Comparison of different interval training methods on athletes’ oxygen uptake: a systematic review with pairwise and network meta-analysis

Abstract

Objective

To (i) examine the effect size and the ranking of the training effects for the high-intensity interval training (HIIT), sprint interval training (SIT), and repeated sprint training (RST) on athletes’ VO₂max through network meta-analysis. (ii) investigate the effects of the training program protocols of the three methods on the improvement of VO₂max through pairwise meta-analysis and meta-regression.

Methods

A systematic search of four databases (Web of Science, PubMed, Scopus, SPORTDiscus) were conducted on April 1, 2025. 51 eligible studies (1,261 athletes), evaluating the direct/indirect effects of HIIT, SIT, RST versus continuous training (CT) and conventional training (CON). Frequentist network meta-analysis quantified effect sizes (NMAs’ g) and probabilistic rankings (P-scores), while three-level meta-regression modeled dose-response relationships for training parameters.

Results

All three interval training methods significantly increased VO₂max compared to CON, with NMA rankings: RST (NMA’s g = 1.04) > HIIT (NMA’s g = 1.01) > SIT (NMA’s g = 0.69) > CT (NMA’s g = 0.29), no significant differences existed among RST, HIIT, and SIT (p > 0.05). Subgroup analyses showed HIIT efficacy was moderated by athlete level, whereas SIT improvements were influenced by intervention duration, training frequency, and training mode. Meta-regression identified inverted U-shaped relationships for HIIT, with peak benefits at 140 s work duration and a work-to-recovery ratio (WRR) of 0.85. For SIT, improvements are not significant when recovery durations exceeding 97 s.

Conclusions

RST, HIIT, and SIT all enhance VO₂max in athletes, with RST demonstrating the strongest probabilistic efficacy. Optimal protocols include 3–6 weeks of running-based HIIT (140 s work, 165 s recovery) or SIT (≤ 30 s sprints, < 97 s recovery) at 3 sessions per week. Conducting RST 3 sessions a week for two weeks is sufficient to achieve improvements.

Systematic review registration

PROSPERO CRD42023435021.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13102-025-01191-6.

Background

Aerobic capacity is crucial for athletes’ competitive performance, with common training methods including Interval Training (IT) and Continuous Training (CT) aimed at enhancing it. In recent years, the field of aerobic training and research has gradually shifted from CT to High-Intensity Interval Training (HIIT) [1]. CT has long been considered an effective strategy for improving aerobic endurance. Extensive research has shown that this approach can significantly increase maximal oxygen uptake, accelerate fat metabolism, and improve cardiovascular capacity when performed over extended periods [2–4]. In contrast, HIIT has gained popularity due to its superior time efficiency and potent training effects. It has been shown to significantly improve both anaerobic and aerobic capacities in a shorter timeframe, particularly when tailored to individual needs. However, the extent of these benefits often depends on factors such as individual fitness levels and the specific parameters of the training protocol [5, 6].

Interval training methods can be broadly classified based on three key characteristics: training intensity, the duration of work intervals, and recovery periods. The most commonly studied types of interval training include HIIT, Sprint Interval Training (SIT), and Repeated Sprint Training (RST) [7–9]. HIIT typically involves work at or near 100% of VO₂max intensity, with the goal of maximizing the number of high-intensity intervals completed [10]. Unlike CT at a steady intensity, HIIT emphasizes repeated bouts of intense effort interspersed with recovery periods [11]. SIT requires maximal or supramaximal efforts—often exceeding 100% of VO₂max—for short durations (typically 10–30 s), followed by longer recovery periods. A common SIT protocol involves 30 s sprints on a Wingate ergometer, followed by 4–4.5 min of passive or low-intensity active recovery [12]. RST also involves supramaximal efforts but is characterized by very short sprints (≤ 10 s) and brief recovery intervals (≤ 60 s). RST often includes repeated sprints over distances of 20 to 60 m, with shorter recovery times that may result in a more pronounced drop in sprint performance across repetitions [9].

Although numerous reviews have examined HIIT, SIT, and RST [13–15], most have focused on the general population and do not adequately address their comparative effects in athletic cohorts. Prior analyses have not found significant differences between HIIT and SIT in improving VO₂max among the general population [16]. However, there is a lack of meta-analyses that simultaneously compare HIIT, SIT, and RST specifically in athletes [14, 17]. This gap highlights the need for a network meta-analysis (NMA), which can integrate both direct and indirect evidence to provide a more comprehensive and robust comparison of multiple training methods [18]. NMAs also offer probabilistic treatment rankings, aiding in decision-making regarding optimal interventions [18]. This study primarily employs systematic review and meta-analysis methods to assess the benefits of various interval training techniques on improving oxygen uptake in athletes. Our specific aims included: (i) to explore the effect size (NMA Hedges’ g) and the ranking (P-score) of the training effects for the three methods through network meta-analysis, (ii) investigate the effects of the training program protocols of the three methods on the improvement of oxygen uptake through three-level paired meta-analysis and meta-regression.

Methods

This study was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [19], and it has been registered in the PROSPERO database (CRD42023435021).

Search strategy

A comprehensive literature search was performed using four electronic databases: Web of Science, PubMed, Scopus, and SPORTDiscus. The search terms included combinations of training interventions (e.g., HIIT, SIT, RST), athlete population terms, and VO₂max-related outcomes. The final search was conducted on April 1, 2025. Only peer-reviewed, English-language studies were included. The detailed strategy was presented in Table S1.

Study selection

The eligibility of the studies for inclusion was independently assessed by two authors (Q.Y. and D.G.), and the final decision regarding inclusion was made after a summary of the comments was prepared and discussed with another author (J.W.). All processes were conducted using Endnote 20. The eligibility criteria for the articles are detailed in Table 1.

Table 1.

Eligibility criteria for systematic review

Inclusion criteria
P	Healthy athletes, with no age restrictions.
I	• HIIT intervention, involving high intensity (< 100%VO2max, 80-95%HRmax), with a mean work duration of < 600 s [20]. • RST intervention, involving maximal intensity sprints (> 95% HRmax, > 100%VO2max), with a mean work duration of 3–7 s or equivalent distance, interspersed with brief (≤ 60 s) recovery periods [21]. • SIT intervention, involving maximal intensity sprints(> 95% HRmax, > 100%VO2max, involve ‘all out’ or ‘supramaximal’ efforts), usually with a mean work duration of ≤ 30s or equivalent distance and interspersed with 2-4 min passive recovery periods [20]. • Training group must have performed the intervention under normal conditions (e.g. usual nutritional intake)
C	• CT intervention, continuous training refers to a form of aerobic exercise that involves continuous, steady-state exercise performed at a moderate intensity. • CON intervention, conventional training, performed a regular training program (specialized technical practice or tactical practice) but without undergoing additional intermittent training or continuous training.
O	Measurements that showed athletic oxygen uptake level (e.g., VO2max, VO2peak)
S	Randomized controlled, parallel and crossover trials.
Exclude criteria
1.	Special populations (e.g. clinical, patients), people with a physical or mental disability or people considered to be injured or returning from injury. Non-athletic populations
2.	Studies incorporated combined experimental training interventions that were outside of their usual training practice (e.g., plus plyometric training)
3.	No relevant outcome measures were reported
4.	Training group was performed under altered or abnormal conditions (e.g., hypoxia, heat stress, and ergogenic aids, different diet)
5.	Grey literature and non-controlled studies.

Data extraction

The initial extraction of relevant data was conducted by two authors (Q. Y. and D. G.) and reviewed and validated by another author (J. W.). The primary data extracted included the following: (1) first author and year of publication, (2) study design, (3) participant characteristics (e.g., athlete level and age), and (4) training protocols for the experimental groups (e.g., intensity and measures of VO₂max). The athlete level was classified based on the study by McKay et al. We classify athletes into Trained/Developmental (Tier 2), Highly Trained/National Level (Tier 3), Elite/International Level (Tier 4), World Class (Tier 5) [22]. Additionally, we categorized the athlete according to their stage of education, such as middle school (< 19 years), college (19–23 years) or graduate (> 23 years). Means and standard deviations (SD) for baseline and endpoint VO₂max were extracted. For outcome measurements presented as images, data were collected using the WebPlotDigitizer online tool [23].

Methodological quality of the included studies

Two authors used the 11-item Physiotherapy Evidence Database (PEDro scale) for independent assessment. The quality of the literature was categorized as excellent (9–10 points), good (6–8 points), fair (4–5 points), or poor (≤ 3 points) based on the composite score of the scale [24].

Statistical analysis

Effect sizes were calculated as standardized mean differences and adjusting for small sample bias (Hedges’ g). A random-effects frequentist network meta-analysis was performed using the netmeta package in R to estimate both direct and indirect treatment effects, considering heterogeneity across studies [25]. The transitivity of the network was assessed by evaluating the consistency between direct and indirect comparisons, as well as through a visual inspection of study characteristics (e.g., age, intervention duration, athlete level). Pairwise heterogeneity was assessed using Cochran’s Q, τ², and I² statistics. Training method rankings were derived using P-scores based on point estimates and standard errors under the assumption of normality, which quantifies the average degree of certainty that a training method is superior to an alternative method [26].

To address dependence among multiple effect sizes from the same study (e.g., different training protocols within HIIT, RST, or SIT), Three-level meta-analyses were conducted using the metafor package in R. This model decomposes variance into within-study (level 2), between-study (level 3), improving accuracy and power by retaining all relevant comparisons [27]. Subgroup analyses explored moderators such as training mode, intervention period, age, and work-to-recovery ratio. Dose-response relationships were modeled using random-effects meta-regression with restricted maximum likelihood estimation. In all meta-regression models, baseline VO₂max was included as a covariate, based on the finding that baseline VO₂max often serves as a significant predictor of subsequent changes in VO₂max. Additionally, for the meta-regression of the work-to-recovery ratio, the duration of a single work bout was included as an additional covariate. Linear, quadratic, cubic, and restricted cubic spline models were compared using the corrected Akaike Information Criterion (AICc), selecting the model with the lowest bias [28].

Risk of publication bias was assessed using funnel plots and Egger’s test [29]. Forest plots illustrated the effect estimates, and the I² statistic was interpreted as follows: < 25%, 25–75%, and > 75% representing low, moderate, and considerable heterogeneity, respectively [30]. A significant threshold of p < 0.05 was applied. The network and pairwise meta-analyses were conducted using the netmeta and metafor packages, with results visualized through ggplot2 [25, 31, 32]. All analyses were performed in R (version 4.4.3; R Core Team, Vienna, Austria) [32]. We assessed the certainty of evidence in the NMA via the Confidence in Network Meta-Analysis (CINeMA) online application [33].

Results

Upon searching, 2971 potentially relevant studies were identified across four databases. Following the removal of duplicates, 1975 potentially eligible studies remained for which abstracts were screened. After excluding studies on the basis of the full text, a total of 51 studies were included in the primary analysis (Fig. 1).

Fig. 1.

PRISMA flow diagram for included and excluded studies

Study characteristics

A total of 1261 athletes were assigned to one of three training groups or two control groups across 62 pairwise comparisons. The pairwise comparisons comprised HIIT versus CON (n = 13 studies) [34–47], SIT versus CON (n = 8 studies) [48–55], RST versus CON (n = 14 studies) [47, 56–68], HIIT versus CT (n = 7 studies) [44, 69–74], SIT versus CT (n = 6 studies) [51, 54, 65, 75–77], HIIT versus SIT (n = 4 studies) [78–81], RST versus HIIT (n = 3 studies) [47, 82, 83], and CON versus CT (n = 1 studies) [54].

The methodological quality of the included studies was generally fair-to-good, as indicated by PEDro scale scores ranging from 4 to 7 (Fig. 2D). A summary of the methodological quality assessment for each included study and individual rating scales was provided in Table S4.

Fig. 2.

(A) Network geometry indicating number of participants in each arm (size of nodes) and number of comparisons between arms (thickness of lines). (B) Forest plot of NMA, effects size of RST, HIIT, SIT on VO₂max compared with CON. (C) Contour-enhanced funnel of NMA. (D) Risk of bias summary plot of included studies. (E) Summary table for credibility assessment using confidence in network meta-analysis (CINeMA). HIIT, high intensity interval training, RST, repeated sprint training, SIT, sprint interval training, CT, continues training, NMA, network meta-analysis

Network meta-analysis

The network meta-analysis included 10 study designs with 62 pairwise comparisons. The RST, HIIT, and SIT interventions included comparisons with at least two other nodes in the network geometry (Fig. 2A). In the 10 study designs, there were no significant differences between direct and indirect comparisons. The random effects model showed moderate heterogeneity (τ² = 0.13, I² = 36%), and the total heterogeneity test was significant (Q = 87.47, p < 0.01). Within-design heterogeneity (Q = 84.42, p < 0.01) suggested differences between studies comparing the same intervention, possibly due to differences in training protocols or population characteristics. The between-design inconsistency test had no statistical significance (Q = 3.05, p = 0.88), indicating that the results of indirect comparison were consistent with those of direct comparison, supporting the transitivity hypothesis of the network.

Compared with CON, RST significantly increased athletes’ VO₂max (NMA’s g = 1.04, 95% CI [0.76, 1.32), p < 0.01]. Similarly, HIIT (NMA’s g = 1.01, 95% CI [0.76, 1.25), p < 0.01] and SIT (NMA’s g = 0.69, 95% CI [0.39, 1.00], p < 0.01) significantly elevated oxygen uptake in athletes. In contrast, the effect size for CT was not significant (NMA’s g = 0.29, 95% CI [-0.07, 0.66], p = 0.11). Ranking on the basis of the P-score analysis: RST > HIIT > SIT > CT > CON (Fig. 2B). Visual inspection of the contour-enhanced funnel plot revealed that studies were concentrated in the center and evenly distributed on both sides (Fig. 2C). The Egger test indicating that the risk of publication bias was not significant (p = 0.42).

Credibility assessment

The credibility assessment revealed a range from high to very low confidence in the network (Fig. 2E). This was primarily attributed to the limited number of direct comparisons and high heterogeneity. Overall, the confidence in the improvements in VO₂max for HIIT, RST compared to CON was moderate and high respectively, and low for SIT verse CON. The evidence for the improvement in VO₂max brought by HIIT, and SIT compared to CT was low and moderate respectively, and very low for RST due to no direct comparisons.

Three-level pairwise meta-analysis

Compared with CON, the RST had a significant combined effect on VO₂max (g = 1.00, 95% CI [0.73, 1.26], p < 0.01, Fig. 3). Similarly, HIIT had a statistically significant combined effect (g = 0.89, 95% CI [0.56, 1.23], p < 0.01). SIT (g = 0.78, 95% CI [0.04, 1.52], p = 0.04), with moderate heterogeneity (I²-level 2 = 0%, I²-level 3 = 70.5%, PI [-1.07, 2.63]). Compared with CT, HIIT demonstrated a statistically significant combined effect on VO₂max (g = 0.69, 95% CI [0.09, 1.29], p < 0.05), with moderate heterogeneity (I²-level 2 = 0%, I²-level 3 = 49.9%, PI [-0.62, 2.00]). SIT was notable (g = 0.39, 95% CI [0.01, 0.77], p < 0.05). HIIT compared to RST (g = 0.16, 95% CI [-0.86, 1.17], p = 0.58), HIIT compared to SIT (g = 0.39, 95% CI [-1.33, 2.11], p = 0.52), with moderate heterogeneity (I²-level 2 = 0%, I²-level 3 = 39.7%, PI [-3.14, 3.91]).

Fig. 3.

Forest plot of the three level meta-analysis. HIIT, high intensity interval training; RST, repeated sprint training; SIT, sprint interval training; CON, convention training; CT, continues training

Risk of publication bias was only observed in the comparison between RST and CON, as indicated by Egger’s test (intercept = -1.09, p < 0.05). However, Duval and Tweedie’s trim and fill method, which added three studies to the left, did not affect the pooled effect. The fail-safe number of additional negative studies was 285. The visual inspection of the funnel plot did not reveal risk of bias (Fig. S1).

Subgroup analysis

In studies utilizing the SIT versus CON study design, subgroup analyses indicated that the running mode (g = 2.44, p < 0.01) resulted in a greater increase in oxygen uptake than the cycling mode (g = 0.69, p < 0.05) and the rowing mode (g = 0.49, p = 0.32), the training mode was identified as a significant modifying factor (p < 0.01, Fig. 4). The duration of the intervention was also a significant modifying factor (p < 0.05), with the 3 to 6 weeks duration (g = 2.44, p < 0.01) resulting in greater increases than durations longer than 6 weeks (g = 0.58, p = 0.08) and less than 3 weeks (g = 0.28, p = 0.56). Intervention frequency was identified as a modifying factor (p < 0.01), with an SIT program of 3 sessions per week exhibiting the significant effect size (g = 0.82, p < 0.05). However, no significant improvement was found in programs lasting more than 3 sessions per week (g = 0.03, p = 0.93), and only one study reported a SIT program with less than 3 sessions per week (g = 2.44, p < 0.01). In studies utilizing the RST versus CON, the training protocol of RST was not identified as a significant moderator (Fig. 5).

Fig. 4.

Subgroup analysis of HIIT and SIT compared to CON. Work-to-recovery ratio for HIIT, a: < 0.5, b: 0.5–1, c: > 1; Work-to-recovery ratio for SIT, a: < 0.125, b: 0.125–0.25, c: > 0.25; Tier 2, Trained/Developmental; Tier 3, Highly Trained/National Level; Tier 4, Elite/International Level

Fig. 5.

Subgroup analysis of RST compared to CON. a: < 0.25, b: 0.25–0.5, c: > 0.5; Tier 2, Trained/Developmental; Tier 3, Highly Trained/National Level; Tier 4, Elite/International Level

In studies utilizing HIIT versus CT group design, the level of the athlete was identified as a significant moderator (p < 0.01, Fig. 6). The Tier 2 class resulted in the largest effect size (g = 1.78) in one study, while the Tier 4 class has a large effect size (g = 0.75, p = 0.21) and the Tier 3 class has a moderate effect size (g = 0.32, p = 0.27). Intervention frequency was identified as a modifying factor (p < 0.01), with a HIIT program of 3 sessions per week exhibiting the largest effect size (g = 0.48, p = 0.10). However, no significant improvement was found in programs lasting less than 3 sessions per week (g = 0.40, p = 0.70), and only one study reported a HIIT program with more than 3 sessions per week (g = 1.77).

Fig. 6.

Subgroup analysis of HIIT and SIT compared to CT. Work-to-recovery ratio for HIIT, a: < 0.5, b: 0.5–1, c: > 1; Work-to-recovery ratio for SIT, a: < 0.125, b: 0.125–0.25, c: > 0.25; Tier 2, Trained/Developmental; Tier 3, Highly Trained/National Level; Tier 4, Elite/International Level

Effect of dose-response

In the meta-regression analysis, the AICc test indicated that the linear model provided an optimal balance between goodness of fit for warm-up duration in HIIT. The warm-up duration in HIIT was identified as the variable significantly associated with changes in athletes’ VO₂max (β = 0.002, 95% CI [0.001, 0.003], p = 0.02), indicating that longer warm-up durations are associated with greater improvements in athletes’ VO₂max (Fig. 7A). The nonlinear meta-regression results demonstrated an inverted U-shaped relationship between work duration per interval and the work-to-recovery ratio in HIIT (Fig. 7B). The highest increase in work duration per interval was observed at 140 s, achieving optimal benefits at a work-to-recovery ratio of 0.85 (Fig. 7C). Meta-regression analysis demonstrated a nonlinear relationship between recovery duration per interval and VO₂max improvement in SIT, when recovery time exceed than 97 s, the effect of SIT on VO₂max was no longer significant (Fig. 7D).

Fig. 7.

Dose-response plot for HIIT and SIT. (A) liner meta-regression results of warm-up time in HIIT. (B) Nonlinear meta-regression results of work duration per interval in HIIT. (C) Nonlinear meta-regression results of work-to-recovery ratio per interval in HIIT. (D) Nonlinear meta-regression results between recovery duration per interval in SIT

Discussion

The primary aim of this review was to quantify and rank the effectiveness of HIIT, RST, and SIT in improving athletes’ VO₂max compared to CT and CON. We also explored the impact of the training protocol on the three training methods. The current analyses revealed that (1) The network meta-analysis indicated that RST (NMA’s g = 1.04), HIIT (NMA’s g = 1.01), and SIT (NMA’s g = 0.69) each resulted in significant increases in VO₂max compared to CON, with prioritized RST > HIIT > SIT > CT > CON, no significant differences emerged among RST, HIIT, and SIT (p > 0.05). (2) Athlete level significantly modified the effectiveness of HIIT. For SIT, significant moderators included intervention duration, training frequency, training mode, athlete level, and work-to-recovery ratio. (3) Meta-regression analyses revealed inverted U-shaped relationships between VO₂max improvement and both work duration and work-to-recovery ratio in HIIT, with peak effects occurring at approximately 140 s of work bout and a work-to-recovery ratio of 0.85. A nonlinear relationship was identified between recovery duration and VO₂max improvement for SIT. When recovery duration exceeded 97 s, the effect of SIT was no longer significant.

Duration of the intervention

Compared to the CON group, we observed that 3 to 6 weeks of HIIT significantly improved the VO₂max of athletes (g = 0.89, p < 0.05), while no significant effects were observed for durations of less than 3 weeks (g = 0.71, p = 0.30) or more than 6 weeks (g = 1.01, p = 0.09). Compared to the CT group, 3 to 6 weeks (g = 0.39, p = 0.24) and more than 6 weeks of HIIT (g = 0.91, p = 0.10) had no significant impact on VO₂max. A recent meta-analysis revealed inconsistent results, showing that HIIT for 1–3 weeks and 4–9 weeks was more effective in improving athletes’ VO₂max, while durations of 10 weeks and above were insignificant. One possible explanation is that Wang et al. categorized SIT as HIIT, which may have somewhat obscured the certainty of the recommended HIIT training duration [84]. Conversely, our results support the findings of Wen et al., indicating that extending the HIIT intervention period does not yield further benefits in athletes [85].It is advisable to consider HIIT training for 3 weeks or more to improve athletes’ VO₂max [86].

In our analysis of SIT, we identified a significant interaction effect related to the intervention duration in SIT compared to CON study designs. Duration of 3 to 6 weeks yielded the largest effect size (g = 2.44, p < 0.01), significantly outperforming durations exceeding 6 weeks (g = 0.58, p = 0.08) and those lasting less than 3 weeks (g = 0.28, p = 0.56). However, when examining SIT versus CT study designs, durations exceeding 6 weeks (g = 0.56, p = 0.27) demonstrated greater effect sizes than those for 3 to 6 weeks (g = 0.39, p = 0.09). Only one study reported an 8-week SIT intervention that produced a moderate effect size (g = 0.56) [54]. In contrast, interventions lasting less than 2 weeks showed small effects (g = 0.30) [76]. This finding contradicts previous systematic review, which indicate that as little as 2-week of SIT is sufficient to achieve significant improvements in VO₂max [86]. One possible explanation for this discrepancy is that the study by Rosenblat et al. included both athletes and non-athletes, which may account for the observed significance of a 2-weeks SIT intervention. Previous research suggests that the optimal improvement in VO₂max from SIT in athletes occurs between 3 and 4 weeks [87]. These findings are consistent with our results, indicating that 3 to 6 weeks of SIT yield better outcomes.

Subgroup analysis revealed that two weeks RST intervention (g = 1.03, p < 0.01) yielded further improvements compared to extending the duration beyond 6 weeks (g = 0.92, p < 0.05). However, only one study reported durations beyond 6 weeks, which limits the generalizability of the results [61]. Our results suggest that two weeks of RST are sufficient to bring about a significant improvement in athletes’ VO₂max. This finding aligns with previous systematic review, which suggest that most adaptations to RST occur within the initial 6-week phase, after which improvements plateau [15, 88]. Additionally, systematic review comparing HIIT and SIT to CT have reported that both HIIT and SIT demonstrate a plateau in VO₂max improvement. HIIT and SIT typically lead to significant gains within a short period of less than 2 weeks, reaching a plateau after a relatively brief duration (6 weeks for SIT and 10 weeks for HIIT), whereas CT can gradually achieve comparable improvements through a longer training duration [89].

Frequency of intervention

An optimized training frequency will yield superior training results for coaches and athletes. The results demonstrated that scheduling the RST 3 sessions per week (g = 1.04, p < 0.01) yielded significant outcomes, whereas 2 sessions per week did not significantly improve VO₂max (g = 0.89, p = 0.22). However, this result is contrary to that of Thurlow et al. [15], Which resulted in an RST of 2 sessions per week had the most beneficial effect, whereas an additional RST sessions per week (i.e., three days) did not cause any conclusive benefits. Given that Thurlow et al.’s systematic review evaluated a combination of outcomes instead of only VO₂max, our study focused exclusively on improvements in VO₂max, which may explain the discrepancy.

Compared to the CON group, we found that a HIIT protocol performed 3 sessions per week (g = 1.21, p < 0.01) yielded superior outcomes compared to 2 sessions per week (g = 0.69, p = 0.07) and more than 3 sessions per week (g = 1.02, p = 0.15). In the HIIT versus CT study design, an additional HIIT training session (g = 0.48, p = 0.10) based on 2 sessions per week (g = 0.40, p = 0.70) did not result in significant improvement. A HIIT training regimen conducted 6 sessions per week reported a larger effect size (g = 1.77), but given the limited number of studies, this result should be interpreted with caution [44]. Notably, in the study by Sarker et al., athletes were classified as Tier 2, which may explain the observation of such a large effect size [44].

In the subgroup analysis of the SIT versus CON study design, only the 3 sessions per week group resulted in a significant improvement (g = 0.82, p < 0.05), whereas the intervention above 3 sessions per week showed a smaller effect size (g = 0.03, p = 0.93). Notably, a 2 sessions per week SIT regimen reported the largest effect size (g = 2.44) [53]. Given the limited number of studies, this result should be interpreted cautiously, and future research is needed to explore this in more detail. In the SIT versus CT studies, 3 sessions per week (g = 0.44, p = 0.11) showed better outcomes than the 2 sessions per week group (g = 0.38, p = 0.48) and those exceeding 3 sessions per week (g = 0.18, p = 0.63). Therefore, setting the training frequency of SIT to 2–3 sessions per week may yield better outcomes.

Work-to-recovery ratio

The training protocols for RST, HIIT, and SIT comprise repeated bouts of intense working interspersed with recovery periods. The total workload of interval training is influenced by the number of intervals, work intensity, duration of work intervals, and recovery duration [11]. The work-to-recovery ratio (WRR) is defined as the ratio of work duration time to recovery duration time for a single interval. The WRR significantly impacts the development of both aerobic and anaerobic capacities, as well as the overall effectiveness of the training program. Biological adaptation mechanisms indicate that the body’s recovery process after strenuous exercise includes the replenishment of phosphocreatine (PCr) (with a half-life of 30 s and complete recovery occurring in approximately 3 min), the removal of hydrogen ions (H⁺), and the restoration of acid-base balance within the primary motor muscle groups (6–10 min) [90, 91]. From a training perspective, sufficient recovery intervals are crucial for facilitating metabolic homeostasis and maintaining optimal training intensity. An imbalance between the demands of the athlete’s body and recovery duration during intervals may lead to premature fatigue, adversely affecting the ability to complete the training program.

In the subgroup analysis of SIT versus CON, a WRR > 1:4 (i.e., less recovery time) resulted in greater improvements in VO₂max (g = 1.13, p = 0.08; average work duration: 21.3 s; average recovery duration: 31.3 s) compared to a WRR of 1:8 (g = 0.41, p = 0.62; average work duration: 26.7 s; average recovery duration: 260 s) or 1:6 (g = 0.29, p = 0.56; average work duration: 20.8 s; average recovery duration: 104 s). Regarding the recovery interval duration for SIT, we found a significant nonlinear dose-response relationship. Specifically, when the recovery duration exceeded 96 s, the effect of SIT on VO₂max improvement was no longer significant, indicating that longer recovery times do not lead to further improvements. This finding supports prior research that recovery periods exceeding 120 s result in low aerobic demand and hinder the ability to induce endurance adaptations [92]. However, other studies indicate that a recovery time of less than 30 s may lead to an inability to achieve the desired intensity in subsequent training [92, 93]. In SIT, a recovery duration of less than 20 s limits the resynthesis of PCr, increasing reliance on aerobic energy sources during subsequent sprints. This may result in athletes being unable to perform subsequent sprint training at an optimal level of quality.

Regarding for HIIT, our analysis comparing HIIT to CON revealed significant improvements for WRR > 1 (g = 1.02, p < 0.05; average work duration: 240 s, average recovery duration: 174 s) and WRR between 0.5 and 1 (g = 0.79, p < 0.05; average work duration: 117.9 s, average recovery duration: 156.4 s). However, no significant improvement was observed in the group with a longer recovery time of < 0.5 (g = 0.93, p = 0.07; average work duration: 30 s, average recovery duration: 170 s). In the HIIT verse CT, we observed that HIIT with a WRR of 1 produced a greater effect size (g = 0.91, p = 0.21; average work duration: 105 s, average recovery duration: 105 s), while the < 0.5 group showed only a moderate improvement (g = 0.60, p = 0.10; average work duration: 50 s, average recovery duration: 180 s). Only one study with a WRR greater than 1 reported a negligible effect size (g = -0.02, p = 0.96) [74].

Interestingly, we found an inverted U-shaped relationship between work duration and the improvement of VO₂max in HIIT, with optimal benefits observed at 140 s. Furthermore, a significant inverted U-shaped association emerged in WRR, with the optimal value of WRR detected at 0.85. Combining these findings, we suggest that a HIIT protocol with a work duration of 140 s and a recovery time of 165 s may yield optimal benefits. Previous study indicates that HIIT with a duration of 2–3 min and a WRR of 1 can more effectively stimulate lactate oxidation and improve VO₂max [93]. Our meta-analysis results corroborate this finding, which indicates that a HIIT WRR of 1 or less (i.e., longer recovery duration) is associated with superior improvements. In HIIT, recovery durations of 2, 3, or 4 min can all achieve their improvement ceilings. However, 2 min recovery induces greater cardiovascular and metabolic stress than 4 min, significantly activating anaerobic glycolysis. Therefore, athletes can utilize shorter recovery duration to create a greater overall physiological challenge [94]. Considering enabling VO₂max to be achieved more rapidly during the subsequent work duration, there is no evidence to suggest that further extending the recovery duration in HIIT enhances benefits by maintaining oxygen uptake levels during recovery [95].

Unfortunately, the training protocol factor was not identified as a significant moderator of RST training effects in the subgroup or dose-response effect analysis.

Training mode

In the study design comparing HIIT to CON, we found that running mode HIIT significantly improved VO₂max (g = 0.96, p < 0.05). One study reported a cycling HIIT (g = 0.34) [34], while another reported a rowing HIIT (g = 0.62) [35]. In the HIIT versus CT design, the effects of the running mode (g = 0.99, p = 0.15) and the rowing mode (g = 0.52, p = 0.20) were not significant. We observed similar results in SIT, in the SIT versus CON design, running mode (g = 2.44, p < 0.01) outperformed cycling (g = 0.69, p < 0.05) and rowing (g = 0.49, p = 0.32), with training mode being a significant moderating factor (p < 0.05). Previous study comparing running with cycling mode SIT found no significant differences between the two modes. Interestingly, improvements in VO₂max were more significant when testing was conducted under the same training modality (i.e., running-based training with treadmill VO₂max measurement) [96]. Similar to the previous systematic review that focused on the results of time trials, the improvements observed in rowing mode were the least significant compared to running and cycling. It remains unclear what causes these performance enhancement differences [97]. One possible explanation is that rowing interval training often reports VO₂peak values from upper-body incremental tests, where a lower ceiling may limit the effect size.

Limitations

Several factors should be considered when interpreting the results of this study. First, the inclusion of nonrandomized trials in the analysis may introduce a risk of bias and imprecision in the findings. Nevertheless, our methodology facilitated a more comprehensive synthesis of the available evidence from RST, HIIT, and SIT, while assessing a lower overall risk of bias. Second, all interventions were conducted in conjunction with regular training, which precluded the ability to isolate the effects of each training method. Additionally, relying on VO₂max or VO₂peak as assessment metrics has certain limitations. For instance, in well-trained older runners, performance improvements are not always associated with changes in maximal oxygen consumption, suggesting that other factors may play a significant role [98]. Furthermore, in the subgroup analyses, some subgroups were limited by a small number of studies, with certain groups comprising only 1–2 studies, which restricts the generalizability of the findings, thus, these results should be interpreted with caution.

In light of these limitations, future practical investigations are essential to address several knowledge gaps. Comparative studies of interval training based on power measurements and load monitoring are necessary, particularly those focusing on heart rate load during work intervals. Investigating potential discrepancies in the efficacy of the three training modalities across various programs would be beneficial. Additionally, future research should explore the optimal recovery periods for active versus passive recovery. Furthermore, additional scientific experimental studies are necessary to provide conclusive evidence regarding the dose-response relationship and the underlying mechanisms involved in enhancing oxygen uptake levels, thereby demonstrating that interval training improves these levels.

Conclusion

This systematic review and meta-analysis of 51 studies enrolling 1,261 athletes aimed to compare the effectiveness of RST, HIIT, and SIT in improving the VO₂max of athletes. The network meta-analysis indicated that RST, HIIT, and SIT each induced significant increases in VO₂max compared to CON, with a ranking of RST > HIIT > SIT > CT > CON as per the P-score, no statistically significant differences were observed among RST, HIIT, and SIT. According to subgroup analysis and meta-regression, a HIIT running regimen conducted 3 sessions a week for more than 3 weeks, with each work interval lasting 140 s and a recovery period of 165 s, tends to yield superior results. Similarly, SIT conducted 3 sessions a week for 3–6 weeks, with a recovery time not exceeding 97 s, tends to yield superior results. Conducting RST 3 times a week for 2 weeks is sufficient to achieve improvements.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Q. Y.: Writing–original draft, Data curation, Conceptualization, Methodology. J. W.: Writing–review & editing, Supervision, Conceptualization. D.G.: Methodology.

Funding

This work was supported by the Major Project of Philosophy and Social Science Research in Colleges and Universities of Jiangsu Province (2021SJZDA173), and the Jiangsu Social Science Fund (24TYB008).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable, this systematic review and meta-analysis was registered in PROSPERO.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.