Heart Rate Variability-Guided Training for Enhancing Cardiac-Vagal Modulation, Aerobic Fitness, and Endurance Performance: A Methodological Systematic Review with Meta-Analysis

e: This systematic review with meta-analysis was conducted to establish whether heart rate variability (HRV)-guided training enhances cardiac-vagal modulation, aerobic tness, or endurance performance to a greater extent than prede ned training while accounting for method- ological factors. Methods: We searched Web of Science Core Collection, Pubmed, and Embase databases up to October 2020. A random-effects model of standardized mean difference (SMD) was estimated for each outcome measure. Chi-square and the I 2 index were used to evaluate the degree of homogeneity. Results: Accounting for methodological factors, HRV-guided training was superior for enhancing vagal-related HRV indices (SMD+= 0.50 (95% con dence interval (CI) = 0.09, 0.91)), but not resting HR (SMD+= 0.04 (95% CI = 0.34, 0.43)). Consistently small but non-signi cant (p> 0.05) SMDs in favor of HRV-guided training were observed for enhancing maximal aerobic

to evaluate the degree of homogeneity. Results: Accounting for methodological factors, HRV-guided training was superior for enhancing vagal-related HRV indices (SMD+= 0.50 (95% con dence interval (CI) = 0.09, 0.91)), but not resting HR (SMD+= 0.04 (95% CI = 0.34, 0.43)). Consistently small but non-signi cant (p> 0.05) SMDs in favor of HRV-guided training were observed for enhancing maximal aerobic capacity (SMD+= 0.20 (95% CI = 0.07, 0.47)), aerobic capacity at second ventilatory threshold (SMD+= 0.26 (95% CI = 0.05, 0.57)), and endurance performance (SMD+= 0.20 (95%CI = 0.09, 0.48)), versus prede ned training. No heterogeneity was found for any of the analyzed aerobic tness and en- durance performance outcomes. Conclusion: Best methodological practices pertaining to HRV index selection, recording position, and approaches for establishing baseline reference values and daily changes (i.e., xed or rolling HRV averages) require further study. HRV-guided training may be more effective than prede ned training for maintaining and improving vagal-mediated HRV, with less likelihood of negative responses. However, if HRV-guided training is superior to prede ned training for producing group-level improvements in tness and performance, current data suggest it is only by a small margin. Keywords: autonomic nervous system; parasympathetic activity; heart rate recovery; resting heart rate; cardiorespiratory tness 1. Introduction Habitual cardiorespiratory endurance exercise improves a variety of markers related to human health and performance [1]. Exercise programs that ef ciently stimulate adaptations are therefore of interest to general, clinical, and athletic populations. Traditional exercise prescription methodology involves prede ned program parameters in which the intensity, volume, frequency, and timing of training are scheduled in advance. Several prede ned training models have been implemented to improve indices of tness and performance in various populations [2,3]. Though group-level improvements in tness-related outcomes Int. J. Environ. Res. Public Health2021,18, 10299.

Int. J. Environ. Res. Public Health2021,18, 10299 2 of 22 support prede ned training, responses at the individual level are mixed [4]. For instance, Bouchard, An, Rice, Skinner, Wilmore, Gagnon, P²russe, Leon, Rao [5] reported an average increase in maximal oxygen uptake ( . V O2max) of 384 202 mL min 1 after a standardized 20-week training program in 720 healthy subjects. However, individual responses ranged from decrements of 100 mL min 1 in some participants to increments of 1000 mL min 1 in others. Thus, individualized exercise prescription that modi es intensity, volume, and timing of exercise according to the evolving status of the participant may increase the effectiveness and ef ciency of exercise training [6]. Cardiac-autonomic functioning, as indexed by vagal-mediated heart rate (HR) variabil- ity (HRV) indices (i.e., the root-mean-square difference of successive normal R-R intervals (RMSSD), the high frequency (HF), and the standard deviation of the instantaneous beat- to-beat R-R interval variability (SD1)) [7], is a non-invasive marker of acute and chronic adaptation to endurance exercise. In the short-term (e.g., within 48 h after exercise), re- covery of HRV to baseline is thought to coincide with restoration of thermoregulatory, metabolic, hemodynamic, and uid-balance related processes that are disturbed by physi- cal exertion [8]. In the long-term (e.g., weeks to months), HRV pro les that re ect higher and/or more stable resting values have been associated with greater improvements in post-intervention tness outcomes among sedentary [9], moderately-trained [9,10], highly- trained [11–13], and clinical populations [14–16]. Recent experiments have compared prede ned training versus HRV-guided training, in which high intensity exercise is pre- scribed when resting HRV is within or above baseline ranges and low intensity exercise (or passive rest) is prescribed when values are suppressed. Some key ndings favoring HRV-guided training include similar or greater improvements in selected tness outcomes despite fewer high intensity sessions, less heterogeneity in tness changes [17,18], and effectiveness in a variety of populations [17–24]. Recent reviews have aimed to consolidate available ndings. Granero-Gallegos, Gonz¡lez-Qu½lez, Plews, Carrasco-Poyatos [25] reported that HRV-guided training had a signi cantly greater effect on . V O2max versus

ndings favoring HRV-guided training include similar or greater improvements in selected tness outcomes despite fewer high intensity sessions, less heterogeneity in tness changes [17,18], and effectiveness in a variety of populations [17–24]. Recent reviews have aimed to consolidate available ndings. Granero-Gallegos, Gonz¡lez-Qu½lez, Plews, Carrasco-Poyatos [25] reported that HRV-guided training had a signi cantly greater effect on . V O2max versus prede ned training. However, this meta- analysis included the training group (i.e., HRV-guided training and prede ned training) as the analysis unit. Therefore, within-group effect sizes (ESs), which exhibit lower internal validity than between-group ESs [26], were estimated. Moreover, these results should be interpreted with caution since testing for subgroup comparisons based on the training prescription method used was not performed. Medellin Ruiz, Rubio-Arias, Clemente- Suarez, Ramos-Campo [27] also compared HRV-guided training to prede ned training for improving aerobic tness and performance (i.e., . VO2max and maximal power output) in endurance-trained athletes and sedentary subjects and reported no differences between training prescription methods. Nevertheless, heterogeneity analyses to test the in uence of methodological approaches and/or individual differences were not performed. Finally, Düking, Zinner, Reed, Holmberg, Sperlich [28] carried out a systematic review on the effectiveness of HRV-guided training and prede ned training in healthy runners. The authors reported that both training prescription methods induce physiological adapta- tions, with effects of HRV-guided training tending to be greater. Thus, collective ndings are inconclusive. Various methodological approaches have been applied in HRV-guided training inter- ventions that may in uence outcomes and may possibly explain the lack of consensus in recent reviews [25,27,28]. Differences in HRV assessment (e.g., body position, pre-recording stabilization period, measurement duration, selection of the vagal-related HRV index, and respiration rate) and the criterion to modify training (e.g., use of single or average HRV values, and static or rolling baseline reference ranges) may in uence HRV values and, consequently, training prescription. Additionally, the training status of the participants may in uence both HRV and the effectiveness of the training program [29]. Highly trained individuals have less room for improvement and a greater tolerance for training stress than recreationally active and sedentary

or average HRV values, and static or rolling baseline reference ranges) may in uence HRV values and, consequently, training prescription. Additionally, the training status of the participants may in uence both HRV and the effectiveness of the training program [29]. Highly trained individuals have less room for improvement and a greater tolerance for training stress than recreationally active and sedentary populations. Thus, a more thorough consolida-

Int. J. Environ. Res. Public Health2021,18, 10299 3 of 22 tion of the original research that accounts for the aforementioned methodological factors is needed. Aerobic tness and performance have been the primary outcomes of interest in recent reviews [25,27,28]. Whether post-intervention changes in markers of cardiac- parasympathetic modulation vary as a function of prescription methodology is unclear. Resting HR and HRV, as well as post-exercise HR recovery (HRR) are various markers of vagal activity, each of which are independent predictors of cardiovascular morbidity and mortality [30]. The ubiquity of mobile devices capable of tracking resting and exercise- related HR metrics has generated widespread interest in these parameters [31], possibly because they are modi able by lifestyle behaviors [32,33]. HRV in particular exhibits con- siderable versatility in informing on health and wellbeing [32–34], longevity [35], tness, and performance [36–40]. Thus, practical and effective interventions that improve HRV are of growing and universal interest [41]. Modi cation of exercise based on daily HRV is now accessible to the masses, but its ef cacy for improving HRV requires clari cation. A comprehensive investigation into the effectiveness of individualized endurance exercise based on daily HRV may be used to guide best practices for future research and inform applied implementation. Therefore, this systematic review with meta-analysis was conducted to establish whether HRV-guided training enhances cardiac-vagal modulation or aerobic tness and performance to a greater extent than prede ned training while accounting for methodological factors. 2. Methods We conducted and reported a systematic review of the literature and a meta-analysis following thePreferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [42]. The systematic review and meta-analysis protocol were prospectively reg- istered in the PROSPERO database (CRD42020218995). 2.1. Data Search and Sources Potential studies were identi ed via a comprehensive strategy. A systematic search was performed in the Web of Science Core Collection, PubMed, and Embase databases from inception to October 2020 using free-text terms based on the PIC (participants, interventions, and comparisons) strategy. Language restrictions were not applied during this phase. The electronic search of individual databases was adapted as necessary (the full search strategy

ed via a comprehensive strategy. A systematic search was performed in the Web of Science Core Collection, PubMed, and Embase databases from inception to October 2020 using free-text terms based on the PIC (participants, interventions, and comparisons) strategy. Language restrictions were not applied during this phase. The electronic search of individual databases was adapted as necessary (the full search strategy is depicted in the supplementary materials, Section S1). Moreover, the reference lists of previous reviews and full-text articles were manually checked to assess for eligibility. Conference proceedings were also searched on the Web of Sciences Core Collection database. Authors of selected studies were contacted via e-mail in an attempt to identify unpublished or ongoing studies that ful lled our selection criteria. These search strategies were used to minimize the risk of publication bias. 2.2. Study Selection Eligibility criteria were established according to the PICOS (participants, interven- tion, comparison, outcomes, and study design) guideline: (a) sedentary healthy people, physically active, and endurance-trained athletes, regardless of training status or sex (par- ticipants); (b) endurance training prescription in the experimental group based on changes in vagal-related HRV indices (intervention); (c) prede ned endurance training prescription in the control group (comparison); (d) cardiac-vagal modulation (i.e., vagal-related HRV indices, HRR, and/or resting HR), aerobic tness parameters (i.e., . V O2max, maximal aerobic capacity, aerobic capacity at second ventilatory threshold (VT2), and/or aerobic capacity at rst ventilatory threshold (VT1)), and/or endurance performance changes after the intervention (outcomes); and (e) randomized and non-randomized controlled trials (study design) written in English or Spanish.

Int. J. Environ. Res. Public Health2021,18, 10299 4 of 22 2.3. Data Extraction, Coding Study Characteristics, and Potential Moderator Variables The following information was extracted from included studies: (a) study charac- teristics (publication year, country, study design (randomized or non-randomized), and journal); (b) baseline participant characteristics (sample size, sex (male, female, or mixed sample), age, . V O2max, weight, athletic status (sedentary, physically active, recreational, or endurance-trained athletes), and sport (if applicable)); (c) exercise characteristics (training mode (endurance training or combined endurance and strength training), intervention length, and prede ned training characteristics); (d) methodological approach characteristics (vagal-related HRV index (RMSSD, HF, or SD1), power spectral density (PSD) method (if applicable), HRV value (single or averaged), number of average HRV values (if applicable), time of the day, device used, body position (sitting, standing, and supine), measurement length, breathing control, smallest worthwhile change (SWC) or reference criterion ( xed or moving), number of average values (if applicable), and criteria for modifying training in the HRV-guided training group). 2.4. Risk of Bias The Cochrane Collaboration's core risk of bias tool was uses to assess risk of selection, detection, attrition, and reporting bias, which were classi ed as high, unclear, or low risk of bias [43]. Two authors (AM and AJ) performed the study selection, data extraction, and risk of bias assessment. Disagreements were settled by consensus and, when consensus was not achieved, a third author (JMS) assessed the study or information to reach an agreement. 2.5. Computation of Effect Size and Statistical Analyses The standardized mean difference (SMD) was used as the ES index to assess changes in cardiac-vagal modulation, aerobic tness parameters, and endurance performance after the intervention. The SMD was calculated by subtracting the mean change in the outcome variables for the HRV-guided training group from the mean change for the prede ned training group divided by the pooled standard deviation (SD) at baseline, corrected by a factor for small samples. SMD positive values indicated that it was favorable to HRV-guided training. In multiple-intervention studies with a shared prede ned training group, the sample size in the prede ned group was

the HRV-guided training group from the mean change for the prede ned training group divided by the pooled standard deviation (SD) at baseline, corrected by a factor for small samples. SMD positive values indicated that it was favorable to HRV-guided training. In multiple-intervention studies with a shared prede ned training group, the sample size in the prede ned group was split-up [44], allowing us to include several analysis units from the same study. Separate analyses were performed for each SMD index according to the outcome measure when it was reported for at least three analysis units to avoid statistical dependence. A random-effects model was applied for each meta-analysis in which the weighting factor was the inverse variance, de ned as the sum of the within-study and the between-studies variance. A conservative value of 0.7 previously proposed by Rosenthal [45] was used to calculate the variance of each study when the studies did not report the correlations between pre- and post-intervention measures. The analysis comprised calculating the mean ES with its 95% con dence interval (CI), a heterogeneity statistical test, chi-square, and theI 2 index to evaluate the degree of homogeneity of the ESs around the average effect. The magnitude of the SMD was classi ed as trivial (<0.20), small (0.20–0.59), moderate (0.60–1.19), large (1.20–1.99), or very large ( 2.00) [46]. We considered a statistically signi cant effect whenp 0.05. Heterogeneity was classi ed as low, moderate, or high at 25%, 50%, and 75%, respectively. In cases of substantial heterogeneity (chi-square test statistically signi cant and/orI 2 index > 50%), moderator variables analyses were performed by assessing the relationship between the ESs and the potential categorical and continuous potential moderator variables using subgroup analysis and simple meta-regressions, respectively. All analyses were carried out using weighted least squares and assuming mixed-effects models. In case of substantial heterogeneity in vagal-related HRV results, tests for subgroup comparisons were performed based on the vagal-related HRV index (i.e., RMSSD, HF, and SD1) and the HRV value (i.e., single HRV value and averaged HRV value) to test the in uence of methodological factors. For subgroup comparisons based

carried out using weighted least squares and assuming mixed-effects models. In case of substantial heterogeneity in vagal-related HRV results, tests for subgroup comparisons were performed based on the vagal-related HRV index (i.e., RMSSD, HF, and SD1) and the HRV value (i.e., single HRV value and averaged HRV value) to test the in uence of methodological factors. For subgroup comparisons based on the vagal-related HRV

Int. J. Environ. Res. Public Health2021,18, 10299 5 of 22 index, RMSSD and SD1were considered the same index (RMSSD/SD1), as previously reported [47]. In cases of substantial heterogeneity regardless of the outcome measure, the in uence of participant and methodological approach characteristics on our ndings were also investigated. Publication bias analyses were performed using a funnel plot with the trim-and- ll method for imputing possible missing ESs [48,49]. Finally, sensitivity analyses were performed to assess the in uence of any individual study by removing each study and performing all analyses. Statistical procedures were performed using STATA software (version 16.0; Stata Corp LLC, College Station, TX, USA). For articles that did not report methodological information (e.g., single or averaged HRV values) or outcome data (i.e., mean or SD), authors were contacted via e-mail to obtain this information. 3. Results 3.1. Study Selection From a total of 3260 studies after removing duplicates, 10 were eligible for full text analysis [17–24,50,51], of which we excluded two studies from qualitative and quantitative synthesis as follows: based on the same sample and other outcome measures reported (n= 1) [50] and training not guided by daily HRV values (n= 1) [51]. Out of all the selected studies, Kiviniemi, Hautala, Kinnunen, Nissilä, Virtanen, Karjalainen, Tulppo [21] included three HRV-guided training groups and two prede ned training groups, allowing us to include three analysis units. Therefore, a total of 10 analysis units were included in the nal qualitative and quantitative synthesis. Although we attempted to locate unpublished studies, all the selected studies had been published in peer-reviewed journals. A Preferred Reporting Items for Systematic Reviews and Meta-analysis ow-chart of our literature search and selection is presented in Figure. Figure 1.Flow chart of the systematic review process.

Int. J. Environ. Res. Public Health2021,18, 10299 6 of 22 3.2. Study Characteristics Study and participant characteristics are summarized in Table. The eight included studies are from four countries and were published between 2007 and 2020. Seven studies (88%) were randomized trials and one (12%) was a non-randomized trial [17]. In total, there were 199 participants (106 participants allocated to the HRV-guided training group and 93 in the prede ned training group) with a mean SD age of 31.8 4.8 years (min-max: 22.5–38.5 years), of which, 120 were males and 79 were females. Out of the 10 included analysis units, ve (50%) were composed exclusively of male participants, three (30%) by female participants, and two (20%) used a mixed sample. Analysis unit sample size at pre-intervention varied from 14 to 40 participants. Based on the authors sample description, one analysis unit (10%) was composed of sedentary participants [19], three (30%) included physically active adults [21], three (30%) recruited recreationally trained athletes [22–24], and three (30%) included well-trained [17,20] and high-level athletes [18]. Out of all the analysis units composed of athletes, two were runners [22,24], two cyclists [17,20], one cross-country and nordic-skiers [18], while one study reported that endurance athletes were included [23]. The average SD weight and . V O2max at pre-intervention were 71.7 7.1 kg (min–max: 62.1–81.5 kg) and 51.3 9.8 mL kg 1 min 1 (min–max:35.5–65.2 mL kg 1 min 1 ), respectively. One study did not report participant weight [24] and another one did not assess . VO2max [19]. Intervention and methodological approach characteristics are reported in Table. Five studies (62.5%) performed the intervention based on endurance training [17,19–22], two (25%) based on combined endurance and strength training [23,24] and one (12.5%) did not report this information [18]. The intervention length ranged from 2 to 8 weeks. Seven studies (87.5%) carried out daily HRV assessments in the morning after awaken- ing [17,18,20–24] and one (12.5%) performed HRV measurements in the afternoon/evening before performing training sessions [19]. Seven studies (87.5%) explicitly reported that a stabilization period was performed before capturing HRV, ranging from 30 s to 5

The review evaluates the effectiveness of HRV-guided training compared to predefined training methods.