Asthma is a heterogeneous, chronic inflammatory airway disease characterised by reversible airflow obstruction, bronchial hyperresponsiveness, and airway remodelling. ¹ In adults, it imposes significant morbidity: persistent dyspnoea, nocturnal symptoms, exercise limitation, and impaired health-related quality of life (QoL). Pharmacological treatment, centred on inhaled corticosteroids (ICS) and long-acting beta-2 agonists (LABAs), remains the cornerstone of management, yet complete asthma control is achieved in fewer than 50% of patients with moderate-to-severe disease. ^{2, 3}

A key contributor to this residual symptom burden is dysfunctional breathing (DB), a chronic alteration of ventilatory patterns that operates independently of airway eosinophilic inflammation. Common manifestations include upper thoracic breathing, mouth breathing, tachypnea, and chronic hyperventilation, which leads to hypocapnia via alveolar CO ₂ depletion. ^{4, 5} Experimentally, hypocapnia increases airway resistance and bronchial hyperresponsiveness in asthmatic subjects independently of inflammatory mechanisms, ⁶ creating a self-perpetuating cycle of dyspnoea, hyperventilation, and worsened symptom perception. ⁷ The prevalence of DB in adults with asthma is estimated at 27–50%, rising to 40–47% in refractory or difficult-to-treat populations, and its presence correlates with impaired QoL and increased healthcare utilisation irrespective of spirometric findings. ^{8, 9, 10}

Breathing retraining comprises structured interventions targeting these dysfunctional ventilatory patterns. The most studied modalities include: (i) the Buteyko breathing technique, based on deliberate hypoventilation and extended breath-holds to normalise CO ₂ levels and reduce bronchial hyperresponsiveness; (ii) Pranayama yoga-based breathing, combining breath control with relaxation; and (iii) physiotherapy-based programmes integrating diaphragmatic breathing, nasal breathing, pursed-lip expiration, and progressive relaxation. ^{11, 12} Recognition of this evidence base has led GINA 2024 and BTS/SIGN guidelines to formally recommend breathing exercises as an add-on intervention in adults with persistent symptoms despite optimal pharmacological therapy. ^{1, 3}

Previous systematic reviews, including the Cochrane review by Santino et al. (2020), reported a trend towards quality of life improvement with breathing exercises but rated overall certainty of evidence as low to moderate, due to methodological heterogeneity and limited study size. ¹² The current evidence base is further strengthened by two large, well-powered RCTs: Bruton et al., demonstrating equivalence between digital and face-to-face delivery in a primary care population, ¹³ and Andreasson et al., demonstrating efficacy specifically in moderate-to-severe asthma under specialist care. ¹⁴ Together, these trials strengthen the evidence base by addressing different clinically relevant contexts: scalable delivery in primary care and incompletely controlled moderate-to-severe asthma under specialist care, making a synthesis of this updated evidence base timely.

The primary objective was to evaluate the efficacy and safety of structured breathing exercises as adjunct therapy on health-related quality of life (AQLQ, Mini-AQLQ, SGRQ) and asthma control (ACT, ACQ) in adults with confirmed asthma. Secondary objectives were to examine effects on spirometric parameters (FEV ₁, PEF), hyperventilation symptoms (Nijmegen Questionnaire), rescue medication use, and healthcare resource utilisation, and to compare the efficacy of different breathing exercise modalities and delivery formats.

This systematic review was not prospectively registered, and no formal protocol was deposited in PROSPERO or another public registry prior to commencement. The review was conducted and reported in accordance with the PRISMA 2020 guidelines. ¹⁵ The PICO framework guided the research question and eligibility criteria.

Studies were eligible if they: (1) were randomised controlled trials (RCTs), controlled quasi-experimental studies, or uncontrolled pre-post pilot studies in which breathing exercises constituted the primary intervention; (2) enrolled adults (≥18 years) with a confirmed asthma diagnosis (GINA criteria); (3) used any structured breathing exercise programme as the primary adjunct intervention alongside standard pharmacological care; (4) reported at least one eligible outcome measure; (5) were published in English, Spanish, French, or Portuguese between January 2013 and March 2025. The date restriction was applied to reflect contemporary breathing retraining approaches and digital delivery models. Uncontrolled pre-post pilot studies were included to capture early feasibility data on novel delivery formats, with findings interpreted accordingly and with appropriate caution. Studies were excluded if they enrolled paediatric populations, used breathing exercises as a minor component within a broader multi-component rehabilitation programme, were non-original research (including systematic reviews and meta-analyses, which were used for contextual reference only), or had no retrievable full text.

A comprehensive electronic search was independently conducted by two reviewers (ACL, EVS) in five databases: Cochrane Library (including CENTRAL), PubMed/MEDLINE, Scopus, Web of Science (Core Collection), and PEDro. The search covered studies published from January 2013 to March 2025 (last searched: March 2025). Reference lists of included studies and relevant systematic reviews were manually screened for additional eligible studies. The search string applied across all databases, adapted as required for each database’s syntax, was:

(“breathing exercises” OR “respiratory therapy” OR “breathing techniques” OR “Buteyko” OR “diaphragmatic breathing” OR “pranayama” OR “breathing retraining” OR “pursed-lip expiration” OR “Papworth method”) AND (“asthma” OR “bronchial asthma”)

Language restrictions were applied (English, Spanish, French, Portuguese). Date and study design filters were applied where available in each database interface. The Boolean search string above was adapted to each database’s specific syntax requirements; core search terms and operators were consistent across all five databases.

Retrieved records were imported into Rayyan for deduplication. Two reviewers (ACL, EVS) independently screened titles and abstracts against the eligibility criteria, then retrieved and assessed full texts of potentially eligible studies. Discrepancies at any stage were resolved by consensus discussion; a third reviewer (LP) was available for arbitration. Reasons for exclusion at full-text stage were recorded for each excluded report and are presented in Table 1.

Data were independently extracted by two reviewers (ACL, EVS) using a standardised form capturing: study identification, design and setting, participant characteristics, intervention and comparator details, follow-up duration, outcome measures, key quantitative results (means, SDs, between-group differences, 95% CIs, p-values), adverse events, and funding sources. Outcome data were extracted and presented as reported by the original authors. No imputation, unit conversion, standardisation of effect sizes, or transformation of summary statistics was performed.

Methodological quality was independently assessed by two reviewers (ACL, LP) using two complementary tools. The PEDro (Physiotherapy Evidence Database) scale ¹⁶ was applied as the primary instrument, given the behavioural and rehabilitative nature of the interventions assessed (maximum score 10; studies scoring ≥6 classified as Good Quality; 4–5 as Fair Quality; ≤3 as Poor Quality). In addition, the Cochrane Risk of Bias 2 (RoB 2) tool ¹⁷ was applied to assess risk of bias across five standardised domains (D1: randomisation process; D2: deviations from intended interventions; D3: missing outcome data; D4: measurement of the outcome; D5: selection of the reported result) in randomised trials. For non-randomised and uncontrolled studies (Agarwal et al. and Karam et al.), methodological limitations were assessed descriptively using PEDro item-level ratings and acknowledged explicitly in the narrative synthesis, as RoB 2 is designed specifically for randomised trials. Certainty of evidence was not formally assessed using GRADE; confidence in the body of evidence is discussed narratively in the Discussion, taking into account study design, risk of bias, sample size, consistency of findings, and outcome type.

Due to significant clinical heterogeneity (diverse breathing exercise modalities, patient populations, and delivery formats), methodological heterogeneity (variability in study design, comparator groups, and supervision levels), and statistical heterogeneity (different outcome measures and assessment instruments across studies), a pooled meta-analysis was not considered appropriate. A pre-specified narrative synthesis was performed, with results presented by outcome domain: (1) health-related quality of life; (2) asthma control; (3) spirometric and physiological parameters; (4) rescue medication use and healthcare resource utilisation; (5) safety. Publication bias could not be formally assessed using funnel plots or regression tests due to the limited number of included studies (fewer than 10), in line with established methodological recommendations. ¹⁸

The database search identified 3807 records; 12 additional records were identified through manual screening of reference lists. Following deduplication, 2941 unique records were screened by title and abstract, of which 2563 were excluded. Of the 378 remaining records, 306 could not be retrieved despite database, institutional library, repository, and author-contact attempts; most non-retrieved reports were conference abstracts, trial summaries, or records without a corresponding full-text publication. This yielded 72 full-text reports assessed for eligibility. Of these, 66 were excluded: 28 due to wrong study design, 12 due to paediatric population, 9 because breathing exercises were not the primary intervention, 6 because the full text was unavailable, 4 because they were outside the specified date range, and 7 because they were systematic reviews or meta-analyses (used for contextual reference only). Six primary studies met all eligibility criteria and were included in the final synthesis. The complete PRISMA flow is presented in Table 2. Full-text exclusions are summarised by exclusion category in Table 1, with representative excluded reports shown for each category.

Six primary studies enrolling a total of 1,098 adult participants were included: four RCTs (Vagedes et al., ¹⁹ Andreasson et al., ¹⁴ Bruton et al., ¹³ Prem et al. ²⁰ ), one uncontrolled pre-post study (Agarwal et al. ²¹ ), and one uncontrolled pilot study (Karam et al. ²² ). Study settings included Germany, Denmark, United Kingdom, India, and the United States. Follow-up durations ranged from 4 weeks (Karam et al.) to 12 months (Bruton et al.; Andreasson et al.). PEDro scores ranged from 3 to 6: three RCTs (Vagedes, Andreasson, Bruton) were classified as Good Quality, one RCT (Prem) as Fair Quality, and the two uncontrolled studies (Agarwal, Karam) as Poor Quality. Detailed study characteristics are presented in Table 3 and methodological quality assessments in Tables 4 and 5.

RoB 2 assessments for the four RCTs are presented in Table 5. Two RCTs (Bruton et al. ¹³ and Andreasson et al. ¹⁴ ) had the most rigorous methodological execution, with Low risk in the randomisation, missing outcome data, outcome measurement, and selection of the reported result domains. However, all four RCTs were rated overall Some concerns, primarily driven by the deviations domain (D2) due to the inherently unblinded nature of behavioural interventions. Vagedes et al. ¹⁹ and Prem et al. ²⁰ had additional concerns in the outcome measurement domain (D4) for patient-reported outcomes. For the two non-randomised studies, RoB 2 was not applied. Agarwal et al. ²¹ (uncontrolled pre-post design) and Karam et al. ²² (uncontrolled pilot, n = 10) were assessed descriptively: both carry a high risk of confounding and should be interpreted with caution, with findings considered hypothesis-generating only. Lack of participant and therapist blinding was a consistent limitation across all six studies, inherent to breathing exercise interventions.

Health-related QoL was the primary or co-primary outcome in four studies. Bruton et al. ¹³ (n = 655) showed AQLQ scores at 12 months were significantly higher in the DVD-and-booklet (DVDB) group versus usual care (adjusted mean difference + 0.28; 95%CI 0.11–0.44; p = 0.002) and in the face-to-face group versus usual care (+0.24; 95%CI 0.04–0.44; p = 0.019). DVDB and face-to-face groups were equivalent (difference 0.04; 95%CI −0.16 to 0.24), confirming non-inferiority of digital delivery.

Andreasson et al. ¹⁴ (n = 193) showed that add-on breathing exercises were superior to usual specialist care for Mini-AQLQ at 6 months (+0.35; 95%CI 0.07–0.62; p = 0.015), with gains sustained at 12 months (+0.38; 95%CI 0.12–0.65). This study specifically recruited patients with incompletely controlled moderate-to-severe asthma, extending evidence to a population not well-served by previous trials.

Agarwal et al. ²¹ demonstrated significant improvements in all SGRQ domain scores following 3 months of Pranayama added to optimal pharmacotherapy (all p < 0.001): symptom score 45.98 → 38.78; activity score 15.45 → 12.34; impact score 17.95 → 12.12; total score 25.83 → 19.20. Given the absence of a control group, these results should be interpreted as indicative rather than confirmatory.

Vagedes et al. ¹⁹ (n = 60, RCT) reported significant improvement in ACQ scores in the Buteyko group versus usual treatment (p < 0.05). Concurrently, Nijmegen Questionnaire scores were significantly reduced (p < 0.05). This improvement occurred alongside a significant reduction in both beta-2 agonist (~20%; p < 0.05) and ICS use (~20%; p < 0.05) without deterioration in asthma control.

Prem et al. ²⁰ (n = 120, RCT) compared Buteyko and Pranayama in a three-arm RCT. Both techniques significantly improved AQLQ versus control. The Buteyko technique showed more consistent ACQ and ACT improvements; Pranayama was associated with greater subjective reduction in perceived breathlessness. Neither technique produced significant between-group differences in FEV ₁.

Andreasson et al. ¹⁴ identified a dose-response relationship: participants practising three times daily showed superior asthma control versus those practising twice daily. A secondary benefit on emotional well-being was observed, with significant improvement in the HADS depression subscale at 6 months (−0.90; 95%CI −1.67 to −0.14).

Static spirometric parameters (FEV ₁, FVC) did not demonstrate consistent or statistically significant improvements across included studies. Bruton et al. ¹³ and Andreasson et al., ¹⁴ both high-quality RCTs, reported no significant between-group differences in FEV ₁ or FeNO. The absence of change in FeNO indicates that clinical benefits are mediated through pathways independent of airway eosinophilic inflammation.

Vagedes et al. ¹⁹ identified a significant increase in the capnovolumetric threshold volume (VDthre) in the Buteyko group (~10 mL/10%; p < 0.05) and improvement in end-tidal CO ₂ (EtCO ₂), indicating normalisation of breathing pattern and measurable reduction in the drive towards chronic hyperventilation.

Vagedes et al. ¹⁹ reported significant reductions of approximately 20% in both beta-2 agonist and ICS daily dosage in the Buteyko group (p < 0.05 for both), absent in the control arm, demonstrating pharmacological de-escalation without loss of clinical control.

Bruton et al. ¹³ demonstrated no significant difference in primary care consultations, asthma attack rates, or asthma-related healthcare costs at 12 months. Andreasson et al. ¹⁴ reported comparable rates of asthma-related adverse events between groups (14.9% vs 18.1%; p = 0.38).

Breathing exercises were well-tolerated across the six included studies. No serious adverse events attributable to breathing exercise interventions were reported in the included studies that assessed safety. In Bruton et al., ¹³ adverse event incidence was numerically lower in active groups versus usual care (39–42% vs 50%), though not statistically significant. These findings support broad clinical applicability.