Reduction of diabetes risk in routine clinical practice: are physical activity and nutrition interventions feasible and are the outcomes from reference trials replicable? A systematic review and meta-analysis

The search was confined to English language articles published between January 1990 and August 2009. Three authors (MC-M, LR, AB) separately interrogated different data sources using the same search terms (see appendix). This was supplemented with hand searches of the reference sections of other systematic reviews [2, 18, 19, 25‐40]. Only studies which investigated at least one of our research questions above, and which were consistent with our inclusion criteria below, were considered in this review (Figure 1).

The review focused on translational research studies where interventions were based on any of the large reference diabetes prevention RCTs mentioned above. These could be: replication studies in the form of RCTs, before/after evaluations, cohort studies with or without a control group, or interrupted time series analyses, where participants have been exposed to a lifestyle intervention of at least 3 months duration and followed up for at least 3 months. Routine clinical practice was defined as a health service setting providing patient care such as primary health clinics, hospital outpatient clinics or specialist medical centres.

Interventions were classified as single (either nutrition or physical activity programs with or without medication), or combined nutrition and/or physical activity programs (structured or unstructured) whether or not they included medication. Structured intervention components were defined as those in which participants received a standard set of sessions with instructions on specific dietary and/or physical activity requirements and goals. In unstructured interventions participants were given generic advice on healthy living without specific goals other than improving diet or physical activity in relation to baseline. The comparison group might be 'no intervention group' or an 'alternative intervention' (single or combined). Prevention programs delivering diabetes education materials only were excluded. Likewise, medication-only studies were excluded. Only programs conducted in routine health services, delivered on-site or in associated facilities, with outcomes measured in healthcare settings by general medical practitioners, specialist physicians, practice nurses, dietitians, physiotherapists, allied health professionals, community health staff, or research staff attached to the health service, were included in this review. Interventions either had to be replications or modification of all or some components of the US Diabetes Prevention Program [DPP] [1] or Finnish Diabetes Prevention Study [DPS] [5] or any other reference trial, or had to include the reduction of diabetes risk or diabetes incidence explicitly as a goal or objective.

Participants were adult men or women with any degree of impaired glucose regulation (impaired fasting glucose or impaired glucose tolerance) or with normal glycaemia but at risk of diabetes as determined by risk factors such as obesity or family history. They may have been recruited from the primary or other healthcare patient clientele or from the general population but had to receive the intervention through routine healthcare services. Participants' risk of diabetes may have been determined by a diabetes risk score, either measured or from self-report, and may have had accompanying blood glucose tests to either identify impaired glucose regulation or exclude diabetes before receiving the intervention. Studies including patients with diagnosed diabetes were included in this review only if they were a replication of the reference trials and whose outcomes were reported separately from participants without diabetes.

Studies were included if they reported at least one of the following main outcome measures of interest:

The secondary outcome examined was:

To assess the potential for bias, and given the heterogeneity of studies included in this review, a generalisability and bias assessment tool covering elements of various checklists and resources from the literature was specifically designed. Items examined included participant recruitment source, selection criteria, treatment allocation, blindness of outcome assessment, simultaneous collection of data for intervention and control groups, measurement error, subgroup analysis and discussion of study limitations (tool available from the authors on request). Reference tools used for this design were STROBE, COCHRANE Collaboration, CLEAR NPT, EQUATOR, PRISMA, TREND and MOOSE [41‐46]. One of the authors (MC-M) conducted the bias and quality assessment of all studies and three other authors (LR, SM & PTE) independently conducted the second bias and quality assessment of some of the studies. Two of the authors (MC-M & SM) independently extracted results and assessed the appropriateness of statistical analyses and conclusions.

Study quality was assessed and graded on the following criteria: (1) evidence of assessment of risk for diabetes at enrolment; (2) explicit eligibility and exclusion criteria; (3) reported participation rate of at least 50% of eligible people; (4) follow-up assessment rates of ≥ 65% of program participants by study conclusion or follow-up; (5) evidence of measurable or explicit outcome assessment; (6) appropriate statistical methods, including adequate control for confounders (in non-RCTs); (7) explicit intervention components; (8) conclusions supported by findings. A numeric score giving equal weight to each of the above criteria was used to determine quality. The maximum possible score was thus 8, indicating highest quality.

The denominator for the effect sizes was the number of subjects in whom the outcome had been assessed. Study results were categorised as positive (statistically significant difference observed), negative (no difference or statistically significant negative effect), or inconclusive (showed no difference but lacked sufficient power to detect a difference). Given the heterogeneity of designs, length of follow-up and outcome measurements of the available studies, pooling of selected results for a meta-analysis was feasible only for four RCTs reporting 12-month follow-up results [47‐50]. The remaining eight studies were critically reviewed but not meta-analysed. Changes in means, and tests of heterogeneity between trials were calculated using random effects models. When not reported in individual studies, standard deviations of mean differences in outcome measures were calculated from supplied study participant numbers and standard errors or from 95% confidence limits, either of before-and-after means or from before-and-after differences in mean values. Meta-analysis was conducted using NCSS software version 7.1.1.9 [51] on the four main outcomes of interest: changes in weight, fasting plasma glucose, waist circumference and 2-hour OGTT.

Sensitivity analysis by study quality was not deemed necessary as all four studies finally selected for meta-analysis had a quality score of 7 or 8 out of the possible 8 maximum score. Our search did not identify unpublished replication studies of diabetes prevention in routine clinical practice. Accordingly, we expected findings not to be significantly affected by publication bias.

All studies included a combined lifestyle intervention but two eligible studies included a medication arm in addition to lifestyle. Seven studies attempted replication of the reference trial approaches from either the U.S DPP [1] or the Finnish DPS [81] with adaptation to routine clinical practice, mostly to cater for limitations in practitioner's time and health service budgets [14, 48, 49, 78, 82‐84]. Modifications included: shorter duration of program (2/7); delivery of group sessions instead of individual face-to-face counselling (4/7); reduced number and frequency of individual or group counselling sessions to which participants were exposed (5/7); and mixed group and individual program approaches (1/7).

Modifications of interventions during the maintenance phase included intermittent support sessions, more economical versions of the resources given to participants, and multidisciplinary teams, either available on site or hired as an additional service. For interventions delivered in a group-based modality, the maximum number of sessions per program was 16, as per the reference trial [1] (median of 6 sessions), but over a shorter period of time. Among the 5 studies reporting delivery of individual counselling sessions, the median number of individual counselling sessions was 13.5.

All dietary interventions were structured and half the physical activity interventions were unstructured (Table 2). While some interventions were delivered with a core intensive phase and an intermittent approach for the maintenance phase, the median duration of intervention was 32 weeks; follow-up periods also varied from 4 to 60 months with a median follow-up duration of 12 months. Delivery of the modified versions of the reference trial interventions was mostly by nurses, psychologists or allied health staff such as health promotion counsellors, dietitians or exercise physiologists alone (8/12) who provided the training, demonstration, counselling or education sessions. Physicians were mainly involved in assessing participant eligibility, referral and outcome measurement (7/12). Two studies did not report the professional background of people delivering the intervention or assessing the participants [49, 80].

Reported measured outcomes of interest were weight (12/12), fasting plasma glucose (9/12) waist circumference (7/12), and 2-hour OGTT (3/12) (Table 3). Six studies had follow-up periods enabling the examination of diabetes incidence or incidence reduction, with the remainder confined to reporting risk improvement via behavioural modification or improvement in metabolic or anthropometric parameters. Self-reported dietary and physical activity outcomes of interest amenable to statistical comparison were not often reported and were confined to mean reduction in fat intake as a percentage of total energy (3/12), and changes in fibre intake (3/12). These are summarised in Table 4. Due to the heterogeneity of units used for measuring and repporting changes in physical activity, it was not possible to meta-analyse these outcomes.

While three of the 12 studies justified their sample sizes on statistical grounds, and not all adjusted for potential confounders, the quality of study design and reporting overall was good in 10 of the 12 studies included, based on quality criteria scores of 7 or 8 out of 8 (Table 1). Two studies were considered suboptimal, with quality scores of 3 or 5 out of 8 respectively [14, 84].

Limited generalisability was identified in five studies, where participants recruited were either self-referred healthy volunteers [50] or a convenience sample of males only [49, 79], or mostly severely obese middle-age women [14, 78]. Two studies reported higher success rates for participants who had already met the goals at baseline [78, 82]. In two studies [14, 84] the intervention incurred charges and out-of-pocket expenses for each session, which lead to differential exposure to intensity and duration of intervention on the basis of participant's ability to pay. Participation rates for the 8 studies reporting them were satisfactory (median 83.5%). However, in one of the studies, where the participation rate was ostensibly 100%, the control group comprised all those people who did not participate due to financial reasons (on whom outcome measures were collected, but possible exposure to other risk reduction regimes was not recorded) [84].

Loss-to-follow up rates in the 12 studies varied greatly, from 5% to 57% (median 14%). Differential withdrawal rates were reported in a further three studies, where a larger proportion of drop-outs were observed: in participants at highest baseline risk [78]; in those from the intensive arm of the intervention [50]; or in subjects who perceived a poor response to the allocated treatment [80].

The meta-analysis showed that the pooled weight loss in the intervention group from the four RCTs yielded a weight loss of 1.82 Kg at one year (Figure 2), less than the 5.6 Kg loss observed in the lifestyle-only group of the reference DPP trial or the 4.2 Kg reported in the intervention group of the Finnish DPS. While all studies showed a positive effect on weight loss, only four of the seven studies, 1 RCT and 3 B-A studies [14, 79, 80, 84] reported weight changes at 1 year similar in magnitude to the DPP in the US (around 5 Kg, Table 5). Excepting the XENical in the Prevention of Diabetes in Obese Subjects [XENDOS] trial [80], studies reporting proportions achieving a pre-defined weight loss goal of 5% or 7% were less encouraging. Most studies reported half or less of participants than in the US reference DPP trial, where 50% of participants achieved 7% weight loss at 6 months [1], or in the Finnish DPS where 43% of participants achieved 5% weight loss at 1 year.

The one-year improvements in fasting plasma glucose were similar to the DPP across several studies but were too small to be clinically important; and reductions of diabetes incidence in the two studies reporting them at 12 months follow-up [79, 83] were somewhat less (37% and 23% respectively) than the reductions apparent from the cumulative risk/incidence plots in the Finnish (~70-80%) and DPP (~70%) trials. For the five studies in this review measuring waist circumference, all concluded that waist circumference reductions were possible with modified lifestyle interventions, but after 1 year only two achieved reductions of sufficient magnitude that cannot be attributed to measurement error (≥ 4 cm) [85]. Decreases in fat consumption and increases in fibre consumption resulting from interventions generally were not reported, and the few studies that did showed no substantial improvements. The exception was the Absetz et al. trial which reported half the participants meeting the fibre goal and achieving the total fat intake goal and a third achieving the saturated fat goal [82].

Nine of the 12 studies explored whether translation of the reference trials into clinical care was feasible. Eight concluded that modification of the original trial approaches for adaptation to real life practice made the lifestyle interventions feasible, affordable or replicable in clinical care settings despite barriers to implementation [14, 49, 78, 82‐84, 86, 87]. The remaining study reported that the transferability of the results from original trials to other settings remains questionable, as the positive effect on outcomes diminishes over time [48].

Seven trials which randomised a total of 4,905 participants to lifestyle intervention or control were identified. The shortest follow-up period was 4 months and the longest follow-up period was six years. Four of these, randomising a total of 1,129 to intervention or control, reported selected outcomes in comparable units at one year [48‐50, 87]. These were meta-analysed, although not all outcomes of interest were available from all these studies (Figure 2). We chose not to meta-analyse outcomes at four [80] or six years [79], as these relate to the maintenance phase of a program rather than the medium term impact and it would be inappropriate to compare them with one-year results.

The systematic review of RCT results at 12-month follow-up showed: mean weight reduction was 1.82 Kg greater in treatment than control groups which was statistically significant (95% CI:-2.7 to -0.99 Kg); pooled mean waist measurement reductions in treatment exceeded control groups by 4.6 cm, and this was also significant (95% CI:-5.8 to -3.4 cm); fasting plasma glucose reduction was 0.19 mmol/l greater in treatment than controls but non-significant (95% CI: -0.44 to +0.06 mmol/l); and a non-significant greater increase in 2-hour oral glucose tolerance test result of 0.04 mmol/l (95%CI: -0.49 to +0.42 mmol/l). From the above, it is apparent that the interventions can achieve significant weight and waist measurement reductions at one year but do not significantly change the main metabolic indicators of diabetes risk such as FPG or OGTT.

Four of the 12 studies achieved the greatest weight loss, i.e. 5 Kg or more at 12 months. As only two of these successful studies had optimal quality scores [79, 80], we further examined the characteristics of these studies to identify common determinants of success in diabetes prevention programs. Common features were RCT design, being based in Sweden, and having long interventions (1 and 4 years) and longer follow-up periods (6 years in Malmo, 4 years in XENDOS). They were not replication studies, and the frequency of participant contact was amongst the highest, with Malmo providing 12 group sessions and XENDOS providing up to 54 individual counselling sessions. Another common feature was that following initial substantial weight loss, the final outcome after several years of follow-up was only an average of 3 Kg weight loss in both studies. We may conclude that the outcomes of these two studies involve social, cultural and health system characteristics unique to that part of Europe that may not be generalisable.

To our knowledge, this is the first attempt to comprehensively compile feasibility and effectiveness of translation of diabetes prevention trials specifically for routine clinical settings.

We used a purpose-built comprehensive quality scoring system based on individual components of relevance from checklists widely used by others in quality assessment of the literature. Our quality criteria allowed for the inclusion of several study types to maximise the chances of identifying and assessing relevant diabetes prevention programs. The search was extensive and individual study authors were contacted to either confirm that their study was conducted under routine clinical care or to exclude any translation study conducted in research settings or under simulated clinical care. Meta-analytic techniques were used when feasible.

Despite the good quality of papers covered in this review, the total number of studies finally included was small; some were exploratory (3 pilots) and many of them had short follow-ups and only modest sample sizes which essentially reflect the financial and time restrictions of real-life interventions in routine clinical practice. We included studies with intervention and follow-up durations of at least 3 months. These are not unusual in routine practice, as modifications to duration and intensity of the strict approaches in the reference trials are common in the replication literature. While longer interventions and follow-up times are ideal, in real-world situations longer studies inevitably are affected by sample attrition and attendant generalisability issues. We wanted to include some measurement of short-term impact and avoid attrition bias and selection bias in our assessment of what is being evaluated in routine practice and therefore we allowed for feasibility and pilot studies to be incorporated.

Analyses from before-and-after studies often did not report on adjustment for confounders. More importantly, the reporting of outcomes of interest was often incomplete or in disparate units of measurement precluding inclusion in the meta-analysis. However, the overall good quality of these studies enabled their inclusion in the broader systematic review.

Many weight-loss-only programs and other lifestyle interventions for reduction of cardiovascular disease risk were excluded as they did not specifically mention replication of the diabetes prevention trials. However, we acknowledge that results from these may also be applicable to diabetes risk reduction, and while examples of reviews of these are available in the literature, their focus is beyond the scope of our review.

Diabetes, Pre-diabetes, Type 2 diabetes, Impaired glucose tolerance OR glucose intolerance, Lifestyle intervention OR lifestyle program OR strategy, Physical activity OR Exercise OR Resistance Training, Healthy eating OR diet OR dietary modification OR weight loss, Behavioural modification, AND (Primary health care, General practi$., clinical practice, routine clinical care), AND (Prevent$. Ti, ab, Translating OR Translation OR Translat$. Ti, ab., Translation research OR translational study OR Replication study).