Early Outcomes From the English National Health Service Diabetes Prevention Programme

The increase in prevalence of type 2 diabetes is a threat to the sustainability of health systems internationally. There is good evidence from randomized controlled trials that behavioral interventions to support people with impaired glucose tolerance to lose weight, adopt a healthy diet, and increase physical activity can significantly decrease the incidence of type 2 diabetes (1–3). Recent systematic reviews and meta-analyses of trials assessing the effectiveness of pragmatic lifestyle interventions for the prevention of type 2 diabetes in routine practice have demonstrated relative risk reductions in incidence of 26% and 29% (4–6).

There is limited experience in the implementation of diabetes prevention programs (DPPs) at scale. The U.S. DPP reported on 14,747 participants (7), the Australian lifestyle intervention program Life! reported on 8,412 participants (8), and the National Type 2 Diabetes Prevention Program in Finland (FIN-D2D) reported on 2,798 participants (9).

In 2016, the National Health Service (NHS) in England established the Healthier You: NHS Diabetes Prevention Programme (NHS DPP) and just over 2 years later has seen England achieve universal population coverage. The NHS DPP was developed to prevent or delay the onset of type 2 diabetes in adults already identified to be at high risk, which is defined as having nondiabetic hyperglycemia (NDH) (HbA_1c 42–47 mmol/mol [6.0–6.4%] or fasting plasma glucose [FPG] 5.5–6.9 mmol/L). The rationale, justification, development, and early implementation of the program have been described previously (10), and an impact analysis has demonstrated the potential for realizing return on investment within 12 years (11). The approach is based on National Institute for Health and Care Excellence Public Health Guidance on Type 2 Diabetes: Prevention in People at High Risk (NICE PH38) (12) and is complemented by primary preventive interventions to tackle obesity (the major modifiable risk factor for type 2 diabetes), such as a levy on sugar-sweetened beverages, outlined in the U.K. government’s Childhood Obesity Plan (13,14).

Using data from the first 2.5 years of activity, we aimed to assess weight and HbA_1c changes in the NHS DPP and to assess whether these changes are comparable to uncontrolled pre/post summary effect sizes reported in the most recent meta-analyses of pragmatic studies on which some of the assumptions in the impact analysis were based. We also aimed to quantify access through uptake and program completion and to assess the impacts of age, sex, ethnicity, baseline BMI, and socioeconomic status.

A service evaluation in England was used to assess the effectiveness of the NHS DPP through prospectively collected national service–level data related to all people referred from program launch in June 2016 to the end of December 2018.

The NHS DPP delivers behavioral interventions that encourage weight loss or the maintenance of a healthy weight; achievement of U.K. dietary recommendations related to fiber, fruits and vegetables, oily fish, saturated fat, salt, and free sugars (15); and achievement of the U.K. chief medical officers’ physical activity recommendations (16). The intervention is delivered according to a national service specification by one of four service providers selected through a national competitive process: Reed Momenta (London, U.K.), ICS Health & Wellbeing (Leeds, U.K.), Ingeus UK (London, U.K.), and Living Well Taking Control (Birmingham, U.K.). The specification was developed by an expert group on the basis of the evidence for clinical and cost effectiveness and on the suggested mechanisms for achieving behavior change described in NICE PH38 (12). These include information provision to raise awareness of the benefits and types of lifestyle changes needed to achieve and maintain a healthy weight, exploration and reinforcement of participants’ reasons for wanting to change and their confidence about making changes, goal setting, action planning, coping plans, and relapse prevention.

Each provider’s service follows the same broad structure of an initial assessment, core sessions, and maintenance sessions, with a minimum total of 13 face-to-face group-based sessions, over at least 9 months, constituting a least 16 h contact time. Each provider must use a known framework for behavior change.

Individuals are eligible if they have a blood test indicating NDH conducted within the previous 12 months and are ≥18 years of age, not pregnant, and not previously diagnosed with type 2 diabetes. Individuals are identified after an NHS Health Check (17), through retrospective searches of general practice records, or through routine clinical practice. Individuals referred are invited to attend an initial assessment at which further program details are provided, and participants are assigned to a group for intervention delivery.

All program providers are contractually required to collect a minimum data set, including demographic and clinical information. Age, sex, postcode, and referral HbA_1c or FPG measurement are recorded at referral receipt. Ethnicity, weight, and height are recorded at initial assessment. Coaches employed by the provider measure the participant’s body weight in light indoor clothing at each intervention session using class 3 scales. Providers assess HbA_1c values for each participant at the initial assessment if the referral HbA_1c or FPG is >3 months old, at 6 months after the first intervention session, and at the end of the program for those still attending. This service evaluation involves assessment of anonymized data collected during routine service delivery; NHS England has published an information governance framework setting out the legal basis for data collection and data flows, ensuring that the service and its evaluation are delivered in compliance with data protection legislation (18).

Individual factors (age, sex, ethnicity, socioeconomic status, baseline BMI, and number of sessions attended) and program factors (provider) were identified as potential outcome moderators. Sex was recorded as male, female, or indeterminate. Participants were grouped into 5-year age bands and self-reported ethnicity as white, Asian, black, mixed, or other. Socioeconomic status was measured using quintiles of the Index of Multiple Deprivation associated with the lower super output area derived from participant postcodes (19). All variables also include an unknown category where either the participant declined to give the relevant information or a value was not recorded. BMI was calculated, and participants were classified as healthy weight/underweight, overweight, or obese as defined according to their reported ethnicity; if their ethnicity was not known or not recorded, participant BMIs were classified according to the white ethnicity group in line with World Health Organization thresholds (20).

The primary outcomes for the evaluation were change in weight and change in HbA_1c analyzed on an intention-to-treat basis. In secondary analyses, data from participants who completed the program were assessed separately.

Weight change, percentage weight change, and the proportion of participants who achieved a weight loss of ≥5% were calculated for all participants associated with cohorts that had had time to finish the program. The baseline measurement was defined as the weight measured at the first intervention session attended to avoid including weight change during the period between initial assessment and intervention commencement. Weight change greater than 5 SDs from the mean was deemed erroneous and recorded as missing.

All providers elected to assess point-of-care tested (POCT) HbA_1c values. POCT devices used by providers were the DCA Vantage (Siemens Healthcare, Guildford, U.K.), Afinion (Abbott Diagnostics, Maidenhead, U.K.), and A1CNow+ (BHR Pharmaceuticals, Nuneaton, U.K.). The same device was used for repeated measures within individuals. Consistent with a recent systematic review and meta-analysis (21), there was a significant negative bias for POCT HbA_1c values compared with referral, laboratory-measured HbA_1c values, greater than could be attributed to regression to the mean and greater than concurrent weight change would suggest was attributable to behavior change between referral and initial assessment. Therefore, mean HbA_1c change was calculated only for the subgroup of participants who had had their HbA_1c measured at initial assessment so that all values for the same individual had been derived using the same device.

Program retention was assessed by following cohorts of participants who attended at least one intervention session; those associated with cohorts where sufficient time had elapsed to have reached the final session were defined as having finished the program. Completion of the program was defined as attendance of at least 60% of sessions (at least 8 sessions for three providers who offered 13 sessions, at least 11 sessions for one provider who offered 18 sessions). This aligns with the a priori criterion used for provider payment, where providers were paid for participants who attended ≥60% at each of five milestones. Completion rates were calculated with the number of participants who had attended at least one intervention session as the denominator.

Intention-to-treat weight change analysis included participants for whom all data fields, except HbA_1c, were complete, with weight change calculated as the weight difference between the first and last sessions attended. HbA_1c change analysis included the subset of these participants who also had an HbA_1c measurement at initial assessment, with HbA_1c change calculated as the HbA_1c difference between initial assessment and the last value recorded. Data for program completers was assessed in secondary analyses.

Sensitivity analyses were conducted using multiple imputation that used multivariate chained equations to impute missing data and then comparing the results to the primary analyses. Both univariate and multivariate analyses were repeated using multiple imputation data sets, and results were compared to ensure that missing data did not introduce bias.

Because of time delays between referral and attendance at initial assessments and first intervention sessions, the proportion of participants who attended either the initial assessment or at least one intervention session was calculated using the number of referrals received up to December 2017 as the denominator, with numbers of corresponding attendees either at initial assessment or at an intervention session, respectively, by December 2018 as the numerator.

A mixed-effects logistic regression model was used to identify characteristics associated with program completion. Age, sex, ethnicity, baseline BMI, deprivation, and provider were considered as fixed effects and local referral area as a random effect, with the contribution of the random effect quantified using the intraclass correlation coefficient (ICC). Local referral areas are only associated with a single provider and, therefore, incorporate the same facilities and facilitators used by that provider. Variation among the four providers was directly accounted for by a fixed effect in the model. Mixed-effects linear regression models were used to identify factors associated with change in weight and change in HbA_1c. The number of sessions attended and provider and participant characteristics were considered as fixed effects and local referral area as a random effect.

Statistical significance was defined as P < 0.05, and CIs were set at 95%. All data were analyzed using Stata 15 software.

Between June 2016 and the end of December 2018, 324,699 people were referred to the program. Of these, 152,294 had attended an initial assessment, and 96,442 had attended at least one of the group-based intervention sessions. Including only those referred up until the end of December 2017, to allow sufficient time to have joined a group, 53% attended an initial assessment, and 36% attended an intervention session. There were 32,665 participants who had attended an intervention session and had had sufficient time to finish the program. Among these participants, 17,252 (53%) attended at least 60% of sessions for an overall completion rate of 19% of those referred. Figure 1 outlines the number of participants at each stage of the program.

Characteristics of participants at each program stage are shown in Table 1: 46% were male, the mean (SD) age was 62 (13) years, and there was broadly equal representation from all deprivation quintiles at referral. Mean HbA_1c at referral was 43.7 (1.5) mmol/mol (6.1% [0.1%]). Ethnicity and weight were not recorded until the initial assessment, at which point 20% of participants were of black, Asian, mixed, or other ethnicity; 69% were white; and 11% were unknown. The mean weight was 83.9 (19.1) kg at initial assessment, and the mean BMI was 30.3 (6.1) kg/m². The largest decrease in the proportions of people retained was between referral and initial assessment, where there were significant decreases in the proportions of males, people aged <60 years, and people from the most deprived quintile (all P < 0.001). Except for the proportion of males (P = 0.52), these decreases continued between initial assessment and program completion (all P < 0.001).

Completion, weight change, and HbA_1c change were assessed for participants associated with cohorts that had finished the program. Of those, 26,753 (82%) had no missing or unknown data (excluding HbA_1c). There were no missing data for age, provider, local referral area, and number of sessions attended. Data were missing for participant postcode (and therefore deprivation quintile) (0.3%), sex (0.5%), BMI (7%), and ethnicity (10%). Data on weight was missing at either baseline or end of program for 7% of participants. There were 19,891 (61%) participants who had their HbA_1c measured using a POCT device at initial assessment, of whom 16,083 had no missing data (49% of all participants).

Univariate analyses of primary outcomes are provided in Table 2 (with secondary outcomes in Supplementary Table 1). The mean (SD) number of days in the program was 179.8 (136) and the mean number of intervention sessions attended was 8.2 (4.6). For the providers offering a total of 13 sessions, the mean number of sessions attended was 7.6 (3.8), and for the provider offering 18 sessions, the mean number of sessions attended was 9.6 (5.8). The regression analysis indicated that participants who were older (up to 70 years of age), from less deprived backgrounds, and with a lower BMI were more likely to complete the program, but there was no effect of sex. Relative to white groups, Asian and mixed ethnic groups had lower completion rates, with no significant differences for other ethnic groups. There were significant differences in completion by provider (Supplementary Tables 2 and 3). Clustering by local referral area made a proportionately small contribution to the outcomes (ICC 3.9% [95% CI 2.1, 7.2]).

Using an intention-to-treat analysis, mean baseline weight was 83.4 kg with a mean weight change of −2.3 kg (95% CI −2.3, −2.2). Mean percentage weight change was −2.7% (−2.7, −2.6), and 24% of participants lost ≥5% of baseline weight. Weight loss increased with the number of sessions attended (Fig. 2). The regression analysis indicated that for each additional session attended, there was a 0.32-kg greater weight loss, and for each 1-kg higher baseline weight, there was an additional 0.03-kg weight loss. Older participants (up to 75 years of age), men, those from areas in the least deprived quintile, and those with a higher BMI lost more weight. Asian and black ethnic groups lost less weight, with no significant differences for other groups. There were significant differences by provider independent of the number of sessions in their program (Supplementary Table 4). The ICC was 0.4% (95% CI 0.2, 0.8).

The mean baseline POCT HbA_1c was 41.8 mmol/mol (6.0%), with a mean change of −1.26 mmol/mol (95% CI −1.31, −1.20) (−0.12% [−0.12, −0.11]). HbA_1c change increased with the number of sessions attended (Fig. 2). The regression analysis indicated that for each additional session attended, there was an additional 0.18 mmol/mol (0.02%) decrease in HbA_1c, and for each 1 kg in weight reduction, there was a 0.15 mmol/mol (0.01%) reduction in HbA_1c. For each 1 mmol/mol (0.09%) increase in baseline HbA_1c, there was a further corresponding decrease of 0.32 mmol/mol (0.03%). There were significantly smaller HbA_1c reductions for older participants, women, those from the most deprived quintile, and those with a higher BMI. There was a significantly smaller HbA_1c reduction for black participants, with no significant differences for other ethnic groups (Supplementary Table 4). The ICC was 1.3% (95% CI 0.7, 2.3).

For completers, the mean baseline weight was 82.4 kg with a mean weight change of −3.3 kg (95% CI −3.4, −3.2). The mean percentage weight change was −4.0% (−4.0, −3.9), and 37% of participants lost ≥5% of weight. The mean baseline POCT HbA_1c was 41.8 mmol/mol (6.0%) with a mean change of −2.04 mmol/mol (−2.12, −1.96) (−0.19% [−0.19, −0.18]). Analysis of characteristics associated with outcomes gave similar results to the intention-to-treat analysis, although weight loss did not differ by sex (Supplementary Table 5).

Sensitivity analysis using imputed data showed that there were no substantive changes in direction and magnitude of the associations (Supplementary Tables 6–9). Statistically significant covariates in the complete-case analysis remained significant in the imputed analysis, with the exception of a single age band (60–64 years) for weight change and mixed ethnicity for completion, the latter placing some uncertainty on the significance of mixed ethnicity. A number of subcategories were significant in the imputed analysis but not in the complete-case analysis. In these cases, there were no substantial differences in the magnitude and sign of the associated coefficient, and no suggested change to the interpretation.

A national program to provide behavioral support to people with NDH in England was associated with a significant reduction in weight and HbA_1c among the 36% of people referred who attended at least one of the group-based intervention sessions. There was a clear dose-response relationship, and people who attended more sessions experienced greater reductions in both weight and HbA_1c.

This report describes the largest cohort of people to our knowledge to be offered an intervention within a DPP, achieving universal population coverage. It includes objective measures of weight, HbA_1c, individual participant data, and flow through the program and assesses the impact on health inequalities. There are some missing data, and we have taken a principled and pragmatic approach to consider the effects on data interpretation. Our sensitivity analyses do not vary in terms of the direction and broad magnitude of the findings in the primary analyses, providing some reassurance that the missing data have not appreciably biased our conclusions. The uncontrolled nature of this analysis means that external confounders cannot be excluded, and there may have been other factors leading to weight loss and HbA_1c reductions, including secular trends.

Data on HbA_1c change are only available for half the participants in whom a baseline measure of POCT HbA_1c was performed so that it was directly comparable to subsequent measurements. However, given the cause of these missing data, it is unlikely that this has introduced a specific bias. Moreover, the characteristics of the subset are similar to the complete data set. At an individual level, the clinical significance of a 1.26 mmol/mol (0.12%) reduction in HbA_1c is difficult to gauge because very few data are available internationally with regard to HbA_1c reductions in response to interventions in the nondiabetic range. However, a left shift in HbA_1c distribution of 1.26 mmol/mol is likely to be significant at the population level.

A current limitation is that the last recorded weight available is that measured at the last session attended, and at present, we do not have data on longer-term outcomes. However, mechanisms for the acquisition of longer-term data have been established, and meanwhile, initial weight loss is a strong predictor of weight loss outcomes in subsequent years (22). From 2017/2018, the National Diabetes Audit in England, which involves extracts from health care data sets held in primary care settings and hospitals, was expanded to include people at high risk of type 2 diabetes, including those with NDH and, hence, eligible for the NHS DPP (23). Data will be systematically extracted for those coded with NDH and linked with the NHS DPP data set, permitting longitudinal tracking of HbA_1c and, therefore, type 2 diabetes incidence, weight, and other recorded cardiovascular risk factors, including blood pressure and lipids, microvascular and cardiovascular disease incidence, and mortality. Recent 30-year follow-up data from the Da Qing Diabetes Prevention Outcome Study demonstrated that lifestyle intervention in people with impaired glucose tolerance, in addition to delaying the onset of type 2 diabetes, also reduced the incidence of cardiovascular events, microvascular complications, and cardiovascular and all-cause mortality and increased life expectancy (3).

The NHS DPP design was based on a Public Health England–commissioned systematic review and meta-analysis assessing the effectiveness of pragmatic lifestyle interventions for the prevention of type 2 diabetes in routine practice over 12–18 months. This found such programs to be associated with weight loss of 2.5 kg (95% CI 1.99, 2.99) and HbA_1c reduction of 0.07% (95% CI 0.01, 0.14) (4,5). The pooled incidence rate ratio of type 2 diabetes among patients attending a DPP compared with those receiving usual care was 0.74 (95% CI 0.58, 0.93), a reduction of 26%. A more recent global systematic review and network meta-analysis of pragmatic DPP studies reporting effects on incidence, weight, and glycemic parameters demonstrated a similar relative risk reduction of 29% associated with 2.5 kg (95% CI 1.90, 3.00) weight loss but no evidence of a reduction in HbA_1c (6). NHS DPP data demonstrate similar weight loss and greater reductions in HbA_1c, providing optimism that this program may lead to reductions in future type 2 diabetes incidence among participants.

The U.S. DPP reported a mean percentage weight reduction of 4.2% (7), greater than the 2.7% weight loss seen in the NHS DPP intention-to-treat analysis, although the U.S. analysis only included participants who had attended at least four intervention sessions. When compared with the 4.0% weight loss seen among completers in the NHS DPP, the results are similar. The Finnish DPP reported weight losses of 1.3 kg in men and 1.1 kg in women (9), and the Australian DPP reported losses of 1.4 kg for participants completing sessions 1–5 and 2.5 kg for participants completing session 1–6 (8). Differences in weight loss across the four programs may reflect differences in intensity, ranging from a median of 14 sessions in the U.S. DPP, a mean of 6 in the Australia DPP, and a mean of 2.9 in the Finnish DPP compared with 8 in the NHS DPP (7–9). Differences in starting weight may also have been contributory: U.S. DPP participants had a mean baseline measurement of 96.8 kg; Australian DPP participants, 87.3 kg; and Finnish DPP, 95.8 kg in men and 83.8 kg in women compared with 83.4 kg for NHS DPP participants. None of the programs in other countries have reported the effects on glycemic parameters, although the Finnish DPP reported beneficial effects on type 2 diabetes incidence and cardiovascular risk factors (9).

Beyond the national DPPs, it is unusual for behavioral programs to take a whole-population approach. We are only aware of one randomized controlled trial of a behavioral intervention for weight loss that was offered opportunistically in primary care. The Brief interventions for Weight Loss (BWeL) trial found that 40% of people offered support attended, and 24% completed, a 12-week program (24), similar to the proportions seen in the NHS DPP, which is a longer program.

People from more deprived areas were less likely to complete the program, lost less weight, and had smaller reductions in HbA_1c. Similarly, the Brief Interventions for Weight Loss trial found that participants from lower socioeconomic backgrounds attended fewer sessions, leading to less weight loss (25).

Black, Asian, mixed, and other ethnic groups are overrepresented in those attending an initial assessment (26), but the adjusted odds ratio of completion among Asian groups is 25% lower than in white groups. Asian and black groups lost less weight, and black groups had smaller reductions in HbA_1c. Uniquely, the effect of ethnicity is independent of socioeconomic status.

Program engagement, retention, and adherence are crucial to attain the desired effects. The findings highlight the need to actively target engagement, retention, and adherence in specific groups to avoid widening inequalities. There has already been a new round of provider procurement for the NHS DPP, with newly appointed providers starting in August 2019. The payment schedule has been adjusted to provide greater incentives to providers to retain participants of black, Asian, mixed, and other ethnicity and those from more deprived backgrounds. Recognizing that a large proportion declined or failed to attend the face-to-face group-based interventions, digital modes of program delivery will be offered for those who decline or fail to attend the face-to-face interventions. Such programs have been shown to be associated with weight loss, although the effects on glycemic control are less clear (27). A large uncontrolled pilot of digital prevention interventions conducted in live service environments in England is currently under way (28).

In summary, reductions in weight and HbA_1c demonstrated in the NHS DPP are encouraging and compare favorably with those reported in recent meta-analyses of pragmatic studies. Furthermore, they are potentially indicative of future reductions in participant type 2 diabetes incidence.

Funding. NHS England funded program development, implementation, and evaluation. K.K. acknowledges support from National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care East Midlands and the NIHR Leicester Biomedical Research Centre. S.J. is funded by the NIHR Oxford Biomedical Research Centre and NIHR Collaboration for Leadership in Applied Health Research and Care Oxford.

Duality of Interest. All authors have completed the International Committee of Medical Journal Editors uniform disclosure form and declare the following: J.V. is the national clinical director for diabetes and obesity at NHS England and is the clinical lead for the Healthier You: NHS DPP. E.B. is the head of health intelligence (diabetes) for Public Health England and leads analysis of the DPP. D.B. is an analyst for NHS England and is actively involved in analysis of the program. C.B. is the primary care advisor to the NHS DPP. J.F. was the diabetes evidence and evaluation lead at Public Health England until September 2018. S.O. is the clinical director at Diabetes UK. B.Y. is clinical lead of the National Diabetes Audit for England and Wales and a trustee of Diabetes UK. N.W. was chair of the program development group for NICE Public Health Guidance on Type 2 Diabetes Prevention: Population and Community-Level Interventions (NICE PH35). K.K. was chair of the program development group for NICE PH38. K.K. is also co-director of the Leicester Diabetes Centre, and one of the program providers, Ingeus UK, provided interventions on the basis of the type 2 DPP developed by Leicester Diabetes Centre. J.S. represents Public Health England on the NHS England Diabetes Programme Board. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors are members or attendees of the Healthier You: NHS Diabetes Prevention Programme Expert Reference Group, chaired by J.V. Members from inception (J.V., S.O., B.Y., N.W., K.K., S.J.) were responsible for developing the service specification for the behavioral intervention and for developing the data fields for the minimum data set. All authors advised on the operational delivery of the program and on the interpretation of data derived via the minimum data set on a quarterly basis since inception in 2015. J.V., E.B., and D.B. were responsible for evaluation design and initial manuscript drafting. E.B. and D.B. performed the data analyses, and J.F. advised on the overall analytic design and multiple imputation. C.B., J.F., and S.J. contributed to early draft development. All authors reviewed drafts of the manuscript and provided constructive feedback and criticism. J.V. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility of the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the Diabetes UK Annual Professional Conference, Liverpool, U.K., 6–8 March 2019, and the European Association for the Study of Diabetes Annual Conference, Barcelona, Spain, 16–20 September 2019.