The protocol has been registered in the PROSPERO registry (CRD42023401649; http://www.crd.york.ac.uk/PROSPERO) and conducted in accordance with the systematic review checklist (PRISMA 2020).

Literature searches were performed in the PubMed, Scopus, Web of Science, Cochrane Library, and China National Knowledge Infrastructure(CNKI) databases (details of the search strategies are reported in Table 1 ).

The search dates were all from the creation of the database to April 1, 2023 The scientific databases were searched according to three criteria: study population (“diabetes mellitus”, “type 2 diabetes mellitus”), medical interventions (“high-intensity interval exercise”, “sprint interval training”, “HIIT”) and outcomes (“blood glucose”, “glucose tolerance disorder”, “lipids”). All search strategies were conducted in the relevant databases using English and Chinese. Two researchers independently completed the initial screening of the articles and the statistics of the basic information in the included literature and the changes in the effect indicators before and after the intervention, and the third researcher negotiated the resolution of disputes when they existed.

The eligibility criteria were established depending on the PICOS (population, intervention, comparison, outcome, study design) items.

(P) The study included patients with type 2 diabetes who met the World Health Organization’s diagnostic criteria (fasting blood glucose≥7.0 mmol/L or OGTT 2-hour glucose≥11.1 mmol/L or HbA1C≥6.5%), and who had an age≥18 years; there were no restrictions on the gender or race of the study participants. Patients with clinically manifest cardiovascular diseases, acute complications, and pregnant or lactating women were excluded.

(I) The intervention group participated in only the HIIT exercise program. HIIT exercise protocol consists of three phases: a warm-up phase, an alternating exercise phase, and an exercise recovery phase (13). The alternating phase alternates high-intensity exercise with low-intensity exercise (the exercise protocol of HIIT is shown in Figure 1 ). During the high-intensity exercise phase, the heart rate should be 75%-95% of the HRmax for 60 s, and during the low-intensity interval phase, the heart rate should be 45%-65% of the HRmax for 60 s. A total of 6–8 sets were completed; HIIT is a form of exercise that can be practiced in a variety of ways (exercise cycle ≥ 2 weeks; frequency ≥ 2 times/week). Exercise types include running, cycling, resistance bands, and unassisted exercise. Studies on joint interventions by combining strength training, diet, medicine, health education, and other means were excluded.

(C) The comparators included moderate-intensity continuous training (MICT, 46%-63% VO2max or 64%-75% HRmax; a duration of usually more than 20 min; types such as running, cycling, walking, body mechanics, and Tai Ji), combined aerobic and resistance training, routine care groups, and static stretching. Studies with control groups practicing low- or moderate-intensity HIIT were excluded.

(O) The outcomes included any of the following indicators. Primary outcomes: fasting blood glucose, glycosylated hemoglobin, fasting insulin, and 2h-PG; secondary outcomes: total cholesterol, triglyceride, high-density lipoprotein, and low-density lipoprotein. Studies that did not contain relevant outcome indicators were excluded.

(S) The study design included randomized controlled trials. Case reports, abstracts, reviews, lectures, commentaries, and data that could not be extracted were excluded.

We used the Cochrane Quality Assessment Tool, as the included articles were randomized controlled trials. The risk of bias and methodological quality of the included studies were assessed by two evaluators using Review Manager 5.3 in terms of selective bias (randomized sampling, grouping), implementation bias (whether the experiment was blinded to the subjects and experimenters), measurement bias (whether the experiment operator was blinded to the endpoints), follow-up bias (completeness of the results), selective reporting bias and other biases were evaluated. The outcomes were expressed as low- risk, unclear, and high- risk. If a dispute arose between the two evaluators during the quality evaluation, a third evaluator was invited to participate to reach harmonization.

After screening the literature, two researchers independently extracted the following data from the eligible literature: the external characteristics of the literature (title, authors, year of publication, nationality of the authors); basic information about the subjects (age, gender, country, sample size, duration of the disease); the experimental design and exercise intervention protocol (training period and frequency, duration, intervention mode, intensity); and the outcome indicators related to the study.

The subgroups were the intensity of exercise (75%-79% HRmax, 80%-89% HRmax and >=90%, HRmax), exercise period (≤8 weeks, 9–12 weeks, and >12 weeks), exercise duration (≤30 min/time and >30 min/time), age (<40 years, 40–59 years, and ≥60 years), and disease duration (<5 years, 5–10 years, and >10 years). In cases of disagreement between the two persons, a third person summarized them and determined their subgroups through a group discussion. The outcome indicators need to be extracted separately for the mean and standard deviation of the pre-test and the mean and standard deviation of the post-test for the intervention and control groups in the study. The mean standard deviation of the difference between the pre-test and post-test was calculated separately.

For some studies (14–24), as they had multiple control groups, we extracted the data across multiple groups and included them in the meta-analysis ( Table 2 ). For FPG, we evaluated 14 studies, of which 21 compared HIIT and control groups. For 2h-PG, 9 studies were evaluated, with 15 comparisons between the HIIT and control groups. For FINS, 8 studies were evaluated, with 14 comparisons between the HIIT and comparison groups. For HbA1c, 18 studies were evaluated, with 30 comparisons between the HIIT and control groups. For TC, 13 studies were evaluated, with 23 comparisons between the HIIT and control groups. For TG, 13 studies were evaluated, and 23 comparisons were made between the HIIT and comparison groups. For HDL, 15 studies were evaluated, with 24 comparisons between the HIIT and control groups. For LDL, 15 studies were evaluated, with 24 comparisons between the HIIT and comparison groups.

To explore the heterogeneity, we conducted a sensitivity analysis, applying a literature-by-exclusion approach. Publication bias was assessed with a visual inspection of a funnel plot. When significant bias was detected, we performed a trim-and-fill analysis.

Stata 16 software was used for meta-analysis. The data included in the study were continuous and were expressed as the mean difference (MD) and 95% confidence interval (CI). The presence of heterogeneity among the studies was tested using the I2 test. Subgroup analyses were performed to analyze the characteristics of the studies and sources of heterogeneity across various classifications.

To reflect the practical value of the effect sizes for clinical purposes, the effect size was calculated according to Hedge s’ g. It is also widely used in meta-analysis. Hedges suggested that g values of 0.2, 0.5, and 0.8 represent small, medium, and large effect sizes, respectively (34).

Through the search strategy, a total of 1076 articles were initially retrieved from 5 databases, and 390 articles that were duplicates or for which the full text could not be accessed were deleted. After reading the article titles and abstracts, 640 articles were excluded, and the full text of the remaining 46 articles was read and screened according to the inclusion and exclusion criteria, resulting in the inclusion of 22 articles. The specific flow of the included studies is shown in Figure 2 .

This study encompassed RCTs exclusively as part of its research design, with all subjects belonging to the type 2 diabetic population, totaling 1268 patients. The experimental group received interventions involving HIIT exercise, which included running (n=8) (17, 26–28, 33, 35, 36), cycling (n=23) (15, 17, 27, 29–31, 35, 37, 38), unassisted exercise (n=5) (16, 20, 25), resistance bands (n=2) (30, 31), and walking (n=2) (22). Meanwhile, the control group was based on aerobic exercise and usual care. The exercise intensity was categorized as moderate in 17 studies, low in 5 studies, and high in 6 studies. The primary outcome indicators were addressed in 32 studies. Details of the basic characteristics of the included literature are shown in Table 2 .

Regarding the quality assessment, 13 studies were identified as having a high risk, and the results are shown in Figures 3 , 4 . In Figure 3 , green means “low risk”, red means “high risk”, and yellow means “unclear”. In Figure 4 , green means “low risk”, red means “high risk”, and yellow means “unclear”. The results were as follows: (1) randomization— 2 studies had high risk, and 4 did not describe the random allocation method; (2) allocation concealment, 16 studies did not describe the specific allocation method; (3) blinding— 3 studies did not blind the subjects, 2 did not blind the evaluators, and most of the studies did not describe the method of blinding; (4) incomplete data reporting— 9 studies had case dropout, and 2 did not mention the number of subjects at the beginning; (5) reporting bias— 2 papers had case dropout, and 2 did not mention the number of starting subjects; and (6) the results were as follows: (1) the randomization method was not described. In reporting bias, two papers had biases.

FPG: There are 21 studies (14–16, 18, 21–23, 25–28, 30, 32, 36) that report the effect of HIIT on the FGB index in a type 2 diabetic population. The results show that HIIT had a large effect size on FGB (MD: -0.55; 95% CI: -0.85- -0.25; Hedges’ g = 0.98; p < 0.01). The left side of the center line is “in favor of HIIT”( Figure 5 ). Because the I² was 76.6%, a random-effects model was chosen for analysis. Notably, there were notable disparities in the exercise duration and disease duration among the subgroups ( Table 3 ), with effect size values for exercise durations of 30 minutes or less surpassing those for durations exceeding 30 minutes (p < 0.01). Furthermore, individuals with a disease duration of less than 5 years (p<0.01) exhibited a significant effect size in terms of reducing FPG. Conversely, no significant differences were observed in the subgroups based on the exercise period or exercise intensity (p=0.267, p=0.194), and the results of fasting glucose were not affected by them.

2h-PG: Fifteen studies (15, 16, 19, 22, 23, 26, 30, 36, 39) evaluated the effect of HIIT on 2h-PG metrics in a type 2 diabetic population. The results show that HIIT had a large effect size on 2h-PG (MD: -0.36; 95% CI: -0.57- -0.14; Hedges’ g = 1.05; p<0.01) ( Figure 6 ). Because the I² was 42.4%, a random-effects model was chosen for analysis. There were no statistically significant differences in the 2h-PG results for the exercise cycle, intensity, duration and disease duration subgroups (p = 0.10, 0.25, 0.32, 0.36, respectively) ( Table 3 ). However, the results were significant for exercise cycles of 8–12 weeks (p = 0.05), exercise intensities of ≥90% (p<0.01), and exercise durations of ≤30 minutes (p <0.01); all had large effect sizes and a good effect on reducing 2h-PG. The effect of 2h-PG reduction was more pronounced for a disease duration of <5 years.

FINS: Fourteen studies (14, 15, 18, 21–23, 30, 35) evaluated the effect of HIIT on FINS metrics in a type 2 diabetes population. The results show that HIIT had a FINS- lowering effect (MD: -0.41; 95% CI: -0.79- -0.03; Hedges’ g =1.07; p<0.01) ( Figure 7 ). Because the I² was 68.8%, a random-effects model was chosen for the analyses. The subgroup results for exercise intensity and duration were significantly different (p<0.01), but their confidence intervals all overlapped ( Table 3 ). Therefore, there were no significant differences among any of the above subgroups, and the FINS results were not affected by exercise intensity or duration. There was no statistically significant difference in age subgroups (p = 0.447), but the results showed a significant effect of HIIT in reducing FINS in the 40–60 year olds (p < 0.01). F or people over 60 years old, the effect was not significant (p = 0.972).

HbA1c: Thirty studies (14–27, 30–32, 35) evaluated the effect of HIIT on HbA1c metrics in a type 2 diabetic population. The results show that HIIT had a large effect size on HbA1c (MD: -0.60; 95% CI: -0.84- -0.36; Hedges’ g = 2.69; p<0.01) ( Figure 8 ). Because the I² was 69.4%, a random-effects model was chosen for analysis. There were significant differences in the exercise cycle subgroups ( Table 3 ), with larger effect size results for ≤8 weeks compared to 8–12 weeks and >12 weeks (p < 0.01). There were no significant differences in the subgroup results for exercise intensity, age or disease duration, so the HbA1c results were not affected by these. From the results, it can be concluded that the group with an age of < 40 years (p < 0.01) had a large effect size, and the results were greater for exercise intensity of 80%-89%. HbA1c reduction was more pronounced for disease duration of <5 years.

TC: Twenty-three studies (14–16, 19, 20, 22–25, 29, 32, 33, 35) assessed the effects of HIIT on markers of TC in a type 2 diabetic population. The results showed that HIIT had a large effect size on TC (MD: -0.58; 95% CI: -0.80- -0.36; Hedges’ g =2.36; p<0.01) ( Figure 9 ). Because the I² was 59%, a random-effects model was chosen for analysis. There was a significant difference in the subgroup results for exercise cycle and disease duration ( Table 4 ), with an exercise cycle >12 weeks having a greater effect size compared to the other two subgroups. The effect size for a disease duration of less than 5 years was greater than that for the subgroup with >5 years of disease duration. However, there were no significant differences between the exercise intensity and age subgroups (p = 0.175, 0.228) and the TC values were not affected by the intensity of exercise or age, but the results show that the exercise intensity 75%-79% and 80%-89% both had large effect sizes.

TG: Twenty-three studies (14–16, 19, 20, 22–25, 29, 32, 33, 35) evaluated the effect of HIIT on markers of TG in type 2 diabetic population. We found that there was a large effect size of HIIT intervention on TG (MD: -0.50; 95% CI: -0.86- -0.14; Hedges’ g =1.50; p<0.01) ( Figure 10 ). Because the I² was 83.9%, subgroup analyses were performed for the exercise period, intensity, duration, age, and disease duration. There were significant differences in the exercise duration subgroups ( Table 4 ), with durations >30 min reducing TG values more than durations ≤30 min. There were no significant differences between the subgroups for the remaining groups, and the TG values were not affected by the exercise cycle, intensity, age, or disease duration. An exercise cycle of 8 to 12 weeks was found to have a large effect size. An age of 40–60 years (p<0.01) had a better effect size for exercise, and an age of >60 years had no significant effect size for exercise (p=0.173). The effect sizes were larger for illness durations of less than 5 years.

HDL: Twenty-four studies (14–16, 19, 20, 22–26, 29, 32, 33, 35, 36) assessed the impact of HIIT on HDL indicators in a type 2 diabetes population. The r esults showed that HIIT promoted HDL with a large effect size (MD: 0.62; 95% CI: 0.29–0.95; Hedges’ g =1.19; p<0.01) ( Figure 11 ). Because the I² was 81.9%, a random-effects model was chosen for analysis. Although the differences between the subgroups for exercise intensity and disease duration were significant (p<0.01) ( Table 4 ), there was an overlap in the confidence intervals, so none of the above subgroups were significantly different, suggesting that the HDL results were not affected by the exercise cycle, intensity, age or disease duration. The results show that exercise cycles of >12 weeks were more effective in promoting HDL in people than exercise cycles of ≤ 8 weeks. Disease durations of less than 5 years had a large effect size. An exercise intensity of 80%-89% (p<0.01) had a large effect size. The group aged 40–60 (p<0.01) and the group over 60 (p<0.01) were both highly effective groups, with greater contributions to HDL.

LDL: Twenty-four studies (14–16, 19, 20, 22–26, 29, 32, 33, 35, 36) assessed the impact of HIIT on LDL metrics in a type 2 diabetes population. The results showed that HIIT had a large effect size on LDL (MD: -0.31; 95% CI: -0.56–0.08; Hedges’ g = 0.91; p<0.01) ( Figure 12 ). Because the I² was 67.6%, a random-effects model was chosen for analysis. Although there was a significant difference between the exercise cycle and disease duration subgroups (p<0.01) ( Table 4 ), the confidence intervals overlapped, so there was no significant difference between the subgroups, suggesting that the LDL results were not influenced by the exercise cycle, intensity, duration, age or disease duration. The results show that an exercise period of >12 weeks (p<0.01) had a large effect size. Exercise durations of >30 min (p<0.01) had a better exercise effect. An age of 40–60 years of age (p<0.01) significantly reduced LDL values with moderate effect size. A disease duration of <5 years (p<0.01) had a large effect size.

Sensitivity analysis of the findings for each indicator showed that the fasting glucose indicator produced a large bias due to the removal of the article by Deng, 2020 (27); the HDL indicator produced a large bias due to the removal of the articles by Adalberto, 2020(2) (20), Mostafa, 2021(1) (14), and Alvarez, 2016 (33); the 2h-PG indicator produced a large bias due to the removal of the articles by Zadeh, 2022 (19); and the FINS indicator produced a large bias due to the removal of the articles by Kamilla, 2017(1) (23) and Kamilla, 2017(2) (23). Therefore, the combined results of the above indicators may be unstable, and the combined results of the other indicators may be more stable (Annex 1).

We examined the publication bias for each indicator based on the funnel plot analysis. It was found that the study sites were found to be basically distributed on either side of the x=0 vertical line, but there was still a small number of study sites scattered regarding the FPG index, LDL index, and TC index. This suggests that there may be some publication bias (Annex 2).