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

Endurance training requires a careful balance of stress and recovery to promote adaptation while avoiding overtraining. The boundary between effective training and overtraining is often unclear due to individual variability in responses. To address this, various psychological and physiological markers are used to monitor training status and readiness, notably heart rate variability (HRV), resting heart rate (RHR), and well-being (WB) scores. HRV, particularly vagally-mediated indices such as RMSSD, reflects autonomic nervous system balance and recovery capacity; reduced resting RMSSD has been associated with fatigue and reduced performance. RHR is a traditional indicator of physical status, often elevated with fatigue or illness, and is mathematically and physiologically intertwined with HRV, warranting normalization considerations. Subjective WB metrics (sleep quality, fatigue, stress, DOMS) are sensitive early indicators of overtraining and relate to expected daily output. However, using physiological or psychological tools in isolation is limited; integration can contextualize HRV trends and improve training precision. Individual differences further necessitate personalized monitoring, as population-level trends can obscure important personal variability. The present study aims to compare interventions that combine vmHRV (RMSSD), RHR, and WB scores to guide training intensity in experienced cyclists, testing whether combined approaches yield greater performance improvements than using vmHRV alone, and exploring the shared and unique information among these measures through correlations and temporal analyses.

HRV is a non-invasive marker of autonomic regulation, reflecting parasympathetic and sympathetic interplay, with baroreflex sensitivity as a key mechanism. Vagally-mediated HRV (vmHRV) is commonly estimated via RMSSD and has been linked to fatigue, overtraining, and performance changes in athletes. Several studies have successfully used HRV-guided training to individualize timing of high-intensity sessions, improving outcomes in running, cycling, skiing, and other endurance contexts (e.g., Kiviniemi et al., Javaloyes et al., Vesterinen et al.). RHR is a widely used physiological marker; elevations often indicate accumulated fatigue or illness. HRV is influenced by average heart rate, and normalization with respect to R–R intervals improves interpretability; combining HRV with RHR may help distinguish recovery from overtraining and capture subtle changes in fitness and fatigue, though more research is needed. Psychological markers—mood disturbance, fatigue, insomnia, irritability—are sensitive early indicators of overtraining and relate to expected training output; concise WB questionnaires covering sleep quality, fatigue, DOMS, and stress are practical tools for monitoring. Neither physiological nor psychological measures alone provide comprehensive insights; integrated approaches balance athlete perception with objective data and have been recommended by consensus statements. Individual differences in autonomic regulation, recovery capacity, resilience, and external stressors underscore the need for individualized monitoring strategies to avoid obscuring important personal variability.

Design: 40-day study comprising a 9-day baseline and a 31-day intervention with daily monitoring and guided training recommendations. Participants: Recruited via social media and HRV4Training newsletter; 119 initially enrolled, 3 excluded (medical conditions), 88 did not complete; final sample n=28 experienced male cyclists (uneven group sizes due to attrition): Group 1 (vmHRV-only) n=8; Group 2 (vmHRV+WB) n=12; Group 3 (vmHRV+WB+RHR) n=8. Ethical approval (UAB CEEAH-5745) and informed consent obtained; participation voluntary and anonymous. Inclusion/exclusion: Injury-free, no medication, no burnout symptoms per ABQ; baseline cycling experience and training history recorded. Instruments: ABQ (15 items, Likert 1–5; higher scores indicate burnout; >70 burnout risk, <50 no risk). vmHRV (RMSSD) via Oura Ring Gen3, WHOOP 4.0, or HRV4Training/EliteHRV with chest strap; devices validated. WB questionnaire: four 7-point items (sleep quality, fatigue, DOMS, stress); WB score computed by summing sleep quality and subtracting fatigue, DOMS, stress (range 10 to −20 AU). RHR via devices. Power meters (e.g., Assioma Duo) for performance tests and training; data uploaded to TrainingPeaks. Performance metrics: Pmax, 1-min, 5-min, 20-min, FTP™ (and FTP™/kg); training intensity monitored via Intensity Factor® (IF®). Procedure: Pre-intervention testing over two days: Day 1 warm-up, 2×1-min maximal efforts (7-min recovery), three ~15 s maximal sprints (7-min recovery), cool-down; Day 2 warm-up, 3×1-min >100 rpm (1-min recovery), 5 min easy, 5 min maximal, 10 min easy, 20 min maximal, cool-down; performed on climbs at constant grade; indoor/outdoor allowed but consistent setting pre/post required. Daily upon waking: record vmHRV (automatic from Oura/Whoop, or 3 min seated/supine chest-strap measurement, fasted), RHR (if available), and WB; data logged via AppSheet. Training recommendations: Group 1 guided by vmHRV; Group 2 by vmHRV+WB; Group 3 by vmHRV+WB+RHR; program advised "High", "Low", or "Rest" days; sessions logged to TrainingPeaks with IF®. Exclusion if non-adherence ≥10 days in intervention. Anthropometrics recorded at testing days; Table 1 reports age, height, weight, BMI, experience, training volume. Statistical analysis: Descriptives as mean±SD. 3×2 MANOVA to compare pre/post changes across groups for cycling power parameters. Within-group changes assessed by Wilcoxon tests. Percentage change (Post–Pre) computed per participant and averaged per group; one-way ANOVA compared percentage changes across groups; significant omnibus tests followed by Bonferroni post-hoc; effect sizes: partial eta-squared (η²: small 0.01, medium 0.06, large >0.14) and Cohen’s d (small 0.2, moderate 0.5, large 0.8). Daily data: Spearman correlations between RMSSD and RHR/WB components per participant; ACF (lags 1–7) on DOMS, fatigue, stress, sleep quality, WB, RMSSD, RHR to assess temporal consistency. Daily analyses used data from 24 participants (RHR incomplete for some in Groups 1 and 2). Significance threshold p<0.05; analyses in SPSS v28; raw data available in OSF.

Training exposure over 31 days: Groups 1, 2, 3 averaged, respectively, high-intensity days ~11, and low-intensity 12, 14, 12 with rest days 8, 6, 8. Baseline comparability: MANOVA indicated no significant pre-intervention differences among groups for any test. Within-group changes (Wilcoxon): Group 1 showed no significant pre–post differences. Group 2 improved significantly in 1-min and 5-min power (p=0.003 for both). Group 3 improved significantly in 5-min (p=0.012), 20-min (p=0.012), FTP™ (p=0.018), and FTP™/kg (p=0.018). Across all groups (Total), significant improvements observed in 1-min (p=0.027, ηₚ²=0.180), 5-min (p=0.001, ηₚ²=0.357), 20-min (p=0.002, ηₚ²=0.315), FTP™ (p=0.000, ηₚ²=0.406), and FTP™/kg (p=0.020, ηₚ²=0.198); Pmax showed no significant change (Total NS; ηₚ²=0.025). Notable numerical improvements: Group 3 increased 5-min power from 310.5 ± 60 to 337.9 ± 71 W and 20-min power from 260.9 ± 55 to 284.5 ± 64 W. Between-group percentage changes (one-way ANOVA): significant differences in 1-min (F(25,2)=4.499, p=0.021, η²=0.265 [large]) and 5-min (F(25,2)=5.082, p=0.014, η²=0.289 [large]); tendency toward significance for FTP™ (F(25,2)=2.541, p=0.099, η²=0.169 [large]). Bonferroni: Group 2 > Group 1 for 1-min (p=0.025, d=9.72 [large]); Group 3 > Group 1 for 5-min (p=0.012, d=6.47 [large]). Daily data: RMSSD and RHR were significantly negatively correlated in 82% of participants (individual Spearman analyses). Significant correlations between RMSSD and WB variables were less frequent: WB in 4 participants, fatigue in 6, DOMS in 5, sleep quality in 4, stress in 1. ACF results: perceived stress showed highest day-to-day consistency at lag-1; sleep quality and RMSSD had moderate autocorrelations at specific lags (e.g., 1 and 5); WB and RHR displayed low autocorrelation across lags.

The integrated approach to guiding training using vmHRV, WB, and RHR produced overall improvements in multiple performance metrics, with the vmHRV+WB+RHR group showing the most consistent gains. Inclusion of WB improved decision-making around high-intensity days, aligning with evidence that wellness predicts training output and contextualizes vmHRV changes (which can indicate both adaptation and fatigue). Adding RHR provided crucial context for interpreting HRV due to HRV’s dependence on average HR and potential HRV saturation at high fitness levels; RHR can help distinguish recovery from overreaching when HRV plateaus. Weekly averaging of HRV and RHR can enhance sensitivity to training adaptations, a principle reflected in the study’s use of rolling averages. The absence of significant changes in Pmax suggests neuromuscular determinants and the likely need for targeted strength training and potentially longer interventions to impact peak power or cycling economy; high baseline fitness may also limit detectable improvements in certain efforts. Individual ACF and correlation analyses underscored substantial inter-individual variability: stress showed persistent day-to-day patterns, while other subjective variables and RHR were more erratic, and relationships between vmHRV and WB components varied across athletes. These findings support personalized monitoring and potentially differential weighting of WB components to tailor training guidance.

Combining vagally-mediated HRV (RMSSD), resting heart rate, and well-being scores to guide training recommendations in experienced cyclists led to significant improvements in key performance metrics, with the triad approach (vmHRV+WB+RHR) yielding the greatest gains. Perceived stress exhibited the most stable day-to-day pattern among subjective measures, while individual variability in correlations between physiological and subjective markers highlights the need for personalized protocols. Integrating physiological and subjective data provides essential context for interpreting vmHRV trends and enhances the precision of training recommendations, helping optimize performance while minimizing overtraining risk. Future work should validate these findings in larger, more diverse cohorts and leverage advanced analytics to further personalize monitoring and training strategies.

Primary limitations include the small final sample (n=28) due to high attrition, reducing statistical power and generalizability; male-only sample limits applicability across sexes. Training recommendations allowed autonomy within "High/Low/Rest" categories, introducing variability in execution; absence of prescribed strength work may have limited improvements in Pmax and endurance economy. IF® quantified external load but did not capture internal strain; future inclusion of %HRmax or similar could balance external and internal load. Heterogeneous, consumer-grade devices (Oura, WHOOP, HRV4Training, EliteHRV) may introduce inter-device variability; standardized instrumentation could improve consistency. The self-administered protocol, despite instructions and contact, lacked direct supervision, potentially affecting adherence and uniformity. Methodological choices (e.g., seated/supine RMSSD without ln transformation) may influence HRV-guided outcomes. Some daily RHR data were incomplete, reducing sample size for time series analyses.