A comprehensive literature search was conducted in the Cochrane Library, PubMed, Embase, Web of Science, MEDLINE, ClinicalKey, CNKI, Wanfang Data, VIP Database, and the Chinese Biomedical Literature Database, covering publications from database inception to March 2023 in both English and Chinese. Search terms included combinations of subject headings and free-text terms related to the population (“diabetic neuropathies”, “diabetic autonomic neuropathies”, “painful diabetic neuropathies”, “diabetic polyneuropathy”), intervention (“exercise”, “aerobic exercise”, “exercise training”, “physical activity”, “physical exercise”), and study type (“randomized controlled trial”, “RCT”). The detailed search strategy for PubMed is provided in Table 1.

The inclusion criteria were as follows: ① study subjects: clearly diagnosed DPN patients, aged ≥18 years. ② interventions: control group received usual healthcare, intervention group received exercise training on the basis of usual healthcare. ③ at least one of the following outcome indicators was included: ankle dorsiflexion range of motion left (ROML), ankle plantar flexion range of motion left (ROML), hallux strength, toes strength, glycated hemoglobin (HbA1c), Michigan diabetic neuropathy score (MDNS), Michigan neuropathy screening instrument (MNSI), five-time sit-to-stand test (FTSST test), body mass index (BMI). ④ randomized controlled trials (RCTs) used.

Exclusion criteria were as follows: ① studies of patients with gestational diabetes. ② literatures that cannot get full text or original data. ③ repeated publications. ④ gray literatures such as conference papers.

Two investigators independently screened the literature. Articles with different opinions were arbitrated by a third investigator. Two independent investigators collected and cross-checked the basic information and extracted data. A standardized data extraction form was designed, including author(s)/year, Country, Mean age, Sample size, Duration of diabetes, HbA1c, Exercise program for intervention group, Control group interventions, Follow-up, Outcomes.

In this study, we used the risk of bias assessment tool recommended by the Cochrane Collaboration Network (25) to assess the risk of bias of the included RCTs. The tool assesses the risk of systematic bias in studies from seven dimensions: randomization method, allocation concealment, blinded implementation, data integrity, selective reporting, and other potential biases. Each dimension was judged as “low risk”, “unclear risk”, or “high risk” based on literature reports. If all dimensions are low risk, the study is rated as low risk overall; if there is at least one high risk dimension, the study is judged as high risk; if there is one or more unclear risks and the remainder are low risk, the study is rated as unclear risk.

In this study, meta-analysis was performed using Review Manager software (RevMan 5.4) as recommended by Cochrane Collaboration. Heterogeneity was assessed by the I² statistic, where I² values of 25%, 50%, and 75% represented low, moderate, and high heterogeneity (26). If the heterogeneity test results showed P ≥ 0.1 and I² < 50%, the fixed-effect model was used; if P < 0.1 and I² ≥ 50%, and the heterogeneity could not be explained by clinical or methodological differences, the random-effects model was used. For the analysis results with high heterogeneity, the possible sources of heterogeneity were further explored through subgroup analysis (with the type of exercise intervention as the grouping variable). When the number of included studies was ≥ 10, Egger's test was used to assess the possibility of publication bias (27). Sensitivity analyses assessed the robustness of the combined results by excluding individual studies one by one. Effect sizes for continuous variables were presented as standardized mean difference (SMD) or weighted mean difference (MD), and 95% confidence intervals (95% CI) were calculated to estimate the precision of the overall effect. The level of heterogeneity was set at α = 0.1, and the remaining tests were set at α = 0.05.

A total of 3,667 relevant studies were retrieved from the databases. After removing duplicates (n = 524) and excluding irrelevant titles and abstracts (n = 3,101), 42 studies were selected for full-text review. Among these, 32 studies were excluded due to not meeting the outcome criteria, not meeting the intervention criteria, or the full text being unavailable. Finally, 10 studies were included in the analysis. The study selection process is illustrated in Figure 1.

The included studies were published between 2013 and 2022, covering six countries. A total of 10 studies were included, comprising seven articles in English and three in Chinese, with a combined sample size of 625 participants (307 in the intervention group and 318 in the control group). The interventions primarily consisted of aerobic and resistance exercises, while other types included tap dance, short-term strength training, and balance training. The frequency of the interventions ranged from two to five times per week, and the duration ranged from 8 to 16 weeks. A brief description of each intervention is provided in Table 2, and detailed information on all included studies is presented in Table 3.

Of the 10 included studies, eight studies reported using random number table or lottery to generate random sequences, and the remaining two studies only mentioned randomization and did not describe specific random implementation methods. Two studies were stored in opaque, sequentially numbered, sealed envelopes using an allocation order. Five studies mentioned blinding of outcome assessors, two mentioned that study reports could not blind study subjects and interventions, and the remaining studies did not mention blinding of subjects, researchers, or outcome assessors. All studies described losses to follow-up and withdrawals in detail and reported pre-specified outcome measures. None of the investigators mentioned other risk of bias situations, such as conflicts of interest. The Cochrane risk of bias summary and graph are presented in Figures 2A, B, respectively.

A total of five studies (29–33) included HbA1c as the outcome measure. The heterogeneity test showed that there was great heterogeneity among the studies (P < 0.001, I² = 91%). The random-effects model was used to combine the effect size. The results (Figure 3A) showed that the intervention group was superior to the control group in reducing HbA1c [SMD = −1.44, 95% CI (−2.30, −0.57), P < 0.01]. Sensitivity analysis of included studies revealed little change, suggesting good stability of the results. However, the final results remained statistically significant (P < 0.01), and the intervention group was superior to the control group in reducing HbA1c.

In the subgroup analysis (Figure 3B), the heterogeneity of resistance exercise was statistically significant among four studies (P < 0.001, I² = 88%), and significant difference detected between the resistance exercise and control groups [SMD = −1.14, 95% CI (−2.00, −0.27), P < 0.05]. One study reported Based on peripheral neuropathy as dominant movement pattern (EPN), the heterogeneity was not applicable. Overall, there was a difference between the subgroups (P < 0.01, I² = 87.3%), indicating that the type of exercise was one of the sources of heterogeneity.

BMI was included as an outcome measure in three studies (31, 32, 35), and there was no statistically significant difference in heterogeneity (P = 0.49, I² = 0%, Figure 4A). The fixed effects model combined effect size showed that the intervention group was superior to the control group in reducing BMI [MD = −0.86, 95% CI (−1.15, −0.57), P < 0.001].

Two studies (34, 36) included ankle dorsiflexion ROML as the outcome measure. The heterogeneity test showed that there was great heterogeneity among the studies (P = 0.06, I² = 72%). The results (Figure 4B) showed that the intervention group was superior to the control group in increasing ankle flexion range of motion [SMD = 0.61, 95% CI (0.13, 1.08), P < 0.05].

Two studies (34, 36) included a ankle plantar flexion ROML, but there was no statistically significant difference in heterogeneity (P = 0.38, I² = 0%, Figure 4C). The random-effects model combined effect size showed that the intervention group was superior to the control group in increasing ankle extension range of motion [SMD = 0.59, 95% CI (0.12, 1.06), P < 0.05].

Two studies (28, 36) included hallux strength and toes strength, and there was no statistically significant difference in heterogeneity between studies (Figures 5A, B). The fixed effects model combined the effect size showed that the intervention group was superior to the control group in increasing hallux strength [MD = 1.89, 95% CI (1.00, 2.78), P < 0.001] and toes strength [MD = 2.51, 95% CI (1.69, 3.33), P < 0.001].

Three studies (19, 29, 34) included FTSST test, and there was no statistically significant difference in heterogeneity between studies (P = 0.27, I² = 24%, Figure 5C). The fixed effects model combined the effect size showed that the intervention group was superior to the control group in enhancing lower extremity functional strength [MD = −2.82, 95% CI (−3.88, −1.76), P < 0.001].

Three studies (28, 30, 36) used MNSI as an outcome measure, and there was no statistically significant difference in heterogeneity (P = 0.14, I² = 48%, Figure 6A). The fixed effects model combined the effect size showed that the intervention group was superior to the control group in improving peripheral neuropathy [MD = −1.20, 95% CI (−2.02, −0.38), P < 0.01].

Other three studies (30, 32, 35) used MDNS as an outcome measure, and there was a statistically significant difference in heterogeneity (P < 0.001, I² = 98%, Figure 6B). However, there was no statistically significant difference between groups in improving peripheral neuropathy [MD = −3.57, 95% CI (– 7.46, 0.33), P = 0.07].

This study showed that exercise intervention, especially ankle joint and distal lower limb muscle training, significantly improved ankle range of motion, toe and lower limb muscle strength in patients with DPN. Meta-analysis of two studies showed that both foot and ankle training and tap dance interventions significantly increased ankle dorsiflexion and plantar flexion range of motion, consistent with previous systematic reviews supporting the effectiveness of physical therapy targeting the foot and ankle in improving joint flexibility (37). The possible mechanism is that this type of training extends joint range of motion by applying mechanical stimulation to the soft tissues around the ankle joint and enhancing its flexibility and neuromuscular control (38). Meta-analysis of two other studies showed that foot and ankle training significantly enhanced hallux and toe muscle strength in patients with DPN, with low heterogeneity among studies, suggesting that it has a consistent effect in enhancing distal muscle strength. This is consistent with the findings of Prókai et al. (39). The latter points out that DPN patients are often accompanied by decreased isokinetic torque of the ankle joint and foot muscle atrophy, especially significantly weakened flexor hallux and minor muscle groups (40). Muscle strength of the hallux and phalanges is key to maintaining gait stability and preventing falls, and its enhancement helps to improve weight-bearing control and dynamic balance, thereby improving overall functional mobility (41).

In addition, a meta-analysis of three studies showed that different types of exercise interventions significantly improve functional strength level of the lower limbs in DPN patients. Among them, progressive resistance training promotes muscle protein synthesis and reverses muscle fiber atrophy by activating the mTOR-p70S6K pathway, delaying muscle loss and enhancing muscle strength more effectively than low-intensity training (42, 43); strength and balance training helps to improve knee extensor strength, motor ability and walking performance (44, 45); while tap dance training improves coordination and balance control ability in addition to enhancing lower limb muscle strength (46). Cruvinel-Júnior et al. (47) further pointed out that foot and ankle exercise training can not only directly improve foot function, but also promote functional recovery by indirect mechanisms such as relieving pain and increasing joint mobility. In summary, exercise intervention, especially foot and ankle training, is effective in improving lower limb function in patients with DPN and can be used as a clinically important rehabilitation strategy.

The clinical manifestations of DPN can be reflected by BMI and HbA1c. This study shows that exercise intervention significantly improved BMI as well as HbA1c in DPN patients. Numerous studies have demonstrated that exercise interventions effectively promote body fat reduction with weight management in different populations (48, 49). This meta-analysis further complements the evidence-based support of exercise for BMI improvement in DPN patients. Exercise lowers BMI by increasing energy expenditure and optimizing metabolic function, particularly by enhancing skeletal muscle insulin sensitivity, promoting lipid metabolism, and reducing fat accumulation (50). The incidence and severity of diabetic neuropathy are positively correlated with the duration of hyperglycemia and blood glucose levels (51), and persistent or fluctuating hyperglycemia can induce irreversible peripheral nerve damage (52). Therefore, active glycemic control is essential for diabetes and its complications, particularly DPN. In this paper, the heterogeneity of different studies of HbA1c is large, but there are statistical differences between the intervention group and the control group, and the intervention group is superior to the control group, which is consistent with the results of previous studies (53, 54). Both EPN and resistance exercise had good effects on improving HbA1c by DPN. Structured resistance exercises >150 min per week in patients with T2DM reduced HbA1c by 0.89% (55). An animal study (56) showed that exercise regulates the metabolic system of the body, and lowers blood glucose levels, thereby reducing the degree of microvascular injury, strengthening nerve conduction function, and ultimately improving peripheral neuropathy symptoms. Therefore, the effect of intensive glycemic control on improving DPN is positive, and the more stringent the glycemic control, the more patients benefit.

MNSI and MDNS are currently among the most commonly used screening tools for assessing DPN. Although studies have shown a significant reduction in the development of abnormal neurological examinations in patients undergoing exercise intervention (57), and improvements in neurological function may be attributed to the strengthening of existing sensory-motor pathways (58), the combined results of this meta-analysis are not fully consistent with previous studies. In the three studies using the MNSI score, the combined results showed a significant advantage of exercise training over usual care in improving neuropathy symptoms. However, in the three studies using the MDNS score, the results did not reach statistical significance and did not clearly indicate a significant effect of exercise training on this scale. The inconsistency in the above results may reflect multiple differences in study populations, intervention protocols, outcome setting, and bias control. Among them, MNSI mainly assesses subjective symptoms and some physical parameters, which are suitable for early screening and are greatly subjectively affected; MDNS focuses on objective neurological function such as muscle strength, reflexes and pain sensation, which is suitable for moderate to severe DPN with more stable results, but requires high study design (59). Although the results of this study suggest that exercise training may help improve neurological function in some indicators, the relevant evidence is still insufficient. In the future, more rigorous, high-quality randomized controlled trials should be conducted in combination with DPN grading and objective indicators (e.g., nerve conduction velocity) to determine the optimal exercise intervention program.

In this study, a detailed search strategy was employed by searching different types of databases and trial registries. In addition, heterogeneity among studies of most outcome measures was low in comparison to different outcome measures, which indicated that the reliability of this meta-analysis was high. We try to explain the source of heterogeneity through sensitive analysis and subgroup analysis, so that the study results are more persuasive. In addition, we evaluated the effect of exercise training on improving DPN through a meta-analysis with a higher evidence-based level, suggesting that it can improve musculoskeletal function and clinical outcome measures in patients. It can provide clinicians with evidence to help them make better clinical decisions.

However, this study still has several limitations, and caution should be exercised when interpreting the results. First, all included studies lacked long-term follow-up data, and the sustained efficacy of exercise training for DPN remains unclear. Additionally, some studies failed to complete routine follow-up due to the impact of the COVID-19 epidemic, limiting a comprehensive assessment of long-term effects. Second, this study covers a variety of exercise intervention modalities, but most of them are single-center studies with small sample sizes, and there are some limitations in the representativeness and external generalizability of the results. Third, the included studies differed in terms of type of exercise, duration and frequency of intervention, which may lead to some degree of clinical heterogeneity. Finally, some studies lacked sufficient reporting on methodological details such as concealment of random assignment, blinded implementation, and outcome measurement, which may have biased the meta-analysis results. In the future, it is urgent to design standardized large-sample, multicenter randomized controlled trials to verify and expand the findings of this study.