlation with RHR for most participants, consistent with its calculations from R-R intervals²⁴. However, correlations between vmHRV and subjective WB variables were considerably less consistent. While several participants displayed significant negative correlations between vmHRV and fatigue or DOMS, as suggested by previous research^34,89,90 these patterns were not universal. Similarly, some individuals exhibited positive correlations between vmHRV and WB, suggesting a link between better autonomic function and subjective well-being, as noted before^91,92, as well as vmHRV and sleep quality^93–95, but it was not the case for all participants.

Overall, these results highlight the need for personalized training programs that monitor key metrics based on an athlete’s unique profiles. In this study, all subjective variables were equally weighted in calculating perceived morning WB. However, the observed variability in correlations suggests that a uniform approach may not adequately capture individual needs, as suggested recently by⁴¹. While vmHRV may be more closely tied to recovery in some individuals, other factors, such as sleep quality, may play a larger role in others. Moving forward, it would be valuable to identify which variables have larger coefficient of variation for each subject and refine the formula to respond to individual dynamics.

Limitations

A main limitation of the study is the small sample size, which reduces the predictive power of the results^59,96. Although the findings are promising, high attrition led to a reduced final sample. While attrition does not inherently invalidate study outcomes if transparently reported and accounted for⁹⁷, the findings should be interpreted with caution until replicated in larger, adequately powered studies. Additionally, the sample only consisted of male athletes, as only three women showed interest, potentially biasing results due to gender differences in HRV and subjective variables^98–100. Another limitation involves the training structure. Participants followed training instructions to perform sessions labelled "High," "Low," or "Rest," which allowed for autonomy but introduced variability in training execution. While this design reflected real-world practice, a more structured plan with prescribed intervals or strength work could have led to more consistent performance outcomes. In this context, training intensity was monitored using IF™, a measure of external training load⁵¹. While IF™ provides an objective indication of work, it does not capture internal physiological strain. Future studies could benefit from integrating alternative parameters such as percentage of maximal heart rate (%HRmax), used in previous studies (e.g.,¹⁶), to balance external and internal load markers.

In terms of instrumentation, data collection relied on a variety of validated consumer-grade tools (e.g., Oura, HRV4Training) to help increase accessibility and recruitment, but which potentially introduced heterogeneity in the data. Future research may benefit from standardising devices across participants to improve inter-device reliability and data consistency. Finally, the protocol of this study was entirely self-administered to replicate real-world training conditions. Although participants received detailed instructions and were in regular contact with the research team, the absence of direct supervision limited the ability to ensure uniform implementation of the protocol. Prior studies have shown that unsupervised training protocols may result in lower adherence rates and less consistent performance outcomes compared to supervised interventions¹⁰¹, and thus future research could consider supervised assessments or hybrid study designs (combining in-person and remote components).

Future directions

This study used physiological and psychological indicators to guide training intensity. In the future, it would be interesting to introduce and investigate more subjective parameters, such as mood, as well as biomarkers like cortisol or hormone levels to further refine monitoring systems. To our knowledge, this is the first study using RHR as a metric to guide training, and given its potential, more research is encouraged, for instance exploring guided-training by LnRMSSD in a ratio with R-R, instead of RMSSD, as did Plews et al.¹².

Additionally, expanding the participant base to include both novice and elite athletes would improve the generalizability of results. Future research should also consider including formal familiarization sessions prior to testing, to improve reliability and reproducibility of performance data¹⁰², a step that was not included in this study. Moreover, retention strategies could also be considered to mitigate high dropout rates, a common challenge in applied longitudinal research⁹⁹. Approaches such as structured onboarding, app-based reminders, and community engagement may help improve adherence and data quality. In these future studies with larger and more diverse populations, demographic factors such as age or BMI should be assessed as potential confounders, as they have shown to influence HRV parameters¹⁰³.

In terms of tools, these monitoring systems should be incorporated into non-invasive, not expensive, time-efficient devices, that could be used by athletes²¹. Advances in methods such as time series analysis or machine learning could be pivotal in understanding the complex interactions between training load, recovery, and the prediction of performance outcomes^31,41,104. For instance, in this study, these methods could help to uncover how daily fluctuations in data affected post-intervention performance, or which variables were most relevant for each athlete. Such insights could help assign differential weights to each well-being component based on their impact on performance. Ultimately, these advances will enable a more individualized approach to athlete management, particularly critical during training and competition phases, when athletes need to perform at their best.

Conclusion

Designing optimal training strategies for endurance athletes is complex. This study evaluated the efficacy of combining vagally-mediated heart rate variability (vmHRV, particularly RMSSD), resting heart rate (RHR), and subjective well-being (WB) scores, including fatigue, stress, DOMS and sleep quality, to guide training recommendations for cyclists. The results showed significant performance improvements, with the combination of vmHRV, RHR, and WB metrics yielding the greatest gains in key performance efforts. Additionally, analysis of daily data revealed perceived stress as the subjective maker with the greatest day-to-day consistency. However, other individual variability in correlations between vmHRV, RHR, and subjective WB variables underscored the importance of personalized training protocols. Overall, the findings reinforce the value of an integrative approach, where vmHRV, RHR, and WB are used collectively to enhance the precision of training recommendations, with RHR and WB scores providing essential context for interpreting vmHRV trends. Such an approach offers a tailored interventions to optimize performance while minimizing the risk of overtraining. Future research should validate these results in larger populations and utilize advanced analytics to further personalize training methodologies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HRV

Heart rate variability

vmHRV

Vagally-mediated heart rate variability

RHR

Resting heart rate

RMSSD

Root mean square of successive differences

HR

Heart rate

WB

Well-being

Author contributions

Conceptualization: CA, LlC; Methodology: CA, LlC; Formal analysis and investigation: CA, LlC; Writing—original draft preparation: CA; Writing—review and editing: CA, LlC, DC; Funding acquisition: LlC; Supervision: LlC.

Funding

This research was funded by MCIN/AEI/10.13039/501100011033 by the Spanish Government (Ref. PID2019-107473RB-C21), and by the Catalan Government (Ref. 2021SGR-00806). Additional funding was provided by the Spanish Ministry of Education through an FPU grant (Ref. FPU2020-05293).

Data availability

The datasets generated during and/or analysed during the current study are available in the Open Science Framework repository, [https://osf.io/w6bzq/?view_only=98c801b0c06a439e9f479f76fabd147a].

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval and Consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Universitat de Barcelona (No. CEEAH-5745). Informed consent for participation in this study was received from all subjects.