Yoga

Yoga

Yoga is sold as a single practice: stretch, breathe, calm the nervous system, and live longer. The literature does not keep one practice. It keeps postures, breath drills, sitting meditation, social classes, and heat — tested as packages against usual care more often than against exercise. This article forces each outcome to keep the object that produced it. It is a research review. It is not medical advice.

Compiled by South Beach Longevity · 20 August 2026 Copyright 2026 Series SBL-41 / SP-YOGA · Register A scientific article Sources peer-reviewed trials, systematic reviews, labelled mechanistic work, practice surveys, and one labelled institutional commentary · verified NCBI records Constraint This document describes published research. It is not medical advice. No human use, dose, route or schedule is recommended anywhere in this document.

How to read this document Every finding is labelled, in the sentence that reports it, by the kind of study that produced it. A wait-list change is not a yoga-specific effect. A sit-and-reach gain is not health. An HRV shift after slow breathing is not a treatment for hypertension. Amounts and durations appear only as reported study parameters, always with the population attached. Nothing here is a recommendation.

Five grades are used in the matrices: STRONG, MODERATE, LIMITED, INCONSISTENT, NOT SUPPORTED. They grade the reviewed record in hand. Sibling title Mobility-and-Flexibility owns stretch architecture; this title owns the mixed cultural-clinical object called yoga.

01 What is being named

A yoga class can be physical activity, a breath protocol, a mindfulness protocol, a heated room, or all of those at once. Cramer, Lauche, and Dobos counted 312 yoga randomised trials through February 2014: 78.2 percent used postures, 74.4 percent used breath control, 49.0 percent used meditation, 10.3 percent used philosophy lectures, and 38.1 percent did not even name a style. The median sample was 59 (Cramer, Lauche, and Dobos, 2014). That mix is the object under test. It is strongly supported as a description of the trial literature. It is not a licence to treat “yoga” as a single mechanism.

This article therefore keeps four working objects, plus heat as a fifth environmental load.

When the comparator is a book, a wait-list, or usual care, the signal is structured attended activity plus attention versus little. When the comparator is stretching, physical therapy, or aerobic exercise, yoga-specific residue is the question — and it usually shrinks (Sherman, Cherkin, Wellman et al., 2011; Anheyer, Haller, Lauche, Dobos, and Cramer, 2022; Chu, Gotink, Yeh, Goldie, and Hunink, 2016). That subtraction is the title’s through-line. A package can still move a Roland-Morris score. It cannot, on that fact alone, be cashed as a unique autonomic, detox, or flexibility-as-health mechanism. The later matrices keep the comparator in the grade.

02 Historical origins

South Asian yoga is a family of soteriological, philosophical, and bodily practices whose modern postural form is not a stable premodern medical system. Alter’s indexed historical work treats twentieth-century Indian yoga as public-health and nationalist reconstruction, not as a physiology paper (Alter, 1997). Telles is entitled to put modern measurements next to older texts; the texts do not become randomised trials by being cited (Telles, 2013). Singleton’s Yoga Body and De Michelis’s modern-yoga historiography are books, not Medline articles; this review returned no PubMed records for those names. Lineage prestige is not an effect modifier in the metas. They already pool Iyengar, Hatha, and Viniyoga.

The WHO commentary of 26 June 2026 narrates “more than two millennia” and then cites Saper and a 2012 mechanism hypothesis (World Health Organization, 2026). That is advocacy for an international day. It is not a trial.

03 Modern forms used in trials

The names that appear in the reviewed record are Iyengar (props, long holds — Sherman, Tiedemann, SAGE), Viniyoga (Tilbrook), Hatha as a catch-all, “yoga breathing,” Power yoga in a German injury survey, and one Bikram fitness trial in healthy young adults (Tracy and Hart, 2013). Cramer’s 2014 inventory already recorded 46 named styles plus a large unnamed remainder (Cramer, Lauche, and Dobos, 2014). Style is a weak sorting key. Dose, standing-balance demand, comparator, and whether breath and meditation were present do more work.

Bikram is a heated room plus a fixed 26-posture series. Tracy and Hart reported fitness changes after Bikram training in healthy young adults (Tracy and Hart, 2013). That is a heat-plus-posture experiment in a selected sample. It is not a detox trial and not a falls trial.

04 Asana as loaded isometric and ROM exposure

A held posture is an isometric at a chosen joint angle, often near end range, with a balance demand if the base is small. Tran, Holly, Lashbrook, and Amsterdam measured health-related fitness after Hatha practice and treated the class as a mixed aerobic-flexibility-strength exposure, not as a stretch-only drill (Tran, Holly, Lashbrook, and Amsterdam, 2001). Tracy and Hart treated yoga as steadiness training and measured motor variability, not muscle architecture (Tracy and Hart, 2008). Sibling physiology still applies: ordinary stretching changes stretch tolerance more than durable tissue length (Folpp, Deall, Harvey, and Gwinn, 2006; Weppler and Magnusson, 2010). A yoga sit-and-reach gain is therefore limited as mobility evidence and not supported as proof that fascia lengthened.

05 Pranayama as a respiratory-behaviour protocol

Named breath drills can change rate, tidal volume, and, in laboratory sessions, heart-rate variability. Raghuraj and Telles reported acute autonomic shifts after alternate-nostril breathing (Raghuraj and Telles, 2003). Nivethitha, Mooventhan, and Manjunath reviewed yoga and HRV and found a literature of small, often single-session studies (Nivethitha, Mooventhan, and Manjunath, 2017). Jerath, Edry, Barnes, and Jerath offered a stretch-receptor account of slow breathing that does not require a Sanskrit label (Jerath, Edry, Barnes, and Jerath, 2006). Streeter, Gerbarg, Saper, Ciraulo, and Brown published a mechanism hypothesis linking yoga, GABA, and the autonomic nervous system (Streeter, Gerbarg, Saper, Ciraulo, and Brown, 2012). Hypotheses are not cashed clinical mechanisms.

Cramer, Haller, Klose, Ward, Chung, and Lauche found COPD effects concentrated in breathing-focused programmes rather than posture programmes (Cramer, Haller, Klose et al., 2019). That is a reminder that “yoga” can mean a respiratory intervention. It is not a licence to spend a COPD breath trial as a back-pain result.

06 Meditation and mindfulness components

About half of yoga RCTs include a meditation block (Cramer, Lauche, and Dobos, 2014). van der Kolk, Stone, West and colleagues tested a trauma-informed package, not isolated asana, against women’s health education (van der Kolk, Stone, West et al., 2014). Telles, Raghavendra, Naveen and colleagues measured attention after yoga meditation in a laboratory design (Telles, Raghavendra, Naveen et al., 2013). Neither paper is MBSR, and neither blinds the participant. Sherman, Eaves, Ritenbaugh and colleagues showed that “expect” is an unstable word in complementary-medicine pain trials: hope and realistic expectancy had to be paired before the word meant anything usable (Sherman, Eaves, Ritenbaugh et al., 2014).

07 What a yoga trial actually delivered

The load-bearing programmes in the reviewed record are multimodal. Sherman’s 2005 and 2011 back-pain classes, Tilbrook’s UK programme, and Saper’s noninferiority trial all combined postures with breath, attention, and a class (Sherman, Cherkin, Erro, Miglioretti, and Deyo, 2005; Sherman, Cherkin, Wellman et al., 2011; Tilbrook, Cox, Hewitt et al., 2011; Saper, Lemaster, Delitto et al., 2017). Hagins, States, Selfe, and Innes found blood-pressure reductions only when postures, meditation, and breathing were all present (Hagins, States, Selfe, and Innes, 2013). No prospectively registered factorial in the reviewed record isolates asana versus pranayama versus meditation versus social contact. The working thesis’s falsifier remains unmet. Cramer’s 2014 inventory already said the median trial was small; a small multimodal class versus a book is not a mechanism study, and it is not improved by giving the package a Sanskrit name (Cramer, Lauche, and Dobos, 2014).

08 Movement and stretch tolerance

Amin and Goodman reported sit-and-reach gains after a six-week Iyengar pilot in 16 women with no control arm (Amin and Goodman, 2014). Polsgrove, Eggleston, and Lockyer found sit-reach and shoulder-flexibility gains in 26 male college athletes against sport training only (Polsgrove, Eggleston, and Lockyer, 2016). Gothe and McAuley randomised 118 sedentary adults, mean age 62, to eight weeks of yoga or stretching–strengthening; both groups improved flexibility, and yoga was “just as effective” (Gothe and McAuley, 2016). Sit-and-reach is a hamstring-biased field test, not a whole-body diagnosis (Mayorga-Vega, Merino-Marban, and Viciana, 2014). Grade for yoga raising sit-and-reach in small or uncontrolled samples: limited. Grade for yoga as a unique mobility method: not supported.

09 Strength and isometric demand

Chair stands and arm curls moved in both arms of Gothe and McAuley’s trial (Gothe and McAuley, 2016). Tracy and Hart measured steadiness, not one-repetition maximum (Tracy and Hart, 2008). Tran’s Hatha fitness study is an uncontrolled health-related battery (Tran, Holly, Lashbrook, and Amsterdam, 2001). Yoga can be a low-to-moderate isometric exposure. It is not supported as a unique strength method against a stretching–strengthening class.

10 Balance and postural control

Tiedemann, O’Rourke, Sesto, and Sherrington’s 12-week Iyengar pilot in 54 community-dwelling older people improved standing balance (+1.52 s), sit-to-stand (−3.43 s), and 4-m walk (−0.50 s) against an education booklet, with blinded assessors and no serious adverse events (Tiedemann, O’Rourke, Sesto, and Sherrington, 2013). Youkhana, Dean, Wolff, Sherrington, and Tiedemann pooled six trials (N = 307, age 60+) and reported Hedges’ g = 0.40 for balance and 0.50 for mobility; they said translation to falls was unknown (Youkhana, Dean, Wolff, Sherrington, and Tiedemann, 2016). Those test gains are moderate. They are not fall prevention.

11 Respiratory mechanics

Slow breathing changes ventilation and can change carbon-dioxide tension. Jerath’s stretch-receptor paper is a mechanism sketch (Jerath, Edry, Barnes, and Jerath, 2006). The COPD review is the clinical neighbour: breathing-focused yoga programmes, not posture programmes, carried the signal (Cramer, Haller, Klose et al., 2019). Respiratory rate is an uncontrolled covariate in most HRV papers (Tyagi and Cohen, 2016).

12 Autonomic and HPA claims

Tyagi and Cohen reviewed 59 HRV studies (2,358 participants). Most were small, Indian, single-session laboratory studies in healthy men; only 15 were RCTs, six with Jadad = 3. The authors called firm conclusions premature and demanded respiration reporting (Tyagi and Cohen, 2016). Zou, Sasaki, Wei and colleagues pooled Tai Chi and yoga for HRV and perceived stress — the contrast is mind–body exercise versus not, not yoga versus paced breathing (Zou, Sasaki, Wei et al., 2018). Pascoe, Thompson, Jenkins, and Ski reviewed yoga and mindfulness-based stress reduction against stress-related physiological measures; the review is a neighbouring stress literature, not a cashed vagal mechanism for clinical endpoints (Pascoe, Thompson, Jenkins, and Ski, 2017).

Streeter, Whitfield, Owen and colleagues randomised 34 people to yoga or metabolically matched walking (19 versus 15). Yoga produced larger mood and anxiety-score changes; thalamic GABA correlated with mood in the yoga arm (Streeter, Whitfield, Owen et al., 2010). That is a small, interesting laboratory RCT. It does not explain Roland-Morris changes in back-pain programmes.

Vague “parasympathetic tone” language is not supported as a cashed mechanism. Acute breath protocols can move HRV in small lab samples (limited). Those lab signals do not licence yoga as vagal medicine.

13 Stress-response language

Pascoe’s review and Zou’s pooled Tai Chi/yoga paper are the reviewed record’s stress-physiology neighbours (Pascoe, Thompson, Jenkins, and Ski, 2017; Zou, Sasaki, Wei et al., 2018). Cortisol, HRV, and perceived-stress scales move in some packages against rest. They do not isolate yoga from slower breathing, expectancy, or fitness. Spending them as the reason blood pressure or depression scores moved is a category error.

14 Flexibility and mobility

The flexibility row is thin and already answered by an active control. Gothe and McAuley’s stretch–strength arm erased a yoga-only claim (Gothe and McAuley, 2016). Amin’s Iyengar pilot had no control (Amin and Goodman, 2014). Flexibility gains are limited. Equating them with health is not supported (see §36).

15 Balance

Test-score evidence is moderate in older samples (Tiedemann, O’Rourke, Sesto, and Sherrington, 2013; Youkhana, Dean, Wolff, Sherrington, and Tiedemann, 2016). Sivaramakrishnan, Fitzsimons, Kelly and colleagues compared yoga with active and inactive controls on physical function and health-related quality of life in older adults; the review is the broader function neighbour, not a falls licence (Sivaramakrishnan, Fitzsimons, Kelly et al., 2019).

16 Strength

Limited as a yoga-specific outcome. Equal to stretching–strengthening where both were tested (Gothe and McAuley, 2016).

17 Chronic low-back pain

This is the strongest clinical file in the reviewed record, and it is also the cleanest specificity failure.

Sherman, Cherkin, Erro, Miglioretti, and Deyo randomised 101 adults. At 12 weeks, yoga improved back-specific function versus a self-care book (−3.4 Roland-Morris points) and showed a smaller, less stable advantage versus exercise classes (−1.8) (Sherman, Cherkin, Erro, Miglioretti, and Deyo, 2005). The later YES trial randomised 228 adults (92 yoga, 91 stretching, 45 book). Yoga beat the book (function −2.5 at 12 weeks) and was not superior to stretching at any time (Sherman, Cherkin, Wellman et al., 2011). That is the load-bearing specificity paper.

Tilbrook, Cox, Hewitt and colleagues randomised 313 UK adults to a yoga programme or usual care plus a booklet. Roland-Morris improved −2.17 at 3 months, −1.48 at 6 months, and −1.57 at 12 months; pain scores were similar; adverse events were 12 in 156 yoga participants versus 2 in 157 usual-care participants, mostly more pain (Tilbrook, Cox, Hewitt et al., 2011). Saper, Lemaster, Delitto and colleagues ran a noninferiority trial in 320 underserved adults: yoga was noninferior to physical therapy on function and pain and not superior to education (Saper, Lemaster, Delitto et al., 2017).

Wieland, Skoetz, Pilkington and colleagues’ Cochrane review (21 trials, N = 2,223) found small, clinically unimportant function (−1.69 Roland-Morris) and pain (−4.53 / 100) changes versus no exercise, little or no difference versus back-focused exercise, and high risk of performance and detection bias in every trial (Wieland, Skoetz, Pilkington et al., 2022). Anheyer, Haller, Lauche, Dobos, and Cramer (27 studies, N = 2,702) found pain and disability improvements versus passive comparators and no significant short- or long-term differences versus active comparators (Anheyer, Haller, Lauche, Dobos, and Cramer, 2022).

Grade versus a book or usual care: moderate (real, small). Grade as yoga-specific versus stretching, physical therapy, or other active care: not supported.

18 Arthritis

Cramer, Lauche, Langhorst, and Dobos reviewed eight RCTs (N = 559) across fibromyalgia, osteoarthritis, rheumatoid arthritis, and carpal-tunnel syndrome and found very-low evidence for pain and disability, with no usable safety data (Cramer, Lauche, Langhorst, and Dobos, 2013). Lauche, Hunter, Adams, and Cramer later pooled nine knee-osteoarthritis trials (N = 640): pain, function, and stiffness improved versus exercise and non-exercise controls, but effects were not robust to bias; hand osteoarthritis did not move; safety was rarely reported (Lauche, Hunter, Adams, and Cramer, 2019). Moonaz, Bingham, Wissow, and Bartlett randomised sedentary adults with arthritis; the trial is a feasibility and symptom-signal paper, not a disease-modifying claim (Moonaz, Bingham, Wissow, and Bartlett, 2015). Ward, Stebbings, Athens, Cherkin, and Baxter’s rheumatoid-arthritis pilot (N = 26) was feasible, with minor adverse events and no group effect on pain or sleep (Ward, Stebbings, Athens, Cherkin, and Baxter, 2018). Grade: limited.

19 Blood pressure

Hagins, States, Selfe, and Innes pooled 17 controlled studies: systolic −4.17 mm Hg and diastolic −3.62 mm Hg overall; larger when postures, meditation, and breathing were all present; not versus exercise; bias high or unclear in all (Hagins, States, Selfe, and Innes, 2013). Cramer, Haller, Lauche and colleagues (7 RCTs, N = 452) reported systolic −9.65 and diastolic −7.22 mm Hg versus usual care (very low quality, I² 90–92 percent), more adverse events than usual care, and no effect versus exercise (Cramer, Haller, Lauche et al., 2014). Hagins, Rundle, Consedine, and Khalsa compared yoga with nonaerobic exercise using 24-hour ambulatory blood pressure (84 enrolled, 68 completers): within-yoga 24-hour diastolic −3.93 mm Hg; between-group only night diastolic was significant (Hagins, Rundle, Consedine, and Khalsa, 2014). Geiger, Cramer, Anheyer, Dobos, and Kohl-Heckl updated the file (30 RCTs, N = 2,283): versus wait-list, systolic −7.95 and diastolic −4.93 mm Hg (very low quality); versus active controls, systolic not significant and diastolic −1.88 mm Hg (Geiger, Cramer, Anheyer, Dobos, and Kohl-Heckl, 2025).

Grade versus inactivity: limited. Grade as a yoga-specific autonomic mechanism versus exercise: not supported.

20 Cardiometabolic outcomes

Cramer, Lauche, Haller and colleagues pooled 44 RCTs (N = 3,168). Versus usual care, systolic pressure, lipids, and HbA1c (−0.45 percent) moved; versus exercise, only HDL differed (Cramer, Lauche, Haller et al., 2014). Chu, Gotink, Yeh, Goldie, and Hunink (37 RCTs) found body-mass index, systolic pressure, LDL, and HDL improved versus non-exercise controls and no significant difference versus exercise (Chu, Gotink, Yeh, Goldie, and Hunink, 2016). Cramer, Langhorst, Dobos, and Lauche found no effect on metabolic-syndrome resolution, diastolic pressure, triglycerides, HDL, or glucose; waist and systolic pressure improved but were not robust to selection bias (Cramer, Langhorst, Dobos, and Lauche, 2016). Innes and Selfe reviewed 25 controlled type-2-diabetes trials (N = 2,170), 13 of them non-randomised, and reported frequent glycaemic and lipid improvements with limited methods (Innes and Selfe, 2016). Cramer, Lauche, Haller, Dobos, and Michalsen found very-low evidence in heart-disease trials and no mortality effect (Cramer, Lauche, Haller, Dobos, and Michalsen, 2015).

Grade versus inactivity: limited to moderate for some risk-factor scores. Grade as yoga-specific versus exercise: not supported. Syndrome resolution: not supported.

21 Anxiety

Cramer, Lauche, Anheyer and colleagues (8 RCTs, N = 319) found SMD −0.43 versus no treatment for elevated anxiety scores and no effect in DSM-diagnosed anxiety disorders (Cramer, Lauche, Anheyer et al., 2018). Kirkwood, Rampes, Tuffrey, Richardson, and Pilkington’s earlier review of eight studies already refused a general-efficacy sentence (Kirkwood, Rampes, Tuffrey, Richardson, and Pilkington, 2005). Grade for scores versus nothing: limited. Grade for diagnosed anxiety disorders: not supported.

22 Depression

Cramer, Lauche, Langhorst, and Dobos (12 RCTs, N = 619) reported short-term SMDs of −0.69 versus usual care, −0.62 versus relaxation, and −0.59 versus aerobic exercise, with no long-term meta-analysis and no safety data (Cramer, Lauche, Langhorst, and Dobos, 2013). In major depressive disorder, Cramer, Anheyer, Lauche, and Dobos (7 RCTs, N = 240) found no difference versus aerobic exercise, worse short-term severity versus ECT, and an unclear risk–benefit (Cramer, Anheyer, Lauche, and Dobos, 2017). Moosburner, Cramer, Bilc, Triana, and Anheyer updated the depressive-disorder file (24 studies): severity SMD −0.43 versus passive controls and not significant versus active controls (−0.22); remission odds ratios favoured yoga against both, with GRADE moderate to very low (Moosburner, Cramer, Bilc, Triana, and Anheyer, 2024).

Uebelacker, Tremont, Gillette and colleagues ran the honest attention-controlled design: 63 yoga versus 59 health-education participants with persistent major depression. The primary 10-week endpoint was null. Over full follow-up, yoga scores were lower (b = −1.38); six-month response was 51 percent versus 31 percent (Uebelacker, Tremont, Gillette et al., 2017). That delayed signal is not nothing. It is also not a primary-endpoint win.

Grade versus passive controls: limited to moderate. Grade versus attention or aerobic exercise: inconsistent, often not supported.

23 Sleep

Khalsa’s 2004 diary study (20 completers, no control) improved sleep-efficiency and related diary metrics (Khalsa, 2004). Khalsa and Goldstein later randomised 20 per arm against sleep hygiene: yoga beat hygiene on total sleep time (d = 0.95), efficiency (d = 1.36), and sleep-onset latency (d = −1.16) on self-report, with no pre-sleep-arousal change and no actigraphy (Khalsa and Goldstein, 2021). Wang, Chen, Dai, Qin, and Wang pooled yoga for sleep quality and insomnia in women with sleep problems (Wang, Chen, Dai, Qin, and Wang, 2020). Cramer’s breast-cancer Cochrane review found sleep SMD −0.25 versus no therapy (Cramer, Lauche, Klose et al., 2017). Grade: limited. Diary expectancy is the standing confounder.

24 Quality of life

In breast-cancer samples, health-related quality of life SMD was 0.22 versus no therapy and −0.04 (not significant) versus exercise (Cramer, Lauche, Klose et al., 2017). In low-back-pain samples, Wieland reported small SF-36 physical (+1.80) and mental (+2.38) changes versus no exercise (Wieland, Skoetz, Pilkington et al., 2022). Grade versus no therapy: moderate in cancer supportive-care samples, limited in back pain. Grade versus exercise: not supported.

25 Older adults and fall risk

Balance and mobility tests can move (moderate). Fall events are a different endpoint. Oliveira, Sherrington, Lord and colleagues’ SAGE trial randomised 700 people (mean age 67) to 12 months of twice-weekly Iyengar-based standing yoga or seated relaxation yoga. The standing programme produced a higher fall rate (0.87 versus 0.64 per person-year; incidence-rate ratio 1.33, 1.01–1.75), six musculoskeletal adverse events, and no serious adverse events. Planned activity and balance confidence rose. The authors said this programme should not be recommended for fall prevention in its current form (Oliveira, Sherrington, Lord et al., 2025). Youkhana had already limited the claim to tests (Youkhana, Dean, Wolff, Sherrington, and Tiedemann, 2016). Swain and McGwin found the highest US emergency-department yoga-injury rate at age 65 and over (Swain and McGwin, 2016).

Grade for fall prevention with that Iyengar-based standing programme: not supported. Grade for “yoga for healthy ageing” as a WHO commentary sentence: advocacy, not a trial (World Health Organization, 2026).

26 Cognition in older adults

Gothe, Kramer, and McAuley, in the same eight-week RCT versus stretching–strengthening (N = 118), found task-switching and n-back favoured yoga (Gothe, Kramer, and McAuley, 2014). That is limited to moderate against this active control, not against aerobic training.

27 Clinical-trial matrix

28 Mental-health matrix

Field-level flag: participants and providers are not blinded; outcomes are usually self-assessed (Wieland, Skoetz, Pilkington et al., 2022).

29 Musculoskeletal matrix

Cramer, Lauche, Hohmann and colleagues randomised yoga against home-based exercise for chronic neck pain and found a function signal against that manual (Cramer, Lauche, Hohmann et al., 2013). A home booklet is not a supervised exercise class.

30 Injury patterns

Yoga is not harmless folklore. Typical reported harm is musculoskeletal. Cramer, Ward, Saper, Fishbein, Dobos, and Lauche reviewed 94 RCTs that reported adverse events (8,430 participants, from 301 identified yoga RCTs). Versus usual care or exercise, intervention-related, nonserious, and serious events did not differ. Versus psychological or educational controls, intervention-related events (OR 4.21) and nonserious events (OR 7.30) were higher. Serious events and dropouts for adverse events were comparable (Cramer, Ward, Saper, Fishbein, Dobos, and Lauche, 2015). Wieland’s Cochrane review found yoga raised the risk of (mostly increased) back pain versus no exercise (RR 4.76; 43 versus 9 per 1,000) (Wieland, Skoetz, Pilkington et al., 2022).

Cramer, Quinker, Schumann, Wardle, Dobos, and Lauche’s German online survey (n = 1,702, 88.9 percent female) found 21.4 percent reporting at least one acute adverse event and 10.2 percent a chronic event; 0.60 injuries per 1,000 hours (Power yoga 1.50); 76.9 percent of acute and 51.6 percent of chronic events recovered fully; self-study without supervision raised risk (Cramer, Quinker, Schumann et al., 2019). Swain and McGwin estimated about 29,590 US emergency-department visits for yoga-related injury from 2001 to 2014; trunk 46.6 percent; sprain or strain 45 percent; the rate highest at age 65 and over (Swain and McGwin, 2016). Campo, Shiyko, Kean, Roberts, and Pappas followed 354 recreational practitioners for one year: 10.7 percent incident pain caused by yoga, more than a third lasting beyond three months or costing practice time (Campo, Shiyko, Kean, Roberts, and Pappas, 2018). Penman, Cohen, Stevens, and Jackson’s Australian survey found 2.4 percent supervised yoga-related injuries in 12 months (Penman, Cohen, Stevens, and Jackson, 2012).

Cramer, Haller, Dobos, and Lauche’s dropout meta-analysis (168 RCTs) put yoga-group dropout at 11.42 percent (10.11–12.73), slightly lower than exercise controls (14.53 percent), and higher after 12 weeks (15.23 percent) (Cramer, Haller, Dobos, and Lauche, 2016). Completers are not a proof of safety.

31 Hypermobility and extreme postures

Handstand, shoulderstand, and headstand accounted for 29.4 percent of practices named as associated with acute adverse events in the German survey (Cramer, Quinker, Schumann et al., 2019). That is moderate evidence that stands cluster reported harm. It is not a per-posture incidence.

Beighton, Solomon, and Soskolne’s nine-point screen is a population description, not a flexibility goal (Beighton, Solomon, and Soskolne, 1973). The 2017 international classification of the Ehlers–Danlos syndromes distinguishes hypermobile EDS as a clinical diagnosis, not a yoga achievement (Malfait, Francomano, Byers et al., 2017). Castori, Tinkle, Levy and colleagues offered a framework that separates joint hypermobility as a trait from related syndromes (Castori, Tinkle, Levy et al., 2017). No yoga-versus-GJH randomised trial sits in the reviewed record. More ROM is not universally desirable. That is strong as sibling connective-tissue classification. It is insufficient as a yoga-specific harm trial.

Harvey, Katalinic, and Herbert’s contracture review still applies: stretch for contracture moves a few degrees, not a life (Harvey, Katalinic, and Herbert, 2017). Chasing a universal athletic range in a hypermobile person is the complementary error to ignoring a task-limiting deficit.

32 Special populations

Older adults: highest emergency-department injury rate (Swain and McGwin, 2016) and a standing Iyengar-based programme that increased falls (Oliveira, Sherrington, Lord et al., 2025). Pregnancy: this review holds a prenatal-depression note (Battle, Uebelacker, Howard, and Castaneda, 2010), not an obstetric safety trial. Inversion–glaucoma and inversion–hypertensive-crisis physiology were not resolved as trials. Heat/Bikram has a young-adult fitness paper (Tracy and Hart, 2013), not a heat-illness incidence file. Those rows stay labelled thin rather than filled from clinic folklore.

Cramer, Sibbritt, Adams, and Lauche found no association between regular yoga or meditation and falls in Australian women aged 59–64 (Cramer, Sibbritt, Adams, and Lauche, 2016). A null cross-section is not a safety proof and is not SAGE.

33 Safety matrix

34 Vague parasympathetic claims

The clinical autonomic claim is not cashed. Tyagi and Cohen refused firm HRV conclusions (Tyagi and Cohen, 2016). Zou pooled Tai Chi with yoga (Zou, Sasaki, Wei et al., 2018). Blood-pressure metas lose their advantage against exercise (Cramer, Haller, Lauche et al., 2014; Geiger, Cramer, Anheyer, Dobos, and Kohl-Heckl, 2025). Streeter’s GABA walking-control RCT is n = 34 (Streeter, Whitfield, Owen et al., 2010). Jerath and Streeter 2012 are hypotheses (Jerath, Edry, Barnes, and Jerath, 2006; Streeter, Gerbarg, Saper, Ciraulo, and Brown, 2012). Not supported as vagal medicine. Limited that acute breath protocols move laboratory signals.

35 Detox

No trial, systematic review, or biomarker paper in this store tests toxin clearance, lymphatic “flush,” or sweat-as-depuration. Heat is an environmental load — core temperature, plasma volume, orthostatic stress — not a purification mechanism (Tracy and Hart, 2013, is a fitness paper). Not supported.

36 Flexibility equated with health

Ordinary stretching changes tolerance more than architecture (Folpp, Deall, Harvey, and Gwinn, 2006; Weppler and Magnusson, 2010). Sit-and-reach is not a whole-body diagnosis (Mayorga-Vega, Merino-Marban, and Viciana, 2014). Hypermobility is a trait, sometimes a syndrome, not a wellness target (Beighton, Solomon, and Soskolne, 1973; Malfait, Francomano, Byers et al., 2017). Yoga ROM papers are small or equal to stretch–strength (Amin and Goodman, 2014; Gothe and McAuley, 2016). Not supported as a health identity. Limited that classes can raise sit-and-reach.

37 Blinding, expectancy, and self-selection

Wieland: every low-back-pain trial high risk of performance and detection bias; outcomes self-assessed; no sham-yoga trial (Wieland, Skoetz, Pilkington et al., 2022). Sherman 2014: patients do not share investigators’ meaning of “expect” (Sherman, Eaves, Ritenbaugh et al., 2014). Uebelacker 2017: primary depression endpoint null versus health education (Uebelacker, Tremont, Gillette et al., 2017). Cramer’s PTSD meta: null versus attention (Cramer, Anheyer, Saha, and Dobos, 2018). Cramer, Ward, Steel, Lauche, Dobos, and Zhang described who already practises yoga in a US survey (Cramer, Ward, Steel, Lauche, Dobos, and Zhang, 2016). Penman, Cohen, Stevens, and Jackson’s Australian respondents were typically tertiary-educated employed women who started for fitness and stayed for stress (Penman, Cohen, Stevens, and Jackson, 2012). Observational “yoga users are healthier” is not supported as treatment evidence. Self-selection is the default alternative. Blinding and expectancy are a strong field-level defect, not a footnote.

38 Yoga-specific effects versus generic activity and social contact

Ross and Thomas’s 2010 comparison review is optimistic and pre-dates Sherman 2011, Chu 2016, and Anheyer 2022 (Ross and Thomas, 2010). Those later active-control papers are the subtraction. Hagins 2013 required the full multimodal bundle for a blood-pressure signal — a bundle that includes class time (Hagins, States, Selfe, and Innes, 2013). Residual signals are condition-specific and small. As a general rule, yoga-specific effects after generic physical activity and social interaction are subtracted are not supported.

39 Unresolved science

A large, prospectively registered factorial that isolates asana, pranayama, meditation, and social contact, with expectancy measured and, where possible, blinded outcomes, is still absent. Dedicated inversion–intraocular-pressure and inversion–hypertensive physiology trials were not resolved. A yoga-versus-generalised-hypermobility harm trial is absent. Pregnancy safety as a systematic trial file is absent. Hot-yoga heat-illness incidence is absent. Those are gaps, not invitations to spend clinic folklore.

What the reviewed record does support, and will not over-spend: some manualized low-back-pain programmes improve back-specific function versus a book or usual care, by amounts that often miss predefined minimum important differences and that disappear against stretching or physical therapy; older adults can gain balance-test scores; blood-pressure and cardiometabolic scores can move versus inactivity and usually do not beat exercise; anxiety and depression scores can move versus wait-lists and often do not beat attention or aerobic exercise; harm is real, usually musculoskeletal, and comparable to other exercise, with stands and older age marking reported and emergency-department harm; a large modern standing Iyengar-based programme increased falls.

40 How to read the grades

STRONG means the reviewed record can carry the sentence against a serious objection. MODERATE means a real signal in named samples with a standing confounder. LIMITED means small, uncontrolled, or very-low-quality. INCONSISTENT means the sign depends on the comparator. NOT SUPPORTED means the reviewed record was searched and the claim failed, not that a future trial is forbidden.

This document does not recommend a style, class, inversion, or schedule for any person. Sibling titles own isolated stretching architecture, physical-activity epidemiology, and exercise-prescription dose–response. Project 05 therapeutic-peptide data were not imported. Project 06 was read only.

99 References