High-Intensity Interval Training (HIIT) in Hypoxia Improves Maximal Aerobic Capacity More Than HIIT in Normoxia: A Systematic Review, Meta-Analysis, and Meta-Regression

The present study aimed to determine the effect of high intensity interval training (HIIT) in hypoxia on maximal oxygen uptake (VO2max) compared with HIIT in normoxia with a Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA)-accordant meta-analysis and meta-regression. Studies which measured VO2max following a minimum of 2 weeks intervention featuring HIIT in hypoxia versus HIIT in normoxia were included. From 119 originally identified titles, nine studies were included (n = 194 participants). Meta-analysis was conducted on change in (∆) VO2max using standardised mean difference (SMD) and a random effects model. Meta-regression examined the relationship between the extent of environmental hypoxia (fractional inspired oxygen [FiO2]) and ∆VO2max and intervention duration and ∆VO2max. The overall SMD for ∆VO2max following HIIT in hypoxia was 1.14 (95% CI = 0.56–1.72; p < 0.001). Meta-regressions identified no significant relationship between FiO2 (coefficient estimate = 0.074, p = 0.852) or intervention duration (coefficient estimate = 0.071, p = 0.423) and ∆VO2max. In conclusion, HIIT in hypoxia improved VO2max compared to HIIT in normoxia. Neither extent of hypoxia, nor training duration modified this effect, however the range in FiO2 was small, which limits interpretation of this meta-regression. Moreover, training duration is not the only training variable known to influence ∆VO2max, and does not appropriately capture total training stress or load. This meta-analysis provides pooled evidence that HIIT in hypoxia may be more efficacious at improving VO2max than HIIT in normoxia. The application of these data suggest adding a hypoxic stimuli to a period of HIIT may be more effective at improving VO2max than HIIT alone. Therefore, coaches and athletes with access to altitude (either natural or simulated) should consider implementing HIIT in hypoxia, rather than HIIT in normoxia where possible, assuming no negative side effects.


Rationale
A combination of reduced barometric pressure (PB), or a reduced effective inspired fraction of oxygen (FiO 2 ), leads to reduced inspired partial pressure of oxygen (PaO 2 ) which ultimately results in the physiological state of hypoxia [1,2]. In human research concerning hypoxia, hypoxia can be examined through two means, firstly hypobaric hypoxia (PB < 760 mmHg; FiO 2 = 20.9%) which generally reflects the state found on earth at altitude, and normobaric hypoxia (PB = 760 mmHg; FiO 2 < 20%) which can be considered simulated altitude [3]. Chronic exposure to natural altitude stimulates renal production of erythropoietin (EPO), driving increased haemoglobin mass (Hb mass) and red blood cell (RBC) count [4,5]. This increases oxygen carrying capacity of the blood, with well reported associations between increased RBC and Hb mass and improved Int. J. Environ. Res. Public Health 2022, 19, 14261 2 of 15 maximal oxygen uptake (VO 2max ) [6,7]. Therefore, athletes have been recommended to spend prolonged periods of time at moderate (2000-3000 m), to high altitude (>3000 m), to stimulate erythropoiesis [8]. Thus, 'altitude training camps' are widely used by professional and recreational athletes alike [9], and the early 1990s saw the popularisation of the 'livehigh, train-low' (LHTL) paradigm by Levine and Stray-Gundersen [6]. In a pooled analysis of six previous experiments, studies adopting the LHTL and LHTH paradigms have shown improved test performance at sea-level, with a~3% increase in VO 2max following altitude training compared to control (i.e., normoxic) participants [10]. More recently, hypoxic training methods have been further developed into live high train high (LHTH), intermittent hypoxic exposure at rest, and live low train high (LLTH) [11][12][13].
Recently, HIIT training has been combined with hypoxia training with the aim of eliciting optimal training adaptations. LLTH methods allow athletes to continue to live at normoxia, whilst exposed to acute periods of hypoxia during training. Within LLTH there are different training methodologies, including continuous hypoxic training (CPT), interval hypoxic training (IHT), and repeated sprint training in hypoxia (RST). Several original investigations have now been conducted examining either HIIT or RST in hypoxia [12,[34][35][36]. Gatterer et al. [37] published results from a pilot study noting HIIT and RST in hypoxia improved sea-level performance of the repeated sprint ability (RSA) and a Yo-Yo intermittent recovery test 2 (YYIR2), in addition to muscle re-oxygenation. Studies investigating physiological adaptations during HIIT in hypoxia have reported that different training methods in hypoxia elicit different training effects, including increased oxidative capacity (CPT), buffering capacity (IHT), and compensatory fiber-selective vasodilation (RST), respectively [12]. Overall, LLTH methods have been shown to stimulate non-haematological peripheral adaptations, such as muscular adaptations which promote energy metabolism, alongside increased perfusion, improving O 2 utilisation and delivery, favouring sporting performance [11].
Despite previous reports suggesting training in hypoxia can augment HIIT-induced adaptations in VO 2max [37], a systematic review by Hopeller et al. [38] reported hypoxia supplementary to exercise training was not consistently advantageous for performance at sea level. Hamlin et al.'s [39] meta-analysis of HIIT-hypoxia training focusses specifically on populations participating in team sports, and on high intensity running (Yo-Yo IRT) after a hypoxic interventions. Therefore, there is a need for a meta-analysis with wider inclusion criteria (i.e., not just team sports players) with VO 2max as the primary outcome variable, as VO 2max is the gold standard of cardiorespiratory fitness measurement. Therefore, for a more coherent interpretation of the effects of HIIT in hypoxia vs. HIIT in normoxia, a quantitative pooled analysis of previous studies was necessary.

Objectives
Despite the abundance of studies investigating and reviewing hypoxia and HIIT separately, there was a lack of literature focusing on the effects of HIIT in hypoxia on VO 2max . Therefore, the aim of this investigation was to conduct a meta-analysis on the effect of HIIT in hypoxia compared to HIIT in normoxia on VO 2max . A secondary aim was to investigate study characteristics (i.e., degree of hypoxia, study duration) on magnitude of effect through meta-regressions.

Eligibility Criteria
This meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. Studies which met the following criteria were included: (1) full text manuscript; (2) not a review; (3) studies were required to have a control group within normoxic/sea-level environment or include pre-exercise intervention measures; (4) healthy participants of any sex aged 16-65 years; (5) studies were required to employ a HIIT intervention/programme for a minimum of 14 days. Furthermore, studies were required to have reported descriptive data, such as mean, standard deviation (SD), and sample size (n). If required, requests for details and full papers were submitted to the author(s). The primary aim was to investigate whether VO 2max was affected by HIIT in hypoxia (environmental or simulated). Therefore, only studies which directly measured (i.e., not estimated) VO 2max (ml·kg·min 1 or l·min −1 preand post-intervention) were included. Within this review both randomised control trials (RCTs) and non-randomised control trials (CTs) were considered. Thus, studies without a control group (i.e., uncontrolled trials with a pre-to post-exposure design) were excluded from analysis.

Study Selection
Following searches, obtained manuscripts were downloaded into a single reference manager (Zotero, 2016, Zotero version 5.0.96.1). Prior to eligibility screening the papers were sorted into a single reference list, with duplicates removed. Title and abstracts for all papers were screened for eligibility by two authors (A.W. and L.D.H.) with those that did not meet inclusion criteria excluded. Any disagreement between both reviewers was discussed in a consensus meeting. Out of the remaining manuscripts, those which examined HIIT in hypoxia were collated. Full text manuscripts were screened in depth and compared against inclusion and exclusion criteria. Following full text eligibility screening authors extracted participant data sets (sample size, n; age, mean ± SD), exercise modality (cycling, running, swimming, etc.), intervention method (HIIT, SIT, multi-component training, and interval training), intervention duration, altitude conditions (FiO 2 , or height above sea level, e.g., 3000 m) and VO 2max analysis method (Douglas bag, breath by breath gas analysis). Furthermore, manuscripts were coded as RCTs or CTs ( Figure 1). Subsequently, all remaining papers were assessed against the Physiotherapy Evidence Database (PEDro) scale. The PEDro scale objectively assesses methodological quality of each study [40].
Subsequently, all remaining papers were assessed against the Physiotherapy Evidence Database (PEDro) scale. The PEDro scale objectively assesses methodological quality of each study [40].

Data Collection Process
Information was imported into a spreadsheet, which was specifically designed for meta-analyses (Jamovi version 2.3.0.0, MAJOR package, https://www.jamovi.org, 2 October 2022). Data from both hypoxic and control groups were extracted from manuscripts: Change in (Δ)VO2max (ml·kg·min −1 or l·min −1 ), effective FiO2, intervention duration, and sample size (n). For clarity, in an RCT or CT, the mean and SD ΔVO2max from pre-to posttraining in the experimental group and control group, plus the n of each group is entered into the spreadsheet (six data items). Where the mean ΔVO2max was not reported, we subtracted pre-training VO2max from post-training VO2max. Where the SD ΔVO2max was not reported, it was calculated thusly:

Data Collection Process
Information was imported into a spreadsheet, which was specifically designed for meta-analyses (Jamovi version 2.3.0.0, MAJOR package, https://www.jamovi.org, 2 October 2022). Data from both hypoxic and control groups were extracted from manuscripts: Change in (∆)VO 2max (ml·kg·min −1 or l·min −1 ), effective FiO 2 , intervention duration, and sample size (n). For clarity, in an RCT or CT, the mean and SD ∆VO 2max from pre-to post-training in the experimental group and control group, plus the n of each group is entered into the spreadsheet (six data items). Where the mean ∆VO 2max was not reported, we subtracted pre-training VO 2max from post-training VO 2max . Where the SD ∆VO 2max was not reported, it was calculated thusly: whereby: corr = correlation coefficient, a value of which describes the relationship between baseline and final VO 2max measurements over time. We used the correlation coefficient from Lawler et al. [41] (0.94) which was the correlation coefficient of pre-and post-intervention VO 2max in a group of trained and untrained male athletes (n = 13). In cases of missing data, authors were contacted via email and asked to provide necessary information. If no response was received, means and SDs were estimated from figures using computer software (Image J, Towson, MD, USA, Imagej.net (accessed on 10 April 2021).

Data Items
Standardised mean differences (SMD) expressed the intervention effect within each study [42] using a restricted maximum-likelihood model estimate. For clarity, the mean change in (∆) VO 2max in the hypoxic group, the SD of ∆VO 2max in the hypoxic group, the n of the hypoxic group, the mean ∆VO 2max in the normoxic (i.e., control) group, the SD of ∆VO 2max in the normoxic group, and the n of the normoxic group were used to calculate SMD. All studies had a control group so no uncontrolled trials were analysed. The alpha level (p) describes the probability of a type I error, and 95% was used as the confidence interval (CI) level. Statistical heterogeneity was quantified using the I 2 statistics. An I 2 value greater than 50% is classified as moderate to high between study heterogeneity. Due to the included studies being considered heterogeneous (I 2 = 63%) a random effects meta-analysis was conducted. Funnel plots and the trim and fill method [43] assessed publication bias. The trim and fill method determines the number of studies necessary to eradicate publication bias from the funnel plot.

Study Selection
Combined results from the three database searches identified 441 articles ( Figure 1). After duplicates were removed a total of 119 titles and abstracts were screened for eligibility using the inclusion and exclusion criteria. We attempted to retrieve 37 records, and 33 reports were successfully retrieved and assessed for eligibility. Of the 33 screened, 24 papers were excluded, leaving nine full text manuscripts included within the final quantitative synthesis.

Study Characteristics
On completion of data pooling, nine studies were included in the analysis: seven were RCTs and two were control trials (Table 1). Within the nine studies, a total of 194 participants (men = 139, women = 55) were included. Studies were 2-13 weeks in duration (Table 2), and included running, cycling, swimming, or multi-component training. The PEDro scale determined quality of studies, and results indicated a mean score of 4 ± 1.

Meta-Analysis
The overall SMD for HIIT in hypoxia was 1.14 (95% CI = 0.56-1.72; p < 0.001; Figure 2). Heterogeneity (I 2 = 67%) justified the use of a random effects model. The Richardson et al. [50] study was weighted the most within the meta-analysis (13%), whereas Zebrowska et al. [52] carried the least weight (9%). Visual inspection of the funnel plot (Figure 3) suggest studies were spread across both sides of the pooled SMD (i.e., without asymmetry) indicating low publication bias. The Trim and Fill method confirmed the number of inputted studies to eliminate publication bias was one, although the overall number of studies was an uneven number. In the next paragraph we describe our sensitivity analysis where we removed one studies to achieve a value of zero (i.e., plot symmetry).
We subsequently performed sensitivity analysis by removing the two studies which fell outside of the funnel [50,51]. This procedure did not cause a qualitative effect (i.e., the direction of overall effect). This resulted in a SMD of 1.51 (95% CI = 1.06-1.96; p < 0.001). We subsequently performed sensitivity analysis by removing the two studies which were not RCTs [44,49]. This procedure did not cause a qualitative effect and the SMD was 1.11 (95% CI = 0.35-1.86; p = 0.004). Finally, we removed the study by Jung et al. [47] to achieve a Trim and Fill value of 0 and the SMD was 1.01 (95% CI = 0.42-1.60; p < 0.01). Therefore, we believe results from the initial meta-analysis are robust against the analysis decisions.

Meta-Analysis
The overall SMD for HIIT in hypoxia was 1.14 (95% CI = 0.56-1.72; p < 0.001; Figure  2). Heterogeneity (I 2 = 67%) justified the use of a random effects model. The Richardson et al. [50] study was weighted the most within the meta-analysis (13%), whereas Zebrowska et al. [52] carried the least weight (9%). Visual inspection of the funnel plot (Figure 3) suggest studies were spread across both sides of the pooled SMD (i.e., without asymmetry) indicating low publication bias. The Trim and Fill method confirmed the number of inputted studies to eliminate publication bias was one, although the overall number of studies was an uneven number. In the next paragraph we describe our sensitivity analysis where we removed one studies to achieve a value of zero (i.e., plot symmetry).   We subsequently performed sensitivity analysis by removing the two studies wh fell outside of the funnel [50,51]. This procedure did not cause a qualitative effect (i.e., direction of overall effect). This resulted in a SMD of 1.51 (95% CI = 1.06-1.96; p < 0.00 We subsequently performed sensitivity analysis by removing the two studies which w not RCTs [44,49]. This procedure did not cause a qualitative effect and the SMD was 1 (95% CI = 0.35-1.86; p = 0.004). Finally, we removed the study by Jung et al. [47] to achi a Trim and Fill value of 0 and the SMD was 1.01 (95% CI = 0.42-1.60; p < 0.01). Therefo we believe results from the initial meta-analysis are robust against the analysis decisio

Overview
The primary aim of this meta-analysis was to test whether VO2max was affected HIIT in hypoxia (environmental or simulated) more than HIIT in normoxia. The m findings were threefold. Firstly, HIIT in hypoxia increased VO2max more than HIIT normoxia. Secondly, meta-regression analysis suggested no relationship between int vention duration in weeks and SMD. Lastly, meta-regression analysis similarly sugges there was no relationship between effective FiO2 and SMD. Given that HIIT has und gone a recent surge in the literature, this meta-analysis provides pooled evidence t HIIT in hypoxia may be more effective in improving VO2max than HIIT in normoxia.

Overview
The primary aim of this meta-analysis was to test whether VO 2max was affected by HIIT in hypoxia (environmental or simulated) more than HIIT in normoxia. The main findings were threefold. Firstly, HIIT in hypoxia increased VO 2max more than HIIT in normoxia. Secondly, meta-regression analysis suggested no relationship between intervention duration in weeks and SMD. Lastly, meta-regression analysis similarly suggested there was no relationship between effective FiO 2 and SMD. Given that HIIT has undergone a recent surge in the literature, this meta-analysis provides pooled evidence that HIIT in hypoxia may be more effective in improving VO 2max than HIIT in normoxia.

The Effect of HIIT in Hypoxia on VO 2max
When the studies were pooled, HIIT in hypoxia displayed a positive effect on VO 2max , compared to normoxia, in which eight of nine studies demonstrated a positive SMD. However, the observed negative SMD in one study [51], was resultant of the normoxic group improving VO 2max (+5.6%), to a greater extent than the hypoxic group (+3.8%). Jung et al. [47] demonstrated the largest effect size of all included studies (SMD = 2.20). The result may not initially be surprising to the reader as participants in the Jung et al. [47] study exercising at 90-95% HR max , with a high training volume which included ten 5 min intervals per session (90 min, 3 d·wk −1 for 6 weeks). Therefore, this large effect may be a repercussion of time spent ≥90% HR max or VO 2max (they are highly related), as time spent at this intensity is known to determine training adaptations [53][54][55]. However, in a meta-analytical approach, the ∆VO 2max from hypoxic group would be compared to the ∆VO 2max in the normoxic group. As both groups underwent the same training regime, it is difficult to associate large SMDs to the training alone, as the 'control' normoxic group underwent analogous training to the 'treatment' hypoxic group.
Upon initial examination of the HIIT interventions, it was surprising that the study of Czuba et al. [45] exhibited a similar effect size (SMD = 2.13) to Jung and colleagues [47], as intervention duration was half that (60-75 min, 3 d·wk −1 for 3 weeks) of Jung et al. [47]. Upon further inspection, although the main aspect of the interval training was 60-75 min, athletes actually spent 2 h per session (6 h wk −1 ) in hypoxia due to additional warm-ups, cool-downs, and general endurance riding. Therefore, this extra time in hypoxia at lower intensities may have provided an additional stimulus to hypoxic HIIT. Mechanisms behind hypoxia-induced improvements in VO 2max compared to volume-and intensity-matched normoxic training is not fully understood with regard to HIIT. However, exposure to acute hypoxia significantly reduces VO 2max [56]. It may be that as VO 2max acutely decreases with increasing altitude [56], the additional challenge of hypoxia increases the relative intensity of exercise, providing an added stimulus for adaptation. VO 2max declines to a larger extent in acute normobaric hypoxia compared to hypobaric hypoxia, occurring alongside a higher VEmax in hypobaric hypoxia [57]. Therefore, this may explain the differences in training adaptations between simulated [45,46,48,[50][51][52] and natural altitude [44,49] studies. As time spent ≥90% HR max or VO 2max determines training adaptations to HIIT [53][54][55], the addition of hypoxia to training may have increased time over this threshold intensity and thus aerobic adaptations. This potential intensity issue has been most recently addressed by Li et al. [58], whereby, it was suggested that the intensity of HIIT in hypoxia can be matched in normoxia by adjusting the relative peak power output and lactate threshold based on graded exercise testing [58]. Unfortunately, given the diverse methods researchers employed to report exercise intensity, we were unable to conduct a meta-regression on the effect of exercise intensity (and thus training load) on ∆VO 2max , or a meta-regression on the effect of time in particular training zones on ∆VO 2max . The exact mechanisms by which HIIT in hypoxia impacts VO 2max compared to normoxia remains controversial [59][60][61]. Increased transcription of skeletal muscle proteins involving redox regulation and glucose uptake has been reported [62], whilst training in hypoxia also improves mitochondrial function and subsequent ATP production [63]. All of these adaptations lead to the improved aerobic capacity shown with acute hypoxia training vs. normoxia training [59,63].

Impact of Altitude Extent and Intervention Duration on VO 2max
While meta-regression analysis did not identify a relationship between altitude extent and SMD, an increase in VO 2max was observed following hypoxic HIIT vs. normoxic HIIT based on the SMD. A statistical explanation of the lack of dose (effective FiO 2 )-response (SMD) may be that variance was small between studies, with six of the nine studies utilising an effective FiO 2 of 15.0-15.3%. Therefore, this limited range would unlikely explain the variance in SMD magnitudes between studies.

Limitations
A key limitation of this systematic review and meta-analysis was the lack of studies concerning HIIT in hypoxia. A greater number of studies would increase robustness of results and add weight to conclusions [64]. Thus, conclusions herein are preliminary until a greater body of literature surrounding HIIT in hypoxia and effect upon VO 2max is available. Moreover, authors sought to examine moderating effect of exercise intensity on the VO 2max , however included studies displayed large variability in exercise prescriptions. Therefore, it remains unknown if different training loads modify the effect of hypoxia on VO 2max . Although a limitation to this review, it is first and foremost a limitation within the field of study, as there are countless methods of measuring exercise intensity. Across studies, there were five different descriptions of intensity; 40% of the studies described intensity as % of maximal heart rate (HR max ), 20% as rating of perceived exertion (RPE), 20% as % of lactate threshold (LAT), 10% as a % of maximum repetition (RM) and 10% as a % of velocity VO 2max (vVO 2max ). Similarly, repetitions, sets, and rest periods were seldom reported. It would improve future research if studies utilised the consensus on exercise reporting template (CERT [65]). Penultimately, as mentioned previously, six of the nine studies utilised an effective FiO 2 of 15.0-15.3%. Moreover, only one of the ten studies examined an effective F i O 2 of <14% (Ref. [48] F i O 2 of 12.6%). Therefore, most studies considered moderate hypoxia, and the lack of variety in effective F i O 2 in the included studies may be considered a limitation and an area for further research. Finally, study quality was low (PEDro scale mean~4), and therefore this must be considered a limitation to the literature base. However, considering three points are awarded for blinding, it is difficult to blind participants to environmental altitude as conscious travel is required.

Conclusions
Findings from the present systematic review, meta-analysis, and meta-regression suggest that participating in HIIT in hypoxia improves VO 2max more than HIIT in normoxia. While there is a lack of association between effective FiO 2 utilised and VO 2max improvement in the hypoxic groups, we believe this was a result of the limit range of FiO 2 examined and therefore a statistical artefact. Thus, we believe it is pertinent to note that results from all but one study demonstrated a positive SMD (i.e., favouring HIIT in hypoxia vs. HIIT in normoxia).