Non-pharmacological interventions to reduce the risk of diabetes in people with impaired glucose regulation: a systematic review and economic evaluation

Prevalence of diabetes, impaired glucose tolerance and impaired fasting glucose in the UK

In 2009 there were estimated to be 2,213,138 people with diabetes in the adult population in England, a prevalence of 5.1%. 3

The Health Profile of England 2009 reports a prevalence of 5.6% of men and 4.2% of women with diabetes in England in 2008. 4 Diabetes was flagged as one of the ‘particular challenges’.

Type 2 diabetes mellitus was estimated to account for 92% of all (diagnosed and undiagnosed) diabetes. The Yorkshire & Humber Public Health Observatory (YHPHO) provides a very useful set of data on diabetes in England. Between 1994 and 2003 the incidence of diabetes doubled from 1.8 to 3.3 per 1000 (age standardised) and the diagnosed prevalence doubled. 5 However, recent data from the Association of Public Health Observatories (APHO) model6 suggest that 26% of men and 22% of women with diabetes are undiagnosed.

The prevalences of doctor-diagnosed T2DM in England in different age and ethnic groups were reported in the Health Survey for England 20047 and are shown in Figures 1 and 2.

FIGURE 1.

Prevalence of doctor-diagnosed T2DM between different ethnic groups in England. Note: figures for the general population were taken from the 2003 survey and represent all ethnic groups in England, which would be predominantly white. Copyright^© 2011, re-used with the permission of The Health and Social Care Information Centre. All rights reserved.

FIGURE 2.

Prevalence of doctor-diagnosed T2DM by age group. Copyright^© 2011, re-used with the permission of The Health and Social Care Information Centre. All rights reserved.

A study by Bagust et al. (2002)8 predicts that by 2036 there will be approximately 20% more cases of T2DM in the UK than in 2000, as a result of the population ageing and increased levels of obesity. 8

The consequences and costs of diabetes

Some key consequences are:

• Life expectancy is reduced. The National Service Framework9 summarised this as being, on average, a loss of up to 10 years of life, but it will vary by age and by gender, with females losing more than males. 5

• Atherosclerosis – arterial disease, which increases the risk of heart attacks, strokes and amputation. The increase is more marked in women. The relative risk (RR) for fatal coronary heart disease (CHD) is 50% higher in women with diabetes than it is in men. 10

• Mortality rates from CHD are up to five times higher for people with diabetes, and the risk of stroke is three times higher. 9

• Nephropathy – disease of the kidneys, which can lead to renal failure requiring dialysis or transplantation. Diabetes is a leading cause of renal failure, and the second commonest cause of lower-limb amputation.

• Retinopathy – disease of the eye, which can lead to visual loss. Diabetes is the leading cause of blindness in people of working age. 9

• Neuropathy – damage to the nerves, which cause a range of problems, including loss of sensation, neuropathic pain, muscle wasting, and altered blood flow.

• Diabetes is estimated to account for 5% of all NHS expenditure and 9% of hospital costs. 5

• The presence of diabetic complications increases the overall NHS spending per patient more than fivefold and increases by five the chance of a person needing hospital admission. 9 Based on TARDIS-2 data, a person in the UK with T2DM incurs average direct costs of > £2000 per year, accounting for 4.7% of all NHS expenditure. 2

• One in 20 people with diabetes incurs social services costs, and for these people the average annual cost was £2450 in 1999. 9

Maintenance of blood glucose levels as close to normal as possible slows or prevents the onset and progression of eye, kidney and nerve diseases caused by diabetes. The evidence for the link between glucose control and arterial disease is less strong, but good control reduces the risk. 11–13

Global prevalence of type 2 diabetes mellitus

Based on current estimates, the global prevalence of T2DM will double from 171 million patients in 2000 to 334 million patients in 2025. The increase in the number of patients will be most pronounced in nations that are currently undergoing socioeconomic development, including increasing urbanisation. Global diabetes prevalence estimates from the International Diabetes Federation (IDF) suggest that the region with the highest prevalence of diabetes is North America at 7.9%, followed by Europe at 7.8%, and the Eastern Mediterranean and Middle East at 7%. When absolute numbers of people with diabetes are considered, it is South East Asia and the West Pacific that are expected to experience the highest increases in prevalence over the coming years. 16

Risk factors for developing diabetes

Age, weight and ethnic group are the three factors that most affect prevalence of T2DM in the UK. The Health Survey for England 20047 reported that the prevalence of T2DM increases sharply with age (see Figure 2). In the general population (2003 data), the prevalence of T2DM in men and women increased from 0.2% and 0.4%, respectively, at age 16–34 years, to 2.2% and 1.7%, respectively, at age 35–54 years and to 9.3% and 6.9%, respectively, in those aged ≥ 55 years. The survey reported that in all but the youngest age groups, the prevalence of T2DM was higher in men than in women.

The risk of diabetes is much higher in people of South Asian ethnicity (i.e. from the Indian subcontinent: Pakistan, India, Bangladesh, Nepal) (see Figure 1). We deal with their situation in greater detail later in Chapter 3.

Lifestyle factors are the main determinants of T2DM but there are strong familial influences:

• In families in which no member has T2DM, the lifetime risk of developing it is approximately 10%.

• If one parent is affected then the risk increases threefold (i.e. 30% lifetime risk).

• If a sibling is affected then the risk increases fourfold (i.e. 40% lifetime risk).

• If both parents are affected then the risk increases sevenfold (i.e. 70% chance).

• If an identical twin is affected then the risk increases ninefold (i.e. 90% chance). 5

These risks are increased if the affected family member had early onset of T2DM. It is likely that the diabetes is preceded by impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT), and that family history could provide opportunities for targeted screening.

Obesity is a strong risk factor. It has been suggested that it is only obesity [body mass index (BMI of > 30 kg/m²)] that increases the risk and not overweight (BMI of > 26–29 kg/m²). Jarrett (1986),17 reviewing older studies, noted that there was little difference in the risk of diabetes at a range of BMIs of < 31 kg/m². However, recent work shows the take-off to diabetes at lower BMIs18 [Figure 3, based on data in the study of Ford etal. (1997)18]. It may be that physical activity is less now than in previous decades. Jarrett (1986)17 was reviewing studies from the early 1980s.

FIGURE 3.

Age-adjusted incidence rates of diabetes as a function of baseline BMI in 30- to 55-year-olds (both sexes).

The National Audit Office suggests that 47% of T2DM cases in England can be attributed to obesity. 19 The risk of T2DM is almost 13 times greater in obese women than in women of normal weight. For men the risk is five times greater. 5

Physical activity reduces the risk of developing T2DM. Jeon et al. (2007),20 in a meta-analysis of cohort studies, found that the risk of T2DM was 30% lower in those undertaking regular moderate-intensity activity than in sedentary people. The reduction in risk associated with increased activity levels is independent of BMI. Sui et al. (2007),21 in a study of mortality in older American adults, found that fitness was a predictor of mortality, independent of adiposity; they also reported that when people were grouped into fitness quintiles (based on time tolerated on treadmills), the fittest quintile had a prevalence of diabetes of 12% compared with 25% in the least-fit quintile.

Physical activity is dealt with in greater detail later in Chapter 2.

The National Service Framework for Diabetes noted socioeconomic differences in the risks of developing T2DM. 9 Those in the most deprived fifth of the population are 1.5 times more likely than average to have diabetes at any given age.

Thomas et al. (2007)22 used data from the 1958 British birth cohort to provide an estimate of the national prevalence of diabetes risk in mid-life, using glycated haemoglobin (HbA_1c) level of ≥ 5.5% as an indicator for possible subclinical alterations in glucose metabolism, which would detect those with IGT as well. The study consisted of 7799 participants born in England, Scotland and Wales and currently living in the UK.

The majority of the population (79.3%) had an HbA_1c level of < 5.5%; 16.7% had an HbA_1c level of 5.5–5.9%; 2.0% had an HbA_1c level of 6.0–6.9%; and 0.6% had an HbA_1c level of ≥ 7.0%. More men than women were found in the higher-HbA_1c level categories and a highly significant inverse socioeconomic gradient was apparent for HbA_1c levels of ≥ 5.5%. When looking at regional variation in HbA_1c levels of ≥ 5.5%, it was found that the east of England had a significantly higher prevalence than the rest of the UK (24.3% vs 19.3%) and Scotland had the second highest prevalence (22%).

On the basis of these data, the authors suggest that a proportion of people are likely to have subclinical elevations in HbA_1c level in their mid-40s, and this proportion is greater in some socioeconomic groups and geographical regions than in others. These individuals are likely to represent the population that has IGT, and at risk of developing T2DM.

Children and adolescents with type 2 diabetes mellitus

A recent alarming development is the emergence of T2DM and glucose intolerance in children and adolescents. 23 The year 1979 saw the first childhood cases of T2DM reported in Pima Indians; however, there are now worldwide reports of children with the disorder. This is thought to be associated with the growing problem of childhood obesity over the last 15 years. In particular, those ethnic groups at high risk of diabetes have reported a close association between the development of diabetes and childhood obesity. 24 Other contributing factors should also be considered, including genetic susceptibility as a result of variations in, for example, insulin sensitivity. In the UK, reports of children with T2DM have appeared only recently. 25 Compared with type 1 diabetes mellitus (T1DM), children with T2DM presented later (12.8 years vs 9.3 years) were usually female, overweight or obese (92% vs 28%) and a large proportion were from ethnic minority groups. In common with their adult counterparts, South Asian children have an increased risk of T2DM (RR = 13.7) compared with white UK children. It is therefore becoming apparent that preventative interventions may also need to be targeted at a younger age group than previously thought in order to prevent the progression to diabetes and heart disease at a later age.

Definitions

Two conditions appear to precede T2DM. The first is IGT, in which fasting glucose is normal but there is post-prandial hyperglycaemia. The definition comes from the oral glucose tolerance test (OGTT). The second is IFG, when the fasting level is raised but the post-prandial level does not reach IGT levels. The normal FPG is currently defined as ≤ 5.5mmol/l.

The World Health Organization (WHO) defines diabetes (Table 1) on the basis of a fasting plasma glucose (FPG) level of ≥ 7 mmol/l or a 2-hour level (in an OGTT) of ≥ 11.1 mmol/l. 34

Impaired glucose tolerance is defined as an FPG level of < 7 mmol/l and a 2-hour level of 7.8–11.0 mmol/l.

Impaired fasting glucose is defined as the range of 6.1–6.9 mmol/l and a 2-hour level of < 7.8 mmol/l.

So those with IFG should exclude IGT, but this is not possible if the 2-hour level has not been measured, whereas those with IGT can have IFG as well.

These criteria differ from those of the ADA, which defined IFG as starting at 5.6 mmol/l. One effect is that a much greater proportion of people with IGT also have IFG. The Danish Inter-99 study35 found that using the WHO criteria, 25% of those with IGT have IFG compared with 60% using the ADA cut-off.

One problem with the WHO criteria is that there is a group of people with FPG levels of 5.6–6.0 mmol/l – above normal but lower than IFG. The ADA definition tidies up this anomaly. 36 There is reluctance in the WHO to do this. First, the prevalence of IFG would greatly increase. In the D.E.S.I.R. study of French men and women aged 30–64 years,37 the prevalence of IFG was 13% in men and 4% in women using the 6.1 mmol/l cut-off, but would rise to 40% and 16%, respectively, using the 5.6 mmol/l cut-off. Second, and more importantly, the risk of progression to diabetes is much higher in people with FPG levels of > 6.1 mmol/l (‘old IFG’) than in those with FPG levels of between 5.6 and 6.0 mmol/l (‘new IFG’).

Progression to type 2 diabetes mellitus from impaired glucose tolerance or impaired fasting glucose

Impaired glucose tolerance and IFG have been called ‘pre-diabetes’ but the term is unsatisfactory because not all people with the two conditions go on to develop diabetes. The proportions of people who progress to diabetes vary among different studies in different populations (Table 2). 38 The 2005 Agency for Healthcare Research and Quality (AHRQ) report on the diagnosis, prognosis and treatment of IGT and IFG concluded that with regard to prognosis of patients following evaluation of a number of trials that reported relevant outcomes:

There is consistent evidence that IFG and IGT are both risk factors for the development of diabetes mellitus. The pooled RR for new DM is 6.02 [95% confidence interval (CI) 4.66 to 7.38] in people with IGT, 4.70 (95% CI 2.71 to 6.70) in people with IFG, and 12.21 (95% CI 4.32 to 20.10) in people with both disorders.

The D.E.S.I.R. study37,39 examined the risks of progression to diabetes from different baseline levels.

UK prevalence of impaired glucose tolerance and impaired fasting glucose

A number of factors influence the prevalence of IGT. As with T2DM, increasing age, obesity levels and ethnicity are risk factors for the development of IGT. As discussed previously, the use of different diagnostic criteria means prevalence rates may not be standardised across studies. 57 Two studies in London58,59 reported prevalences of IGT in over-40-year-olds of 3% and 4.1%, respectively, whereas a study conducted in a slightly older age group of 59- to 70-year-olds in Hertfordshire reported a dramatically increased prevalence of 18%. 60 The UK prevalence of IFG was reported as being approximately 17% in men and women aged > 60 years from a sample of over 7000. 61

In terms of variation among ethnic groups, a study in Coventry62 has reported higher prevalence of IGT in Asians than in the Caucasian population; they reported prevalences of 11.2% and 8.9% in Asian men and women, respectively, compared with 2.8% and 4.3% in Caucasian men and women.

Global prevalence of impaired glucose tolerance and impaired fasting glucose

The Diabetes Atlas, 2nd edition, 2002, compiled by the IDF, collated global regional estimates of prevalence of IGT in the 20- to 79-year age group using data from 212 countries and territories. 16 In 2003, the South East Asian region had the highest prevalence rate of IGT (13.2%), compared with 10.2% in Europe, 7.3% in both Africa and South/Central America, 7.0% in North America, 6.8% in the Eastern Mediterranean and Middle East and 5.7% in the Western Pacific region. These figures relate to single, broad age categories and did not divide prevalence according to age group and sex. This publication predicts that by 2025 an additional 7 million people in Europe will have IGT.

Studies in other developed countries, such as Australia, support the observation that the rate of IGT varies among studies. Dunstan et al. (2002)63 reported 9% and 12% in men and women, respectively; however, this was much higher than the 3.4% seen in two (smaller) previous Australian studies (Busselton and Victoria, in 1981 and 1992, respectively). 64,65 In the USA, 15.4% of a cross-section sample of adults aged 40–74 years (tested from 1988 to 1994) had IGT. The potential effect of migration on prevalence of IGT may be illustrated by the results from studies on the Chinese population both in the UK and in China. In a population of 375 Chinese people in Newcastle aged between 25 and 64 years, the prevalence of IGT was 8.0% (4.0% to 12.0%) in men and 16.1% (11.0% to 21.2%) in women66 compared with a large study of 92,187 people in the Da Qing province of China,67 which reported prevalence of 0.9% in both sexes. This shows that prevalence rates in source populations may not always be relevant to migrants to the UK.

The prevalence of IFG is not as widely reported as IGT. Dunstan et al. (2002)63 reported that the prevalence of IFG in Australia was 8% in men and 3.4% in women. In the USA, 33.8% of a cross-section sample of adults aged 40–74 years had IFG; however, more recent estimates from 1999 to 2002 indicate that, among US adults aged ≥ 20 years, 26% had IFG, which was similar to the prevalence in 1988–94 (25%). In India, both IFG (8.7%) and IGT (8.1%) show high prevalence with an overlap in one-third.

Does ethnicity affect the association between obesity and diabetes?

Diaz et al. (2007)84 used cross-sectional data for eight ethnic groups from the 2003–4 National Health and Nutrition Examination Survey and the 2003–4 Health Survey for England to examine the relationships of BMI, WC and waist–height ratio (WHR) with diabetes risk and across different ethnic groups.

The presence of diabetes was ascertained in 11,624 adults ≥ 20 years old, self-reported as US white, US black, Mexican American, English white, English black, Bangladeshi, Pakistani, Indian or Chinese. Diabetes was defined as self-report of doctor diagnosis or HbA_1c level of > 6.1%.The crude prevalence of total diabetes for individuals of < 40 years old did not differ between the ethnic groups, but for individuals ≥ 40 years old there were significant (p < 0.01) differences.

The percentage of individuals with normal BMIs (< 25 kg/m²) who had diabetes was lowest (3.4%) in English white people. Higher prevalences were seen in other ethnic groups (5.0–10.9%). Mexican Americans (10.9%) and Bangladeshis (10.8%) have the highest prevalence of total diabetes, followed by Indians (8.7%) and Pakistanis (8.0%). Receiver operating characteristic curves for the association of total diabetes with BMI, WC and WHR for adults ≥ 40 years old showed that overall, for both genders, WHR and WC have larger areas under the curve than BMI, reflecting a statistically significantly (p < 0.05) higher discriminating ability for these two measures than for BMI. When stratifying by ethnic group it was generally found that WC and WHR generally have larger areas under the curve than BMI. However, there were differences between ethnic groups and between genders. WC and WHR were both significantly better predictors of future diabetes for both men and women in US white people and in women in US black people. In English white people, WHR in men and both WC and WHR were significantly better than BMI. WC and WCR showed no significant difference to BMI in either gender in Bangladeshis and Chinese.

Therefore, it appears that the association of diabetes and BMI may differ among ethnic groups, and different thresholds may be necessary. Adding other anthropomorphic measures, such as WC, may improve risk assessment. An important finding was that diabetes was found in many individuals who were classed as being of normal weight by BMI.

Frequency

Evidence suggests that exercise needs to be taken regularly for preventative effect. 95 The effect on insulin sensitivity of a single bout of aerobic exercise lasts between 24 and 72 hours, depending on the duration and intensity. 96 As the duration of increased insulin sensitivity is generally not more than 72 hours, the ADA recommends no more than two consecutive days without aerobic exercise. 94

In men there is a clear dose–response relationship between total energy expenditure in physical activity and the prevention of T2DM. Burchfiel et al. (1995)97 reported that the findings of the Honolulu Heart Program, in a large cohort of over 6800 Japanese American men, indicated a lower risk of diabetes in men who were taking the most exercise. Similarly, in the Physicians’ Health Study, the age-adjusted RR for diabetes gradually fell from 0.77 (CI 0.55 to 1.07) in men who exercised only once a week, when compared with men who did not exercise at all, to 0.58 (CI 0.40 to 0.84) in men who exercised five or more times per week (p = 0.0002 test for trend). 98

The evidence on the frequency of exercise in relation to the risk of diabetes in women is less clear. The Nurses’ Health Study found that regular exercise (at least once a week) was associated with a reduced incidence of T2DM (RR = 0.67, 95% CI 0.6 to 0.75, p < 0.001) but it found no clear trend according to frequency of exercise. 99 Compared with sedentary women (nurses who did not exercise weekly), RRs of T2DM are 0.74 (95% CI 0.6 to 0.91) for those who exercised once a week, 0.55 (95% CI 0.44 to 0.68) for those who exercised twice a week, 0.73 (95% CI 0.59 to 0.90) for those who exercised three times a week, and 0.63 (95% CI 0.53 to 0.75) for those who exercised four or more times a week. 99 In the Nurses’ Health Study, exercise was defined as ‘vigorous’ exercise, i.e. whether participants engaged in ‘any regular activity similar to brisk walking, jogging, bicycling, etc., long enough to work up a sweat’.

Intensity

A Cochrane review100 identified evidence for improvements in glycaemic control over a range of exercise intensities. Analyses conducted using the data from the Nurses’ Health Study cohort found that after controlling for intensity of exercise the RRs of T2DM were 0.67 (95% CI 0.55 to 0.81) for < 1 hour/week and 0.56 (95% CI 0.46 to 0.69) for ≥ 1 hour/week (the test for trend being significant at p < 0.001). 101 In a 14-year cohort of almost 6000 American men, it was found that for every 2.1-megajoule (MJ) (500 calories) increment in weekly energy expenditure, there was a 6% decrease in diabetes risk. 91 The age-adjusted risk of diabetes in American men who played vigorous sport was lower than in men who played moderate sport only. In this study, the age-adjusted risk of diabetes gradually fell from 1.00, 0.90, 0.69 to 0.65, as the levels of sports activity (self-reported at baseline) increased from no sports, moderate sports (energy expenditure 5 kcal/minute), vigorous sports (10 kcal/minute) to a combination of moderate and vigorous sports (p = 0.02 for trend).

Although the majority of evidence suggests that the most vigorous activity is protective, some evidence suggests little additional benefit in exceeding moderate-intensity activity. 95

Moderate- and high-intensity exercise may confer comparable benefits. 102 The Insulin Resistance Atherosclerosis Study reported that increased participation in non-vigorous as well as overall and vigorous physical activity was associated with increased insulin sensitivity. 103 Improvements in insulin sensitivity and glucose tolerance are seen in sedentary individuals who incorporate moderate levels of activity into their lifestyle but the change appears to occur far more slowly and less dramatically compared with higher-intensity activity. 104

In a 10-year cohort study Finnish women who engaged in vigorous activity [a metabolic equivalent of task (MET) value of ≥ 6] less than once a week were found to have a greater age-adjusted risk of diabetes of 2.23 (95% CI 0.95 to 5.23; p = 0.043) than women who engaged in vigorous activity at least once a week. 105 A similar trend was found for men, although the association was weak (age-adjusted RR = 1.63, 95% CI 0.92 to 2.88; p = 0.082). However, the association between intensity of activity and subsequent incidence of diabetes was no longer statistically significant for both genders after adjusting for total amount of activity (total energy expenditure).

In a prospective study of 7577 British men, the risk of diabetes decreases progressively as the intensity of physical activity increased from light to moderate, but the risk was not decreased any further in men who regularly performed vigorous activity. 106 Similar results were reported from a more recent prospective study of 7142 British men. 107 The study reported the age-adjusted risk of developing diabetes was lower in men who engaged in moderate physical activity (RR = 0.53) than in those who engaged in light (RR = 0.66) physical activity. However, no further reduction in risk was seen in men who engaged in moderately vigorous activity (RR = 0.53).

Jeon et al. (2007)20 carried out a systematic review of moderately intense physical activity and the risk of diabetes. ‘Moderately intense’ was defined as requiring a MET task score of 3.0–6.0 but not as high as vigorous (defined as requiring more than six times resting metabolic rate), and studies that involved mixtures of moderate and vigorous were excluded. Using MEDLINE, EMBASE and checking reference lists of retrieved studies, they identified 10 cohort studies involving 310,221 participants, of whom 9367 developed diabetes.

The risk of diabetes was reduced by 31% by moderate physical activity (RR = 0.69; 95% CI 0.58 to 0.83) compared with being sedentary. However, adjusting for BMI reduced the reduction to 17% (RR = 0.83; 95% CI 0.76 to 0.90).

One concern in debates about the prevention of diabetes has been to distinguish between exercise (e.g. an activity taken in addition to activities of daily living) and physical activity (with a broader meaning including all forms of activity). Taking exercise may incur time and financial costs, whereas increasing physical activity could be done as part of daily living, such as simply walking more. Another concern is whether those who take fixed periods of exercise may do less at other times.

Five of the studies in the review by Jeon et al. (2007)20 examined the role of walking, usually defined as at least 2.5 hours a week of brisk walking [e.g. walking at 5.6 km/hour (3.5 miles/hour) on a flat surface requiring 3.8 MET]. 20

Those who walked regularly had a 30% reduction in diabetes (RR = 0.70; 95% CI 0.58 to 0.84), though, again, adjusting for BMI reduced the reduction to 17% (RR = 0.83; 95% CI 0.75 to 0.91). However, this implies that moderate-intensity physical activity can reduce progression to diabetes even in those who do not lose weight. Compared with the minimal amount of walking in the reference category, the highest category was at least 10 MET hours/week, which is equivalent to ∼2.5 hours/week of brisk walking.

Time

The review by Bassuk and Manson (2005)108 identified evidence that shows physical activity sessions lasting as little as 10 minutes can improve the metabolic and cardiovascular risk profile of otherwise sedentary individuals. 102 The US Diabetes Prevention Program (DPP) reported a 58% decrease in the development of diabetes in high-risk individuals who exercised for a minimum of 150 minutes per week (approximately 20 minutes per day).

The same total amount of exercise taken in several instalments appeared to be more effective in a small study by Eriksen et al. (2007). 109 Eighteen patients with T2DM were randomised to either three 10-minute sessions or one 30-minute session daily, of home-based cycle training, for 5 weeks. Both groups got fitter, but the 3 × 10-minute group had lower fasting and post-load glucose levels. The authors speculate that more energy is expended in short bursts.

Type

It is evident that alternative forms of physical activity that produce similar metabolic improvements to aerobic exercise may also be beneficial in the management of T2DM. The review by Eves and Plotnikoff (2006)110 identified evidence to suggest that resistance training may be effective in improving glycaemic control. They reported that results of three randomised controlled trials (RCTs) showed the effectiveness of resistance exercise to improve insulin sensitivity was comparable with that reported for aerobic exercise. These three RCTs compared resistance training with control (either flexibility exercise or no exercise), but the review suggests that the reported effect was comparable with that reported for aerobic exercise in other studies. There is also some evidence that the duration of effect of resistance training on insulin sensitivity is somewhat longer, perhaps as some of its effects are mediated by increases in muscle mass. 111 One RCT randomised 43 individuals with T2DM to either resistance or aerobic training for 4 months. 112 The authors reported that HbA_1c level was significantly reduced with resistance training (from 8.3% ± 1.7% to 7.1% ± 0.2%; p < 0.001), but not with aerobic training (from 7.7% ± 0.3% to 7.4% ± 0.3%; p > 0.05). Fasting blood glucose and insulin resistance were also reduced and lipid profile improved with resistance but not aerobic training.

Although both aerobic and resistance exercise increase glucose disposal, resistance exercise tends to increase muscle mass and therefore glucose storage space. 94 The review by Eves and Plotnikoff (2006)110 found that increases in skeletal muscle mass are related to decreases in HbA_1c level. 110 This supports the hypothesis that resistance training improves glycaemic control by augmenting the skeletal muscle storage of glucose. In one of the studies cited in the Eves and Plotnikoff review (2006),110 Castaneda et al. (2002)113 reported that resistance training significantly improved glycaemic control, increased fat-free mass and reduced abdominal adiposity, whereas body weight, and total, arm and leg fat mass did not change between intervention and control groups. The review also found ‘no significant relationship between the improvements in insulin sensitivity and the losses in either visceral or subcutaneous fat’ with regard to resistance training. Improvements to functional status and body composition often occur more rapidly with resistance compared with aerobic training, and therefore might be more immediately rewarding. ‘As there is substantial evidence that supports both aerobic and resistance training, it is possible that a combination of both may be the optimal intervention’. 110

A recent meta-analysis of 27 studies to examine the effects of different modes of exercise training on measures of glucose control concluded that all forms of exercise training (aerobic, resistance and combined) produce reductions in HbA_1c levels (reductions 0.8% ± 0.3%), a measure of glucose control, with combined training generally being more effective than aerobic or resistance training alone. 114 (Note: These reductions are as great as seen with many glucose-lowering drugs. 30) The authors emphasise that the effects of exercise are similar to those of dietary and drug interventions and the combined effects might be moderate or large. Furthermore, evidence also suggested small beneficial effects on related risk factors for complications of diabetes and those with more severe disease were found to benefit most.

Sigal et al. (2007)115 carried out a RCT in 251 people with T2DM, aged 39–70 years, with four arms:

• aerobic training

• resistance training

• both the above

• a sedentary control group.

Exercise was carried out three times weekly for 22 weeks, in community facilities. The primary outcome was HbA_1c level at 6 months. Other outcomes included plasma lipids and blood pressure. HbA_1c level was reduced in the aerobic group by 0.51% (p = 0.007), in the resistance group by 0.38% (p = 0.038) and in the combined group by 0.9% compared with the control subjects. The bigger reductions in the combined group of 0.46% compared with aerobics alone, and 0.59% compared with resistance alone, were also statistically significant (p = 0.014 and 0.001, respectively). This may simply reflect the greater amount of exercise in the combined group. No significant differences were seen in blood pressure and lipids at 6 months.

Some of the conflicting results above may be related to baseline risk. Gill and Cooper (2008)116 carried out a systematic review of both observational studies and prevention trials. In brief, they noted that the marked reduction in diabetes incidence was attenuated once adjusted for BMI but that even after such adjustment physical activity reduced the risk by 20–30%. They also examined the evidence for thresholds, finding that there was uncertainty about whether there was a minimum level to achieve benefit, especially in women, but also that there appeared to be a level above which additional activity conferred no additional benefit. However, perhaps the most useful aspect of their review was the examination of the benefit according to baseline risk. Thus, in lean individuals with very low risk of diabetes, physical activity made little difference, whereas in people at high risk (high BMI, family history), physical activity was clearly beneficial, although especially so if combined with weight loss. They suggest a BMI threshold of 27 kg/m² (23 kg/m² in those of Asian descent) for prescribing higher levels of physical activity than the current health guidelines recommend (150 minutes of moderate activity or 60–90 minutes of vigorous activity per week).

Adherence

It is often those who would benefit the most from aerobic exercise that have the greatest difficulty performing it. Adherence with an exercise programme is important for successful outcomes. However, there are considerable problems of adherence with formal exercise programmes. 117 Around 50% of participants commonly drop out of supervised exercise programmes, and as many as 90% of participants drop out after 1 year. 118 Even though the importance of exercise is stressed more to diabetic people than non-diabetic individuals, diabetic individuals are no more likely to exercise than non-diabetic individuals. 117 Those more likely to adhere to exercise programmes appear to be elderly individuals, females, self-referred patients and patients participating in programmes along with a spouse. 117 The type of exercise programme also affects adherence, as described below.

Physical activity and exercise patterns

It may be useful to distinguish between physical activity, which includes both activity as part of daily life (e.g. walking or cycling to work, occupational activity) and exercise, from the sorts of exercise that require participation in sports or other activities which are not part of everyday life (e.g. hillwalking, going to gyms).

If we try to categorise physical activity patterns over a lifetime, we would need quite a complicated classification, including:

1. always active – active at school and continuing throughout life, albeit with different sports at different ages

2. never active after school, and possibly not even at school

3. active in younger ages, but stopping – girls probably earlier than boys; boys may go on to play sports, such as football or rugby, into their late 20s or longer

4. active in younger age groups then stopping because of work or family commitments, but then starting again for social reasons or after retirement

5. as for (4) but starting again for health reasons

6. mixed stops and starts.

For our purposes, we could think a slow progress towards IGT over many years, with groups (2) and (3) being the most at risk.

Given that the average age at diagnosis of IGT may be in the fifties, we may be identifying a group of people who are not only inactive and overweight, but also have been so for decades. The implication of that is that unhealthy habits may be hard to change. It may be unrealistic to expect them to take up exercise, and the main thrust may have to be to try to encourage an increase in physical activity, such as walking.

This raises two issues. The first is about efficacy compared with effectiveness, with efficacy referring to results in trials (volunteers, perhaps high level of input) and effectiveness referring to results in real life. The second is whether modest increases in physical activity, such as walking, are sufficient to reduce diabetes. Is there a threshold level which must be exceeded to get an effect, or is there a continuum in which all activity has some effect? It may be that strenuous exercise is better metabolically, but modest will achieve greater uptake. A review by the Health Development Agency concluded that132 ‘Interventions that promote moderate-intensity physical activity, particularly walking, and are not facility dependent, are also associated with longer-term changes in behaviour.’

Di Loreto et al. (2003)133 addressed the issues of real-life applicability and achieving adherence in a randomised trial of a carefully designed intervention to promote physical activity in unselected people with T2DM. Having noted the benefits of exercise in diabetes, they go on to comment that133 ‘many physicians do not spend time making an effort to convince type 2 diabetic subjects to exercise, probably because older adults comply poorly with their recommendations.’

They then designed an intervention that was based on a number of factors, including motivation, self-efficacy, family support, removing impediments, enjoyment, and checking on understanding of benefits. They took care not to suggest radical increase in exercise, but used a staged approach so that people were not discouraged. Nevertheless, these small steps achieved, over time, a considerable increase in physical activity. Patients might start with only a 20-minute walk daily but would increase this at weekly intervals. From a baseline activity level of about one MET hour/week, the intervention group increased to 27 METs. The control group were given standard advice, including on exercise, but increased to only four METs/week. The target level of over 10 METs hours/week was achieved by 69% of the intervention group and 18% of the control group. One weakness in the report is that the method of randomisation was not given. Another is that results were given at 24 months, when it appeared that the intervention was continuing. It would be interesting to know if the increase continued after it was stopped.

The authors attribute success partly to their own enthusiasm for exercise – the counselling was given by physically active physicians. They also emphasise the importance of not deterring people: ‘This step-by-step approach intentionally avoided goals that the patient was unable to imagine attaining’.

Ogilvie et al. (2007)134 carried out a systematic review of interventions to promote walking but found a wide range of results, and many very short-term studies. It might have been better to exclude short-term studies. Ogilvy et al. (2007),134 noting the short durations, commented that the review might be showing proof of efficacy rather than effectiveness. But they did conclude that some interventions appeared to increase walking time by 30 to 60 minutes per week. The keys to success seemed to be targeting (usually of people motivated to try to increase activity) and tailoring. However, the authors comment that ‘Few studies in this review found unequivocal improvements in health, risk factors for disease, or even overall levels of physical activity’.

Prevalence of type 2 diabetes mellitus in South Asians

We found a number of studies on ethnic differences in prevalence. These are summarised in Appendix 1. Most studies reported higher prevalence of T2DM in South Asians and African immigrants than in Europeans, the exception being that of Davies et al. in 1999140 (30% in Asians and 34% in Caucasians, p < 0.001). The age-adjusted prevalence in South Asians ranged from 4.6% in 1985 [Mather and Keen (1985)141] to 25% in 1997 compared with 1.2% in 1985 and 6.7% in 1997 among Caucasians. 136 A more recent study by Riste et al. (2001)142 reported an age-adjusted prevalence of 33% in Pakistanis compared with 20% in Europeans. Seven times as many Asians as Europeans had been diagnosed between the ages of 30 and 54 years but similar numbers were diagnosed in those aged < 25 years. In a review, Chowdhury and Hitman (2007)143 noted the fourfold risk of T2DM in South Asians compared with white people, and that South Asians had a 1 : 3 risk of developing diabetes in their lifetimes.

South Asians were significantly younger than Europeans at first recorded diagnosis of diabetes (average of 14 years) and had significantly lower BMIs. 144 Median time to referral to hospital clinic is longer for South Asians. 144–146 Differences in fasting glucose concentration, insulin levels and insulin resistance are well advanced by adolescence according to Whincup et al. (2005),147 with a prevalence of IFG markedly higher in 13- to 16-year-old South Asians compared with European young adults (5.6 vs 1.5%; OR 3.9; 95% CI 1.4 to 10.9; p = 0.004).

Diabetes is more common in men than in women in all of the ethnic minorities except Pakistani women. 7,58,62,136,142,148,149 The Southhall study showed that males have high prevalence compared with females despite lower BMIs, which was not observed in the European population. 141 However, the prevalence of IGT seems to be higher in females than males. Simmons et al. (1991)62 reported a 9.8% prevalence of IGT in of males compared with 11.2% in females, and Cruickshank et al. (1991)149 reported a prevalence of IGT of 25% in Gujarati Indian males compared with 32% in Gujarati females.

Within South Asians there is great variation among different subgroups. In a study by Simmons et al. (1992),150 Punjabi Sikh males had a higher prevalence of diabetes than females (89/1000 in males vs 75/1000 in females), but among Hindus (Gujarati and Punjabi) and especially Muslims (Pakistani and Gujarati), prevalence was higher in women. 150 However, these differences were statistically non-significant. Gujarati Muslims had the highest prevalence of diabetes mellitus in this study, despite having a similar diet to Gujarati Hindus and Pakistani Muslims. Simmons et al. (1992)150 wondered if that could be due to either previously diagnosed diabetes or frequent consanguineous marriages in this community.

The prevalence of diabetes may be underestimated, depending on the criteria used to diagnose diabetes in South Asians. A study by Harris et al. (2000)151 assessed the impact of new ADA and WHO diagnosis criteria for diabetes on subjects from three ethnic groups (South Asians, Caucasians and those of African descent) and showed that in South Asians, overall 31/340 (9.1%) qualified for newly diagnosed diabetes mellitus using WHO criteria compared with 17/340(5.0%) by ADA criteria. 151 Overall, the proportion of individuals with impaired glucose homeostasis was 13.7% with WHO criteria (IGT + IFG) compared with 3.8% with ADA criteria (IFG). This difference was greatest for South Asians with 20.3% with impaired glucose regulation under full WHO criteria compared with 4.4% under ADA criteria. Thus, failure to identify people with IGT under the ADA criteria could underestimate the scale of the problem and be detrimental given the greater risk of CHD in people with IGT, especially South Asians.

Waist–hip ratio

Waist–hip ratio, reflecting central obesity, is higher in South Asians than Europeans. 58,142,145,149,152,153 Logistic regression analysis of univariate association between glucose tolerance and various anthropometric variables showed a stronger association with waist–hip ratio than BMI. However, the waist–hip ratios were not different among any of the South Asian subgroups. 137,152 Conversely, Yajnik et al. (2008)154 found that once adiposity had been taken into account, waist measurement did not contribute anything further. They studied insulin resistance and adiposity in groups of rural, poor urban and middle-class urban men in or near the city of Pune. The found that adiposity explained two-thirds of the differences in insulin resistance, which was commonest among the prosperous middle-class men. Half of the urban middle-class men were centrally obese. A third were overweight or obese if a BMI threshold of 25 kg/m² was used. However, that threshold may be too high for South Asians. The mean BMI in the often adipose and centrally obese middle-class men was only 23.6 kg/m².

Cholesterol

South Asians, in particular females, have lower levels of total cholesterol58,136,145,155 and high-density lipoprotein cholesterol-C (HDL-C)58,153 than Europeans. Cappuccio et al. (1997)136 reported 67% of South Asians having total cholesterol level of > 5.2 mmol/l compared with 78% of the Europeans. However, more recent studies have reported increasing levels of total cholesterol in South Asians and the difference narrowing, so that there was no significant difference in cholesterol between South Asians and Europeans. 144,152,156 Among the subgroups of Indians, McKeigue et al. (1991)58 reported high levels of total cholesterol in Sikhs (6.06 mmol/l) and the lowest levels in Gujarati Hindus (5.45 mmol/l; p < 0.001). However, the HDL-C is high among the Sikhs (1.22 mmol/l) and lowest in Muslim men (1.04 mmol/l; p < 0.001).

Diet

Cultural factors play a part in dietary habits. Grace et al. (2008)157 found that among the Bangladeshi population in Tower Hamlets, it was regarded wrong to serve healthier curries with reduced fat content to guests, because it might be seen as ‘inhospitable’. The same study noted that the community was well aware of diabetes and its risks. 157

Effects of migration on risk factors

Studies have also compared the prevalence of diabetes and its risk factors in populations from the same community who have emigrated to westernised cultures, with those who still live in their own countries. Ramaiya et al. (1995)158 reported higher prevalence of IGT (in both sexes) newly diagnosed diabetes (in women) and hypercholesterolaemia (in men) in the Asian Indian Bhatia community from Gujarat living in Tanzania, and the same community, living in the UK. A more recent study by Patel et al. (2006)159 studied the cardiovascular risk factors among Gujaratis living in Britain and compared it with the non-migrant Gujaratis in India. Although there was no significant difference in prevalence of diabetes, the most striking factor between the migrants and indigenous population was on nutrition. There was increased dietary energy intake in the migrants with significant contribution by fat intake. Serum cholesterol, triglycerides, BMI, and waist–hip ratio were all higher in the Gujarati immigrants to the UK than those in India. This illustrates the risks imposed by migration and cultural adaptation among people from the same cultural, geographic and genetic background.

Physical activity

Physical inactivity is identified as an important risk factor for diabetes and CHD. Physical activity is much lower in South Asians than in Europeans. 142,155,158,160 Fischbacher et al. (2004)160 reviewed 12 studies of levels of physical activity and fitness in South Asians (Indians, Pakistanis and Bangladeshis). Results in adults consistently showed lower levels of physical activity in South Asians than in the general population or white groups (South Asians’ activity levels were ∼ 50–75% of those of Europeans), regardless of the diverse sampling methods, mode of physical activity assessment and criteria for activity levels.

However, there were differences among the Asian groups. Fischbacher et al. (2004)160 reported that Bangladeshis had the lowest physical activity level. Similarly, the Health Survey for England7 reported that Indian, Pakistani and Bangladeshi men were 14%, 30% and 45% less likely than the general population to meet current guidelines for physical activity.

Greater differences were found in activity levels between South Asians groups and the general population in women than in men. Bangladeshi women had very low levels of physical activity compared with the general population, with only 21% achieving recommended levels of physical activity. A lower level of activity was also reported in older respondents in all ethnic groups. This difference was greater among Bangladeshi men and women and Indian women than among corresponding general population groups (13% and 18%, respectively, in Bangladeshi and Indian women aged 16–34 years compared with 1% and 2%, respectively, in the ≥ 55 years age group and 26% and 11%, respectively, in the general population). Fast and brisk walking, and participation in sports and exercise, were less commonly reported in South Asian women.

This reduced level of physical activity is in part due to cultural norms. Grace et al. (2008),157 in a qualitative study among Bangladeshi women in Tower Hamlets, found that exercise (as opposed to physical activity, such as walking) was seen as alien to the culture, and inappropriate behaviour, especially among women and older people. This was less so in later generations. The reasons included views about appropriate dress.

Comparing immigrants from the same community (the Bhatia community in Gujarat) in Tanzania and the UK, diabetes mellitus and cardiovascular risk factors except hypertension were high in people living in Tanzania compared with immigrants to the UK. The most striking difference was the levels of physical activity in the communities despite similar BMIs. Sedentary lifestyle was observed in 63–84% of Gujaratis in Tanzania compared with 26–29% in the UK. High levels of physical activity were observed in 8% of Gujaratis in Tanzania compared with 36% in the UK (p < 0.001). 158

An older study by Samanta et al. (1991)155 from Leicester noted a marked difference in activity between South Asians and white people: 8% active compared with 33% active.

Carroll et al. (2002)161 noted that the barriers to physical activity in South Asian Muslim women were culture, language, religion, age and socioeconomic status. However, a pilot scheme of ‘exercise on prescription’ suggested that these barriers could be overcome.

Smoking and alcohol

Two studies have reported that smoking and alcohol consumption levels are lower in South Asians than in Europeans,58,146 although Bhopal et al. (2004)162 noted that Bangladeshi men have high smoking rates (57%) compared with 32% in Pakistani and 14% in Indian men. A study by Chowdhury et al. (2006)156 reported a non-significant difference in smoking between South Asians and European (23.6% vs 22%; p = 0.46). Among Indians, smoking is more prevalent among Muslims and lower in Hindus and Sikhs because of their religious beliefs. 58,136

Insulin resistance

Many of the features noted above are related to insulin resistance. In his 2007 Bloom Lecture, Felix Burden (2007)163 summarises the effects of the increased insulin resistance in South Asians as:

• early development of IGT (but not so much IFG)

• more rapid transition from IGT to diabetes

• earlier onset of diabetes

• higher prevalence of diabetes.

Knight et al. (1992)164 compared male manual workers of Asian origin (64% Muslims from Pakistan and the Punjab, 31% mainly Gujarati Hindus) and non-Asian origin in two textile factories in Bradford. Diabetes was observed in 13% of Asian and 4.5% of non-Asians. 164 Insulin resistance was much commoner in the Asians. The serum insulin level at 2 hours after glucose load in Asians was double that in white people. Asians had more central obesity, but lower BMIs (24 kg/m² vs 25 kg/m²).

Inclusion and exclusion criteria

Search strategy

Electronic databases were searched for published systematic reviews, RCTs, economic evaluations and ongoing research up to September 2007. The databases searched were MEDLINE, EMBASE, The Cochrane Library, Science Citation Index, the National Research Register and the UKCRN. Appendix 2 shows the databases searched and the strategy in full. Updating searches were carried out in February 2011, mainly to identify more papers from the main studies. Auto-alerts on MEDLINE were run until September 2011. Selective updating searches in MEDLINE and EMBASE were carried out in January 2012, focusing on new cost-effectiveness analyses and recent reviews.

Abstracts returned by the search strategy were examined independently by two researchers (AS and NW) and screened for inclusion and exclusion. Disagreements were resolved by discussion and consultation with a third researcher (PR). Full texts of the identified studies were obtained. Two researchers (AS and MI) examined these independently.

Data extraction

Two reviewers (AS and MI) extracted data regarding study design and characteristics, details of the intervention, and patient characteristics and outcomes into a specially designed form. Differences in data extraction were resolved by discussion, referring back to the original paper and in consultation with PR and NW.

Quality assessment

To assess the quality of the RCTs, the following criteria were used:

1. method and description of randomisation

2. description of attrition/losses to follow-up

3. specification of eligibility criteria

4. blinding

5. power calculation

6. robustness of outcome measurements

7. similarity of group participants at baseline

8. data analysis.

Overall study quality was rated as follows: A (all quality criteria met), B (one or more of the quality criteria only partially met) or C (one or more criteria not met).

Internal validity:

• sample size

  1. – power calculation at design

• selection bias

  1. – explicit eligibility criteria

  2. – proper randomisation and allocation concealment

  3. – similarity of groups at baseline

• performance bias

  1. – similarity of treatment other than the intervention across groups

• attrition bias and intention-to-treat (ITT) analysis

  1. – all patients accounted for

  2. – number of withdrawals specified and reasons described

  3. – analysis undertaken on an ITT basis

• detection bias:

  1. – blinding

  2. – objective outcome measures

  3. – appropriate data analysis

  4. – any potential conflict of interest was noted.

Systematic reviews

Five good-quality systematic reviews were identified. These were ones that scored highly using the five quality criteria as used for the NHS Centre for Reviews and Dissemination Database of Abstracts of Reviews of Effects (DARE). Four of these met five quality criteria,38,95,179–181 and the other met four and was uncertain on one. 182 These reviews do not include all of the trials now available, nor do they include recent papers from some trials which were available.

The AHRQ published a review of the evidence on the diagnosis, prognosis and treatment of IGT and IFG in 2005, which included six trials of lifestyle interventions. 38 The authors included studies of > 6 months’ duration. As will be reported later, that seems too short because changes are often achieved with lifestyle interventions in the short term which do not persist. We preferred a minimum follow-up of 2 years. However, five of the six trials had a duration of ≥ 2 years. A meta-analysis of four trials of combined diet and exercise gave a RR of 0.54 (95% CI 0.42 to 0.70). The review concluded that there was good evidence that lifestyle interventions could reduce the risk of diabetes in people with IGT.

The Diabetes Australia Guideline covered a wider range of topics, but included consideration of the roles of physical activity and diet in reducing the risk of diabetes. 95 The guideline review concluded that exercise can slow the progression from IGT to T2DM, and that reduction in dietary fat, especially saturated fat, also reduces the risk of diabetes. It also concluded that a combined diet and exercise programme was more effective than either alone.

A review by Gillies et al. (2007),179 published during the preparation of this review, examined the evidence for both pharmacological and lifestyle interventions. 179 They noted that the trials were heterogeneous in the interventions, ethnicity, weight and age, but they carried out meta-analyses that gave the following HRs:

• diet 0.67 (95% CI 0.49 to 0.92)

• exercise 0.49 (95% CI 0.32 to 0.74)

• combined diet and exercise 0.49 (95% CI 0.40 to 0.59)

• all studies pooled 0.51.

The studies that carried most weight in their meta-analysis were the Finnish DPS183 and the US DPP,108 which will be described in detail later. The longer term results from the Finnish trial were not then available.

The Cochrane review by Norris et al. (2005)180 focused on weight loss or control. In four studies of at least 1 year’s duration, mean weight loss was 2.8 kg, and in the two studies of 2 years’ duration, it was 2.7 kg. The weight loss results from the DPP were larger (4.9 kg at 2 years) but were not included in the meta-analysis because of lack of data on distribution.

Yamaoka et al. (2005)182 reviewed trials on ‘long-term’ (but included studies of ≥ 6 months) non-pharmacological weight loss interventions in pre-diabetes. They concluded that the incidence of T2DM could be reduced by about 50%. They excluded the DPP study,108 but concluded that the meta-analysis of smaller studies matched the results from DPP.

Two Cochrane reviews have looked at separate elements of lifestyle change. Nield et al. (2008)184 set out to assess the effects of dietary advice for preventing T2DM but included only two trials, one being the Da Qing trial described later in this chapter. The other trial did not appear to be restricted to people with IGT. Nield et al. (2008)184 concluded that there was a lack of good data on prevention of T2DM by diet alone.

Orozco et al. (2008)185 assessed the effects of exercise alone or exercise and diet compared with standard advice. They concluded that the combination was effective, but not exercise alone.

Randomised trials

Nine RCTs comparing lifestyle intervention with standard care of IGT were identified: DPP,108 Kosaka et al. (2005),186 Liao et al. (2002),187 Mensink et al. (2003),188 Oldroyd et al. (2006),189 Da Qing,190 Indian DPP,173 Finnish DPS183 and Wein et al. (1999). 191

Some trials gave rise to many papers, only some of which are cited below. Where there are multiple papers from a study, we cite the website from where, in most cases, details of the protocol, copies of the papers and slides can be downloaded.

The Diabetes Prevention Program

The DPP study repository website contains full details of the DPP, including a complete protocol, results, and list of publications. 192

Description and quality of study

The DPP included 3234 participants with IGT, and compared intensive dietary and physical activity advice with standard advice. 108 The sample size necessary to achieve 90% statistical power was estimated to be 2279 participants. This was based on two main assumptions, (1) an expected conversion rate to diabetes of 6.5 per 100 person-years among participants assigned to the standard lifestyle recommendations plus placebo, and (2) that for participants assigned to intensive lifestyle or metformin intervention groups, the diabetes development HR is reduced by ≥ 33%, i.e. to < 4.33 per 100 person-years. The primary outcome was development of diabetes by ADA criteria, using an OGTT.

Participants were recruited using a variety of methods including volunteering in response to advertisements, open screening and referral by health-care providers with the aim of recruiting ≥ 50% women, ≥ 50% ethnic minority and roughly 20% aged ≥ 65 years old. Participants were randomised (stratified by clinical centre) into three treatment groups:

• intensive lifestyle intervention

• standard advice plus metformin

• standard advice plus placebo.

A fourth arm with randomisation to troglitazone was discontinued in 1998 once the toxicity of that drug was realised.

The treatment groups were similar at baseline with respect to age, sex, race, weight and BMI. Baseline physical activity was reported; one of the inclusion criteria for the DPP was being able to walk one-quarter of a mile in 10 minutes. Baseline leisure physical activity levels of the DPP participants using the modifiable activity questionnaire (MAQ) showed that women reported being less active than men (p < 0.0001), and older individuals (> 60 years of age) reported more leisure physical activity (p < 0.0001) than younger age groups.

Baseline diet in terms of total energy intake and fat consumption was reported for only the intensive lifestyle intervention group. 193 Treatment regimens were reported in detail. Staff were provided with ongoing training and provision of intervention materials. Drug administration was double blind; however, if a diagnosis of diabetes was made then participants, investigators and primary care providers were unblinded to the diagnosis and measurements. It was not reported whether any concurrent medication (apart from metformin) was taken; however, participants were excluded at screening if they were using medications known to impair glucose tolerance. Patients were assessed over a mean follow-up of 2.8 years (range 1.8 to 4.6 years) using self-reporting of diet and physical activity and using objective methods for all other measurements. Study attrition was 8% by the end of the study (92.5% had attended a scheduled visit within 5 months of the close of the study). AEs were reported according to treatment group.

Participants

The DPP trial recruited 3234 overweight participants with IGT of ≥ 7.8 to < 11.1 mmol/l (2-hour plasma glucose detected by a 75 g OGTT) and a FPG of between 5.3 and 6.9 mmol/l.

Participants came from a range of ethnic groups: 54.7% were white, 19.9% were African American, 15.7% were Hispanic, 5.3% were American Indian and 4.4% were Asian. In total, 67.7% were female and 69.4% had a family history of diabetes. The minimum age was 25 years. The mean age (± SD) of the 3234 participants was 50.6 ± 10.7 years and the mean weight was 94.2 ± 20.3kg. The minimum BMI was 24 kg/m². The majority of participants had a BMI of < 40 kg/m²; however, the BMI was ≥ 40 kg/m² in 8% of men and 21% of women. The overall mean BMI was 34 kg/m²: 30.8% had BMIs of 30 kg/m² to < 35 kg/m² and 36.9% had BMIs of ≥ 35 kg/m², so overall 67.7% had BMIs of ≥ 30 kg/m². In men, 56.5% had BMIs of ≥ 30 kg/m², and 73% of women had BMIs of ≥ 30 kg/m².

A history of and/or treatment for hypertension was present in 27% of participants. More than 37% of men and 33% of women reported a history of and/or treatment for high cholesterol. The authors noted: ‘the DPP cohort includes individuals who are more overweight and hyperinsulinaemic and less hypertensive than the subjects in other studies’. As such, ‘DPP participants may be less susceptible to hypertension-related morbid events that may confound the secondary CVD outcomes attributed to IGT or hyperglycaemia per se’. 192

One issue in interpreting trials is the representativeness of the recruits. Figure 5, below, shows the stages of recruitment to the DPP.

FIGURE 5.

Screening and recruitment for the DPP. Reprinted from Controlled Clinical Trials, volume 23, Rubin et al. , authors, The Diabetes Prevention Program: recruitment methods and numbers. pp. 157–171. Copyright (2002) with permission from Elsevier.

About 80% of the 158,177 potential participants were eliminated between steps 1 and 2. About one-third gave no reason for exclusion, but for those for whom a reason was available it was found that the five primary reasons for stopping after step 1 included (1) choosing not to have an OGTT (18%); (2) being excluded because the finger-stick glucose reading was outwith the entry criteria (17%); (3) BMI (12%); (4) being excluded because taking medications, such as thiazide diuretics, was likely to confound assessment for diabetes (7%); and (5) being given a diagnosis of diabetes (6%).

There was another large reduction (26,266 individuals) at step 2, which involved administration of an OGTT to determine glucose tolerance criteria for eligibility. About two-thirds of these were people whose fasting or 2-hour OGTT results were not in the eligible range. Many people were excluded at this stage for more than one reason. Step 3 was a 3-week run-in period; 81% (3819) who started the run-in were eventually randomised to one of the study arms.

Most of the exclusions of the initial volunteers, therefore, were based on plasma glucose results. However, many people selected themselves out, and this should be borne in mind when considering the generalisability of the results.

Intervention

In the DPP trial, the intensive lifestyle intervention group was assigned a goal of achieving and maintaining at least 7% weight loss through low-calorie, low-fat diet and moderate-intensity physical activity (at least 150 minutes/week). Participants were encouraged to achieve the weight loss (through reduction in dietary fat intake to < 25% of calories) and exercise goals within the first 24 weeks. Sixteen individual (one-to-one) sessions with a case manager (‘lifestyle coach’) within this time covered general information about diet and exercise and behaviour strategies, such as self-monitoring, goal-setting, stimulus control, problem-solving and relapse prevention training (i.e. resource intensive). During maintenance, group courses were offered every 3 months on topics related to exercise, weight loss or behaviour issues. Two comparator groups were randomised to standard lifestyle advice with either placebo (tablets twice daily) or metformin (850 mg twice daily). The standard lifestyle advice (given to all participants including the intensive lifestyle intervention group) consisted of written information and a 20- to 30-minute individual advice session recommending 5–10% weight loss and 30 minutes of physical activity 5 days a week. In addition, participants were advised to avoid excessive alcohol intake and stop smoking. Participants were reviewed annually. The average duration of intervention and follow-up was 2.8 years (range 1.8 to 4.6 years).

Results

Primary outcomes

Progression to diabetes

The DPP assessed the incidence of diabetes by annual OGTT testing or semi-annual FPG testing over 4 years. In addition, testing was prompted if symptoms arose that were suggestive of diabetes. Results were expressed in four ways: (1) the number of cases per 100 person-years; (2) per cent reduction in incidence; (3) estimated cumulative incidence at 3 years; and (4) number needed to treat (NNT) one patient with diabetes.

The number of cases per 100 person-years was significantly lower in the lifestyle intervention and metformin groups than in the placebo group (4.8, 7.8 and 11.0 in the lifestyle, metformin and placebo groups, respectively) equating to 58% (95% CI 48% to 66%) and 31% (95% CI 17% to 43%) lower incidence in the lifestyle and metformin groups, respectively, compared with the placebo group. At 3 years the estimated cumulative incidence of diabetes was 14.4%, 21.7% and 28.9% in the lifestyle, metformin and placebo groups, respectively (so they were quite a high-risk group).

The NNT to prevent one case of diabetes during a 3-year period was estimated to be 6.9 for the lifestyle intervention and 13.9 for metformin. The authors stated:

the incidence of diabetes in the placebo group was higher than expected perhaps owing to greater frequency of glucose testing or selection of persons at higher risk. 108

Subgroup analysis (note: the study had inadequate power for this analysis) found that treatment effects did not differ significantly according to sex, race or ethnic group; however, the effect of the lifestyle intervention was significantly greater among participants with lower baseline glucose concentrations 2 hours after glucose load. Similarly, the effect of the lifestyle intervention over metformin was greater in older participants and those with lower BMIs. The authors stated:

The racial and ethnic-group differences in incidence of diabetes were perhaps reduced by the selection of participants who were overweight, and had elevated fasting and post-load glucose concentrations which are three of the strongest risk factors for diabetes. 108

After adjustment for weight change in the lifestyle group, no independent effects of increased physical activity or decreased per cent fat on diabetes risk were found. 193

Several different measures of body size at baseline were predictive of the subsequent development of diabetes. 83 When analysed at 3.2 years, large WC at baseline was a better predictor of risk for developing diabetes in both sexes than other measures in the placebo and lifestyle groups. The HR for WC was 1.29 (p < 0.01) and 1.53 (p < 0.01) for women in the placebo and lifestyle groups, respectively, and 1.43 and 1.49 for men (adjusted for age and self-reported race/ethnicity) relative to smaller waists. A graded increase in the risk of developing diabetes was seen as the tertile of WC increased in both lifestyle and placebo groups.

The age of participants had an impact on progression to diabetes. 194 Diabetes incidence rates did not vary by age in the placebo group (11, 10.8 and 10.3 cases per 100 person-years in young, middle-aged and older groups, respectively, p = 0.71). In contrast, intensive lifestyle intervention was more effective with increasing age (6.3, 4.9 and 3.3 cases per 100 person-years, in the 25- to 44-year, 45- to 59-year and 60- to 85-year age groups, respectively, p = 0.007). Those in the oldest age group lost more weight and were more physically active.

Diabetic retinopathy was detected in 12.6% of the participants who developed diabetes during the DPP compared with 7.9% of those who did not (p = 0.03). The only characteristics reported to be different between those who developed retinopathy and those who did not were HbA_1c level (6.4% ± 0.55% vs 6.2% ± 0.63%; p < 0.05 – as reported in the paper, although the difference of only 0.2% seems trivial) and systolic blood pressure (SBP) (128 mmHg vs 125 mmHg; p < 0.05). 195

Regression to normal glucose values

The DPP measured FPG and HbA_1c levels in all participants every 6 months for up to 4 years. Both FPG and HbA_1c levels changed significantly over the time of the study. In the first year, the FPG of the lifestyle and metformin groups decreased; however, in subsequent years the FPG values increased and returned to baseline levels by 2.5 years; further increases were observed thereafter up to 4 years (significant difference between groups; p < 0.001). In the placebo group, FPG values increased at each time point from baseline up to 4 years (significant difference between groups 0.5–3 years; p < 0.001). HbA_1c values showed a similar initial decrease in the lifestyle and metformin groups; however, both groups showed an increase thereafter with the metformin group lying between lifestyle and placebo values, as shown in Figure 6. At the start, the HbA_1c level was 5.9%. At 4 years, the means were, approximately, placebo 6.1%, lifestyle 6.0% and metformin 6.0%.

FIGURE 6.

Mean change in HbA1c level vs years from randomisation in the DPP. Reprinted with permission from the Massachusetts Medical Society. The Diabetes Prevention Program Research Group. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N Engl J Med 2002;346:393–403. 108

The placebo group saw a constant increase in values from baseline to 4 years. The percentage of participants with normal glucose concentrations was greater at all time points up to 4 years in the lifestyle intervention group compared with both metformin and placebo groups. At 4 years, the percentages with normal fasting glucose were lifestyle 54.1%, metformin 45.1% and placebo 43.8%. The percentages with normal 2-hour glucose were lifestyle 37.8%, metformin 27.9% and placebo 24.2%. The percentages with both normal fasting and 2-hour glucose levels were lifestyle 29.6%, metformin 19.7% and placebo 17.2%. There were no differences in the percentages reverting to NGT among the age groups. 164

The 10-year results showed continuing but reduced benefit, with the cumulative incidence of diabetes reduced by 34% in the lifestyle group (and by 18% in the metformin group). This was despite all groups being offered a modified form of the original intervention (in groups rather than 1 : 1) after the end of the trial, and despite the metformin group being allowed to continue on the drug.

Adverse events

Gastrointestinal symptoms were less frequent in the lifestyle group than in the placebo group (12.9/100 person-years vs 30.7/100 person-years, p < 0.02), whereas the rate of GI symptoms was significantly higher in the metformin group than in the placebo group (77.8/100 person-years vs 30.7/100, p < 0.02). The rate of musculoskeletal symptoms was significantly higher in the lifestyle group (24.1/100 person-years) than in both the metformin and placebo groups (20.0 and 21.1/100 person-years, respectively). No significant differences between groups were seen in hospitalisation or deaths.

Secondary outcomes

Weight loss

Data on weight change were available for 3085 participants at 6 months, 3064 participants at 1 year, 2887 at 2 years and 1510 at 3 years. Participants assigned to the lifestyle intervention had significantly greater weight loss than participants assigned to receive metformin or placebo. Over 4 years the average weight loss was 5.6 kg, 2.1 kg and 0.1 kg in the lifestyle, metformin and placebo groups, respectively (p < 0.001).

Weight loss, reduction in WC and percentage of participants who achieved the 7% weight loss goal all increased with age. 194 The percentages of participants who achieved the weight loss goal (7%) of body weight were 25–44 years = 33%, 45–59 years = 39% and 60–85 years = 55%. Participants aged 60–85 years had the most weight loss.

At the 10-year follow up, the lifestyle group had regained most of the weight, going from a nadir of a mean 7 kg reduction at 1 year back up to a mean weight loss of 2 kg.

What makes the difference?

One of the papers from the DPP explored the relative contributions of the changes in weight and exercise achieved by studying the intensive lifestyle group. 193 Dietary and exercise data were collected by questionnaires. The number of participants with ≥ 3-year or longer follow-up was only 638 because recruitment went on until May 1999 and the data were collected at end of July 2001.

Hamman et al. (2006)193 reported that the mean weight loss at 3 years was 4.1 kg. Physical activity increased at each year end compared with baseline. The percentage of calorie intake from fat fell from 34% at baseline to 28% at the end of year 1. A total of 153 participants developed diabetes – a rate of 5 per 100 person-years. When all of the changes were examined in a multivariate Cox model, weight loss was the dominant factor. Even small amounts of weight loss helped – on average there was a 16% reduction in the risk of diabetes per kilogram of weight loss. Weight loss reduced diabetes incidence similarly across all race/ethnicity groups, for both sexes, for all ages, and for several levels of physical activity, regardless of the level of initial obesity.

Hamman et al. (2006)193 then looked at the effects of meeting the various goals at year 1, dividing participants into eight subgroups. The subgroup that met all of the goals had the lowest risk of diabetes, with an 89% reduction in risk (HR 0.11, 95% CI 0.05 to 0.24; p < 0.0001) compared with the group meeting none of the goals. Weight loss was responsible for much of the reduction, but there was halving of the incidence of diabetes in those who met the physical activity goal (150 minutes of moderate-intensity activity per week) as shown in the comparison of subgroups 3 and 4 compared with 1 and 2. The reduction was 46% after adjusting for baseline variables and 44% (HR 0.56; 95% CI 0.36 to 0.89; p = 0.012) after adjusting for weight loss over follow-up. There is of course an interaction, in that increased physical activity helped to sustain weight loss. However, among 495 recruits who did not achieve the 12-month weight loss target (7% or 6.6 kg on average), the incidence of diabetes was reduced by 44% in those who achieved the physical activity target. Note that this group did lose some weight (mean 2.6 kg) but adjustment for this had little impact on the effect of physical activity.

The results may also indicate that at least ∼150 minutes per week of moderate activity are required before an effect on diabetes risk is achieved.

Adherence

Dietary intake

DPP examined the adherence of participants to intervention. Diet adherence was assessed at 1 year only and showed that the lifestyle group had a significantly greater reduction in daily energy intake over the first year than metformin and placebo groups (–450, –296 and –249 kcal, respectively; p < 0.001). This was associated with a significantly greater change in the average intake of fat from baseline to 1 year (–6.6%, –0.8% and –0.8% of total calories, respectively; p < 0.01). Data for longer time points were not reported. There was no significant difference in reduction of reported caloric intake among different age groups, although older (60–85 years) recruits reduced calorie intake less (by 10% at 12 months) than younger recruits (25–44 years; 18%). 194

Physical activity

Compared with baseline, all study groups had increased their physical activity at 1, 2, 3 and 4 years. However, participants in the lifestyle intervention group had significantly greater change in physical activity at all time points than both the metformin and placebo groups, the physical activity of which increased only slightly over baseline values. 108 The authors stated:

After adjustment for weight change no independent effect of increased physical activity or decreased per cent fat on diabetes risk was found. 193

Paradoxically, those participants who met the physical activity goal of 150 minutes/week of moderate activity had a 44% reduction in diabetes risk, independent of weight loss.

In a representative sample of DPP lifestyle participants (n = 274; 94% of the final 293 lifestyle participants randomised), characteristics that correlated with high levels of baseline, 1-year and end-of-study physical activity were (1) being a man and (2) having lower BMI and lower perceived stress, depression and anxiety scores at baseline. Higher baseline BMI and being a woman correlated with lower baseline, 1-year and end-of-study physical activity levels, with women having significantly higher BMIs and higher levels of depression, anxiety and perceived stress than men. 196

Older age was an independent predictor of achieving the goal of 150 minutes of physical activity at 1 year and 2 years. Lifestyle participants aged > 60 years achieved greater minutes of physical activity and greater per cent weight loss and greater risk reductions for developing diabetes (71% risk reduction compared with 48% risk reduction in persons aged 25–44 years). Higher levels of baseline physical activity correlated with greater readiness to change physical activity levels (p < 0.0001) and lower levels of perceived stress (p = 0.009), depression (p < 0.003), and anxiety (p = 0.03) at baseline, 1-year and end-of study levels. 196

Perhaps time is a factor, with those who are older having more time post retirement. The older age groups had the highest MET-hours of activity and did better with lifestyle change than younger groups, although it should be noted that they started leaner and with better insulin sensitivity. 194

Metabolic syndrome

Recruitment to DPP was on the basis of IGT, but a later paper by Orchard et al. (2005)197 reported the prevalence of metabolic syndrome at baseline and the effects of the interventions.

The metabolic syndrome was defined by the criteria from the National Cholesterol Education Program’s Adult Treatment Panel III, with three or more of the following:

• WC of > 102 cm in men and > 88 cm in women

• serum triglyceride level of at least 1.7 mmol/l

• HDL-C level of < 1.03 mmol/l in men and < 1.3 mmol/l in women

• blood pressure of ≥ 130/85 mmHg

• FPG level of 6.2 mmol/l.

The metabolic syndrome was present in 53% (n = 1711) of the 3234 participants at baseline, with little variation by age. After 3 years, the prevalence of the metabolic syndrome increased from 55% at baseline to 61% in the placebo group (p = 0.003), did not change (54–55%) in the metformin group (p > 0.2) and decreased from 51% to 43% (p < 0.001) in the lifestyle group. The decrease in the lifestyle group correlated most strongly with decreases in WC and blood pressure.

Fujimoto et al. (2007)198 examined the relationship between changes in body fat and progression to diabetes by 1 year. At baseline, recruits were mainly obese, with mean BMI of 32.1 kg/m² in men and 33.0 kg/m² in women. Visceral and subcutaneous fat was measured by CT, as well as by standard measurements such as BMI. The lifestyle group had big reductions in both visceral fat (reduced at L2–3 by 24% in men and 18% in women at 1 year) and subcutaneous fat (again at L2–3, reduced by 16% in men and 11% in women). Progression to diabetes was associated with fat changes differently in the arms of the study. In the metformin group, the reduced diabetes risk was independent of body fat changes. In the placebo group, only the subcutaneous fat changes correlated with diabetes risk, and then only in men. In the lifestyle group, all of the fat variables correlated with diabetes reduction in men. In women, weight, BMI and WC were significant predictors but the association with visceral fat did not quite reach statistical significance.

After the end of the randomised trial, all three groups (lifestyle, metformin and placebo) were offered the lifestyle intervention, albeit in groups rather than the original individualised provision. At the 10-year follow-up, the original lifestyle group had regained about 5 kg of their original (by 12 months) 7 kg weight loss, so that by 10 years their weight was little different from the metformin group, who lost about 2.5 kg. 199 However, the incidence of diabetes by 10 years remained lower in the former lifestyle (by 34%) group and metformin (by 18%) group than in the former placebo group. The onset of T2DM was delayed by about 4 years with lifestyle and by about 2 years by metformin. One feature of note was that not only was the best effect of lifestyle seen in the 60- to 85-year age group, but also that this group had no significant response to metformin. 199

Perreault et al. (2009)200 noted that in some diabetes prevention trials, glucose tolerance regressed to normal, and used DPP data to examine the factors that were associated with this. The factors included baseline evidence indicating a milder condition (lower baseline FPG and 2-hour plasma glucose) and those indicating response to the intervention, especially greater weight loss. As reported by Hamman et al. (2006)193 for every 1 kg weight loss there was a 16% reduction in diabetes risk, and it was the most important predictor of regression. 193 However, Perreault et al. (2009)200 noted that intensive lifestyle interventions appeared to have other components that were independent of weight loss, most probably physical activity.

In another paper, Perreault et al. (2008)201 noted that meeting the goals of the lifestyle intervention was a strong predictor of reduction in progression to diabetes. However, men in the lifestyle arm met more goals than women but had the same progression to diabetes. Perreault et al. (2009)200 explain this on the basis that men were at higher risk from baseline.

The DPP also examined changes in cardiovascular risk factors over time, and found that among those who progressed to diabetes, there were rises in blood pressure and triglycerides, and a fall in HDL-C. 202 These changes were statistically significant, although quite small. Those who regressed from IGT to NGT showed reductions in blood pressure and triglycerides, and improvements in HDL-C levels. These improvements in HDL-C levels were more marked in the lifestyle group than in the metformin one, and LDL cholesterol (LDL-C) also fell in the lifestyle group. Goldberg et al. estimated that in the lifestyle group these changes should bring about a 10–13% reduction in heart disease.

The effect of metformin has been examined in several of the DPP papers. In the main study report it was noted that metformin reduced the risk of diabetes by 31%. 108 However, the effects varied among groups. Metformin had little effect in the oldest age group, whereas lifestyle change was more effective with increasing age. 194 Crandall et al. (2006)194 suggest that this may be related to the oldest age group (60–85 years) being leaner, and having less insulin resistance, at baseline. Metformin may work by reducing insulin resistance and the older group will have less to gain. Another DPP paper reported that metformin was effective mainly in those with BMIs of > 35 kg/m². In this BMI subgroup, metformin was as effective as lifestyle. 83

Some of the benefits of metformin were the result of weight loss rather than improvements in insulin sensitivity. 203 Lachin et al. (2007)203 estimated that 64% of the metformin benefit was mediated through weight loss. Interestingly, they note that the effect of weight loss achieved with metformin appears to be less than the same weight lost by lifestyle change.

Most DPP papers report prevention of diabetes, but the 10-year follow-up199 also reported that diabetes was delayed, with the time at which 40% of the (high-risk) groups progressed to diabetes being 4 years later with lifestyle and 2 years later with metformin compared with placebo.

Ackermann et al. (2009)204 estimated the direct utility of weight loss, after adjusting for a range of possible confounding factors. They used the SF-6D utility measure, and estimated that there was a gain in utility of 0.007 (on a scale of 0.29–1.00) for every 5 kg of weight loss. Although this fell below the minimum effect size that was deemed to be clinically important (0.04), they considered that the change could improve cost-effectiveness if sustained for years.

In summary, DPP found that intensive diet and exercise intervention in those with IGT significantly reduced the incidence of diabetes by 58% compared with standard lifestyle advice. The reduction was sustained, with a cumulative incidence of diabetes at 6 years of 23% in the intervention group, compared with 38% in the control groups. At 10 years, the incidence of diabetes was 34% lower in the lifestyle group and 18% lower in the metformin group compared with the control group. This was despite the lifestyle group regaining much of the 7 kg weight lost in the first year.

Inclusion and exclusion criteria

To be included, studies had to assess the long-term costs and consequences of treating people with lifestyle interventions to prevent progression to diabetes. Studies that only assessed short-term costs and consequences were excluded because many of the benefits and cost savings that may be associated with diabetes prevention efforts are likely to occur over a long time horizon. Studies reported in languages other than English were not included.

Data extraction

Data were extracted from the long-term modelling studies using headings that were consistent with those used in the previous HTA review on screening for diabetes:72

1. author and year

2. decision problem (comparators, population, setting, objectives)

3. cohort information (characteristics and numbers)

4. model structure, perspective and scope (basic structure and assumptions)

5. modelling of disease progression (details of structure and assumptions used for modelling diabetes progression)

6. modelling of diabetes complications (details of structure and assumptions used to model progression of diabetes complications)

7. mortality (details of how mortality was modelled)

8. costs (details of costs considered in the model)

9. outcomes (outcomes reported and methods for calculating life-years or QALYs)

10. findings (reported results of the base-case analysis)

11. sensitivity analysis (details and results of any sensitivity analysis conducted).

Quality assessment of included studies

One reviewer critically appraised identified studies using the BMJ guidelines for reviewers of economic evaluations240 in conjunction with a checklist for good practice in decision-analytic modelling in HTA. 241 When the individual reviewer was uncertain about the appropriateness of methods or assumptions used in any included study, the opinions of two further reviewers were sought.

Decision problem

As in previous studies,243,244 Eddy et al. (2005)248 assessed the cost-effectiveness of providing either the DPP lifestyle or metformin intervention for people with IGT compared with a baseline strategy of no treatment other than dietary advice. They assumed individuals would enter into an intensive management programme if their HbA_1c levels rose to > 7%. However, Eddy et al. (2005)248 also included an alternative strategy in which the lifestyle intervention described in the DPP trial commenced at time of onset of diabetes rather than prior to onset. This strategy is a feasible option for people who have been identified as having IGT/IFG, as their progression to diabetes would be closely monitored.

Cohort information

As in previous studies, Eddy et al. (2005)248 analysed the costs and effects of the alternative management strategies for a cohort with the characteristics of those enrolled in DPP trial – 10,000 people who would meet the inclusion criteria for this trial were simulated to receive each alternative and followed over a period of 30 years.

Model structure, perspective and scope

The type of model developed and used by Eddy et al. (2005,248 2003249) is very different to the Markov models previously used to address this issue. It is very complex and not widely used in health-care decision modelling. It was therefore difficult to appraise the model using the published guidelines for good practice in decision-analytic modelling that were used in the previous review. 241 However, a number of validations have been performed in which the model has been used to simulate existing randomised trials. 252 The results of these provide a high degree of confidence in the model’s ability to predict clinical outcomes over the short to medium term. Very briefly, the model uses object-oriented programming and relies on a large number of differential equations to model the physiological mechanisms and interactions that underlie the development and progression of diabetes and its complications. There are no discrete disease states or time cycles in the model, rather the underlying biological variables are continually interacting in a very large number of possible combinations, and these variables can give rise to clinical events at any point in time. The authors provide a detailed description of how some variables and equations were derived from empirical sources in an appendix and separate technical paper. 249,253

In their analysis relating to the prevention of diabetes, Eddy et al. (2005)248 present annualised rates of change in key biological variables and the cumulative incidence of important clinical events. As mentioned above, the costs and effects of the alternative treatment strategies were tracked over a time horizon of 30 years, somewhat shorter than the 70-year time horizon used by Herman et al. (2005)244 in a previous analysis. Like Herman et al. (2005),244 Eddy et al. (2005)248 apply utility decrements to events that impact upon QoL, and estimate the cost per QALY for the alternative strategies. Eddy et al. (2005)248 also assess the problem from three different perspectives: the individual patient (in terms of individual risk), a 100,000-member health plan, and society as a whole (i.e. for the entire US IGT/IFG population).

Modelling diabetes progression

As already mentioned, Eddy et al. ’s model (2005)248 differs substantially from the Markov models used in previous analyses. Markov models consist of a limited number of discrete disease states. Individual patients or cohorts are simulated to transit between these based on annual transition probabilities calculated from epidemiological or randomised studies. The more complex of the Markov models used in previous analyses of diabetes prevention allow key biological variables, such as HbA_1c level, to mirror annual changes observed over time in several diabetes trials. The modelled HbA_1c levels in turn influence the probabilities of developing micro- and macrovascular complications as represented by several discrete state submodels. The model of Eddy et al. (2005)248 on the other hand is much more complex allowing multiple interactions between many continuous biological variables and clinical events. For example, the development of diabetes is influenced by age, sex, race/ethnicity, BMI, and a factor that registers the effect of glucose intolerance. The authors explain how a host of variables – representing FPG in people without diabetes, hepatic glucose production, the effect of insulin resistance, the amount of insulin produced, the efficiency with which the body uses insulin, and patient characteristics (age, sex, race/ethnicity, BMI) are used to model the physiological mechanisms underlying rising FPG levels and the development of diabetes. 249

Diabetes complications

Eddy et al. (2005)248 include coronary artery disease (MI and congestive heart failure), stroke, nephropathy, neuropathy and retinopathy in their model. Coronary artery disease is modelled through the occurrence of coronary artery occlusions that occur as a result of atherosclerotic plaque formation and rupture, and/or the development of occlusive thrombi. Many underlying biological variables are used to determine the progression and timing of these events through sets of differential equations. The authors state that stroke is modelled in a similar way. The progression of nephropathy is modelled as a function of the person’s FPG, blood pressure, and glycaemic load (a variable that represents the degree of elevation in FPG combined with the duration of time it has been elevated to different degrees). Retinopathy is modelled in a similar way to nephropathy. Here, it is the clinical manifestations of retinopathy (microaneurysms, haemorrhages and exudates) that are modelled as a function of a person’s FPG, blood pressure, and glycaemic load. Finally, diabetic neuropathy (loss of sensation) is modelled as a function of a person’s FPG, blood pressure and glycaemic load. Diabetic foot ulcers and amputations are further functions of the neuropathy feature.

Mortality

Eddy et al. (2005)248 state that they model death from diabetes and its complications based on the extent of failure in the underlying physiological systems:

A person in the model will die if a coronary artery is occluded and the subsequent infarction reduces their myocardial function to the point that cardiac output and blood pressure cannot be maintained. 248

Deaths from other causes are also included in the model.

Resource use and costs

The direct and non-direct medical costs associated with the interventions to prevent progression to diabetes, were taken as those collected alongside the DPP trial. 108 Eddy et al. (2005)248 also state that they included a set of algorithms in their model that describe the treatment processes providers follow for patients with a complete range of clinical circumstances. The model also includes system resources, and use of these resources is triggered whenever a patient encounters the health system as a result of a clinical event or treatment process being implemented. The model calculates cost by tracking all resource use and adding up the cost for each use. The itemised costs used in the model come from a detailed micro-costing survey of an integrated managed care organisation that provides comprehensive care for people with diabetes with no deductibles or co-payments. All costs reported in the analysis are in US dollars (US$) for the year 2000. This additive costing approach used by Eddy et al. (2005)248 is somewhat different to Herman et al. ’s (2005)244 approach of using a multiplicative costing model. It may be more accurate, as Herman et al. (2005),244 to some extent, rely on assumptions to estimate the multipliers for their model.

Outcomes

In terms of outcomes used to establish cost-effectiveness, Eddy et al. (2005)248 estimated QALYs using utility weights reported in the DPP trial108 for people with IGT. For people with diabetes they used weights from another published survey. Both surveys used the Quality of Well-Being Index to measure utility decrements associated with IGT, diabetes and complications. These are the same surveys used by Herman et al. (2005)244 to estimate QALYs in their model of diabetes prevention. The QoL decrements were assumed to be additive for people with more than one complication.

Findings

Eddy et al. (2005)248 estimated that, from an individual perspective, the intensive lifestyle intervention of the DPP108 would reduce the 30-year probability of a person with DPP characteristics (see Table 7) developing diabetes by 11 percentage points compared with the baseline strategy (from 72% to 61%). From the perspective of a 100,000-member health plan, with 4% of its population at high risk of diabetes, the model predicted that the intensive lifestyle intervention would reduce the number of people developing diabetes by 434 compared with the baseline strategy (2887–2453). The expected value of the cost per QALY came to US$143,000. The cost per QALY was higher over shorter time horizons. From the societal perspective (i.e. providing a national DPP lifestyle programme to all high risk people in the USA), the lifestyle intervention cost about US$62,000 per QALY compared with the baseline strategy. However, when the authors assessed the cost-effectiveness of implementing the intensive lifestyle intervention only when individuals actually developed diabetes, they predicted this strategy would result in a lower 30-year cost per QALY of US$24,500 compared with the control group. Using this as the reference strategy, the incremental cost per QALY for the DPP strategy increased to US$202,000. But as people with IGT are at about 1.8 times the risk of heart disease, this approach would allow some to die – those who had MIs before being diagnosed as diabetic. It is unclear whether or not this increased risk is captured by Eddy et al. ’s model. Using this as the reference strategy, the incremental cost per QALY for the DPP strategy increased to US$202,000. Thus, Eddy et al. ’s243,244 findings are much less favourable than those of previous analyses that assessed the cost-effectiveness of the DPP interventions.

Sensitivity analysis

Two approaches have been used to assess the validity and robustness of the Archimedes Diabetes Model249 and its findings relating to the diabetes prevention interventions. First of all, the model has been validated by simulating 19 clinical trials, including the DPP,108 and comparing the results with those found in the actual trials. This provides a high degree of confidence in the validity of the model over short to medium time horizons. However, uncertainty remains over the ability of the model to accurately predict outcomes beyond the follow-up period of available clinical trials. To further address uncertainty relating to the values of variables used in the model, the authors state that distributions were assigned to variables and individuals were created by simultaneously drawing a value for each variable from each distribution. To explore the uncertainty relating to the effectiveness of the DPP intervention, the authors incorporated a distribution for this parameter based on the 95% CI reported in the original trial. They observed that this source of uncertainty substantially increased the width of the CI for the 30-year chance of developing diabetes. However, the assumption of a linear increase in diabetes prevalence does not fit with the data from the GPRD (presented later) and it is more likely that most of those who will progress to diabetes will do so within 10 years, and so the Eddy model243,244 will underestimate person-years of diabetes. Based on their simulations, the authors present a probability distribution of the cost per QALY for the lifestyle intervention and conclude that it is extremely unlikely (< 0.1% chance) that the cost per QALY would fall below the threshold of US$50,000 for a 100,000-member health plan. The incremental cost per QALY was also found to be sensitive to the discount rate, and most notably, the cost of the lifestyle intervention. If, for example, the cost of delivering the DPP lifestyle intervention could be reduced to US$217 per person per year, without reducing its effectiveness, then the cost per QALY ratio would be much more favourable.

Gillies 2008

In a very thorough analysis, Gillies et al. (2008)265 assessed four screening and prevention strategies:

• no screening

• screening for T2DM

• screening for T2DM and IGT, with lifestyle intervention

• screening for diabetes and IGT, with pharmacological intervention.

Hence, their decision problem was broader than that in this review, which starts with people already diagnosed. Gillies et al. (2008)265 studied the whole pathway from screening to death. Their main aim was to examine whether screening should be for T2DM alone, or for diabetes and IGT. However, the latter pathway will be more effective than screening only for diabetes alone, if intervening in those with IGT is cost-effective.

The base-case scenario was one-off screening at the age of 45 years, in people with above-average risk of diabetes. Risk was based on known history of ischaemic heart disease, hypertension, dyslipidaemia, family history of T2DM, and BMI score of > 25kg/m². They base their effectiveness estimates for the prevention of diabetes on a previous review of pharmacological and non-pharmacological interventions of their own. 179

The modelling was done using a mixed decision tree and Markov modelling approach. Because their costs per QALY are for the whole pathway they are not comparable with ones in this review, but the key finding for our purposes was that screening for both diabetes and IGT, with lifestyle intervention in the latter, was more cost-effective than screening for diabetes alone, with costs per QALY of £6242 and £14,150, respectively, both compared with no screening. They assumed that lifestyle intervention would cost about £400 in the first year, and £280 per year thereafter. These costs were derived from a review of interventions in obesity266 rather than from the diabetes trials.

Irvine et al. (2011)267 carried out a cost-effectiveness analysis based on the University of East Anglia Impaired Fasting Glucose programme. This programme screened almost 4000 high-risk individuals and identified 209 with IFG. Of these, 85% entered the lifestyle intervention trial, in which they were randomised to the intervention (dietary advice and physiotherapist-led exercise groups, the aim being to achieve 7% weight loss over 6 months) or to the control group, which received a 2-hour session of diet and exercise advice. Costs were assessed for the intervention and control arms. Recruits completed European Quality of Life-5 Dimensions (EQ-5D) questionnaires and ICERs were calculated.

However, there were major weaknesses in this study. First, follow-up was short, a mean of around 7 months. Second, and more importantly, benefits were assessed only during the trial, and ICERs were therefore very high because the duration of benefit was short. The intervention was assessed as having an ICER of £67,163 per QALY. When a subgroup of participants (number not given) with longer follow-up was analysed, the cost per QALY fell to £17,075. No modelling of the long-term effects of weight loss was carried out.

Attempts are being made to adapt the DPP for delivery by community organisations. Ackermann and Marrero (2007)268 are collaborating with the YMCA of Greater Indianapolis to evaluate the effectiveness, adoption, implementation, and maintenance of a modified DPP lifestyle intervention for delivery in YMCA branch facilities. The intervention is called the Group-Organized YMCA Diabetes Prevention Program (GO-YDPP). The GO-YDPP core curriculum involves the same 16-lesson approach as the original DPP, but lessons are delivered in groups of 10 to 12 participants and are held over just 16 weeks. GO-YDPP has retained the same physical activity and weight loss goals of the original DPP lifestyle intervention. The YMCA has estimated the total costs during the first year to be US$275 to US$325 per participant. This compares favourably with the original DPP intervention, estimated to cost more than US$1400 per participant.

Ackermann et al. (2009)204 used data from the DPP to assess the utility benefit from weight loss. Data on QoL were collected using SF-36, from the start of the trial, in 3206 subjects. From this, the DPP investigators derived the SF-6D, and assessed the effects of weight loss on that, at 12 and 24 months, adjusted for other variables. SF-6D gives a range of values from 0.29 to 1.0, with 1.0 being best possible. Mean BMI among DPP patients was 34 kg/m² and mean weight was 94 kg. The lifestyle group lost an average 6.7 kg by 1 year. Their SF-6D scores rose (improved) by 0.010 for every 5 kg of weight lost. Meaningful improvements in health-related QoL appeared only once subjects had lost at least 5–10 kg. The DPP investigators noted that, if weight loss is sustained for years, the small increase in utility would become significant in cost-effectiveness analysis.

Bertram et al. (2010)269 populated a microsimulation model with data from several sources, including AusDiab (Australian Diabetes Obesity and Lifestyle Study). In AusDiab, people were screened at the age of > 55 years, or at age > 45 years if they had risk factors such as hypertension, family history of diabetes of a high BMI, or if they were deemed to be high risk, such as having had previous GDM. The model had three lifestyle interventions: diet alone, exercise alone or both. It also had three pharmacological interventions, including metformin. The effectiveness of interventions was based on the meta-analysis by Gillies et al. (2007)179 from the UK. The modelling showed that the most cost-effective interventions were diet and exercise combination, and metformin. Cost-effectiveness was taken to be an ICER of < AU$50,000 (AU$, Australian dollars) per disability-adjusted life-year (DALY). Adding metformin to diet and exercise was not cost-effective because the cost per DALY rose to around AU$81,000, though with very wide CI (14,000 to 130,000). Exercise alone and diet alone had ICERs of AU$30,000 and AU$38,000 but the upper confidence limits were well above affordable levels, whereas the upper confidence limit for the combination was AU$35,000. A number of assumptions were made, including that results in general population would be as good as in trials such as DPP and DPS.

Another Australian study by Colagiuri and Walker (2008)270 used a different model – the Diabetes Cost–benefit Model – to assess the cost–utility of screening for undiagnosed diabetes. However, although the main analysis was on diabetes, some results are provided for detecting and intervening in IGT. The model uses AusDiab data for two groups: all of those aged 55–74 years, and those aged 45–54 years but with one or more risk factors, such as obesity, family history of diabetes, or hypertension. Those with IGT would be offered a lifestyle programme estimated to cost AU$500 (but not specified, so technically its effectiveness is unproven) but they also assessed the effect of using a more costly AU$1000 intervention based on the DPS lifestyle arm. They used the DPP and DPS results of a 58% reduction in progression to diabetes but rounded it to 60%. They estimated savings in health-care costs based on the UK Prospective Diabetes Study (UKPDS) 61 paper271 (which showed that diagnosis at lower PG had better outcomes and hence lower costs – Colaguiri was first author of that paper). They pessimistically assumed that only about half of those offered the lifestyle intervention would accept it, and that about one-third of those who did would thereby avoid progression to diabetes. The cost per DALY avoided is around AU$50,000, which is considered borderline acceptable. Limiting screening and intervention to the over-55s would make it more cost-effective (ICER AU$48,000).

Johansson et al. (2009)272 assessed the cost-effectiveness of the Stockholm Diabetes Prevention Program, but this was not restricted to people with IGT and so not relevant to this review. The results were mixed: in some groups the programme appeared cost-effective, in others not.

Li et al. (2010)273 reviewed cost-effectiveness studies of prevention of diabetes. Most were in people with IGT, and most have been included in this review. Some were not in people with IGT. Li et al. (2010)273 conclude that intensive lifestyle interventions to prevent diabetes in people with IGT are ‘very cost-effective’. Most of the included studies were based on the DPP results.

Zhuo et al. (2012)274 reported that in the USA, a nationwide community-based lifestyle programme could break even within 14 years. However, the programme they used was a hypothetical one, based on the small pilot study by Ackermann et al. (2008),275 whereas the reduction in diabetes used in their modelling was based on the results from DPP and DPS.

Data source

The GPRD is one of the largest longitudinal primary care databases, and provides anonymised data for research. It was established in the late 1980s and, since then, more than four million residents of the UK have registered with GP practices contributing data. General practices from across England, Wales, Northern Ireland and, to a lesser extent, Scotland voluntarily contribute anonymised patient data to the research database. GPRD covers approximately 6% of the UK population. Age and sex structures have been shown to closely reflect the UK population as a whole. 276–282 The database is currently held and maintained by the UK DH.

Details of patient characteristics, prescribed treatments and clinical diagnoses are recorded in the database. Clinical diagnoses are recorded using two disease coding systems. Until the late 1990s (with a gradual replacement), diagnoses were recorded using a modified version of the Oxford Medical Information System (OXMIS). Later, this was replaced by the READ coding system, now in widespread use in the UK.

The GPRD has been used extensively for research in drug therapy and outcomes and people with diabetes mellitus have been studied previously. The estimated prevalence of diabetes has been reported to be in keeping with other studies and national surveys. There was evidence of over recording of codes related to diabetes in children and young adults but, in older age groups, the recording was similar to national survey data. 283

Study population

For this study, we used data from approximately 300 GP practices contributing to the GPRD that have been well validated and used in research previously. 284 We identified a study population of all patients who were registered with a GPRD practice at any time between 1 January 2000 and April 2005, and with a minimum of 6 months of follow-up in the database. People with < 6 months’ follow-up were excluded.

Identifying people with impaired glucose tolerance/impaired fasting glucose

To identify people recorded as having a diagnosis of IGT/IFG (cases), we searched the computer records of each person in the study population for any of the codes in Table 8. IGT/IFG diagnosis was based on the reported measurement of glucose in blood or urine following glucose loading or from random samples, or specific codes that suggested IGT/IFG. The date of the first recording of one of these codes was taken as the ‘index date’. People with a diagnosis of diabetes mellitus, or a prescription for diabetes mellitus treatment recorded before the index date, were excluded.

To validate the appropriateness of the codes selected, we undertook a pilot study where we reviewed a random sample of 20 complete computer patient records from among those with one of the IGT/IFG codes provided in Table 8. From this, it was apparent that GPs often used these codes as initial consultation codes and that they were soon upgraded to a diagnosis of diabetes mellitus after appropriate diagnostic tests. We therefore classified potential cases of IGT/IFG as follows:

• IGT/IFG cases Those with no subsequent diabetes mellitus diagnosis for at least 1 month after the index date.

• Excluded Those with a subsequent diabetes mellitus diagnosis < 1 month after the index date.

For IGT/IFG cases, we examined their complete computerised records to obtain details of characteristics: age, sex and BMI. For estimation of BMI, we used the last recording of weight before the index date. We searched patients’ records for evidence of hypertension, CVD, stroke or obesity – present either before or after the diagnosis of IGT/IFG. We followed the IGT/IFG patients’ records prospectively, from the index date, to ascertain if they subsequently developed diabetes mellitus. Finally, we sought evidence of prescribing of metformin or glitazones to treat IGT/IFG and whether any cases were already receiving medicines important in reducing risk of vascular disease (i.e. statins or aspirin). This was done because it is likely that the diagnosis of IGT or IFG, both of which are associated with an increased risk of heart disease, might trigger the use of statins, and we wanted to know the extent to which such patients were already on a statin. If a high proportion was already being treated because of other risk factors, the benefits of screening for and treatment of IGT would be reduced.

Using a further random sample of 50 full computerised records, two researchers checked the data extraction for the above variables against the original record. This was found to be satisfactory with no errors in the data extracted.

Analysis

A descriptive analysis of those diagnosed with IGT/IFG was undertaken, reporting the proportion by sex, age and BMI. Prevalence rates for IGT/IFG were reported per 100,000 persons. The proportion and timing of progression to diabetes mellitus was estimated. Analysis was conducted using Stata version 9.2 (StataCorp LP, College Station, TX, USA).

Prevalence of recorded impaired glucose tolerance/impaired fasting glucose

The trend in IGT/IFG prevalence increased during the 5-year study period from 17 per 100,000 in 2000 to 31 per 100,000 in 2004 (Figure 7).

FIGURE 7.

Annual prevalence of IGT/IFG.

Characteristics of people with impaired fasting glucose/impaired glucose tolerance

Of the people with IGT/IFG, 4298 (47.3%) were male and 4798 (52.7%) were female. The median age was 60.5 years [interquartile range (IQR) 22.8 years]. Table 10 provides the distribution by age. The BMI, based on the last recorded weight before the index date (a median of 3 years before the index date), is summarised in Table 10. Only 225, 2.5%, had been given a coded diagnosis of ‘obesity’. The high percentage with BMI scores of < 25 kg/m² is somewhat surprising. Metformin and glitazones have been suggested as potential treatments for IGT/IFG, but in this group only 253 (2.8%) of those with IGT/IFG were prescribed metformin or glitazones.

Cardiovascular risk

Prior to diagnosis of IGT/IFG 3.7% of people had experienced a previous cardiovascular event and 12.5% had a history of hyperlipidaemia (Table 11). A total of 1364 people (15%) were already receiving a prescription for aspirin at the time of their diagnosis and 990 (10%) were already receiving a prescription for statins, whereas 582 (6%) were prescribed both aspirin and statins.

Progression to diabetes mellitus

We followed each person with IGT/IFG from their index date to establish if they went on to develop diabetes. In total, 2662 (29.3%) developed diabetes mellitus during their follow-up. Of the 6434 people who did not develop diabetes mellitus, 506 died. The remaining 5928 had not developed diabetes mellitus at the time of their last follow-up. The median duration of follow-up was 2.9 years (minimum 0.9 years, maximum 53.6 years).

Among those who progressed to develop diabetes mellitus, 778 (29.2%) had a weight and height recording that indicated they had been obese (BMI ≥ 30 kg/m²). Table 12 summarises the distribution of BMI, gender and age among those who progressed to develop diabetes mellitus.

Of the 2662 who went on to develop diabetes, 799 (30.0%) progressed within 6 months after first diagnosis of IGT/IFG. A further 323 (12.1%) progressed within 12 months. By 5 years, 2256 (85% of those who went on to get diabetes mellitus) had progressed to develop diabetes mellitus (Table 13).

Summary of the Sheffield Type 2 Diabetes Model

The Sheffield Type 2 Diabetes Model is an integrated health-state simulation model of the natural history of diabetes and the lifetime cost-effectiveness of different treatments for T2DM. The model replicates patients’ risk of progression through five comorbidities: retinopathy, nephropathy, neuropathy, CHD and cerebrovascular disease. The intensity of management and monitoring can be varied by altering targets, such as those for glycaemic control, requirement for insulin, blood pressure control, and intensity of lipid-lowering therapy. For microvascular complications, the model is largely based on the Eastman models,285 using results from the Diabetes Control and Complications Trial (DCCT). For macrovascular complications, the model uses equations from the UKPDS. 286–288 The time spent by patients in each state for each comorbidity is recorded, for example years spent on dialysis, severe vision loss, etc., together with transitions between states. The effects of treatments on complications are modelled either via a RR (e.g. for the effect of photocoagulation on risk of severe vision loss) or via the effect on underlying risk factors (e.g. the effect of antiglycaemic medication on HbA_1c level). Complications are driven by individual demographic and modifiable characteristics at each time period, and the model includes diabetes and other-cause mortality. Total costs are obtained by adding the costs of therapy, the costs of one-off treatments (e.g. cost of amputation), and ongoing treatment of complications (e.g. treatment following stroke). The health benefit, the incremental QALYs, is obtained by applying QoL measures to the time spent in the various diabetic health states. Cost-effectiveness estimates for potential interventions are obtained by dividing the total costs by the incremental QALYs.

Model structure

The structure of the model used to assess the cost-effectiveness of lifestyle interventions to prevent diabetes is shown in Figure 8. ‘Control parameters’ include HbA_1c level switching thresholds for oral hypoglycaemic agents (OHAs) and insulin, targets for blood pressure, and assumptions regarding lipid-related therapy.

FIGURE 8.

Model structure.

Inputs and parameters specific to this assessment of interventions for preventing diabetes include:

• incidence curves over time for diabetes in patients with IGT

• the RR for diabetes of lifestyle interventions

• the cost of lifestyle interventions

• the effect of lifestyle interventions on other measures, such as weight and blood pressure.

Baseline characteristics

Table 14 below shows the baseline characteristics, some of which are discussed further below.

Characteristics were generated for 40,000 patients who progressed to diabetes. Matching between the two arms was undertaken (using ranking of duration to progression to diabetes) so that the same patient could not progress to diabetes faster in the intervention arm than in the control group.

Description and cost of intervention

Modelling a ‘real-life’ scenario

Modelling based on what happens in trials may not reflect what happens in routine care. Trials are protocol driven, and patients are expected to adhere to the treatment to which they are randomised. However, in routine care, clinical judgements are applied throughout, and if a treatment is not working, it should be stopped promptly. This can make treatments much more cost-effective, because only those people who are responding will continue to incur the cost.

The effect of this was illustrated in the review of new drugs for lung cancer for NICE. 297 In the trials, patients continued the drugs for the full series of chemotherapy courses, if they could tolerate them. In routine care, the effectiveness could be assessed after one or two courses, and the drug was stopped if there was no response. This considerably reduced the cost per QALY.

Hence, while modelling the results, as seen in the trials, has the merit of being based on observed data, it may be misleading as a guide to cost-effectiveness in practice.

Lifestyle measures are effective in preventing or delaying diabetes if adhered to but not all patients will adhere. This will apply even when those starting lifestyle interventions have been self-selected by, first, volunteering to be screened knowing what intervention would follow, and, second, coming back for intervention after being found screen positive.

There will probably be a range of adherence, rather than a dichotomy into full adherence and none. But, broadly speaking, some people will succeed and others will not. There is little point in pursuing lifestyle measures in those who do not adhere sufficiently, especially when we have an effective and inexpensive alternative – metformin. (However, lifestyle will always be the first choice, as it is more effective. In the DPP, lifestyle changes reduced the development of diabetes by 55%, and metformin by 30%, although the advantage of lifestyle changes over metformin was most evident in the > 45-year age groups.)

If we abandon lifestyle interventions in the non-adherent group, those left in the intervention group will be those doing best, and their results will be better than the average in the lifestyle groups in the trials. We will also reduce the cost, because the numbers left in the lifestyle group will be smaller (so we will need fewer groups, assuming that the intervention was group based). Simultaneously, the non-adherent group will be doing better on metformin (on the assumption that most will take tablets but not exercise and diet: in the DPP adherence to metformin was 71%, defined as taking 80% or more of the prescribed dose. 298

The question for modelling is how we define failure and at what interval. Various studies have shown that the effect of interventions wear off after they are stopped, or reduced in intensity.

Figure 9 shows the effect of discontinuation [based on data from Swinburn et al. (2001)299], and Figure 10 shows the effect of a reduction in intensity [based on data from Wing et al. (1998)238]. In the former, Swinburn et al. (2001)299 provided the reduced-fat diet intervention for 12 months, and the effect lasted for another 12 months. Weight loss was 3 kg at 6 months, 3.3 kg at 12 months and 3.2 kg at 2 years. However, weight gain resumed at this point, and the intervention group weight curve meets the control subjects after 5 years.

FIGURE 9.

Change in weight (kg) over time: reduced-fat diet intervention.

FIGURE 10.

Change in weight (kg) vs time: 2-year lifestyle interventions in overweight individuals. LI, lifestyle intervention; BA, basic advice.

In the latter, Wing et al. (1998)238 provided weekly meetings for the diet groups for the first 6 months, reducing to fortnightly for the second 6 months. As the graph shows, reduction in intensity was accompanied by an increase in weight. By 6 months, 30–40% of recruits had ceased attending. By 12 months, most had ceased to attend.

However, in this case the non-adherers probably adhered for a short period and then relapsed. It seems best therefore to base discontinuation on the results from the larger trials, such as DPS, where there was initial improvement (weight loss and physical activity) in the first 6 months, maintained to 12 months, followed by a loss of some of that gain by 18 months and beyond. In most trials,108,238,299 results are good for 6 months, after which they tend to plateau for a while before the subject starts to regain weight. These results reflect the mix of adherers and non-adherers. We will use that interval as the point at which non-adherers will leave the lifestyle arm and switch to metformin.

Trials such as DPP and DPS did not provide scatterplots of results at time intervals so we do not have distributions showing how many succeeded. However, the DPS reported progression to diabetes by ‘success score’ based on how many of the five lifestyle goals were met. Diabetes did not develop (at the 12-month visit) in any of those who achieved all five goals. If we take a success score of three or more as the indicator of adherence then only 37% of the intervention group adhered. 13% of the control group also achieved this level of success, so standard care works for some: the marginal effect of intervention was 24%.

Therefore, assumption 1 is that only about 40% will do well enough to justify continued intervention.

In addition to the interval, we also need to decide how to define failure. The key result in the trials is weight loss, except in the Indian trial,173 in which the intervention groups did not show significant changes in weight from baseline (although the control group gained weight). The key indicator of failure, therefore, should be not losing weight or not losing enough weight. That raises the question of defining ‘enough’, which has to be applied at the time point of switching treatment: 12 months. In the DPS, the average amount of weight lost was modest (3.5 kg at 2 years, vs 0.8 kg in the control group – but with a large SD of 5.5 kg, indicating that some had lost a lot of weight), but was clearly enough to reduce progression to diabetes. The target weight loss in the DPS was ≥ 5 kg and this was achieved by 43% at 12 months and 42% at 2 years. The group that lost most weight showed the greatest gain in insulin sensitivity after four years. 218 At the 7-year follow-up, weight loss was the strongest predictor of success, and in multivariate analysis, weight loss was the only factor that remained significant. 213,214

In the DPP, the intervention group lost on average 7 kg by 6 months. In the DPP, the target was weight loss of ≥ 7%, and that was achieved by 50% at the 24-week follow-up point (when the 16-lesson curriculum ended). Weight loss was the main determinant of diabetes prevention in the DPP. 193

With good compliance, patients should be losing weight slowly but steadily, about 2 kg per month, or about 1 lb a week. That would suggest a loss of about 12 kg at 6 months. The DPS and DPP results show much less than that in the lifestyle groups, but their means reflect the full range of compliance, so those who adhere best will achieve more. In the DPS, those in the highest tertile of weight loss achieved losses of 8–17%, equivalent to 7–15 kg. Those who lost less had very little improvement in insulin sensitivity.

The definition of failure might therefore be taken as weight loss, being liberal, of < 5% at 6 months, which would exclude 60% of those in the DPS, namely those who did not achieve three or more of the targets. (See Projected incidence of diabetes: intervention arm. )

Projected incidence of diabetes: control arm

As a base case, we assume that adherence to the intervention was in line with the DPS achieving the same effectiveness over 8 years.

Projected incidence of diabetes: intervention arm

In the DPS,183 each subject was (1) given advice on physical activity, (2) offered circuit-type training and (3) offered seven sessions with a nutritionist during the first year of the study and one session every 3 months thereafter.

We assume that, after the initial 4-year course, those in the lifestyle intervention arm benefit from some sustained reduction in diabetes risk as observed in the DPS183 follow-up, although it is unclear how much is prevention compared with delay. It may be too optimistic to assume that the benefit of having prevented some patients progressing to diabetes is sustained indefinitely. The 20-year results224 from the Chinese Da Qing study also add weight to an assumed mix of prevention and delay.

Projecting incidence of diabetes for responders and non-responders

Responders

For the overall lifestyle group, the RR for having developed diabetes compared with the overall control group increased over time from 0.48 at year 4 to 0.65 at year 8 (although the absolute difference in incidence remained about the same, suggesting that there is some sustained protective effect after 4 years). Fitting curves to the data for the intervention period and follow-up period suggests a logarithmic form that is consistent with the concept of an increasing RR but one that increases at an ever slower rate (because of some sustained preventative effect).

Among those in the lifestyle intervention group of the DPS achieving a success score of ≥ 3 at the end of year 1, the incidence of diabetes during the intervention period was 2.2% (2/87) compared with 24% (56/233) in the overall control group, giving a RR of 0.10 (95% CI 0.02 to 0.38). The RR was 0.67 for non-responders in the lifestyle arm compared with the overall control group.

In order to estimate how the RR in the responders changes between year 4 and year 8, we explored alternative scenarios for how the RR in responders and non-responders might change, while the combined RR changes from 0.48 to 0.65. A conservative but plausible scenario is for the RR in the non-responders to change from 0.67 at year 4 to 1.0 at 20 years. The corresponding change in the responders’ RR is to increase from 0.1 after 4 years to 0.46 at year 20 as shown below (Figure 11).

FIGURE 11.

Change in RR for having developed diabetes: responders vs control group.

Non-responders

For non-responders, after switching to basic advice we assume that these patients are at the same risk as the corresponding least successful group in the control arm of the DPS, i.e. we are assuming that non-response is partly a result of non-adherence to the intervention and advice. Based on the risk of those achieving a goal success score of 0 or 1 in the control arm of the DPS183 (55% of patients), we obtained a RR of 1.07 for the non-responders, after switching to basic advice, compared with all control subjects.

Combined incidence of diabetes with a switching strategy for non-responders

The estimated and projected cumulative incidence of diabetes obtained, under our base-case strategy involving switching those who do not achieve targets, is shown in Table 17 and Figure 12. Actual values are estimated from the incidence curves. We assume no further incidence of diabetes in either arm after year 20.

TABLE 17 - Estimated and projecteda cumulative incidence of diabetes

Shown in italic text.

Year	Lifestyle intervention (%)	Control (%)
0	0.0	0
1	2.1	7
2	10.1	15
3	14.5	22
4	18.4	28
5	24.0	34
6	29.0	40
7	32.0	43
8	36.5	49
9	40.2	53
10	43.4	56
11	46.0	59
12	48.2	62
13	50.1	64
14	51.7	65
15	53.0	67
16	54.1	68
17	55.1	69
18	55.9	69
19	56.6	70
20	57.2	70

FIGURE 12.

Calculated and projected (dashed) cumulative incidence of diabetes. BA, basic advice; LI, lifestyle intervention.

Under these assumptions, for the 57% of patients in the control arm who are projected to progress to diabetes over the 20 years, the average time to diabetes is approximately 4.9 years compared with 7.4 years for the first 57% of patients to progress in the intervention arm, a delay of 2.5 years.

At year 20, there is a difference between treatment arms in the proportion of patients who have progressed to diabetes. Although it is not known whether, as we have assumed, these patients have been prevented from developing diabetes indefinitely, this assumption is less important for the overall results because:

• patients will be aged 75 years on average by then so that life expectancy beyond is much reduced

• any convergence of the incidence curves beyond year 20 is likely to be gradual

• costs and benefits beyond the 20-year horizon will be heavily discounted.

Alternative treatment pathway for non-responders

In a sensitivity analysis, non-responders are switched to metformin rather than basic lifestyle advice, in line with IDF guidance. 303 The DPP showed that the RR for progression to diabetes with metformin increased with age at baseline, although there was some benefit even in the oldest subgroup. 194 For our population starting with an average age of 55 years, we estimated the RR compared with placebo to be 0.75. Although we assumed that metformin would be given indefinitely (given its very low cost), over an intervention period spanning 20 years, the RR is likely to be higher than that observed in the DPP, with an average follow-up of only 3.2 years. For the purposes of the modelling, we considered it more conservative to use the RR for a 65-year-old at baseline, which we estimated to be of 0.83.

Adherence to interventions

Non-progressors and regressors to normal glucose tolerance

In patients who do not progress to diabetes, there may still be some benefit from lifestyle intervention through improved cardiovascular risk profile, especially reduced SBP. Lower glucose levels, better lipid profile (e.g. HDL-C improvement through exercise) and the effect of weight loss or increased physical activity on CVD (independent of the above risk factors) may also confer benefits but appear to be less significant or unknown. Weight loss is also known to improve QoL.

However, in the absence of any evidence on differential CVD risk profiles between treatment arms for those who remain with IGT, we have not included any such effects.

Similarly, we have not modelled rates of regression to NGT because there is a lack of evidence to accurately quantify:

• timing of regression

• the benefits of relatively small differences in glycaemia at relatively low levels (i.e. below the threshold for IGT but above the normal level).

Not including such effects has a conservative effect on the cost-effectiveness of lifestyle interventions.

Other treatment and monitoring assumptions for impaired glucose tolerance patients

Some patients with IGT have a high enough risk of a CVD event that they should be taking statin therapy. This is because of a combination of their age, presence of features of the metabolic syndrome, or presence of established CVD. We estimated that approximately 20% of patients should already have a prescription for a statin at the point of diagnosis of IGT.

For this assessment, assumptions in respect of the benefit from statin therapy are important in setting a baseline level of CVD risk. We assume an average dose of 40 mg generic simvastatin at a cost of £3.80 per 28 [British National Formulary (BNF) 54]. 304 Only 5.5% of patients in DPS were treated at baseline with lipid-lowering therapy, so the baseline LDL-C level of 3.56 mmol/l as per Table 14 is effectively the pre-statin level. With an underlying adherence rate of 86% in statin trials,305 an LDL-C fall of 37% with 40 mg of simvastatin is expected based on the analysis of Law et al. (2003),306 i.e. 1.32 mmol/l. This reduction would lead to a RR of CHD with statins of 0.50 taking the rate for 60-year-olds to be most appropriate for this purpose. It is assumed that primary and secondary prevention RRs are similar307 and that these are similar in diabetic people and non-diabetic people. 308. Given a 10% fall in stroke per 1.0 mmol/l fall in LDL-C, a 1.32 mmol/l fall in LDL-C is assumed to give a RR of stroke of 0.91.32/1 = 0.87. Adjustment to CVD risk is made for the difference between the real-world statin adherence level [which we assumed to be 60% guided by Kopjar et al. (2003)305 and Penning-van Beest et al. (2007)309] and that in the trials (on which the above RRs were based).

In both arms, we assume an additional appointment with a GP nurse each year to monitor IGT status.

No testing for microvascular complications is assumed until a patient becomes diabetic (even though some retinopathy and albuminuria develops in patients with IGT).

Effect of intervention on weight and systolic blood pressure

Weight loss is not only an important determinant of the differences in diabetes risk already discussed, but also of changes in QoL and CVD risk factors. The reported and expected changes in weight over the long term can guide assumptions about differences between groups in QoL and CVD risk factors.

Effect of weight loss on cardiovascular risk

We are unaware of any clear evidence of whether weight is an independent risk factor for CVD in diabetic people after taking account of the risk factors in the UKPDS equations.

Effect of intervention on blood pressure

In the DPS, patients’ blood pressure (and cholesterol) was checked at annual visits, and they were advised to contact their physician for treatment and follow-up if levels were abnormal. However, use of antihypertensive drugs was unchanged from baseline to year 1 in each arm, so the difference between treatments in the reduction in SBP at the end of year 1 (5 mmHg vs 1 mmHg; p = 0.007) can be related to the intervention effect that probably arises from the effect of weight loss on blood pressure. Similarly, a meta-analysis undertaken by the Cochrane group (which included the DPS results) reported a mean difference in blood pressure of 4.4 mmHg between groups. 180

For the 40% of patients who achieve the goals of the lifestyle intervention, the fall in blood pressure would be expected to be more than 5 mmHg. Taking the reported 14 mmHg SD in the fall in the DPS, applying a truncated normal distribution and assuming that the best achievers obtain the greatest falls, gives a mean fall in SBP of 8.5 mmHg in this subgroup. However, the correlation between weight loss and SBP would not be perfect, so we have adopted a more conservative estimate of 6.5 mmHg for the average SBP fall in the lifestyle achievers. We assume that this reduction is maintained throughout the intervention period. Beyond this, partial regain of weight may lead to loss of some of this benefit, so we conservatively assumed that this is lost between years 5–8.

For the 60% of patients who switch away from the lifestyle intervention after 12 months, the average fall in this subgroup at year 1 would be 4 mmHg to agree with the average fall of 5 mmHg of the group. It is assumed that this is lost by year 4, as it is assumed that without active treatment there would be no sustained weight loss to maintain the effect on SBP.

The net effect of the above in the overall lifestyle arm (responders and switchers) is that the initial fall in SBP of 5 mmHg is lost at an assumed approximate constant rate of 0.63 mmHg per year.

For controls, we assume that the 1 mmHg fall in SBP at year one is gradually lost by the end of year four given that weight was regained. For patients in the control arm that become diabetic, we assume that negligible HbA_1c reductions will be achieved through increased diet and exercise, given that they are likely to be asymptomatic with a low baseline HbA_1c level of 6.2%. We also assume that they will not achieve sufficient diet & exercise changes to derive any SBP change at diagnosis. Alternative assumptions are tested out in a sensitivity analysis.

Effect of intervention on lipids

Although the Cochrane review by Norris et al. (2005)180 reported a slightly greater fall in total cholesterol with lifestyle interventions (0.18), it is unclear that this would be replicated in a setting of routine statin therapy, so we have assumed no reduction in cholesterol. Reported HDL-C effects are negligible. We assume that lipid levels do not change significantly from diagnosis of IGT to diagnosis of diabetes. This is reasonable given our assumptions on statin use.

Effect of intervention on quality of life

Weight loss has been shown in several studies to improve QoL as shown in Table 19.

A simple weighted average of the above gives a utility change of 0.0025 per kilogram change.

The incorporation of weight-based utility effects in the model is further supported by a recent review by Dennett et al. (2008). 315

In the DPP,108 the average overall QoL over 3 years in the lifestyle group was 0.02 higher than in the placebo group (0.70 vs 0.68). This is more than might be predicted based on weight changes alone, so lifestyle interventions may impact QoL in other ways, for example a feeling of well-being from exercise.

Some aspects of the DPP108 were more intensive than the DPS183 (e.g. monthly follow-up, target weight loss of > 7% rather than 5%), so we have based our QoL assumptions on weight differences. This gives the effects shown in Table 20.

These effects apply only while patients are pre-diabetic.

A sensitivity analysis assuming a more conservative effect of weight on QoL, 0.001 per kg, was undertaken.

Relationship between glycaemia and cardiovascular disease risk

Microvascular complications during impaired glucose tolerance period

Studies have reported incidence of retinopathy and albuminuria in the pre-diabetic period, albeit low rates. 316 We have used the same risk equations as used in the full diabetes model, as these will generate low rates of incidence for patients with IGT because of the exponential relationship between HbA_1c level and events.

Management of diabetes: parameters and assumptions

• HbA_1c threshold for switching (to OHAs): 7.4%. 317

• HbA_1c threshold for switching (to insulin): 8.5%.

[Although earlier use of insulin might be advocated, response to insulin is known to be relatively poor in many overweight patients, who may therefore be more reluctant to start. Rubino et al. (2007)318 suggest that, in practice, many wait 5 years or more with an HbA_1c level of > 8% before starting, and many above 9% insulin, which suggests switching at a higher threshold. 318]

• assumed HbA_1c level cap under insulin therapy: 9.2%.

• blood pressure control target (SBP, mmHg): 140 mmHg for IGT, 135 mmHg for diabetes

• OHA treatment strategy for diabetes: metformin, then metformin + sulphonylurea, then metformin + insulin.

Effect of treatment following diagnosis of diabetes

Other model parameters and assumptions

Sensitivity analyses

We have undertaken some sensitivity analyses around key uncertainties in the evidence base, as shown in Table 26. Given that the intervention is highly cost-effective in the base-case analysis, we have mostly focused on sensitivity analyses to demonstrate the effect that more conservative assumptions would make on the cost-effectiveness of the intervention.

Base-case results

Table 27 provides a summary of the base-case results.

Table 28 gives a detailed breakdown of the base-case results.

The total incremental cost (per patient with IGT) of the intervention amounts to £121. This can be attributed largely as follows:

• an additional £449 cost of the intervention (although this is partly a result of longer duration in the pre-diabetic state, on average, hence longer duration of both active intervention and follow-up treatment)

• £220 saving: a mix of preventing and delaying the need for initiating expensive insulin therapy (a slightly lower but similar effect would have been observed if triple OHA therapy had been an option within the treatment pathway before insulin)

• £78 saving: avoiding the excess monitoring costs (including retinal screening) that arise once diabetes has been diagnosed.

Savings relating to treatment of complications were only £19.

The QALY benefits of the lifestyle intervention can be attributed largely as follows:

• 0.0309: improved survival driven partly by lower average blood pressure and HbA_1c levels

• 0.0348: QoL impact of the significant differences between treatment arms, achieved through both

  1. – weight loss during the 4-year period of the intervention for IGT

  2. – preventing or delaying the weight gains that arise with sulphonylureas and especially insulin.

Although the incremental QALY gain seems small, in our experience this is expected for diabetes-related interventions that target a fairly broad patient group that is more at risk of future adverse health consequences than imminent ones. There is nevertheless a tangible and cost-effective population benefit.

The table above shows that, although estimated cardiovascular mortality is reduced in the intervention arm, the total number of CHD and stroke events is almost the same. This seems counterintuitive but can be explained as follows:

• The lower blood pressure levels, and to a lesser extent HbA_1c levels, during the earlier years reduces the case fatality rate so that the proportion of events that are fatal is lower in the intervention arm.

• A delayed diagnosis of diabetes results in delayed initiation of metformin therapy, which is likely to be cardioprotective – this has an upwards effect on the total number of CHD events in the intervention arm relative to the control arm.

• Lower blood pressure levels in the intervention arm during the earlier years result in lower CVD risk and fewer patients crossing the threshold for starting statin therapy – this also has an upward effect on the relative number of events in the intervention arm compared with the control arm.

• The reduced mortality in the intervention arm leads to greater secondary non-fatal events (because more patients are alive and at risk of recurrent events).

Even though total CHD and stroke events may not be reduced as expected, there is a reduction in CVD mortality and less heart failure, as well as other significant benefits – largely a delay in the need for insulin (or other expensive antiglycaemic therapy) and improved QoL due to weight loss.

In the sensitivity analysis (see Sensitivity analysis results, below) in which those not achieving lifestyle goals with the intervention subsequently switch to metformin rather than basic lifestyle advice, both non-fatal and fatal CHD events were lower in the intervention arm. This is because of the expected effect of metformin on CHD risk and is one of the factors that needs to be considered in evaluating the merits of using lifestyle or metformin therapy for prevention of diabetes.

Affordability

Even if the intervention is cost-effective in the long run, an equally important outcome is how affordable the intervention is initially, as often cost savings are generated in the longer term. For a lifestyle intervention the net 1-, 3-, 5- and 10-year cumulative outlay per patient diagnosed with IGT is shown in Table 29. Ten-year cumulative outlay is estimated at approximately £346 per patient, i.e. significant investment is needed in the early years, even though the intervention is likely to result in small annual cost savings in the long run through reduced costs of monitoring diabetes, and lower therapy costs (especially insulin).

Figure 13 shows than annual net investment over the first 10 years. Within 5 years, annual cash flows are predicted to be positive.

FIGURE 13.

Annual net investment over time.

Compared with drug therapies, lifestyle intervention can be relatively labour intensive and the staff resourcing needs of a national programme, whether through the NHS or external contract arrangements, would need to be carefully assessed.

Sensitivity analysis results

The following key sensitivity analyses (Table 30) have been undertaken (as per Sensitivity analyses, above).

Critical success factors of the Diabetes Prevention Study

To our knowledge, it is not known which components of the DPS were critical in achieving the required weight loss that seems to be a threshold for reducing risk of diabetes. Replicating the effectiveness of the DPS in a real-world UK setting may be dependent on specific goals, perhaps significant uptake of exercise classes, as weight loss is known to be frequently less effective when pursued through dietary approaches alone.

It might be possible to design a cheaper intervention than that used in the DPS but with the same effectiveness, perhaps taking into account any advances, since the DPS, in knowledge on optimal diets for reducing diabetes risk, or using counselling to re-enforce behavioural changes.

Maintaining lifestyle changes in the real world

Ability to comply with, and respond to, lifestyle changes varies from patient to patient, particularly the ability to lose weight. Even in the DPS, only 43% achieved the weight reduction goal, and 36% of subjects increased their physical activity. Furthermore, the Good Ageing in Lahti Region (GOAL) Lifestyle Implementation Trial demonstrated that replicating, in the real world, the same degree of physical activity and weight reduction as that observed in trials may be difficult. 335 Re-enforcement of lifestyle changes through effective counselling strategies might be possible nevertheless. 133

Durability of benefit in responders

In the DPS, the HR for diabetes in the intervention arm increased steadily between years 4 and 8. This is probably owing to a mix of participants not sustaining the intensity of lifestyle changes, a wearing-off of the protective effect of the intervention, and age-related beta cell loss. In those that successfully sustain weight loss and other lifestyle changes, the extent to which the benefit of preventative interventions is sustained in the long term will determine whether diabetes can be truly prevented rather than just delayed.

Affordability of the lifestyle intervention

Well-supported lifestyle interventions are typically not cheap, more than the cost of first-line drug treatment for diabetes including the monitoring cost. If lifestyle interventions are not affordable because of the initial investment needed, or are beyond available manpower resources, then metformin, although less effective in trials, might be a cheap and convenient alternative (metformin costs only £44 per year). This is particularly relevant to non-responders to a lifestyle intervention. A sensitivity analysis with a strategy that switches non-responders to metformin significantly increased the cost-effectiveness of intervening to prevent diabetes.

Probabilistic sensitivity analysis

We did not undertake a probabilistic sensitivity analysis (PSA) owing to time constraints and the emerging conclusions concerning the cost-effectiveness of the intervention (i.e. the intervention remaining cost saving across a range of one-way sensitivity analyses and a couple of multiway pessimistic scenarios). Given this, the probability of the intervention not being cost-effective is unlikely to reach a level that would influence decision-making.

Threshold for discontinuing the lifestyle intervention

We have examined the incremental cost-effectiveness of switching those who do not achieve lifestyle targets away from the intervention, to either routine ‘basic advice’ or to metformin. The exact threshold at which to apply this switching, and the subsequent treatment of those who are switched, would have to be further evaluated. The intervention strategy may still be cost-effective if only those failing to achieve a DPS goals success score of at least two were switched. It would be worth undertaking further research to examine the costs and benefits of continuing the lifestyle intervention in those with a goal success score of exactly 1 and exactly 2. The weight loss and reduction in diabetes risk from the lifestyle intervention in these subgroups might still be sufficient to be cost-effective.

Using a DPS score of at least 2 after 12 months as the criteria for continuing the lifestyle intervention would mean approximately 25% more patients continuing with the lifestyle intervention beyond the first year than if the criteria was at least 3 (possibly meaning 25% fewer patients requiring medical treatment with metformin to reduce the risk of diabetes).

Perverse consequence of delaying diabetes

There may be some adverse consequences of delaying a diagnosis of diabetes and thereby the intensification of treatment that follows as a result of it. In particular:

• The initiation of metformin is delayed, thereby delaying the potentially substantial reduction in CVD risk with metformin (see Base-case results, above).

• More intensive but cost-effective treatment of blood pressure and lipids may be delayed – this depends on whether lipids are initially treated with a ‘fire-and-forget’ policy or treated to targets or based on CVD risk. Intensification may lead to what seem like relatively small improvements in risk factors but some analyses undertaken suggest these can significantly influence the results.

• Two-hour glucose may slowly drift upwards towards 11 mmol/l, leaving patients exposed to increased CVD risk over a long period.

However, given how cost-effective the results show the intervention to be, these factors are unlikely to outweigh the benefits.

Other benefits of treating obesity in patients with impaired glucose tolerance

There are likely to be some other benefits from lifestyle improvements that are not included in the model, such as a reduction in obesity-related cancer and other problems (e.g. osteoarthritis in knees and hips). This together with the weight-related QoL benefit may be enough to justify lifestyle intervention in lower risk populations in which a significant proportion may never progress to diabetes. Furthermore, CVD risk equations may not fully capture the benefits of lifestyle changes, especially the benefits of exercise and weight loss. In non-diabetic people, at least, the QRISK equations suggested an association between weight and CVD risk, independent of blood pressure and lipids. 336

Comparison with other cost-effectiveness assessments of interventions for impaired glucose tolerance

It is difficult to compare our results directly with previous economic assessments because:

• Our treatment pathway involves patients who do not achieve lifestyle goals switching off the intensive treatment. This greatly improves the cost-effectiveness.

• We have assumed that lifestyle changes become habitual in those who achieve targets within the 4-year intervention period, so that ongoing intensive intervention is not needed after 4 years.

The fact that our results are cost-effective is nevertheless consistent with an assessment of the DPS in Swedish setting, and assessments by Palmer et al. (2004)243 and Herman et al. (2005). 244 Our decision rule, by which those not achieving weight loss after 12 months no longer continue with the intervention, is a key factor in the intervention being cost-effective. An assessment by Eddy et al. (2005)248 did, however, produce less favourable results. Possible reasons for this are multiple and were discussed in detail in Chapter 5.

Implications for further research and modelling

As discussed above (see Mix of impaired glucose tolerance/impaired fasting glucose and risk profile of cohort in the model), these results apply to patients that have IGT (either with or without IFG). Further research may be needed in order to determine the effectiveness of lifestyle interventions in patients without isolated IFG, particularly in respect of:

• the reduction in risk of progression to diabetes

• the relationship between IFG and risk of CVD.

The exact threshold for determining treatment failure with lifestyle intervention and pursuing alternative means of prevention (probably with metformin) requires further modelling and analysis. It is also often asserted, based on the Indian DPP,173 that there is no added benefit from combining metformin with lifestyle intervention above lifestyle or metformin alone. There was considerable overlap between the CI around the results for each treatment arm in the Indian DPP – a more accurate interpretation is that is was not possible to establish whether there was any added benefit, and further research results are needed.

More evidence is required on the extent to which lifestyle intervention and metformin lead to sustained benefits in the long term. It is also unknown whether it would be worthwhile continuing with some form of active intervention over a longer period that that used in prevention trials to date (typically 4–6 years).

The cost-effectiveness of any preventative or screening-related diabetes intervention is affected by the large uncertainty in the CVD risk reduction of metformin. Future economic assessments and clinical risk assessment in practice would benefit from a more precise estimate of metformin’s effect. The DPP Outcomes study (follow-up to the main DPP study) may provide some evidence for this, or it could be incorporated into the design of a future trial of interventions for IGT.

The cost-effectiveness of lifestyle interventions is determined by various cost and QoL impacts, often underpinned by an uncertain evidence base, so there is uncertainty in the results and conclusions. However, we have carried out a range of sensitivity analyses, each suggesting that lifestyle intervention (sustained only in those achieving weight loss/lifestyle targets) is cost-effective. Some further (but probably less important) sensitivity analyses that could be undertaken are summarised in Appendix 6.

Would trial interventions work in ‘real life’?

Seidel et al. (2008)351 did not think that the DPP intervention could easily be replicated in community settings. Therefore, they set out to test a group-based lifestyle balance intervention in urban areas of Pittsburgh, described as ‘medically underserved’, and being a socioeconomically depressed area following the decline of the steel industry, with an ageing population owing to out-migration. A community approach was taken, with notices in sites such as churches, shops, local newspapers and radio. Only short-term (6-month) results are as yet available, but look promising, with about half of the recruits losing at least 5% of body weight. Only 88 people joined but most of these completed the course. It would be useful to see longer-term follow-up with larger numbers.

The Greater Green Triangle Diabetes Prevention Project352 aimed to evaluate whether a structured group programme of lifestyle intervention, set in an Australian primary health-care setting, would give similar reductions in risk factors to that found in the RCTs.

The 12-month study used before and after testing. Patients from general practices, who were at high risk of developing T2DM, were screened opportunistically using The Diabetes Risk Score tool. 352 The intervention, based on the Finnish GOAL study353 consisted of six structured 90-minute group sessions over 8 months.

Participants were aged 40–75 years with moderate or high risk of developing T2DM. Only 76% attended both the baseline and 12-month clinical tests and at least one group session.

After 12 months statistically significant improvements were observed in participants’ mean weight, WC, fasting and 2-hour glucose lipids, DBP, and most psychological measures. However, follow-up is still short, and only time will tell if the benefits are sustained.

From this, it would appear that the results of the trial can be replicated in routine primary care.

In Finland, the national diabetes programme (DEHKO) includes a subprogramme to prevent diabetes (the FIN-D2D project),354 involving three strategies:

• a population-based approach to prevent obesity and diabetes

• the high-risk strategy – identification and screening of people at high risk, followed by lifestyle measures

• early diagnosis of people with T2DM and prompt treatment to prevent complications.

The 1-year follow-up reported that the risk of diabetes was 0.31 (95% CI 0.16 to 0.59) in those who lost 5% or more of weight relative to those who maintained weight. 355 This was in a group identified by a high (≥ 15) FINDRISC score.

Good results were reported from a DPP-style intervention in Montana, with 39% of respondents reporting that they had maintained or achieved 7% weight loss. 356 However, there was considerable attrition in this study. Of 591 initial recruits, only 79% (466) completed the programme, and only 40% of these (188) responded to the follow-up questionnaire. So only 12% of the initial cohort had lost 7% of weight at about 18 months after entry. It should also be noted that the data were self-reported.

The primary goal of the IDF Taskforce on Prevention and Epidemiology consensus workshop in 2006 was the prevention of T2DM in both the developed and developing world. 303 The IDF plan is aimed at simultaneously controlling modifiable risk factors with lifestyle modification in two target groups: (1) people at high risk of developing T2DM and (2) the entire population.

An editorial by Simmons et al. (2007)357 on the IDF consensus statement questioned whether individualistic approaches to diabetes prevention would be effective, and felt that the emphasis should be more on population-based approaches. They made the point that the rising prevalence of diabetes seen in increasingly obese populations is a result of a shift in the entire distribution of glucose in that population, rather than simply an increase in the number of people at the tail of the distribution. Therefore, the most effective strategy for lowering the mean glucose level of the population, thus lowering the population burden of CVD attributable to hyperglycaemia, is to attempt to increase average levels of activity and reduce obesity in the large number of people with moderately raised levels of glucose.

The evidence base for individual approaches to diabetes prevention is stronger than for population-based approaches, as the former are much easier to test in RCTs. However, Simmons et al. (2007)357 feel that in order to correct this imbalance and to be able to evaluate population-based approaches, researchers need to be more open to other study designs, such as natural experiments in local communities using quasi-experimental designs.

The GOAL Lifestyle Implementation Trial335 aimed to test whether the findings achieved in the DPS trial183 could be replicated in a ‘real world’ setting. 335 The study used a longitudinal pre-test and post-test study design, and focused on the five key lifestyle changes derived from the DPS. 183

The 352 participants, mean BMI > 32 kg/m² and aged between 50 and 65 years, were recruited from primary health-care centres in Finland. Risk status for T2DM was determined using a standardised risk questionnaire. The inclusion criterion was set at risk score of ≥ 17% 10-year risk; 25% of the participants had IGT at baseline. The intervention included six group counselling sessions given over 12 months.

Only 57% of the participants attended all six counselling sessions, and 33 participants dropped out of the study. The number of participants in this study who attained four or five of the lifestyle objectives was 20%, which was similar to the DPS (18%). However, the > 5% weight loss goal was significantly less frequently achieved in this study (12%) than in the DPS (43%). The physical activity goal was significantly less frequently achieved in this study (66% in GOAL335 vs 86% in DPS183), whereas the fibre objective was significantly higher in this study. After 1 year, several clinical risk factors decreased significantly, including DBP, weight and BMI (only men), and WC (both sexes). The results at 36 months showed that the weight loss was maintained. 358

Therefore it would seem that this lower-intensity (and hence lower-cost) 12-month intervention was successful in decreasing diabetes risk in a ‘real world’ setting. However, it is not known whether these results will be sustained over the longer term.

One general practice in Glasgow tried to implement healthy eating and exercise over a 3-year period, with group sessions, an exercise scheme, and dietetic and medical time. However, maintenance of gains proved difficult unless the intervention was continued. As Guthrie comments:359

In the end all of these patients required a continuous personal input to maintain their weight loss, regular exercise, or healthy eating, and it simply became unsustainable.

Perhaps we need to look at interventions that are less ‘medical’ in nature. Truby et al. (2006)360 carried out a randomised trial of four commercial weight loss programmes, including WeightWatchers. All four diet programmes had good results, with an average weight loss of almost 6 kg at 6 months. We need much longer follow-up to see if weight loss continues or is sustained.

Can exercise alone reduce progression to diabetes?

Yates et al. (2007)361 conducted a systematic review of controlled trials to establish whether increasing physical activity, independent of changes in diet or weight loss, can reduce the risk of T2DM in people with pre-diabetes (IGT and/or IFG).

The review included eight trials (seven randomised and one non-randomised) in individuals with IGT. Seven of the studies used a multicomponent lifestyle intervention and one used a structured gym-based exercise training intervention. Four studies included the incidence of diabetes as the main outcome and found that diabetes incidence was reduced by 42–64% compared with the control group. These studies reported only small changes in physical activity. The other four studies used 2-hour plasma glucose levels as the primary indicator of glucose control, and only one reported a significant improvement. Three of these studies reported small to moderate increases in maximal oxygen uptake, suggesting adherence. All but one of the studies included in the review reported significant weight loss among participants.

The conclusion of the review was that the role of physical activity independent of other lifestyle changes in the treatment of pre-diabetes remains uncertain. Given the relatively modest increases in physical activity, the success of the interventions is probably due to the weight loss.

Laaksonen et al. (2007),362 in response to the Yates review, did not agree that the 9 minutes/day increase in moderate to vigorous physical activity reported in the intervention group of the Finnish DPS183 was insubstantial. They found that the percentage of sedentary individuals (< 1 hour/week of moderate to vigorous physical activity) in the intervention group of the DPS183 decreased from 37% to 15%, and those engaging in at least 2.5 hours/week increased from 41% to 62%, and felt that this was likely to result in health benefits.

They also reported that individuals who engaged in at least 2.5 hours/week of brisk walking or other forms of moderate to vigorous physical activity were 44–69% less likely to develop diabetes than individuals engaging in < 1 hour/week. These findings were based on post hoc analyses from the intervention arm.

Burns et al. (2007)363 examined the effects of a 3-month aerobic exercise training programme in young obese insulin-resistant subjects with and without T2DM. They recruited 13 subjects with T2DM and 18 non-diabetic control subjects for the baseline study. The two groups were matched for age, BMI, body fat and physical fitness. An exercise intervention (involving 1 hour of exercise training four times per week for 12 weeks) was completed by seven of the subjects with T2DM and 14 of the obese control subjects. The overall mean age of the completers was 26 years and mean BMI was 34 kg/m².

The authors had hypothesised that exercise alone, while maintaining a stable diet, should improve insulin sensitivity in these severely insulin-resistant subjects, but found to their surprise that neither group showed metabolic improvements after the aerobic exercise intervention. The authors comment that this result raises interesting new questions about the pathogenesis and treatment of early-onset T2DM in obese young people.

Carnethon (2007)364 (in an editorial on the Yates 2007361 review) comments that it is important to understand whether it is the diet or physical activity component of the lifestyle intervention that is the key to its success. Physiological responses to, and compliance with, the diet and exercise components of the lifestyle intervention vary greatly between individuals. By focusing the interventions on the most effective component, or the component that is easiest to adopt, we might be able to improve compliance with the intervention.

A Cochrane review by Orozco et al. (2008)185 examined the role of exercise and diet in preventing T2DM. 185 The eight studies they included differed from our inclusions. They included Bo et al. (2007),365 which recruited patients with metabolic syndrome. They did not include three studies that were included in this review. One was Wein et al. (1999),191 which appears to have been suitable for inclusion according to their criteria because it focused on women with previous GDM. It is not listed as an exclusion, or indeed anywhere in the Cochrane review. It may have been missed. They listed Mensink et al. (2003)188,233,234 as awaiting assessment because the final 3-year results had not been published – they were published a few months later and have been included in this review. They also excluded Liao et al. (2002)187 because ‘the control group received an intervention that differed from standard recommendation’. It is not clear what was meant by this.

The final conclusion of the Cochrane review was that more evidence was required on the effects of exercise alone.

Would UK populations comply with increased activity?

In Finland, a combined strategy of general population and targeting of high-risk individuals is being used. For this to work in the UK, we would need to encourage the population to be more physically active. Current statistics suggest that this might be difficult.

Data from the Information Centre ‘Statistics on Obesity, Physical Activity and Diet: England 2006’366 show that:

• In 2004, only 35% of men and 24% of women reported achieving the physical activity targets of at least 30 minutes of moderate activity five times a week.

• The main reasons given were that health was not good enough (50%, which seems implausible), lack of time (18%) and lack of interest (15%).

The lowest levels of activity were seen in those with the highest BMI. The proportions who were physically active fell from 44% of men with a good BMI, to 31% in the obese (BMI 30–40 kg/m² and to 16% in the morbidly obese (BMI > 40 kg/m²). The same trend was seen in women: 30% with high activity levels in those with normal BMI to 18% in the obese. Figure 14 shows the proportions achieving the target level of at least 30 minutes of physical activity 5 days per week.

FIGURE 14.

Percentage participation in physical activity vs BMI.

In the Norfolk cohort of the EPIC,290 only 20% of the participants met three or more of the diabetes prevention goals (similar to those in the Finnish DPS183), and only 1% achieved all five; 10% achieved none. 367

The results of the PREPARE (Pre-diabetes Risk Education and Physical Activity Recommendation and Encouragement) trial368–370 give some grounds for optimism. The aim of the PREPARE trial was to see if structured education could increase physical activity and improve glucose tolerance. (It was not a trial of preventing T2DM and so was not eligible for inclusion in Chapter 4). It recruited individuals with IGT who were randomised to usual care, or to the PREPARE education programme, with or without pedometer use. The trial was quite small with initially 103 recruits, falling to 73 by the 2-year follow-up. 368–370 Only 32% of those invited to take part did so. The 12-month follow-up showed no difference in weight, but did show a significant decrease in fasting and 2-hour blood glucose in the pedometer group compared with the control group. The education-only group did not show any difference. The 2-year results showed a continuing benefit in terms of blood glucose with a reduction of 1.6 mmol/l in the pedometer group, although by this time there were only 22 patients in this arm. It would be worth repeating this trial with larger numbers and longer-term follow-up. The intervention was inexpensive.

Could genetic studies help?

Genotyping participants might help to target interventions to those who are at high risk and who will best respond to lifestyle interventions.

Laaksonen et al. (2007)211 examined the interactions of the physical activity, dietary, and weight loss components of the intervention with the 12Glu9 polymorphism of the ADRA2B gene in the development of T2DM in the combined intervention and control groups of the Finnish DPS. 183 Average follow-up time was 4.1 years.

Increased LTPA decreased the likelihood of diabetes more in those with the 12Glu allele of the ADRA2B gene. The RRs of upper compared with lower tertiles for increased LTPA were Glu12/12 = 0.12 (0.03–0.53); Glu12/9 = 0.30 (0.11–0.79) and Glu9/9 = 1.05 (0.32–3.47).

Favourable dietary changes reduced the risk of diabetes more in those who were homozygous for the 9Glu allele. The RRs of upper compared with lower tertiles for dietary changes were Glu12/12 = 0.65 (0.23–1.84); Glu12/9 = 0.85 (0.28–2.27) and Glu9/9 = 0.21 (0.06–0.75).

Weight reduction seemed to decrease the risk of diabetes more in 12Glu9 heterozygotes, but the interaction was not significant. The RRs of upper compared with lower tertiles were, for decrease in BMI, Glu12/12 = 0.48 (0.14–1.60); Glu12/9 = 0.11 (0.05–0.28) and Glu9/9 = 0.81 (0.26–2.48).

A better understanding of the genetic factors responsible for the different responses to various components of lifestyle interventions might allow us to predict who will respond to lifestyle interventions, and to tailor the intervention according to the individual’s genotype.

Ongoing research

Given the success of the big trials, some of the research now is around ‘real-life’ delivery.

The Norfolk DPS371 is an ambitious project, which will screen 10,000 people at risk of diabetes over 5 years, and randomise 950 people with ‘pre-diabetes’ into a 36-month RCT of a novel diet and lifestyle intervention. The intervention contains three arms, and will be delivered by health-care professionals in group settings. One arm will be part delivered by lay mentors who have existing T2DM.

To take part, participants must be > 40 years old, live in the county of Norfolk or be registered with a GP in the county of Norfolk, and meet at least one of the following inclusion criteria: BMI of ≥ 30 kg/m², family history of T2DM, history of CHD or previous GDM, or previous IGT or IFG.

Ambitious although that may be, it is dwarfed by the Qingdao Diabetes Prevention Project. 372,373 This study, sponsored by Helsinki University, aims to translate the trial experience to real-life settings with goals to (1) raise the public awareness of diabetes and diabetes risk factors, and promote healthy diet and physical activity; (2) reduce the number of people at high-risk of developing diabetes through lifestyle counselling; (3) attain early diagnosis of diabetes; and (4) evaluate the clinical effectiveness, cost-effectiveness, feasibility, acceptability and sustainability of the programmes. The project involves community-based targeting of the entire population of 1.94 million people living in four administration districts of the city of Qingdao in China.

The project applies both a population approach and a high-risk approach. In the first phase of the project (2005–8) the work emphasis was on health promotion. In the second phase (2008–12) lifestyle counselling sessions will be provided to about 242,112 high-risk individuals identified, and the efficacy and the cost of the project will be evaluated at the end of the project in 2012.

In Bournemouth, an 8-month intensive lifestyle programme is being tested in obese people without diabetes but with a family history of the condition. The study size is quite small, 66 patients, but the main interest is in levels of glucagon-like peptide 1. 374 The aim is to investigate whether gradual weight loss achieved with healthy lifestyle changes influences hormonal factors affecting appetite and blood glucose control in obese people without the presence of diabetes.

In Canada, a study called PREPARE (Prediabetes Research and Education Promoting Activity and Responsible Eating)375 is being run from Brescia University College in London, Ontario. PREPARE375 is a 6-month community-based pre-diabetes lifestyle and behavioural change programme for adults aged ≥ 30 years with pre-diabetes. It includes a series of six interactive education sessions, of 2 hours each, on healthy eating and physical activity. Individuals self-selecting the control arm receive the current standard of care for pre-diabetes, which is a one-time 2-hour group education session. The primary outcome measure is the average number of vegetable and fruit servings consumed per day, measured at 6 months and 12 months after the baseline assessment.

In the USA, a pilot RCT will compare the clinical effectiveness and cost-effectiveness of two programmes – the DPP system and a community-based Health Living Program – delivered in primary care. 376 The primary outcome measure is 7% reduction in participant weight at 22 weeks, and the estimated enrolment is 200 people. Participants must be aged ≥ 18 years and diagnosed with pre-diabetes.

In Colorado, a RCT called Adaptation of the Diabetes Prevention Program for Primary Care377 aims to assess the efficacy of adding in-person visits to the use of portion-controlled foods for long-term weight loss, and to assess the use of trained lay counsellors for the maintenance of weight loss. Participants will be recruited primarily from primary care practices at the University of Colorado. Up to 200 patients will be provided with 6 months of high-intensity weight loss counselling. Those participants remaining after the first 6 months will be randomly assigned to either standard maintenance or intensified maintenance during months 7–18. The standard maintenance group will receive information handouts regarding weight maintenance, whereas those in the intensified maintenance group will continue to have monthly in-person visits with the weight loss counsellor (‘weight coach’). The primary outcome is weight change.

In East Harlem, in New York, a community-based peer-led programme, HEED (Help Educate to Eliminate Diabetes),378 will randomise overweight adults with pre-diabetes to intervention or usual care, with weight loss as the primary outcome. HEED is a 10-week course. 378 The intervention group will participate in an eight-session course held over a 10-week period. The comparator group will receive a delayed intervention, i.e. they will be offered the chance to participate in the course 1 year after enrolment into the trial. The estimated enrolment is 400 participants.

Also in the USA, the Diabetes Prevention Program Outcomes Study (DPPOS)379 is an observational extension to the DPP. The primary outcome is the development of diabetes, and secondary outcomes include composite microvascular and macrovascular measures. It aims to have 3250 participants.

A small feasibility study from Emory University380 is looking at developing a culturally appropriate lifestyle intervention in South Asians (people with origins in India, Pakistan, Bangladesh, Nepal, Sri Lanka, etc.) living in or near Atlanta, GA. This study will test the acceptability of a culturally appropriate lifestyle intervention for the prevention of diabetes in the South Asian community. The intervention will be based on the DPP but tailored to the needs of the community, based on feedback gathered in focus groups. Participants will be required to attend one group exercise class per week, based on traditional Indian dances and other culturally appropriate activities.

In Catalonia, Spain, a two-stage study, Diabetes in Europe – Prevention Using Lifestyle, Physical Activity and Nutritional Intervention in Catalonia (DE-PLAN-CAT),381 is combining screening methods (FINDRISC vs OGTT) with a later intervention study comparing usual care with individual- or group-based education. A total of 2082 people have been screened, and one-third are expected to present high-risk criteria. They will choose one of three possible interventions to modify their lifestyle (informative approach, one-to-one or group training). Final results are expected in 2016.

The DE-PLAN approach is also being studied in a cluster RCT, conducted by Osakidetza382 in the Basque Country in high-risk (FINDRISC > 14) populations seen in 14 primary care centres. The plan is to recruit over 2500 subjects. The intervention group will receive a structured educational intervention on healthy lifestyles (diet and physical activity) and the control group will receive standard care for the prevention and treatment of T2DM. The primary outcome is the incidence of diabetes at 24 months, and results are expected at the end of 2013.

The Prevention of Diabetes and Obesity in South Asians (PODOSA) study383,384 has screened 1300 people of Indian and Pakistani origin for hyperglycaemia, and has recruited 170 at high risk to an intervention study of healthy eating and increasing physical activity. It is due to report in 2013 (see progress at www.podosa.org/progress.html). Some findings have been published. Gill et al. (2011)383 reported that there was an association between time spent sitting and the 2-hour PG, but not with fasting PG, among all people who were screened for possible inclusions. Douglas et al. (2011)384 reported on the difficulties of recruitment of this group through the usual channels such as GPs and diabetes registers, but noted that there was much greater success using community associations and encouraging recruits to bring in others (‘snowballing’).

Research needs

The highest priority appears to be research into ways of improving adherence to lifestyle measures.

However, given that at least some people will adhere, the next priority is to refine the interventions, addressing questions such as:

• What level of provision or contact is necessary? Could the gains seen in the DPS and DPP be achieved at lower cost?

• How long do interventions need to be continued for?

• What type (or types) of physical activity – in terms of frequency and intensity – will give the best balance between efficacy and adherence?

• How can physical activity be increased?

• How can adherence be improved?

• Does genetic testing have a role to play in determining which intervention should be used?

• What benefits are accrued by those on lifestyle interventions who return to NGT, or who remain in IGT, other than avoiding diabetes?

• Would an intervention, such as the Finnish DPS,183 achieve the same benefits in the UK? Are progression rates similar? We know that at least one UK group, the South Asians, do worse.

• What interventions would be most effective in the highest-risk groups?