A randomised trial of the effect and cost-effectiveness of early intensive multifactorial therapy on 5-year cardiovascular outcomes in individuals with screen-detected type 2 diabetes: the Anglo Danish Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care (ADDITION-Europe) study

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/116/300. The contractual start date was in July 2010. The draft report began editorial review in August 2014 and was accepted for publication in December 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Torsten Lauritzen reports unrestricted grants from Novo Nordisk, AstraZeneca, Pfizer, GlaxoSmithKline, Servier and HemoCue during the conduct of the study. Torsten Lauritzen also owns stock/shares in Novo Nordisk. Guy EHM Ruttem reports grants from Novo Nordisk, GlaxoSmithKline and Merck during the conduct of the study and personal fees from Novo Nordisk and AstraZeneca outside the submitted work. Melanie J Davies has acted as a consultant, advisory board member and speaker for Novartis, Novo Nordisk, Sanofi-Aventis, Eli Lilly and Company, Merck Sharp & Dohme, Boehringer Ingelheim and Roche. Melanie J Davies has also received grants in support of investigator and investigator-initiated trials from Novartis, Novo Nordisk, Sanofi-Aventis, Eli Lilly and Company, Pfizer, Merck Sharp & Dohme and GlaxoSmithKline. Kamlesh Khunti has acted as a consultant and speaker for Novartis, Novo Nordisk, Sanofi-Aventis, Eli Lilly and Company and Merck Sharp & Dohme and has received grants in support of investigator and investigator-initiated trials from Novartis, Novo Nordisk, Sanofi-Aventis, Eli Lilly and Company, Pfizer, Boehringer Ingelheim and Merck Sharp & Dohme. Simon J Griffin reports grants from the Wellcome Trust, the Medical Research Council, NHS Research and Development, the National Institute for Health Research and the University of Aarhus (Denmark) and non-financial support from Bio-Rad Laboratories during the conduct of the study. Simon J Griffin has also received personal fees from Eli Lilly and Company, the Royal College of General Practitioners and Astra Zeneca outside the submitted work.

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Simmons et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Burden of diabetes

Type 2 diabetes is a common chronic disease, which affected 382 million people worldwide in 2013. 1 It is estimated that there will be 592 million individuals living with diabetes by 2035. Diabetes is a major cause of premature death. Global mortality attributable to known diabetes in adults aged 20–79 years in the year 2013 is estimated at 5.1 million deaths, which is 8.4% of total world mortality. 1 Diabetes is ranked among the leading causes of blindness, renal failure and lower limb amputation in virtually every developed society. People with diabetes have an increased risk of cardiovascular, cerebrovascular and peripheral vascular disease (PVD) and have a reduced risk of survival after suffering from silent ischaemia and myocardial infarction (MI) (compared with those without diabetes). Around 65% of individuals with type 2 diabetes die of cardiovascular disease (CVD). Diabetes is a very expensive chronic condition. It imposes a huge burden on national economies and health-care systems, as well as costs for individuals with diabetes and their families. Expenditure related to diabetes in 2010 was estimated to be approximately 10% of total health-care budgets in the UK and is projected to rise to 17% in 2035. 2

Preventing diabetes

The high social and economic cost of diabetes makes a compelling case for prevention of the disease. But where should governments and health services focus their attention? Most research has been completed on the tertiary prevention of diabetes, that is, the treatment of people with established disease. There have been significant improvements in the treatment of individuals with diabetes3 and there is good evidence that the development of long-term complications of diabetes can be significantly decreased by intensive treatment (IT) (see Screening and early treatment for early diabetes). We also have long-term evidence that diabetes can be prevented among those at high risk of developing the disease (primary prevention). 4 Intensive lifestyle and pharmacological interventions reduce the rate of progression of type 2 diabetes among people with impaired glucose tolerance (IGT). In a meta-analysis of published diabetes prevention trials, Gillies et al. 4 reported pooled hazard ratios of 0.51 [95% confidence interval (CI) 0.44 to 0.60] for lifestyle interventions compared with standard advice and 0.70 (95% CI 0.62 to 0.79) for oral diabetes drugs compared with the control. Longer-term follow-up of these trials provides evidence of the sustained prevention of diabetes. 5–7 Delaying or preventing diabetes in this way also reduces the risk of microvascular complications and reduces cardiovascular and all-cause mortality. 8,9

Secondary prevention strategies (i.e. earlier detection, e.g. by screening) have received little attention in the past. The current evidence base for the recommendation of screening and early treatment for diabetes is limited.

Screening and early treatment for diabetes

Type 2 diabetes meets many of the formal criteria for a disease for which screening is justified. 10 The condition is an important health problem associated with a substantial burden of suffering and health service cost. The natural history of the disease is well characterised. 11,12 The condition is frequently asymptomatic,13 with the true onset occurring several years before diagnosis. 11,14 Although detection of the condition may be improving in some parts of the world,15 nearly half of all people with diabetes remain undiagnosed. 1 When patients are diagnosed, many already have complications, such as CVD, chronic kidney disease and heart failure, retinopathy and neuropathy. 16–18 This suggests a potential window for earlier detection and treatment. Furthermore, there are a number of screening tests that are simple, safe and validated and that perform reasonably well when evaluated against recommended diagnostic criteria. 19 Modelling studies suggest that a programme of screening for diabetes would reduce both all-cause and diabetes-related mortality. 20–22 However, these estimates depend on a number of key assumptions. The only published trial of screening to date did not show an effect of population-based screening on mortality over 10 years of follow-up. 23

The National Screening Committee states that there should be an effective treatment for individuals identified through early detection, with evidence of early treatment leading to better outcomes than late treatment. 10 A screening programme for diabetes is most likely to seek to prevent CVD, the leading cause of premature death and disability among patients with diabetes. There is good evidence that the development of long-term complications of diabetes can be significantly decreased by IT. Results from the UK Prospective Diabetes Study (UKPDS) and the Kumamato trials have demonstrated the benefits of tight glycaemic24,25 and blood pressure control,26 whereas several trials, including the Collaborative Atorvastatin Diabetes Study27 and the Heart Protection Study (HPS),28 have confirmed the benefits of lipid-lowering drugs. The Steno study is one of the few trials to compare the benefits of targeted intensive multifactorial treatment with those of routine care (RC) for risk factors for CVD in individuals with established diabetes. 29 After 13 years of follow-up, patients receiving the IT had a 59% (95% CI 0.25% to 0.67%) lower risk of a CVD event and a 46% (95% CI 0.32% to 0.89%) lower risk of all-cause mortality than those receiving RC. When we started our research in 2010, there was no trial evidence that intensive multifactorial treatment improved CVD outcomes when commenced in the lead time between detection by screening and diagnosis in routine clinical practice.

In terms of preventing microvascular complications, treatment of individual risk factors such as blood pressure and glucose level reduces the risk of microvascular complications among clinically diagnosed patients. 24,30–32 IT of multiple risk factors in the Steno study was associated with a 61% (95% CI 13% to 83%) lower risk of nephropathy, a 58% (95% CI 14% to 79%) lower risk of retinopathy and a 63% (95% CI 21% to 82%) lower risk of autonomic neuropathy. 33 However, the effects on microvascular outcomes of starting multifactorial treatment earlier in the course of the disease are uncertain. Data from trials of IT of hyperglycaemia suggest that beneficial effects can be seen for microvascular outcomes in the short term, whereas cardiovascular benefits are evident only with longer follow-up. However, there remains some uncertainty about the merits of tight glycaemic control. 34,35

Little research has been completed on the potential effects of intensive multifactorial treatment on patient-reported outcome measures (PROMs) early in the course of the disease. For largely asymptomatic patients, such a treatment regime might be burdensome. Health practitioners might be reluctant to offer IT including the prescription of several medications and recommendations to change several lifestyle behaviours; this might lead to psychosocial stress and reduced satisfaction with treatment. 36 When assessing the effectiveness of early treatment, PROMs are a valuable complement to hard outcomes such as mortality and cardiovascular events. They are also increasingly being used as key performance indicators in chronic illness. PROMs reflect a patient’s assessment of his or her own health and well-being and involve questions about physical and social functioning and mental well-being. They may include both generic and disease-specific questions. The reliability of PROMs is similar to that of clinical measures such as blood pressure or blood glucose monitoring. 37 The use of PROMs has been recommended in the evaluation of health-care services and in regulatory decision-making. 38 Their use provides an opportunity to help drive changes in how health care is organised and delivered. 37

Evidence from the UKPDS study suggests that IT of blood pressure and blood glucose among newly diagnosed type 2 diabetes patients is not associated with adverse effects on quality of life. 39 The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial of intensive glycaemic control in US patients with long-standing type 2 diabetes was associated with modest improvements in satisfaction with diabetes treatment and did not lead to an increase in health-related quality of life. 40 The effects of intensive multifactorial treatment on PROMs among people with type 2 diabetes detected by screening are not known.

In addition to a lack of information on the effects of intensive multifactorial intervention on macrovascular, microvascular and PROMs early in the diabetes disease trajectory, little is known about the cost-effectiveness of such treatment. It is expected that increasing numbers of new patients will be identified as governments introduce national assessment programmes, such as the NHS Health Checks programme. 41 The balance of benefits, harms and costs of IT may not be the same for screen-detected individuals as for those with clinically diagnosed and long-standing diabetes.

The ADDITION-Europe trial

In 2001 a group of colleagues from Cambridge in the UK, Utrecht in the Netherlands and Copenhagen and Aarhus in Denmark came together to answer some of the outstanding questions about screening and early treatment for diabetes. Colleagues in Leicester were later invited to join the collaboration. The main aim of the Anglo–Danish–Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care (ADDITION-Europe) was to investigate whether or not intensive multifactorial treatment improves outcomes compared with RC when commenced in the lead time between detection by screening and clinical diagnosis. This was a two-phase study consisting of a screening phase and a pragmatic, cluster-randomised, parallel-group trial. Results from the screening phase of the study have previously been reported. 42–46 This report concerns the results from the 5-year follow-up of the trial, which was funded in the UK by the National Institute for Health Research (NIHR) Health Technology Assessment programme.

Aims

To examine the effect of intensive multifactorial treatment compared with RC on (1) cardiovascular outcomes, (2) microvascular outcomes and (3) self-reported health status, well-being, diabetes-specific quality of life and treatment satisfaction after 5 years’ follow-up. A further aim was to assess the cost-effectiveness of intensive multifactorial treatment compared with RC in the UK setting.

Design

The ADDITION-Europe trial was set up to evaluate the effectiveness of intensive multifactorial treatment with regard to macrovascular, microvascular and PROMs among individuals with screen-detected diabetes. It consisted of two phases – a screening phase and a pragmatic cluster-randomised parallel-group trial – in four centres (Denmark, Cambridge, UK, the Netherlands and Leicester, UK). This report concerns the results of the treatment trial (see Chapter 3). The main trial was supplemented with an economic evaluation to consider the cost-effectiveness of the intervention in the UK (see Chapter 4). A description of the trial protocol has already been published. 47

Ethical approval and research governance

The study was approved by local ethics committees in each centre and participants provided informed consent.

Randomisation, concealment and blinding

Of 1312 general practices invited to participate, 379 (29%) agreed and 343 clusters (26%) were randomised. Practices were randomly assigned by statisticians independent of the measurement teams to screening plus routine diabetes care or screening followed by intensive multifactorial treatment in a 1 : 1 ratio. Randomisation included stratification by county and number of full-time family physicians in Denmark and by single-handed or group status in the Netherlands. In Cambridge, randomisation included minimisation for the local district hospital and the number of patients per practice with diabetes. In Leicester, randomisation included minimisation for practice demographics, deprivation status and prevalence of type 2 diabetes.

Population-based stepwise screening took place between April 2001 and December 2006 among people aged 40–69 years (50–69 years in the Netherlands) without known diabetes, as previously described. 43,48–50 Screening programmes, which varied by centre (Table 1), consisted of a risk score51 (Cambridge) or self-completion questionnaires (Denmark52 and the Netherlands54) followed by capillary glucose testing or an invitation to attend an oral glucose tolerance test without prior risk assessment (Leicester). Individuals were diagnosed with diabetes according to the World Health Organization (WHO)’s 1999 criteria,56 including the requirement for confirmatory tests on separate occasions.

Inclusion and exclusion criteria

All patients newly diagnosed with type 2 diabetes were eligible to participate in the treatment study unless their family physician indicated that they had contraindications to the proposed study medication, an illness with a life expectancy of < 12 months or psychological or psychiatric problems that were likely to invalidate informed consent. Overall, 3057 eligible participants with screen-detected diabetes agreed to take part (Denmark, n = 1533; Cambridge, n = 867; the Netherlands, n = 498; Leicester, n = 159).

Intervention

The characteristics of the interventions to promote IT in each centre have been described previously47–49,57 and are summarised in Table 1. We aimed to educate and support general practitioners (GPs) and practice nurses in target-driven management (using medication and promotion of a healthy lifestyle) of hyperglycaemia, blood pressure and cholesterol, based on the stepwise regimen used in the Steno-2 study. 29 Treatment targets and algorithms (Table 2) based on trial data in people with type 2 diabetes24,26,29,58,59 were the same for the IT groups in all centres. GPs were advised to consider prescribing an angiotensin-converting enzyme (ACE) inhibitor for patients with blood pressure ≥ 120/80 mmHg or a previous cardiovascular event58 and 75 mg of aspirin daily to patients without specific contraindications. Although treatment targets were specified and classes of medication recommended, prescribing decisions, including choice of individual drugs, were made by practitioners and patients. Following publication of the results of the HPS,28 the treatment algorithm included a recommendation to prescribe a statin to all patients with a cholesterol level of ≥ 3.5 mmol/l.

Intensive treatment was promoted through the addition of several features to existing diabetes care. Small group or practice-based educational meetings were arranged with GPs and nurses to discuss the treatment targets and algorithms and lifestyle advice, including supporting evidence. Audit and feedback were included in follow-up meetings up to twice per year (the total number of practice meetings ranged from 2 to 10) or co-ordinated by post. In the Netherlands patients were seen in general practice by diabetes nurses who were authorised to prescribe medication and adjust doses under GP supervision. In Denmark and Cambridge practice staff were provided with educational materials for patients. In Denmark and the Netherlands patients were sent reminders if annual measures were overdue. In all centres practices received additional funding to support the delivery of care (up to the equivalent of three 10-minute consultations with a GP and three 15-minute consultations with a nurse per patient per year for 3 years). Leicester patients were referred to the Diabetes Education and Self-Management for Ongoing and Newly Diagnosed (DESMOND) structured education programme53 and were offered 2-monthly appointments with a diabetes nurse or physician in a community peripatetic clinic for 1 year and 4-monthly appointments thereafter. Clinic staff were prompted to contact patients defaulting from appointments.

In the RC group, GPs were provided only with the diagnostic test results. Patients with screen-detected diabetes received the standard pattern of diabetes care according to the recommendations applicable in each centre. 60–63

Data collection

Centrally trained staff undertook health assessments at baseline and after 5 years, including biochemical and anthropometric measures, and administered questionnaires, following standard operating procedures. Staff were unaware of study group allocation. Follow-up examinations took place from September 2008 until the end of December 2009. The mean [standard deviation (SD)] follow-up period was 5.3 (1.6) years. All biochemical measures were analysed in five regional laboratories at baseline and follow-up. Standardised self-report questionnaires were used to collect information on education, employment, ethnicity, lifestyle habits (smoking status, alcohol consumption), prescribed medication and health status. Questionnaires were completed at the same health assessment visit as the anthropometric and biochemical measurements. If participants did not complete follow-up questionnaires or measurements then the most recent values were obtained from general practice records along with information on prescribed medication.

Primary end point

The primary end point was a composite of first cardiovascular event, including cardiovascular mortality, cardiovascular morbidity (non-fatal MI and non-fatal stroke), revascularisation and non-traumatic amputation. All-cause mortality and each of the individual components of the primary end point were secondary outcomes. In each centre participants’ medical records or national registers were searched for potential end points by staff unaware of group allocation. For each possible end point, packs containing relevant clinical information (such as a death certificate, post-mortem report, medical records, hospital discharge summary, electrocardiographs and laboratory results) were prepared and sent to two members of the expert committees, who were unaware of group allocation, for independent adjudication according to an agreed protocol using standardised case report forms. Committee members met to reach consensus over discrepancies. The date of completion of follow-up for the primary end point was deemed to be the date of the first primary end point, the date of remeasurement at 5 years if no end point occurred or the date that the end-point search was undertaken if the participant did not experience an event or attend follow-up.

Secondary end points

Sample size

An individually randomised trial would have required a total of 2700 individuals (1350 per group) to detect a 30% reduction in the cumulative risk of the primary end point at a 5% significance level and with 90% power, allowing for 10% loss to follow-up and assuming an event rate in the RC group of 3% per year (based on results from the UKPDS24). We expected a minimal effect of clustering within general practices, with an estimated intracluster correlation coefficient of 0.01; assuming an average of 10 participants per general practice, the design effect was 1.09 and so we inflated the estimated sample size for this cluster trial to 3000.

Statistical analysis

The analysis and reporting of this trial were undertaken in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines [see www.consort-statement.org/ (accessed 27 January 2016)]. Statistical analysis was undertaken in Stata 11.1 (StataCorp LP, College Station, TX, USA), SAS v9.2 (SAS Institute Inc., Cary, NC, USA) and Review Manager v5.1 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark), following a predefined analysis plan, which was finalised before preparation of the end-point data set, agreed with the Trial Steering Committee and deposited on the study website (see www.addition.au.dk/). The primary comparative analyses between the randomised groups were conducted on an intention-to-treat (ITT) basis without imputation of missing data.

Patient and public involvement

In the development of the ADDITION-Cambridge study, patients were involved in two pilot studies. The first was to assess the feasibility and uptake of the diabetes screening programme and to examine the effects of invitation to diabetes screening on anxiety, self-rated health and illness perceptions. 79 The second was a qualitative study of patients’ experiences of being screened for diabetes. 80 Further qualitative work was undertaken exploring practitioners’ experiences of taking part in the main ADDITION-Cambridge study. 81 In terms of patient involvement during the 5-year follow-up phase of the trial, we sent a newsletter to all ADDITION-Cambridge participants in 2008 outlining the results to date and informing them of our intention to invite them back for the 5-year follow-up assessment. We said to participants: ‘[a]ny ideas that you may have on the conduct and planning of the 5-year follow-up health check would be welcomed – please get in touch if you have any comments’. Responses fed into our 5-year planning. We also sent participants a newsletter in May 2012 with the 5-year study results and invited them and a guest to attend a local meeting during June/July 2012 where they had the chance to meet with ADDITION staff and hear about the results of the study. The five meetings were very well received and gave this special group of patients a chance to ask questions about the study and their diabetes treatment. Comments ranged from ‘very informative’ and ‘the content was useful and excellent’ to ‘the results of the study will motivate me to work harder at controlling my diabetes’. There was very strong support for being involved with a 10-year follow-up study. Similar feedback to study participants took place in the Netherlands and Denmark.

In ADDITION-Leicester, the intervention model was particularly suited to patient and public involvement, with ongoing contact with the clinical care team facilitating continuity and dialogue. IT participants were consulted and offered continued post-trial care delivered in a setting of their choice (primary or secondary care). A number of participants have become members of the local patient and public involvement group and are actively involved in steering local research. A newsletter describing the final results and thanking participants for their involvement was sent following completion of the 5-year outcome ascertainment for the entire cohort (July 2014). A significant number of participants were identified as being at high risk of diabetes during the ADDITION-Leicester study and have subsequently been involved with other studies aimed at the prevention of diabetes. We have also developed a black and minority ethnic panel that has contributed more widely to diabetes research in hard-to-reach groups.

Primary analyses: cardiovascular outcomes

Primary end-point data were available for 99.9% (3055/3057) of participants. In total, 238 first cardiovascular events occurred during a mean (SD) follow-up period of 5.3 (1.6) years (Table 4). There were 121 events among 1678 participants (7.2%) in the IT group (incidence 13.5, 95% CI 11.3 to 16.1 per 1000 person-years) and 117 events among 1377 participants (8.5%) in the RC group (incidence 15.9, 95% CI 13.3 to 19.1 per 1000 person-years). The hazard ratio for the IT group compared with the RC group was 0.83 (95% CI 0.65 to 1.05; p = 0.12). The cumulative probability for the primary CVD end point appeared to diverge after 4 years of follow-up (Figure 5). In predefined subgroup analyses there were no interactions between the intervention and age or previous cardiovascular event (p > 0.1). The p-value was calculated using Cox regression and fixed-effects meta-analysis.

FIGURE 5.

Cumulative incidence of the composite cardiovascular end point in the RC and IT groups of the ADDITION-Europe trial (p = 0.12). Source: reproduced open access from Griffin et al. 55

However, estimated hazard ratios were 0.70 (95% CI 0.52 to 0.95) in those aged ≥ 60 years and 1.12 (95% CI 0.70 to 1.79) in those aged < 60 years. Hazard ratios for individual components of the composite end point all favoured the IT group (see Table 4), although none achieved statistical significance (Figure 6). There were no amputations as first events. The intracluster correlation coefficient for the primary end point was 0.002 (Denmark 0.014, UK 0.0000016, the Netherlands 0.025). This suggests that the cluster design had little effect on study power.

FIGURE 6.

The relative risk of the development of cardiovascular death, non-fatal MI, non-fatal stroke and revascularisation as a first event and the composite cardiovascular end point by country and overall in the IT group compared with the RC group. Source: reproduced open access from Griffin et al. 55

In total, there were 196 deaths (n = 60 cardiovascular, n = 97 cancer and n = 39 other), 104 (6.2%) in the IT group (mortality rate 11.6, 95% CI 9.6 to 14.0 per 1000 person-years) and 92 (6.7%) in the RC group (mortality rate 12.5, 95% CI 10.2 to 15.3 per 1000 person-years) (see Table 4). The combined mortality hazard ratio for the IT group compared with the RC group was 0.91 (95% CI 0.69 to 1.21). A Kaplan–Meier plot is shown in Figure 7 and country-specific hazard ratios in Figure 8. The heterogeneity of results between countries was not statistically significant. In the UK there were significantly fewer deaths in the IT group, whereas in Denmark there was a non-significant reduction in cumulative risk in the RC group.

FIGURE 7.

Kaplan–Meier estimates of all-cause mortality in the RC and IT groups in the ADDITION-Europe trial. Source: reproduced open access from Griffin et al. 55

FIGURE 8.

The relative risk of all-cause mortality by country and overall in the IT group compared with the RC group in the ADDITION-Europe trial. Source: reproduced open access from Griffin et al. 55

Secondary analyses: microvascular outcomes

Of the 2861 patients still alive at 5 years, 2493 (87.1%), 2710 (94.7%) and 2312 (80.8%) had data for urinary ACR, eGFR and peripheral neuropathy, respectively. Retinal photographs were retrieved for 2190 (76.5%) participants.

At 5 years’ follow-up any albuminuria was present in 316 (22.7%) participants in the IT group and 269 (24.4%) participants in the RC group. Macroalbuminuria was present in 56 (4.0%) and 37 (3.4%) participants in the IT and RC groups, respectively. Centre-specific ORs for any albuminuria favoured the IT group, but the pooled OR was not statistically significant (0.88, 95% CI 0.72 to 1.07; Figure 9). The pooled OR for macroalbuminuria was 1.15 (95% CI 0.76 to 1.74). In both groups the urinary ACR increased between baseline and follow-up. In the IT group the mean (SD) increase was 1.45 (SD 0.60) mg/mmol and in the RC group the mean (SD) increase was 1.30 (0.66) mg/mmol. The overall difference in means was –0.02 (95% CI –0.96 to 0.91) mg/mmol. There were no significant interactions between study group and any of the subgroups.

FIGURE 9.

Odds ratios and frequencies of any retinopathy, any peripheral neuropathy and any albuminuria at follow-up by study group in the ADDITION-Europe trial. Source: reproduced from Sandbæk et al. ,83 under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 licence (CC BY-NC-ND 3.0).

The mean (SD) eGFR increased between baseline and follow-up in both the IT group and the RC group [IT 4.31 (0.49) ml/minute, RC 6.44 (0.90) ml/minute; difference in means –1.39 (95% CI –2.97 to 0.19) ml/minute]. There were no significant interactions between treatment group and any of the subgroups concerning eGFR. The number of missing data was equally distributed between the two groups.

Retinopathy was present in 125 (10.1%) participants in the IT group and 116 (12.1%) patients in the RC group. Centre-specific ORs favoured the IT group but the pooled OR was not statistically significant (0.84, 95% CI 0.64 to 1.10) (see Figure 9). Imputation of missing values did not affect the estimates.

Individuals without retinal images at follow-up had significantly higher baseline mean HbA_1c levels than individuals with retinal data [7.18% (55 mmol/mol) vs. 6.99% (53 mmol/mol), respectively; p = 0.044]. However, there was no difference between randomised groups (interaction p-value = 0.78). There was a significant interaction between retinopathy, randomised group and baseline HbA_1c level (p = 0.007). IT appeared to be more effective among individuals with HbA_1c ≥ 6.6% (≥ 49 mmol/mol) at baseline (OR 0.65, 95% CI 0.45 to 0.93) than among individuals with HbA_1c < 6.6% (< 49 mmol/mol) at baseline (OR 1.17, 95% CI 0.75 to 1.82). There was no evidence of an interaction with either age or sex. Severe retinopathy was present in one participant in the IT group and in seven participants in the RC group.

Peripheral neuropathy was present in 63 (4.9%) participants in the IT group and 60 (5.9%) participants in the RC group (pooled OR 0.95, 95% CI 0.68 to 1.34) (see Figure 9). Non-responders to the neuropathy questionnaire had a higher body mass index (p = 0.055) and were more likely to be from a ethnic minority group (p = 0.004) than responders. Imputation of missing values did not affect the estimates.

The overall intracluster correlation coefficient values were as follows: retinopathy 0.014 (95% CI 0.00017 to 0.55), albuminuria 0.025 (95% CI 0.0066 to 0.091) and neuropathy 0.011 (95% CI 4.6 × 10^–7 to 1). These results indicate that the impact of clustering on study power was small.

Secondary analyses: patient-reported outcomes

Data from 2861 participants were included in the multiple imputation analyses. Patients who completed questionnaires at baseline and follow-up (n = 2217) were more likely than those who did not complete questionnaires (n = 644) to be male (59.5% vs. 50.8%), of white ethnicity (95.2% vs. 89.9%) and employed (44.8% vs. 32.7%). Questionnaire completers were also less likely to smoke (24.3% vs. 34.0%), had higher levels of alcohol consumption (median 5 units per week vs. 3 units per week) and had higher EQ-5D scores (median 0.85 vs. 0.81) at baseline. They also had lower systolic blood pressure levels (148.5 mmHg vs. 151.4 mmHg). Other characteristics were comparable between treatment groups (data not shown).

Table 5 shows the PROM scores at follow-up, separately for each centre and by randomised group. EQ-5D values did not change between diagnosis and follow-up: the median (interquartile range) score was 0.85 (0.73 to 1.00) at baseline and 0.85 (0.73 to 1.00) at 5 years’ follow-up.

The mean differences in PROMs comparing the IT group with the RC group are shown in Figures 10–13. There were no statistically significant differences in health status (see Figure 10), well-being (see Figure 11), diabetes-specific quality of life (see Figure 12) and satisfaction with diabetes treatment (see Figure 13) between the two groups. There was some heterogeneity between centres. The I²-statistic varied between 41% (W-BQ-positive) and 73% (EQ-VAS).

FIGURE 10.

Mean difference in health status between the IT group and the RC group by centre (boxes) after 5 years’ follow-up in the ADDITION-Europe trial and pooled estimates (diamonds) calculated by random-effects meta-analysis. Horizontal bars and diamond widths denote 95% CIs and box sizes indicate relative weight in the analysis. MCS, mental component summary; PCS, physical component summary. Source: reproduced from van den Donk et al. ,84 under the Creative Commons Attribution Noncommercial licence.

FIGURE 11.

Mean difference in W-BQ12 scores between the IT group and the RC group by centre (boxes) after 5 years’ follow-up in the ADDITION-Europe trial and pooled estimates (diamonds) calculated by random-effects meta-analysis. Horizontal bars and diamond widths denote 95% CIs and box sizes indicate relative weight in the analysis. Source: reproduced from van den Donk et al. ,84 under the Creative Commons Attribution Noncommercial licence.

FIGURE 12.

Mean difference in ADDQoL scores between the IT group and the RC group by centre (boxes) after 5 years’ follow-up in the ADDITION-Europe trial and pooled estimates (diamonds) calculated by random-effects meta-analysis. Horizontal bars and diamond widths denote 95% CIs and box sizes indicate relative weight in the analysis. Source: reproduced from van den Donk et al. ,84 under the Creative Commons Attribution Noncommercial licence.

FIGURE 13.

Mean difference in DTSQ scores between the IT group and the RC group by centre (boxes) after 5 years’ follow-up in the ADDITION-Europe trial and pooled estimates (diamonds) calculated by random-effects meta-analysis. Horizontal bars and diamond widths denote 95% CIs and box sizes indicate relative weight in the analysis. Source: reproduced from van den Donk et al. ,84 under the Creative Commons Attribution Noncommercial licence.

Multiple imputation analyses resulted in slightly different point estimates. However, the overall patterns remained the same and there were no statistically significant differences in any of the PROMs (results not shown).

Validating the UK Prospective Diabetes Study outcomes model

Accurate prediction of cardiovascular risk among individuals with diabetes is important for providing prognostic information and targeting treatment to those at highest risk. It is also a key component of economic evaluations of interventions aimed at reducing the disease burden of diabetes and improving the quality and length of life for those with this chronic condition.

The UKPDS-OM is a cost-effectiveness analysis tool. It was derived from the UKPDS, a multicentre randomised controlled trial among 5102 individuals with newly diagnosed type 2 diabetes. 85 Participants were recruited from 23 UK centres. 86 The UKPDS-OM has been validated for the prediction of CVD and other complications in internal and external cohorts of people with clinically diagnosed diabetes. The observed and predicted cumulative incidence of CVD complications from diagnosis to 12 years’ follow-up were very similar when using an internal cohort. 85 Results varied from poor to moderate in terms of discrimination, calibration and absolute risk for external cohorts,87,88 with the UKPDS-OM tending to overestimate CVD risk. 89

The UKPDS-OM was developed using data from individuals recruited between 1977 and 1997. The treatment and costs of diabetes and related complications have changed since this time. The variables assessed by the model are also likely to vary over time and between countries. As such, we aimed to evaluate the performance of the UKPDS-OM in the ADDITION-Europe study, which included a multinational contemporary cohort of individuals with screen-detected diabetes.

Cost-effectiveness analysis

In our second analysis we aimed to examine the short- and long-term cost-effectiveness of IT compared with RC among people with screen-detected type 2 diabetes.

Long-term cost-effectiveness analysis

Intensive treatment was associated with positive incremental QALYs in the long term (0.0465 by 30 years, statistically significant at 20 years and beyond). The incremental cost of IT compared with RC at 10, 20 and 30 years also increased over time to £1745 at 30 years, yielding a point estimate ICER of £82,250 at 10 years, which fell to £35,000 at 20 years and rose slightly to £37,500 at 30 years (see Table 11). The unadjusted results suggest a lower point estimate QALY gain in the IT arm, which is reversed once adjustment is made for baseline differences. The result suggests that, although cost-effectiveness improves over time, it is still above commonly accepted thresholds in the UK (defined as reaching an ICER of £30,000 per QALY gained),105 even over a 30-year time horizon (Figure 14).

FIGURE 14.

Broken line chart showing the simulated ICERs at 10, 20 and 30 years for the ADDITION-Europe intervention. Source: reproduced from Tao et al. ,99 under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).

Analysis of uncertainty

Uncertainty and sensitivity analyses were performed on the predicted 30-year results. The cost-effectiveness plane based on bootstrap sampling shows the scatterplot of cost and QALY pairs under the base case (intervention cost of £981 per person) and two alternative scenarios with lower IT intervention costs (Figure 15) (£750 and £500 per person). Under these scenarios, 30-year point estimate ICERs are £37,503, £32,550 and £27,178. The cost at which the ICER falls to £30,000 is £631. If the intervention could be delivered ≤ £631 per patient, it may be considered cost-effective.

FIGURE 15.

Cost-effectiveness plane showing pairs of incremental costs and QALYs from bootstrap samples using three different costs of delivering the IT intervention. The points to the right of the lines are cost-effective. Source: reproduced from Tao et al. ,99 under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).

Under all three scenarios, the majority of points are in the north-east quadrant, suggesting that IT is nearly always both more expensive and more effective (generates more QALYs) than RC, although the proportion of the probability mass in the north-west quadrant suggests that there is greater uncertainty around whether or not incremental QALYs are positive. The probability of cost-effectiveness according to the three different treatment costs (£981, £750 and £500) is 51.1%, 60.9% and 70.4% at a £20,000 per QALY gained threshold and 65.1%, 71.1% and 77.0% at a £30,000 per QALY gained threshold, respectively (Figure 16).

FIGURE 16.

Cost-effectiveness acceptability curves showing the probability of IT being more cost-effective than RC based on net benefit values from bootstrap samples using three different costs of delivering IT. The two dotted lines show the cost-effectiveness acceptability thresholds of £20,000 and £30,000 per QALY. Source: reproduced from Tao et al. ,99 under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).

One-way sensitivity analyses varying unit treatment costs, utility decrements and the discount rate showed that the discount rate had the biggest impact on the ICER, with the impact of variations in utility decrements and treatment costs being minimal (Figure 17).

FIGURE 17.

Tornado diagram showing the influence of changing different parameters that contribute to the ICER in the long-term cost-effectiveness modelling analysis. The choice of discount rate has the greatest impact on the ICER (lower discount rate, unit costs and utility decrements all associated with a higher point estimate ICER). Source: reproduced from Tao et al. ,99 under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).

Change in treatment and risk factors

Results from the ADDITION-Europe trial show that screening for type 2 diabetes and early intensive multifactorial treatment of individuals found to have screen-detected diabetes is feasible in primary care. Screening identifies people with high levels of modifiable cardiovascular risk factors. The proportion of participants prescribed one or more blood pressure-, lipid- and glucose-lowering drugs and aspirin increased significantly from baseline to 5 years’ follow-up in both study groups. There were small differences in treatment between groups at 5 years favouring the IT group.

In both the IT group and the RC group, intermediate outcomes such as blood pressure and cholesterol improved considerably in the 5 years following diagnosis by screening; weight and glycaemia levels did not increase. In terms of change from baseline, and absolute levels of risk factors at 5 years, our results compare well with those of other trials in which patients were recruited at diabetes diagnosis and followed for 6 years. 110,111 However, the between-group differences observed at 1 year,49 which were clinically significant, were not maintained at 5 years and differences were smaller than those achieved in similar studies. 29,110 It is possible that adherence to treatment targets were suboptimal in this pragmatic trial. In three of the centres, the screening algorithm included the use of risk scores encompassing treatment for high blood pressure as one of the variables to predict diabetes risk. This limited the achievable differences in blood pressure levels between groups. Indeed, at baseline in the UKPDS, which began recruitment 20 years before the ADDITION-Europe study, 12% of patients were prescribed antihypertensive medication and 0.3% were prescribed lipid-lowering medication. In the ADDITION-Europe study, at baseline, 45% of patients were prescribed antihypertensive medication and 16% were prescribed lipid-lowering medication. Our trial was undertaken during a time of improvements in the delivery of diabetes care in general practice. This included the introduction of evidence-based guidelines in Denmark112 and the Netherlands113 and the Quality and Outcomes Framework for primary care in the UK. 114 Again, these guidelines and frameworks are likely to have reduced the achievable differences in treatment and risk factors between groups.

Despite increasing age and duration of diabetes there was a significant decline in 10-year modelled CVD risk in the 5 years following diagnosis in both the RC group [–5.0% (SD 12.2%)] and the IT group [–6.9% (SD 9.0%)]. Within all four centres the modelled CVD risk was lower in the IT group than in the RC group at 5 years. Comparable improvements in the CVD risk factors that drive modelled CVD risk have been reported among newly diagnosed patients enrolled in lifestyle interventions over 12 months. 54,115 Such changes were also observed among clinically diagnosed diabetes patients in the UKPDS trial at 6 years of follow-up. 110

Cardiovascular outcomes and all-cause mortality

The intervention to promote intensive target-driven management of individuals with screen-detected diabetes was associated with a non-statistically significant 17% relative reduction in the incidence of a composite cardiovascular event end point over 5 years. For all components of the primary end point, differences favoured the IT group. Differences were smallest for stroke and largest for MI. Cardiovascular event rates began to diverge after 4 years of follow-up. We found no evidence of an increased risk of adverse events including self-reported hypoglycaemia or mortality. The incidence of cardiovascular events in the ADDITION-Europe study (8.5% in the RC group) was lower than expected and less than event rates among newly diagnosed patients in the UKPDS (12.1%). 110

The combined mortality hazard ratio for the IT group compared with the RC group was 0.91 (95% CI 0.69 to 1.21). The heterogeneity in results between countries was not statistically significant. In the UK there were significantly fewer deaths in the IT group. In Denmark there was a non-significant reduction in cumulative risk in the RC group. The mortality rate (6.7% in the RC group) was also lower than expected. In the Hoorn study116 the mortality rate was 25% over 10 years in screen-detected patients and in newly diagnosed patients in Denmark111 the mortality rate was 33% over 7.4 years. The mortality rate in the ADDITION-Europe trial was similar to that seen in the Danish general population (without diabetes) of the same age between 1995 and 2006. 117 As before, this is most likely because of the high quality of care that was delivered to patients early in the course of the diabetes disease trajectory in both the RC group and the IT group.

Microvascular outcomes

Trial results for microvascular outcomes were similar to those for the cardiovascular outcomes. At 5 years, increases in prescribed treatment and improvements in cardiovascular risk factors were not associated with significant reductions in microvascular events. Differences between study groups favoured the IT group, with differences smallest for neuropathy and greatest for retinopathy. As for the CVD outcomes, the frequency of microvascular complications at 5 years’ follow-up was lower than expected. It was also lower than the reported frequencies among individuals with diabetes for a similar length of time. 32 We concluded that this was most likely because of the early detection of diabetes and the high-quality treatment that we observed in both study groups.

In terms of the nephropathy outcomes, 18.5% of ADDITION-Europe participants had any albuminuria at diagnosis. 50 This figure is not directly comparable to the estimate of 7.2% of patients with microalbuminuria at baseline in the UKPDS as a higher level of urinary albumin excretion was used to define microalbuminuria in the UKPDS. 32 The rate of progression of albuminuria in the ADDITION-Europe study, from 18.5% at baseline to 23.5% (1% per year), was half that observed in the UKPDS, despite ADDITION-Europe participants being older (mean age 60 years vs. 52 years in the UKPDS) and more likely to have high blood pressure at diagnosis. 32 Similarly, higher progression rates were reported in the VADT (Veterans Affairs Diabetes Trial),118 ACCORD119 and ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation)120 trials. However, these studies recruited older people who had long-standing diabetes.

We observed a beneficial increase in eGFR in both the IT group and the RC group. This is probably linked to the treatment of multiple cardiovascular risk factors, including lipid-lowering treatment, glucose-lowering treatment and ACE or ARB treatment.

As with the nephropathy outcomes, retinopathy frequency was lower than expected. Just 12% of ADDITION-Europe RC patients developed retinopathy after 5 years of follow-up. In the UKPDS, 37% of patients had retinopathy at diagnosis and a further 22% developed the condition during the next 6 years. 31 Again, this is likely to reflect early treatment in the lead time between screen detection and clinical diagnosis.

A similar trend was observed for neuropathy, with 5% of the population reporting peripheral neuropathy at 5 years, with no difference between groups. This is markedly lower than the 11% reported in a population with an average diabetes duration of 16 years. 67 However, caution should be exercised in the interpretation of this result. We used the Michigan neuropathy questionnaire, which may not be sufficiently sensitive121 or reliable to detect early progress of neuropathy and/or differences between the IT group and the RC group. 122 There was heterogeneity in the effect size estimates for peripheral neuropathy across all centres, which supports this view.

Patient-reported outcomes

Compared with RC, IT was not associated with differences in any PROMs (health status, well-being, quality of life and treatment satisfaction) after 5 years of follow-up. Health status, as measured by the EQ-5D, did not change between diagnosis and follow-up. Again, we conclude that the time period within which our trial was undertaken, for example when targets for cholesterol and blood pressure became stricter for diabetes patients, resulted in smaller than expected differences between study groups.

Baseline health-related quality of life was high in our cohort (median EQ-5D score of 0.85). EQ-5D estimates remained stable over follow-up in spite of participants’ age increasing by 5 years. This is an important finding, particularly as the burden of treatment increased over this time. In a study among 1136 Dutch people with type 2 diabetes (mean age 64.9 years), the average EQ-5D score was 0.74 (SD 0.27). 123 Older patients, women and those with a longer duration of diabetes reported lower levels of health-related quality of life. In a cluster-randomised trial in 55 Dutch general practices, there was no change in diabetes-related health over 1 year between standard and intensive multifactorial therapy, although a negative effect on the SF-36 social functioning score could not be ruled out. 124 In the UKPDS, intensively treating blood pressure and blood glucose in newly diagnosed diabetes patients did not adversely affect health status. 39 In an ACCORD substudy, there was a difference in change in the SF-36 physical component score and the DTSQ scale over 4 years between the standard group and the intensive glucose-lowering group. 40 However, the absolute difference in the SF-36 score was only 0.5 units, considerably less than the general threshold of 3–5 points recognised as an important difference. ACCORD participants differed from ADDITION-Europe patients in that they were older, they had had clinically diagnosed diabetes for an average duration of 9–10 years, 35% were on insulin treatment and they had a different ethnic background. Overall, the findings from other studies are largely in line with the results from the ADDITION-Europe trial, that is, no significant differences in PROMs between treatment groups.

It is worth noting that, although pooled analyses from all centres showed no differences between the treatment groups, for some PROMs the results in Leicester clearly favoured the IT group. This might be linked to the nature of the intervention. This was delivered by an intermediate care team at a community-based diabetes specialist care facility. Patients also received weight management advice, a glucometer at diagnosis and the DESMOND structured education programme. 53 Other ADDITION-Europe centres relied on existing primary care teams to achieve the IT targets.

Validation of the UK Prospective Diabetes Study outcomes model

The UKPDS-OM had moderate discriminatory ability to predict non-fatal MI and stroke among ADDITION-Europe participants. However, the model overestimated absolute risk. As diabetes patients begin to be diagnosed earlier in the disease trajectory, and their CVD risk is therefore lower, the UKPDS-OM may need updating for CVD risk prediction in contemporary diabetes populations.

The accuracy of the UKPDS-OM (version 1.3) in estimating the risk of non-fatal MI and stroke varied between countries. It performed particularly poorly in the Netherlands compared with the UK and Denmark. This could be because of a difference in baseline characteristics between the ADDITION-Europe cohorts and the UKPDS cohort. The UKPDS population had some baseline characteristics associated with a lower risk of CVD, including lower mean age, mean body mass index and mean systolic blood pressure, and some characteristics associated with a higher risk, for example a higher percentage of men, a lower mean HDL cholesterol level and a slightly higher mean HbA_1c level. The UKPDS also excluded people with previous CVD. However, the purpose of the UKPDS-OM is to provide coefficients for the inputted risk factors. As such, the accuracy of the predicted risk should be relatively insensitive to the baseline characteristics. We could reject a model on the basis of baseline characteristics only if the two populations were very different, for example predicting the risk of CVD based on a cohort of individuals aged 80 years and applying these risks to a cohort of individuals aged 20 years.

The most likely reason for the overestimation of risk is because of improvements in treatment between the time that the UKPDS data were collected (1977–97) and the time of ADDITION-Europe data collection (2002–6). The ADDITION-Europe intervention also included a much more comprehensive set of preventative treatments offered from an earlier stage in the disease. Our results are consistent with those of other studies examining the predictive ability of the UKPDS and Framingham models in populations with lower disease rates. 88,125

As AF and PVD data were not collected at baseline, these variables were set to zero in the UKPDS-OM. PVD was a risk factor only for amputation in the UKPDS-OM and so the omission of these data should not affect our results. AF is a risk factor for stroke in the UKPDS-OM;85 missing data for this variable are likely to underestimate stroke risk. However, in one study the prevalence of AF ranged from 1.2% to 2.8% among people aged 60 years126 and so missing data for this variable are unlikely to have a large effect on our results.

The UKPDS-OM performed differently in predicting the differences in CVD risk factors between the intervention groups. It underestimated the incremental risk of MI between groups and slightly overestimated differences for stroke. Assessment of the cost-effectiveness of health-care interventions necessitates knowledge of the increment between the comparator groups, rather than the difference in absolute levels of each group. It is ultimately a subjective decision whether or not these differences are sufficient to render the UKPDS-OM unsuitable for extrapolating the ADDITION-Europe data. The insight from our work shows that, if the UKPDS-OM is unadjusted, it will underestimate any reductions in MI risk attributable to IT and consequently potentially underestimate the cost-effectiveness of the intensive intervention (i.e. overestimate the ICER). The small disparity in the predicted compared with observed risk of stroke may be of limited consequence because of the effects of discounting.

As stated above, the UKPDS-OM had moderate discriminatory ability for predicting non-fatal MI and stroke. The aROC curve ranged from 0.65 to 0.79 and goodness of fit was acceptable. The UKPDS-OM is based on a set of Weibull survival regression equations. For the model to perform successfully, it is important to choose a suitable baseline distribution and relevant covariates. The best-fit model for the ADDITION-Europe data was an exponential distribution. Our analyses showed that age, sex and HbA_1c level were important covariates and should be included in the UKPDS-OM.

Cost-effectiveness results

Cumulative costs and QALYs over a time horizon of 1–5 years indicated that IT was not cost-effective over the short term. This result may be linked to the improvements in cardiovascular risk factors that were observed in both trial groups between baseline and follow-up. 55 In the long-term modelling analysis, the 30-year simulated ICER was greater than the recommended UK National Institute for Health and Care Excellence threshold (£20,000–30,000 per QALY),105 suggesting that interventions to promote delivery of intensive multifactorial treatment in the ADDITION-UK study were not cost-effective compared with RC in the long term.

The intervention becomes cost-effective only if it can be delivered at a cost of < £631 per participant. The cost of delivering IT in our analysis (£981 per participant) was calculated based on the study protocol,47 which covered all aspects of the intervention. This figure may overestimate the cost of reproducing the intervention as trial materials would not require complete redevelopment (although some adaptation to local needs would be likely). Alternative approaches to influencing practitioner behaviour, such as point-of-care reminders and decision aids, might be cheaper than practice visits by specialists. Furthermore, treatment costs were based on payments made to GP practices according to predicted costs based on the study protocol rather than actual costs. These costs may have been overestimated as GPs prescribed less medication than expected. 55 Finally, as per our validation analysis, the UKPDS tends to overestimate the risks of cardiovascular events, which could have led to overestimation of the ICER.

Our finding of a general improvement in cost-effectiveness over the long term is exemplary of the nature of preventative treatments. As most of the costs of such interventions are borne ‘up front’, and chronic disease complications take a long time to develop, health and cost benefits are usually seen only over the longer term. We used a 30-year modelling horizon in our analyses as the average age of participants at baseline was 60 years. We therefore assumed that a 30-year time horizon would equate to a lifetime horizon for the majority of individuals. A key question for policy-makers is whether or not they are prepared to consider a lifetime horizon in their decision-making. The optimal time horizon for any economic evaluation is one that is sufficient to capture all impacts on incremental costs and outcomes. 127 In our case, this suggests a lifetime horizon. However, investing in an intervention now means spending the money from this year’s budget without seeing any benefit for many years. During this time personnel, governments and government policy may change. Introducing such polices is therefore a challenge given pressures (political or otherwise) to show the benefits from investment decisions over the short to medium term.

The DESMOND study examined the benefits of a structured education and self-management programme for people with newly diagnosed diabetes in the UK from 2004 to 2006. 53,128 The authors found that the intervention was likely to be cost-effective compared with usual care over a lifetime horizon (ICER £5387). One reason for the difference between our results and the results of the DESMOND study and other studies might be the early disease stage of participants in our study. Early detection might help prevent complications in the future by attenuating disease progression. However, intervening ‘too early’ might be less cost-effective. For example, intervening when an individual presents with diabetes-related symptoms may reduce the risk of an event that would otherwise have occurred in 5 years’ time. Intervening at an earlier stage, after detection of IGT, for example, may prevent an event that would not otherwise occur for 20 years. In general, individuals and society have a preference for benefits now rather than in the future. As such, they would value preventing an event in the next 5 years more highly than preventing the same event in 20 years’ time.

Strengths and limitations

The ADDITION-Europe study participants were drawn from representative population-based samples in three different European countries. Practices were recruited from a large geographical area in each country. Randomised practices (26% of those invited) were nationally representative in terms of key clinical and sociodemographic characteristics. 43,48,49 However, recruitment was non-random and the generalisability of our findings to other settings should be considered with caution. Furthermore, participants were predominantly white (93%), which might limit the extrapolation of our data to more ethnically diverse regions. Individuals were identified using a range of different screening programmes and were diagnosed according to 1999 WHO criteria. 56

The study intervention included algorithms and targets based on well-established trial data and we used a range of methods associated with changes in practitioner and patient behaviour. We randomised general practices rather than individuals (as the intervention was delivered mainly by practitioners) to minimise the risk of contamination. We achieved high levels of participant retention and independently adjudicated end-point ascertainment in both trial groups (99.9%). Trained staff were unaware of study group allocation. We assessed clinically important outcomes using standardised equipment and protocols. For the modelled CVD risk analysis, we used the latest version of the UKPDS risk score (version 3). This was derived from > 40,000 patient-years of data and 1115 CVD events. 75 For PROMs we used both generic and diabetes-specific measures.

There was variation between centres in the approaches to population-based screening for diabetes. However, all participants were diagnosed according to WHO guidelines. 56 Furthermore, the differences in screening programmes do not affect the internal validity of the study as they apply equally to both study groups in each centre. Rather, they increase the external validity or generalisability of the trial by including patients detected using a range of different screening programmes. There was also variation in the methods used to encourage the general practice teams to follow the treatment algorithms. However, there was one common set of treatment targets and algorithms for the IT group in all centres and, as stated above, this heterogeneity should increase generalisability without diminishing internal validity. We pooled centre-specific estimates using fixed-effects meta-analysis and formally tested for heterogeneity between centres using the I²-statistic. Although we acknowledge that this test has limited power, the primary outcome and all of its components did not exhibit significant between-centre heterogeneity and so we have not formally explored in this report explanations for any observed heterogeneity. Furthermore, there are a number of other centre-level covariates other than those previously mentioned relating to the different health systems in each country that may have contributed to variation in participant characteristics and response to the intervention. The different approaches taken in each centre to promote IT of multiple risk factors represent complex interventions. As the study was pragmatic, it is not possible to determine which aspects of the interventions might have been effective and which aspects might have been less so.

Cluster randomisation occurred before screening and patient recruitment. This design feature can introduce a difference between groups in participant characteristics. Although individuals in the IT and RC groups were well matched overall, in Denmark the intervention may have influenced the number and type of patients recruited to the IT group. Staff undertaking opportunistic screening may have conducted more tests in individuals at high CVD risk as they were aware of the potential benefits of early detection and treatment. This may have attenuated the effects of the intervention and might explain some of the differences in results between centres. Differences appeared to be present but might not have been statistically significant because of the limited power of the test of heterogeneity. Our intracluster correlation coefficients were small55 and did not adversely affect study power. Methods of screening, laboratory testing and outcome assessment were standardised across study groups in each centre in this pragmatic trial but did differ between centres. This may also have contributed to some of the observed heterogeneity. Measurement of biochemical outcomes differed slightly between baseline and follow-up in two centres but not between groups. Reassuringly, interlaboratory validation suggested that methods were consistent between centres. Primary end points were adjudicated using a standardised manual and case report form in each centre. Other potential sources of heterogeneity in our trial arise from variation between centres in participant characteristics. This might be because of differences in the screening programmes and underlying populations, different modes of delivering the intervention and other unmeasured factors.

For the CVD outcomes analyses there were few differences between those with and those without follow-up data. For the modelled CVD risk analysis, 27% of data were missing. Sensitivity analyses suggested that the primary analysis was likely to represent an accurate ITT analysis. We accounted for missing data at baseline using the missing indicator method and assumed that those lost to follow-up had a 10% higher risk than those with available data at 5 years. There was qualitatively no change to the final results and the difference in modelled CVD risk between groups remained significant. It is likely that we overestimated the level of modelled CVD risk in our contemporary trial cohort as the UKPDS model is based on a historical cohort diagnosed 20 years before the ADDITION-Europe study participants. However, this would not have affected the effect size estimate differentially by group. There was no difference between self-reported AF at 5 years between the RC group (13.3%) and the IT group (14.0%), suggesting that the exclusion of this variable from the UKPDS model was unlikely to affect our findings.

For the microvascular outcomes analysis there were differences in the proportions of individuals with data for different outcomes at 5 years. For example, 77% had retinopathy data and 95% had eGFR data. There was some evidence of healthy volunteer bias when examining the characteristics of participants with and without microvascular outcome data; however, absolute differences were small [e.g. 0.19% (2 mmol/mol) for HbA_1c]. There were no differences between the IT group and the RC group in terms of the proportion with missing microvascular data. However, the prevalence of retinopathy and neuropathy might have been underestimated considering the adverse risk profile of individuals with missing data for these outcomes. Microvascular outcome assessment and laboratory testing were standardised across participants from the IT and RC groups in each centre but did differ between centres; for example, there was no standardised procedure for retinal photography. Some centres used routine secondary data sources, whereas others collected retinal data during remeasurement of participants at the 5-year follow-up examination. This variation might have led to differences in the precision of outcome assessment between centres but not between study groups. Furthermore, we assessed only gradable photos and they were coded using a standard scale (ETDRS65) by three experienced ophthalmologists who were unaware of study group allocation.

For the PROMs, patients who completed the questionnaires differed from those who did not. However, we believe that these differences are unlikely to have had an important effect on the outcomes as the differences were small, were not in one direction and were similar for both the IT group and the RC group. Our multiple imputation analysis confirmed the results of the main analysis, which supports our conjecture that selective dropout of patients did not introduce significant bias. As most of the PROMs were not measured at baseline, we were not able to examine changes in PROMs over time, nor adjust for baseline measurement of each PROM. This would not have been possible for diabetes-specific outcome measures such as the DTSQ and ADDQoL. We adjusted for a number of possible confounding variables, which were similar between the IT group and the RC group at baseline. We also adjusted all models for the baseline EQ-5D score as a proxy for baseline health status.

For the UKPDS-OM validation work we used multiple imputation, which allows the best use of the available data and takes into account the uncertainty in missing values. Both the imputed data set and the complete-case data set produced similar results, confirming the robustness of our results. Our validation work was limited by a lack of annual CVD risk factor measurement. In the UKPDS cohort, biochemical data were collected every year. We had to average baseline and follow-up values for the year of event in regression analyses as they were not available for every year. We did not have information on baseline AF, PVD, years since pre-existing CVD, blindness and renal failure. This might have introduced some bias in model prediction and influenced the efficiency of the model, most likely contributing to an underestimation of risk of CVD. We validated the UKPDS-OM using 5-year data and extrapolated the model to a longer time horizon in our cost-effectiveness work to estimate the 30-year incremental cost-effectiveness of the IT intervention. As follow-up continues more events will occur. We will revise our estimate of the cost-effectiveness of the intervention once longer-term data are available.

For the cost-effectiveness work we examined outcomes over both the short term and the long term and performed sensitivity analyses to test the robustness of our findings. We used data from the UKPDS, a UK study of individuals with diabetes and long-term CVD outcomes to calculate unit costs, utility decrements and modelled CVD risk. We adjusted for centre (Cambridge/Leicester) rather than practice as the intervention differed slightly between the centres; however, within the same centre practices shared standardised practitioner guidelines. Intraclass correlation coefficient values for GP practice for the primary outcome were very small in the main ADDITION-Europe trial, supporting our decision to adjust for centre rather than practice.

There were some limitations. First, we focused on only macrovascular outcomes (CVD and associated acute events). Microvascular outcomes such as retinopathy and nephropathy are also important in assessing the impact of type 2 diabetes on both quality of life and costs. Exclusion of these is likely to underestimate the benefit from any preventative intervention and thus underestimate cost-effectiveness (i.e. overestimate the ICER), particularly in the longer term as patients avoid CVD and live with diabetes long enough to develop microvascular complications.

Second, although the treatment protocol was identical, the lifestyle intervention was different in Leicester and Cambridge. To represent the cost of implementing an ‘ADDITION-like’ intervention that reflected a degree of diversity across the country we simply averaged the costs of the two interventions. We did, however, include centre in the adjusted analyses estimating incremental costs and QALYs. Results were similar when we adjusted for cluster (GP practice) and when running analyses separately by centre.

Third, we used an additive method to calculate treatment costs and utility decrements for individuals with multiple events. This is a commonly used method and was applied in the UKPDS-OM,100,129 but it is unclear if the cost and utility decrements of subsequent complications should be additive or multiplicative. A review of health state utilities in patients with type 2 diabetes published after completion of our analysis130 recommends the use of the Clarke data set for most comorbid events, or health state utilities very close to those used in our analysis, with one notable exception: we assigned a disutility of 0.263 for renal failure, whereas Beaudet et al. 130 recommend values between 0.204 and 0.164. Using a higher disutility will overestimate the health gain from averting renal failure. However, we estimated the difference in risk of renal failure over 30 years between the IT group and the RC group to be –0.05% (SE 0.06%). The impact of this on the incremental QALY gain per patient in each arm is therefore likely to be negligible.

Fourth, most of the equations in the UKPDS-OM predict future risk factors and events based on baseline values, time since diagnosis and the value of biometrics such as HbA_1c level in the previous time period (year). It is an individual-level model in which patient data are calculated individually (rather than as mean values of a cohort). Thus, the model assumes that any trends present at 5 years will continue into the future, effectively assuming a continuation of the effect of the small differences in treatment generated by the intervention at 5 years. The 10-year follow-up of the UKPDS cohort found that, although between-group differences in HbA_1c level and blood pressure were both lost within 12 months of the end of the active phase of the study, those who had previously achieved tighter control over HbA_1c still had a lower event rate at 10 years than those who had not, while any benefit from lower blood pressure was not maintained. 131,132 Current guidelines and practice have changed since the initiation of the ADDITION-Europe trial, with patients recommended to receive at least as IT as the ‘intensive’ arm. Trial patients in the RC arm received a higher level of treatment than observed in ‘standard practice’, which may have diluted the observed incremental health gain. However, after 5 years of treatment, there was room for improvement in the prescription of cardioprotective treatment in both groups. Current evidence suggests that delays in treatment intensification in people with type 2 diabetes (clinical inertia) are still very common.

On the cost side, the official duration of the IT intervention was 3 years, for which practices were reimbursed for additional activity and prescriptions. We projected additional prescribed drug costs in the IT arm over the full 30 years under the assumption that patients in the RC arm would not increase their medication and patients in the IT arm would not decrease their medication over time. We may, therefore, have overestimated the cost of the IT arm. Whether this is true or not will be assessed once the 10-year follow-up data from the ADDITION-Europe cohort are available for analysis.

Finally, although we used the most appropriate CVD risk model available, as outlined earlier, the UKPDS-OM was derived using data from a historical cohort of people with clinically diagnosed diabetes. The ADDITION cohort included people with screen-detected diabetes. Other research125,133 has shown that the UKPDS-OM tends to overestimate absolute CVD risk, a finding replicated in our own investigation of the suitability of the UKPDS-OM to extrapolate ADDITION study data. 96 Furthermore, the utility decrements for MI and stroke derived from the ADDITION study were smaller than those in the UKPDS, a finding that is consistent with contemporary patients receiving better care and hence reporting higher quality of life. This underlines the importance of our sensitivity analyses, which showed that changes to these parameters did not substantially affect the ICER. Of the discount rate, intervention unit cost and utility decrement, variations in the discount rate had the biggest impact on the ICER.

Conclusion

Improvements in the quality of diabetes care in general practice during the duration of the ADDITION-Europe trial meant that treatment in the RC group was better than expected and not very different from the care provided in the IT group. 55 Cardiovascular risk factors (weight, blood pressure, cholesterol and modelled CVD risk82), including health-related behaviours (smoking), improved substantially following detection of diabetes by screening, even among those receiving RC. The small differences between groups in the intensity of treatment during the 5 years following diagnosis were associated with a non-significant 17% reduction in the primary composite cardiovascular end point in favour of the IT group. 55 There was no difference between the groups at 5 years in the prevalence of microvascular outcomes83 or in health status, well-being, quality of life or treatment satisfaction. 84 Recent trials of IT of glycaemia among patients with long-standing diabetes suggest that adverse effects occur early35,134 and that any benefit in terms of a reduction in cardiovascular risk takes longer than 5 years to be realised. 131 Whether or not the small difference in the intensity of treatment of multiple CVD risk factors early in the course of the disease in the ADDITION-Europe study will translate into a significant reduction in cardiovascular events and mortality over the longer term is uncertain.

We demonstrated that it was feasible to use the UKPDS-OM model to extrapolate the ADDITION-Europe data. 96 We took account of the possible underestimation of any reductions in MI and used the model in our cost-effectiveness analysis. Cost-effectiveness work showed that promotion of IT compared with RC was not cost-effective over a short- to medium-term time horizon for people with screen-detected diabetes in the UK. However, the intervention has the potential to be cost-effective if it can be delivered for approximately £630 per patient rather than £981. Such savings may be plausible through adaptation of predeveloped materials and economies of scale in delivery. Although there were few differences between screened and unscreened populations, the small proportion (2.9%) of individuals whose diabetes was detected by screening did appear to benefit. 23

Contributions of authors

The following authors were involved in the conception of the original ADDITION-Europe study design: Knut Borch-Johnsen (Professor of Medicine), Melanie J Davies (Professor of Diabetes Medicine), Simon J Griffin (Professor of General Practice), Kamlesh Khunti (Professor of Primary Care Diabetes and Vascular Medicine), Torsten Lauritzen (Professor of General Practice), Guy EHM Rutten (Professor of General Practice), Annelli Sandbæk (Professor of General Practice) and Nicholas J Wareham (Professor of Epidemiology)

For Chapters 3 and 4, the following authors contributed to the design of the work; the acquisition, analysis and interpretation of data; drafting the work and revising it critically for important intellectual content; and preparing the results for publication: Knut Borch-Johnsen, Melanie J Davies, Simon J Griffin, Kamlesh Khunti, Torsten Lauritzen, Guy EHM Rutten, Annelli Sandbæk, Stephen J Sharp (Senior Statistician), Rebecca K Simmons (Senior Investigator Scientist, Diabetes Epidemiology), Maureen van den Donk (Scientific Staff Member, Epidemiology) and Nicholas J Wareham.

For the modelled CVD risk section of Chapter 3, James A Black (PhD student, Epidemiology) completed the data analysis, drafted the work and revised it critically for important intellectual content and prepared the results for publication.

For Chapter 4, the following authors contributed to the design of the work; conducted the economic analysis and interpretation of data, drafted the work and revised it critically for important intellectual content; and prepared the results for publication: Libo Tao (Research Fellow, Health Economics) and Edward CF Wilson (Senior Research Associate, Health Economics).

All authors gave final approval for this version of the work to be published. Professor Simon J Griffin agrees to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Publications

Lauritzen T, Griffin S, Borch-Johnsen K, Wareham NJ, Wolffenbuttel BH, Rutten G. The ADDITION study: proposed trial of the cost-effectiveness of an intensive multifactorial intervention on morbidity and mortality among people with Type 2 diabetes detected by screening. Int J Obes Relat Metab Disord 2000;24(Suppl 3):6–11.

Griffin SJ, Borch-Johnsen K, Davies M, Khunti K, Rutten GEHM, Sandbæk A, et al. Effect of early intensive multifactorial therapy on 5-year cardiovascular outcomes in individuals with type 2 diabetes detected by screening (ADDITION-Europe): a cluster-randomised trial. Lancet 2011;378:156–67.

van den Donk M, Sandbæk A, Borch-Johnsen K, Lauritzen T, Simmons RK, Wareham NJ, et al. Screening for type 2 diabetes. Lessons from the ADDITION-Europe study. Diabet Med 2011;28:1416–24.

Simmons RK, Echouffo-Tcheugui JB, Sharp SJ, Sargeant LA, Williams KM, Prevost AT, et al. Screening for type 2 diabetes and population mortality over 10 years (ADDITION-Cambridge): a cluster-randomised controlled trial. Lancet 2012;380:1741–8.

Simmons RK, Sharp SJ, Sandbæk A, Borch-Johnsen K, Davies MJ, Khunti K, et al. Does early intensive multifactorial treatment reduce total cardiovascular burden in individuals with screen-detected diabetes? Findings from the ADDITION-Europe cluster-randomized trial. Diabet Med 2012;29:e409–16.

Lauritzen T, Borch-Johnsen K, Davies MJ, Khunti K, Rutten GEHM, Sandbæk A, et al. Screening for diabetes. What do the results of the ADDITION trial mean for clinical practice? Diabetes Management 2013;3:367–78.

Tao L, Wilson ECF, Griffin SJ, Simmons RK. Performance of the UKPDS outcomes model for prediction of myocardial infarction and stroke in the ADDITION-Europe trial cohort. Value Health 2013;16:1074–80.

van den Donk M, Griffin SJ, Stellato RK, Simmons RK, Sandbæk A, Lauritzen T, et al. Effect of early intensive multifactorial therapy compared to routine care on self-reported health status, general wellbeing, diabetes specific quality of life and treatment satisfaction in screen-detected type 2 diabetes patients: the ADDITION-Europe study. Diabetologia 2013;56:2367–77.

Black JA, Sharp SJ, Wareham NJ, Sandbæk A, Rutten GE, Lauritzen T, et al. Change in cardiovascular risk factors following early diagnosis of type 2 diabetes: a cohort analysis of a cluster-randomised trial. Br J Gen Pract 2014;64:e208–16.

Black JA, Sharp SJ, Wareham NJ, Sandbæk A, Rutten GEHM, Lauritzen T, et al. Does early intensive multifactorial therapy reduce modelled cardiovascular risk in individuals with screen-detected diabetes? Results from the ADDITION-Europe cluster randomised trial. Diabet Med 2014;31:647–56.

Kuznetsov L, Griffin SJ, Davies MJ, Lauritzen T, Khunti K, Rutten GE, et al. Diabetes-specific quality of life but not health status is independently associated with glycaemic control among patients with type 2 diabetes: a cross-sectional analysis of the ADDITION-Europe trial cohort. Diabetes Res Clin Pract 2014;104:281–7.

Long GH, Cooper AJ, Wareham NJ, Griffin SJ, Simmons RK. Healthy behavior change and cardiovascular outcomes in newly diagnosed type 2 diabetic patients: a cohort analysis of the ADDITION-Cambridge study. Diabetes Care 2014;37:1712–20.

Sandbæk A, Griffin SJ, Sharp SJ, Simmons RK, Borch-Johnsen K, Rutten GEHM, et al. Effect of early multifactorial therapy compared with routine care on microvascular outcomes at 5 years in people with screen-detected diabetes. A Randomized Controlled Controlled Trial: the ADDITION-Europe study. Diabetes Care 2014;37:2015–23.

Black JA, Long GH, Sharp SJ, Kuznetsov L, Boothby CE, Griffin SJ, et al. Change in cardio-protective medication and health-related quality of life after diagnosis of screen-detected diabetes: results from the ADDITION-Cambridge cohort. Diabetes Res Clin Pract 2015;109:170–7.

Black JA, Simmons RK, Boothby CE, Davies MJ, Webb D, Khunti K, et al. Medication burden in the first 5 years following diagnosis of type 2 diabetes: findings from the ADDITION-UK trial cohort. BMJ Open Diabetes Res Care 2015;3:e000075.

Herman WH, Ye W, Griffin SJ, Simmons RK, Davies MJ, Khunti K, et al. Early detection and treatment of type 2 diabetes reduce cardiovascular morbidity and mortality: a simulation of the results of the Anglo-Danish-Dutch study of intensive treatment in people with screen-detected diabetes in primary care (ADDITION-Europe). Diabetes Care 2015;38:1449–55.

Kuznetsov L, Long GH, Griffin SJ, Simmons RK. Are changes in glycaemic control associated with diabetes-specific quality of life and health status in screen-detected type 2 diabetes patients? Four-year follow up of the ADDITION-Cambridge cohort. Diabetes Metab Res Rev 2015;31:69–75.

Tao L, Wilson EC, Wareham NJ, Sandbæk A, Rutten GE, Lauritzen T, et al. Cost-effectiveness of intensive multifactorial treatment compared with routine care for individuals with screen-detected type 2 diabetes: analysis of the ADDITION-UK cluster-randomized controlled trial. Diabet Med 2015;32:907–19.

All scientific papers arising from this report have been published as Open Access articles. As such, the papers are covered by a Creative Commons Attribution (CC BY 3.0) licence or the Creative Commons Attribution Non Commercial licence (CC BY-NC 3.0).

Data sharing statement

The ADDITION-Europe data are available to share via http://epi-meta.medschl.cam.ac.uk/ (accessed 28 January 2016).

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.