Dose-response relationship of normobaric hypoxia training on body composition and metabolic health in obese adults: a systematic review and meta-analysis

Abstract

Background

Normobaric hypoxia training (NHT) has emerged as a potentially superior exercise intervention for obesity management, theoretically offering enhanced metabolic stress and body composition benefits compared to normoxic exercise. However, optimal dose-response parameters remain undefined, limiting clinical translation and standardization of hypoxic exercise protocols.

Objective

To comprehensively investigate dose-response relationships between NHT parameters and body composition/metabolic outcomes in adults with obesity through systematic review and meta-analysis.

Methods

A systematic search of five databases (PubMed, Web of Science, Scopus, SPORTDiscus, CINAHL) was conducted from January 2014 to June 2025. Inclusion criteria comprised randomized controlled trials comparing NHT versus normoxic exercise in adults with BMI ≥ 25 kg/m². Primary outcomes included body weight, body fat percentage, BMI, fat mass, and waist circumference. Random-effects meta-analysis and univariate/multivariate meta-regression were employed for dose-response modelling. Dose parameters included fractional inspired oxygen (FiO₂), session duration, training frequency, intervention duration, and composite hypoxia dose scores.

Results

Ten studies involving 301 participants were included. Contrary to theoretical expectations, NHT demonstrated no superior body composition benefits compared to normoxic exercise across primary outcomes: body weight (MD = 0.12 kg, 95% CI: -1.99 to 2.22, p = 0.89), body fat percentage (MD = 0.21%, 95% CI: -3.00 to 3.41, p = 0.87), BMI (MD = -0.34 kg/m², 95% CI: -0.16 to 0.85, p = 0.15), and waist circumference (MD = -1.26 cm, 95% CI: -9.38 to 6.87, p = 0.66). Fat mass increased in NHT groups (MD = 1.10 kg, 95% CI: 0.24 to 1.95, p = 0.02). None of the primary outcomes achieved pre-defined clinical significance thresholds. Comprehensive dose-response meta-regression examining six hypoxia parameters (FiO₂, session duration, frequency, intervention weeks, total exposure hours, composite dose score) revealed no statistically significant relationships with any body composition outcome (all p > 0.05, all R² = 0.0-15.5%). Moderate multicollinearity among dose variables (r = 0.688–0.995, max VIF = 6.45) precluded reliable multivariate modelling, though univariate analyses consistently demonstrated null dose-response effects across all parameter-outcome combinations, indicating absence of clear dose-response gradients within examined ranges. NHT showed a non-significant trend toward improved cardiovascular fitness (VO₂peak: MD = 1.43 mL/kg/min, 95% CI: -0.86 to 3.72, p = 0.16) though with moderate heterogeneity across studies (I² = 62%).

Conclusions

This systematic review and meta-analysis found no evidence that normobaric hypoxia training produces superior body composition outcomes compared with equivalent normoxic exercise in adults with obesity. The absence of superior body composition benefits, combined with paradoxical dose-response relationships and increased intervention complexity, suggests that NHT cannot be recommended as a superior alternative to conventional exercise training for obesity treatment. The non-significant trend toward cardiovascular benefits with inconsistent responses across studies indicates that further research is needed to determine whether NHT may have value for specific applications focused on aerobic capacity enhancement. These findings emphasize the importance of rigorous evidence evaluation before widespread implementation of novel exercise interventions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13102-025-01457-z.

Introduction

Obesity represents one of the most pressing global public health challenges of the 21 st century, with its prevalence nearly doubling between 1999 and 2021 [1, 2]. Current projections indicate that by 2030, over 3.3 billion adults aged 25 and older worldwide will be affected by obesity, with particularly alarming increases in developing nations [1, 2]. This epidemic extends far beyond individual health consequences, imposing substantial economic burdens on healthcare systems and society at large. Obesity is intrinsically linked to a cascade of serious health complications including cardiovascular disease, type 2 diabetes, metabolic syndrome, and premature mortality, fundamentally altering physiological function and quality of life [3]. The pathophysiology of obesity involves complex interactions between genetic predisposition, environmental factors, and behavioural patterns, creating a multifaceted challenge requiring innovative therapeutic approaches.

Traditional exercise interventions have long served as the cornerstone of evidence-based obesity management, demonstrating consistent efficacy in promoting weight loss, improving metabolic health, and reducing cardiovascular risk factors. Systematic reviews and meta-analyses have established that structured exercise programs, particularly when combined with dietary modifications, can produce clinically meaningful reductions in body weight, body fat percentage, and metabolic dysfunction markers [4, 5]. However, the effectiveness of conventional exercise approaches in obese populations faces significant practical limitations that often compromise long-term success. These constraints include mechanical limitations imposed by excess body weight, reduced exercise tolerance and capacity, increased risk of musculoskeletal injury, and persistently poor adherence rates that undermine sustained weight management efforts [3, 4, 6]. Additionally, many individuals with obesity experience exercise-related discomfort, social anxiety, and motivational barriers that further impede successful intervention implementation.

The physiological rationale for normobaric hypoxia training emerges from decades of altitude medicine research demonstrating that reduced oxygen availability triggers profound adaptive responses that may be particularly beneficial for metabolic health and body composition management [1, 4, 7, 8]. Normobaric hypoxia training involves exercise performed under artificially reduced oxygen conditions, typically achieved through specialized chambers, masks, or tent systems that lower inspired oxygen fraction (FiO₂) to levels equivalent to simulated altitudes of 1,500-6,000 m [3, 4, 6, 8]. Unlike hypobaric hypoxia encountered at actual altitude, normobaric systems maintain atmospheric pressure while selectively reducing oxygen concentration, providing greater accessibility, safety, and precise environmental control for clinical applications.

The mechanistic foundation for hypoxic exercise benefits centres on the activation of hypoxia-inducible factor-1 (HIF-1), a master transcriptional regulator that orchestrates cellular responses to oxygen limitation [4, 6, 8]. HIF-1 activation under hypoxic conditions stimulates enhanced glucose metabolism, improved insulin sensitivity, increased fat oxidation, and modifications in appetite-regulating hormones including leptin and ghrelin. Furthermore, hypoxic exercise appears to generate superior metabolic stress compared to equivalent workloads under normoxic conditions, potentially allowing obese individuals to achieve greater physiological adaptations while exercising at reduced mechanical loads [3]. This phenomenon may address fundamental barriers to exercise participation in obese populations by reducing joint stress, minimizing injury risk, and improving exercise tolerance and adherence.

Emerging evidence suggests that normobaric hypoxia training produces additive benefits beyond those achieved through normoxic exercise alone, with several studies reporting superior improvements in body composition, weight loss, and metabolic parameters when exercise is performed under hypoxic conditions [4, 6, 8]. The proposed mechanisms underlying these enhanced effects include increased energy expenditure during hypoxic exercise, enhanced post-exercise oxygen consumption, improved mitochondrial biogenesis and function, and favourable alterations in substrate utilization patterns that promote fat oxidation over carbohydrate metabolism [4, 6]. Additionally, hypoxic exposure may influence circulating concentrations of adipokines and inflammatory markers, potentially creating a more favourable metabolic milieu for sustained weight management.

Despite growing research interest and promising preliminary findings, the current evidence base for normobaric hypoxia training in obesity management lacks standardization and consensus regarding optimal intervention parameters. Existing studies employ vastly heterogeneous protocols with oxygen concentrations ranging from 10 to 18% FiO₂, session durations varying from 20 min to over 90 min, training frequencies spanning 2–7 sessions per week, and intervention periods extending from 2 weeks to 8 months [1, 4, 6, 9]. This substantial protocol variability creates significant challenges for clinical translation, evidence synthesis, and the development of standardized treatment guidelines. Furthermore, the absence of systematic dose-response analysis prevents clinicians and researchers from determining optimal hypoxic exposure parameters that maximize therapeutic benefits while ensuring patient safety and treatment adherence.

Previous systematic reviews have examined the general effects of hypoxic exercise on various health outcomes in diverse populations, but none have specifically investigated dose-response relationships or provided evidence-based recommendations for optimal protocol design in obesity management [7]. The knowledge gap regarding dose-response relationships represents a critical barrier to the clinical implementation and standardization of normobaric hypoxia training interventions. Understanding how specific hypoxic dose parameters, including oxygen concentration, exposure duration, training frequency, and total intervention length, influence therapeutic outcomes is essential for developing evidence-based clinical guidelines, optimizing treatment protocols, maximizing therapeutic efficacy, and ensuring patient safety.

The dose-response relationship in normobaric hypoxia training encompasses multiple interacting variables that collectively determine intervention effectiveness. The level of hypoxic stimulus, quantified by FiO₂ percentage or equivalent simulated altitude, represents a fundamental parameter influencing the magnitude of physiological stress and adaptive responses. Session duration and training frequency determine the cumulative hypoxic exposure dose, while intervention length affects both acute and chronic adaptations. The interaction between these dose components likely creates complex, non-linear relationships with therapeutic outcomes that require sophisticated analytical approaches to characterize effectively [1, 4, 6, 10]. Understanding optimal dose-response relationships holds profound implications for clinical practice, research methodology, and public health policy. From a clinical perspective, evidence-based dosing guidelines would enable healthcare providers to prescribe hypoxic exercise interventions with confidence in their safety and efficacy, similar to established exercise prescription principles for traditional training modalities. For researchers, standardized protocols would facilitate more meaningful comparisons between studies, enhance the interpretability of meta-analytic findings, and guide the design of future investigations. From a public health standpoint, optimized protocols could maximize the cost-effectiveness of hypoxic training interventions and inform resource allocation decisions for implementing these technologies in clinical and community settings.

This systematic review and meta-analysis addresses this critical knowledge gap by providing the first comprehensive dose-response analysis of normobaric hypoxia training parameters in obesity treatment. The investigation aimed to synthesize existing evidence from randomized controlled trials to identify optimal hypoxic exposure parameters that maximize improvements in body composition and metabolic health outcomes while maintaining safety and feasibility for clinical implementation. Through rigorous meta-regression analyses and subgroup investigations, this review generates clinically actionable evidence for optimal protocol design, contribute to the standardization of hypoxic exercise interventions, and establish a foundation for future research and clinical practice guidelines in this rapidly evolving field.

Methods

Protocol registration and reporting guidelines

This systematic review and meta-analysis protocol was registered with the International Platform of Registered Systematic Review and Meta-Analysis Protocols (INPLASY) on July 2, 2025, under registration number INPLASY202570012 (DOI: 10.37766/inplasy2025.7.0012). The protocol registration ensures transparency, prevents duplication of research efforts, and enables tracking of any protocol modifications throughout the review process. The completed systematic review strictly adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement guidelines for comprehensive and transparent reporting of systematic reviews and meta-analyses [11, 12]. All phases of the review process, including search strategy development, study selection procedures, data extraction protocols, quality assessment methods, and statistical analysis approaches, follow established PRISMA 2020 reporting standards to ensure methodological rigor and reproducibility.

Search strategy

A comprehensive search strategy was implemented across multiple electronic databases to identify all relevant randomized controlled trials investigating normobaric hypoxia training in adults with obesity. The systematic search was conducted in five primary databases spanning January 1, 2014, to 2025, ensuring comprehensive coverage of contemporary literature in this emerging therapeutic field [13]. The databases selected encompassed biomedical literature (PubMed/MEDLINE), multidisciplinary scientific databases (Web of Science Core Collection and Scopus), exercise and sports science literature (SPORTDiscus), and nursing and allied health literature (CINAHL Complete), providing broad disciplinary coverage relevant to hypoxic exercise interventions.

The search strategy incorporated four primary concept blocks systematically combined using Boolean operators to maximize sensitivity while maintaining specificity. The hypoxia exposure concept included terms such as “normobaric hypoxia,” “simulated altitude,” “artificial hypoxia,” “hypoxic training,” “hypoxia conditioning,” “intermittent hypoxia,” “hypoxic environment,” “altitude simulation,” “reduced oxygen,” and “oxygen restriction.” Exercise and training concepts encompassed “exercise,” “training,” “physical activity,” “aerobic exercise,” “resistance training,” “strength training,” “interval training,” “HIIT,” “high-intensity interval training,” “exercise intervention,” “physical training,” and “exercise therapy.” Population-specific terms included “obesity,” “obese,” “overweight,” “body weight,” “body composition,” “body mass index,” “BMI,” “weight loss,” “fat mass,” “adiposity,” “metabolic syndrome,” “metabolic health,” and “waist circumference.” Study design filters incorporated “randomized controlled trial,” “randomised controlled trial,” “RCT,” “randomized,” “randomised,” “controlled trial,” “clinical trial,” “intervention study,” and “experimental study.”

Search parameters were restricted to English language publications from peer-reviewed journals, focusing exclusively on randomized controlled trials involving human adults aged 18 years and older. The search strategy was systematically adapted for each database while maintaining consistency in core concepts and ensuring comprehensive coverage across different indexing systems. Medical Subject Headings (MeSH) terms were utilized in PubMed searches, while equivalent controlled vocabularies and subject headings were employed in other databases as appropriate. Complete database-specific search strategies, including exact search strings and results, are provided in Appendix 1 in Supplementary Material.

Eligibility criteria

Study eligibility was determined using the Population, Intervention, Comparator, Outcomes, and Study design (PICOS) framework to ensure systematic and reproducible selection criteria [12, 14, 15]. Two independent reviewers conducted the eligibility assessment, with disagreements resolved through discussion or consultation with a third reviewer when consensus could not be reached.

Inclusion criteria

Studies were included if they met the following criteria: the study population comprised adults aged 18 years or older with a body mass index of 30 kg/m² or greater (obese) or 25 kg/m² or greater when studies specifically focused on metabolic health outcomes (overweight). Participants included generally healthy adults or those with obesity-related metabolic conditions such as type 2 diabetes, metabolic syndrome, hypertension, or insulin resistance, with both male and female participants eligible. Studies could be conducted in any setting including laboratory, clinical, or community-based environments. The intervention required normobaric hypoxia training defined as exercise training performed under artificially reduced oxygen conditions using normobaric hypoxia systems, with oxygen concentration between 10 and 18% (equivalent to approximately 1,500-6,000 m simulated altitude), session duration of at least 20 min per session, frequency of at least two sessions over the study period, and delivery through normobaric hypoxia chambers, masks, tents, or other validated hypoxia delivery systems [11, 12]. Exercise components must include any form of structured exercise training such as aerobic, resistance, high-intensity interval training, or combined modalities, performed concurrently with hypoxic exposure using progressive or standardized exercise protocols.

The comparator group must involve exercise training performed under normoxic conditions with fractional inspired oxygen approximately 21% or sea-level equivalent, utilizing identical exercise protocols to the intervention group including same exercise modality, intensity, duration, and frequency, same intervention duration, same population characteristics, and performed under normoxic conditions only. Primary or secondary outcomes must include at least one measure of body weight, body mass index, body composition, or fat mass. Study design criteria required randomized controlled trials only, including parallel-group or crossover designs, single-blind, double-blind, or open-label trials, and multi-centre or single-centre studies [11, 12]. Additional requirements included intervention duration of at least two weeks, documented hypoxia parameters specifying fractional inspired oxygen level or equivalent altitude simulation, structured exercise protocol with specified intensity, duration, and frequency, publication in English language, publication between January 1, 2014, and 2025, publication in peer-reviewed journals, and sufficient data availability for meta-analysis including means, standard deviations, or extractable effect sizes.

Exclusion criteria

Studies were excluded based on the following criteria: study design limitations including non-randomized studies, observational studies, case reports, case series, reviews, or conference abstracts. Population exclusions comprised children or adolescents under 18 years of age, highly trained athletes or competitive athletes defined as more than 10 h training per week or competitive sport participation, pregnant participants, individuals with acute illness or unstable medical conditions at baseline, participants with previous altitude acclimatization within three months before study commencement involving exposure above 2,500 m, and studies exclusively involving clinical populations with severe comorbidities unrelated to obesity such as cancer, severe cardiovascular disease, or chronic obstructive pulmonary disease [1–3]. Intervention exclusions included hypobaric hypoxia involving actual altitude exposure, passive hypoxia without concurrent exercise, blood flow restriction techniques, fractional inspired oxygen levels below 10% or above 18%, interventions lacking appropriate normoxic exercise control groups, and studies with different exercise protocols between intervention and control groups [4, 6, 8]. Outcome-related exclusions involved studies not reporting any primary or secondary outcomes of interest. Duration exclusions comprised interventions lasting less than two weeks and acute studies examining single session effects only. Publication-related exclusions included studies published before 2014 or after 2025, and data-related exclusions encompassed insufficient data for analysis, duplicated publications, and non-English language studies.

Study selection process

Study selection was conducted using a systematic three-phase approach following established best practices for systematic reviews [11, 12, 16]. All search results from the five databases were imported into Zotero reference management software for proper referencing and organization. Zotero facilitated the verification of complete reference information and enabled systematic exportation of verified references to EPPI-Reviewer Web Version 6.16, where comprehensive duplicate removal and subsequent screening procedures were performed [17] The screening process was performed by two independent reviewers working in parallel to minimize selection bias and ensure comprehensive evaluation of all retrieved records. The first phase involved title and abstract screening, where each reviewer independently evaluated all deduplicated records against the predetermined eligibility criteria [11, 12, 16]. During this phase, reviewers made one of three decisions for each record: include for full-text retrieval when the study appeared to meet al.l inclusion criteria based on available information, exclude when the study clearly did not meet inclusion criteria or met one or more exclusion criteria, or uncertain when insufficient information was available in the title and abstract to make a definitive decision [11, 12, 16]. Records categorized as uncertain were automatically advanced to the next phase to ensure no potentially relevant studies were inadvertently excluded due to inadequate abstracts or unclear reporting.

The second phase involved systematic full-text file retrieval for all records that passed the initial title and abstract screening or were marked as uncertain. Complete manuscripts were obtained through institutional subscriptions. The third phase comprised full-text screening, where both reviewers independently assessed the complete manuscripts against all PICOS criteria to make final inclusion or exclusion decisions [11, 12, 16]. For each excluded study at the full-text level, reviewers documented the primary reason for exclusion using standardized categories aligned with the exclusion criteria to facilitate transparent reporting and PRISMA flow diagram completion [11, 12, 16].

Throughout both screening phases, reviewers worked independently with decisions masked from each other until individual assessments were completed. Inter-rater agreement was calculated using Cohen’s kappa coefficient to quantify the level of concordance between reviewers. Disagreements between the two primary reviewers at either screening phase were systematically identified and resolved through a structured consensus process. Initial conflicts were addressed through direct discussion between the two primary reviewers, allowing for clarification of eligibility criteria interpretation and consideration of additional study details [11, 12, 16]. When consensus could not be reached through discussion, a third reviewer was consulted to provide an independent assessment and facilitate resolution. In cases involving complex methodological considerations or borderline eligibility decisions, a fourth reviewer was available to participate in team discussions until unanimous agreement was achieved. The screening process incorporated quality control measures to ensure consistency and accuracy throughout study selection. Prior to commencing formal screening, all reviewers participated in a calibration/coding exercise involving a representative sample of records to align understanding of eligibility criteria and screening procedures.

Data extraction

Data extraction was conducted using standardized forms designed specifically for normobaric hypoxia training dose-response analysis. Two independent reviewers extracted data from all included studies using pre-defined extraction tables covering nine domains: study and participant characteristics, hypoxia intervention parameters, exercise protocol details, primary body composition outcomes, secondary body composition measures, metabolic health outcomes, cardiovascular and performance variables, safety and adherence data, and risk of bias assessment. Study characteristics extracted included publication year, country, study design, intervention duration, and sample sizes with gender distribution using the format “NHT = N (M = n, F = n), CON = N (M = n, F = n).” Participant demographics were recorded as group-specific means and standard deviations following the convention “NHT = Mean ± SD, CON = Mean ± SD” for age, baseline BMI, and body weight, alongside population type, inclusion criteria, baseline fitness levels, and dropout rates.

Critical dose-response parameters included fractional inspired oxygen levels (FiO2), session duration, training frequency, total hypoxic exposure time, delivery methods, and progressive protocols. Exercise characteristics encompassed modality, intensity, duration, frequency, supervision, and compliance rates. Primary outcomes focused on body weight, BMI, body fat percentage, fat mass, and waist circumference measured at baseline and post-intervention. Secondary outcomes included additional body composition measures, metabolic biomarkers, cardiovascular parameters, and performance indices. Disagreements between reviewers were resolved through discussion or third-party consultation. All extracted data were verified for accuracy, unit consistency, and completeness before statistical analysis.

Quality assessment

Methodological quality of included studies was assessed using the Cochrane Risk of Bias Tool 2.0 (RoB 2) for randomized controlled trials [18–20]. Two independent reviewers evaluated each study across five key domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each domain was rated as “low risk,” “high risk,” or “some concerns” based on specific signalling questions and algorithm guidance provided in the RoB 2 framework [12, 19]. The randomization process domain assessed adequacy of sequence generation and allocation concealment procedures. Deviations from intended interventions examined adherence to assigned treatments and appropriateness of analysis methods. Missing outcome data evaluation focused on completeness of data and handling of dropouts. Measurement of outcome domain considered objectivity of assessment methods and potential for measurement bias, particularly relevant given the difficulty of blinding participants to hypoxic interventions. Selection of reported results assessed potential for selective reporting of favourable outcomes.

An overall risk of bias judgment was assigned to each study based on the combination of domain-specific assessments, with studies classified as low risk only when all domains showed low risk of bias. Inter-rater agreement was calculated using Cohen’s kappa coefficient, and disagreements were resolved through discussion or consultation with a third reviewer when consensus could not be reached [11, 21, 22]. Publication bias was assessed through visual inspection of funnel plots when ten or more studies were available for a given outcome, supplemented by Egger’s regression test for statistical detection of small-study effects [21, 22]. Asymmetry in funnel plots or significant Egger’s test results (p < 0.10) indicated potential publication bias requiring sensitivity analysis.

Statistical analysis

All statistical analyses were conducted using R statistical software version 4.2.2 with specialized packages for meta-analysis (metafor, meta), network meta-analysis (netmeta), and dose-response modelling (dmetar, rms) [23–25].

Effect size calculation and data preparation

For multi-arm studies, a single intervention arm closest to moderate hypoxia intensity (14–16% FiO₂) was selected for comparison against the normoxic control to maintain independence. Effect sizes were calculated as mean differences (MD) for outcomes on identical scales or standardized mean differences (SMD) using Hedges’ g for different measurement methods [11, 26, 27]. Dose variables included: FiO₂ level (%), session duration (minutes), training frequency (sessions/week), intervention duration (weeks), and cumulative exposure time. A composite hypoxia dose score was calculated as: (21 - FiO₂) × session duration × frequency × weeks. Categorical variables were created for hypoxia intensity: mild (16–18%), moderate (14–16%), severe (10–14%).

Meta-analysis and dose-response modelling

Random-effects meta-analysis employed the Restricted Maximum-Likelihood (REML) method with Knapp-Hartung (Hartung-Knapp-Sidik-Jonkman - HKSJ) adjustments to account for uncertainty in between-study variance estimation [11, 28, 29]. Heterogeneity was quantified using I² statistics, interpreted as low (< 25%), moderate (25–50%), substantial (50–75%), or considerable (>75%), with corresponding τ² values and 95% confidence intervals reported. Prediction intervals were calculated to estimate the range of true effects in future studies.

Dose-response relationships were investigated through systematic meta-regression analyses examining six hypoxia dose parameters: fractional inspired oxygen concentration (FiO₂, %), session duration (minutes), training frequency (sessions/week), intervention duration (weeks), total cumulative exposure time (hours), and a composite hypoxia dose score. The composite dose score integrated multiple exposure parameters using the formula: (21 - FiO₂) × session duration × frequency × intervention weeks, where 21 represents normoxic oxygen concentration (21% FiO₂). For example, a protocol implementing 60-minute training sessions at FiO₂ = 15%, three times weekly for 8 weeks yields a composite dose score of (21 − 15) × 60 × 3 × 8 = 8,640 arbitrary units [30–32]. This formula quantifies cumulative hypoxic stimulus by integrating intensity (oxygen reduction from normoxia), session duration, training frequency, and intervention duration into a single metric.

Univariate meta-regression analyses modelled each dose parameter separately using the specification θ_i = β₀ + β₁ × dose_i + ε_i, where θ_i represents the treatment effect in study i, β₀ the intercept, β₁ the dose-response coefficient, dose_i the dose value, and ε_i the residual error [25]. The coefficient β₁ represents the change in mean difference per unit increase in the dose parameter, with positive values indicating diminished treatment effects at higher doses and negative values indicating enhanced effects at higher doses. The proportion of between-study heterogeneity explained by each dose parameter (R²) was calculated as R² = (τ²null - τ²model)/τ²null × 100%, where τ²null represents heterogeneity from the null model without covariates.

Multivariate meta-regression models incorporating multiple dose parameters simultaneously were planned but required adequate sample sizes (k ≥ 5 studies) and absence of substantial multicollinearity. Multicollinearity was assessed using variance inflation factors (VIF), with VIF < 5 indicating acceptable collinearity, 5–10 indicating moderate concern requiring cautious interpretation, and >10 indicating severe multicollinearity necessitating univariate-only analyses(Table 1) [33, 34]. Pairwise correlations between dose variables were examined using Pearson correlation coefficients, with |r| >0.7 indicating high correlation potentially affecting multivariate model stability.

Table 1.

Study characteristics summary

Study ID	Study Design	Intervention Duration	Sample Size by Group	Age	BMI Baseline	Weight Baseline	Baseline Fitness Level
Camacho-Cardenosa 2018	RCT	12 weeks	IHT = 13 (M = 0, F = 13), INT = 15 (M = 0, F = 15)	IHT = 44.43 ± 7.18, INT = 43.14 ± 7.67	IHT = 30.03 ± 6.37, INT = 29.59 ± 5.25	IHT = 80.10 ± 18.88, INT = 80.41 ± 16.27	Sedentary (< 2 bouts of 30 min exercise/week)
Camacho-Cardenosa 2020	RCT	12 weeks	NHT = 41 (M = 0, F = 41), CON = 41 (M = 0, F = 41)	NHT = 36.6 ± 9.5, CON = 31.3 ± 8.8	NHT = 29.78 ± 6.23 kg/m², CON = 29.31 ± 5.21 kg/m²	NR	Sedentary
Chacaroun 2020	RCT	8 weeks	NHT = 12 (M = 11, F = 1), CON = 11 (M = 8, F = 3)	NHT = 52 ± 12 years, CON = 56 ± 11 years	NHT = 31.2 ± 2.4 kg/m², CON = 31.8 ± 3.2 kg/m²	NR	Sedentary (< 2 h low-intensity physical activity/week)
Gatterer 2015	RCT	32 weeks	NHT = 16 (M = 4, F = 12), CON = 16 (M = 6, F = 10)	NHT = 50.3 ± 10.3 years, CON = 52.4 ± 7.9 years	NHT = 37.9 ± 8.1 kg/m², CON = 36.3 ± 4.0 kg/m²	NHT = 105.5 ± 20.0 kg, CON = 103.2 ± 15.1 kg	NR
Ghaith 2022	RCT	8 weeks	NHT = 16 (M = 10, F = 6), CON = 15 (M = 13, F = 2)	NHT = 51.0 ± 8.3 years, CON = 52.0 ± 7.5 years	NHT = 31.5 ± 4.0 kg/m², CON = 32.4 ± 4.8 kg/m²	NHT = 95.4 ± 19.4 kg, CON = 99.9 ± 15.5 kg	Sedentary (< 2 h low-intensity physical activity/week)
Jiao 2024	RCT	4 weeks	NHT = 13 (M = 5, F = 8), CON = 9 (M = 4, F = 5)	NHT = 36.62 ± 9.54, CON = 31.33 ± 8.75	NHT = 28.08 ± 2.10 kg/m², CON = 29.28 ± 5.81 kg/m²	NHT = 80.70 ± 9.95 kg, CON = 84.98 ± 19.39 kg	Sedentary
Kong 2014	RCT	4 weeks	NHT = 9 (M = 5, F = 4), CON = 9 (M = 5, F = 4)	NHT = 19.8 ± 2.2, CON = 22.3 ± 1.7	NHT = 34.7 ± 5.3 kg/m², CON = 33.8 ± 5.6 kg/m²	NHT = 99.0 ± 19.5 kg, CON = 103.4 ± 24.7 kg	Sedentary
Mai 2019	RCT	6 weeks	NHT = 12 (M = 12, F = 0), CON = 9 (M = 9, F = 0)	NHT = 57.0 ± 7.5, CON = 58.0 ± 8.5	NHT = 33.5 ± 4.3 kg/m², CON = 34.2 ± 2.6 kg/m²	NHT = NR, CON = NR	Sedentary
Morishima 2014	RCT	4 weeks	NHT = 9 (M = 9, F = 0), CON = 11 (M = 11, F = 0)	NHT = 30 ± 2, CON = 32 ± 3	NHT = 25.6 ± 1.2 kg/m², CON = 25.4 ± 0.9 kg/m²	NHT = 74.4 ± 4.2 kg, CON = 73.8 ± 4.0 kg	Sedentary
Park 2019	RCT	12 weeks	NHT = 12 (M = 12, F = 0), CON = 12 (M = 12, F = 0)	NHT = 66.5 ± 0.9, CON = 66.5 ± 0.67	NHT = 26.0 ± 0.61 kg/m², CON = 25.63 ± 0.35 kg/m²	NHT = 70.72 ± 5.45 kg, CON = 68.39 ± 5.03 kg	Low activity levels

All meta-regression analyses employed random-effects specifications to account for residual heterogeneity after including dose covariates. Statistical significance was set at p < 0.05 for all analyses. Non-linear dose-response relationships were not formally tested due to the limited number of studies. All analyses were performed using the metafor and meta packages in R statistical software version 4.2.2 [23, 24].

Sensitivity analyses

To assess the robustness of pooled effect estimates, two complementary sensitivity analyses for all primary outcomes we conducted. Leave-one-out analysis systematically removed each study individually and re-calculated pooled effect estimates with the remaining studies (k-1) using identical random-effects models (REML with Knapp-Hartung adjustment) [11, 35, 36]. A finding was considered robust if removal of any single study did not change the statistical significance, direction of effect, or clinical interpretation of results. Substantial influence was defined as: (1) change in statistical significance (p-value crossing 0.05 threshold), (2) change in direction of effect (confidence interval crossing zero), (3) >20% relative change in point estimate magnitude, or (4) I² change >25% points. Influence diagnostics were conducted using multiple statistics to identify studies exerting disproportionate influence on pooled estimates. We calculated Cook’s distance (threshold >1.0 for high influence), DFFITS (threshold >2√(p/n)), hat values for leverage assessment, and standardized residuals to identify outliers (|residual| >2.5) [11, 37, 38]. All sensitivity analyses were conducted using the meta and metafor packages in R version 4.2.2.

Results

Study selection

The systematic search across five electronic databases yielded a total of 410 records (PubMed: 55, Scopus: 211, Web of Science: 94, SPORTDiscus: 39, CINAHL: 11). After removal of 131 duplicate records, 279 unique records proceeded to title and abstract screening (see Appendix 1 in Supplementary Material & Fig. 1). During this initial screening phase, 252 records were excluded for the following reasons: not target population (183 records), wrong intervention type (47 records), not primary research (11 records), no relevant outcomes (8 records), and wrong study design (3 records) (Fig. 1). Twenty-seven reports were deemed potentially eligible and retrieved for full-text assessment. Following comprehensive full-text evaluation against the predetermined inclusion and exclusion criteria, 17 reports were excluded due to: outcomes not relevant (1 report), intervention does not meet criteria (5 reports), population does not meet criteria (8 reports), and insufficient and duplicate data (3). No additional records were identified through supplementary search methods. The final selection process resulted in 10 studies meeting all eligibility criteria for inclusion in the systematic review and meta-analysis [39–48]. The complete study selection process is illustrated in the PRISMA flow diagram (Fig. 1), which demonstrates adherence to systematic review reporting standards and ensures transparency in the identification and selection of relevant literature.

Fig. 1.

PRISMA 2020 flow diagram for the systematic review

Study characteristics

The 10 included studies (Table 1) conducted between 2014 and 2024 across six countries: Spain [39, 40], Germany [42, 46], China [44, 45], France [41, 43], Japan [47], and South Korea [48]. Intervention durations ranged from 4 weeks to 32 weeks, with most studies (6/10, 60%) implementing interventions lasting 8–12 weeks (Table 1). The total sample included 301 participants across all studies, with individual study sample sizes ranging from 18 to 82 participants.

Gender distribution varied considerably across studies. Three studies exclusively recruited females [39, 40, 46], two studies exclusively recruited males [47, 48], and five studies included mixed populations with varying gender ratios [41–45]. Participant ages ranged from 19.8 to 66.5 years, representing diverse adult age groups from young adults to older populations (Table 1). Baseline BMI values ranged from 25.4 to 37.9 kg/m², with most participants classified as overweight or obese according to WHO criteria [9, 49, 50]. Body weight at baseline varied from 65.04 to 105.5 kg across studies.

All studies recruited sedentary or low-activity participants, with consistent inclusion criteria requiring minimal regular physical activity engagement (typically < 2 h of low-intensity exercise per week or < 2 bouts of 30-minute exercise sessions weekly). This homogeneity in baseline activity levels strengthens the comparability of findings across studies. Geographic diversity spanning six countries across Europe, Asia, and East Asia, combined with population heterogeneity in age, gender, and obesity severity, provides broad representativeness for the target population of sedentary adults with excess body weight seeking exercise-based interventions for obesity management.

A comprehensive study-to-analysis mapping table is provided in Appendix 2 in Supplementary Material, which documents which studies contributed data to each meta-analysis, including sample sizes per outcome, measurement approaches, and timepoints used. This mapping ensures transparency regarding the independence of effect estimates and facilitates identification of potential unit-of-analysis issues across multiple outcomes.

Intervention characteristics

Exercise protocol details

The included studies employed diverse exercise modalities across the intervention spectrum (Table 2). Aerobic exercise was the predominant training mode, implemented in 6 studies through cycling [39–41, 43, 47] or treadmill walking [46] while rest combined exercises majorly treadmills and cycling. High-intensity interval training (HIIT) protocols were utilized in 2 studies [39, 43], featuring interval durations ranging from 30 s to 3 min at intensities between 80 and 130% of maximum workload. Combined exercise approaches incorporating both aerobic and resistance components were implemented in 3 studies [44, 45, 48], with [48] employing sequential treadmill, bicycle, and elastic resistance training within single sessions. Exercise intensities varied considerably across studies, with moderate-intensity protocols targeting 50–75% of maximum heart rate [41, 42, 47] and high-intensity protocols reaching 80–130% of maximum workload [39, 40, 43]. Session durations ranged from 29.6 min to 180 min, with most studies (8/10) implementing sessions between 40 and 90 min. All studies employed supervised exercise protocols with experienced research personnel, ensuring proper technique and safety monitoring throughout interventions.

Table 2.

Exercise protocol details

Study	Body Composition Method	Exercise Type	Exercise Mode	Exercise Intensity	Intensity Monitoring	Exercise Duration	Progression Protocol	Control Group Exercise
Camacho-Cardenosa 2018	BIA	IHT/INT: Interval training	Cycling (Ergoselect series 100/200; Ergoline GmbH)	IHT/INT: 90% Wmax (3 min) + 55–65% Wmax recovery (3 min	HR monitor (Team System, Polar), RPE	IHT/INT: 41.5 min average	Yes - weeks 1–2: 3 intervals, weeks 3–5: 4 intervals, weeks 6–8: 5 intervals, weeks 9–12: 6 intervals	Identical to intervention but in normoxia (FiO2 = 20.9%)
Camacho-Cardenosa 2020	BIA	Aerobic interval	Cycling	90% Wmax intervals, 55–65% Wmax recovery	HR monitor	42 min	Yes (3–6 sets)	Yes
Chacaroun 2020a	MRI	Moderate continuous	Cycling	75% HRmax	Heart rate	45 min	No	Identical protocol at normoxia
Gatterer 2015	BIA	Moderate continuous	Cycle ergometer/treadmill/cross trainer	65–70% HRmax	Heart rate	90 min	Yes (HR target adapted after 3 months)	Identical protocol at normoxia
Ghaith 2022	BIA	HIIT	Cycling	80% Wpeak (NHT) vs. 100% Wpeak (CON)	Heart rate and RPE	Variable (30–60 s intervals)	Yes (volume and duration increased)	Identical protocol at normoxia
Jiao 2024	BIA	Combined	Multiple modalities	≥ 3 METs, ≥ 70% VO2max	NR	60 min	No	Yes
Kong 2014	BIA	Combined	Running, cycling, stepping, strength	60–70% HRmax	HR monitor	120 min	Yes	Yes
Mai 2019	ADP	Moderate-intensity aerobic exercise	Treadmill walking	50–60% HRmax	HR monitor	60 min	No	Yes
Morishima 2014	DXA	Aerobic	Cycling	55% VO2max	Heart rate monitor	60 min	No	Identical protocol in normoxia (FiO2 = 21%)
Park 2019	BIA	Combined (aerobic + resistance)	Treadmill (30 min) + bicycle (30 min) + elastic resistance (30–40 min)	60–70% HRmax, 6–7 OMNI-RES scale	Heart rate using Tanaka formula (208 - [0.7 age])	90–120 min	No	Identical protocol in normoxia

ADP Air Displacement Plethysmography, BIA Bioelectrical Impedance Analysis, MRI Magnetic Resonance Imaging, DXA Dual-Energy X-ray Absorptiometry

Hypoxia intervention parameters

Normobaric hypoxia delivery methods varied across studies, with environmental chambers being the predominant approach used in 7 studies [39, 40, 42, 44–47] and nitrogen-enriched gas masks utilized in 2 studies [41, 43]. Fractional inspired oxygen levels ranged from 12.0% to 17.2%, corresponding to simulated altitudes between 2,000 and 4,500 m (Table 3). The majority of studies (8/10) employed moderate hypoxia levels between 14.0 and 16.4% FiO₂, equivalent to 2,500-3,500 m altitude simulation. Training frequency was consistent across most interventions, with 8 studies implementing 3 sessions per week [39–41, 43, 45–48] while 2 studies utilized twice-weekly protocols [42] and 1 study employed 5 sessions per week [44]. Total hypoxic exposure time varied substantially from 12 h [47] to 156 h [42], reflecting differences in session duration, frequency, and intervention length. Hypoxia monitoring was implemented through pulse oximetry in 5 studies [39–41, 43, 45], with target oxygen saturation levels typically maintained at 80–92%. Environmental conditions were standardized in 4 studies, with temperature maintained at 22–24 °C and relative humidity controlled at 40–60% [39, 40, 44, 48]. All control groups received identical exercise protocols under normoxic conditions (FiO₂ 20.9–21%), ensuring appropriate comparative assessment of hypoxic versus normoxic exercise effects.

Table 3.

Hypoxia intervention parameters

Study ID	Hypoxia Delivery Method	FiO2 Level (%)	Simulated Altitude (m)	Session Duration (min)	Sessions/Week	Total Hypoxia Exposure	Hypoxia Monitoring
2018	Normobaric hypoxia chambers (CAT 310, Louisville, CO)	17.2 ± 0.3%	2500 m	IHT: 41.5 min average, RSH: 29.6 min average	3	IHT: 24.9 h, RSH: 17.8 h	Electronic device (HANDI+; Maxtec)
Camacho-Cardenosa 2020	Normobaric hypoxic chambers	17.2	NR	42 min	3	25.2 h	SpO2
Chacaroun 2020	Mask (nitrogen-enriched gas mixture)	13	3,700 m	45 min	3	18 h	SpO2 target 80%
Gatterer 2015	Normobaric hypoxic chambers	Exercise: 14.0 ± 0.2%, Rest: 12.2 ± 0.3%	Exercise: 3500 m, Rest: 4500 m	180 min (90 min exercise + 90 min rest)	2	156 h	NR
Ghaith 2022	Mask (nitrogen-enriched gas mixture)	0.12	4200 m	90 min	3	36 h	SpO2 target 80 ± 2%
Jiao 2024	Normobaric hypoxic chambers	15	2700 m	60 min	5	20 h	Electronic device
Kong 2014	Normobaric hypoxic chambers	16.4–14.5%	2000–3000 m	120 min	3.5	28 h	Pulse oximeter
Mai 2019	Normobaric hypoxic chambers	15	2500 m	60 min	3	18 h	SpO2
Morishima 2014	Normobaric hypoxia chamber	15	3000 m	60 min	3	12 h	Not specified
Park 2019	Normobaric hypoxia chamber	14	3,000 m	90–120 min	3	54–72 h	Not specified

Appendix 2 in Supplementary Material provides a comprehensive overview of which studies contributed data to each meta-analysis, including sample sizes, measurement approaches, and analytical notes. The table facilitates transparency in reporting by clearly documenting the independence of effect estimates and ensuring readers can trace the contribution of individual studies across multiple outcomes. This mapping is essential for understanding the evidence base supporting each pooled estimate and for identifying potential unit-of-analysis issues or overlapping data sources. For studies with multiple intervention arms (e.g [39])., with both interval training and repeated-sprint protocols under hypoxia), included only the arm most representative of continuous or moderate-intensity hypoxic training to maintain consistency with other included studies and avoid unit-of-analysis errors.

Risk of bias assessment

The risk of bias assessment using the Cochrane Risk of Bias 2.0 tool revealed methodological quality patterns characteristic of behavioural exercise interventions across the 10 included studies (Figs. 2 and 3). The assessment identified domain-specific concerns that warrant careful interpretation of findings. Four studies (40%) demonstrated low risk with adequate sequence generation and allocation concealment clearly described [39, 40, 42, 44]. Six studies (60%) raised some concerns due to insufficient detail regarding randomization methods or allocation concealment procedures, though all studies explicitly stated random assignment of participants [41, 43, 45–48].

Fig. 2.

Summary of the risk of bias assessment for included studies

Fig. 3.

Risk of bias assessment for included studies

All 10 studies (100%) demonstrated some concerns in deviations from intended interventions domain, reflecting the inherent impossibility of blinding participants and personnel to hypoxic versus normoxic environmental conditions (Figs. 2 and 3). This universal limitation is characteristic of behavioural exercise interventions rather than indicating methodological flaws. Participants necessarily knew whether they were exercising in reduced oxygen environments, creating potential for performance bias and differential behavioural responses between groups. Despite this limitation, all studies maintained protocol fidelity through supervised sessions and standardized exercise prescriptions, minimizing the impact of awareness on intervention delivery. All 10 studies (100%) demonstrated low risk of bias for missing outcome data, with complete or near-complete outcome reporting and appropriate handling of any attrition (Figs. 2 and 3). Dropout rates were generally low (ranging from 0 to 15%) and balanced between groups, with intention-to-treat or complete-case analyses appropriately applied. This consistency across studies strengthens confidence in the completeness of reported outcomes.

Six studies (60%) achieved low risk ratings through objective measurement methods with blinded outcome assessors [39, 40, 42, 44, 47, 48]. These studies employed gold-standard body composition assessment techniques (DXA or calibrated BIA devices) with assessors masked to group allocation. Four studies (40%) raised some concerns, primarily related to reliance on self-reported measures for some outcomes or lack of explicit confirmation of assessor blinding [41, 43, 45, 46]. No studies were rated as high risk in this domain.

Six studies (60%) demonstrated low risk with pre-registered protocols or comprehensive reporting of all pre-specified outcomes [39, 40, 42, 44, 47, 48]. Four studies (40%) raised some concerns due to lack of accessible protocols or potential selective outcome reporting, though there was no clear evidence of omission of unfavourable results [41, 43, 45, 46]. Based on the domain-specific assessments, 9 studies (90%) were rated as having low overall risk of bias when considering that the universal blinding concerns in the deviations domain are inherent to the intervention type and do not reflect poor methodological design [39–42, 44–48]. One study (10%) raised some concerns due to cumulative issues across multiple domains beyond the expected blinding limitations [43]. No studies were judged at high risk of bias across all domains.

The predominant methodological limitation across all studies, the inability to blind participants to hypoxic exposure, was systematically acknowledged in the GRADE certainty assessments, resulting in downgrading for risk of bias across all outcomes (Table 4). This transparent handling of inherent intervention limitations provides appropriate context for interpreting effect estimates and clinical recommendations. Beyond the universal blinding constraint, the included studies generally demonstrated adequate methodological rigor in randomization, outcome measurement, and data completeness, supporting the validity of pooled effect estimates while recognizing the certainty limitations imposed by open-label study designs.

Table 4.

Summary of findings for NHT versus control in obese adults

Outcome and follow-up	Patients (studies), N	Absolute effects (95% CI)	Certainty	What happens
		Difference
BMI	170 (7 RCTs)	0.34 (−0.16 to 0.85)	⨁⨁⨁◯ Moderatea,b	NHT likely results in little to no difference in BMI.NHT probably results in little to no difference in BMI.
Body Fat Percentage	121 (5 RCTs)	0.21 (−3 to 3.41)	⨁◯◯◯ Very lowa,c,d,e	The evidence is very uncertain about the effect of NHT on Body Fat Percentage.
Body Weight	125 (5 RCTs)	0.12 (−1.99 to 2.22)	⨁⨁◯◯ Lowa,f	NHT may result in little to no difference in Body Weight.
Fat Mass	114 (5 RCTs)	1.1 (0.24 to 1.95)	⨁⨁⨁◯ Moderatea,g	NHT likely increases Fat Mass slightly.
Waist Circumference	114 (4 RCTs)	−1.26 (−9.38 to 6.87)	⨁⨁◯◯ Lowa,h	The evidence suggests that NHT results in little to no difference in Waist Circumference.

CI Confidence interval, MD Mean difference

^aSome concerns across majority of studies due to inability to blind participants to hypoxic intervention

^bCI approaches clinical threshold but remains non-significant; moderate sample (n=170)

^cI²=71%, substantial heterogeneity, p=0.001

^dMixed measurement methods (DXA, BIA, skinfolds) with different accuracy profiles

^eVery wide CI (−3.00 to 3.41) crosses both clinical thresholds; small sample

^fWide CI crosses null and includes both benefit and harm; small sample size (n=125)

^gMixed measurement methods

^hExtremely wide CI (−9.38 to 6.87) crosses null AND both clinical decision thresholds; very small sample (n=114)

Primary outcomes: body composition

Body weight

Five studies [42, 43, 45, 47, 48] involving 125 participants (62 NHT, 63 control) provided body weight data (Fig. 4). Individual study mean differences ranged from − 7.10 kg [45] to 1.90 kg [42]. The random-effects meta-analysis showed a small, non-significant effect with a mean difference of 0.12 kg (95% CI: −1.99 to 2.22, p = 0.89). The test for overall effect yielded T = 0.15, df = 4 (p = 0.89). No heterogeneity was detected, with τ² = 0.00 (95% CI: 0.00 to 75.14), Chi² = 1.52, df = 4 (p = 0.82), and I² = 0%. The 95% prediction interval was [−1.99, 2.22], reflecting the confidence interval due to zero between-study variance.

Fig. 4.

Forest plot for body weight change

Body fat percentage

Five studies [41, 42, 44, 47, 48] with 121 participants (62 NHT, 59 control) contributed to the body fat percentage analysis (Fig. 5). Individual study effects showed considerable variation, ranging from − 2.69% [48] to 2.78% [44]. The pooled random-effects analysis yielded a small, non-significant mean difference of 0.21% (95% CI: −3.00 to 3.41, p = 0.87). The test for overall effect demonstrated T = 0.18, df = 4 (p = 0.89). Substantial heterogeneity was observed with τ² = 5.13 (95% CI: 0.48 to 44.77), Chi² = 17.93, df = 4 (p = 0.001), and I² = 71%. The 95% prediction interval extended from − 6.85 to 7.26, indicating considerable uncertainty in the expected range of true effects across populations.

Fig. 5.

Forest plot for body fat percentage change

Body mass index

Seven studies [41–45, 47, 48] comprising 170 participants (87 NHT, 83 control) provided data for BMI analysis (Fig. 6). Individual study effects ranged from − 1.50 kg/m² [45] to 1.20 kg/m² [42]. The random-effects meta-analysis demonstrated a small, non-significant effect favouring NHT with a mean difference of 0.34 kg/m² (95% CI: −0.16 to 0.85, p = 0.15). The test for overall effect yielded T = 1.66, df = 6 (p = 0.15). Heterogeneity analysis revealed τ² = 0.00 (95% CI: 0.00 to 2.66), with Chi² = 2.94, df = 6 (p = 0.82), indicating I² = 0%, suggesting no heterogeneity among studies. The 95% prediction interval was [−0.16, 0.85], identical to the confidence interval due to the absence of between-study variance.

Fig. 6.

Forest plot for BMI change

Fat mass

Five studies [41, 43–45, 47] with 114 participants (59 NHT, 55 control) contributed to fat mass analysis (Fig. 7). Individual study effects varied from − 1.70 kg [45] to 1.40 kg [43]. The pooled analysis demonstrated a significant effect favouring NHT with a mean difference of 1.10 kg (95% CI: 0.24 to 1.95, p = 0.02). The test for overall effect showed T = 3.57, df = 4 (p = 0.02). Heterogeneity assessment revealed τ² = 0.00 with confidence intervals not estimable, Chi² = 0.41, df = 4 (p = 0.98), and I² = 0%, indicating no heterogeneity among studies. The 95% prediction interval was [0.24, 1.95], identical to the confidence interval.

Fig. 7.

Forest plot for fat mass change

Waist circumference

Four studies [39, 41–43] comprising 114 participants (57 NHT, 57 control) provided waist circumference data (Fig. 8). Individual study mean differences ranged from − 8.60 cm [43] to 2.02 cm [39]. The random-effects meta-analysis yielded a small, non-significant effect with a mean difference of −1.26 cm (95% CI: 9.38 to 6.87, p = 0.66). The test for overall effect demonstrated T = 0.49, df = 3 (p = 0.66). Moderate heterogeneity was present with τ² = 12.81 (95% CI: 0.00 to >128.13), Chi² = 5.61, df = 3 (p = 0.13), and I² = 47%. The 95% prediction interval extended from − 15.25 to 12.73, indicating moderate uncertainty in the range of true effects.

Fig. 8.

Forest plot for waist circumference change

Clinical significance assessment

Applying pre-defined clinical significance thresholds, none of the primary outcomes achieved clinically meaningful changes. For BMI, the mean difference of 0.34 kg/m² fell below the 1.0 kg/m² threshold for clinical significance. Body fat percentage showed a 0.21% reduction, which was substantially below the 2.0% clinical threshold. Body weight demonstrated a minimal 0.12 kg difference, far below typical clinical significance thresholds of 3–5% body weight reduction. Fat mass showed a significant statistical effect of 1.10 kg; however, this represented a gain rather than the expected loss. Waist circumference reduction of 0.68 cm was minimal compared to the 5.0 cm threshold typically considered clinically meaningful. The confidence intervals for all outcomes except fat mass included zero, indicating uncertainty about the direction of effect.

Certainty of evidence (GRADE assessment)

The certainty of evidence for primary body composition outcomes ranged from very low to moderate based on GRADE criteria (Table 4). Universal concerns regarding risk of bias reflected the open-label nature of all included trials, with participants and personnel necessarily aware of hypoxic versus normoxic conditions, an inherent limitation of exercise-based hypoxia interventions.

Body mass index (BMI) demonstrated moderate certainty evidence, downgraded once for serious risk of bias. The evidence suggests NHT likely results in little to no clinically meaningful difference in BMI, with the mean difference of 0.34 kg/m² (95% CI: −0.16 to 0.85) falling substantially below the pre-defined clinical significance threshold of 1.0 kg/m². Despite methodological limitations in blinding, the consistency across studies (I²=0%) and adequate sample size (n = 170) support moderate confidence in this null finding.

Body fat percentage showed very low certainty evidence, downgraded for serious risk of bias, serious inconsistency (I²=71%), and serious indirectness related to heterogeneous measurement methods (DXA, BIA, skinfolds with differing accuracy profiles in obese populations). The evidence remains very uncertain regarding NHT effects on body fat percentage, with the trivial mean difference of 0.21% (95% CI: −3.00 to 3.41) obscured by substantial between-study variability and measurement heterogeneity.

Body weight demonstrated low certainty evidence, downgraded for serious risk of bias and serious imprecision characterized by wide confidence intervals (−1.99 to 2.22 kg) crossing null and encompassing both potential benefit and harm with a small sample (n = 125). The evidence suggests NHT may result in little to no difference in body weight, though the broad confidence interval precludes definitive conclusions.

Fat mass showed moderate certainty evidence, downgraded once for serious risk of bias. Notably, this was the only primary outcome demonstrating statistical significance, with NHT associated with a 1.1 kg increase in fat mass (95% CI: 0.24 to 1.95, p = 0.02)—an unfavourable effect direction contrary to hypothesized benefits. The absence of heterogeneity (I²=0%) and statistically significant result support moderate confidence, though the unexpected direction and modest sample size (n = 114) warrant cautious interpretation.

Waist circumference demonstrated low certainty evidence, downgraded for serious risk of bias and serious imprecision. The extremely wide confidence interval (−9.38 to 6.87 cm) crosses null and encompasses both clinically meaningful benefit and harm thresholds (± 5.0 cm), rendering the evidence insufficient to determine NHT effects on central adiposity. The very small sample (n = 114, 4 studies) and wide prediction interval (−15.25 to 12.73 cm) indicate substantial uncertainty for future study findings.

The predominantly moderate to very low certainty ratings across outcomes reflect systematic limitations in the evidence base beyond the universal blinding constraints, including small sample sizes (114–170 participants per outcome), heterogeneous measurement methodologies for body composition assessment, and inconsistent effects across studies for certain outcomes. These certainty assessments indicate that true effects may differ substantially from observed estimates, and future research incorporating larger samples, standardized measurement protocols, and longer intervention durations could materially change conclusions regarding NHT efficacy for obesity management.

Sensitivity analyses

Leave-one-out sensitivity analyses revealed mixed robustness patterns across outcomes. For BMI (k = 7 studies), sequential removal of individual studies resulted in mean differences ranging from 0.21 to 0.41 kg/m² (maximum relative change: 39.9%), though all iterations remained non-significant (p >0.05) with confidence intervals crossing zero. While the magnitude of change exceeded the 20% threshold when removing [48] (39.9% change) and [47] (27.3% change) (Table 5), the clinical interpretation remained unchanged, all estimates fell below the clinical significance threshold of 1.0 kg/m², confirming the null finding is robust.

Table 5.

Leave-one-out sensitivity results

outcome		Study Omitted	No. of studies	NHT (n)	CON (n)	MD (CI)	P value	I2	Change MD	interpretation
BMI	Chacaroun 2020		6	75	72	0.36 (−0.24, 0.96)	0.182	0%	4.40%	Minimal change
	Gatterer 2015		6	71	67	0.33 (−0.24, 0.9)	0.2	0%	−4.30%	Minimal change
	Ghaith 2022		6	71	68	0.41 (−0.06, 0.87)	0.075	0%	17.70%	Minimal change
	Jiao 2024		6	74	74	0.39 (−0.13, 0.9)	0.11	0%	11.80%	Minimal change
	Kong 2014		6	78	74	0.38 (−0.14, 0.89)	0.117	0%	9.80%	Minimal change
	Morishima 2014		6	78	72	0.25 (−0.47, 0.97)	0.411	0%	−27.30%	Substantial change
	Park 2019		6	75	71	0.21 (−0.56, 0.98)	0.519	0%	−39.90%	Substantial change
Body Fat Percentage	Chacaroun 2020		4	50	48	0.46 (−3.94, 4.85)	0.763	78.80%	121.10%	Substantial change
	Gatterer 2015		4	46	43	−0.13 (−4.3, 4.03)	0.926	76.20%	−164.50%	Substantial change
	Jiao 2024		4	49	50	−0.15 (−4.26, 3.96)	0.915	76%	−172.50%	Substantial change
	Morishima 2014		4	53	48	−0.47 (−5, 4.05)	0.761	49.10%	−329.80%	Substantial change
	Park 2019		4	50	47	1.75 (0.22, 3.29)	0.036	0%	751%	Substantial change
Body Weight	Gatterer 2015		4	46	47	0.05 (−2.71, 2.8)	0.962	0%	−60.90%	Substantial change
	Ghaith 2022		4	46	48	0.33 (−1.72, 2.37)	0.647	0%	181.90%	Substantial change
	Kong 2014		4	53	54	0.23 (−2.04, 2.49)	0.772	0%	94.80%	Substantial change
	Morishima 2014		4	53	52	−0.6 (−3.95, 2.74)	0.607	0%	−618.80%	Substantial change
	Park 2019		4	50	51	0.35 (−3.28, 3.98)	0.781	0%	199.40%	Substantial change
Fat Mass	Chacaroun 2020a		4	47	44	1.11 (−0.05, 2.28)	0.056	0%	1.30%	Significance changed
	Ghaith 2022		4	43	40	1.08 (−0.08, 2.24)	0.059	0%	−1.80%	Significance changed
	Jiao 2024		4	46	46	1.15 (0.02, 2.28)	0.048	0%	4.50%	Minimal change
	Kong 2014		4	50	46	1.21 (0.78, 1.65)	0.003	0%	10.60%	Minimal change
	Morishima 2014		4	50	44	0.47 (−1.4, 2.34)	0.481	0%	−57.10%	Substantial change
Waist Circumference	Camacho-Cardenosa 2018		3	44	42	−2.21 (−16.2, 11.77)	0.566	58.20%	−75.80%	Substantial change
	Chacaroun 2020		3	45	46	−2.36 (−16.96, 12.25)	0.559	54.40%	−87.10%	Substantial change
	Gatterer 2015		3	41	41	−1.73 (−16.59, 13.12)	0.665	62.90%	−37.70%	Substantial change
	Ghaith 2022		3	41	42	1.39 (−0.31, 3.09)	0.072	0%	210.50%	Substantial change

MD Mean difference, N Sample size, CI Confidence interval, I²Heterogeneity

Body fat percentage demonstrated substantial sensitivity to individual studies. Removal of [48] changed the effect from non-significant (p = 0.867) to significant (p = 0.036), with the mean difference shifting from 0.21% to 1.75%. However, both estimates remained clinically trivial (below the 2.0% threshold). This study contributed disproportionately to heterogeneity, with I² ranging from 0% to 78.8% across leave-one-out iterations (Table 5). The substantial heterogeneity (I²=70.8%) and sensitivity to single studies indicate considerable uncertainty in body fat percentage effects.

For body weight (k = 5 studies), effect estimates ranged from − 0.60 to 0.35 kg across leave-one-out iterations, with the main finding (MD = 0.12 kg, p = 0.886) showing large relative changes (up to 618.8% when removing [47] (Table 5). This extremely high percentage change reflects the near-zero baseline effect size; however, all iterations remained non-significant with wide confidence intervals crossing zero, and all point estimates were clinically trivial (< 1 kg). The null finding for body weight is therefore considered robust despite high relative percentage changes.

Fat mass, the only outcome demonstrating statistical significance in the main analysis (MD = 1.10 kg, p = 0.023), showed substantial sensitivity to study inclusion. Removal of Morishima 2014 resulted in loss of statistical significance (MD = 0.47 kg, p = 0.481), representing a 57.1% reduction in effect magnitude. Additionally, significance was lost when removing [41] (p = 0.056) or [43] (p = 0.058) (Table 6). Statistical significance was maintained in only 2 of 5 leave-one-out iterations (40%), indicating this finding is NOT robust and depends heavily on the inclusion of specific studies. Combined with the small effect size and unfavourable direction (increased fat mass), this sensitivity pattern substantially weakens confidence in this as a reliable finding.

Table 6.

Influence diagnostics results

Outcome	Study	Cook’s D	DFFITS	Hat Value	Std. Residual	Weight (%)	Influential	Classification
BMI	Chacaroun 2020	0.01	−0.07	0.06	−0.21	5.90%	No	Normal
	Gatterer 2015	0.01	0.07	0.02	0.38	1.70%	No	Normal
	Ghaith 2022	0.09	−0.34	0.04	−1.07	3.60%	No	Normal
	Jiao 2024	0.04	−0.21	0.02	−0.87	2.40%	No	Normal
	Kong 2014	0.03	−0.17	0.02	−0.84	1.80%	No	Normal
	Morishima 2014	0.21	0.43	0.38	0.41	37.70%	No	Normal
	Park 2019	0.44	0.63	0.47	0.49	46.90%	No	Normal
Body Fat Percentage	Chacaroun 2020	0.05	−0.20	0.11	−0.34	11%	No	Normal
	Gatterer 2015	0.09	0.28	0.13	0.68	13%	No	Normal
	Jiao 2024	0.10	0.30	0.13	0.72	12.60%	No	Normal
	Morishima 2014	0.35	0.57	0.30	0.77	30%	No	Normal
	Park 2019	1.80	−7.94	0.33	−1.51	33.40%	YES	High DFFITS
Body Weight	Gatterer 2015	0.01	0.08	0.04	0.29	3.80%	No	Normal
	Ghaith 2022	0.08	−0.33	0.04	−0.85	3.90%	No	Normal
	Kong 2014	0.02	−0.16	0.02	−0.72	1.50%	No	Normal
	Morishima 2014	0.90	0.94	0.48	0.61	47.80%	No	Normal
	Park 2019	0.09	−0.27	0.43	−0.22	43%	No	Normal
Fat Mass	Chacaroun 2020a	0.00	−0.04	0.07	−0.06	6.80%	No	Normal
	Ghaith 2022	0.00	0.06	0.06	0.08	6%	No	Normal
	Jiao 2024	0.03	−0.15	0.08	−0.18	7.50%	No	Normal
	Kong 2014	0.14	−0.87	0.04	−0.59	4%	No	Normal
	Morishima 2014	4.15	2.16	0.76	0.37	75.60%	YES	High DFFITS
Waist Circumference	Camacho-Cardenosa 2018	0.14	0.34	0.22	0.67	22.40%	No	Normal
	Chacaroun 2020	0.18	0.38	0.30	0.69	29.90%	No	Normal
	Gatterer 2015	0.04	0.16	0.21	0.35	21.30%	No	Normal
	Ghaith 2022	1.08	−7.92	0.26	−1.68	26.40%	YES	High DFFITS

Waist circumference (k = 4 studies) demonstrated considerable variability, with effect estimates ranging from − 2.36 to 1.39 cm across leave-one-out iterations (maximum change: 210.5%). Removal of [43] reversed the direction of effect from favourable (−1.26 cm) to unfavourable (+ 1.39 cm) (Table 5), though all iterations remained non-significant. The moderate heterogeneity (I²=47%) and wide confidence intervals indicate substantial uncertainty, though the consistent lack of statistical significance supports the null interpretation.

Influence diagnostics identified several studies with elevated influence statistics, though most remained below critical thresholds (Table 6). Cook’s distance exceeded 1.0 for three studies: [48] for body fat percentage (Cook’s D = 1.80, DFFITS=−7.94) [47], for fat mass (Cook’s D = 4.15, DFFITS = 2.16), and [43] for waist circumference (Cook’s D = 1.08, DFFITS=−7.92). No studies showed standardized residuals exceeding ± 2.5, indicating absence of statistical outliers, though the high DFFITS values for Park 2019 and Ghaith 2022 confirm substantial influence on fitted values for their respective outcomes.

Overall, sensitivity analyses reveal important nuances in robustness where the null findings for BMI and body weight are robust despite some large relative changes. For body fat percentage, the findings were sensitive to [48] making it the source of the heterogeneity. Fat mass shows critical sensitivity concerns in that the significant finding is not robust due to the dependence on specific studies. Lastly. waist circumference’s null finding are robust despite directional changes These patterns indicate that the primary conclusion, NHT does not produce superior body composition outcomes, is generally supported, though confidence is reduced for fat mass given its sensitivity to individual studies. The unfavourable fat mass finding should be interpreted cautiously given it achieves significance in fewer than half of leave-one-out iterations.

Dose-response analysis

Multicollinearity assessment

Correlation analysis revealed substantial interdependence among hypoxia dose parameters (Table 7). The strongest correlations emerged between total exposure hours and the composite hypoxia dose score (r = 0.995), session duration and total exposure hours (r = 0.892), and session duration and the composite dose score (r = 0.898), indicating these variables capture overlapping variance. Intervention duration demonstrated strong positive correlations with total exposure hours (r = 0.945) and the composite dose score (r = 0.932), while showing strong negative correlation with training frequency (r = −0.688), reflecting the inverse relationship typical in exercise prescription where longer interventions often employ lower frequencies. Fractional inspired oxygen (FiO₂) showed weak correlations with other dose parameters (|r| = 0.032 to 0.270), suggesting it captures relatively independent variance related to hypoxic intensity rather than exposure volume.

Table 7.

Multicollinearity diagnostics for hypoxia dose parameters – a pearson correlation matrix

Variable	FiO₂ %	Session Duration	Frequency	Weeks	Total Hours	Composite Dose
FiO₂ %	1.000
Session Duration	−0.270	1.000
Frequency	0.158	−0.249	1.000
Weeks	−0.032	0.710***	−0.688***	1.000
Total Hours	−0.138	0.892***	−0.514	0.945***	1.000
Composite Dose	−0.235	0.898***	−0.524	0.932***	0.995***	1.000

Pairwise Pearson correlation coefficients among dose variables

***p < 0.001 indicating high correlation (|r| >0.7)

Variance inflation factor analysis demonstrated moderate multicollinearity for body mass index (BMI), the only outcome with sufficient data (k = 7 studies) for multivariate assessment (Table 7). Maximum VIF of 6.45 (intervention weeks) exceeded the threshold of 5, indicating moderate collinearity concerns. Individual VIFs were: FiO₂ = 1.25, session duration = 3.82, training frequency = 2.92, and intervention weeks = 6.45. The elevated VIF for intervention weeks reflects its strong correlations with multiple dose parameters. All other outcomes (body weight, body fat percentage, fat mass, waist circumference) included fewer than 5 studies per outcome (k = 4–5), precluding reliable multivariate VIF calculation. Given these multicollinearity patterns and limited study numbers, univariate meta-regression models were prioritized as the primary analytical approach to avoid unstable parameter estimates characteristic of collinear multivariate models.

Univariate dose-response relationships

Fractional inspired oxygen (FiO₂)

FiO₂ concentrations across included studies ranged from 12.0% (severe hypoxia, equivalent to ~ 4,500 m altitude) [43] to 17.2% (mild hypoxia, equivalent to ~ 1,200 m altitude) [39]. Univariate meta-regression revealed no statistically significant dose-response relationships between FiO₂ level and any body composition outcome (Table 8). For BMI (k = 7), the coefficient was 0.371 (95% CI: −0.486 to 1.228, p = 0.396), indicating a non-significant trend where lower FiO₂ (higher hypoxic intensity) associated with slightly greater BMI reductions, though this relationship explained no between-study heterogeneity (R² = 0.0%). Body weight (k = 5) demonstrated a similar non-significant positive coefficient (β = 1.850, 95% CI: −2.182 to 5.882, p = 0.369, R² = 0.0%), as did body fat percentage (β = 1.172, p = 0.602), fat mass (β = 0.039, p = 0.972), and waist circumference (β = 1.484, p = 0.262, R² = 15.5%) (Table 8). Despite waist circumference showing 15.5% heterogeneity explained, wide confidence intervals and non-significance indicated substantial uncertainty. These null findings suggest that within the examined FiO₂ range (12.0–17.2.0.2%), hypoxic intensity does not meaningfully predict body composition outcomes.

Table 8.

Univariate meta-regression examining dose-response relationships between hypoxia parameters and body composition outcomes

outcome	dose variable	k	coefficient	se	CI lower	CI upper	pval	R squared	direction
BMI	Fio2 level %	7	0.371	0.437	−0.486	1.228	0.396	0.000	Lower FiO2 → Greater benefit
BMI	Session duration	7	0.002	0.009	−0.015	0.020	0.789	0.000	Higher dose → Less benefit
BMI	Sessions per week	7	−0.880	0.696	−2.245	0.485	0.206	0.000	Higher dose → Greater benefit
BMI	Intervention weeks	7	0.029	0.061	−0.090	0.149	0.630	0.000	Higher dose → Less benefit
BMI	Total exposure hours	7	0.004	0.009	−0.013	0.021	0.678	0.000	Higher dose → Less benefit
BMI	Hypoxia dose score	7	0.000	0.000	0.000	0.000	0.735	0.000	Higher dose → Less benefit
Body Weight	Fio2 level %	5	1.850	2.057	−2.182	5.882	0.369	0.000	Lower FiO2 → Greater benefit
Body Weight	Session duration	5	−0.011	0.037	−0.082	0.061	0.766	0.000	Higher dose → Greater benefit
Body Weight	Sessions per week	5	−2.902	3.925	−10.594	4.791	0.460	0.000	Higher dose → Greater benefit
Body Weight	Intervention weeks	5	0.001	0.204	−0.399	0.401	0.996	0.000	Higher dose → Less benefit
Body Weight	Total exposure hours	5	−0.002	0.030	−0.061	0.057	0.943	0.000	Higher dose → Greater benefit
Body Weight	Hypoxia dose score	5	0.000	0.000	0.000	0.000	0.914	0.000	Higher dose → Greater benefit
Body Fat Percentage	Fio2 level %	5	1.172	2.246	−3.230	5.574	0.602	0.000	Lower FiO2 → Greater benefit
Body Fat Percentage	Session duration	5	−0.008	0.032	−0.071	0.055	0.809	0.000	Higher dose → Greater benefit
Body Fat Percentage	Sessions per week	5	0.535	1.785	−2.963	4.032	0.765	0.000	Higher dose → Less benefit
Body Fat Percentage	Intervention weeks	5	0.012	0.162	−0.306	0.329	0.943	0.000	Higher dose → Less benefit
Body Fat Percentage	Total exposure hours	5	0.001	0.025	−0.047	0.050	0.961	0.000	Higher dose → Less benefit
Body Fat Percentage	Hypoxia dose score	5	0.000	0.000	0.000	0.000	0.937	0.000	Higher dose → Less benefit
Fat Mass	Fio2 level %	5	0.039	1.138	−2.191	2.270	0.972	0.000	Lower FiO2 → Greater benefit
Fat Mass	Session duration	5	−0.029	0.067	−0.161	0.103	0.664	0.000	Higher dose → Greater benefit
Fat Mass	Sessions per week	5	−0.772	1.502	−3.716	2.173	0.608	0.000	Higher dose → Greater benefit
Fat Mass	Intervention weeks	5	0.011	0.718	−1.396	1.418	0.988	0.000	Higher dose → Less benefit
Fat Mass	Total exposure hours	5	−0.052	0.123	−0.294	0.190	0.675	0.000	Higher dose → Greater benefit
Fat Mass	Hypoxia dose score	5	0.000	0.000	−0.001	0.000	0.769	0.000	Higher dose → Greater benefit
Waist Circumference	Fio2 level %	4	1.484	1.324	−1.111	4.080	0.262	15.510	Lower FiO2 → Greater benefit
Waist Circumference	Session duration	4	−0.013	0.057	−0.124	0.099	0.824	0.000	Higher dose → Greater benefit
Waist Circumference	Sessions per week	4	−2.233	7.341	−16.622	12.156	0.761	0.000	Higher dose → Greater benefit
Waist Circumference	Intervention weeks	4	0.133	0.313	−0.481	0.746	0.672	0.000	Higher dose → Less benefit
Waist Circumference	Total exposure hours	4	0.008	0.045	−0.080	0.095	0.860	0.000	Higher dose → Less benefit
Waist Circumference	Hypoxia dose score	4	0.000	0.000	0.000	0.000	0.957	0.000	Higher dose → Less benefit

Coefficients (β₁) represent the change in mean difference per unit increase in each dose parameter. Positive coefficients indicate that higher doses associate with smaller treatment effects (less favourable outcomes) while negative coefficients indicate that higher doses associate with larger treatment effects (more favorable outcomes). Statistical significance set at p < 0.05. R² represents the proportion of between-study heterogeneity explained by each dose parameter. 

k  number of studies contributing to each analysis, SE  Standard error, CI  Confidence interval

Session duration

Session durations varied from 41.5 [39] to 180 min [42] across protocols (median = 60 min). Session duration showed no significant associations with body composition changes. For BMI (k = 7), the coefficient was near-zero (β = 0.002, 95% CI: −0.015 to 0.020, p = 0.789, R² = 0.0%), indicating that each additional minute of session duration associated with a trivial 0.002 kg/m² change. (Table 8). Body weight (β = −0.011, p = 0.766), body fat percentage (β = −0.008, p = 0.809), fat mass (β = −0.029, p = 0.664), and waist circumference (β = −0.013, p = 0.824) all demonstrated non-significant relationships with similarly minimal coefficients and zero heterogeneity explained (Table 8). The consistently null findings across outcomes suggest that extending individual session duration within the observed range does not enhance body composition benefits, contradicting the theoretical expectation that prolonged hypoxic exposure would amplify metabolic stress and adaptation.

Training frequency

Training frequencies ranged from 2 [42] to 5 sessions per week [44, 45] (median = 3 sessions/week). No significant dose-response relationships emerged between training frequency and body composition outcomes. BMI showed a non-significant negative coefficient (β = −0.880, 95% CI: −2.245 to 0.485, p = 0.206), suggesting higher frequencies might associate with greater reductions, though this relationship explained no heterogeneity (R² = 0.0%). Body weight (β = −2.902, p = 0.460), body fat percentage (β = 0.535, p = 0.765), fat mass (β = −0.772, p = 0.608), and waist circumference (β = −2.233, p = 0.761) all demonstrated non-significant associations with highly variable coefficient directions and zero to minimal heterogeneity explained (Table 8). The lack of consistent frequency-response patterns indicates that increasing training frequency within the studied range (2–5 sessions/week) does not reliably enhance body composition adaptations.

Intervention duration

Intervention durations spanned 4 [45, 47] to 32 weeks [42] (median = 8 weeks). Univariate meta-regression revealed no significant relationships between intervention length and body composition outcomes. For BMI (k = 7), each additional week associated with a trivial 0.029 kg/m² change (95% CI: −0.090 to 0.149, p = 0.630, R² = 0.0%) (Table 8). Body weight approached but did not achieve statistical significance (β = 0.001, 95% CI: −0.399 to 0.401, p = 0.996), with effectively zero coefficient indicating no relationship. Body fat percentage (β = 0.012, p = 0.943), fat mass (β = 0.011, p = 0.988), and waist circumference (β = 0.133, p = 0.672) similarly showed non-significant null relationships (Table 8). The absence of duration-response effects suggests that extending interventions beyond 4 weeks does not produce incremental body composition benefits, contrary to expectations that chronic adaptation requires prolonged exposure.

Total exposure hours

Total cumulative hypoxic exposure ranged from 12 [47] to 192 h [42] (median = 25.2 h). No significant dose-response relationships emerged between total exposure hours and any body composition outcome. Coefficients were consistently small and non-significant across BMI (β = 0.004, p = 0.678), body weight (β = −0.002, p = 0.943), body fat percentage (β = 0.001, p = 0.961), fat mass (β = −0.052, p = 0.675), and waist circumference (β = 0.008, p = 0.860), with all R² values at 0.0% (Table 8). Given that total exposure hours strongly correlates with other dose parameters (r = 0.886–0.995) (Table 7), these null findings reinforce that cumulative hypoxic stimulus, regardless of how distributed across sessions, frequency, or duration, does not predict body composition outcomes.

Composite hypoxia dose score

The composite dose score, integrating FiO₂, session duration, frequency, and intervention duration, ranged from 4,320 to 80,640 arbitrary units (median = 7,920). Despite capturing multidimensional hypoxic exposure, the composite score showed no significant associations with any outcome. Coefficients approached zero across BMI (β < 0.001, p = 0.735), body weight (β < 0.001, p = 0.914), body fat percentage (β < 0.001, p = 0.937), fat mass (β < 0.001, p = 0.769), and waist circumference (β < 0.001, p = 0.957), with all R² = 0.0% (Table 8). The composite score’s strong correlations with component variables (r = 0.900–0.995.900.995) suggest it primarily reflects total exposure volume, yet the null findings indicate that neither individual dose parameters nor their integration predicts body composition efficacy.

Absence of multivariate dose-response models

Multivariate meta-regression incorporating multiple dose parameters simultaneously was precluded by two methodological constraints. First, moderate multicollinearity (max VIF = 6.45) among dose variables for BMI indicated shared variance would inflate standard errors and produce unstable parameter estimates in multivariate specifications. Second, insufficient study numbers for all outcomes except BMI (k = 4–5 versus required k ≥ 5) limited statistical power for reliable multivariate modelling (Table 7). The strong correlations between dose parameters (r = 0.688–0.995) further contraindicated multivariate approaches, as highly correlated predictors yield unreliable individual coefficient estimates even when the overall model fits adequately. Consequently, univariate models represent the most appropriate analytical strategy given the available data, though they cannot assess independent effects of dose parameters while controlling for other parameters.

Interpretation of null dose-response findings

The systematic absence of statistically significant dose-response relationships across all examined parameters and outcomes indicates that within the ranges studied, hypoxia dose parameters do not meaningfully predict body composition efficacy. All 30 univariate dose-response models (6 dose variables × 5 outcomes) yielded non-significant relationships (all p > 0.05), with 29 of 30 models explaining zero between-study heterogeneity (R² = 0.0%) and only one model (waist circumference–FiO₂) explaining modest heterogeneity (R² = 15.5%, p = 0.262, non-significant) (Table 8). The consistent null findings hold across multiple dose conceptualizations, intensity (FiO₂), session characteristics (duration), training prescription (frequency), intervention length (weeks), cumulative exposure (total hours), and integrated dose (composite score), suggesting that dosing parameters within studied ranges do not modulate NHT efficacy for body composition adaptation Table 9.

Table 9.

Variance inflation factors

Outcome	k	FiO₂ VIF	Duration VIF	Frequency VIF	Weeks VIF	Max VIF	Assessment
BMI	7	1.25	3.82	2.92	6.45	6.45	Moderate concern

Variance inflation factors (VIF) for multivariate meta-regression model predicting BMI. VIF < 5 indicates acceptable collinearity; 5–10 indicates moderate concern requiring cautious interpretation; >10 indicates severe multicollinearity precluding reliable multivariate estimation. Only BMI had sufficient studies (k ≥ 5) for VIF calculation.

The failure to identify dose-response gradients may reflect genuine biological phenomena where hypoxia dose-response relationships are non-monotonic, threshold-based, or contingent on individual characteristics not captured in available moderators. Alternatively, the limited sample sizes (k = 4–7 per outcome), restricted dose ranges examined, and substantial measurement heterogeneity across studies may have precluded detection of true dose-response effects. The strong correlations among dose parameters (r = 0.688–0.995) indicate protocols typically manipulate multiple parameters simultaneously rather than systematically varying individual components, further limiting ability to isolate specific dose-response effects. These methodological constraints underscore the preliminary nature of dose-response evidence and the need for future trials employing factorial designs that systematically vary individual dose parameters while holding others constant.

Secondary outcomes

Metabolic health parameters

The metabolic effects of normobaric hypoxia training extend beyond body composition changes to encompass fundamental alterations in glucose and lipid metabolism. Exposure to hypoxic conditions during exercise may enhance metabolic adaptations through multiple mechanisms, including improved insulin sensitivity, enhanced glucose uptake, and favourable modifications in lipid metabolism pathways [1, 4, 8]. The activation of hypoxia-inducible factor-1 (HIF-1) under reduced oxygen conditions stimulates metabolic reprogramming that may optimize substrate utilization and improve overall metabolic health [2, 3]. Understanding the impact of normobaric hypoxia training on key metabolic biomarkers provides crucial insights into the therapeutic potential of this intervention for managing metabolic dysfunction associated with obesity. Three metabolic health outcomes demonstrated sufficient data availability (≥ 4 studies) for meta-analysis: fasting glucose, total cholesterol, and triglycerides. These biomarkers represent critical indicators of metabolic health, with fasting glucose reflecting glucose homeostasis and insulin sensitivity, while lipid parameters indicate cardiovascular risk and metabolic syndrome status. The analysis of these outcomes provides essential evidence for the metabolic benefits of normobaric hypoxia training beyond traditional body composition improvements.

Fasting glucose

The meta-analysis of fasting glucose included 4 studies [39, 41, 43, 47] 103 participants (50 NHT, 53 control). Given that studies used different laboratory assay methods and units (mg/dL vs. mmol/L), standardized mean differences were calculated to enable comparison across studies. The pooled analysis revealed a trivial, non-significant effect with an SMD of −0.14 (95% CI: −0.87 to 0.59, p = 0.59) (Fig. 9). The test for overall effect yielded T = 0.60, df = 3 (p = 0.59). Low heterogeneity was detected among studies (I² = 30%, τ² = 0.07, 95% CI: 0.00 to 2.61, Chi² = 4.08, df = 3, p = 0.25), indicating relatively consistent findings across investigations. The prediction interval [−1.25, 0.97] suggests that future studies would likely demonstrate similarly modest effects on fasting glucose levels. Individual study effects ranged from − 0.59 [43] to 0.42 [39], with most studies showing minimal impact on glucose homeostasis. According to Cohen’s effect size conventions (0.2 = small, 0.5 = moderate, 0.8 = large), the observed SMD of −0.14 represents a trivial effect size that falls well below the threshold for clinical meaningfulness [11, 51]. The lack of significant glucose-lowering effects may reflect the relatively healthy metabolic status of participants at baseline, with mean fasting glucose values within normal ranges across most included studies, or insufficient intervention intensity and duration to induce meaningful glucose metabolism adaptations. The low heterogeneity suggests that this null finding is consistent across different study populations, intervention protocols, and hypoxic exposure parameters.

Fig. 9.

Forest plot for fasting glucose change following normobaric hypoxia training

Total cholesterol

Four studies [39, 41, 43, 47] comprising 103 participants (50 NHT, 53 control) contributed to the total cholesterol analysis. Due to different laboratory assay methods and measurement units across studies, standardized mean differences were calculated. The meta-analysis demonstrated a small, non-significant effect with an SMD of 0.33 (95% CI: −0.17 to 0.82, p = 0.13). The test for overall effect yielded T = 2.08, df = 3 (p = 0.13). No heterogeneity was observed between studies (I² = 0%, τ² = 0.00, 95% CI: 0.00 to 1.09, Chi² = 1.83, df = 3, p = 0.61) (Fig. 10), indicating consistent effects across investigations. The narrow prediction interval [−0.17, 0.82] reflects minimal uncertainty in expected treatment effects for future studies, though the confidence interval crossing zero indicates uncertainty about the direction of effect. Effect sizes ranged from 0.00 [43] to 0.63 [39], with all individual studies showing effects favouring control or no difference. The pooled SMD of 0.33 represents a small effect size according to Cohen’s conventions, suggesting a slight trend toward higher total cholesterol in the NHT group, though this did not reach statistical significance [11, 51]. Contrary to the hypothesized cholesterol-lowering benefits of hypoxic exercise training, these findings indicate that NHT does not provide advantages over normoxic exercise for improving lipid profiles. The absence of heterogeneity (I² = 0%) suggests this finding is robust across different intervention protocols, populations, and study durations. The consistent direction of effect across studies, while not statistically significant, raises questions about whether hypoxic exposure during exercise may interfere with the cholesterol-lowering adaptations typically observed with aerobic training under normoxic conditions.

Fig. 10.

Forest plot for total cholesterol change following normobaric hypoxia training

Triglycerides

The triglycerides analysis included 4 studies [39, 41, 43, 47] with 103 participants (50 NHT, 53 control). Standardized mean differences were calculated to account for different laboratory assay methods and measurement units across studies. Results showed a small, non-significant effect with an SMD of 0.40 (95% CI: −1.91 to 2.71, p = 0.62). The test for overall effect yielded T = 0.55, df = 3 (p = 0.62). Substantial heterogeneity was observed among studies (I² = 90%, τ² = 1.77, 95% CI: 0.38 to 31.00, Chi² = 19.97, df = 3, p = 0.0002), indicating considerable variation in triglyceride responses across investigations (Fig. 11). The wide prediction interval [−4.42, 5.22] reflects substantial uncertainty in expected treatment effects for future studies, with the range encompassing both large reductions and large increases in triglycerides. Individual study effects demonstrated marked variability, ranging from − 0.65 [39] to 2.67 [47]. This represents a spread of over 3 standard deviations, indicating fundamentally different responses across study populations or protocols. The substantial heterogeneity (I² = 90%) suggests that triglyceride responses to normobaric hypoxia training are highly variable and likely influenced by factors such as baseline triglyceride levels, dietary intake patterns, intervention duration, hypoxic exposure parameters, or specific population characteristics that were not consistently controlled or reported across studies [47]. study showed a large effect size favouring control (SMD = 2.67), indicating substantially higher triglycerides in the NHT group, while [39] showed a moderate reduction in the NHT group. This inconsistency prevents drawing definitive conclusions about the effect of NHT on triglycerides and indicates the need for further research to identify potential moderators of this response.

Fig. 11.

Forest plot for triglycerides change following normobaric hypoxia training

The metabolic health outcomes reveal a pattern of null to potentially adverse effects from normobaric hypoxia training. None of the three analysed metabolic parameters showed statistically significant improvements with NHT compared to normoxic exercise. Fasting glucose demonstrated a trivial effect size (SMD = −0.14) with low heterogeneity, indicating consistent lack of benefit across studies. Total cholesterol showed a small, non-significant effect favouring control (SMD = 0.33) with no heterogeneity, suggesting NHT may not provide the lipid-lowering benefits of conventional exercise. Triglyceride responses were highly variable (I² = 90%) and non-significant, indicating inconsistent and unpredictable effects that may depend on unmeasured study or participant characteristics. These findings contrast with the hypothesized metabolic benefits of hypoxic exercise and suggest that NHT does not offer metabolic advantages over normoxic exercise for adults with obesity. The absence of improvements in key metabolic biomarkers, combined with the null findings for body composition outcomes, strengthens the conclusion that current evidence does not support NHT as a superior intervention for obesity management.

Cardiovascular and performance parameters

Cardiovascular and exercise performance adaptations represent fundamental physiological responses to normobaric hypoxia training that extend beyond metabolic and body composition changes. The cardiovascular system demonstrates remarkable plasticity in response to hypoxic stimuli, with adaptations including improved cardiac efficiency, enhanced oxygen delivery capacity, and favourable blood pressure modifications [1, 2]. Exercise performance improvements under hypoxic conditions may result from enhanced oxygen utilization efficiency, increased cardiac output, and improved peripheral oxygen extraction [3, 4, 52]. These cardiovascular adaptations are particularly relevant for individuals with obesity, who often present with elevated cardiovascular risk factors and reduced exercise capacity. The physiological mechanisms underlying cardiovascular improvements during normobaric hypoxia training involve activation of the sympathetic nervous system, enhanced nitric oxide production, and improved endothelial function [4, 52]. Hypoxic exposure stimulates angiogenesis and mitochondrial biogenesis, leading to improved tissue oxygenation and enhanced aerobic capacity. Additionally, the reduced mechanical stress associated with exercising at lower intensities under hypoxic conditions may provide cardiovascular benefits while minimizing injury risk in obese populations. Three cardiovascular and performance outcomes demonstrated sufficient data availability (≥ 4 studies) for meta-analysis: diastolic blood pressure, systolic blood pressure, and VO₂peak. These parameters represent critical indicators of cardiovascular health and functional capacity, providing essential insights into the therapeutic potential of normobaric hypoxia training for cardiovascular risk reduction and performance enhancement.

Diastolic blood pressure

The meta-analysis of diastolic blood pressure included 5 studies [39, 41, 42, 44, 45] with 123 participants (63 NHT, 60 control). The pooled analysis revealed a modest, non-significant effect with a mean difference of 3.56 mmHg (95% CI: −1.52 to 8.63, p = 0.12). Moderate heterogeneity was observed among studies (I² = 35%, τ² = 5.82, Chi² = 6.19, p = 0.19), suggesting some variability in diastolic blood pressure responses across investigations. Individual study effects ranged from − 1.08 mmHg [44] to 7.40 mmHg [42], with most studies showing favourable trends toward control (Fig. 12). The prediction interval [−4.86, 11.97] indicates considerable uncertainty in expected treatment effects for future studies. While the overall effect did not reach statistical significance, the consistent direction of effect across most studies suggests potential cardiovascular benefits that may become more apparent with larger sample sizes or longer intervention durations.

Fig. 12.

Forest plot for diastolic blood pressure change following normobaric hypoxia training

Systolic blood pressure

Five studies [39, 41, 42, 44, 45] comprising 123 participants (63 NHT, 60 control) contributed to the systolic blood pressure analysis (Fig. 13). The meta-analysis demonstrated a small, non-significant effect with a mean difference of 3.25 mmHg (95% CI: −5.51 to 12.01, p = 0.36). Moderate heterogeneity was detected between studies (I² = 45%, τ² = 21.32, Chi² = 7.16, p = 0.13), indicating variability in systolic blood pressure responses across different study populations and intervention protocols. Individual study effects showed considerable variation, ranging from − 6.80 mmHg [44] to 10.90 mmHg [42]. The wide prediction interval [−12.43, 18.94] reflects substantial uncertainty in expected treatment effects. While some studies demonstrated favourable systolic blood pressure reductions, others showed increases, suggesting that individual responses to normobaric hypoxia training may be influenced by factors such as baseline blood pressure, intervention intensity, or participant characteristics.

Fig. 13.

Forest plot for systolic blood pressure change following normobaric hypoxia training

VO₂peak

The VO₂peak analysis included 5 studies [39, 41–43, 47] with 186 participants (94 NHT, 92 control). Results showed a small, non-significant effect with a mean difference of 1.43 mL/kg/min (95% CI: −0.86 to 3.72, p = 0.16). The test for overall effect yielded T = 1.74, df = 4 (p = 0.16). Moderate heterogeneity was observed among studies (I² = 62%, τ² = 1.94, 95% CI: 0.00 to 28.11, Chi² = 9.55, df = 4, p = 0.05) (Fig. 14), indicating variability in aerobic capacity responses across different study populations and intervention protocols. Individual study effects showed considerable variation, ranging from − 0.20 mL/kg/min [41] to 4.42 mL/kg/min [40]. The wide prediction interval [−3.06 to 5.93] reflects substantial uncertainty in expected treatment effects for future studies. While the pooled estimate suggests a potential improvement in VO₂peak with NHT of approximately 1.43 mL/kg/min, this effect did not reach statistical significance and the confidence interval included both clinically meaningful improvements and trivial effects. The moderate heterogeneity (I² = 62%) suggests that VO₂peak responses to normobaric hypoxia training may be influenced by factors such as baseline fitness levels, hypoxic exposure parameters, intervention duration, or population characteristics that require further investigation. The observed heterogeneity indicates inconsistent findings across studies, with some demonstrating substantial improvements (Camacho-Cardenosa et al., 2020: 4.42 mL/kg/min; Ghaith et al., 2022: 2.80 mL/kg/min) while others showed minimal or no benefit (Chacaroun et al., 2020: −0.20 mL/kg/min; Gatterer et al., 2015: 0.10 mL/kg/min).

Fig. 14.

Forest plot for VO₂peak change following normobaric hypoxia training

The cardiovascular and performance outcomes demonstrate mixed effects of normobaric hypoxia training. While the improvement in VO₂peak did not reach statistical significance (MD = 1.43 mL/kg/min, p = 0.16), the direction of effect and magnitude of improvement in some studies suggest potential cardiovascular benefits that may warrant further investigation with larger sample sizes or longer intervention periods. The moderate heterogeneity (I² = 62%) indicates that aerobic capacity responses to NHT are inconsistent across studies, suggesting that effectiveness may depend on specific intervention characteristics or population factors. Blood pressure effects similarly did not reach statistical significance, though consistent trends toward reduction across most studies suggest potential cardiovascular benefits. These findings indicate that normobaric hypoxia training may have some value for cardiovascular conditioning in specific contexts, though the current evidence base is insufficient to support definitive conclusions about its superiority over normoxic exercise for aerobic capacity enhancement in obesity management programs.

Discussion

Summary of findings

This systematic review and meta-analysis represents the first comprehensive dose-response investigation of normobaric hypoxia training in obesity management, synthesizing evidence from 10 randomized controlled trials involving 301 participants. The findings reveal no superior body composition benefits from normobaric hypoxia training compared with equivalent normoxic exercise across all primary outcomes. Body weight (MD = 0.12 kg, 95% CI: −1.99 to 2.22, p = 0.89), body fat percentage (MD = 0.21%, 95% CI: −3.00 to 3.41, p = 0.87), BMI (MD = 0.34 kg/m², 95% CI: −0.16 to 0.85, p = 0.15), and waist circumference (MD = −1.26 cm, 95% CI: −9.38 to 6.87, p = 0.66) all demonstrated trivial, non-significant effects. Fat mass showed a statistically significant increase in NHT groups (MD = 1.10 kg, 95% CI: 0.24 to 1.95, p = 0.02), though sensitivity analyses indicated this finding was not robust to individual study inclusion. None of the primary outcomes achieved pre-defined clinical significance thresholds. Comprehensive dose-response meta-regression examining six hypoxia parameters revealed no statistically significant relationships with any body composition outcome, with 29 of 30 univariate models explaining zero between-study heterogeneity (R² = 0.0%). Moderate multicollinearity among dose variables (max VIF = 6.45) precluded reliable multivariate modelling. Cardiovascular fitness showed a non-significant trend toward improvement (VO₂peak: MD = 1.43 mL/kg/min, 95% CI: −0.86 to 3.72, p = 0.16) with moderate heterogeneity (I² = 62%).

Comparison with previous meta-analyses

The null findings for body composition outcomes align with recent systematic reviews examining hypoxic training in obesity [3]. reported no significant differences between hypoxic and normoxic exercise training for body composition outcomes (SMD = −0.10, 95% CI: −0.20 to −0.01, p = 0.90, I² = 0%), suggesting theoretical benefits of hypoxic exercise may be systematically offset by reductions in training quality or exercise intensity under oxygen-restricted conditions. Similarly, the current findings corroborate [1], who found inconsistent body composition effects across diverse hypoxia protocols, with benefits appearing highly dependent on specific population characteristics, intervention parameters, or outcome measurement methods not adequately captured in available moderator analyses.

The non-significant trend toward VO₂peak improvement (MD = 1.43 mL/kg/min) shows directional consistency with previous meta-analyses reporting cardiovascular benefits from hypoxic training, though the magnitude and statistical significance differ [1, 3, 6]. This discrepancy may reflect the current review’s focus on adults with obesity specifically, whereas prior reviews included broader populations with higher baseline fitness levels potentially more responsive to hypoxic stimuli. The moderate heterogeneity (I² = 62%) observed for VO₂peak suggests response variability that warrants investigation of population-specific moderators.

Mechanistic disconnect between theory and practice

The absence of superior body composition benefits exposes a critical disconnect between acute physiological responses observed in laboratory settings and chronic adaptations in real-world interventions. While acute studies consistently demonstrate enhanced fat oxidation, increased energy expenditure, and activation of hypoxia-inducible factor-1 pathways during hypoxic exercise [4], these mechanistic advantages appear insufficient to overcome practical limitations imposed by reduced exercise capacity under hypoxic conditions. The theoretical premise that hypoxic exposure enhances metabolic stress beyond normoxic exercise at equivalent absolute intensities may be undermined by participants’ inability to maintain training intensity, duration, or adherence when exercising under oxygen restriction.

He et al. [8] demonstrated age-specific variations in hypoxic training responsiveness, with middle-aged and older adults showing significant BMI reductions (MD = −0.92, 95% CI: −1.28 to −0.55, p < 0.00001) following hypoxia conditioning. These findings suggest population characteristics, particularly age-related differences in metabolic flexibility, cardiovascular reserve, and baseline fitness, may critically determine intervention effectiveness. The current analysis, spanning ages 19.8 to 66.5 years with considerable heterogeneity, may obscure age-specific effects that could emerge with adequately powered subgroup analyses.

Dose-response relationships and protocol optimization

The systematic absence of statistically significant dose-response relationships across all examined parameters challenges fundamental assumptions about optimal hypoxic training prescription. All 30 univariate dose-response models (6 dose variables × 5 outcomes) yielded non-significant relationships (all p >0.05), with 29 of 30 models explaining zero between-study heterogeneity (R² = 0.0%). The failure to identify clear dose-response gradients may reflect genuine biological phenomena where hypoxia effects are non-monotonic, threshold-based, or contingent on individual characteristics not captured in available moderators [3, 6, 53]. Alternatively, limited sample sizes (k = 4–7 per outcome), restricted dose ranges examined, and substantial measurement heterogeneity across studies may have precluded detection of true dose-response effects.

The moderate multicollinearity observed among dose parameters (r = 0.688–0.995 for some variable pairs, max VIF = 6.45) indicates protocols typically manipulate multiple parameters simultaneously rather than systematically varying individual components while holding others constant. This collinearity pattern limits ability to isolate specific dose-response effects and suggests future trials employing factorial designs are needed to disentangle independent contributions of FiO₂ level, session duration, training frequency, and intervention duration to body composition adaptations [32, 53, 54].

Methodological limitations and clinical implications

Several methodological limitations warrant consideration when interpreting findings. The heterogeneity in study methodologies, including FiO₂ levels (12.0–17.2.0.2%), session durations (41.5–180 min), training frequencies (2–5 sessions/week), and intervention periods (4–32 weeks), creates a diverse methodological landscape that both strengthens external validity and complicates identification of optimal protocols. The predominance of bioelectrical impedance analysis for body composition assessment introduces measurement error that may obscure small treatment effects, particularly for body fat percentage [55–57]. Gold-standard dual-energy X-ray absorptiometry was employed in only a subset of studies, limiting precision of effect estimates for the most clinically relevant outcomes.

The relatively short intervention durations in most studies (median 8 weeks) may be insufficient to capture meaningful body composition changes. Traditional exercise interventions typically require 12–16 weeks to produce clinically significant improvements [58, 59], and hypoxic stress addition may necessitate longer adaptation periods due to complex physiological adjustments required for optimal performance under oxygen-restricted conditions. The universal blinding limitation, all studies demonstrated “some concerns” regarding deviations from intended interventions, reflects the inherent impossibility of masking participants to hypoxic versus normoxic conditions. This limitation, systematically acknowledged in GRADE assessments resulting in certainty downgrades, represents a fundamental constraint of behavioural interventions rather than methodological failure.

GRADE assessment revealed low to very low certainty evidence for most outcomes, primarily due to risk of bias (inherent blinding limitations), imprecision (wide confidence intervals crossing null), inconsistency (substantial heterogeneity for some outcomes), and indirectness (heterogeneous measurement methods). These certainty ratings indicate true effects may differ substantially from observed estimates [1, 60, 61], and future research with larger samples, standardized protocols, and longer interventions could materially change conclusions regarding NHT efficacy.

Sensitivity of the results

Comprehensive sensitivity analyses provide important context for interpreting the study findings. Leave-one-out analyses and influence diagnostics revealed that while most null findings were robust across analytical approaches, the statistical significance of fat mass increase was notably sensitive to individual study inclusion, achieving significance in only 40% of leave-one-out iterations. This sensitivity, combined with influence diagnostics showing [47] exceeded Cook’s distance threshold (D = 4.15), indicates the unfavourable fat mass finding warrants cautious interpretation.

The identification of [48] as highly influential for body fat percentage (Cook’s D = 1.80, contributing substantially to I²=70.8%) highlights the impact of measurement heterogeneity in this outcome [48]. showed the largest beneficial effect (−2.69%) using a specific body composition method, while other studies using different methods showed minimal or opposite effects. This pattern underscores the critical need for standardized body composition assessment protocols in future hypoxia research, as different measurement techniques (DXA, BIA, skinfolds) may show different sensitivities to intervention effects in individuals with obesity.

Despite these sensitivities, the overarching conclusion that NHT does not produce superior body composition outcomes compared to normoxic exercise remains supported. For the four outcomes showing null effects in main analyses (body weight, BMI, body fat percentage, waist circumference), no leave-one-out iteration changed the clinical interpretation of “no meaningful difference.” This consistency across sensitivity approaches, combined with GRADE assessments indicating low to very low certainty evidence, strengthens confidence that current evidence does not support NHT superiority for obesity management.

The sensitivity patterns also provide direction for future research. The fat mass finding requires replication in larger samples with standardized measurement protocols before being considered a reliable adverse effect. Studies should report detailed body composition methodologies and consider multiple measurement approaches to assess consistency of findings across techniques.

Clinical translation and implementation considerations

The absence of superior body composition benefits demonstrated in this meta-analysis has important implications for clinical decision-making and resource allocation. Normobaric hypoxia training systems require substantial capital investment, specialized equipment maintenance, and trained personnel for safe operation, costs that cannot be justified based on current evidence of equivalent outcomes to conventional exercise training. The consistent cardiovascular fitness improvements observed across studies, while not achieving statistical significance in the current analysis, suggest NHT may offer value for specific applications where aerobic capacity enhancement is the primary objective rather than body composition improvement [3, 8, 62].

Healthcare providers implementing NHT should maintain realistic expectations about body composition outcomes and consider hypoxic training as a potential adjunct for exercise variety or adherence enhancement rather than a superior alternative to established approaches. Comprehensive medical screening to identify contraindications, gradual acclimatization protocols, and continuous monitoring during initial sessions remain essential safety considerations. The lack of clear dose-response relationships suggests conservative approaches using moderate hypoxic exposures (15–16% FiO₂) are appropriate from both safety and effectiveness perspectives until additional evidence clarifies optimal dosing parameters.

Future research directions

Future research should prioritize large-scale randomized controlled trials with standardized protocols, validated outcome measures, and intervention durations sufficient to capture meaningful body composition changes (minimum 16 weeks). Mechanistic studies investigating interactions between hypoxic exposure parameters and exercise training variables are essential for understanding optimal protocol design. Research examining energy expenditure, substrate utilization patterns, hormonal responses, and appetite regulation under various hypoxic training conditions could identify factors determining individual responsiveness.

The development of personalized approaches predicting individual responses to hypoxic training represents a critical research priority. Genetic polymorphisms affecting hypoxic sensitivity, baseline fitness levels, metabolic flexibility, and other physiological characteristics may determine who benefits most from hypoxic interventions. Cost-effectiveness analyses comparing hypoxic training with established obesity interventions are needed to inform healthcare policy decisions, considering not only direct intervention costs but also long-term health outcomes, quality of life measures, and healthcare utilization patterns.

Conclusion

This systematic review and meta-analysis provides comprehensive dose-response evaluation of normobaric hypoxia training for obesity management, synthesizing evidence from 10 randomized controlled trials involving 301 participants. The findings demonstrate no superior body composition outcomes from normobaric hypoxia training compared with equivalent normoxic exercise in adults with obesity. Across all primary outcomes, body weight, BMI, body fat percentage, fat mass, and waist circumference, effect sizes were trivial and clinically non-meaningful, with none achieving pre-defined minimal clinically important differences. The statistically significant increase in fat mass observed among NHT participants was not robust to sensitivity analyses, limiting confidence in this as a reliable finding.

Comprehensive dose-response meta-regression examining six hypoxia parameters (FiO₂, session duration, training frequency, intervention duration, total exposure hours, and composite dose score) revealed no statistically significant relationships with any body composition outcome. The systematic absence of dose-response gradients, combined with moderate multicollinearity among dose variables, indicates current evidence cannot identify optimal hypoxic exposure parameters for body composition improvement. The non-significant trend toward cardiovascular fitness improvement (VO₂peak: MD = 1.43 mL/kg/min, p = 0.16) with moderate heterogeneity (I² = 62%) suggests potential benefits for aerobic capacity enhancement, though further research is needed to establish efficacy and identify responsive populations.

GRADE assessment revealed low to very low certainty evidence for body composition outcomes, primarily due to inherent blinding limitations in behavioural interventions, imprecision with wide confidence intervals, and measurement heterogeneity. These certainty ratings indicate future research may substantially alter effect estimates and clinical recommendations. The absence of body composition benefits, combined with increased intervention complexity, cost, and equipment requirements, indicates current evidence does not support normobaric hypoxia training as a preferred alternative to conventional exercise training for obesity management.

The findings underscore the importance of rigorous evidence evaluation before widespread implementation of novel exercise interventions. While normobaric hypoxia training may have value for specific applications focused on cardiovascular conditioning or in populations demonstrating particular responsiveness (e.g., older adults, as suggested by subgroup findings in previous reviews), conventional exercise approaches remain the evidence-based standard for body composition improvement in adults with obesity. Future research employing factorial designs to systematically vary individual dose parameters, longer intervention durations (≥ 16 weeks), standardized body composition assessment methods (DXA), and adequately powered sample sizes is essential to definitively characterize the role of hypoxic training in obesity treatment and identify potential responder subgroups who may derive meaningful benefits from this intervention approach.

Authors’ contributions

K.YZ and W.JF was responsible for the overall design and coordination of the study. He led the data analysis and was involved in interpreting the results and drafting the manuscript.K.YZ as the corresponding author, was responsible for the final review of the manuscript and ensuring the academic integrity of the research. He also participated in the study design and discussion of the results.Q.YF and Y.TW and L.YQ was involved in the study design and was responsible for data collection. He also analyzed the results and assisted in writing.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Acknowledgements.

Data availability

All data comes from references and is publicly available.

Declarations

Ethical approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.