2.1 Ethics statement

Ethics approval and consent to participate were not applicable for a systematic review and meta-analysis. We did not include personal data or patients in this systematic review and meta-analysis. All authors have agreed to publish the results of this work.

2.2 Literature search

We reviewed all studies reporting the effect of exercise training on HRV in T2DM patients. Animal studies were excluded. The PubMed, Cochrane Library, Science Direct and Embase databases were searched until September 20^th 2020, with the following keywords: diabetes AND (exercise OR physical) AND (“heart rate variability” OR HRV). The search was not limited to specific years and no language restrictions were applied. To be included, studies needed to describe our primary outcome variables i.e. HRV data before and after exercise training in T2DM patients, with or without a control group (no physical activity intervention). We excluded studies that assessed the effects of other intervention (such as dietary or psychological intervention) in combination with exercise training. Conferences, congresses or seminars, were excluded. In addition, reference lists from all publications meeting the inclusion criteria were manually searched to identify any further studies not found through the electronic search. Ancestry searches were also completed on previous reviews to locate other potentially eligible primary studies. Two authors (Mathilde Picard, Dutheil Frédéric) conducted the literature searches, reviewed the abstracts, and based on the selection criteria, decided the suitability of the articles for inclusion, and extracted the data. When necessary, disagreements were solved with a third author (Valentin Navel). We followed the guidelines outlined by PRISMA [37] (S1 Checklist).

2.3 Data extraction

The primary outcome analysed was HRV parameters. Time-domain parameters were RR–intervals (or Normal to Normal intervals–NN), standard deviation of RR intervals (SDNN), percentage of adjacent NN intervals varying by more than 50 milliseconds (pNN50), and root mean square of successive RR-intervals differences (RMSSD). Frequency-domain parameters were total power (TP), low frequency (LF), high frequency (HF) and LF/HF ratio. The RMSSD and pNN50 are associated with HF power and hence parasympathetic activity, whereas SDNN is correlated with LF power. Although LF power is an index of both sympathetic and parasympathetic activity, LF power is commonly considered as a measure of sympathetic modulations, particularly when expressed in normalised units. In practical terms, an increase of the LF component is generally considered to be a consequence of an increased sympathetic activity [38]. HF power represents the most efferent vagal (parasympathetic) activity to the sinus node [8, 39–42]. Therefore, an increase of the HF component reflects an increased parasympathetic activity. The LF/HF ratio represents the sympathovagal balance (Table 1). Secondary outcomes included HRV parameters before exercise training, characteristics of training (modalities of exercise such as endurance, resistance, or high intensity interval training; supervised or not; duration and frequency of sessions; duration of training; intensity), characteristics of T2DM (duration of T2DM, HbA1c, treatments), clinical (body mass index, blood pressure, VO₂max or VO₂peak, treatments) and biological (total cholesterol, triglycerides, LDL-cholesterol, HDL-cholesterol) parameters, and sociodemographic (age, sex, smoking).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive characteristics of HRV parameters.

https://doi.org/10.1371/journal.pone.0251863.t001

2.4 Quality of assessment

We used the Scottish Intercollegiate Guidelines Network (SIGN) criteria to check the quality of included articles, both for randomized and non-randomized clinical trials, with the dedicated evaluation grids. Checklists consisted of 10 and 7 items, respectively. We gave a general quality score for each included study based on the main causes of bias evaluated in section 1 of the checklist through 4 possibilities of answers (yes, no, can’t say or not applicable) [43] (S1 Appendix). In addition, we also used the 0–10 Physiotherapy Evidence Database (PEDro) scale for a complementary overview of the quality of the studies (S2 Appendix).

2.5 Statistical considerations

We conducted meta–analyses on the effect of exercise on HRV parameters in T2DM. P values less than 0.05 were considered statistically significant. For the statistical analysis, we used Stata software (version 16, StataCorp, College Station, US) [44–48]. Main characteristics were synthetized for each study population and reported as mean ± standard-deviation (SD) for continuous variables and number (%) for categorical variables. We conducted random effects meta–analyses (DerSimonian and Laird approach) when data could be pooled (more than five data for the same outcome) [49]. Particular attention was paid towards short recordings (1 minute) of HRV parameters [50]. First, we calculated the effect size (ES, standardised mean differences–SMD) [51] of each HRV parameter after exercise compared to baseline (before exercise) in T2DM. ES is a unitless measure centered at zero if HRV does not differ between measures before and after exercise. A positive ES denotes higher levels of the tested HRV parameter in T2DM patients after exercise. An ES of 0.8 reflects a large effect, 0.5 a moderate effect, and 0.2 a small effect. As ES is a unitless measure and as we compared data after and before exercise, frequency-domain HRV parameters measured in ms² or in normalized unit (nu) were combined. In addition, we conducted meta-analyses stratified on type of exercise (endurance, resistance, mixed, HIIT), supervision of exercise or not. We also conducted a meta-analysis on controls to verify the absence of changes within each HRV parameter. We searched for potential publication bias using funnel plots of all aforementioned meta–analyses and we evaluated heterogeneity by examining forest plots, confidence intervals (CI) and I-squared (I²). A low heterogeneity is reflected by I² values <25%, modest for 25–50%, and high for >50%. We verified the strength of our results by conducting further meta–analyses (sensitivity analyses) after exclusion of studies that were not evenly distributed around the base of the funnel. Lastly, we reperformed the aforementioned meta-analyses only using the studies with the highest level of proof, i.e. only on randomised studies and only on randomized controlled studies. When possible (sufficient sample size), meta–regressions were proposed to study the relationship between changes in HRV parameter (RR intervals, RMSSD, pNN50, SDNN, total power, LF, HF, LF/HF) and clinically relevant parameters such as characteristics of intervention (type of exercise, supervised or not, duration and number of sessions, frequency, intensity), clinical parameters (time from T2DM diagnosis, HbA1c, etc.), sociodemographic (age, sex, etc.), or methods of measurement of HRV, and their changes when pertinent. Results were expressed as regression coefficients and 95% CI.

3.1 Quality of articles

The assessment of the quality of these 21 studies was performed using the score SIGN. Results varying from 20% [24] to 80 [23] for Yes responses, with a mean score of 45.7 ± 14.7. Few studies showed a high level of proof mainly due to the lack of a control group or a poor method of randomisation (S2 Appendix and S1 Fig). Using the PEDro scale, scores ranged from 2 [19, 20, 24] to 6 [23, 54–56] out of 10 (S2 Appendix and S2 Fig).

3.2 Study designs and objectives

Included studies were published from 2003 to 2019 and conducted in various geographic locations. All the 21 included studies measured HRV parameters in T2DM patients before and after physical exercise program. Seven studies had a control group of T2DM patients without exercise: five were RCT [25, 52, 54, 56, 57] and two were non-RCT [19, 58]. Five studies compared different T2DM groups of exercise (and had no group without exercise): four were randomized [23, 53, 55, 59], and one not [22]. They assessed the influence of the type [23, 53, 55] or frequency/duration [22, 59] of exercise. Four studies had a control group of non-T2DM patients undergoing the same exercise training than the T2DM [20, 21, 60, 61]. One study compared T2DM patients with or without cardiac autonomic neuropathy [62]. Four were pre-post single group studies [24, 63–65].

3.3 Study characteristics: Inclusion and exclusion criteria

Included patients had to have T2DM, without further details for most studies. Inclusion criteria for TD2M patients were biological in five studies (fasting glucose > 126 mg/dl [19, 22, 52, 60, 61] or glucose levels > 200 mg/dl after an oral glucose tolerance test [60, 61]). Most studies included patients with sedentary behavior or low level of physical activity [21–23, 56, 60–62, 65, 66], i.e. less than 3 hours of physical activity per week [65] or less than 60 min moderate vigorous activity per week [25], or not being involved in regular physical activity [20–23, 60–62]. Some studies required patients to be aged over 35 [53], 40 [23, 58, 60, 61] or 65 [59] years old, or under 60 [53, 60, 61], 65 [55], 70 [23], 74 [59] years old, or between 12 and 19 years old [20]. Nine studies included sex-specific populations: men [22, 53, 56, 57, 63, 64] or women [21, 54, 59]. Six studies included patients according to their body mass index (BMI): >30 [21, 56, 60, 61, 64] or between 24 and 36 kg.m^-2 [23]. The main exclusion criteria were: smoking [21, 23, 24, 55, 56, 59–61, 63–65], exogenous insulin [21, 22, 25, 55, 56, 59, 63], beta blockers or arrythmia [19, 21, 23, 25, 52, 55, 56, 59–61, 63, 64], and cardiovascular disease [21, 23–25, 52, 53, 55, 56, 58, 60, 61, 63–66].

3.4 Characteristics of population

3.5 Intervention

3.6 HRV measures

3.7 Meta-analysis on the effect of physical exercise on HRV

After exercise training several time domains indices significantly improved in T2DM patients (Fig 2) i.e. an increased SDNN (effect size = 0.59, 95% CI 0.26 to 0.93) [22, 24, 25, 54–57, 59, 62–64, 66, 68], RMSSD (0.62, 0.28 to 0.95) [20, 22, 54, 56, 59, 62–64, 68], PNN50 (0.62, 0.23 to 1.00) [22, 57, 62–64, 66] (S3–S6 Figs). For frequency domains (Fig 3), LF decreased (-0.37, -0.69 to -0.05) [20, 21, 23, 25, 54, 56, 57, 59, 62–65, 68], HF (0.58, 0.16 to 0.99) [20, 21, 23, 25, 54, 56–59, 62–65], and LF/HF (-0.52, -0.79 to -0.24) [20, 21, 23, 25, 54, 56–59, 62–65] increased. All aforementioned meta-analyses had a high degree of heterogeneity (> 60%). TP did not differ between groups (0.03, -0.18 to 0.23) [23, 58, 60, 61, 65, 66] (S7–S10 Figs). None of these parameters varied in control groups from RCTs [25, 54, 56–58] (S3–S10 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Summary of meta-analysis on the effect of exercise training on time domain parameters of HRV in T2DM patients—stratified by type of exercise and type of supervision.

https://doi.org/10.1371/journal.pone.0251863.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Summary of meta-analysis on the effect of exercise training on frequency domain parameters of HRV in T2DM patients—stratified by type of exercise and by type of supervision.

https://doi.org/10.1371/journal.pone.0251863.g003

3.8 Meta-analysis stratified by type of exercise

After endurance training, all time and frequency domains measures were improved in T2DM patients: SDNN (effect size = 0.65, 95%CI 0.19 to 1.10) [20, 22, 55, 59, 62–64], RMSSD (0.66, 0.21 to 1.11) [22, 59, 62–64, 66], PNN50 (0.87, 0.57 to 1.18) [20, 22, 62–64], and HF (0.56, 0.18 to 0.94) [20, 21, 23, 59, 62–65] were significantly higher; LF (-0.55, -0.95 to -0.15) [20, 21, 23, 59, 62–65] and LF/HF (-0.49, -0.74 to -0.24) [21, 23, 59, 62–65] were significantly lower. After resistance training, only LF (-0.9, -1.56 to -0.30) [23] and LF/HF (-0.96, -1.59 to -0.33) [23] were significantly lower, without any changes for other time (SDNN, RMSSD, and PNN50) and frequency (HF) domain parameters. After combined endurance and resistance training, there were no changes in any of HRV parameters. After HIIT, only two of the aforementioned parameters were improved: RMSSD (1.26, 0.58 to 1.94) [56] and LF/HF (-1.63, -2.64 to -0.62) [25, 56] (Figs 2 and 3).

For comparisons between type of exercises, metaregressions showed that LF/HF was more improved after endurance (0.55, 0.11 to 0.98) or resistance (1.022, 0.23 to 1.81) training compared with combined endurance and resistance training, and was more improved after HIIT compared with endurance (-1.19, -1.89 to -0.48) or combined endurance and resistance training (-1.73, -2.49 to -0.98). PNN50 was more improved after endurance training compared with combined endurance and resistance training (-1.12, -2.07 to -0.161) (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Metaregressions on factors influencing the effect of exercise training on HRV parameters in T2DM patients.

https://doi.org/10.1371/journal.pone.0251863.g004

3.9 Meta-analysis stratified by type of supervision

After a supervised training, SDNN (effect size = 0.84; 95%CI 0.42 to 1.25) [22, 24, 54, 54–56, 62–64], RMSSD (0.82, 0.43 to 1.21) [22, 54, 56, 62–64], pNN50 (0.85, 0.55 to 1.15) [20, 22, 62–64] were significantly higher; LF (-0.61,-1.01 to -0.21) [23, 54, 56, 62–65] and LF/HF (-0.72,-1.07 to -0.37) [23, 54, 56, 62–65] were significantly lower, HF (0.82,0.26 to 1.37) [23, 54, 56, 62–65] was significantly higher. After a partially supervised training, there were no changes in any HRV parameters except an increase in LF (0.33, 0.01 to 0.65) [21, 57, 58]. After unsupervised training, only LF/HF (-1.09,-1.99 to -0.19) [25] was significantly lower (Figs 2 and 3).

For comparisons between type of supervision, metaregressions showed that PNN50 (1.082, 0.10 to 2.06), LF (0.96, 0.08 to 1.83) and LF/HF (0.70, 0.02 to 1.38) were more improved after supervised exercise training compared to partially supervised (Fig 4).

3.10 Others metaregressions

Frequency of exercise session per week, as well as session duration or total duration of training were not significantly associated with the variation of any HRV parameters. Patients who improved the most their HRV parameters following exercise training were those with the longest time from diagnosis of T2DM (associated with an increased RMSSD and PNN50 following exercise training: 0.09, 0.03 to 0.15, p = 0.010, and 0.06, 0.001 to 0.11, p = 0.046, respectively; and a decrease in LF: -0.09, -0.15 to -0.03, p = 0.006), highest total cholesterol levels at baseline (associated with an increase SDNN and RMSSD following exercise training: 0.02, 0.01 to 0.03, p = 0.003, and 0.01, 0.001 to 0.08, p = 0.044, respectively), triglycerides (associated with a decrease in LF: -0.01, -0.02 to -0.006, p = 0.039), diastolic blood pressure (associated with an increase in RMSSD: 0.08, 0.02 to -0.16, p = 0.022). Patients using biguanides improved less LF and LF/HF after exercise training (0.04, 0.01 to 0.06, p = 0.008 and 0.04, 0.01 to 0.06, p = 0.009, respectively), as well as those using calcium channel blockers and ACE inhibitors (0.08, 0.004 to 0.15, p = 0.042 and 0.01, 0.001 to 0.03, p = 0.043 both for LF/HF, respectively), whereas users of beta blockers had a greater increase in TP (0.14, 0.02 to 0.25, p = 0.028) (Fig 4). Other variables (age, gender, body mass index, smoking, Hba1c, VO₂peak, insulin therapy, duration or condition of ECG recording–deep breathing or not) were not significantly associated with the variation of any outcomes.

An improvement in BMI after exercise training was significantly associated with an improvement in LF (0.13, 0.01 to 0.24, p = 0.036). Similarly, improvement of Hba1c and of SDNN were linked (-0.08, -0.12 to -0.05, p < 0.001), as well as for V0₂ peak and SDNN (0.06, 0.03 to 0.09, p = 0.001), RMSSD (0.05, 0.01 to 0.09, p = 0.017) and LF/HF (-0.05, -0.08 to -0.02, p = 0.005). Improvement of systolic and diastolic blood pressure were linked to RMSSD improvement (-0.24; -0.43 to -0.05, p = 0.023; and -0.17; -0.31 to -0.02, p = 0.032, respectively) (Fig 4).

3.11 Sensitivity analyses

Funnel plots of meta–analyses are presented in S11 Fig. Meta–analyses were reperformed after the exclusion of studies that were not evenly distributed around the base of the funnel and showed similar results. The few studies with the highest level of proof (maximum 6 randomized studies per parameter and maximum 4 RCT per parameter) demonstrated an increase in HF and a decrease in the LF/HF ratio (S12 Fig).

4.1 The benefits of exercise training on HRV in T2DM

There are overwhelming evidence that regular physical activity is associated with a reduced risk for all-cause mortality, and several chronic medical conditions [69]. Most international physical activity guidelines recommend to meet the goal of 150 min/week of moderate-to-vigorous intensity physical activity (MVPA) or 75 min/week of vigorous intensity physical activity [70]. In T2DM patients, exercise leads to better glycemic control, insulin signaling, and blood lipids, reduced low-grade inflammation and improved vascular function [71]. To prevent cardiac autonomic neuropathy, a multifactorial approach is recommended [14], as it can reduce the risk of cardiac autonomic neuropathy progression by 68% [11–13]. Considering that cardiac autonomic neuropathy is a predictor for cardiometabolic events in T2DM patient [72], our meta-analysis showed strong evidence that physical exercise training can improve HRV, both in time and frequency domains. Then, physical exercise training could be a cost-effective intervention to prevent or slow down the cardiac autonomic neuropathy progression in T2DM [73]. Exercise training can improve vagal tone and hence decrease lethal arrhythmias. Even though mechanisms are not yet fully understood, angiotensin II and nitric oxide (NO) are potential mediators of the effects of exercise on vagal tone improvement [74].

4.2 Which type of exercise training?

Opinions differ over the exercise modalities that best limit cardiovascular risk [75]. In patients with metabolic syndrome, it has been shown that mixed training with high-intensity increased visceral fat loss, and that training with high-resistance intensity resulted in faster improvement [76]. Greater improvements in sympathovagal balance were demonstrated for patients with metabolic syndrome following a moderate intensity of training; whereas greater decreases in mean 24-hour heart rate were shown for high-intensity resistance training [39]. But the vast majority of the literature about the effects of exercise on glycemic parameters in T2DM has been centered on interventions involving aerobic exercise and there is ample evidence that aerobic exercise is a tried-and-true exercise modality for managing and preventing T2DM [71, 77]. Resistance training showed also benefits for T2DM patients including improvements in glycemic control, insulin resistance, fat mass, blood pressure, strength, and lean body mass [78]. We demonstrated that endurance training led to an improvement of all parameters of time and frequency domain measures, whereas resistance training and HIIT improved only some outcomes. However, the lack of significant results for other modalities than endurance training can be mainly due to insufficient number of studies reporting those modalities of training.

4.3 Supervision, frequency, and duration of exercise training

In general, there are strong evidences on the benefits of supervised training [79, 80]. In T2DM, we showed that after supervised exercise training, all HRV parameters were significantly improved whereas none of them was improved after partially supervised exercises and only the LF/HF ratio after unsupervised exercises. Our results are also in accordance with literature in T2DM demonstrating the benefits of supervised training on various parameters such as Hba1c, BMI, blood pressure, dyslipidemia or fitness, in comparison with non-supervised training [15, 81]. Interestingly, those benefits were demonstrated independently of dietary intervention. Despite only including studies assessing an exercise intervention alone, none of the included studies except two [53, 60] followed dietary intake and consequently this might have impacted our results. Furthermore, frequency and volume of exercise have been shown to be linked with metabolic improvement in T2DM [82]. Each aerobic exercise session added within a week may produce an additional reduction of 0.39% in HbA1c level [82]. Nevertheless, we did not find any association between frequency, duration of sessions or duration of interventions and HRV improvements. However, studies did not differ considerably between them (duration of sessions were mainly around 45 minutes, frequency of sessions were mainly around 3 sessions per week, and intensity of exercises were around 60–70% for most studies) without any study assessing the impact of low-intensity training.

4.4 Predictors of HRV improvements

Patients who benefited the most from exercise training were those with a longer term diagnosis of T2DM and dyslipidaemia (higher level of total cholesterol and triglycerides), suggesting that benefits might be higher in the most severe patients. Moreover, characteristics of patients such as age or gender may also influence the benefits of exercise training [32, 33, 83, 84]. Some studies reported greater training-induced improvements of HRV in older than in young adults [32, 33], others did not identify any differences [84]. In our meta regressions, age and gender were not associated with any improvement with respect to HRV parameters, BMI or smoking. This suggests that, to some extent, beginning exercise, even late in life can be effective. Finally, it is well known that hypertension is linked with poor HRV [85]. We demonstrated that benefits of exercise training were lower in T2DM patients taking antihypertensive medications, even if literature showed significant improvement in HRV parameters after exercise training in hypertensives patients [86, 87]. T2DM patients using metformin improved less their HRV after exercise compared with T2M patients that did not use metformin. This result may seem contradictory, as metformin has been shown to improve HRV [88]. We also note that most studies (14/21) did not report the use of metformin. Considering that metformin is the first medication to treat T2DM, our results may suffer from a bias of reporting.

4.5 Clinical and biological improvements associated with benefits of exercise on HRV

There is strong evidence showing that physical activity is associated with a reduction in all-cause mortality [89]. Exercise training is known to improve several metabolic parameters in T2DM patients such as HbA1c%, serum insulin and glucose, VO₂peak [28, 90] but physiology of exercise benefits on HRV remains unclear. For example, it remains unclear to what extent changes in blood lipids contribute to the cardiovascular benefits of exercise [16, 91]. In our meta-analysis, an improvement in HDL, LDL, total cholesterol or triglycerides serum levels after exercise was not linked with any improvement in HRV parameters, suggesting that improvement in lipid levels would not be associated with exercise benefits [16, 91]. Conversely, we showed association between HRV improvements and improvement of Hba1c, BMI, and VO₂peak after exercise suggesting that these parameters could be key contributors of exercise benefits on HRV [80]. Increased HRV were therefore linked with a better control of T2DM, and with fitness improvement. Beta blockers are known to affect HRV [92]. We cannot conclude that beta blockers influenced response in HRV to exercise, as only one study reported its use [58] and beta blockers being explicitly an exclusion criteria in most studies [19, 21, 23, 25, 52, 55, 56, 59–61, 63, 64].

4.6 Limitations

We inherited the limitations of all meta-analyses [93] and the limitations and biases of the individual studies investigated. Furthermore, we conducted the meta-analyses on only published articles, so they are theoretically exposed to publication bias. While the meta-analysis is based on a moderate number of studies, the use of broader keywords in the search strategy limits the number of missing studies. In addition, some studies were monocentric, limiting the generalizability of our results. Moreover, the generalizability of our results (improvement of HRV following exercise training) may also be limited to patients who have a T2DM rather well controlled (as they have an Hba1c 7.5%). Data collections and inclusion/exclusion criteria were not identical within each study, which may have affected our results, as well as heterogeneity due to different study designs. To reduce bias of measures, when a study reported HRV in different positions [94], we limited data to decubitus measures, as position and conditions of measure may influence HRV. Most studies included were not RCT, precluding robust conclusions for our meta-analyses. For LF and HF, some studies reported measures in both ms² and normalized units. These reported measures have been included in our meta-analysis and therefore could affecting the weighting of studies. However, we conducted sensitivity analyses with only one or the other unit to verify that it did not affect the results. We also limited the influence of extreme results and heterogeneity by repeating analyses after the exclusion of studies with results not evenly distributed around the funnel plots. Finally, even though in some studies [23, 24, 53, 55, 60, 62–65] patients were asked not to change their dietary intake, an assessment of their dietary intake should have been conducted to verify that the exercise intervention did not modify their eating habits (that could be a confounding factor).