Effectiveness of wearable technology-based physical activity interventions for adults with type 2 diabetes mellitus: A systematic review and meta-regression

2.1 Eligibility criteria

The inclusion of studies followed the eligibility criteria (Table S2): Participants diagnosed with T2DM were 18 years of age or older, regardless of the severity of the condition or mode of treatment (oral or parenteral medication). The interventions incorporated wearable technology to specifically target the promotion of PA through wearable devices. The comparators in this study included usual care, waitlist, no treatment, usual care, use of the wearable device alone, coaching, or educational advice. Primary outcomes included PA and glycemic level outcomes, whereas secondary outcomes included BP, BMI, and serum lipid profile. We included only RCTs, without any date restrictions, and excluded studies with inaccessible full texts.

2.2 Information sources

We searched eight bibliographic databases, including PubMed, Embase, Cumulated Index to Nursing and Allied Health Literature, Cochrane, Web of Science, Scopus, Institute of Electrical and Electronics Engineers, and ProQuest Dissertation & Theses, for published and unpublished studies. We searched the Cochrane Central Registry of Controlled Trials and CinicalTrials.gov for potentially relevant trial documents. Existing similar SRs and relevant articles' reference lists were also hand-searched. For studies with inaccessible full texts or missing information, we contacted the authors. Most authors did not respond, leading to insufficient information for an assessment of reporting bias.

2.3 Search strategy

Two concepts, “diabetes” and “wearable devices,” were combined using a three-step search strategy. We identified initial key terms from PubMed and Embase by analyzing the titles, abstracts, and index terms of relevant articles. Concepts were then combined with the appropriate Boolean operators, and database-specific searches were conducted for each database (Table S3). Validated RCT filters were utilized where available. The first reviewer (CR) refined the search strategy after consultation with a senior research librarian, with the final search strategy reviewed by the research team.

2.4 Selection process

The selection process was guided by the PRISMA statement³¹ and the Cochrane Handbook for SRs.³⁰ All results were exported into Covidence software,³² duplicates were removed. Reviewer 1 (WWA) and reviewer 2 (YL) screened the titles and abstracts independently to exclude ineligible studies. We obtained the full texts of all potentially relevant studies. Two independent reviewers then pilot-tested the eligibility criteria with eight articles and independently screened the full texts for inclusion. Disagreements between two reviewers (WWA and YL) that could not be resolved through discussion were resolved through consulting a third reviewer (STL).

2.5 Data extraction

We used a modified data extraction form, developed following the requirements of the Cochrane Handbook for SRs of Intervention,³³ to obtain information on study characteristics, including authors, year of publication, country of study, setting, study design, population, mean age, sample size, interventions, wearable used in the intervention, model, comparators, outcomes, measures, attrition rates, use of intention-to-treat analysis (ITT) or missing data management (MDM), and grant. Intervention-specific items include medium, materials, co-interventions, duration, theoretical basis, reminder, feedback, and follow-up. Piloting of the form and subsequent data extraction were done independently by between two reviewers (CR and AB) with any dispute resolved through discussion, or consultation with reviewer 3 (STL).

2.6 Quality appraisal of included studies

The quality assessment of included studies was independently evaluated by independent reviewers (WWA and YL) to minimize errors and provide different perspectives, using the Cochrane Collaboration's risk of bias tool version 2 (RoB 2).³⁴ RoB 2 was assessed based on the randomization process, deviations from the intended intervention, missing outcome data, measurement of outcome, and selection of the reported result. Each domain was rated “yes,” “probably yes,” “probably no,” “no” or “no information.” The highest risk rating across all five domains determined the overall risk. To maximize reliability, WWA and YL piloted the RoB 2 on three articles to better understand the rating criteria and ensure consistency. We then assessed the domains and resolved any disputes through discussion or consultation with STL.

2.7 Interrater reliability

The study examined the inter-rater reliability of two reviewers, WWA and YL, in the areas of trial selection, data extraction, trial quality rating, and overall evidence certainty using Cohen's Kappa (κ).³⁵ The following ranges of values for κ were interpreted: 0.01–0.20, none to minor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, significant; and 0.81–1.00, practically perfect agreement.³⁵

2.8 Data synthesis

The R software version 4.3.1's package meta was applied to do the meta-analysis, subgroup analysis, and meta-regression studies.³⁶ To get objective estimates, the limited maximum likelihood estimator for random-effect meta-analysis was used.³⁷ The Hartung–Knapp adjustment involved estimating the between-study variance and treating it as if it were known and fixed.³⁸ This modified method justified using quantiles from a t statistic rather than a Z statistic for making inferences about the average effect.³⁸ This method provided actual coverage probabilities, which was more accurate than the conventional method.³⁸ To assess the genuine effects that might be anticipated for 95% of future similar trials, a 95% prediction interval (PI) was calculated.³⁹ There was a statistically significant effect in the future if all 95% of values fall on the same side of the null.³⁹ There was no appreciable impact in the future if all values fall inside both sides of the null.³⁹ The continuous data were presented as mean difference (MD) or standardized MD (SMD) with standard deviation (SD) and 95% confidence intervals (CIs) using the inverse-variance statistical approach.^{40, 41} A t-test with p < 0.05 was used to assess the WT-BPAI's overall impact.³⁶

As most studies had small sample sizes, the Hedges' g was used to measure the effect size.⁴² The effect size was defined as very small (g = 0.01), small (g = 0.2), medium (g = 0.5), large (g = 0.8), and very large (g = 1.2).^{43, 44} Data transformation was done using formulas by Higgin et al. (2022).⁴⁵ Intervention efficacy for each outcome was presented with forest plots. I² statistics was used to find heterogeneity.⁴⁰ I² is the percentage of the observed variance that corresponds to the variance in true effects. The following criteria are used to interpret I² values: 0%–40% small, 30%–60% moderate, 50%–90% substantial, and 75%–100% significant.⁴⁰

2.9 Additional data analysis

This review included subgroup and meta-regression analyses to investigate the possibility of covariates impacting the treatment effect.³⁰ For categorical covariates, subgroup analyses were conducted based on the wearable device type, the involvement of a goal-setting component, and the intervention function. If p was less than 0.1, the subgroup effect was statistically significant.⁴⁶ For continuous covariates, a series of univariate meta-regression analyses were performed.⁴⁷ A random-effects meta-regression model was used to investigate the influence of publication year, sample size, mean age, and duration of intervention. Regression coefficients (β) were the estimated change in the effect of the size units of the covariates on health outcomes.⁴⁸ A significant effect was shown by a p < 0.5.⁴⁸ We carried out a narrative synthesis if the data from selected trials were insufficient for meta-analysis.

2.10 Certainty of evidence and publication biases

The overall quality of the evidence from the included studies was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria.⁴⁹ The GRADE consists of five domains, including risk of biases, heterogeneity, indirectness of evidence, imprecision of results, and risk of publication bias.⁴⁹ Certainty of evidence was graded as “high,” “moderate,” “low,” or “very low”⁵⁰ using GRADEpro online software.⁵¹ Funnel plots and Egger's test were used to test for publication bias when 10 or more studies were included in a forest plot.⁵² With the aid of the funnel plot's asymmetry and an Egger test p-value of less than 0.05, publication bias was determined.⁵²

3.1 Study selection

A flow diagram of PRISMA (Figure 1) describes the search procedure. A search of eight electronic databases using Covidence³² produced 2851 records between the start date and January 16, 2024. By hand searching, no further records could be found. Covidence³² was used to eliminate 1881 duplicates, and YL and WWA screened 1936 entries based on their titles and abstracts. Thirty-four studies were eliminated with reasons (Table S4) after the remaining records were evaluated against the qualifying criteria. Ultimately, this SR included 19 RCTs.^{14, 53-70}

3.2 Study characteristics and description of intervention

Table 1 summarizes the characteristics of the 19 RCTs involving 2547 participants from 14 countries including Australia (n = 1),⁵⁸ Belgium (n = 1),⁵⁷ Brazil (n = 1),⁶⁵ Canada (n = 1),⁷⁰ China (n = 1),⁶¹ India (n = 1),⁶⁸ Japan (n = 2),^{63, 64} Mexico (n = 1),⁵³ Netherlands (n = 1),⁶⁰ Norway (n = 1),⁵⁶ Oman (n = 1),⁵³ Turkey (n = 1),⁶⁹ the United Kingdom (n = 1),⁵⁴ and the United States (n = 5).^{14, 55, 62, 66, 67} The trials were conducted from 2004⁷⁰ to 2022,⁶³ with a mean age range of 44.2⁵³ to 68.7⁶⁵ years. Sample sizes ranged from 30^{55, 59} to 593.⁵⁴ Of the 19 RCTs, five trials^{53, 60, 63, 69, 70} adopted a theoretical basis for developing their interventions. The types of wearables used included pedometers (n = 11), accelerometers (n = 1),⁶³ smartwatches (n = 3),^{60, 62, 69} a combination of pedometers and accelerometers (n = 1),⁵⁹ activity trackers (n = 2),^{64, 66} and a heart rate monitor (n = 1).⁶¹ The intervention duration ranged from 5 weeks⁵⁹ to 1 year.^{53, 54, 66, 67} All interventions included a monitoring function, and three of them included a reminder function.^{62, 66, 69} Intervention details are reported in Table S5.

3.3 Risk of bias in studies

The risk of bias summary according to the RoB 2 criterion is displayed in Figure S1. A Cohen's kappa score of 0.89, which indicates nearly perfect agreement, was obtained from two reviewers.³⁵ Twelve RCTs used per-protocol analysis, and seven ITT studies were included in the total of 19 RCTs. Seventy trials (89.5%) did not provide information on allocation concealment. As all participants knew their allocated intervention during the trial, there were some concerns regarding deviations from the intended intervention. A total of 10 trials (52.6%) did not perform to register in clinical trial registries or publish a protocol. As a result, there were reservations regarding the overall risk of bias in all the included trials. The attrition rates ranged from 0%⁵⁵ to 36.70%.⁷⁰ An attrition rate of less than 5% was seen in just three studies. Eight of them used either ITT or MDM.

3.4 Glycemic outcomes

Seventeen trials^{14, 53-56, 58, 60-70} in 20 arms involving 2407 participants reported glycated hemoglobin (HbA1c) levels and six trials^{55, 56, 63, 65, 68, 70} involving 224 participants reported fasting blood glucose (FBG) as the glycemic outcome. The HbA1c levels measure glycemia over 90 days,⁷¹ and FBG readings measure glycemia at the point of care.⁷² Figure 2 showed that interventions did not lead to a significant change in HbA1c (t = 0.49, p = 0.63), where HbA1c decreased by 0.04% (95% CI = −0.21 to 0.13) with a very small effect size (g = −0.05) (Figure S2) when compared with the comparator. Similarly, interventions did not lead to a significant change in FBG (t = −0.77, p = 0.48), with a small effect size (g = −0.34, 95% CI = −1.50 to 0.81) (Figure S3). Moderate and substantial heterogeneities were detected (I² = 54%–86%).

3.5 Physical activity

We measured the outcomes of PA in this review in step count (steps per day) and moderate-to-vigorous physical exercise (MVPA) using pedometers or accelerometers. (Table 2). We found 11 arms of 9 RCTs^{53, 55, 56, 62, 63, 65-67, 70} with 1076 participants for steps per day (Figure 3). Meta-analysis revealed a significant (t = 3.11, p < 0.05) increase of 1583 steps/day (95% CI = 477–2719), with medium effect size (g = 0.54) favoring intervention (Figure S4). The PI ranged from −0.56 to 1.64, with values found on either side of the null, suggesting that the intervention did not substantially decrease the number of steps taken daily when compared to a comparator in comparable future research. However, substantial heterogeneity was detected (I² = 83%). Only two trials^{54, 57} involving 432 participants were found for MVPA. Meta-analysis reveals that interventions resulted in a nonsignificant increase (t = 3.10, p = 0.20) in MVPA by 7.69 min per day (95% CI = −23.83 to 39.20), with a small effect size (g = 0.34). (Figure S5) Unimportant heterogeneity was observed (I² = 29%).

3.6 BP, BMI, and serum lipid profile

Other outcomes included BP involving 1158 participants from 11 RCTs,^{14, 53-56, 58, 63, 65, 67, 68, 70} BMI involving 900 individuals from 9 RCTs,^{53-55, 58, 60, 61, 63, 65, 68} high-density lipoprotein (HDL) involving 740 participants from 7 RCTs,^{53-56, 63, 65, 70} and low-density lipoprotein (LDL) involving 993 participants from seven arms of five RCTs.^{53-55, 66, 70} For BP, meta-analyses showed that intervention resulted in a significant decrease (t = −3.28, p < 0.01) in systolic BP (SBP) by 2.46 mm Hg (95% CI = −4.13 to −0.79) but not a significant decrease (t = −0.75, p = 0.470) in diastolic BP (DBP) by 1.00 mm Hg (95% CI = −3.95 to 1.96) when compared with the comparator (Figures S6 and S7). The effect sizes for SBP and DBP were very small effect size (g = −0.15 to −0.13). The PI (95%) was discovered on both sides of the null, suggesting that the intervention did not substantially decrease the SBP when compared with a comparator in similar studies in the future. No heterogeneity (I² = 0%) was observed for SBP, whereas substantial heterogeneity (I² = 75%) was detected for DBP. There were no significant differences between the intervention and control groups on the MDs of the BMI, HDL, or LDL (Figures S8–S10). However, Figure S8 shows that the intervention significantly reduced the effect size of BMI (t = −2.83, p = 0.02) with a very small effect size (g = −0.18) when compared to the comparator. No heterogeneities (I² = 0%–9%) were detected.

3.7 Subgroup and meta-regression analyses

We ran a series of subgroup analyses and meta-regression analyses for dichotomous and continuous covariates, respectively, where the results came from ten or more RCTs. Table 3 illustrates the results of subgroup analyses of wearable types, goal setting, and function on HbA1c, steps per day, SBP, and DBP (Figures S11–S20). The intervention for the sole monitoring function showed a significantly larger effect size (p = 0.02) on HbA1c reduction (g = −0.14, 95% CI: −0.36 to 0.08) compared with the intervention for the monitoring and reminder function (g = 0.15, 95% CI: −0.06 to 0.35). The effect sizes on increasing the number of steps taken each day were significantly larger (p < 0.01) when the intervention was only for monitoring function (g = 0.80, 95% CI: 0.28–1.31) compared with when it was both for monitoring and reminder function (g = 0.15, 95% CI: −0.12 to 0.41). There were no significant subgroup differences between the wearable device used, and involvement in goal setting on steps per day, HbA1c, SBP, and DBP.

Table 4 shows the results of meta-regression analyses to explore whether the year of publication, sample size, mean age, and duration of intervention influence intervention effects. Results revealed that the intervention duration and participants' mean age were significant covariates for the steps per day (β = −0.08, p = 0.03) and SBP (β = 0.02, p = 0.03), respectively. The findings revealed that the shorter duration intervention significantly increased daily steps in comparison to the longer duration intervention. The result also indicated that younger adults significantly decreased their SBP after interventions compared to older adults. However, the year of publication and sample size did not influence steps per day, HbA1c, SBP, and DBP.

3.8 Narrative synthesis

Among 19 RCTs, one study⁵⁹ did not provide sufficient data for steps per day, BP, glucose level, and serum lipid profile after intervention in the report. Our team emailed the authors, requesting supplementary data, but there was no response after three reminders. Hence, narrative analysis was done. Participants in the WT-BPAI group achieved a significantly higher level of PA after 1 and 5 weeks than those in the control group. There were no changes in BP, BMI, glycemic level, and serum lipid profile. These patterns of findings resembled the results of the meta-analysis.

3.9 Certainty of evidence and publication bias

Following the GRADE criteria, all outcomes had very low levels of evidential confidence (Table S6). The randomization process, deviations from the intended intervention, and missing outcome data all contributed to the downgraded RoB. High heterogeneity of outcomes downgraded inconsistency. The different types of wearables led to a downgrade in the domain of indirectness. The small sample size and wide confidence intervals have induced a downgrade in the domain of imprecision. Publication bias was assessed for steps per hour, HbA1c, SBP, DBP, and BMI, as more than 10 studies were included in their meta-analyses Based on Egger's test and visual inspection of funnel plots, no publication bias was detected for step count (p = 0.20), HbA1c (p = 0.93), SBP (p = 0.59), and DBP (p = 0.48) (Figures S21–S24).

4.1 Summary of findings

This review included a total of 19 trials involving 2574 adults with T2DM across 14 countries. Meta-analyses identified that the WT-BPAI significantly increased steps per day and SBP when compared with the comparator group. Subgroup analyses revealed that interventions with monitoring functions had a significantly greater effect on lowering HbA1c and increasing the number of steps taken each day than interventions with monitoring and reminder functions. Random-effects meta-regression revealed that the shorter the length of the intervention and the younger the person were found to be significant covariates for increasing the number of steps taken each day and lowering SBP, respectively. Based on RoB 2, there were some concerns in over half of the studies because of the randomization process, deviations from the intended intervention, and missing outcome data. Publication bias was not detected. According to the GRADE criteria, there was very low overall certainty in the evidence.

4.2 Glycemic outcomes

Our meta-analyses showed that the WT-BPAI reduced HbA1c and FBG when compared with the comparators, but the differences were not statistically significant. This is like findings from similar SRs^16-18 that reported no significant effect on glycemic levels. Intervention may help to increase PA by increasing muscular contractions to promote glucose uptake,⁷³ resulting in decreased glycemic levels.⁷⁴ However, solely PA may not significantly reduce glycemic levels. Notably, one study has attrition rates as high as 36.70%.⁷⁰ The WT-BPAI may have become less novel and challenging over time, making it harder for users to track their PA and achieve their health objectives. As a result, it may have become more important for users to stay motivated to engage in PA beyond simply understanding their PA levels.¹⁰ Hence, the lack of statistical significance is probably due to the individuals' struggles to maintain consistent PA as the intervention advanced, rather than the ineffectiveness of interventions in lowering glycemic levels. Furthermore, adults with T2DM were administered different medications to control glycemic levels according to their situations.¹⁶ That could explain the lack of effect on HbA1c and FBG.

4.3 Physical activity

Based on meta-analyses, WT-BPAI significantly increased PA level with a medium effect size for step count, where step count increased by around 1583 steps/day. This is similar to findings by Baskerville et al.,¹⁶ Franssen et al.,¹⁷ Hodkinson et al.,¹⁹ Kirk et al.,²⁰ and Wang et al.²⁴ The possible explanation is that people with diabetes were aware of PA levels through immediate quantitative feedback from wearables,^{75, 76} where this awareness promotes self-evaluation and self-reinforcement of the behavior change.⁷⁷ It may explain the increased steps per day after the intervention. However, we cannot find significant differences between intervention and comparator on MVPA. The nonsignificant results may be due to the limited sample size of the selected two trials in a meta-analysis. A small sample size would have insufficient statistical power to differentiate the differences between the two groups.⁷⁸ Therefore, we should interpret MVPA findings cautiously. In line with a previous review,¹⁶ the effect sizes from the accelerometer and pedometer were comparable in our subgroup analyses. There was no significant difference in all outcomes between wearable types. Given that pedometers are a cost-effective means to monitor the steps per day, future studies can consider pedometers as wearable devices for monitoring physical activities. Our meta-regression showed that the short duration had a better effect on increasing steps per day when compared with a long duration of intervention. The findings align with a prior SR²¹ that the wearable device-based intervention with a shorter duration had a better effect. This might account for a drop in motivation to participate and adherence over an extended period.²¹

4.4 Blood pressure

The intervention had statistical significance in reducing SBP, but not DBP. These results were like a recent SR¹⁷ on a significantly decreased SBP after wearable activity tracker-based interventions. These findings can be explained by PA may maintain healthy BP through the body's adaptive mechanisms involving blood circulation and metabolic changes.⁷⁹ Given that a difference in statistical significance was observed between SBP and DBP reduction, the intervention can reduce SBP more effectively than DBP.^{80, 81} There is a correlation between a decrease in SBP and a decrease in stroke mortality.⁸² Hence, the WT-BPAI may lower the risk of cardiovascular disease.¹⁷

4.5 Body mass index

Our meta-analyses revealed that adults in the WT-BPAI group did not decrease the MD in BMI, but significantly reduced the effect size of BMI when compared with the comparator group. Considering that a high level of sedentary behavior was found in adults with T2DM,⁸³ WT-BPAI might act as a motivational tool by increasing the motivation and encouragement of physical activities and translating the goals into practice.⁸⁴ Physical activities could help adults in the WT-BPAI increase their energy expenditures.¹⁶ Hence, the WT-BPAI has an indirect effect by increasing weight loss, which could lower the effect size of BMI.

4.6 Serum lipid profile

Our SR revealed that there are no appreciable variations in LDL or HDL levels between the WT-BPAI and comparator and groups. The lipid profile showed no significant differences, which is like a previous SR¹⁶ of the impact of accelerometer and pedometer use in people with T2DM. Notably, another SR¹⁷ found that wearable activity tracker-based interventions showed significantly decreased LDL among individuals with any chronic disease. The different populations may be the cause of these discrepancies. Comparable to glycemic results, varying individuals may be taking various drugs to regulate their lipid profiles, which might partially account for the absence of impacts on HDL and LDL.¹⁶ It is necessary to do more research on the WT-BPAI across different types of populations.

4.7 Function and duration of intervention

Our subgroup analysis observed that the WT-BPAI with just the monitoring function had a greater effect on lowering HbA1c and increasing daily steps than the WT-BPAI with both the monitoring and reminder functions. The reminder function might not promote engagement in the intervention because participants had difficulties understanding the messages and notifications, which caused them to ignore the reminders.⁸⁵ However, only three trials^{62, 66, 69} included both monitoring and reminder features; further study was required to validate the outcome. We observed significantly larger effect sizes when the intervention was in younger adults on SBP reduction than older adults in subgroup analyses. One possibility is that young adults regularly engage in PA to improve BP because they are familiar with wearable functions.^{86, 87} Another possible explanation for this finding is that older adults may not adhere to the use of wearable technology.⁸⁸ As a result, the WT-BPAI has the potential to lower SBP more effectively in young adults than in old adults.

4.8 Strengths and limitations

There are various advantages to our review. This review was able to minimize bias and overcome the methodological limitations of past comparable reviews by thorough, rigorous, and systematic searches and analyses.²⁷ The present review identified a greater number of RCTs and no evidence of publication bias. Subgroup and meta-regression analyses were employed to investigate the potential covariates affecting PA levels, glycemic levels, BP, BMI, and serum lipid profile in this study. We used objective outcomes in our review to prevent social desirability bias. To avoid paradoxical effects, a random-effects meta-analysis was conducted with the Hartung–Knapp adjustment.³⁸ Furthermore, the 95% PI of the meta-analyses was supplied to predict the outcome in future situations.³⁹

However, there are some limitations. First, there were substantial heterogeneities created by many inter-study differences, such as the variety of co-interventions, length of intervention, staff participation, and comparators, which might have an impact on the actual outcomes in different trials. Hence, the total effect may have been incorrectly calculated due to the significant heterogeneity of several outcomes. Second, the fact that only English-language research was included might restrict how far the results can be applied. Third, there was a lack of information from the selected trials about how active participants were and mortality severity at baseline, which may have affected their PA levels. Fourth, there is uncertainty regarding the intervention's durability because all outcomes were utilized after the intervention. Finally, the rating of the overall evidence as very low quality raises the possibility of eliminating the internal validity of the findings.

4.9 Implications to clinical practice

Because the overall evidence of all outcomes was very low in this review, we provided limited information for clinical decision-making. However, the WT-BPAI can consider supplementing interventions to increase daily steps and reduce SBP in addition to routine diabetic care. Based on our subgroup and meta-regression results, the findings can provide direction for future WT-BPAI design. The short duration of the WT-BPAI is preferable, and the function of the intervention is solely for monitoring, which could be an effective and workable way to boost PA. The target population for the WT-BPAI could benefit young adults with T2DM by decreasing their SBP. This would transform DM self-management in the future by empowering patients to take better care of and control their health via wearable technology.

4.10 Future research

First, future trials could investigate the use of a short-duration WT-BPAI with only a monitoring function for young adults with T2DM. This may increase the efficacy of therapies in encouraging adherence and allow for a long-term evaluation of their impact on glucose levels in subsequent reviews. Second, trials should consider making the availability of their protocols to prevent unnecessary ambiguity in reporting bias. Third, the very low overall quality of evidence in our review reminds us that future trials should provide information for allocation concealment and design a way for blinding participants. Lastly, as PA intervention mainly affects lean mass rather than fat mass,⁸⁹ both lean and fat mass must be included in the outcomes. Nevertheless, only three trials^{14, 55, 61} reported body fat (%) and two trials^{63, 65} reported body fat mass (kg) in this review. A study showed that increased glycemic levels were hindered by low skeletal muscle mass and poor skeletal muscle tissue quality.⁹⁰ Given that skeletal muscular mass plays a role in T2DM,⁹⁰ further investigations are needed.