From previous studies evaluating changes across different SmO₂ signals following endurance training interventions, we hypothesized that endurance training would induce greater desaturation to a lower SmO₂min at maximal task tolerance during an IET. As for the effects of training on the change in SmO₂min relative to Wpeak and V˙O₂peak, we hypothesized that both Wpeak and V˙O₂peak would increase with training and that a lower SmO₂min would be correlated with higher post-training Wpeak and V˙O₂peak. The aim of this meta-analysis was to provide an estimate of the effects of endurance training on SmO₂min during an IET in healthy participants.

The online databases MEDLINE (OVID), EMBASE (OVID), and SPORTDiscus (EBSCOHost) were searched for published, full-text articles in English up to and including July 2024. The keywords used to identify relevant studies can be found in Supplementary 1. Further records were identified by manual citation searching through additional online searches. The inclusion criteria were as follows: (1) were in English; (2) endurance training intervention (31); peripheral muscle NIRS (32); IET performed pre/post training (33); SmO₂ or analogous saturation parameter measured. Exclusion criteria were as follows: (1) No clinical populations were included in the analysis, but clinical studies that had healthy controls with an endurance training intervention were included, with only the control data included in the extraction; (2). Studies were excluded if they had with no measure of muscle oxygen saturation or provided no data that allowed for calculation of saturation.

If reports included partial data, respective authors were contacted to provide additional data by the lead author (AY). If no response was received, a follow-up email was sent, and studies were excluded if no data were provided after 30 days. In the case where necessary data were only presented in a graphical format, PlotDigitizer (https://plotdigitizer.com, 3rd version, August 2022) was used to extract relevant values.

All records screened independently using Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia, for eligibility by two authors (AY, JA, or HN), any disagreements were discussed and evaluated by the group (AY, JA, HN, and MK). Similarly, reports assessed for eligibility were independently reviewed by two authors (AY, and HN) and reason for exclusion was decided collaboratively (AY, and HN).

Data were initially extracted for a systematic review from reports assessed for eligibility by three authors (AY, HN, and JA) (Figure 1). However, for the five papers included in the quantitative synthesis, further data extraction and analysis was conducted by AY, HN, and MR.

The extracted data from each included article were: participant characteristics (sample size, sex, age, body mass, and BMI), intervention (exercise type, exercise mode, duration of training, frequency of training, and number of sessions), NIRS (device, baseline IET SmO₂ pre-training, IET maximal task tolerance SmO₂ pre-training, baseline IET SmO₂ post-training, IET maximal task tolerance SmO₂ post-training), Wpeak (IET Wpeak pre-training, IET Wpeak post-training), V˙O₂peak (IET V˙O₂peak pre-training, IET V˙O₂peak post-training). Baseline IET SmO₂ (%) showed no significant differences pre- to post-training in any of the included studies. Since there were no baseline differences, we normalized baseline SmO₂ to 100% and differences from baseline during exercise were reflected as a decrease from 100% to the SmO₂min (i.e., the lower the SmO₂min value the more oxygen desaturated). To estimate the magnitude and direction of change in the standardized SmO₂min between pre-training and post-training, the standardized SmO₂min values were subtracted from each other to yield the change (ΔSmO₂min). For Wpeak and V˙O₂peak, if a p-value was not provided for change following training, it was estimated using the means and standard deviation (SD) from pre-training to post-training. When IET was longer than 12 min, a correction factor was used to standardize Wpeak (34).

Risk of bias for individual studies was evaluated using the PEDro scale and the completed scale can be found in Supplementary 1. Two reviewers independently assessed studies for bias (AY and HN), and any disagreements were resolved by the reviewers and a third author (MK).

Data analysis and statistical analysis were carried out in R (v4.1.2, R Foundation for Statistical Computing, Vienna, Austria). Group data were reported as means and SDs. The effect measure of the continuous primary outcome variable (ΔSmO₂min) was presented as the standardized mean difference (SMD) and its 95% confidence intervals for pooled data, following standardization of the data as described above. When SD was not provided, standard error of the mean (SE) was converted using the following formula: SD=SE⋅n. When SE was not reported, the p-value was used to estimate the SD using the following formula: SD=n⋅(x_1−x_2t). To provide a conservative estimate of the SD, a p-value expressed as an inequality (i.e., “<”) was changed to an equality (i.e., “=”) (27). A forest plot was produced to visually display the quantitative results; a random effect meta-analysis model (DerSimonian-Laird random effects method) was used to synthesize the individual studies (31). Statistical heterogeneity was quantified using the I² statistic. I² values of 25%, 50%, and 75% were considered as low, moderate, and high degrees of statistical heterogeneity (35). To determine the relationship between the effect size and the sample size, a funnel plot was used. An Egger's Test was used to account for any small sample size bias (36).

We conducted 5 subgroup analyses to estimate the effect of endurance training on ΔSmO₂min in relation to the following modifiers: continuous variables (Wpeak and V˙O₂peak) and categorical variables (sex, training type, and exercise mode). The expected effect of endurance training on the continuous variables is an increase, which was explored using meta-regressions and visually displayed as bubble plots. For these variables, the data were converted to a percentage change from pre-training to post-training. Categorical variables of the retrieved studies were evaluated based on sex (female or male), training type (continuous and interval exercise), and mode (leg cycling, combined arm and leg cycling, and unsupervised training). They were analyzed using a mixed-effects model to estimate an effect of endurance training on the relationship between ΔSmO₂min and each of the categorial variables separately.

A combined total of 869 records were identified in the search prior to the removal of 366 duplicates by an automatic review application (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia) (Figure 1). After title and abstract screening, 25 reports were retrieved for full-text text review. One could not be retrieved for full text (37). Nineteen papers were excluded at the full text review stage. Eleven were excluded due to exercise protocol used (32, 38–47). Eight were excluded due to ineligible reported outcome measures (33, 48–54). Identification of studies via other methods yielded six studies. Five were ineligible due to exercise protocol used (55–59). Five studies were included in the final quantitative analysis (19–23). Zinner et al. had two eligible groups from the same study, meaning that 6 groups were present in the analysis (23).

A total of 58 participants were included (22 females, age = 26.7 ± 4.0), all were either sedentary or recreationally active, healthy individuals (Table 1) (60). The participants from Rissanen, et al. (21), were a healthy group used as control for a larger study on diabetic patients. Training type included continuous training (n = 17), and interval training (n = 41). Exercise mode included cycling (n = 40), arm and leg cycling exercise (n = 10), and unsupervised training (i.e., no reporting of exercise mode) (n = 8).

The completed PEDro scores for the included studies can be found in Supplementary 1. No form of blinding was used in any of the included studies. All studies included were sufficiently similar at baseline and used measures of point estimates and variability for the primary outcome measure. The included participant data from Rissanen et al., was a control healthy group from a larger clinical study, and five participants were excluded post hoc from the control to match the clinical group in anthropometric and V˙O₂peak data (21). The data were excluded both at baseline and post-training and did not affect the comparison. Therefore, there was not sufficient follow-up for those initially allocated to the training intervention.

For individual studies, the effect of endurance training on SmO₂min was presented as means and SDs in Table 2, and as effect measures and confidence intervals in Figure 2.

No significant effect of endurance training on ΔSmO₂min was observed. The pooled effect size was −1.92 (95% CI = −5.56 to 1.72, p < 0.3), indicating no effect of training on ΔSmO₂min. There was a high degree of heterogeneity in the pooled results (I² = 70%) for changes in ΔSmO₂min. To investigate the source of heterogeneity in ΔSmO₂min, we considered all the variables that were extracted. However, a meta-regression was done on Wpeak and V˙O₂peak only due to their relevance to our question.

For analyses relating to Wpeak, only 4 studies and 5 groups were included; Kime et al. 68) did not report any measures of power. The effect of endurance training on the relationship between Wpeak and ΔSmO₂min resulted in a tendency for SmO₂ to be lower with increased Wpeak (Figure 3). The effect of training on Wpeak showed a significant improvement from 286 ± 45 watts (W) to 308 ± 53 W (p < 0.01). The heterogeneity for ΔSmO₂min increased (I² = 73%) when grouped with Wpeak. Peak power output data can be found in Supplementary 1.

The relationship between V˙O₂peak and ΔSmO₂min showed no effect of training. Keramidas et al. (20) did not state a p-value for V˙O₂peak for the difference between pre-training to post-training. As such, we were unable to calculate the SD of the difference in V˙O₂peak between the two time points, and it was excluded from the meta-regression. The heterogeneity for ΔSmO₂min slightly decreased but remained moderate (I² = 68%) when grouped with V˙O₂peak. V˙O₂peak (ml · kg⁻¹ · min⁻¹) values from individual studies can be found in the Supplementary 1.

Our analysis detected no effect of training on ΔSmO₂min with regards to sex reported for participants (female or male), training type (continuous or interval), and exercise mode (cycling, arm and leg cycling, or unsupervised training). The heterogeneity for all stratified analyses increased from moderate to high, suggesting that these characteristics would likely not influence the change in ΔSmO₂min following a training program. For sex, training type, and exercise mode, the I² increased to 93%, 91%, and 90%, respectively. A summary table of the results, including the p-value for interactions for each of the modifiers, can be found in Supplementary 1.

Visual inspection of a funnel plot (Figure 4) observing changes in ΔSmO₂min yielded no signs of asymmetry. Additionally, an Egger's test was used to confirm the lack of small sample size bias and test for funnel plot asymmetry (z = −1.55, CI = −5.07 to 9.74, and p = 0.12) (36).

Heterogeneity increased across all stratified subgroup analyses, with no effect of training. With regards to the effect of sex, our data included 22 female participants out of 58, in four of the six included groups (20, 23, 61). Previous studies looking at sex-based differences indicated that higher adiposity at the quadricep muscles seen more commonly in female participants, was an important factor affecting muscle oxygenation outcomes (15, 62, 63). None of the included studies reported adiposity for males and females independently, which may explain the increased heterogeneity when stratifying the studies by sex.

The exercise type was divided into continuous or interval training. This analysis showed no effect of training on ΔSmO₂min, with a greater heterogeneity relative to the primary analysis. The continuous training intervention was used in two studies (21, 61). Rissanen et al., performed a year-long training intervention that included both endurance and resistance training. The exercise modalities used in the study were not specified. Therefore, we cannot conclude whether interval training was used, or to what degree the intervention differed from other studies that used interval training as their primary mode of training.

For exercise mode, no effect of training was found. The dominant mode of exercise in the included studies was cycling, both in training and during the IET. Zinner, et al., was the only study that included both leg and arm cycling (23). They included two groups that were exposed to a similar training load, one with leg and arm cycling, and the other with leg only. Interestingly, despite a lack of significant change in their SmO₂min response after training, their results for the leg and arm group were the only ones showing directionally different findings (increase in SmO₂min post-training) in the vastus lateralis muscle oxygenation response during an IET. Their report raises an interesting consideration in light of the findings presented in this study, concerning what NIRS can reveal about local skeletal muscle adaptations across muscle groups, or between upper and lower body muscles, during whole-body exercise. The ability of NIRS to quantify SmO₂ across multiple muscle sites is currently underexplored and may shed light on the integrated systemic and local responses to training interventions. The ability to critically evaluate the effect of exercise mode on SmO₂ is currently limited due to most studies using cycling as their primary mode of IET, even if the training intervention included a different mode of exercise.

Despite not finding a significant effect of training on ΔSmO₂min associated with an increase in Wpeak in an IET after training, a trend was detected showing that with improvements in Wpeak, ΔSmO₂min tended to be greater (i.e., greater desaturation to a lower SmO₂min post-training). This finding corresponds with previous studies that investigated training interventions and their effects on muscle oxygenation measured by non-IET protocols (43, 56, 64). The most recent study by Paquette et al., investigated the effect of a 3-week training camp on national-level kayakers using muscle oxygen saturation as a primary outcome. Their results showed a greater ability to desaturate to a lower SmO₂min after training, implying enhanced muscle oxygen extraction in conjunction with an increase in performance (14). The increase observed in Wpeak in the current meta-analysis shows that IET performance was improved with training, with longer duration to maximum exercise tolerance. Thus, despite only observing a tendency for decreased SmO₂min at the end of the IET (p = 0.06), the rate of SmO₂ desaturation during the IET was slowed after endurance training, producing the same or lower minimum at a higher Wpeak.

It is important to note that despite showing a trend for a greater SmO₂ desaturation at a higher Wpeak, V˙O₂peak did not present the same trend. This meta-analysis did not attempt to evaluate how SmO₂ changes might be related to either peripheral skeletal muscle or systemic cardiopulmonary adaptations. Our results do raise an important question about implications concerning SmO₂ changes in relation to improvements in performance without evidence of an effect of endurance training on the relationship between peripheral muscle measurements and systemic measures of maximal aerobic capacity.

From the meta-regression of ΔSmO₂min and V˙O₂peak, our results show no relationship between the change in ΔSmO₂min to the change in V˙O₂peak from pre- to post-training. As mentioned previously, SmO₂min is associated with maximal task tolerance during an IET (11, 12). This is due to the progressive disruption of local metabolic milieu with increasing workload (2). Our results suggest that a similar SmO₂min value will be achieved at maximal exercise tolerance after a training intervention, regardless of improvements in fitness. This suggests that a higher SmO₂ value observed at a similar absolute submaximal workload after a training intervention may be associated with enhanced fitness and predictive of increased performance. For athletes and performance specialists, instead of requiring maximal testing to confirm fitness changes, SmO₂ could be evaluated more frequently at submaximal workloads to suggest directional performance changes. This is like the current use of heart rate monitors in-field training relative to objective, external load measures such as running pace or cycling power output. Ideally, NIRS could be used in conjunction with heart rate monitors during regular training bouts instead of dedicated testing sessions, to improve tactical decision-making by practitioners and athletes. In this context, by observing the relationship of SmO₂ to cycling power output, performance capacity could be estimated on a session-to-session basis. Before doing so, future studies should investigate the test-retest reliability of NIRS signals, especially when choosing wearable NIRS sensors during training sessions, to determine the minimal worthwhile change that could indicate fitness improvements.

Our meta-analysis includes a small number of studies, and limited number of participants. Despite not presenting a risk of bias between studies, a lack of standardization of training intervention may have influenced our results and contributed to the heterogeneity of the analysis (65). In line with this limitation, inherent differences in exercise intensity during the training period may have effected training adaptations. One of the studies defined their intervention as endurance training but did not standardize their training intervention (21). For exercise mode, as indicated previously, cycling is a common exercise mode for IET; however, SmO₂ responses may be different for running, rowing, and other exercise modalities related to different muscle recruitment demands. Future studies should use the same mode for IET as for training. Additionally, NIRS signals are sensitive to the absorbance characteristics of biological tissues (15, 62, 66, 67). Therefore, for better quantification of training intervention effects on SmO₂, reporting adiposity is critical. Lastly, the focus of the analysis was to quantify the effects of training on SmO₂min during an IET; however, future studies should consider investigating the ability of other testing protocols to quantify muscle oxygenation, such as constant workload protocols across intensity domains and intermittent IETs, among other protocols.