The review study followed the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) 2020 statement (Page et al., 2021). A protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) ID: CRD42023436799.

The PICOS (populations, interventions, comparator interventions, outcomes and study design) framework guided the eligibility criteria selection (Cochrane Collaboration, 2024).

Three-step search for relevant randomized controlled trials (RCTs) as recommended by the Cochrane Handbook for Systematic Reviews of Interventions was conducted (Cochrane Collaboration, 2024). Four electronic databases (PubMed, Web of Science, EBSCO, Cochrane Library) were searched for articles published up to 9 September 2023. Then, searching was also done in published trial articles. Medical Subject Headings (MeSH) terms and keywords were chosen based on study design (“Randomized Controlled Trials”), exposure (“Exercise” OR “Training” OR “Physical activity”), outcomes (“Postural Control” OR “Gait Performance” OR “Biomechanics”) and participants (“Diabetic Peripheral Neuropathy”). The full search strategy for PubMed can be found in the online Supplementary Appendix B. Reference lists of included studies were also searched for relevant articles.

All returned titles were screened by the first author (DW) to exclude duplicate or non-relevant studies. The abstract of each remaining study was then independently reviewed by DW and XP during the literature search. Then the full texts of the remaining studies were independently reviewed by the two authors against the inclusion and exclusion criteria. Disagreements were discussed and consensus was reached among the authors in all cases. All studies in the systematic review were eligible for inclusion in the meta-analyses.

We collected data on authors, year of publication, number of participants allocated to the intervention and control groups, participant-based information (age, gender, BMI, duration of diabetes, HbA1c), type and duration of exercise intervention for the experimental and control groups (time, frequency), outcome measures, and study duration. This study gathered the outcomes including RCTs in exercise intervention, encompassing statistical metrics of continuous variables such as sample size, mean, standard deviation, and others for each study.

Two authors (DW and XP) independently assessed the risk of bias at the study level of included RCTs following the Cochrane Risk of Bias Tool (RoB 2) (Sterne et al., 2019). The seven items considered for the risk of bias included: the randomization process, bias arising from period and carryover effects, deviations from intended interventions, missing outcome data, measurement of the outcome, selection of the reported result and overall bias. Open and apply the macro using the Excel tool (website of Cochrane Methods Bias: https://www.riskofbias.info/) to assess the risk of bias for each article (Liu et al., 2021).

This study employed RevMan 5.3 (The Cochrane Collaboration, Copenhagen, Denmark) software to perform meta-analysis, subgroup analysis, and generate forest plots. The extracted data were all continuous variables, expressed as mean difference (MD) and their 95% confidence intervals (CI). If the units were inconsistent, standardized mean difference (SMD) was used. Heterogeneity of outcome measures was assessed using the I² statistic and p-values. If heterogeneity was low (I² ≤ 50%, p ≤ 0.01), a fixed-effects model was applied; if heterogeneity was high (I² > 50%, p > 0.01), a random-effects model was used (Higgins et al., 2003). Subgroup analyses focused on three key factors: postural control, gait, and muscle strength. Each factor was further subdivided based on different variables to minimize heterogeneity in the study. If significant heterogeneity persisted, sensitivity analysis was conducted to identify its sources. If the sources of heterogeneity could not be determined, descriptive analysis was performed. The significance level ɑ for the pooled effect size was set at 0.05.

After the initial search, a total of 532 articles were included. After removing 128 duplicate articles, 280 articles were excluded based on title and abstract screening. Further full-text retrieval, reading, and quality assessment led to the exclusion of 98 articles that did not meet the inclusion criteria. Ultimately, 23 articles were included in the study. The process of article inclusion and exclusion is illustrated in Figure 1. The characteristics of the study participants, intervention, control, and outcome measures were shown in the review (Table 1).

The 10 studies included in this research all exhibited a low risk in the randomization process (Kruse et al., 2010; Song et al., 2011; Lee et al., 2013; Sartor et al., 2014; Waheed et al., 2021; Dixit et al., 2016; Khan et al., 2022; Monteiro et al., 2022; Hizomi arani et al., 2023; Silva et al., 2023), with no instances of missing data. Among these, one study was an open-label randomized controlled trial (Rodrigues et al., 2022), resulting in a high risk in intervention allocation; two studies did not implement allocation concealment for participants or assessors (Melai et al., 2014; Venkataraman et al., 2019), leading to some risk in intervention allocation; and two studies failed to conceal allocation from treatment providers, leading to a high risk of bias in intervention compliance (Rodrigues et al., 2022; Yoosefinejad et al., 2015). The ten studies have not clearly described the concealment of the allocation of treatment and testing personnel, introducing some risk to intervention adherence (Melai et al., 2014; Abdelaal and El-Shamy, 2022; Zhao et al., 2021; Perrin et al., 2021; Waheed et al., 2021; Ahmad et al., 2019; Venkataraman et al., 2019; Allet et al., 2010; Mueller et al., 2013; Abdelbasset et al., 2020). The risk of bias assessment results for the included studies are presented in Table 2.

Impaired Balance is a strong risk factor for falls with a medium to large effect size (Chantanachai et al., 2021). In previous studies, these methods of balance testing had a high heterogeneity within or across settings (Perell et al., 2001), such as COP, TUG, BBS, OLS, and so on (Blodgett et al., 2022; Perell et al., 2001). Objective evaluation of postural control is usually based on the analysis COP using a force platform (Blodgett et al., 2022; Quijoux et al., 2020), as the gold standard for evaluation of balance. The velocity and sway area of COP were the best features for discriminating between fallers and non-fallers (Perell et al., 2001), and had a high correlation with the severity of neuropathy (Brown et al., 2015; Perell et al., 2001). In this review, three studies (Melai et al., 2014; Enders et al., 2023) included meta-analysis to estimate postural control between experimental and control groups and observed that could significantly decrease AP sway amplitude with EO and EC after balance training. Although studies presented stability in the ML direction as a key factor to differentiate high fall risk among older population (Brown et al., 2015), Figure 2 showed only COP sway path with EC had improved compared with EO.

Otherwise, five studies with other exercises were not included in the meta-analysis (Sartor et al., 2014; Zhao et al., 2021; Dixit et al., 2016; Venkataraman et al., 2019; Yoosefinejad et al., 2015). Dixit and coworkers only found that eight-week aerobic exercise could reduce COP sway velocity along the x-axis and increase ML displacement on foam with EC (Dixit et al., 2016). Zhao et al. also validated the effect of aerobic exercise significantly affected COP trajectory and elliptical area (Zhao et al., 2021). A study described a relatedly lower COP velocity measured by plantar pressure system after 12-week strength training (Sartor et al., 2012), but the number and quality of each study were relatively less. However, a meta-analysis reported that balance exercise intervention changed the COP parameters in either eyes or closed condition among older adults (age > 60), while resistance and multi-component exercise did not (Low et al., 2017). What are the possible explanations for the relatively fewer changes observed in COP sway indicators of ML direction? A study showed that ML sway path length/velocity did not change in older adults after balance exercises (Low et al., 2017), similar to this review. Moreover, another review showed that ML data were found to be more discriminatory than AP features (Quijoux et al., 2020). The possible reasons were the small sample size and inconsistencies in the data collection protocols, leading to high heterogeneity (69%) of the ML sway path.

In conclusion, exercise interventions revealed a moderate to high effect on balance performance parameters compared to the control group, and balance training significantly improved postural control to a greater extent compared to other exercises. Balance exercise, in this review, included regular stability training, sensorimotor training, proprioception training, and combined training above. A review presented that sensorimotor training also plays a crucial role in targeting balance control and existing sensory and motor signs and symptoms of DPN (Streckmann et al., 2022).

It is well known that gait performance, a natural daily activity heavily reliant on the synergy between the nervous and musculoskeletal systems, can be easily affected by the pathological process in older adults with DPN (Wang et al., 2022). Adults with DPN usually adopt a conservation gait with lower gait velocity, shorter stride length, longer stride time and stance time, compared with non-neuropathy diabetes (Wang et al., 2022). A study showed that exercise could significantly improve gait velocity compared to a control group, especially balance exercise (Allet et al., 2010). Some described that velocity and stride length effectively increased after 8-week sensorimotor exercise in older adults with DPN regardless of age (Reeves et al., 2021), similar to this review. In general, studies reported that lower walking speeds accompanied by increased falling risk in older adults, significantly discriminate between fallers and non-fallers (Reeves et al., 2017). Based on meta-analysis, balance exercise could more greatly increase gait velocity under both self-gait speed and fast-gait speed conditions than resistance training. Otherwise, this review reported that stride length only had an increase trend after exercise interventions. There was a potential factor that the sample size was small only 75 in the interventional group and 70 in the control group. Another reason was that cadence contributed 80% whereas stride length only contributed 20% to this change of gait velocity, resulting in difficulty regulating stride length (Allet et al., 2010).

Diabetes is responsible for the deterioration of muscle strength and becomes more severe as DPN progresses, leading to altered gait biomechanics, impaired stability and increased fall risks (Orlando et al., 2022). Some suggested that reduced strength of the knee and ankle may cause a disturbance in perturbation response in balance, potentially increasing the risk of falling. Although these results had a low level of quality and evidence, in this review, exercise could increase muscle strength and joint mobility in people with DPN, especially after foot-ankle functional training. Regarding the foot-ankle exercise, the exercise protocol is designed to consist of the same set: warm-up exercise, strengthening of the intrinsic foot muscles, strengthening of the extrinsic foot muscles, and functional exercise (e.g., balance and gait training), to manage the musculoskeletal complications related to diabetes (Silva et al., 2023; Monteiro et al., 2020). The observed phenomenon of reducing foot-ankle joint mobility and intrinsic muscle strength may be attributed to symptoms of distal peripheral neuropathy (Mustapa et al., 2016; Wang et al., 2022). Strength combined with functional training provided a better effect on increasing muscle strength of the lower extremity, compared to resistance training alone. Although resistance exercise was an effective intervention to combat muscle loss and delay some neurological symptoms (Melai et al., 2014; Venkataraman et al., 2019), people with neuropathy should be wary of weight-bearing exercise because of plantar loading links to potential foot ulcer development (Li and Hondzinski, 2012). Combinational exercises could more greatly improve metabolic control of the organism than aerobic and resistance training, similar to the previous study (Orlando et al., 2022). Improving foot functionality has a positive impact on people’s overall physical activity levels and quality of life (Streckmann et al., 2022). Foot-ankle functional training significantly increased ankle joint ranges of motion including dorsiflexion and plantarflexion in older adults with DPN. That is the foot-ankle exercise program focused mainly on the foot joints of muscle strength and mobility in gait or dynamic activities.

This review has some limitations. The primary limitation is the focus on short-term outcomes, without accounting for the potential long-term benefits or risks associated with the interventions. Moreover, this study did not conduct subgroup analyses based on exercise types or exercise prescriptions (such as intensity, duration, frequency, etc.), and failed to provide detailed guidance for the implementation of specific exercise programs in the daily rehabilitation of DPN patients. Furthermore, although 23 studies were included in this review, the exercise modalities assessed were not evenly distributed. Most studies focused on balance training, followed by multi-component and foot-ankle functional exercises. In contrast, strength training and whole-body vibration training were represented by fewer studies. This uneven distribution may contribute to increased heterogeneity across the meta-analysis and limit the precision of the confidence intervals. Therefore, clinicians should interpret the results with caution when considering them for decision-making.