Description of the condition
Diabetes mellitus is a metabolic disorder resulting from a defect in insulin secretion, insulin action, or both. A consequence of this is chronic hyperglycaemia (that is, elevated levels of plasma glucose) with disturbances of carbohydrate, fat and protein metabolism. Long-term complications of diabetes mellitus include retinopathy, nephropathy and neuropathy. The risk of cardiovascular disease is increased. For a detailed overview of diabetes mellitus, please see under 'Additional information' in the information on the Metabolic and Endocrine Disorders Group in The Cochrane Library (see 'About', 'Cochrane Review Groups (CRGs)'). For an explanation of methodological terms, see the main glossary in The Cochrane Library.
Type 2 diabetes mellitus is the most common type of diabetes. In Western countries the disease affects up to 7% of the population (Harris 1998; WHO 1994). Its incidence is associated with the 'Westernised lifestyle', mainly in terms of dietary habits and physical activity (ADA 1996). Special attention is to be paid to the increasing incidence of the disease in newly industrialized and developing countries in which lifestyle changes have occurred (WHO 1994). Also, an increasing incidence in adolescents, especially in the US, has been shown particularly in the non-white population (Rosenbloom 1999). Obesity and decreased physical activity are associated with the development of glucose intolerance and type 2 diabetes (Eriksson 1991). The events leading to the development of the disease are mainly an abnormal insulin secretion and insulin resistance. Thus, an insufficient insulin secretion combined with a reduced capacity of peripheral tissues to utilise glucose and an increased glucose production by the liver leads to the progressive development of hyperglycaemia (Lillioja 1993). However, the exact sequence of events leading to the development of the disorder is still to be fully characterised. As the progression from normoglycaemia to overt hyperglycaemia is slow, a significant proportion of individuals remain undiagnosed during the initial period of the disease. A majority of individuals later developing the disease present with obesity, which further contributes to an increased insulin resistance (Beck-Nielsen 2000). Other known factors are also associated with the appearance of the disease, for example increasing age and lack of physical activity. As there is a strong familial predisposition, first-degree relatives of known type 2 diabetic patients are at increased risk of developing the disorder (Rewers 1995).
Impaired glucose tolerance (IGT) and impaired fasting glucose (IFG) are generally recognised as an expression of abnormal glucose metabolism regulation. These two conditions can be considered as an intermediate stage between normal glucose tolerance and diabetes mellitus. They should be regarded as risk factors for the development of diabetes rather than a disease as they do not produce any symptoms. People with IGT or IFG are at increased risk of developing type 2 diabetes and cardiovascular disease (even before the onset of diabetes). They are also associations to other risk factors for diabetes such as obesity, unfavourable dietary habits and lack of exercise. The term IGT (sometimes referred as ‘prediabetes’) was introduced in 1979 (NDDG 1979). The term IFG has been introduced much later (ADA 1997). IGT and IFG represent different pathophysiological mechanisms: IGT is seen as a characteristic of peripheral insulin resistance and IFG is seen as an expression of raised hepatic glucose output and a defect in early insulin secretion. Currently, the criteria for IGT and IFG are as follows (plasma venous glucose concentrations): IGT - fasting blood glucose less than 7.0 mmol/L and two-hour post-load blood glucose 7.8 to 11.0 mmol/L; and IFG - fasting blood glucose 6.1 to 6.9 mmol/L (two-hour post-load blood glucose less than 7.8 mmol/L, if measured) (ADA 1999; WHO 1999). However, in 2003 the American Diabetes Association recommended to change these latter criteria to 5.6 to 6.9 mmol/L (ADA 2003).
Description of the intervention
It has been shown that weight reduction and an increase in daily energy expenditure decrease insulin resistance and increase glucose tolerance (Ross 2000). Indeed, advice on diet and exercise are an important part of the treatment of type 2 diabetes. Nutritional advice usually consists of caloric restriction if the patient is overweight, low total fat content (especially saturated fat) and high (predominantly unrefined) carbohydrate content. It may be hypothesised that interventions aimed at preventing the development of obesity and at increasing physical activity will lower the incidence of type 2 diabetes in those individuals at high risk. Further, the adoption of a healthy lifestyle, that is avoiding being overweight and exercising regularly, may provide a protective effect against other elements of the metabolic syndrome usually associated with impaired glucose tolerance and type 2 diabetes mellitus, such as hypertension and hyperlipidaemia, which may subsequently increase morbidity and mortality from cardiovascular disease (Stamler 1999; Stampfer 2000).
How the intervention might work
Diet and exercise as preventive measures
Primary prevention covers the activities aimed at preventing diabetes from occurring in susceptible individuals or in the general population. This can be done through modification of environmental, behavioural determinants (for example dietary habits) or both, or by any specific intervention (for example pharmacological). Diet and exercise interventions are sometimes termed 'lifestyle interventions'.
There are prospective cohort studies that have shown that increased physical activity, independent of other risk factors, has a protective effect against the development of type 2 diabetes (Helmrich 1991; Manson 1992). These epidemiological prospective studies demonstrated that various levels of regular physical activity one to several times a week were associated with a decreased incidence of the disease at long-term follow-up (14 years and five years respectively), in both men and women of different age groups (Helmrich 1991; Manson 1992).
Pan et al. concluded that diet or exercise or both interventions produced a 31% to 46% reduction in the incidence of diabetes during a period of six years in those with impaired glucose tolerance (Da Qing 1997). After the Da Qing study (Da Qing 1997), some studies based on randomised controlled trials for high-risk persons revealed the potential of lifestyle modifications for preventing type 2 diabetes: The Diabetes Prevention Program (DPP 2002) was a multicenter randomised controlled trial in the United States with the three intervention arms lifestyle intervention (diet and exercise), metformin and placebo. This study reported that the lifestyle intervention was more effective than metformin in reducing the incidence of diabetes in persons at high risk; whereas in Finland, the Diabetes Prevention Study found that an intensive lifestyle intervention reduced the diabetes risk compared with a general diet and exercise advice (DPS 2003). A recent meta-analysis of randomised controlled trials showed that a lifestyle intervention reduced the 2-h plasma glucose by 0.84 mmol/L (95% confidence interval (CI) 0.39 to 1.29), and the one year incidence of diabetes was reduced by approximately 50% (relative risk (RR) 0.55, 95% CI 0.44 to 0.69) compared with the control group (Yamaoka 2005). However, this meta-analysis published in 2005 considered only papers published in English in the two databases MEDLINE and ERIC. There is also a Cochrane review (Norris 2005) that addresses the issue of lifestyle changes in people with prediabetes (IGT or IFG) on weight that concluded that these interventions produce a positive effect in terms of weight loss. Recently, another systematic review and meta-analysis showed a positive effect of lifestyle interventions in terms of type 2 diabetes prevention in people with IGT (Gillies 2007). Another upcoming Cochrane review will address the question on the effect of diet only interventions in the prevention of type 2 diabetes (Moore 2005). Therefore, it is advised that the current review is read alongside this latter review.
Adverse effects of the intervention
Exercise or diet interventions are not generally considered to be associated with any serious adverse event. However, physical activity may cause traumatic injuries of variable severity depending on the type and intensity of exercise. Additionally, exercising may produce adverse effects on the cardiovascular system in those people with insufficient training or unfavourable cardiovascular fitness (even cardiovascular events and death may potentially occur while exercising). Also, the implementation of dietary measures may produce several deficiencies in the nutritional status if restrictive low-calorie diets are used. Further, dieting may produce a derangement in the quality of life of persons under this treatment. Unfortunately, very few information on these issues is available from randomised controlled trials.
Why it is important to do this review
As mentioned above, type 2 diabetes mellitus affects an important proportion of the population in most countries. Additionally, an increasing incidence of the disease is already seen in both in industrialized and developing countries. Therefore, type 2 diabetes mellitus is an important health care issue worldwide.
Although other reviews are available (Gillies 2007; Norris 2005; Yamaoka 2005;), the scope of the current review is a wider one and covers not only those people with IFG or IGT but interventions on all other type 2 diabetes at-risk populations. Also, this review offers un up-to-date literature search and provides detailed description of the identified studies.
To assess the effects of exercise or exercise and diet for preventing type 2 diabetes mellitus.
Criteria for considering studies for this review
Types of studies
Randomized controlled clinical trials of interventions that followed-up participants for at least six months.
Types of participants
Participants of any age, sex or ethnicity belonging to any of the major risk groups for the development of type 2 diabetes (ADA 2004b):
- impaired glucose tolerance according to the World Health Organisation criteria (WHO 1999);
- impaired fasting glucose according to the American Diabetes Association criteria (ADA 2004);
- previous gestational diabetes;
- hypertension equal to or greater than 140/90 mmHg;
- family history of type 2 diabetes in first degree relatives;
- obesity (that is a body mass index (BMI) equal or greater than 30 kg/m
- dyslipidaemia (that is HDL-cholesterol equal or less than 35 mg/dl, triglycerides equal or more than 250 mg/dl, or both);
- high risk ethnic groups (for example African-Americans, Hispanic-Americans, native Americans, Asian-Americans, Pacific Islanders).
Types of interventions
- exercise versus standard recommendations or no intervention;
- exercise and diet versus standard recommendations or no intervention;
- exercise versus diet.
Trials where the intervention or control group comprised the administration of any pharmacological agent were excluded.
Types of outcome measures
- development of type 2 diabetes mellitus (incidence);
- diabetes and cardiovascular related morbidity (for example angina pectoris, myocardial infarction, stroke, peripheral vascular disease, neuropathy, retinopathy, nephropathy, erectile dysfunction, amputation).
To be consistent with changes in the classification and diagnostic criteria of type 2 diabetes mellitus through the years, the diagnosis should have been established using the standard criteria valid at the time of the beginning of the trial (for example ADA 1997; ADA 1999; ADA 1999; WHO 1980; WHO 1985; WHO 1998). Ideally, diagnostic criteria should have been described in the publication. When it was necessary, authors' definition of diabetes mellitus was used. Diagnostic criteria were planned to be eventually subjected to a sensitivity analysis.
- development of impaired glucose tolerance (plasma glucose two hours after a 75 g oral glucose load equal or greater than 140 mg/dl (7.8 mmol/L) and less than 200 mg/dl (11.1 mmol/L) (WHO 1999));
- development of impaired fasting glucose (fasting plasma glucose equal or greater than 100 mg/dl (5.6 mmol/L) and less than 126 mg/dl (7.0 mmol/L) (ADA 2004));
- anthropometric measures: body weight, body mass index (BMI) and waist-to-hip-ratio;
- systolic and diastolic blood pressure;
- lipid levels: total cholesterol, LDL- and HDL-cholesterol, triglycerides;
- quality of life, ideally measured with a validated instrument;
- adverse effects (for example traumatic injuries secondary to leisure physical activity, nutritional deficits);
- all-cause mortality;
Covariates, effect modifiers and confounders
Timing of outcome measurement
Outcomes were planned to be assessed in the middle (up to two years of follow-up) and long term (more than two years of follow-up) according to clinical criteria.
Search methods for identification of studies
See Cochrane Metabolic and Endocrine Disorders Group methods used in reviews.
We used the following sources for the identification of trials:
- The Cochrane Library (issue 1, 2008);
- MEDLINE (until March 2008);
- EMBASE (until March 2008);
- CINAHL (until March 2008);
- LILACS (until March 2008);
- SocioFile (until March 2008)
We also searched databases of ongoing trials: Current Controlled Trials (www.controlled-trials.com - with links to other databases of ongoing trials).
The described search strategy (see for a detailed search strategy under Appendix 1 was used for MEDLINE. For use with EMBASE, The Cochrane Library and the other databases the search strategy was slightly adapted. For the database LILACS, we used a simplified search strategy because the database does not permit to introduce so many search terms.
Searching other resources
We tried to identify additional studies by searching all the reference lists of relevant trials, reviews and meta-analysis. Studies published in any language were included.
Data collection and analysis
Selection of studies
To determine the studies to be assessed further, three authors (DM, GGP, AMB) independently scanned the abstracts or titles, or both sections and the keywords of every study identified. Each study was evaluated independently by two authors (DM in combination with GGP or AMB). All potentially relevant articles were investigated as full text. Where differences in opinion existed, the third author who initially did not evaluate the article reviewed it to reach a final decision between the three authors. An adapted QUOROM (quality of reporting of meta-analyses) flow-chart of study selection is attached (Moher 1999).
Data extraction and management
For studies that fulfilled inclusion criteria, two authors (LJOS, AMB) independently abstracted relevant population and intervention characteristics using standard data extraction templates (for details see Characteristics of included studies and Appendix 2; Appendix 3; Appendix 4; Appendix 5 and Appendix 6) with any disagreements resolved by discussion, or if required by a third party. We attempted to contact the original authors of the articles for missing data.
Assessment of risk of bias in included studies
The risk of bias of the included trials was assessed by examining sequence generation, allocation concealment, addressing of incomplete outcome data, selective reporting and other potential bias. Blinding of participants and caregiver or treatment administrator to group assignment is not feasible in trials on lifestyle interventions (diet and/or exercise) but blinding of outcome assessors to group assignment and blinding of participants and caregivers or treatment administrators to outcomes are evaluated. Risk of bias assessment of all included studies was performed independently by two authors (AMB, MR). When differences in opinion between these two authors were found, a third author (DM) evaluated the study to reach a final decision between the three reviewers. Studies were not excluded on the basis of high risk of bias; a sensitivity analysis was performed to compare results between studies with potential bias and those without.
Measures of treatment effect
The effect sizes for dichotomous data were expressed as risks ratios in all trials except the Da Qing 1997.
As the Da Qing 1997 study was randomised at the clinic level, the consequential clustering effect was adjusted for by re-analysing the reported data by fitting a Poison regression model, with clinic included as a random effect. As number of events is modelled, rather than rate, person years of follow-up are entered as an offset in the linear predictor. The output was an incidence rate ratio for each comparison adjusted for clustering. The incidence rate ratio can be assumed to be equivalent to a hazard ratio and included in the meta-analyses as such. The analysis was carried out in STATA (Gillies 2007).
For continuous outcomes, weighted mean differences and 95% confidence intervals (CI) were calculated.
Dealing with missing data
Relevant missing data were planned to be obtained from authors. Evaluation of important numerical data such as screened, eligible and randomised patients as well as intention-to-treat (ITT) and per-protocol (PP) population was carefully performed. Attrition rates, for example drop-outs, losses to follow-up and withdrawals were investigated. Issues of missing data, ITT and PP were critically appraised and compared to specification of primary outcome parameters and power calculation.
Dealing with duplicate publications
In the case of duplicate publications and companion papers of a primary study, we tried to maximise yield of information by simultaneous evaluation of all available data. In cases of doubt, the original publication (usually the oldest version) obtained priority.
Assessment of heterogeneity
In the event of substantial clinical, methodological or statistical heterogeneity study results were not planned to be combined by means of meta-analysis. Heterogeneity was examined with I
Assessment of reporting biases
The assumption of publication bias was planned to be visually evaluated by using a funnel plot, whereby effect estimates of the common outcome measure were plotted against trial sample size. There are a number of explanations for the asymmetry of a funnel plot, including true heterogeneity of effect with respect to study size, poor methodological design of small studies and publication bias (Sterne 2001). Therefore, we carefully used this tool (Lau 2006).
All data were summarised statistically under a random-effects model meta-analysis. Continuous outcomes were combined with the Der-Simonian method, and diabetes incidence with the inverse-variance method. Statistical analysis were performed according to the statistical guidelines referenced in the newest version of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008).
Subgroup analysis and investigation of heterogeneity
Subgroup analyses were planned to be performed, if feasible, to assess whether some groups of participants at risk for type 2 diabetes could obtain more benefit from exercise or exercise and diet interventions. Subgroups were planned to be analysed according to risk factors for diabetes mellitus (impaired glucose tolerance, abnormal fasting glucose, previous gestational diabetes, family history of type 2 diabetes, obesity, dyslipidaemia, hypertension, high risk ethnic groups). Additionally, sex and age subgroups and different types of diets (for example weight reduction versus emphasis on 'healthy eating') or exercise schedules (for example daily exercise versus twice a week exercise, types of exercise), were planned to be analysed. Finally, intervention duration was considered in the subgroup analysis.
Sensitivity analyses took into account the influence of the following factors on effect size:
- repeating the analysis to explore the influence of risk of bias, as described above;
- repeating the analysis excluding very large or long studies to evaluate how much they influence the results;
- repeating the analysis excluding trials, if any, supported by industry;
- changes in the diagnostic criteria of type 2 diabetes mellitus through the years could produce significant variability in the clinical characteristics of the patients included as well as in the results obtained.
The robustness of the analysis was planned to be explored further by repeating the analysis using different measures of effects size (odds ratio, risk difference) and different statistical models (fixed- and random-effects models).
Description of studies
Results of the search
The initial search identified 4875 records. After revising all titles and abstracts, full papers were obtained from all potentially relevant studies (46 in total) for further examination. Finally 25 publications describing eight studies met the inclusion criteria (Bo 2007; Da Qing 1997; DPP 2002; DPS 2001; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998). Four of the included papers were found by hand searching. The other studies were excluded on the basis of their abstracts or full texts because they were not relevant to the question under study (see Figure 1 for details of the amended QUOROM (quality of reporting of meta-analyses) statement). No non-published studies were identified. One ongoing trial was identified (EDPS). Five papers describing three studies (Kinmonth 2008; Mensink 2003; Savoye 2007) had not published yet the diabetes incidence, these studies might be included in further updates.
|Figure 1. QUOROM (quality of reporting of meta-analyses) flow-chart of study selection.|
For details about the included studies see 'Characteristics of included studies'. Heterogeneity was found in terms of the inclusion criteria, the intervention, ethnic groups, age, weight and body mass index (BMI). There was also heterogeneity in the diagnostic criteria to define impaired glucose tolerance and type 2 diabetes.
Agreement in study selection, that is qualifying a study as 'included' or 'potentially relevant', was complete among authors.
All included publications focused their interventions on improving physical activity and encouraging weight loss. Two of them (Da Qing 1997; Wing 1998) separated the intervention in four study arms: exercise only, diet only, exercise plus diet and control group. All publications included diet and exercise interventions in the same group.
Two of the publications included pharmacological interventions in separated arms (DPP 2002; IDPP 2006), in these cases, and for the purpose of this review, the results of these arms are not reported. None of the studies focused exclusively on exercise interventions for the prevention of diabetes.
The exercise interventions differed largely between trials, from the advise to promote physical activity (Bo 2007; Da Qing 1997; IDPP 2006; Kosaka 2005; Oldroyd 2005), to a few times weekly supervised exercise programmes, that differed in intensity (DPP 2002; DPS 2001; Wing 1998). Most of the programmes included exercises like walking, jogging and cycling, with different intensities.
The diet interventions were based mainly on caloric restriction, reduced fat intake and increased fibre intake (Bo 2007; Da Qing 1997; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). In one of the publications, advise was given to decrease 5% to 10% the amount of each meal, depending on the body mass index of each individual, when this was equal or greater than 24 kg/m2, and the avoidance of weight gain if BMI was lower than 24 kg/m
Two of the publications did not report any behavioural intervention (Bo 2007; Da Qing 1997). All other publications included different forms of behavioural interventions. The record of the physical activity or the dietetic conduct for self-feedback was used in two of the publications (DPS 2001; Wing 1998). In the remaining publications, motivational strategies and setting up of goals were used (DPP 2002; DPS 2001; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998).
In one of the included studies, the control group did not receive any intervention (Oldroyd 2005). In the seven remaining studies, the control group received habitual recommendations, advice or education on how to increase physical activity and have a healthier diet to achieve weight loss.
Number of study centres
Three publications report the number of centres where the trial was performed, five in DPS 2001, 27 in DPP 2002 and 33 in Da Qing 1997. In the remaining publications the number of study centres was not clearly reported.
Two of the trials were performed in the United States (DPP 2002; Wing 1998); one in Italy, one in Finland, one in the UK, one in Japan, one in China and one in India, respectively (Bo 2007; DPS 2001; Oldroyd 2005; Kosaka 2005; Da Qing 1997; IDPP 2006).
Duration of the interventions
The interventions lasted from one year in Bo 2007 to six years in Da Qing 1997. The number of contacts with the individuals in the interventions ranged from five in Bo 2007 to 51 in Wing 1998. The intervention was always given directly to the participants, in groups or on an individual basis. In most of the publications the intervention facilitators were a physiotherapist, an exercise physiologist and/or a dietitian.
Duration of follow-up
Follow-up duration differed largely between trials; it ranged from one year in Bo 2007 to seven years in DPS 2001. Follow-up times from Da Qing 1997; DPP 2002; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998 were 6, 2.8, 3, 4, 2 and 2 years respectively. Intervention in DPS 2001 was prematurely terminated by an independent end-point committee on the basis of the results of the first data analysis. Median duration of the intervention in that study was four years (mean 3.2 years). Participants were followed up until a median of seven years. In the data analysis, results of DPS 2001 are presented at mean follow-up time of 3.2 years (end of intervention).
For details of compliance measures see Table 1.
All studies reported some compliance measurement. Seven studies used self-reporting methods (Bo 2007; Da Qing 1997; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). Self-reporting methods comprised three day food diaries (Da Qing 1997; DPS 2001; Wing 1998), questionnaires (Bo 2007; Da Qing 1997; Wing 1998), self-reporting (DPP 2002; IDPP 2006; Oldroyd 2005) and individual interviews (Da Qing 1997; DPS 2001). One study measured improvement in physical condition with a half-mile walk test (Wing 1998) measuring time to completion and predicting VO
Language of publication
Two of the initially selected studies were published in Chinese (Fang 2004; Tao 2004) and one of the studies was published in Japanese (Sakane 2006). After data extraction was performed by original language speakers these studies were excluded because they did not meet the inclusion criteria (see table 'Characteristics of excluded studies'). All included studies were published in English.
In total, the eight included studies had 5956 participants (range 78 to 3234), the mean age was 50.3 years and the mean BMI was 31.2 kg/m
Most of the included individuals were recruited from the community (Da Qing 1997; DPS 2001; IDPP 2006; Wing 1998), from a clinic population (Kosaka 2005), from a combination of the community and a sample of clinical persons in DPP 2002, from a combination of research studies, local hospital biochemistry laboratory databases and general practitioner surgeries in Oldroyd 2005 and persons aged 45 to 64 from family physicians, representative of the local health districts (Bo 2007).
In one study the inclusion criterion was to have a metabolic syndrome or two components of the metabolic syndrome plus high-sensitivity CRP (hs-CRP) serum values equal or greater than three mg/L (Bo 2007). In one study the inclusion criterion was to be overweight (30% to 100% ideal body weight), nondiabetic and to have one or two biological parents with diabetes mellitus (Wing 1998). In the other six studies an inclusion criterion was to present some kind of glucose tolerance alteration. Nonetheless, the definition of glycaemic values varied. Two of the studies had the additional inclusion criterion to have a BMI equal or greater than 24 kg/m
Generally, exclusion criteria were the presence of chronic diseases that could interfere with the participation in the intervention group or to complete follow-up, and the use of medications that could interfere with the results (Bo 2007; DPP 2002; DPS 2001; Kosaka 2005; Oldroyd 2005). Three studies did not report any exclusion criteria (Da Qing 1997; IDPP 2006; Wing 1998).
The way to define glycaemic values to determine glucose intolerance and the diagnosis of diabetes varied among studies: In one study (Kosaka 2005) impaired glucose tolerance (IGT) was defined as a fasting plasma glucose (FPG) value below 140 mg/dl and a plasma glucose two hours after a 100 g glucose load between 160 and 239 mg/dl, these values were described by the authors as roughly corresponding to the 140 to 199 mg/dl range in the 75 g oral glucose tolerance test (75 g OGTT) and thus corresponding to IGT according to the World Health Organization in 1980 (WHO 1980).
Four studies (Bo 2007; Da Qing 1997; DPS 2001; Oldroyd 2005) used the criteria defined in 1985 (WHO 1985). The criteria of the American Diabetes Association in 1997 (ADA 1997) was used in DPP 2002. The IDPP 2006 study used the glycaemic diagnostic criteria of WHO 1999.
Wing 1998 used the criteria of WHO 1985, diabetes incidence was tried to be assessed retrospectively using the new criteria for diabetes in WHO 1997.
For details on secondary outcomes see Appendix 6.
Non of the studies reported the incidence of impaired glucose tolerance or impaired fasting glucose. Most of the studies reported the change from baseline to follow-up of fasting plasma glucose values (Bo 2007; Da Qing 1997; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998) or values at two hours after a glucose load (Da Qing 1997; DPS 2001; IDPP 2006; Oldroyd 2005). Seven studies reported changes from baseline to follow-up in body weight or body mass index (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998). Waist circumference or waist-to-hip ratio was available from six studies (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). Data on lipid profiles were reported in five studies (Bo 2007; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). Systolic and diastolic blood pressure were available from six studies (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998).
Three studies (DPP 2002; IDPP 2006; Da Qing 1997) described adverse effects. Cost-effectiveness of the intervention was investigated in two studies (DPP 2002; IDPP 2006). None of the studies reported the results of quality of life measures.
Sixteen papers had to be excluded after careful evaluation of the full publication. There were different reasons for exclusion (for details see Characteristics of excluded studies).
Risk of bias in included studies
|Figure 2. Risk of bias summary: review authors' judgments about each risk of bias item for each included study.|
|Figure 3. Risk of bias graph: review authors' judgments about each risk of bias item presented as percentages across all included studies.|
All included studies were randomised controlled clinical trials of parallel design. One study was randomised by clinics (Da Qing 1997) and all other studies randomised individuals. Adequate sequence generation was performed in two studies (Bo 2007; Oldroyd 2005), the other studies did not report how the sequence generation was performed. Three studies reported how allocation concealment was performed (Bo 2007; DPP 2002; Oldroyd 2005). In DPS 2001 allocation concealment was not performed and in the rest of the studies it was not mentioned in the publications. Kosaka 2005 was the only study that specified a randomisation ratio other than 1:1; the randomisation ratio of this study was 1:4 (intervention:control). The control group was divided into three subgroups: a group in which body weight increased by 1.0 kg or more, a group in which body weight remained unchanged, and a group in which body weight decreased by 1.0 kg or more.
Double-blinding is not possible or practical in these studies because of the type of intervention. Two studies had a medication treatment arm; in DPP 2002 double-blinding in the medication and control (placebo) arms was stated. Three studies reported the blinding of outcome assessors (Bo 2007; DPS 2001; Wing 1998) and two studies the blinding of investigators and/or participants to some of the results (DPP 2002; IDPP 2006).
Incomplete outcome data
All studies reported discontinuation rates. Four studies described some details about discontinuing participants (Da Qing 1997; DPS 2001; IDPP 2006; Oldroyd 2005). Discontinuing rates in the exercise plus diet group ranged from 8.7% (DPS 2001) to 23% (Oldroyd 2005). In the control group attrition rates varied between 2.2% (IDPP 2006) to 38% (Oldroyd 2005). Attrition rates between intervention and control groups were dissimilar in three studies (IDPP 2006; Oldroyd 2005; Wing 1998). Two studies did not report discontinuation rates of each arm separately (Da Qing 1997; DPP 2002). In the Bo 2007 study participants signed the informed consent after randomisation, all discontinuing participants were participants who did not sign the informed consent before the beginning of the intervention. In the Oldroyd 2005 study discontinuation rates were very high and dissimilar between groups. In two studies (Kosaka 2005; Wing 1998) no reasons for missing data were provided.
Selective reporting was unclear in all studies. One study (DPP 2002) had published a protocol, but data were presented in many publications, most of them are not included in this review. Therefore, it was not possible to track if all pre-specified outcomes were reported and if they were reported in the pre-specified way.
Other potential sources of bias
In one study (Da Qing 1997) physical exercise, expressed in units per day, was significantly higher at baseline in the diet plus exercise group than in the control group. In the Oldroyd 2005 study a significantly larger proportion of control participants reported engaging in regular physical activity at least once a week compared with intervention participants (53% versus 24%) and there were fewer women (10/32 (32%)) than men (22/32 (69%)) in the control group compared with the intervention group.
Effects of interventions
For details of baseline characteristics see Appendix 3.
Two studies demonstrated clinically relevant differences between intervention and control groups. In one study, the number of participants reporting engaging in regular physical activity sufficient to get their heart thumping at least once a week, was more prevalent in the control group as was the the proportion of men (Oldroyd 2005). In another study, physical exercise expressed in units per day, was significantly higher at baseline in the diet plus exercise group than in the control group (Da Qing 1997).
In most studies the proportion of female participants was higher than the proportion of males. In three studies the proportion was around 50% in the intervention group (Bo 2007; Da Qing 1997; Oldroyd 2005) and in one study there were no female participants (Kosaka 2005).
The mean age of patients randomised to intervention groups ranged from 44.2 to 58.1 years. The main ethnic groups participating in the trials consisted of Asian and Caucasian participants. One study included other ethnic groups (DPP 2002). Three studies did not state the ethnic group of the participants (Bo 2007; DPS 2001, Wing 1998).
Most study participants were overweight or obese, the mean body mass indices (BMI) in patients randomised to intervention groups ranged between 24.0 and 36.1 kg/m
For details of primary outcomes see Appendix 5.
Five of the included studies had as primary outcome the incidence of diabetes. Two studies had diabetes incidence as a secondary outcome (Bo 2007; Wing 1998) and one study did not have diabetes incidence as a primary or secondary outcome, but reported it (Oldroyd 2005). Overall, 4573 participants provided information on the incidence of diabetes.
For the analysis of the data an intention-to-treat (ITT) analysis was performed in all studies where data were presented in a per-protocol analysis when sufficient data were available (Bo 2007; DPS 2001; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998). Diabetes incidence of discontinuing participants was considered to be the same as in the control group. In one study there were not enough data to perform an ITT analysis: randomised participants in each group werrrrrrrrr not specified in the publication (Da Qing 1997).
All eight studies included an exercise and diet group, a standard recommendation or no intervention group. The total number of events was 339 of 1976 in the exercise plus diet groups and 616 of 2252 in the control groups. Pooling of the eight studies by means of random-effects meta-analysis revealed a risk ratio of 0.63 (95% CI 0.49 to 0.79). The test for heterogeneity indicated an I
We repeated the analysis excluding the largest study (DPP 2002) which had a weight of 26% in the random-effects model and had a low risk of bias. The risk ratio in the random-effects model was then 0.69 (95% CI 0.55 to 0.87). Heterogeneity decreased to an I
Two studies had a diet only and an exercise only arm (Da Qing 1997; Wing 1998). The total number of events was 62 of 178 in the exercise groups and 92 of 173 in the control groups. Data were combined in a random-effects meta-analysis comparing the exercise group versus the control group resulting in no statistical significant differences between the groups. The test for heterogeneity indicated an I
The total number of events was 62 of 178 in the exercise groups and 72 of 167 in the diet groups. Data were combined in a random-effects meta-analysis comparing the exercise group versus the diet group. No statistical significant differences between the groups were found. The test for heterogeneity indicated an I
Diabetes and cardiovascular related morbidity
One study reported cardiovascular related morbidity (DPP 2002). The IDPP 2006 study mentioned the number of cardiovascular events, four in the exercise plus diet intervention group and two in the control group. These outcomes were not the primary objective of these studies.
For details of secondary outcomes see Appendix 6.
Development of impaired glucose tolerance
No study investigated the development of impaired glucose tolerance. Three studies reported changes from baseline in the 2-h plasma glucose (DPS 2001; IDPP 2006; Oldroyd 2005). The Da Qing 1997 study reported data on 2-h plasma glucose but it was not included in the meta-analysis because participants were cluster randomised. Overall, 756 participants provided information on the changes from baseline of 2h PG. In the random-effects meta-analysis no significant difference between groups was found. The test for heterogeneity indicated an I
Heterogeneity could only be reduced to an I
Data of 2-h plasma glucose for the comparison of the exercise group versus the control group and the exercise versus the diet group were only available from Da Qing 1997. Therefore, it was not combined by means of a meta-analysis.
Development of impaired fasting glucose
No study investigated the development of impaired fasting glucose. Six studies reported changes from baseline in fasting plasma glucose (FPG) (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). The Da Qing 1997 study reported data on changes from baseline in FPG but it was not included in the meta-analysis because participants were cluster randomised. In total, 3315 participants provided information on the changes from baseline of FPG in the comparison exercise and diet versus control. Combining the data in a random-effects meta-analysis resulted in statistically significant differences between groups, favouring the exercise plus diet intervention group. The test of heterogeneity indicated an I
Data of FPG for the comparison of exercise group versus control group and exercise versus diet group were available from Da Qing 1997 and Wing 1998. Because the Da Qing 1997 study was cluster randomised, data were not combined by means of a meta-analysis.
For the comparison exercise and diet group versus control group there were six studies reporting body mass index (BMI) (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998). In total, 3315 participants provided information on the changes from baseline of BMI in the comparison exercise plus diet versus control. When data were combined in a random-effects model, statistically significant differences were found favouring the exercise plus diet intervention group. The test for heterogeneity indicated an I
Seven studies reported body weight (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Kosaka 2005; Oldroyd 2005; Wing 1998). Overall, 3773 participants provided information on the changes from baseline of body weight in the comparison exercise plus diet versus control. Pooling these seven studies by means of a random-effects meta-analysis resulted in statistically significant differences in the mean weighted difference favouring the exercise plus diet group. The test for heterogeneity indicated an I
Four studies reported the waist-to-hip-ratio (WHR) (DPP 2002; IDPP 2006; Oldroyd 2005; Wing 1998). In total, 2546 participants provided information on the changes from baseline of WHR in the comparison exercise plus diet versus control. No significant differences were found for WHR. The test for heterogeneity indicated an I
Four studies reported waist circumference (Bo 2007; DPP 2002; DPS 2001; Oldroyd 2005). Overall, 2983 participants provided information on the changes from baseline of waist circumference in the comparison exercise plus diet versus control. In the random-effects meta-analysis the weighted mean difference was statistically significant, favouring the treatment group. The test for heterogeneity indicated an I
We could not analyse body weight, BMI, WHR and waist circumference in the comparison of groups of exercise versus control group and diet versus exercise group because data were only available from one study.
Comparing exercise and diet groups versus control groups, five studies reported total cholesterol (Bo 2007; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998), five studies reported HDL-cholesterol (Bo 2007; DPS 2001; IDPP 2006; Oldroyd 2005; Wing 1998), three studies reported LDL-cholesterol (IDPP 2006; Oldroyd 2005, Wing 1998) and four studies reported triglycerides (Bo 2007; DPS 2001; IDPP 2006; Oldroyd 2005). Overall, 1154, 1154, 385 and 1091 participants provided information on the changes from baseline of total cholesterol, HDL-cholesterol, LDL- cholesterol and triglycerides, respectively. No statistically significant differences were found in total, HDL- and LDL-cholesterol between groups when pooling the data by means of a random-effects meta-analysis. When combining the data for triglycerides by means of a random-effects meta-analysis, it resulted in a weighted mean difference of -0.14 mmol/L (95% CI -0.22 to -0.05). No important heterogeneity was found in any of the lipid level analyses except in the HDL-cholesterol analysis were a heterogeneity of I
The DPS 2001 study reported changes in serum total cholesterol-to-HDL cholesterol. At one year of follow-up changes in the intervention group were -0.4(0.8) mmol/L and in the control group -0.1(0.8) mmol/L. At three years of follow-up changes were -0.6(0.9) mmol/L and -0.3(0.8) mmol/L in the control group (data ar presented a means (SD)).
Two studies reported the use of pharmacological therapy for dyslipidaemia. In the DPP 2002 study, at baseline 5.2% of the participant reported taking medication for dyslipidaemia. At three years of follow-up, 12% in the exercise plus diet group compared to 16% in the control group of the participants were taking medication for dyslipidaemia. In the DPS 2001 study 6% and 5% of the participant were taking cholesterol-lowering drugs in the control and exercise plus diet group respectively at baseline. By the end of year one, 8% of the participants in the control and 6% in the exercise plus diet group were taking cholesterol-lowering drugs.
In the comparison of the groups exercise versus control group and diet versus exercise group there was just one study reporting lipid levels (Wing 1998).
Systolic and diastolic blood pressure
Data of systolic and diastolic blood pressure were obtained from six studies (Bo 2007; DPP 2002; DPS 2001; IDPP 2006; Oldroyd 2005 Wing 1998). Overall, 2521 participants provided information on the changes from baseline of systolic and diastolic blood pressure in the comparison exercise plus diet versus control.
Combining the data of systolic blood pressure in a random-effects meta-analysis, the test for heterogeneity resulted in an I
Combining the data of diastolic blood pressure by means of a random-effects meta-analysis resulted in a weighted mean difference of -2 (95% CI -3 to -1). No statistical heterogeneity was found.
Two studies reported the use of antihypertensive medication (DPP 2002; DPS 2001). In the DPP study the use of antihypertensive medication at baseline was 17% in both groups, at three years of follow-up the use of hypertensive pharmacological therapy was 23% in the exercise plus diet group and 31% in the control group. In the DPS study at baseline 31% of the participants in the control group and 30% of the participants in the intervention group were taking antihypertensive drugs. These values did not vary at one year of follow-up.
Quality of life
No study reported measurements of quality of life.
For details on adverse effects see Appendix 4.
One study mentioned musculoskeletal symptoms (DPP 2002), 728 events in the exercise plus diet intervention group and 639 events in the control group. The same study mentioned hospitalisation, 168 in the exercise plus diet intervention group and 174 in the control group. The IDPP 2006 study mentioned a total of 25 cases of hospitalisation for various surgical procedures.
For details see Appendix 4.
Four studies made some statement about the number of participants who died during the course of the trial (Da Qing 1997; DPP 2002; IDPP 2006; Oldroyd 2005). The overall percentage of deaths was comparable between the intervention and control groups. These outcomes were not the primary objective of these studies.
Two studies published a within trial cost-effectiveness analysis of the exercise and diet intervention (DPP 2002; IDPP 2006). Both studies concluded to be cost-effective from perspective of a Health-Care System.
In the overall analysis of diabetes incidence the comparison between exercise plus diet versus control groups heterogeneity was found as indicated by I
In the analysis of 2h plasma glucose, heterogeneity was observed by an I
High heterogeneity was found in the overall analysis of fasting plasma glucose (I
We also found high heterogeneity in the analysis of all anthropometric measures. Heterogeneity could be substantially reduced in the BMI and body weight analysis by excluding from this the DPP 2002 and IDPP 2006 study. In the WHR analysis heterogeneity could be substantially reduced by excluding from the analysis either the DPP 2002 or the IDPP 2006 study. Mean baseline BMI differed between studies. In the meta-analysis of all anthropometric measures studies are sorted by baseline BMI. A general correlation between baseline BMI and decrease in BMI, body weight and WHR at follow-up could be observed. The IDPP 2006 and the IDPP 2006 study are the studies with the lowest and the highest median BMI from the studies that presented data on anthropometric measures if not taking in account the Wing 1998 study that falls out of the general tendency and the Kosaka 2005 study in the body weight analysis. See Analysis 1.5; Analysis 1.6 and Analysis 1.7.
Important heterogeneity was found in the HDL-cholesterol analysis ( I
Combining the data of systolic blood pressure in a random-effects meta-analysis, the test for heterogeneity resulted in an I
We did not perform subgroups analyses because covariates were unevenly distributed and there were not enough studies to estimate an effect in various subgroups.
For the planned BMI subgroup analysis, there were confounding characteristics in the different studies (studies with lower mean baseline BMI as well a studies mainly with Asian participants and studies with higher mean BMI with mainly Caucasian participants). One study presented data of study participants separated in lean (BMI less than 25 kg/m
Mean age at baseline was very similar between studies. Only the DPP 2002 study investigated the influence of age on the effects of exercise and diet interventions in the prevention of diabetes. Three age groups were established: 25 to 44, 45 to59 and 60 to85. The results show that the exercise and diet intervention was more effective with increasing age (6.3, 4.9, and 3.3 cases per 100 person-years, in the 25 to 44, 45 to 59, and 60 to 85 year age groups, respectively). There were other baseline differences between the three age groups. With increasing age there were more male (21%, 31%, 49% respectively) and Caucasian participants (47%, 55%, 66% respectively). The 60 to 85 year age group also had the lowest baseline BMI. Baseline fasting plasma glucose and 2-hour glucose were similar in all 3 groups.
Sensitivity analyses were performed for every element on the risk of bias table by excluding studies that had a high risk of bias. No statistically significant diferences in the effects were observed when comparing these results with the overall analysis.
Publication and small study bias
Not performed due to insufficient amount of data.
Summary of main results
This systematic review shows that educational interventions based on exercise and diet are effective in reducing the incidence of type 2 diabetes mellitus in people presenting with impaired glucose tolerance and the metabolic syndrome. Many of the individuals included in the studies had an additional risk factor for the development of diabetes, i.e. they were overweight or obese. Therefore, the conclusions of this review apply to this type 2 diabetes risk category of individuals. Also, it must be pointed out that although the interventions were heterogeneous in nature they were effective in different settings. However, as interventions aimed at changing the behavioural pattern of people are complex we do no presently know how these interventions perform outside a trial setting. It must be stressed that the effects of these interventions are consistently seen in all the studies except for two (Oldroyd 2005; Wing 1998). In the study by Oldroyd, the absence of effectiveness of the intervention may be at least in part explained by the fact that more individuals in the control group were physically active at inclusion in the trial. Additionally, the study by Wing found no effect of exercise alone or in combination with diet in the prevention of diabetes in the long term (two years), participants of this study were obese first-degree relatives of persons with type 2 diabetes. The results may be partly explained by the low proportion of individuals demonstrating long-term behavioural changes in this study. From the available information on follow-up of participants of some of the studies, the exercise and diet interventions show an effect that lasts even after the intervention had ceased. However, both studies (Oldroyd 2005; Wing 1998) had a high risk of bias (Figure 2).
No firm conclusion can be drawn about the effectiveness of exercise alone in preventing diabetes (see further comments under 'Limitations') as only two studies are available. The combined data from these two studies showed no significant difference when compared to the control participants, although the larger trial alone showed a positive effect of exercise in preventing diabetes. Further, in these same studies no difference was seen between the groups of exercise alone and diet alone in terms of diabetes incidence.
The results of this systematic review are consistent with those of previously issued reviews (Gillies 2007; Norris 2005; Yamaoka 2005;). Additionally, one of these reviews already showed that lifestyle interventions are at least as effective as pharmacological interventions in terms of type 2 diabetes prevention (Gillies 2007).
Individuals suffering from type 2 diabetes mellitus have a lower quality of life and a higher risk of cardiovascular morbidity and mortality. It may be stated that any diabetes-free period of life attained through these preventive strategies may be associated with an improved quality of life in people at risk. However, a major issue is the increased cardiovascular morbidity and mortality associated with diabetes; we do not know whether the evaluated interventions are able to prevent cardiovascular events in the population at risk. Additional cardiovascular risk factors associated with type 2 diabetes mellitus are high body weight and increased waist-to-hip ratio, dyslipidaemia and high blood pressure. The combined exercise plus diet intervention has favourable effects on weight reduction, waist circumference and blood pressure (systolic and diastolic). It may be predicted that any improvement in these risk factors may be associated with a favourable cardiovascular outcome. However, exercise and diet have a very modest effect on the lipid profile. Concerning the use of medications for lipid and blood pressure control, a conclusion can not be drawn although, if any, an effect of a lower use of these medications will favour the intervention arms of the two major trials (DPP 2002; DPS 2001).
In terms of cost-effectiveness, the available data of one of the large trials (DPP 2002) show that 'lifestyle programs' are associated with modest incremental costs compared with the placebo group (DPP 2002). The analysis showed that lifestyle interventions are cost-effective from the perspective of the Health-Care System (DPP 2002). The Markov simulation model showed that the delay in the development of diabetes and its associated complications was highly cost-effective in all age groups and dominated the pharmacological intervention with metformin (Herman 2005).
Limitations of the review
Concerning the effectiveness of exercise alone the results of the current review are insufficient to draw a final conclusion. Only two studies were identified addressing this question. The combination of the data coming from these two studies showed no significant effect of exercise alone in terms of prevention of type 2 diabetes mellitus. One of the studies (Da Qing 1997) showed a positive effect of exercise alone in terms of diabetes prevention as compared to standard recommendations. The other study (Wing 1998), as previously noted, showed no significant effect of such a strategy, probably as the intervention failed to keep participants compliant with the exercise intervention. However, the comparison of interventions based on exercise alone against those based on diet alone yielded important differences although again only a small number of individuals were included in these studies.
Only two studies with low risk of bias were identified (Bo 2007; DPP 2002). Exclusion of these trials in the analyses resulted in a decreased effect of the interventions under assessment.
Unfortunately, publication and small study bias could not be assessed at present due to the insufficient amount of the data. However, updating of this issue is planned in future versions of this review.
Implications for practice
Overall, interventions aimed at increasing exercise combined with diet are able to decrease the incidence of type 2 diabetes mellitus in participants with impaired glucose tolerance or the metabolic syndrome. There are insufficient data on exercise alone for diabetes prevention. Also, there are no data providing evidence of the effect of these interventions on morbidity and mortality. Further, no firm conclusions can be drawn from the available evidence on which strategy to follow when trying to induce behavioural changes in people at risk. These results should be taken into account by health-care policy makers when planning the implementation of these strategies in real-life settings. Additionally, the favourable cost-effectiveness of lifestyle measures over pharmacological intervention is to be taken into account when planning implementation of prevention programmes into routine clinical practice.
Implications for research
There is a need for studies exploring the effect of interventions on exercise alone or combined with diet on morbidity and mortality, with special focus on cardiovascular outcomes.
The authors would like to thank Mr Ivan Solà for his useful advice in the development of the protocol of this review.
We are grateful to Dr Ramachandran and Dr Oldroyd for providing additional non-published data, and to Dr Gilllies for her useful advise on the analysis of the cluster randomised trial.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Differences between protocol and review
- Index terms
Appendix 1. Search strategy
Appendix 2. Details of study features
Appendix 3. Baseline characteristics
Appendix 4. Adverse events
Appendix 5. Primary outcome data
Appendix 6. Secondary outcome data
Last assessed as up-to-date: 29 February 2008.
Protocol first published: Issue 2, 2001
Review first published: Issue 3, 2008
Contributions of authors
OROZCO LJ: searching for trials, quality assessment of trials, data extraction, data analysis.
BUCHLEITNER AM: searching for trials, quality assessment of trials, data extraction, data analysis, review development.
GIMENEZ-PEREZ G: protocol development, searching for trials, review development.
ROQUE M: quality assessment of trials, data extraction, data analysis, review development.
RICHTER B: protocol development, quality assessment of trials, data analysis, review development.
MAURICIO D: protocol development, searching for trials, quality assessment of trials, data analysis, review development.
Declarations of interest
Sources of support
- Corporacio Parc Taulí, Spain.
- Hospital de la Santa Creu i Sant Pau, Spain.
- Hopital Universitari Arnau de Vilanova, Spain.
- Institut de Recerca Biomèdica de Lleida, Spain.
- Agència d'Avaluació de Tecnologia i Recerca Mèdiques, Departament de Salut de la Generalitat de Catalunya, Spain.The review was supported by Grant No. 075/23/06
Differences between protocol and review
Background has been updated significantly.
Risk of bias of studies assessment has substituted the quality of studies assessment defined in the protocol. This was done for adapting the review to RevMan 5 and the Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.0 (Higgins 2008).
Medical Subject Headings (MeSH)
MeSH check words
* Indicates the major publication for the study