Introduction

Pre-diabetes mellitus is defined as a status between normal individuals and diabetes and involves impaired glucose tolerance (IGT) or impaired fasting glucose (IFG).1, 2, 3, 4 Individuals with pre-diabetes mellitus typically develop type 2 diabetes within 10 years as well as a high risk for cardiovascular disease;1 and approximately 5–10% of individuals with pre-diabetes progress to diabetes each year. Prevalence of pre-diabetes is increasing worldwide, and experts estimate that >470 million people will become pre-diabetic by 2030.5 In 2010, the American Centers for Disease Control and Prevention (CDC) reported that approximately one in three US adults aged ⩾20 years (an estimated 79 million persons) had pre-diabetes mellitus.6 In 2013, Xu et al.7 reported that 493.4 million Chinese adults were pre-diabetic. Interventions during the pre-diabetes stage are crucial to avoid development of full-blown diabetes and other adverse cardio-metabolic risk factors.8

Many studies have examined medication efficacy or lifestyle interventions among type 2 diabetes patients,9,10,11although few have reported on physical activity or exercise-diet interventions among a pre-diabetic population.12, 13 The effect of lifestyle interventions among a pre-diabetic population remains unclear, particularly for exercise-only interventions. To address this research gap, we conducted a systematic review and a meta-analysis of published randomized controlled trials (RCTs) and non-RCTs to examine the efficacy of exercise-only and exercise-diet interventions in the prevention of type 2 diabetes.

Materials and methods

Search strategy

A literature search was conducted for English articles in PubMed from January 1980 to December 2013, EMBASE from January 1988 to December 2013 and Cochrane databases from January 1995 to December 2013. Articles pertaining to dietary or exercise interventions for pre-diabetic individuals were identified.

Key search words included the following: pre-diabetes/IGT/IFT, intervention (lifestyle, exercise, sport and physical activity) and RCT/non-RCT (see Supplementary Appendix 1). Data extraction was performed independently by two investigators (J Wu and L Zheng) using a predefined form (Table 1).

Table 1 Characteristic of included studies
Full size table

Inclusion and exclusion criteria

Inclusion criteria were as follows: (1) the study population was diagnosed as pre-diabetic and over 40-year old; (2) the study design included RCT or non-RCT; and (3) interventions included exercise-only or exercise-diet. Exclusion criteria were as follows: (1) those who had a history of type2 diabetes mellitus; (2) those who used diabetes medication as part of the intervention.

Subjects

Pre-diabetic patients were defined as IGT or IFG using one of the established criteria from the WHO14 or the American Diabetes Association.15 Studies that defined IGT or IFG using other criteria were included for review if the mean value of the individuals’ plasma glucose levels fell within the range of IGT or IFG as defined by the WHO or American Diabetes Association criteria.14, 15 We also included studies whose participants had a fasting blood glucose of <7.8 mmol/l and their 2-h plasma glucose was between 7.8 and 12.5 mmol/l; or their fasting blood glucose was between 5.3 and 6.9 mmol/l.

Assessment scale

Two researchers (J Wu and L Zheng) independently conducted quality assessment of all the included articles using the Cochrane Collaboration tool.16, 17 If there were discrepancies, a third researcher (GW) took part in the discussion to make the final assessment decision. The Cochrane Collaboration tool assessed the risk of bias in the following domains: selection bias, performance bias, detection bias, attrition bias, reporting bias and other bias. Each domain was classified as ‘low risk,’’unclear’ or ‘high risk.’

Statistical methods

The primary outcome of this meta-analysis was mean change in fasting plasma glucose (FPG) levels. Calculation of mean change in FPG between baseline and follow-up visits was conducted. Heterogeneity between studies was analyzed by means of I2=[(Q–df)/Q] × 100%, where Q is the χ2 heterogeneity statistic and df is the degrees of freedom.18, 19 Meta-regression20 was undertaken to explore the impact of risk of bias for the included studies, such as intervention style and intervention period. A random effects meta-analysis was used to estimate the overall mean changes. I2 >0.5 indicated substantial heterogeneity (statistical heterogeneity).21 A forest plot was conducted to explore the relationship between interventions (exercise-only or exercise-diet) and mean change of FPG in the pre-diabetes population. Publication bias was examined using a funnel plot (Begg’s test).22 All statistical analyses were conducted using the statistical software package Stata12.0 (Stata Corp., College Station, TX, USA).23

Results

Search results

Overall, 206 studies were indexed in the primary screen and search, 12 of which were duplicates. On the basis of our criteria, 13 studies were eligible for a full-text review. Exclusion of 193 articles is shown in Figure 1. One article12 was excluded because the s.d. value of FPG was missing. A total of 12 studies 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35 were included in the present meta-analysis.

Figure 1
figure 1

Flow diagram of search strategy.

Full size image

Baseline characteristics of participants

Eleven of the 12 studies were RCT. The number of participants in each study ranged from 8 to 256. For the intervention and control groups, there were 2045 and 1976 participants, respectively. Three studies reported the mean change of FPG for exercise-only. The remaining nine studies reported the pooled effect of exercise-dietary interventions (Table 1). Duration of follow-up ranged from 12 weeks to 6 years for the intervention arms.

Quality assessment

Using the Cochrane Collaboration Tool, detailed information of quality assessment is shown in Table 2. Eight27, 28, 29, 31, 32, 33, 34, 35 of the included studies have more than five domains classified as a low risk of bias, whereas four studies24, 25, 26, 30 written by Katula,24 Oldroyd,25 Lu HY26 and Steven30 were deemed unclear in the ‘detection bias’ and other bias domain. Allocation concealment was not mentioned in the studies with the exception of Oldroyd’s article.25 In fact, all of the included studies were open-label trials, whose performance bias was defined as ‘high risk’. Several studies 22, 30, 33 indicated that FPG outcome assessment was blinded to the laboratory technicians.

Table 2 Cochrane Collaboration’s tool for assessing risk of bias
Full size table

A meta-regression was carried out to examine reasons for heterogeneity. Regression results revealed that intervention style was the main reason for heterogeneity. Heterogeneity remained present despite the creation of two subgroups for intervention style, and overall I2 was 94.5% (P<0.05). Therefore, random effects models were used to calculate the total effect and subgroup effect. Exercise-diet interventions showed a significant effect on decreasing FPG levels (weighted mean difference (WMD)=−0.22 mmol/l, 95% confidence interval (CI): −0.25, −0.18, Z=12.06, P<0.05). However, the exercise-only intervention did not demonstrate a significant effect (WMD=−0.09 mmol/l, 95% CI: −0.18, 0.00, Z=1.91P>0.05). The overall WMD and its 95% CI were −0.19 mmol/l and (−0.22, −0.15) (Figure 2).

Figure 2
figure 2

Forest plot according to intervention style (group 1: exercise-diet interventions; group 2: exercise-only interventions).

Full size image

According to four different intervention periods, the shortest period intervention (<1 year) did not display a significant effect for glucose control (WMD=−0.12 mmol/l, 95% CI: −0.29, 0.05, Z=1.35, P>0.05). There was a significant effect (WMD=−0.20 mmol/l, 95% CI: −0.25, −0.14, Z=7.19, P<0.05) in the 1-year subgroup. The longer the intervention period, the higher the subtotal effect (Figure 3). The pooled effect in the ⩾2-year subgroup was the highest with WMD (95% CI)=−0.24 mmol/l (−0.43, −0.05).

Figure 3
figure 3

Forest plot according to intervention period (group 1: <1 year; group 2: 1 year; group 3: 1–2 years; group 4: >2 years).

Full size image

As for different regions, significant heterogeneity existed among studies conducted in America (I2=98.30%) and China (I2=94.50%) but not in Europe (I2=47.00%). The subtotal effect was varied in different regions. The studies conducted among the European population displayed a higher subtotal effect, and its WMD was −0.22 mmol/l and its 95% CI was (−0.27, −0.17) with Z=8.01 and P<0.05. WMD (95% CI) in the American population was −0.17 mmol/l (−0.21, −0.12).

When participants were divided into two age groups (40–55 years and ⩾55 years), no significant subgroup pooled effects were found in the younger subgroup (WMD=−0.27 mmol/l, 95% CI: −0.60, 0.05, Z=1.65, P>0.05). However, in the elderly subgroup, there was a significant effect (WMD=−0.19 mmol/l, 95% CI: −0.22, −0.15, Z=10.31, P<0.05). Among the two groups, the subtotal effects were −0.27(−0.60, 0.05) and −0.19 (−0.22, −0.15), respectively. The Begg funnel plot showed an asymmetrical distribution and a Kendall score of 7 (P=0.865), indicating no statistical evidence of publication bias in the present meta-analysis (Figure 4).

Figure 4
figure 4

Funnel plot for publication bias.

Full size image

Discussion

The current meta-analysis calculated the pooled effect of exercise-only and exercise-diet interventions on FPG control. Meta-regression analysis demonstrated significant heterogeneity among the included studies. However, the clinical heterogeneity among articles was available as the effect of each study had the same direction. Subgroup analysis and random effects models were used to examine the pooled effect based on different intervention style, period, region and age group.

Results reported from a Da Qing study displayed that the exercise-only intervention may decrease the risk of diabetes.28 However, the findings of our current analysis showed that the change in FPG levels was not significant in the exercise-only interventions. Exercise-diet interventions had more significant subtotal effects in decreasing FPG. One possible explanation for these findings could be due to the status of weight loss, which had a strong relationship with the FPG level.36, 37, 38 It was clear that exercise-only had a weaker effect compared with the exercise-diet interventions in decreasing body weight.39 Furthermore, there was limited data available for exercise-only interventions, thereby decreasing reliability. Among the included studies, the effect of exercise-only intervention was inconsistent. In addition, some of the studies reported on exercise time, whereas others reported exercise intensity, causing difficultly in calculating a standard value of exercise quantity. Therefore, more studies focusing on exercise-only interventions are needed to explore the relationship of FPG changes in a pre-diabetic population.

To our knowledge, this is the first meta-analysis assessing intervention length, as it relates to change in FPG levels. According to findings reported by Lindstrom et al.,40 the lifestyle intervention (exercise-diet) group had a significant change in FPG levels during the first year. There was almost no significant effect after follow-up from 3 to 4 years. However, our present meta-analysis findings are to the contrary, indicating that the longer the intervention the greater the intervention effect (including exercise-only and exercise-diet), with no significant effect when the intervention period was <1 year.

The subgroup analysis indicated that the subtotal effect was significant in the 55-year age group but not among the 40- to 55-year age group. These results showed that the intervention effect (exercise-only and exercise-diet intervention) may be stronger in the elderly population. Our findings indicated that age was also correlated with glucose metabolism and were consistent with several previous studies conducted in human 41, 42 and animal models.43 The current results also showed that the intervention group had a better outcome among the European and the American population than in the Asian population. Research by Danaei et al.44 found that glucose metabolism varied by nations and regions.

The strength of our investigation was that the majority of included studies were RCTs, with the exception of one article written by Steven et al.30 In addition, large numbers of participants and varying geographic locations contributed to a diverse and a comprehensive set of data. The present analysis also had several limitations as studies included in this meta-analysis were limited to articles published in English. In addition, 12 of the studies were heterogeneous, and despite subgroup analysis most of the statistical heterogeneities remained. Therefore, further rigorous studies are needed to confirm this finding. Furthermore, the intervention period range varied from 12 weeks to 6 years, which may add to the heterogeneity in our analysis. Finally, the quantitative information of exercise-diet was not assessed between studies according to the same standard, which could potentially modify the association between intervention effect and the FPG level.

Both exercise-only and exercise-diet interventions have displayed effects on decreasing FPG, with better results in the later group. The current data also showed that the pooled effect was significant in longer intervention periods (⩾1 year), the elderly population (⩾55 years age) and Europeans/Americans. More studies are needed to assess the efficacy of exercise-only interventions among a pre-diabetic population.