By Guy William Fincham, Clara Strauss, Jesus Montero-Marin & Kate Cavanagh, Scientific Reports, http://www.nature.com 

Breathwork comprises various practices which encompass regulating the way that one breathes, particularly in order to promote mental, emotional and physical health (Oxford English Dictionary). These techniques have emerged worldwide with complex historical roots from various traditions such as yoga (i.e., alternate nostril breathing) and Tibetan Buddhism (i.e., vase breathing) along with psychedelic communities (i.e., conscious connected breathing) and scientific/medical researchers and practitioners (i.e., coherent/resonant frequency breathing). Recently, breathwork has been garnering public attention and popularity in the West due to supposed beneficial effects on health and well-being in addition to the breathing-related pathology of covid-19, however it has only been partly investigated by clinical research and psychiatric medical communities.

Slow-paced breathing practices have gained most research attention thus far. Several psychophysiological mechanisms of action are proposed to underpin such techniques: from polyvagal theory and interoception literature along with enteroception, central nervous system effects, and increasing heart-rate variability (HRV) via modulation of the autonomic nervous system (ANS) and increased parasympathetic activity. ANS activity can be measured using HRV, the oscillations in heart rate connected to breathing (i.e., the fluctuation in the interval between successive heart beats). Fundamentally, as one inhales and exhales, heart rate increases and decreases, respectively. Higher HRV, arising from respiratory sinus arrhythmia, is typically beneficial as it translates into robust responses to changes in breathing and thus a more resilient stress-response system.

Stress-response dysfunction, associated with impaired ANS activity, and low HRV are common in stress, anxiety, and depression. This may explain why techniques like HRV biofeedback can be helpful, however, it is possible that simply pacing respiration slowly at approximately 5–6 breaths/minute, requiring no monitoring equipment, can elicit similar effects. Polyvagal Theory, for instance, posits that vagal nerves are major channels for bidirectional communication between body and brain. Bodily feedback has profound effects on mental states as 80% of vagus nerve fibres transmit messages from body to brain. Further, the neurovisceral integration model states that high vagal tone is associated with improved health along with emotional and cognitive functioning. Vagal nerves form the main pathway of the parasympathetic nervous system, and high HRV indicates greater parasympathetic activity.

Modifying breathing alters communication sent from the respiratory system, rapidly influencing brain regions regulating behaviour, thought and emotion. Likewise, respiration may entrain brain electrical activity, with slow breathing resulting in synchrony of brain waves, thereby enabling diverse brain regions to communicate more effectively. It has been observed that adept long-term Buddhist meditation practitioners can achieve states where brain waves are synchronised continuously.

Breathwork and stress

Stress, anxiety and depression have markedly exceeded pre-covid-19 pandemic population norms. Thus, research is needed to address how this can be mitigated. A recent survey based on more than 150,000 interviews in over 100 countries suggested that 40% of adults had experienced stress the day preceding the survey (Gallup, US). Prior to the pandemic, mental health difficulties were already a significant issue. For instance, stress has been identified by the World Health Organisation as contributing to several non-communicable diseases and a 2014 survey, led in collaboration with Harvard, of over 115 million adults showed that 72% and 60% frequently experienced financial and occupational stress, respectively (Robert Wood Johnson Foundation, US).

Chronic stress is associated with, and can significantly contribute to, many physical and mental health conditions, from hypertension and cardiovascular disease to anxiety and depression. For common mental health problems such as anxiety and depression, cognitive behavioural therapy (CBT) is widely recommended in treatment guidelines worldwide, yet many do not recover and waiting times can be long, in addition to extensive professional training and ongoing supervision being required for therapists. Moreover, such treatment is typically individualised and offered on a one-to-one basis making it resource intensive. The present state of global mental health coupled with the access barriers to psychological therapies requires interventions that are easily accessible and scalable7, and manualised practices such as breathwork may meet this remit.

Breathing exercises can be easily taught to both trainers and practitioners, and learned in group settings, increasingly via synchronous and asynchronous methods remotely/online. Therefore, given the need for effective treatments that can be offered at scale with limited resources, interventions focusing on deliberately changing breathing might have significant potential. Indeed, some government public health platforms already recommend deep breathing for stress, anxiety and panic symptoms (NHS and IAPT, UK). However, the evidence underlying this recommendation has not been scrutinised in a comprehensive systematic review and meta-analysis and this is the aim of the current study.

Moreover, it is not only slow-paced breathing which may help reduce stress. Fast-paced breathwork may also offer therapeutic benefit as temporary voluntarily induced stress is also known to be beneficial for health and stress resilience. For example, regular physical exercise can improve stress, anxiety and depression levels, along with HRV. Similarly, fast-paced breathing techniques can induce short-term stress that may improve mental health, and have also been shown to volitionally influence the ANS, promoting sympathetic activity. There are countless breathwork techniques—and such variation in their potential modalities and underlying principles warrants exploration.

Review aims

It is important that hype around breathwork is grounded in evidence for efficacy—and effects are not overstated to the public. Whilst some previous reviews of breathwork have been published, it is not possible to conclude the effectiveness of breathwork for stress (nor mental health in general) based on previous meta-analyses, since they have been restricted by certain factors. These include focusing on populations with impaired breathing (i.e., chronic obstructive pulmonary disease—COPD, and Asthma), insufficient focus on the breathwork intervention itself (i.e., including interventions where breathwork is combined with several other intervention components) making it hard to elicit separate effects, along with spanning more literature on self-reported/subjective anxiety and depression compared to stress. On the other hand, systematic reviews with narrative syntheses of quantitative data may have overlooked key studies because of too much focus on a specific technique (i.e., slow breathing or diaphragmatic breathing), an absence of randomised-controlled trials (RCTs), scanter literature on self-reported/subjective stress compared to self-reported/subjective symptoms of anxiety and depression, along with limited databases, or exclusion of unpublished studies and grey literature (i.e., theses/dissertations).

Furthermore, in keeping with the participant, intervention, control, outcome and study design (PICOS) framework, there is an absence of examining dose–response correlates with effects and subgroup analyses evaluating differential effects of different breathwork interventions and how they were delivered, what controls were used, effects on populations with differing health statuses and, finally, the psychological outcome measures used. All of these are crucial for an adequate ethical, precautional and practical implementation of breathwork interventions. Accordingly, subgroup analyses were explored to account for these, for the primary outcome of stress. It could be relevant to investigate potential sources of heterogeneity in terms of effects on stress, and this might be related to how some subgroups (such as mental/physical health populations, along with nonclinical/general populations) receive the intervention. Moreover, other subgroups such as the type of breathwork intervention (i.e., slow/fast) and how it is delivered (i.e., online/in-person or individual/group-based), along with the type of comparator (active/inactive control) and outcome measure (questionnaire) used to self-report on stress, may be sources of heterogeneity and thus warrant investigation.

So far, there is no existing meta-analysis of RCTs on the effect of breathwork on psychological stress. Thus, to fill this research gap, the aim of our meta-analysis was to estimate the effect of breathwork in targeting stress. Because prolonged stress can significantly contribute to anxiety and depressive symptoms and there is considerable overlap between them, we included these two common mental health issues as secondary outcomes, to provide a bigger picture and greater context around the findings on stress. The primary outcome was pre-registered as stress since it is a transdiagnostic variable, relevant in a variety of disorders, and also in people without a diagnosis but suffering from high levels of psychological distress. This makes stress a very interesting target for breathwork-based interventions.

In brief, our research question was the following: do breathwork interventions lead to lower self-reported/subjective stress (primary outcome), anxiety, and depression (secondary outcomes) in comparison to non-breathwork control conditions? We propose this work as a first comprehensive systematic review and meta-analysis exploring the effects of breathwork on stress and mental health, to help lay a solid foundation for the field to grow and evolve in an evidence-based manner.

Methods

We focused solely on RCTs reporting psychological measures, to gauge any potential efficacy or effectiveness of breathwork. We also explored sub-analyses for stress outcomes depending on the health status of the study population, technique, and delivery of breathwork, along with types of control groups and stress outcome measures used. Finally, we examined dose–response effects of breathwork on stress.

Pre-registration and search strategy

Our meta-analysis was pre-registered on the international prospective register of systematic reviews PROSPERO (2022 CRD42022296709). Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards were applied throughout. We searched published, unpublished, and grey literature in the following five databases: PsycInfo, PubMed, ProQuest, Scopus, and Web of Science, along with two clinical trial registers: ClinicalTrials.gov and ISRCTN. The search was run up to February 2022 for all seven electronic repositories, with no date restrictions, in line with the search criteria pre-registered on Prospero, including keywords such as: breath*, respir*, random*, RCT, and stress. For purposes of feasibility in conducting the search, we maintained our focus on the pre-registered primary outcome, following Cochrane Collaboration guidelines to meet the highest criteria for self-reported/subjective stress outcomes by searching trial registers for unpublished studies. There is limited search functionality on trial registers and time involved in contacting researchers for trial data. Moreover, as mentioned above, some previous reviews have not searched unpublished, grey literature before and there are less data available on breathwork and self-reported/subjective stress, in comparison to self-reported/subjective anxiety and depression. In brief, given our focus on stress (paired with time and resource constraints), we conducted the most robust search possible for the primary outcome whilst secondary outcomes only included published data—and we were explicit about this from pre-registration onwards.

Inclusion and exclusion criteria

Inclusion criteria were that studies: (1) were published in the English language, (2) included a breathwork intervention where breathwork formed 50% or more of the intervention (and home practice/self-practice, if any), (3) were RCTs, (4) included an outcome measure of self-reported/subjective stress, anxiety, or depression, (5) included an adult participant sample 18 + years of age. For the five databases, studies with abstracts that did not include either the primary outcome keyword (stress), or a secondary outcome keyword (anxiety or depression), were excluded. For the two registers, if it was clear from the summary information that trials did not comprise the primary outcome of stress, they were excluded. As mentioned above, stress is a transdiagnostic health variable, relevant across various (clinical and nonclinical) populations and conditions, hence it was our primary interest. Additional rationale included the fact that there is far more limited research literature available on self-reported/subjective stress and breathwork (as opposed to anxiety and depression) and, since this was the primary outcome, because fewer (published) data were available, and to make the secondary search (which was only used in the present study to contextualise findings) more feasible, we used the referred search strategy, as this allowed us to find more information on stress from unpublished sources.

For all electronic repositories, studies with control conditions that comprised components of breathwork were excluded, except for studies which had time-points wherein data were collected before controls participated in breathwork (i.e., crossover RCTs). Only non-breathwork controls were used as post-intervention comparisons. Studies with interventions that comprised of equipment (oronasal or otherwise) which physically altered and/or assisted breathing activity were excluded. Breathwork was operationalised as techniques which involved conscious and volitional control or manipulation of one’s breath (depth, pattern, speed or otherwise) through deliberate breathing practices. Interventions that affected breathing as a by-product, e.g., mindfulness, singing, and aerobic exercise, were excluded.

Review strategy and study selection

The first author conducted the search and initial screening against eligibility criteria along with full-text screening. Records were then screened, excluding reports based on review of titles and keywords in abstracts or summary information (for trials), or if the inclusion criteria were not met. Remaining reports were sought for retrieval and the full-text reports assessed for eligibility, before final eligibility decisions were made. Further identification of studies comprised forward and backward citation searching via Google Scholar and reference lists, respectively, of the final reports included from the database/registry search. For inter-rater consistency purposes, one of the authors (JMM) checked a random sample (10% of reports) after duplicates had been removed. Furthermore, where GWF was unsure after full-text screening, they consulted authors KC and CS to come to a collective decision on eligibility. Any discrepancies between authors were resolved by discussion and reaching consensus.

Data extraction

Our primary outcome was self-reported/subjective stress. Secondary outcomes were self-reported/subjective anxiety, depression, and global mental health (where two or more of stress, anxiety and depression were combined into a total measure without providing subscale data). We extracted the following data across the studies’ conditions: sample sizes, means, and standard deviations of outcome scores post-intervention (timepoint 1—T1, where T0 is pre-intervention/baseline) along with at latest follow-up where possible (a true follow-up was classed as when participants no longer received any instruction for the breathwork intervention). Where studies involved crossover designs, the midpoints were categorised as post-intervention (before the control group started the breathwork given initially to the intervention group). For studies which required multiple groups’ mean and standard deviation (M ± SD) scores to be combined, or for just SDs to be calculated, these were calculated in accordance with the Cochrane Collaboration handbook. For example, calculating SDs from Ms and 95% confidence intervals (CIs) or combining multiple groups’ M ± SD scores if two or more groups completed an intervention that involved breathwork (but the study still comprised a non-breathwork control).

Risk of bias and quality assessment

The most recent, revised Cochrane Collaboration’s tool for assessing risk of bias in randomised trials (RoB 2) was used for analysing studies on the primary outcome measure of self-reported/subjective stress. The studies were analysed across the following five domains for the stress outcomes: randomisation process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each domain produced an algorithmic judgement of “low risk of bias”, “some concerns”, or “high risk of bias”, resulting in an overall risk of bias judgement. For further inter-rater consistency purposes, both JMM and GWF completed bias scoring using RoB 2 on all included studies for stress, with any discrepancies resolved via discussion.

Data synthesis and analysis

To evaluate whether breathwork can effectively lower stress compared to non-breathwork controls and to quantify the estimation we ran a quantitative synthesis meta-analysis using standardised mean differences and a random-effects model. This used aggregate participant data of M ± SD scores on stress outcome measures for intervention and control conditions of each study at post-intervention (T1), along with the groups’ sample sizes. We also conducted a sensitivity analysis by removing one study at a time, to evaluate the robustness of effects. Separate random-effects meta-analyses were run for the secondary outcomes. The software Review Manager (RevMan) version 5.4 was used. For the between-group effect sizes (ESs) we computed Hedges’ g, based on the standardised between-group difference at post-intervention considering sampling variance among groups; an ES of 0.2 is classed as small, 0.5 medium and 0.8 large. For each separate outcome, the ESs were calculated via comparison of post-breathwork intervention scores between the conditions. Intention-to-treat data were chosen over per-protocol data where available, since the former provides a more conservative estimate of between-group differences.

Heterogeneity of ESs variance was assessed using Cochran’s Q based on a chi-square distribution (χ2) and Higgins’ I2. If χ2 is significant and an I2 index value is around 50%, this implies variance may be explained by variables other than breathwork and such statistical heterogeneity is moderate, respectively. A funnel plot was produced to examine publication bias for the primary outcome, and the software R (version 4) was used to explore asymmetry of the funnel plot via the Egger’s test (i.e., correlations between standard error and ESs). Moreover, Rosenthal’s fail-safe N was calculated (to estimate how many further studies yielding zero effect would be required to make the overall ES non-significant for stress). Kendall’s tau-b (τB) correlations were used to detect any potential relationships between ESs of breathwork on stress and: estimated total duration of intervention/home practice, total number of intervention/home practice sessions, and intervention/home practice session frequency. If intervention time was not provided by a study (where participants only had home practice), we used the minimum estimated home practice duration (recommended in the study) to gauge the approximate time taken for participants to ‘learn’ the breathwork technique. Minimum recommended duration was used for most conservative estimates, helping account for common attrition found across behavioural studies.

Lastly, subgroup analyses were run for stress, again using a random-effects model. These subsets included: health status of population (physical, nonclinical, or mental health), technique type (fast or slow-paced breathing) and delivery method of the breathwork intervention (individual, group, or a combination of both, and remote (self-help), in-person, or combination) along with the type of control group (active or inactive; in line with Cochrane Collaboration guidelines), and outcome measure used (scale).

Results

The search produced 1325 results: 1175 and 150 records from databases and registers, respectively. After duplicates were removed, the titles and abstracts (or summary information for registers) of 679 records were screened. During screening, the eligibility of 11% of reports were decided collectively among GWF, KC, and CS. All studies included by GWF were checked by KC and CS to ensure none were incorrectly included. One particular study57 that comprised a global mental health measure only had to be excluded as there were insufficient studies to reliably interpret results (n < 5)—the only other available was Goldstein et al. (which also included a measure of self-reported/subjective stress). Accordingly, the global mental health secondary outcome was dropped from the analysis.

Further data were required for eight reports; corresponding authors were contacted, and data from four studies were retrieved, but not the remaining half subsequently excluded from the analysis. Thus, a total of 104 reports were screened and 81 were excluded, leaving 23. As a result of citation searching, a further three studies were included. Of the 26 total reports included in the quantitative synthesis meta-analyses, stress comprised 12 studies. Secondary outcomes of self-reported/subjective anxiety and depression comprised of 20 studies, and 18 studies, respectively. for more information on the secondary outcomes.

Summary of findings for stress

In terms of data extraction, all studies provided raw M ± SD scores apart from two, where estimated marginal M ± SDs were given (raw data was requested from corresponding authors but could not be obtained). One study required SDs from Ms and 95% confidence intervals (CIs) provided, both of which were calculated in accordance with Cochrane Collaboration guidelines. Furthermore, another study required two groups’ M ± SD scores (there was one control group and two intervention groups) to be combined and two further studies involved crossover designs (hence data were extracted at the midpoints of each study before controls started the breathwork intervention). Analyses of follow-up scores were not possible for self-reported/subjective stress as there were insufficient studies for results to be reliably interpreted.

The 12 studies included in the meta-analysis for the primary outcome of stress were completed from 2012 to 2021 (seven, or 60%, were conducted from 2020 onwards). Half of these studies were conducted in the US, two in India, one globally, and one each in: Israel, Turkey, and Canada. The average age was 41.7 (± 8.47) and 75% identified as female, since the largest study was for women only. Attrition rates (after the breathwork intervention began) ranged from 3 to 40%. Participant sample sizes ranged from 10 to 150, with the total number of participants analysed being 785. The number of participants randomised to a breathwork intervention or control condition was 417 and 368, respectively. The minimum total estimated durations of an intervention/home practice ranged from 80 to 5625 min.

Half of the studies comprised physical health, five nonclinical, and one mental health samples. Ten and two studies comprised interventions with a primary focus on slow-paced breathing and fast-paced breathing, respectively. Seven were individual-based interventions, four taught to groups, and one a combination of both modes. Half were remote/self-help interventions, five in-person, and one combination. Seven and five studies had inactive and active control groups, respectively. Eight studies used the perceived stress scale (PSS), three used the stress subscale from the depression anxiety stress scale (DASS), and one used the perceived stress questionnaire (PSQ).

Risk of bias for stress

Risk of bias scoring for the 12 studies on the primary outcome is reported using RoB 2. Three studies’ overall assessment were algorithmically scored as being at high risk of bias, with domain two (deviations from the intended interventions) contributing to most bias. The remaining nine studies’ overall risk of bias were algorithmically scored as having some concerns. Only one study did not disclose how randomisation was conducted. Most of the domains, from randomisation to selection of the reported result, were scored as having some concerns or low risk of bias. We did not find reported adverse events or lasting bad effects directly attributed to breathwork interventions; four studies (six in total including secondary outcome studies) actively reported on this. Nonetheless, regarding safety and tolerability, a small subgroup of participants in Ravindran et al.’s study focusing on fast-paced breathwork in unipolar and bipolar depression reported side effects such as hot flushes, shortness of breath and/or sweating. However, these participants opted to continue the intervention and no participants dropped out of the breathwork group due to adverse effects.

Breathwork and stress

The random-effects meta-analysis (k = 12) displayed a small-medium but significant post-intervention between-group ES, g = − 0.35 [95% CI − 0.55, − 0.14], z = 3.32, p = 0.0009, denoting breathwork was associated with lower levels of self-reported/subjective stress at post-intervention than controls. There were insufficient studies including follow-up measures for a meta-analysis. Heterogeneity was moderate but non-significant, χ211 = 19, p = 0.06, I2 = 42%. Via removing one individual study at a time, the ES of breathwork on stress ranged from − 0.27 to − 0.39 and remained significant in all cases. Initial visual inspection of the funnel plot suggested some skew due to studies with small samples; however, the Egger’s test was non-significant, z = 0.03, p = 0.947, indicating a low chance of publication bias. Fail-safe N analysis denoted that a further 69 studies yielding zero effect would need to be added to make the overall ES non-significant for stress. On removal of the one potential outlier the ES remained significant but became smaller: − 0.27. On removal of the two studies using estimated marginal M ± SDs, the ES remained significant and became larger: − 0.40.

Subgroup analyses for stress

We conducted five sub-analyses for the primary outcome self-reported/subjective stress. There were no significant differential effects between subgroups.

There was a significant effect of breathwork on stress in nonclinical samples, but not in mental (only one study) or physical health populations. Moreover, significant effects were yielded when breathwork was primarily focused on slow-paced breathing (but not for fast-paced breathing), taught to individuals alone, and when taught to groups (but not in combination, which comprised only one study). There were also significant effects of breathwork on stress when the intervention was taught remotely, in-person, and using a combination of these two delivery methods. Significant effects existed for both active and inactive control groups. There were significant effects for studies which used PSS and DASS measures (but not the PSQ, used by only one study).

Heterogeneity was high for studies with physical health samples, slow-paced breathwork, when breathwork was taught to groups and in-person, plus those studies with inactive controls, and when stress was measured by using the DASS, suggesting potential moderating factors that were not accounted for by the subgroup analyses. There was no significant correlation between estimated total duration of breathwork intervention/home practice and ES (n = 12) τB = − 0.05, p = 0.418, number of intervention/home practice sessions and ES for stress (n = 12) τB = − 0.28, p = 0.107, nor for intervention/home practice session frequency and ES (n = 12) τB = − 0.17, p = 0.224.

Breathwork and secondary outcomes

In terms of data extraction, one study had a measure with positively scored anxiety and depression subscales; accordingly, we subtracted the subscale score from the maximum score to reverse the polarity of the measure without changing the magnitude of difference. Another study required two groups’ M ± SD scores to be combined. Analysis of follow-up scores were not possible for secondary outcomes as there were insufficient studies (n < 5). Random-effects analysis for anxiety (k = 20) showed a significant small-medium between-group ES in favour of breathwork, g = − 0.32 [95% CI − 0.48, − 0.16], z = 3.90, p < 0.0001, with moderate and significant heterogeneity, χ219 = 38.62, p = 0.005, I2 = 51%. Sensitivity analysis showed ESs ranging from − 0.29 to − 0.34, significant in all cases. No individual study was responsible for the significant heterogeneity. Random-effects analysis for depression (k = 18) displayed a significant small-medium ES in favour of breathwork, g = − 0.40 [95% CI − 0.58, − 0.22], z = 4.27, p < 0.0001, and heterogeneity was moderate and significant, χ217 = 40.5, p = 0.001, I2 = 58%. Sensitivity analysis showed ESs ranging from − 0.35 to − 0.44, significant in all cases. On removal of two potential outliers, the ES remained the same. No single study was responsible for the significant heterogeneity.

Keywords; breathwork, stress