The Effects of Mindfulness and Meditation on Vagally-Mediated Heart Rate Variability: A Meta-Analysis

Abstract

Objective

Heart Rate Variability (HRV) is a marker of autonomic nervous system function associated with both physical and mental health. Many studies have suggested that mindfulness and meditation-based interventions (MBIs) are associated with improvements in HRV, but findings are mixed and to date no comprehensive meta-analysis has synthesized results.

Methods

Systematic literature searches were conducted using PsycINFO, Embase, Medline, CINAHL, ERIC and Scopus to identify randomized controlled trials (RCTs) investigating the effects of predominantly seated MBIs on resting-state vagally-mediated HRV. Risk of bias was judged using the Cochrane Risk of Bias tool.

Results

Nineteen RCTs met criteria for inclusion in the meta-analysis. Random-effects meta-analysis found that MBIs were not efficacious in increasing vagally-mediated resting-state HRV relative to control conditions (Hedges’ g = 0.38, 95% CI = − 0.014 to 0.77). When removing an outlier (g = 3.22), the effect size was reduced, confidence interval narrowed, and findings remained non-significant (g = 0.19, 95% CI = −0.02 to 0.39). High heterogeneity in results (I² = 89.12%) could not be explained by a priori determined moderators including intervention duration, study setting and control type.

Conclusion

There is currently insufficient evidence to indicate that MBIs lead to improvements in vagally-mediated HRV over control conditions. Future large, well-designed RCTs with low risk of methodological bias could help add to the current evidence to elucidate any role MBIs might play in impacting HRV.

INTRODUCTION

Mindfulness and meditation-based interventions (MBIs) have become increasingly popular (1), given their potential to facilitate adaptive physical and psychological change (2). MBIs are a class of psychological intervention that focus on developing the skill of non-judgmental awareness of the present moment (i.e. mindfulness) and attentional training (3).The mechanisms underlying potential benefits of MBIs remains unclear and are the focus of increasing attention (4, 5). While there has been much focus on self-reported mechanisms, potential changes to physiological functioning are of particular interest (6, 7) and could provide more robust evidence of the utility of MBIs.

One such physiological marker, attracting rapidly growing attention, is heart rate variability (HRV). HRV, defined as beat-to-beat variations in heart rate, reflects the interplay between parasympathetic and sympathetic influences on heart rate. Vagally-mediated HRV indices that specifically reflect parasympathetic modulation of heart rate include time-domain measures (e.g. the root-mean-square of successive R-R-interval differences; RMSSD), high-frequency (HF) HRV and respiratory sinus arrhythmia (RSA) (8). Indices of vagally-mediated cardiac activity and reactivity are of particular clinical interest because vagal tone is indicative of the sensitivity of the nervous system to everchanging physical and emotional demands in the environment (9). Longitudinal studies have found that high HRV is a powerful indicator of health outcomes such as morbidity and risk of mortality in both medical (10, 11) and community settings (12-14). Moreover, because the prefrontal cortical activity crucial to top-down regulation of emotions also produces inhibitory input to the sinoatrial node by way of the parasympathetic nervous system, vagally-mediated HRV can be considered a peripheral biomarker of emotion regulation (15), though viewpoints are mixed (16). Regardless of what process it reflects, HRV is correlated with indicators of mental health and well-being, including greater psychological flexibility, social approach behaviors, and reduced risk of mental illness (15, 17).

Self-reported mindfulness, as well as MBIs, are associated with enhanced emotion regulation and cognitive control (5, 18), however prior findings are largely reliant on self-reported measures of emotion regulation, which can have notable construct overlap with measures of mindfulness (19). Therefore, there is a need to ascertain the extent to which MBIs are associated with underlying changes in physiological indicators, such as vagally-mediated HRV. MBIs appear to enhance the regulation of emotion by facilitating reduction of evaluative processing, which is supported by structures along the cortical midline, in favor of recruitment of a nonconceptual sensory pathway, supported by a limbic pathway (20, 21). In other words, mindfulness may offer an alternative to effortful top-down cognitive control of negative emotions by instead promoting a more nonconceptual, body-based way of experiencing emotions (20, 21). Thus MBIs may result in improvements in cognitive control and emotion regulation that, by proxy, would be evidenced by increases in HRV (15, 20, 21).

There is some evidence that MBIs are associated with adaptive increases in HRV (22, 23), however findings are mixed (24, 25) and a comprehensive quantitative synthesis of findings is lacking. A scoping review was recently published (26) but did not include any quantitative assessment. To date just one meta-analysis has considered HRV changes associated with MBI participation (27). Synthesizing results from four studies, Radmark and colleagues (27) found no evidence of an association between MBI participation and changes in HRV. Unfortunately, the inclusion of only four studies limited the statistical power to detect a true effect and limited the generalizability of results. Additionally, the analysis was not limited to randomized controlled trials (RCTs), which can increase the risk of bias (28). Finally, results for four different – yet highly inter-correlated measures of HRV were reported, including the low-frequency/high-frequency ratio whose mechanism of action is not well understood (29). To reduce risk of type I statistical error and improve interpretability of results, we respond to these issues by limiting our reporting to vagally-mediated HRV, which has a clear mechanism of action (i.e. parasympathetic down regulation of the heart). We also conduct a comprehensive literature search of RCTs to help synthesize current evidence on the relationship between mindfulness-based-training and potential improvements in HRV.

Methods

Search strategy and selection criteria

The review followed The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (30), and was pre-registered on the PROSPERO database for systematic reviews (# CRD42018108456). Systematic literature searches were conducted using five electronic databases: PsycINFO (Ovid), Embase (Ovid), Medline (Ovid), CINAHL, ERIC and Scopus, covering the period up until December 4^th 2019. Broadly, the search aimed to identify studies that simultaneously related to both mindfulness/meditation and HRV.

Search terms relating to mindfulness and meditation were based on those used in an earlier meta-analysis on meditation (31) and included “mindful*”, “meditat*”, “Vipassana”, “Zen”, “Sahaj yoga”, “Dhyan yoga”. We added the terms “body scan”, “breath awareness”, “focused attention”, (“open-monitoring” or “open monitoring”) and “choiceless awareness”, given that these terms are used widely in the meditation literature. Meditation terms were paired with HRV related terms including: “heart rate variability”, “HRV”, “vagal” and “respiratory sinus arrhythmia”. Only peer reviewed journal articles published in English were considered in this review. A sample search strategy can be found in the Appendix A, Supplemental Digital Content. After removing duplicates, search results were then screened manually by two of three reviewers (authors LB, AR and KE) to identify relevant studies that met our inclusion criteria.

Studies were considered eligible for inclusion if they were RCTs of an MBI that delivered instruction for predominantly formal and seated practice in adults (aged 18+) and included outcome data on at least one index of vagally-mediated HRV. Movement-based contemplative practices such as tai chi and yoga were excluded to reduce heterogeneity, and to limit the confounding factor of physical exercise, which is a known predictor of HRV (32). To delineate our findings from brief interventions and induction studies, eligible interventions were required to have a minimum total duration of four hours and include face-to-face contact with an instructor. To reduce the risk of inflating study findings through the inclusion of very small samples (33), studies were required to have a minimum of at least 10 participants allocated to each arm of the study. A comprehensive list of inclusion and exclusion criteria can be found in Appendix B, Supplemental Digital Content.

HRV Measurement

Time domain measures are the simplest measure of vagally-mediated HRV, and of these RMSSD has been recommended for use because it is more independent of respiratory influences on heart rate than other vagally-mediated measures such as RSA (34, 35). For this reason, RMSSD was selected as the preferred HRV metric to include in our analyses where available, or alternatively another time-domain measure such as the standard deviation of NN intervals (SDNN) was used. If a time-domain measure was unavailable, HF-HRV or RSA was used. These vagally-mediated HRV measures are highly correlated and have been directly compared in previous meta-analyses (36-38).

Data extraction

Data on the year and place of publication, population, sample size, HRV outcome measure and intervention characteristics (name, duration, total contact hours, prescription of homework, mode of delivery, instructor qualifications, control type, number of study arms) were extracted by two independent authors (LB and AR) with excellent inter-rater reliability ( κ = .90) and discussed until consensus was achieved.

Risk of Bias

The Cochrane Collaboration Risk of Bias Tool (39) was utilized to identify potential sources of bias in included studies on four domains. These include 1) method of randomization and allocation concealment (the domain of selection bias); 2) blinding of outcome assessment (the domain of detection bias); 3) quantity and management of missing data (the domain of attrition bias); and 4) selective outcome reporting (the domain of reporting bias). Risk of bias was assessed by two independent judges with good inter-rater agreement (κ = .95), and summary scores were visualised using the robvis tool (40).

Statistical analyses

Random effects meta-analyses were conducted using the Metafor package in R, version 3.5.0. A random effects approach was chosen as a conservative alternative to fixed effects modelling. Random effects modelling is appropriate when there is some between-study variation, such as study population (41). Hedges’ g was used as the effect size measure, which is similar to Cohen’s d but has the advantage in correcting for upwards bias. Hedges’ g was computed based on between-group post-intervention differences in HRV, as baseline data was only available for a subgroup (k = 13) of studies on the primary outcome. To test the robustness of the primary result, differences in change scores were also considered. Where studies included two control conditions (e.g. both an active and waitlist control), we used active control data as this represents a more rigorous control. Post-intervention resting state vagally-mediated HRV was the primary outcome measure. Since statistically non-significant effects can be caused by inadequate statistical power to detect a real effect (e.g. type II error), it is important to consider effect size estimates alongside tests of statistical significance (e.g. if the confidence interval captures the null hypothesised value) (42). Thus, in addition to statistical significance determined by the confidence interval, we also consider the effect size of estimates when interpreting results. Hedges’ g values of 0.25, 0.5 and 0.9 were determined to indicate evidence of small, medium and large effects, as recommended for use in HRV research (43).

Between-study heterogeneity was tested using the Q and I² statistics, and influential cases were determined based on the criteria of Cooks distance being greater than one. Any significant variance was modelled using the a priori determined moderators including intervention-related factors (gold standard versus other interventions and dose), study setting, and control type (active versus passive). Publication bias was examined based on visual examination of funnel plots, as well as significance of the Egger regression and Kendall rank correlation tests.

Results

Our search generated a total of 1045 unique studies. After screening against inclusion criteria, 19 RCTs (combined n analysed = 1465 including 606 allocated to the mindfulness treatment arm) met criteria for inclusion in this review (see figure 1 for a flowchart of included studies).

Figure 1.

Flow chart of included studies in this meta-analysis

Study Characteristics

Characteristics of included studies are found in Table 1. Five of the included studies employed mindfulness based stress reduction (MBSR), which is a gold standard mindfulness-based intervention (1), and the remaining 14 studies used other meditation-based interventions. Seventeen studies compared the MBI to a control condition (k = 13 passive; k = 4 active). Three studies made comparisons to an active treatment, cognitive behaviour therapy(44-46). Three studies measured HRV using 24-hour Holter recordings (24, 47, 48), and the remaining studies used short-term recordings ranging from 2 – 30 minutes (mean = 7.8 minutes).

Table 1.

Study Characteristics of Randomized Controlled Trials

Reference	Country	N	Mean Age	% female	Population	Intervention	Mode of Delivery	Total contact hours	Homework	Instructor Qualifications	Control type	Study arms	HRV measure	Duration of measurement (mins)
(Carroll & Lustyk, 2018)	USA	34	43.4	27	Substance use disorder	MBRP	Group	16	Not reported	Experienced health professional	Passive	3	RMSDD	15
(Crosswell et al., 2017)	USA	71	47	100	Breast cancer survivors	MAP	Group	12	17.5	Mindfulness teacher with >20 years experience	Passive	2	RMSSD	10
(Delgado et al., 2010)	Spain	36	Range 18-24	100	College students high in worry	Mindfulness practices	Group	10	Encouraged	Health professional with > 10 years experience	Treatment: CBT techniques	2	HF-HRV	5
(Delgado-Pastor et al., 2015)	Spain	45	21.5	100	College students high in worry	Mindfulness interoceptive training	Group	6	Encouraged	Not reported	Passive	3	RMSSD	5
(Dunne et al., 2019)	Ireland	42	41	74	Emergency department personnel	ABT	Group	16	32.5	Not reported	Passive	2	RMSSD	24 hours
(Faucher, Koszycki, Bradwejn, Merali, & Bielajew, 2016)	Canada	38	38	38.5	Social anxiety disorder	MBSR	Group	27.5	Prescribed	MBSR instructor with > 20 years experience	Treatment: CBT	2	HF-HRV	30
(Grossman et al., 2017)	Switzerland	168	52.5	100	Fibromyalgia	MBSR	Group	27	42	MBSR instructors with > 7 years experience	Active and Passive	3	RSA	24 hours
(Hunt, Al-Braiki, Dailey, Russell, & Simon, 2018)	USA	119	19.3	74	College students	MBI	Group	4	Encouraged	Not reported	Active and Passive	5	SDNN	5
(Kok et al, 2013)	USA	71	37.5 (median)	66	University staff	LKM	Group	6	Encouraged	Experienced health professional	Passive	2	HF-HRV	2
(Krick & Felfe, 2019)	Germany	267	26.0	21.3	Police officers	MBI	Group	12	Prescribed	Experienced instructor	Passive	2	RMSSD	5
(Muthukrishnan, Jain, Kohli, & Batra, 2016)	India	74	22	100	Pregnant women	MBI	Unclear	10	17.5	Not reported	Passive	2	Unclear	unclear
(Nyklicek et al., 2013)	Netherlands	88	46.1	71	Stressed community sample	MBSR	Group	20	42	MBSR instructor with 1 year experience	Passive	2	RMSSD	5
(Oken et al., 2017)	USA	128	59.8	80	Stressed adults age 50+	MBI	Individual	7.5	Encouraged	Experienced instructor	Passive	2	SDRR	5
(Owens et al., 2016)	USA	19	49.4	89	Those with benign heart palpitations	MBSR	Group	27	42	Not reported	Passive	2	RMSSD	24 hours
(Price et al., 2019)	USA	217	35	100	Substance use disorder	MABT	Individual	12	Encouraged	Not reported	Active and Passive	3	RSA	5
(Sekar et al., 2019)	India	30	28.5	100	Nurses with perceived stress	Mahamantra Chanting	Individual	15	Not clear	Research assistant	Passive	2	SDNN	5
(Wahbeh, Goodrich, Goy, & Oken, 2016)	USA	114	52.1	6	Combat Veterans	Mindful body-scan	Individual	2	14	Trained research assistant	Passive	4	Unclear	5
(Wolever et al., 2012)	USA	239	42.9	77	Insurance employees	Mindfulness at Work	Group*	14	Encouraged	Experienced instructor	Passive	3	RR Intervals	10
(Zimmermann-Schlegel et al., 2018)	Germany	110	58.8	22.1	Type 2 Diabetes Patients	MBSR - modified	Group	20	Encouraged	Health professionals	Passive	2	RMSSD	5

ABT: Attention Based Training. CEB: Cultivating Emotional Balance. H: High. HRV: Heart Rate Variability. L: Low. LKM: Loving Kindness Meditation. M: Medium. MABT: Mindful Awareness in Body Oriented Therapy. MAP: Mindful Awareness Practices. MBRP: Mindfulness Based Relapse Prevention. MBSR: Mindfulness Based Stress Reduction. MORE: Mindfulness Oriented Recovery Enhancement. MRBWT. Mindfulness and Resource Based Worksite Training. TM Transcendental Meditation.

^*Group was delivered in either a conventional classroom or virtual classroom allowing for real-time bi-directional communication with no significant differences in effect sizes across modalities. Note. N represents the total study sample size of those randomized to a condition. Mean age and proportion female are calculated based on the total sample size.

The total duration of interventions ranged from 4 to 28 hours, with a mean of 14 hours. Additionally, seventeen of the nineteen studies encouraged or assigned home mindfulness practice between sessions. About half of the included studies (k = 9) were conducted in the United States. Study samples were diverse including medical (k = 5) mental health (k = 4) and community (k = 10) settings.

Resting State HRV

Sixteen RCTs compared MBI treatment to a control condition and used resting state HRV as an outcome measure. Twelve of these studies used passive control conditions and the remaining four used an active control. There was a non-significant difference in resting state HRV between meditation and control groups, g = 0.38, 95% confidence interval (CI) = −0.014 to 0.77, p = .059 (see figure 2 for the forest plot). This small to medium sized non-significant effect favored the MBI condition. This result was close to significance. Thus, to test the robustness of this result, we also considered meta-analyses using i) group differences in pre-post change scores and ii) pre-post difference in HRV in the treatment group alone. For the group differences in pre-post change analysis, a subgroup of studies (k = 13) were included that had both pre and post intervention data on resting-state HRV. Echoing findings from our primary analysis, this effect was non-significant with evidence of a small to medium effect size, g = 0.29, 95% CI = −0.19 to 0.76. For the less-rigorous analysis of pre-post change in the intervention arm alone (excluding data from the control condition), there was evidence of a significant small-medium increase in HRV post MBI intervention relative to baseline, g = 0.33, 95% CI = 0.11 to 0.55.

Figure 2.

Forest plot of post-intervention group differences in resting-state HRV between intervention and control conditions.

There was a high degree of inter-study heterogeneity in results of the primary analysis Q (15) = 105.39, p < 0.001, with I² indicating that 89.12% of observed variance being attributable to study heterogeneity. While all Cook’s d values were less than 1 indicating no influential cases, the study by Muthukrishnan and colleagues (2016) had a standardized difference of beta value of 1.91, indicating that the pooled effect shifts by 1.91 standard deviation units when the study is removed from the model. When the effect associated with this study was removed, the pooled main effect was smaller and remained non-significant (g = 0.19, 95% CI = −0.02 to 0.39). Heterogeneity was attenuated but remained significant (Q (14) = 33.63, p = 0.002, I² = 56.15%).

Four studies reported on resting HRV at follow-up (mean follow-up = 23.5 weeks). There was no difference in HRV between treatment and control groups at follow-up (g = − 0.18, 95% CI = − 0.62 to 0.26; see forest plot, Figure S1, Supplemental Digital Content.

Mindfulness versus cognitive behavior therapy (CBT)

Three studies compared MBIs to CBT. In this subset of studies, no post-intervention group differences in resting-state HRV were observed, g = 0.44, 95% CI = −0.23 to 1.10. The non-significant difference favored MBI treatment, indicating a potential small to medium effect favoring MBIs over CBT. There was no evidence of heterogeneity in results Q (2) = 4.46, p = 0.11, I² = 55.32%.

Do results vary as a function of intervention related factors?

We considered whether results from studies of the gold standard MBSR intervention (k = 5) differed from studies of non-gold-standard interventions on the primary outcome. Overall, intervention type (MBSR versus other) was not a significant moderator of results QM (2) = 3.93, p = 0.14. In a subgroup analysis, non-gold standard interventions were associated with significantly higher resting-state HRV post-intervention g = .44, 95% CI = .003 to .88. In contrast, the effect was not significant for MBSR interventions, g = 0.09, 95% CI = −0.084 to 1.01.

We also considered if dose was associated with between-study heterogeneity in results. We found that intervention duration was not a significant moderator of results, QM (1) = 0.041, p = 0.84. Longer interventions did not lead to greater improvements in HRV relative to shorter interventions.

Do results vary across study settings?

Study setting (medical, mental health, community) was also a non-significant moderator of results, QM (df = 3) = 4.93, p = 0.18. The non-significant trend favored studies conducted in a medical setting. These studies exhibited a large but non-significant pooled effect, g = 0.81, 95% CI = −0.005 to 1.62. It should be noted that the effect within medical settings included the outlier study by Muthukrishnan and colleagues, that individually reported a very large effect of the intervention on resting state HRV of g = 3.22 (49). When removed, the non-significant effect within medical setting studies was no longer present g = − 0.037, 95% CI = −0.49 to 0.42.

Do results vary between studies with active versus passive control conditions?

Control type (active versus passive) was not a significant moderator of results, QM (2) = 5.16, p = 0.076. The non-significant trend favored studies with a passive control condition. In a subgroup analysis, our primary outcome reached significance among studies with a passive control condition g = 0.52, 95% CI = .07 to .96. There was no effect among studies with an active control g = − 0.03, 95% CI = −.78 to .73.

Risk of bias

Seven studies were assessed as having low risk of bias, scoring low risk on at least four of six domains of study bias (24, 48, 50-54). Six studies were assessed as having high risk of bias, with at least one domain rated at high risk (44, 46, 47, 55-57). Most studies were rated high risk due to having high levels of missing data (more than 20% missing or between 5-20% missing but with no clear indication of appropriate handling of missing data). The remaining six studies were assessed as having unclear risk of bias, with fewer than four domains being rated as low risk, and the remaining domains rated as unclear (22, 45, 49, 58-60). A traffic light plot of assessments is found in Figure S2, Supplemental Digital Content.

A moderator analysis comparing studies at low risk of bias to those of unclear or high risk of bias was significant QM (2) = 6.41, p = 0.041. Studies rated low risk of bias reported no significant post-intervention group differences in resting state HRV g = 0.044, 95% CI = −0.51 to 0.60. The effect was significant, however, for studies rated unclear or high risk of bias, with evidence of a medium to large effect g = 0.64, 95% CI 0.14 to 1.14.

While considering the potential role of duration of HRV measurement was not an a priori determined moderator, we also considered if the primary result varied as a function of HRV duration given that recording duration is a relevant variable that can influence HRV measures (35). Three studies used 24 hour Holter monitor recordings (24, 47, 48) and the remaining studies used short-term recordings with a mean recording duration of 7.8 minutes.., This effect was not significant QM (2) = 3.63 p = 0.16.

Publication bias

The funnel plot (see figure 3) shows asymmetrical bias, as there was an absence of studies to the middle right of the funnel. Bias was also indicated by a significant rank correlation test τ = 0.43, p = .02, but not by an eggers regression which was non-significant, Z = 1.12, p = .26. The trim and fill test estimated zero studies missing on the left side of the funnel and six studies missing on the right side (favoring MBIs). Thus, there is no evidence of publication bias in the expected direction (i.e. non-publication of non-significant findings) (61). A Funnel plot for the follow-up timepoint is found in Figure S3, Supplemental Digital Content.

Figure 3.

Funnel plot for meta-analysis of post-intervention group differences in the primary outcome, resting-state HRV.

Discussion

This is the first comprehensive meta-analysis to examine the effects of mindfulness-based training on HRV relative to a control or active treatment comparison. Our primary analysis found that MBI training was not significantly related to higher resting-state vagally-mediated HRV relative to control conditions at either post-intervention or follow-up. Our primary analysis was robust to meta-analytic method, as it was replicated when considering group differences in change scores, an approach which is more likely to detect a significant effect (62). However, despite the absence of a statistically significant overall result, results were nearing significance and we found evidence of small-medium sized non-significant effects in resting-state HRV which is noteworthy (42). The effect is, however, diminished considerably when a single study that shows evidence of effect size inflation is removed. Taken together, these results suggest a small, but potentially meaningful effect that warrants further investigation.

A recent scoping review has identified HRV as a potentially useful biomarker, capable of demonstrating MBI effects (26) but the authors failed to quantitatively synthesize study results and they included uncontrolled studies in their review which are lower quality and more prone to bias (28). Here, we address these limitations by examining objective evidence from RCTs linking MBI training with changes in HRV. Overall, we found little evidence of improved resting-state vagally-mediated HRV following mindfulness training.

Our primary outcome of group differences in resting-state HRV was approaching significance (p = .059), and was associated with small-medium effect favoring the intervention group (g = 0.38) so it is important to consider the robustness of this result. Firstly, we found evidence of high levels of between-study heterogeneity (I² = 89.12%). Thus, it is plausible that differences in study design could explain why a subgroup of included studies might exert salutary effects on HRV whereas others might not (1). If there was a real effect of mindfulness on HRV, we might expect to see evidence that more rigorous interventions – such as the gold standard MBSR program – were more efficacious in increasing HRV than shorter and less regulated interventions. However, intervention type (MBSR versus other) was a non-significant moderator of the resting-state HRV. Moreover, when investigating the subgroups separately, other interventions (g = .44) but not MBSR (g = .09) were associated with significantly higher HRV post-intervention. Similarly, there was no evidence of a dose response, meaning that longer interventions were no more likely to lead to improvements in HRV than shorter interventions. Taken together, our findings show that gold standard and more intensive MBIs are no more likely to lead to improvements in HRV than shorter and less intensive interventions. This evidence points to the fact that our non-significant main finding cannot be explained by the inclusion of brief or non-gold standard interventions. On the contrary, the gold standard MBSR program was associated with a trivial effect size, as well as a non-significant result.

A second way to investigate the robustness of our finding is to consider alternate methods of conducting the meta-analysis. Thirteen studies had data on pre- and post-intervention resting state HRV, enabling us to conduct a meta-analysis of change scores in this subgroup. Change score meta-analyses may be more likely to detect significant effects than analyses based on study arm group differences alone (62). However, our analysis of change scores echoed our main result; there was no difference in pre-post resting HRV change scores in MBI versus control conditions, although the small to medium effect size (albeit non-significant) was replicated (g = 0.29). It is important to note that when we removed control data and considered pre-post change in the intervention group alone, there was evidence of a significant effect (g = .33). This finding agrees with prior systematic reviews investigating associations between meditation and HRV which have reported on evidence of significant effects (26, 63). Both of these review papers included one-arm trials in their descriptive summaries, and no quantitative synthesis was conducted. This indicates that any improvement in HRV observed in one-arm trials of MBIs may be due to common factors that are not specific to meditation (64). Non-randomized studies cannot add to our current knowledge of the efficacy of MBIs in improving HRV, and only large well-designed RCTs will be helpful in adding to the evidence at this stage (1, 65).

A third consideration of the robustness of our results is to look at the pooled effect when outliers and/or influential cases are removed from the model. As visualized in the forest plot (figure 2), we found evidence that the effect reported by Muthukrishnan and colleagues (49) was significant and substantially higher than other studies (g = 3.22). It is unclear why this intervention differed from the others and exerted a very large effect on resting HRV. When this study was removed from analysis, the pooled effect was attenuated (g = .19) and remained non-significant.

We cannot definitively determine if our non-significant result is primarily due to limited data, methodological differences between studies, or a combination of the two. However, we recommend interpreting the non-significant overall effect size (g = 0.38) with caution because this effect was not observed in studies using the gold-standard MBSR intervention (g = 0.11), and it was not observed in studies rated as having a low risk of bias, assessed using the Cochrane collaboration tool (g = 0.044). Additionally, studies with a higher risk of bias tended to report larger effect sizes (g = 0.64). Nonetheless, it is important to acknowledge that group differences did (non-significantly) favor the MBI intervention over control conditions post-intervention, but not at follow-up. Therefore, it is possible that MBIs might contribute to short-term improvements in HRV in some instances, but more research would be needed to understand this issue, and specifically when and for whom MBIs might exert salutatory benefits on HRV.

While we limited our analyses to vagally-mediated metrics of HRV including RMSSD, other time-domain measures, as well as HF-HRV and RSA, it is important to note that these measures, while highly inter-correlated, are not synonymous with vagal tone. Numerous factors can confound the association between these measures and underlying vagal tone, including interindividual differences, physical activity and, importantly, respiration (16). A key caveat in the methodological design of the included studies that only three of the 19 studies (>16%) controlled for respiration rate (24, 44, 55). While we mitigated this issue through adopting RMSSD as the primary HRV metric used when available, which is less affected by respiration than other measures such as RSA, RMSSD is nonetheless influenced by respiration (35). This is problematic, as respiration rate changes can influence HRV metrics without affecting vagal tone (35). We excluded studies of paced or controlled breathing interventions for this reason, however it is possible that mindfulness-based practices with natural breathing may nonetheless lead to changes in respiration rate that could subsequently influence HRV, independent of any underlying changes in vagal tone and associated influence of the parasympathetic nervous system. Thus, we recommend that future studies control for respiration rate, in accordance with current recommendations (66). The predictive value of HRV may be improved when controlling for respiration, so it is possible that this methodological consideration could help identify any true effect of MBIs on HRV (67).

While we found no evidence of a dose-response, it should be noted that this review of RCTs does not include coverage of expert or longer-term meditators, a group wherein physiological changes associated with mindfulness might be more pronounced. It is difficult to conduct RCTs investigating very high doses of mindfulness, due to time and resource limitations, and non-randomized studies of expert meditators run the risk of confounding factors influencing results (68). Nonetheless, group comparison studies (controlling for differences in demographics and other factors known to influence HRV) of HRV indices in expert versus novice meditators could be helpful to shed light on this question, and help ascertain the extent to which mindfulness might influence HRV at very high doses, and thus help determine if a true effect of mindfulness on HRV exists, while acknowledging that there are some lifestyle factors that may be incredibly difficult to control (68).

The American Heart Association has recently recommended meditation as an adjunct treatment to reduce cardiovascular risk (69). HRV is an index of cardiac health, so it is important to consider the extent to which it might specifically be improved through meditation training. The results of this meta-analysis show that there is currently insufficient evidence to indicate that meditation improves HRV, and further research is needed. In contrast, there is stronger evidence that physical interventions such as exercise (32) and mindful movement-based practices such as tai chi and yoga (70) lead to salutary HRV changes. Therefore, physical activity-based interventions may offer a more plausible pathway to improved HRV than seated meditation practice.

To our knowledge, only one prior study has conducted a meta-analysis of MBIs and HRV, and it included data from only four studies (27). In this small earlier synthesis, authors found no evidence that MBIs were associated with improvements in resting state HRV relative to control conditions. We expand on this finding to include a larger sample of studies . In agreement with Radmark and colleagues (27), we also found insufficient evidence that meditation training is associated with higher resting-state HRV relative to control conditions. HRV may not be a key mechanism that helps explain the wide-ranging health benefits of meditation practice, and so future small trials of MBIs should not prioritize including HRV as an outcome measure. Only large, well-designed RCTs of high quality MBIs (e.g. gold standard or other rigorously developed interventions) designed with low risk of methodological bias could help add to the current evidence to elucidate any role MBIs might play in improving HRV.

Glossary

HRV

heart rate variability

HF-HRV

high-frequency heart rate variability

MBI

mindfulness based intervention

RSA

respiratory sinus arrhythmia