A randomised controlled trial of cognitive behaviour therapy and motivational interviewing for people with type 1 diabetes mellitus with persistent sub-optimal glycaemic control: A Diabetes and Psychological Therapies (ADaPT) study

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 01/17/05. The contractual start date was in February 2003. The draft report began editorial review in May 2008 and was accepted for publication in August 2009. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. 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 referees 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

None

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and 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: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

Type 1 diabetes

Type 1 diabetes is characterised by an absolute lack of insulin production secondary to pancreatic beta cells being selectively destroyed. This report focuses on adults with type 1 diabetes, previously known as insulin-dependent, juvenile or childhood-onset diabetes, who are having difficulties in achieving optimal diabetes control. Chronic hyperglycaemia leads to micro- and macrovascular complications, affecting many of the body’s systems, especially the nervous and vascular systems, resulting in increased mortality. 1

There are two forms of type 1 diabetes: type 1A is by far the most common and is secondary to an autoimmune attack on beta cells; and type 1B is much less common, the cause is still not known and it is most common in people of black African/Caribbean or Asian descent with varying degrees of insulin deficiency. The pathogenesis of type 1A diabetes is understood to be a combination of a varying genetic susceptibility to type 1 diabetes and exposure to one or more environmental triggers, such as viral infection, toxins or food allergens initiating β cell destruction. 2 The classic triad of symptoms that patients experience include excessive urination (polyuria), thirst (polydypsia) and weight loss. Other associated symptoms include visual changes, fatigue and constant hunger.

There are an estimated 2.35 million people with diabetes in England and there is an epidemic of diabetes. 3 Type 1 diabetes accounts for 10–15% of all cases of diabetes. The prevalence of type 1 diabetes in the UK is about 1 in 800 children up to the age of 16. There is a marked variation in the incidence of type 1 diabetes according to geographical location and ethnicity; a child in Finland is nearly 350 times more likely to be diagnosed with diabetes than a child in China. 4 The incidence of type 1 diabetes is increasing annually worldwide at around 2–5% and in the UK there has been an increase from 7.7 to 13.5 children per 100,000 from the early 1970s to the late 1980s. 5 Around 50% of all cases present before the age of 18 years, and the remainder present at a low rate throughout adulthood.

Microvascular complications include retinopathy which can lead to progressive blindness; nephropathy and progressive renal failure; and neuropathies such as peripheral neuropathy which is associated with diabetic foot disease and limb amputation and autonomic neuropathy such as gastroparesis. Macrovascular complications include coronary heart disease, cerebrovascular disease and peripheral vascular disease. While there are differences in the susceptibility to such complications between individuals, perhaps reflecting genetic influences, these complications are linked to sub-optimal glycaemic control, early onset and longer duration of diabetes, smoking, obesity and sedentary lifestyle. 2 Diabetes complications usually take many years to become clinically manifest. This is a core challenge for people with diabetes because they may not acutely suffer or be aware of symptoms of hyperglycaemia and yet are asked to modify their behaviours for a potential complication in the distant future.

Measuring glycaemic control

There are many ways to measure glycaemic control. The most common is random blood glucose testing, which measures the current level of glucose (mmol/l) and glycated (or glycosylated) haemoglobin or HbA_1c (%). The random blood glucose test is measured using small portable self-testing kits and allows for a rapid respond to high or low readings using a range of options such as insulin dose or dietary adjustment and this is one of the important diabetes self-care skills. Glycated haemoglobin is an indication of the percentage of red blood cells have been exposed to glucose during their 120-day life cycle and this depends on how much glucose is circulating in the blood. Once a haemoglobin molecule is glycated, it remains that way and thus can serve as a proxy marker for the level of glycaemic control especially in the last 6 weeks and less so over the last 12 weeks. The normal range of glycated haemoglobin is between 4% and 6%, and the ideal target for people with diabetes is to get as close to the non-diabetes range as possible without problematic hyperglycaemia. The current national guidelines have set a practical target of between 7% and 8% taking into account the circumstances of the individual patient. 6

Management of diabetes

At present there is no cure for diabetes, and management is predominantly self-care with administration of exogenous insulin for the rest of the person’s life. The main aim of diabetes treatment is to optimise glycaemic control to minimise the risk of micro- and macrovascular complications. People with diabetes need to adopt a diabetes-specific diet (predominantly monitoring the amount of carbohydrates), exercise, monitor blood glucose levels and titrate their insulin injections accordingly several times a day, take additional medication to reduce their risk of macrovascular complications, self-examine injection sites and injuries to their feet and stay in touch with their diabetes team. One of the landmark studies, the Diabetes Controls and Complications Trial (DCCT) showed that intensive insulin therapy regimes involving multiple injections and frequent monitoring by diabetes teams over 6–10 years were associated with an improvement in glycaemic control sufficient to reduce the risk of diabetes complications,7,8 although not necessarily an improvement in quality of life. 9 These interventions continued to be effective in reducing long-term complications after the study had ended and the glycaemic control in the intensively treated group had gradually returned to the baseline, suggesting that a period of good glycaemic control is associated with a resetting of ‘metabolic memory’. 10,11

Since the DCCT, intensive structured education programmes such as DAFNE (Dose Adjusted for Normal Eating) that involve titrating insulin doses with carbohydrate intake, to support people in leading flexible dietary lifestyles, are also effective in improving glycaemic control. 12,13

Continuous subcutaneous insulin infusions, sometimes called the ‘external pancreas’, attempt to match endogenous insulin rhythms and may be effective in those individuals who have difficulties in managing multiple injections. 14

While a variety of treatment options are available, they are not always preferred by the patient, and how individuals manage their self-care determines to a large extent the course of the illness.

There have also been recent hopeful advances in islet cell transplantation,15 but these are indicated for specific groups of individuals with problematic hypoglycaemia and are still emerging technologies.

The problem of sub-optimal glycaemic control

Despite the evidence for effective intensive insulin regimes and structured education programmes, between 25% and 50% of adults with type 1 diabetes have sub-optimal glycaemic control. 16 In the DCCT, the majority of the intensive group did not achieve or sustain the target HbA_1c of 6.0%. 7 Despite recent reductions in the national average glycated haemoglobin, the average HbA_1c in most diabetes clinics is still around 9%. 17,18 In a prospective observational study of adolescents, sub-optimal glycaemic control persisted into young adulthood,19 and a similar observation was found in Scottish university students with type 1 diabetes. 20

Socioeconomic impact of diabetes

Diabetes, like any other chronic disease, has an impact on work, income, quality of life and social relationships. Certain jobs are excluded such as joining the military and large goods vehicle driving, and people with diabetes worry about stigma in the workplace, forming intimate relationships and their relationships with peers. Type 1 diabetes is a disease of the young and consequently affects those who are still at school and have an adult life’s worth of being economically productive. Young people with diabetes are less likely to achieve academically. 21 People with diabetes are twice as likely to be admitted to hospital as the general population, and the presence of complications increases the cost to the NHS more than fivefold. 22

Factors associated with sub-optimal glycaemic control

There are cross-sectional and a few landmark large-scale prospective studies that have investigated the sociodemographic, biological and other lifestyle factors associated with glycaemic control in adults with type 1 diabetes.

A large European cross-sectional survey (n = 2387) of adults (age 25–60 years), the EURODIAB Insulin Dependent Diabetes Mellitus complications study,23 showed that the mean HbA_1c was higher in adults with lower socioeconomic status as defined by age at completion of education.

The frequently cited Düsseldorf study, which evaluated the effectiveness of an intensified 5-day insulin treatment and teaching programme24 in 697 type 1 diabetes adults with advanced diabetes complications, found that sub-optimal glycaemic control at 3 years was associated with smoking, younger age at onset of diabetes, less frequent self-monitoring, lower socioeconomic status (composite score depending on income per month, household composition, employment and educational status), less diabetes-related knowledge and perceived abilities to cope, and being female, representing 17% of the variance. 25

In the Pittsburgh Epidemiology of Diabetes Complications study (n = 657),26,27 worse glycaemic control was associated with younger age, lower income, lower educational attainment and low frequency of self-monitoring of blood glucose.

In a subgroup of participants (n = 623) allocated to intensive treatment in the DCCT, those who reported adhering to prescribed meal plans and adjusted food and/or insulin in response to hyperglycaemia had significantly lower HbA_1c results than those who did not. 28

In 84 newly diagnosed adults with type 1 diabetes alcohol consumption and knowledge of diabetes at 4 months after diagnosis were found to be independent predictors of glycaemic control at 12 months, explaining 16% of the variance. 29 Other factors such as the General Health Questionnaire, diabetes-specific quality of life, cognitive ability and personality were not but these findings may reflect the short follow-up.

Psychological factors and their association with glycaemic control

People with diabetes are at higher risk of psychological problems such as depression and anxiety than the general population and have psychological issues specific to diabetes such as fear of hypoglycaemia.

Potential role of psychological treatments

Considering the limits of intensive medical interventions, the high rates of psychological problems in people with diabetes and their association with sub-optimal glycaemic control and other adverse outcomes, there is an a priori role for psychological treatments as adjuncts in helping to improve glycaemic control. Psychological therapies utilise the therapeutic alliance between the patient and the therapist in which the patient’s problems are described in terms of his or her emotions, cognitions (or thinking) and/or behaviours, and in some therapies these are linked to early life experiences with the overall aim of improving psychological functioning. Psychological treatments are used widely in mental health settings to treat a range of emotional disorders such as depression,75 anxiety disorders and psychosomatic disorders. The potential for psychological treatments in chronic disease setting to improve psychological and/or biological outcomes is an emerging field. 76

A person with diabetes needs skills in, firstly, managing the practical daily routine of diabetes self-management and, secondly, coping with the burden of living with a chronic condition. The former requires diabetes knowledge delivered through diabetes education programmes such as DAFNE. 13 The latter involves multiple psychological processes that, if they go wrong, lead to psychological problems which are best managed using a psychotherapeutic approach.

Psychological interventions should be distinguished from educational interventions although they are not mutually exclusive. Educational interventions are based on didactic and social learning theory (sometimes also called collaborative, therapeutic and behavioural) to improve diabetes self-management by increasing knowledge. 77,78 Facts and knowledge are imparted through advice, lectures and written material, and problem solving via a process of memory and testing and retesting. The relationship is based on the teacher or educator–pupil paradigm. Diabetes education is a core component of usual diabetes care but is not always sufficient in achieving glycaemic control. 79

Psychological treatments are based on the principle that the therapist and the patient are in a collaboration and develop a therapeutic alliance within the context of which psychological processes variously associated with conscious and/or unconscious emotions, the transference, past experiences, thoughts and behaviour are evaluated using a variety of techniques such as: listening and reflecting; history taking; formulating the problem(s); giving meaning or interpretations to thoughts, feelings, and behaviours; challenging of unhelpful beliefs and assumptions; and setting goals. Psychological treatments are not a core component of usual diabetes care, although national guidelines for type 1 diabetes state that psychological care should underpin all aspects of diabetes care.

Systematic reviews of the effectiveness of psychological treatments in improving glycaemic control

Just over 10 years after insulin therapy was introduced, the first attempts at a theory-based psychotherapy in managing neurosis in diabetes were published,80 but there was little progress for the next 40 years as the emphasis was on diabetes education. In the early 1990s, a case series of successful psychoanalytical treatments for adolescent female inpatients with brittle diabetes heralded the potential of psychotherapy for complex cases. 81 The first RCTs appeared in the early 1980s. There have been several systematic reviews of RCTs comparing the effectiveness of psychological treatments in improving diabetes control. One review included only children and adolescents with type 1 diabetes as the population of interest and did not distinguish between psychological and educational interventions. 82 They found a standardised pooled effect size of 0.33 and they interpreted this as small to medium. Another review did not distinguish between type 1 or type 2 diabetes which does not seem appropriate as the epidemiology, natural history, sociodemographic profile and treatments are different. 83

We conducted a Cochrane Collaboration-based systematic review of RCTs comparing the effectiveness of psychological treatments for improving glycaemic control in people with type 1 diabetes. The full details of the rationale, methodology and results have been published. 84 The purpose of the review was to assess, quantify and critique the current evidence in order to model our proposed interventions. We aimed to focus on those interventions that were either solely or predominantly psychotherapeutic.

Psychological treatments were categorised into those most commonly used in health-care settings as follows: supportive or counselling therapy;85,86 cognitive behaviour therapy (CBT) or cognitive and behavioural techniques;87,88 psychoanalytically informed therapies;89–91 and family systems therapy. 92 Studies that did not explicitly label their intervention as above were still included if they used one or more psychological techniques that could be coded into one of the above categories. Techniques such as relaxation, activity scheduling, problem solving, goal setting, contract setting, cognitive restructuring and stress management were categorised as variants of the CBT model. 93,94 Techniques such as motivational interviewing (MI) were categorised under the counselling model. 95

There were 13 RCTs for adults included in the systematic review. Most RCTs had small sample sizes (< 100 participants) and did not adequately describe the study progress according to the CONsolidated Standards of Reporting Trials (CONSORT) guidelines. 96 The most common clinical subgroup was sub-optimal glycaemic control;97–100 there was one RCT each for newly diagnosed diabetes,101 complications,102 and obesity. 103 The mean duration of diabetes was 14.12 [standard deviation (SD) 6.85] years. Eight studies used either groups or a combination of group and individual format. The majority of RCTs assessed CBT techniques, one RCT tested cognitive–analytical therapy and another was based on psychodynamic techniques, 98,101 and two RCTs tested counselling. 104,105

There were 11 adult studies (n = 516) with data that could be pooled. Using random effects meta-analyses, there was a small pooled estimate of the mean standardised effect sizes which was non-significant [–0.17, 95% confidence interval (CI) –0.45 to 0.10; p = 0.22] that, translated into absolute reductions in HbA_1c, represented a 0.22% reduction (–0.13% to 0.56%) for adults. In a sensitivity analysis, restricting the adults to CBT worsened the pooled standardised effect size for adults (0.02, CI –0.41 to 0.44; p = 0.95).

Treatment manuals and treatment fidelity

In psychotherapy intervention studies, assessment of treatment integrity (whether treatment was delivered adequately and as intended) and treatment discrimination or specificity (whether techniques from other therapies were included) is important to validate and translate the intervention. Methods include specifying the techniques and the condition to be treated, standardising these in a manual and testing the therapists’ abilities. 106–109 Poor integrity may be associated with poor outcomes and study hypotheses cannot be validly tested.

In our systematic review, only two RCTs in adults with type 1 diabetes reported using a manual. 110,111 Three RCTs assessed treatment fidelity. Halford and colleagues110 videotaped the treatment sessions for adherence to the treatment manual procedures, but the results were not reported. Didjurgeit and colleagues102 stated that the therapist was supervised by one of the co-authors. In Van der Ven and colleagues,111 one of the authors observed the intervention and control group sessions through a one-way mirror.

Assessment of moderators (predictors)

The joint report by the Department of Health (DoH) and the Medical Research Council (MRC) and the MRC framework for evaluating complex interventions identified a need for theory-based research to identify key moderators and mediators of different behaviours associated with optimal and sub-optimal glycaemic control. 112,113 We identified two studies that attempted a moderator analysis.

Glasgow and colleagues104 found no significant correlations between baseline variables (sex, age, education, duration of diabetes, insulin taking, type of diabetes, number of comorbid chronic diseases, perceived barriers and perceived importance to dietary self-care, perceived seriousness of disease, subjective desire for involvement in diabetes management) and glycated haemoglobin at 3 months.

Didjurgeit and colleagues102 found that their intervention was more effective for those with high (> 8.0%) baseline HbA_1c levels 6 months later, but they did not adjust for baseline differences in HbA_1c.

Health professionals as therapists

Conventionally, psychological treatments are delivered by mental health professionals who have had training of varying intensity and quality depending on the type of therapy being offered. There are several problems for using the conventional mental health therapist in the chronic disease seeing. First, it tends to be too costly and, with the increasing prevalence of certain conditions such as obesity, diabetes and cardiovascular disease, there are insufficient numbers of therapists to meet the need. Second, the therapist may not be sufficiently knowledgeable of chronic disease and its manifestation to help the patient tackle disease-specific cognitions and behaviours, although this can be overcome with experience. A diabetes specialist on the other hand is already delivering a care package, and adding skills may be a better use of resources and relational continuity. Third, the patient may have additional concerns about seeing mental health experts or being stigmatised especially as he or she may appraise his or her difficulties as a consequence of having to live with a chronic condition rather than a separate mental health problem and, anecdotally, patients prefer to have all their care within one setting.

A model for psychological treatments for adults with type 1 diabetes

Based on this review and clinical experience, we developed two manual diabetes-specific psychological interventions which included elements of motivational enhancement therapy (MET) and CBT technique that targeted beliefs and behaviours that maintained poor glycaemic control.

The rationale for conducting an RCT of psychological treatments to improve glycaemic control is justified on the following grounds: sub-optimal glycaemic control in adults with type 1 diabetes is a common problem despite the patient’s and the diabetes team’s best endeavours; and it is associated with multiple psychological problems (depression, anxiety, eating problems and diabetes-specific coping problems), increased morbidity and mortality, and reduced quality of life. A systematic synthesis of RCTs of psychological treatments to help people improve their diabetes self-care and subsequently their diabetes control found that the evidence for their effectiveness in improving glycaemic control was limited, but the explanation for this may be due to methodological limitations in reducing biases, and the nature and validity of psychological techniques. The lack of psychotherapists trained in diabetic medicine limits the availability for people with diabetes, but there is potential to increase the skills of diabetes professionals which has not been evaluated.

Two psychological treatments widely practised are MET and CBT.

Motivational enhancement therapy

Cognitive behaviour therapy

Summary

Cognitive behaviour therapy and MI-based interventions have yet to be fully evaluated in adults with type 1 diabetes. MET and CBT have a number of differences. Very briefly, MET does not explicitly socialise patients into a specific model of behaviour change and it does not introduce thought and activity homework. MET and the creative writing tasks are focused on resolving ambivalence about behaviour change, whereas in CBT a variety of cognitive and behavioural interventions are used to identify, reality test and correct distorted conceptualisations and the dysfunctional beliefs underlying these cognitions.

The literature review appeared to suggest that a period of motivational work may be a pre-requisite for effective CBT. 168 This had clinical validity as, when CBT is delivered as a standalone intervention, its initial assessment phase often includes issues of motivation and ambivalence to change. MET and CBT share common features; both are patient-centred approaches requiring a strong therapeutic alliance aiming to nurture the willingness to change. This novel method of integrating the two approaches has yet to be tested in diabetes settings.

In designing the current trial we trained diabetes nurses to deliver the treatments, as psychologists are a scarce resource in the diabetes setting. We added CBT to MET as one of the interventions rather than testing CBT alone, as we were aiming to reach a group with persistent problems with diabetes control and likely to be ambivalent about change. The primary aim of this RCT was to determine whether MET + CBT was more effective than usual diabetes care in improving glycaemic control in adults with type 1 diabetes and persistent sub-optimal glycaemic control. The second aim was to assess whether MET was more effective than usual diabetes care in improving glycaemic control.

Main aims and objectives

The study project is titled ‘A Diabetes and Psychological Therapies Study (ADaPT)’. Our study population was derived from adults with type 1 diabetes. We selected two psychological treatments, MET and CBT, to test in an RCT. The control group was usual diabetes care. The treatments were adapted to be diabetes-specific and were manual. We used a range of diabetes, psychological and economic measures. The main statistical approach was analysis of covariance (ANCOVA) and we also used methods to impute missing data. The main aims were as follows:

1. To test the effectiveness and cost–utility of MET + CBT compared with MET and compared with usual care in helping patients with type 1 diabetes improve their glycaemic control and quality-adjusted life-years (QALYs).

2. To examine the cost-effectiveness of MET + CBT and MET compared with usual care for improving glycaemic control.

3. To identify cognitive, behavioural and biological predictors of glycaemic control.

4. To assess the effectiveness of MET + CBT compared with MET and with usual care in other secondary outcomes (depression, quality of life, diabetes cognitions and diabetes self-care activities).

Hypotheses

The main hypothesis was that MET + CBT would be more effective than usual care at improving glycaemic control in adults with type 1 diabetes at 12 months’ follow-up.

The subsidiary hypotheses were that (1) MET would be more effective than usual care at improving glycaemic control at 12 months’ follow-up and (2) MET + CBT will be more effective than MET at improving glycaemic control at 12 months’ follow-up.

Design

We used a three-arm parallel RCT as the gold standard design to test the effectiveness of psychological treatments. Following randomisation, participants remained in the study for 12 months. We followed the CONSORT guidelines to inform the conduct of the study and in the reporting of the trial. 169

Setting

ADaPT was co-ordinated by the Clinical Trials Unit, Institute of Psychiatry, King’s College London and registered with Current Controlled Trials (www.controlled-trials.com; International Standard Randomised Controlled Trial Number 77044517). The recruiting centres were based in south-east London (King’s College Hospital, Guy’s and St Thomas’ Hospitals, Lewisham University Hospital and Mayday University Hospital) and Greater Manchester (Manchester Royal Infirmary, North Manchester General Hospital and Stockport General Hospital/Stepping Hill Hospital). The advantage of two geographical sites was that this increased the generalisability of the study findings and focused resources in recruitment in high population density areas. According to the Commission for Racial Equality these sites represent some of the most ethnic and socioeconomic diverse populations in England (www.census.gov.uk).

Ethics approval

Approval was obtained from the South-west Multi-Centre Research Ethics Committee, UK (reference 02/6/101) and the ethics committees of all participating centres. A Trial Steering Committee and Data Monitoring and Ethics Committee oversaw the conduct of the study. All participants provided signed informed consent being given a three-page patient information sheet, a summary leaflet, a face-to-face information giving session and an opportunity to consider with a follow-up telephone call.

Study population, case definition and study criteria

Diabetes mellitus was defined according to the World Health Organization criteria. 170 Participants with type 1 diabetes were recruited between September 2003 and August 2005. The study population consisted of adults (18–65 years) registered as having type 1 diabetes with one previous HbA_1c of between 8.2% and 15.0%, identified by the local investigators using the clinic diabetes database and resident within the recruiting hospital’s health authorities. As there were variations in the administrative and procedural organisation of patient registers between the study sites, the screening and recruitment methods were adapted to each site.

The target population who were assessed for trial eligibility were adults with type 1 diabetes, defined by (a) onset at younger than 35 years of age and (b) onset of insulin therapy within 6 months of diagnosis or ketones in the urine, for a minimum duration of 2 years and their current (measured at time of screening or 1 week before or after the screening assessment) HbA_1c was between 8.2% and 15.0%

Participants were excluded if they:

1. were not fluent in English as this was necessary for psychotherapeutic communication

2. were pregnant or attending a pre-pregnancy clinic

3. had an antidepressant initiated less than 2 months ago to reduce the bias of recovery from depression

4. had an acute or serious medical illness as defined by treating physician

5. had advanced diabetes complications (such as registered blind or serum creatinine values > 300 mmol/l)

6. had known haemaglobinopathy or severe mental disorder; were in psychotherapy or within 3 months of having completed a structured diabetes education programme; or were participating in another trial.

Analysis of glycated haemoglobin

The HbA_1c was measured by ion-exchange high-pressure liquid chromatography using the following analysers at each participating clinic: Menarini HA-8140, HA-8121 or HA-8160 (Menarini Diagnostics, Florence, Italy) or Tosoh 2.2 Plus (Tosoh Medics, Foster City, California, USA) or Variant II HPLC System (Bio-Rad Laboratories, Hercules, California, USA) using methodology aligned to the DCCT.

Baseline measures

Prior to randomisation, data on the five following baseline characteristics were collected.

Randomisation

Randomisation was conducted by the Clinical Trials Unit at the Institute of Psychiatry, King’s College, London. The researchers gave the following information: clinic name and patient initials, hospital number, date of birth and sex. A randomisation list stratified according to centre using minimisation and blocks of random sizes (three, six, nine and twelve) was prepared in advance to ensure a roughly equal number of patients allocated in each of the three arms of the trial while avoiding possible predictability associated with blocks of fixed sizes. If randomised to either the MET or MET + CBT intervention, that participant was assigned to a nurse therapist depending on her availability. One nurse was allocated to the Manchester sites, and at any one time between one and three nurses were allocated to the London sites. Allocation concealment was ensured as the Clinical Trials Unit held the randomisation list in a password-locked computer and a password-locked access program. Once a participant was recruited, the researcher would contact the Clinical Trials Unit data manager who would only then reveal the allocation to himself and to the researcher. Researchers contacted participants by telephone to inform them of the randomisation allocation, to clarify study participation issues and concerns and to allocate a nurse therapist to those receiving therapy sessions. A standard letter was also sent with the dates of their 3-, 6-, 9- and 12-month follow-ups for the HbA_1c blood tests.

Outcome measures

Adverse events

A list of adverse events if and when they are voluntarily reported was compiled. Potential adverse events presently identified include death, psychiatric admission, medical admission, and onset of complication secondary to rapid glycaemic control (painful neuropathy, accelerated retinopathy, hypoglycaemic episodes).

Blinding

At baseline all measures were collected before randomisation. The nurses and technicians who conduct the anthropometry and laboratory analysis were blind to allocation and therefore blind to the main outcome measure of A_1c at each time point. All psychological assessments were included in a self-report questionnaire, therefore blinding of participants was not possible. The nature of psychological treatments as a talking therapy means that participants and therapists cannot be blind to their allocation.

Strategies used to maximise follow-up rate

A number of strategies were used to optimise response rates such as reminder telephone calls and letters of blood test and missed appointments; liaising with GPs for blood test and results and changes in contact details; checking hospital registers and local health authorities for changes in contact details; participants who had dropped out of therapy were still contacted for the final follow-up blood test; handwritten personalised Christmas and birthday cards and newsletters were sent to trial participants.

Adverse events monitoring

All participants were asked about the number of severe hypoglycaemia episodes requiring third-party assistance in the 6 months preceding randomisation and the 6 months preceding the 12-month follow-up. In addition, an open-ended question on any adverse events was asked at each 3-month HbA_1c follow-up.

Sample size calculation

This was based on a hypothesised 0.8% difference in HbA_1c in the MET + CBT (or MET) group compared with usual diabetes care. We assumed that the SD of the changes was approximately 1.65 based on systematic reviews we had previously conducted. 84 At a power of 90%, a type 1 error rate of 0.05 (two-tailed), a randomisation ratio of 1:1:1 and a 20% drop-out rate, we estimated that a sample size of 339 participants (n = 113 in each group) was required.

Statistical analysis

Data were analysed using stata 9 (Stata, College Station, Texas, USA), R (www.r-project.org) and sas version 9.1 (SAS Institute Inc, Cary, NC, USA). Baseline characteristics were compared to assess the effectiveness of randomisation. Patients were analysed according to randomised groups, following the intention-to-treat principle. For the primary outcome of 12-month HbA_1c we used ANCOVA to estimate the differences in intervention group means, adjusting for the baseline glycated haemoglobin based on those who completed their 12-month HbA_1c measurement. We calculated mean within-group changes by subtracting the mean glycated haemoglobin at 12 months from the mean HbA_1c at baseline for completers. We repeated this for the intermediate quarterly HbA_1c outcomes.

We used logistic regression to estimate the odds for any severe hypoglycaemia episode at 12 months, adjusting for whether participants had any severe hypoglycaemia episodes or not at baseline in each intervention group compared with usual care. We assessed whether 12-month HbA_1c varied according to therapist in the intervention arms by fitting ANCOVA models in the two intervention arms, allowing for a therapist effect.

To assess the sensitivity of the results for glycated haemoglobin to missing data, we used multiple imputation to impute missing three, six, nine and twelve measurements. 176 A general location model was used for imputations, assuming multivariate normality for continuous variables and a log linear model for categorical variables, which was fitted using Markov chain Monte Carlo in R. 176 Imputation was performed separately in the three arms and used information from variables associated with missing glycated haemoglobin measurements and those strongly associated with glycated haemoglobin, including ethnicity, employment status, depression, age and, in the intervention arms, whether patients completed their therapy. 177 The distribution of observed HbA_1c at follow-up time points was approximately normal, suggesting our imputation model was reasonable.

For the assessment of baseline moderators or predictors of outcome, we used the ‘glm’ (general linear model) command to fit the ANCOVA models, where 12-month HbA_1c was the dependent variable and baseline HbA_1c was a covariate. An interaction term between the potential moderator and treatment group was used to estimate effect moderation. The ANCOVA model assumes that the residuals of the model predictive of 12-month HbA_1c adjusted for the covariates (baseline HbA_1c and the moderators) are normally distributed and that the variance is the same in each treatment group. The ‘lmatrix’ command was used to estimate the difference in 12-month HbA_1c (adjusted for baseline HbA_1c) at specific levels of the moderators. The treatment effects are reported for each category of each categorical moderator, such as gender. For continuous moderators the treatment effects are reported at a range of levels. The tests for the interaction between treatment group and moderator were carried out for each of the following three pairs: MET + CBT versus usual care; MET versus usual care; and MET + CBT versus MET. The results are reported in the following format: F-ratio value, degrees of freedom for effect of model, degrees of freedom for the residuals of the model, 95% CIs and p-value. Full details of the statistical analysis plan approved by the Trial Steering Committee and the Data Monitoring and Ethics Committee are given in Appendix 1.

Delivery of interventions

Prior to randomisation, all participants were given a fact sheet containing the minimum level of diabetes knowledge expected in people with type 1 diabetes as recommended by UK guidelines and were then randomised to one of the following three.

Protocol changes

We made changes to the protocol that was submitted for funding. All changes were approved by the Trial Steering Committee and given ethics approval.

1. 1. Minimum age for participate in the study increased from 16 to 18 years.

2. 2. The minimum duration of type 1 diabetes at recruitment was increased from 1 to 2 years to reduce the bias of a protracted honeymoon period during which there is fluctuation in pancreatic insulin secretion and exogenous insulin administration.

3. 3. Upgrade to minimum level of disease on entry into the trial. Patients were screened for end-stage diabetes-related complications and serious health problems.

4. 4. We added excluding women who were actively receiving medical help in planning to become pregnant during the trial as this requires intensive input from the diabetes health-care team. We clarified that we did not withdraw participants who become pregnant during the trial unless there was a medical reason to do so.

5. 8. We did not assess perceived social support from health professionals at baseline as the questionnaire was too long.

6. 9. It was not possible to measure total dose of insulin every 3 months. Dosage, type of insulin and number of injections per day were assessed at baseline, and at 6 and 12 months. We replaced social support at 6 months with an assessment of psychological well being (General Health Questionnaire-12)183 and items from the Summary of Diabetes Self-Care Activities Questionnaire172 in order to potentially test any process effects of health technologies on glycaemic control.

Introduction

To the best of our knowledge, there was no formal training programme for teaching general nurses psychological skills that are specific to diabetes. We recruited six nurses and developed a training programme for the purposes of ADaPT. A key component was to assess the nurses competency and ensure this was deemed satisfactory prior to and during the delivery of MET and CBT. We established that general nurses can be trained to deliver a range of diabetes-focused psychological techniques.

Training programme

Six nurses were recruited and trained to deliver the study interventions. Three nurses were H Grade diabetes specialist nurses (DSNs) with experience as DSNs for 4–10 years prior to joining the study. The fourth and fifth nurses collectively had previous working experience in eating disorders and nutrition, community rehabilitation and family planning. The sixth nurse was a CBT nurse therapist who had extensive experience in working with type 1 and 2 diabetes, she was also the main CBT supervisor for the study and trained in MET. All nurses were based in London except one H Grade DSN who saw all patients recruited in the Manchester area. We aimed to develop a training programme that was brief, focused on skills transferable to primary and secondary care settings and with components that would ensure competency and adherence to the study protocol.

Nurses underwent training in MET and CBT simultaneously and in parallel. Training in each of the therapies included a 2- to 5-day course followed by self-directed learning and regular supervision and coaching. This ensured that nurses were given a holistic training incorporating theory, practice and reflection.

Training also involved studying written material and the trial manuals, watching standard MI video training tapes, scoring session transcripts and role play. Each nurse was assigned a caseload of 10–11 practice patients with type 1 diabetes and sub-optimal glycaemic control in keeping with the study population. Supervisors gave feedback on audio-taped sessions and at least one videotaped session. Competency levels were reached before the onset of trial recruitment using the Motivational Interviewing Treatment Integrity (MITI)184 Rating Scale and the Revised 12-item Cognitive Therapy Scale (CTS-R). 185

During the training programme, nurses attended weekly individual and group supervision sessions, using video conferencing to include the nurse based in Manchester and this continued during the trial.

Nurses were trained specifically in the following skills:

• Basic psychotherapy skills, such as an ability to communicate and empathise with people from different backgrounds, using a non-judgemental approach during all interactions.

• Assessing the burden of diabetes and the presence of depression using the Accu-Chek.

• Being able to reflect on their interactions with patients and having the ability to recognise and keep to appropriate boundaries.

• MET skills in how to use complex and simple reflections to demonstrate warmth and empathy while attending to patient levels of motivation and self-efficacy; refrain from confronting and criticising; attend to signs of change; and strengthen the alliance by agreement of tasks and therapy goals.

• Assessment of suicide risk and signs for deteriorating mental state.

• CBT techniques taught to the nurses included identifying and evaluating negative automatic thoughts, developing alternatives to unhelpful rules and assumptions, behavioural experiments, activity scheduling, continuums, responsibility pie charts, examining advantages and disadvantages of different types of coping, anxiety management, assertiveness training with role play and problem solving. We involved significant others when appropriate. Nurses were trained in how to structure each session. This involved agreeing an agenda, reviewing homework, summarising previous sessions and creating a bridge with the current session and reverting to their MI style when resistance interrupted the process of change.

• Studying a written curriculum of key texts and landmark papers and the trial protocol, watching standard MI video training tapes, scoring session transcripts and role play.

• A training caseload of 10–11 patients with type 1 diabetes and sub-optimal glycaemic control who were similar to the trial participants, but not participating in the study. Supervisors gave feedback on audio-taped sessions and at least one video-taped session.

• Skills needed in participating in clinical supervision (see below).

Clinical supervision during conduct of the study

During the training programme, nurses attended weekly group and individual supervision sessions including the use of video conferencing and telephone supervision to include the nurse based in Manchester. As the skills of the nurse therapists and their case loads increased, weekly individual supervision eventually replaced the group sessions. Nurses prepared for supervision choosing a specific difficulty or question (supervisory road map). 186,187 The supervisors modelled reflective practice and nurses were asked to listen to their own audio-taped therapy sessions and to reflect on their strengths and weaknesses. Informal peer supervision and support contributed towards the development of their skills. The aim was to help them develop the ability to think more about the process of change. Nurses’ beliefs about their patients and the therapy were explored to highlight their own contribution to the session and enhance a deeper sense of knowledge of the CBT techniques. We supported them to resist the ‘righting reflex’114,120 which would normally enable them to provide advice and education on improving glycaemic control.

Assessing treatment fidelity

There are many different methods and techniques for the assessment of treatment fidelity. We developed a framework using the following steps:

Results

One hundred and four (88.9%) and 83 (78.3%) participants attended the third sessions in the MET and MET + CBT treatment groups respectively and 70 (66.0%) from the seventh session in the MET + CBT group only (Figure 1). We randomly selected a total of 72 tapes, of which 55 were usable, 17 (23.6%) were unusable due to poor or no sound. Fifteen of the 55 tapes were used for the inter-rater agreement stage.

FIGURE 1.

Study flowchart. A primary analysis based on linear-mixed models used data from all participants, as all participants had at least baseline glycated haemoglobin measured.

Discussion

The ADaPT study demonstrated that general medical nurses can be trained to clinically satisfactory levels of competency in psychotherapy skills based on MET and CBT.

Motivational enhancement therapy and CBT had shared and specific techniques. They shared techniques of pacing and use of time, eliciting of emotional expression and interpersonal effectiveness, and pragmatic use of open and closed questions and of simple and complex reflections.

The two therapies were broadly distinguishable in that practically all CBT and 70% of the MET sessions were accurately recognised as such. MET as expected included more MI-adherent behaviours and CBT included more CBT-relevant techniques such as eliciting key cognitions and application of change methods. Overall, for the group that received MET + CBT it appeared that a more accurate description of what was delivered was that CBT was combined with MET rather than being completely separate from it.

Previous research on MI skills193,194 assessed with the MITI and the MISC show that MI experts exceed the five-point level on the seven-point (range 1–7) Likert rating scale for the MISC global dimensions such as acceptance and egalitarianism, and have a reflection-to-question ratio > 2 and a percentage of complex reflections > 50. In our study the mean MISC global dimensions for the MET tapes ranged from 4.8–5.3. The mean reflection-to-question ratio was close to 2 and the mean percentage of complex reflections was above 60. Our results show the nurses were above average in almost all MITI and MISC components. Our findings are similar to the skills scores obtained in a trial of MET delivered by trained clinicians in the field of substance abuse. 106

Previous research has suggested that the cut-off point for competency in CT/CBT using the CTS-R is 39. 195 As the nurses scored a mean of 52.1 with an SD of 7.5, we can assume they probably delivered CBT skilfully.

The assessment of fidelity is further strengthened for the following reasons. First, the statistical program we used to select the subset of the audio-taped sessions for the inter-rater and treatment fidelity assessment ensured that the results are likely to be representative of the therapy delivered throughout the trial. Second, the raters were blind to the intervention being delivered. We rated a total of 40 tapes (around 20% of all the approximately usable 80% of third MET and seventh CBT sessions). Because of the sampling scheme used to select taped sessions (balanced with respect to rater/therapy type), observed differences between therapy types are less likely to be due to rater differences. Similarly, differences between therapy types are unlikely to be confounded by nurse differences (therapy type and nurse factors were reasonably balanced by design). The estimated ICC for the total CTS-R score was similar to the Pearson’s product moment correlation (r = 0.67) reported by Reichelt and colleagues196 who trained cognitive therapy supervisors to use the CTS-R to rate cognitive therapy video-taped sessions. The ICCs for the two global MITI components in our study were more satisfactory than those published by other researchers in the field. Moyers and colleagues188 obtained ICCs of 0.52 and 0.58 for the ‘empathy/understanding’ and ‘spirit’ items respectively, although this could be due to greater variability in treatment delivery, and half of the ICCs for the behaviour counts in our trial were also similar.

The ICC did not have uniformly acceptable values across all MI and CBT domains. The ICC for the number of MI adherent behaviours was low and for CBT, eliciting and planning behaviours and eliciting key cognitions was borderline low. This raises the question as to whether longer training, enhanced training or more careful selection of health professionals could have demonstrated better ICCs and larger effects on glycaemic control. The sample of six nurses was not a representative sample of diabetes nursing, they are likely to have self-selected as being more psychologically aware, although this is not the same as being more skilled in psychological care. These caveats suggest that further study of the training required and the level of competency that needs to be achieved to elicit the largest effect is needed.

There is no standardised consensus on fidelity assessment, therefore we designed our own approach for this RCT. In view of resources and project milestones we developed an initial protocol to allow us to assess fidelity within the resources and project milestones. Our results may have been affected by the sessions and the session segments we chose to rate. Carroll and colleagues197 rated all their MI and standard treatment taped sessions with the help of 15 independent raters. Also, as the raters rated more tapes it is possible they became unblinded through recognition of the voices of the therapists. Rating more tapes may have increased the power to detect further differences between MET and CBT and also examine differences between MET delivered in the MET only and the MET + CBT groups. Finally, the statistical power needed to detect such differences was limited for some components by their low inter-rater reliability. If raters had used transcripts of the sessions as well as the audio-tapes, reliability may have been higher and further differences may have been found. Further work in this area is ongoing and clearly needed in order to best inform the interpretation of this study and modifications to the next generation of technologies.

Trial CONSORT diagram

We identified a target population of 1659 potentially eligible patients of adults (age 18–65 years) with a probable diagnosis of type 1 diabetes who had at least one HbA_1c in the previous year between 8.2% and 15.0% (Figure 1). A third (n = 578) refused to consent to screening (having their current HbA_1c checked) or their eligibility status could not be completed due to clinic non-attendance and appointment cancellations. A total of 1081 patients underwent further screening. We excluded patients (n = 574) who did not meet our case definition of type 1 diabetes (type 2 diabetes patients, gestational diabetes, latent autoimmune diabetes in adults, maturity-onset diabetes of the young), and people with type 1 diabetes who fell into one of the following categories: current HbA_1c of lower than 8.2%, or diagnosed within the previous 2 years and aged over 35 years at the time of diagnosis.

There were 507 patients with persistent sub-optimal glycaemic control diagnosed with type 1 diabetes in the previous 2 years and aged over 35 years at the time of diagnosis who constituted the eligible sample, and these were fairly proportionately distributed amongst the sites (Table 4). From these, 344 patients were randomised to MET (n = 117), MET + CBT (n = 106) and to usual care (n = 121).

Baseline characteristics

Therapy session attendance

Most participants allocated to either MET or MET + CBT attended at least one session (n =  211; 94.6%) (Figure 1). The majority of participants allocated to MET (82.9%) attended all four sessions compared with just over half of those allocated to the MET + CBT group who completed all 12 sessions (55.7%). The proportion of patients in the MET + CBT group who completed the MET component was also lower than the proportion of patients in the MET group who completed all their sessions (72.6% versus 82.9%), although not statistically significant [χ² (1) = 3.42, p = 0.08]. The average duration of MET, calculated from the date of the first session to the fourth, was 9.7 weeks (SD 6.5) and likewise the average duration of MET + CBT, calculated from the date of the first session until the twelfth, was 26.5 weeks (SD 11.0) for those who completed all allocated sessions. It took approximately 1 month longer than anticipated to deliver the interventions [mean 3.0 months (SD 1.8) for MET and mean 6.8 months (SD 2.6) for MET + CBT].

Quarterly follow-up rates for glycated haemoglobin

Participants with missing glycated haemoglobin values at 12 months (n = 39) did not differ from completers (n = 305) on baseline characteristics except for employment status, significant others and general health perceptions. In particular, missing HbA_1c values at 12 months were not statistically associated with baseline HbA_1c (p = 0.95). Participants who were unemployed were more likely to not complete their 12-month blood test than those in full- or part-time employment [16.7% versus 8.3% respectively, χ² (1) = 5.47, p = 0.02]. Participants who attended the final assessment reported on average a significantly higher number of people they felt close to [t(327) = 1.89, p = 0.006] and better positive general health perceptions [t(327) = 1.90, p = 0.05] than participants who did not attend. Non-completers were also younger [t(342) = 1.77, p = 0.08] and suffered from greater role limitation due to physical problems as assessed by the SF-36 (Short Form-36 Health Survey Questionnaire) [t(334) = 1.97, p = 0.06], but these findings were not statistically significant.

Protocol recruitment violations

After randomisation it became apparent that two study participants did not meet the criteria for type 1 diabetes. One participant was diagnosed with gestational diabetes and later with type 2 diabetes that required insulin therapy. The second participant was diagnosed with steroid-induced diabetes. A third participant who had denied a history of serious mental disorder had a relapse of his manic depression and became lost to follow-up. All three patients remained in the study and were included in the analyses.

Mean change in glycated haemoglobin between groups

The mean HbA_1c at quarterly intervals in each group is shown in Table 8. For the main outcome in the intention-to-treat analysis in those who completed their 12-month follow-up, the 12-month glycated haemoglobin levels adjusted for baseline HbA_1c levels were significantly lower in the MET + CBT group (n = 95) than for usual diabetes care (n = 105) (adjusted mean difference 0.45%, 95% CI 0.12% to 0.79%; p = 0.008); non-significantly lower in the MET group (n = 105) than for usual care (n = 105) (adjusted mean difference 0.16%, 95% CI –0.20% to 0.51%; p = 0.38); and non-significantly lower in the MET + CBT group than in the MET group (adjusted mean difference 0.30%, 95% CI –0.07% to 0.66%; p = 0.11).

Mean change in glycated haemoglobin within each group

There was a significant reduction in the mean HbA_1c from baseline to 12 months within the MET + CBT group (mean difference 0.59%, 95% CI 0.31% to 0.87%; p = 0.0001), but not within the MET (mean difference 0.24%, 95% CI –0.04% to 0.52%; p = 0.10) or the usual care (mean difference 0.12%, 95% CI –0.10% to 0.35%; p = 0.28) groups.

Sensitivity analysis

For the sensitivity analysis, the mean glycated haemoglobin values based on multiple imputed glycated haemoglobin data at quarterly intervals in each group is shown in Figure 2. The glycated haemoglobin values tended to reduce in all groups at 3 months but only in the MET + CBT group was there a persistent and, at 12 months, a significant reduction in HbA_1c. The 12-month HbA_1c levels adjusted for baseline HbA_1c levels were significantly lower in the MET + CBT group than for usual diabetes care (adjusted mean difference 0.43, 95% CI 0.10 to 0.77); non-significantly lower in the MET group than in usual diabetes care (adjusted mean difference 0.20, 95% CI –0.15 to 0.55); and non-significantly lower in the MET + CBT group than in the MET group (adjusted mean difference 0.23%, 95% CI –0.13% to 0.59%; p = 0.21). The mean HbA_1c based on the observed measurements is generally lower than when based on the imputation, suggesting that participants who missed the 3-, 6- and 9-month assessments tended to have higher HbA_1c.

FIGURE 2.

Mean glycated hemoglobin levels based on available measurements at each time point (n = 344)

Harmful events monitoring

In the 233 participants with data, there was no significant difference in the reporting of one or more severe hypoglycaemia episodes at 12 months in the MET + CBT group or the MET group than in the usual care group [adjusted odds ratio 0.50 (95% CI 0.17 to 1.45; p = 0.20) and adjusted odds ratio 1.15 (95% CI 0.43 to 3.08; p = 0.78) respectively].

Assessment of potential predictors associated with main outcome

The following factors were assessed: age (in years) at randomisation, gender, marital status, ethnicity, educational qualifications, employment status, age at diagnosis of type 1 diabetes, duration of diagnosis (in years), baseline HbA_1c, diabetes complications (neuropathy and retinopathy only as too many missing values for nephropathy), depression (continuous), anxiety (categorical), somatising disorder (categorical), eating disorders (categorical), fear of hypoglycaemia (HFS-II behaviour and worry subscales), dietary and exercise on the Summary of Diabetes Self-Care Activities, smoking status (ex- and non-smokers versus current smokers), and number of alcohol units consumed per week. The results are summarised in Tables 9–11 and for the purposes of parsimony only the statistically significant findings are highlighted in the text.

There was weak evidence that the effectiveness of MET + CBT compared with usual care depended on baseline HbA_1c (p = 0.062), but not for MET when compared with usual care (p = 0.50) (n = 305). For those who completed their 12-month follow-up, the reduction in 12-month HbA_1c in the MET + CBT group compared with usual care was estimated to increase by 0.32% (95% CI –0.02% to 0.66%) for each 1% increase in baseline HbA_1c.

There was tentative evidence that age was an effect modifier: the younger the participant the larger the reduction in HbA_1c in those who received MET + CBT than in those under usual diabetes care (p = 0.0008), but this was not observed when comparing MET with usual care (p = 0.93).

While MET + CBT compared with usual care did not appear to be any more effective for white than for non-white participants, MET compared with usual care was significantly better for white participants with a 1.03% lower HbA_1c than for non-white participants (p = 0.03).

Participants who reported more worry about hypoglycaemia were more likely to benefit from MET + CBT by a 0.04% decrease in HbA_1c than from usual care [F(1, 156) = 0.05, p = 0.82]. For every one point increase in worry about hypoglycaemia, the MET group had a significantly higher HbA_1c by 0.04% than usual care [F(1, 165) = 4.90, p = 0.03]. This was not paralleled by high risk behaviours aiming to reduce risk of hypoglycaemia.

For every additional day in the week that participants in the MET + CBT group checked their glucose levels there was a 0.22% increase in their HbA_1c (p = 0.001) compared with the control group. For every additional day in the week that participants spaced their carbohydrates evenly through their day, MET + CBT was associated with a 0.16% increase in HbA_1c [F(1, 191) = 5.35, p = 0.02] compared with the control group and the difference was statistically significant. For every additional day per week that participants in the MET + CBT group engaged in 30 minutes of physical exercise there was a 0.12% decrease in their HbA_1c [F(1, 193) = 2.78, p = 0.09] compared with the control group and this difference was marginally significant.

There was no evidence that the effectiveness of MET + CBT or MET on 12-month glycated haemoglobin varied according to gender, marital status, educational status, employment status, duration or age of diagnosis of type 1 diabetes, BMI, diabetes complications, baseline PHQ-9 depression score, presence of syndromal anxiety, eating disorder or somatoform disorders. After adjusting for baseline glycated haemoglobin, 12-month glycated haemoglobin did not vary according to therapist in either the MET + CBT (p = 0.63) or the MET (p = 0.34) groups.

Change in other secondary outcomes

Introduction

As previously discussed in Chapter 1 (Socioeconomic impact of diabetes), diabetes and its complications incur substantial costs for health services. Costs can also fall upon patients themselves and wider society. For example, given that people with type 1 diabetes commonly develop the condition at a young age, they may experience various employment-related impacts throughout their lives, which could translate into important productivity losses for society. Such high costs, alongside high prevalence and the chronic nature of the condition, necessitate a need for cost-effective approaches to treatment and long-term management. To the best of our knowledge, there have been no previous RCT-based economic evaluations of psychological interventions for adults with type 1 diabetes. This chapter reports a comprehensive economic evaluation carried out as part of this trial to assess the cost-effectiveness and cost–utility of MET and MET + CBT in addition to usual care compared with usual care alone.

Methods

Results

Cost-effectiveness and cost–utility

Of the 18 cost–outcome combinations examined for the cost-effectiveness and cost–utility analyses, only one showed statistical between-group differences for both cost and outcome elements: the MET + CBT group had higher health and social care costs (mean difference = £790, 95% CI £507 to £1072) and greater HbA_1c improvement (mean difference = 0.45 points, 95% CI 0.11 to 0.80) than usual care. This translated into an ICER of £1756 (Table 23).

TABLE 23 - Cost-effectiveness ratios

	MET vs usual care	MET + CBT vs usual care	MET vs MET + CBT
Cost per additional point improvement on HbA1c, health/social care perspective	3821	1756	636
Cost per additional point improvement on HbA1c, societal perspective	4593	1809	514
Cost per additional QALY (EQ-5D), health/social care perspective	48,636	311,970	MET dominates
Cost per additional QALY (EQ-5D), societal perspective	160,750	271,333	MET dominates
Cost per additional QALY (SF-36), health/social care perspective	133,750	3,950,000	MET dominates
Cost per additional QALY (SF-36), societal perspective	160,750	4,070,000	MET dominates

In other cost–outcome combinations, both intervention groups had numerically higher costs and better outcomes than usual care. The MET + CBT group had higher costs and better HbA_1c outcomes than the MET group, but the MET group dominated with regard to QALY outcomes. Where relevant, indicative ICERs are presented for information, but should be interpreted with caution given that they are based on point estimates which (a) do not represent any uncertainty surrounding these and (b) showed that, aside from the MET group having higher health and social care costs than usual care and the MET + CBT group having better HbA_1c outcomes than usual care, all other costs and outcomes were not statistically different.

Incremental cost-effectiveness ratios for HbA_1c improvements were lower for MET + CBT compared with usual care than for MET alone. ICERs for QALYs notably differed for the two interventions and all far exceeded the acceptable threshold implicitly suggested in the National Institute for Health and Clinical Excellence’s decision-making. 206 In comparisons of the two interventions against each other, MET + CBT cost an additional £514 or £636 per additional point improvement over MET, depending on the cost perspective taken.

Figures 3–8 show probabilities of cost-effectiveness for each intervention compared with usual care and for the two interventions against each other, again for all cost–outcome combinations. Probabilities of cost-effectiveness for both interventions were generally higher when based on the outcome of HbA_1c improvements than on QALY gains, reaching over 0.7 at thresholds of £5000 or higher for each additional point improvement on HbA_1c for MET + CBT compared with usual care from both health and social care (Figure 5) and societal (Figure 6) perspectives. Equivalent probabilities for MET compared with usual care did not exceed 0.51 at the thresholds of £45,000 per additional point improvement in HbA_1c. MET + CBT had high probabilities of cost-effectiveness compared with MET alone, reaching 0.8 at thresholds of £5000 per additional point improvement in HbA_1c from both cost perspectives.

FIGURE 3.

CEACs for MET versus usual care, health/social care perspective.

FIGURE 4.

CEACs for MET versus usual care, societal perspective.

FIGURE 5.

CEACs for MET + CBT versus usual care, health/social care perspective.

FIGURE 6.

CEACs for MET + CBT versus usual care, societal perspective.

FIGURE 7.

CEACs for MET versus MET + CBT, health/social care perspective.

FIGURE 8.

CEACs for MET versus MET + CBT, societal perspective.

Probabilities of cost-effectiveness based on QALYs were generally very low for all comparisons. At thresholds of £20,000 per additional QALY, probabilities of cost-effectiveness for MET compared with usual care were 0.08 (EQ-5D) and 0.03 (SF-36) from the health and social care perspective; probabilities from the societal perspective were higher than this, but still low. At the same threshold, MET + CBT had zero probability of cost-effectiveness compared with usual care from the health and social care perspective and 0.06 (EQ-5D) and 0.11 (SF-36) probability of cost-effectiveness from the societal perspective. Again for the same threshold of £20,000, probabilities of cost-effectiveness for MET + CBT compared with MET were 0.11 (for both the EQ-5D and SF-36) from the health and social care perspective and 0.13 (EQ-5D) and 0.24 (SF-36) from the societal perspective. Although the two approaches to QALY estimation produced very different ICERs, they resulted in very similar CEACs.

Limitations

The economic evaluation had some limitations. The cost data carried a risk of error due to participant recall bias. CSRI questions asked about resource use and other economic impacts for the previous 3 months at baseline, and for the previous 6 months at the 6- and 12-month assessments. The reliability of self-reporting over such durations is unclear. However, this data collection approach seemed appropriate for three reasons. Firstly, given the multisite nature of this study and the broad evaluation perspective taken (due to the breadth of health and other impacts of diabetes), it was infeasible to examine the records of multiple care providers, and some participant self-reporting still remained necessary (e.g. for measuring time off work and out-of-pocket expenses). Secondly, measuring resource use for less than a 1-year period risked finding artificially differential costs simply because of differences in the timing of patient care reviews. There is a potential for a temporary ‘flurry’ of health-care activity around the time of these reviews, and cost estimates could vary depending on whether they included or excluded such a period. Using a 1-year measurement period without breaks reduced the likelihood of this. Thirdly, we tried to maximise quality by collecting these data by interview (in person at the baseline assessment and by telephone at 6 and 12 months), rather than self-complete and, due to resource constraints, it was not possible to collect the data any more frequently than we did. It could generally be assumed that the impact of such bias was even across the randomisation groups, and the lack of difference in costs when the intervention costs were excluded supports this assumption. However, we cannot discount the possibility that costs were double-counted if intervention group participants mistakenly reported therapy sessions as part of their usual diabetes care, in which case health-care costs for those two groups may be over-stated and any potential savings resulting from the interventions concealed.

Table 15 suggested that the partial sample used for the economic analyses may not have been fully representative of the full study sample. However, when those potentially differing characteristics were used to impute missing costs and outcomes to generate a full sample with complete data, the findings suggest that values obtained from the partial sample were a good representation of the full sample. Further analyses, perhaps based on alternative imputation techniques, may be needed to check the robustness of this conclusion.

Finally, the time horizon of the evaluation is likely to have been insufficient to identify relevant longer term outcomes for this patient group. For example, reductions in HbA_1c may result in fewer complications in the future, which may in turn impact on longer term quality of life. A cost–utility analysis within a longer term evaluation may therefore suggest quite different conclusions.

Discussion and conclusions

Outline of discussion

The main aim of this study was to compare the effectiveness of two types of psychological treatments to be delivered by nurses specially trained in these technologies, with usual diabetes care in improving glycaemic control in a group of people with persistent difficulties with diabetes control. In this chapter, the main findings will be summarised, the advantages and limitations of the study design will be examined, this study will be compared with other published literature and suggestions for future research will be put forward.

Main findings

Our first finding was that MET + CBT was associated with a statistically significant reduction of nearly 0.5% of HbA_1c compared with usual diabetes care. The second finding was that the reduction in HbA_1c in the MET group compared with usual diabetes care was not significant, but the upper confidence limit did suggest that some people improved. The quarterly measurements of HbA_1c showed that there were initial reductions in the HbA_1c in all three groups, but these persisted at 12 months in only the MET + CBT group. The pattern of results did not change when we conducted sensitivity analyses by using imputed data for missing values of HbA_1c.

Our third finding was that baseline HbA_1c and age were significant moderators of the treatment effect, but only when MET + CBT was compared with usual diabetes care. The worse the glycaemic control at baseline, the greater was the reduction in HbA_1c. For instance, participants with HbA_1c of 12% on average had a reduction in the region of 1.5–2% if they were in the MET + CBT group compared with usual care. Those in their 20s had larger reductions in HbA_1c, again in the region of 1–2% points compared with usual diabetes care than those in their 40s who had reductions in the region of 0.3%.

The fourth finding was that MET was associated some improvement in body weight which almost reached significance, but otherwise neither of the experimental technologies were associated with any significant improvement in any of the other secondary outcomes.

The fifth finding was that neither MET nor MET + CBT is an undisputedly cost-effective treatment approach compared with usual care alone in the short term. Both interventions led to significantly higher health and social care costs over 1 year because the additional costs of MET and CBT did not appear to be offset by savings elsewhere. MET + CBT had greater probabilities of cost-effectiveness compared with usual care than did MET, if value was placed on HbA_1c outcomes (over 0.7 at thresholds of £5000 per additional point improvement in HbA_1c); but MET had a greater chance of cost-effectiveness if value was placed on QALY outcomes [although at a threshold of £20,000 per additional QALY, probabilities only reached 0.31 (based on the SF-36)]. MET + CBT had a good probability of cost-effectiveness compared with MET alone based on HbA_1c outcomes but, based on QALYs, it was dominated by MET and had low probabilities of cost-effectiveness. Therefore, cost-effectiveness conclusions are dependent on the relative importance of these two outcomes. These broad conclusions apply from both a health/social care and societal perspective.

Methodological issues

Comparison with other intervention studies

The RCTs identified in our review84 that perhaps seem most similar appear to be those by Glasgow and colleagues,104 Fosbury and colleagues,98 Pouwer and colleagues105 and van der Ven and colleagues. 111

In terms of the target population, Fosbury and colleagues98 and van der Ven and colleagues111 were the only studies that to the best of our knowledge included adults with persistent sub-optimal glycaemic control. Persistent sub-optimal control was defined as having HbA_1c > 9.0% for the former study and ≥ to 8.0% for the latter on two or more consecutive clinic visits over a period > 12 months.

Glasgow and colleagues’104 intervention is the most similar to the MET intervention we evaluated. They developed a diabetes-specific intervention which used the transtheoretical model of stages of changes, the concept of self-efficacy component and a computer-based assessment and feedback, which included the standardised Summary of Diabetes Self-Care Scale, patient beliefs about barriers to self-care and readiness to change, and a one-page feedback form. The main themes were then discussed and advice was given. In our intervention nurses also reviewed the feedback form with their patients but, in keeping with the MI spirit, they were advised to refrain from giving advice unless they first requested and received permission. This intervention was similar to our MET arm.

Pouwer and colleagues105 trained DSNs to use a computer-based well-being questionnaire and to explore the results with their patients in a non-judgemental way using active listening and exploration of feelings. This intervention was similar to our MET arm and likewise they did not find a significant reduction in glycaemic control, but this group did not have significant sub-optimal glycaemic control at baseline and were not selected on this basis and this was not the main outcome.

Fosbury and colleagues98 compared CAT with individual diabetes-specific counselling and education sessions with a DSN. An experienced CAT therapist saw all patients allocated in the intervention group for between 16 and 20 50-minute sessions. CBT techniques, mainly cognitive such as diary keeping and self-monitoring, were integrated with approaches theoretically derived from object relations that aimed to address resistance to therapy and the persistence of ‘seriously (self-)damaging behaviours’. CAT is a unique intervention, but if it has to be grouped it was most similar to our MET + CBT arm. They did find a significant difference between CAT and nurse education, again supporting our findings that the more intense (by either type or duration) therapy package is associated with better outcomes.

Our findings contrast with a recent study that found that six sessions of a group CBT intervention did not significantly reduce glycated haemoglobin in adults with type 1 diabetes. 111 Our study suggests that a therapy tailored to the individual may be more effective than a group setting, which may explain the difference in the results of the two studies.

An important aspect of our study that distinguishes it from others was whether with appropriate training, psychological techniques specific to problems with diabetes self-care could be delivered by non-mental health-care professionals to improve diabetes outcomes, as access to both specialist nurses and psychotherapists is usually not possible nor always desirable. We demonstrated that specialist nurses could be trained to deliver diabetes-specific psychological treatments competently and effectively in terms of reducing glycaemic control. Whether extended or enhanced understanding of therapist styles would further improve competencies and lead to better effect on glycaemic control warrants further inquiry.

Another difference between our study and the majority of previous RCTs was that we compared two psychological treatments with usual care. MET failed to improve glycaemic control compared with usual care, but when combined with CBT the combination was effective. This could be due to the techniques specific to CBT. Alternatively it may be that longer rather than shorter treatment is required to improve glycaemic control; the possibility that treatment effect is confounded by treatment dose cannot be tested in this study and therefore cannot be ruled out. In the clinical setting, CBT is most effective when patients are motivated to change their behaviours, and supporting motivational change initially in a group of people who have found it difficult to change their diabetes self-care behaviours has face validity. There have been recent similar advances in the medical and behavioural management of alcohol abuse and dependence where a state-of-the-art individualised therapy titled ‘COMBINE’ that begins with motivational enhancement techniques and is followed by cognitive behavioural techniques is currently being evaluated. 210

Clinical significance of findings

There are several clinical implications of our study considering that there are national and international recommendations supporting psychological care in type 1 diabetes6,211 and yet there is a severe shortage of psychotherapists to meet this demand. First, we achieved a clinically meaningful reduction in glycated haemoglobin. In the DCCT study, although normalisation of blood glucose values was not achieved in the intensive treatment group, the reduction in glycaemic control that was achieved was associated with a 40–60% reduction in the development of microvascular complications over the 7 years of follow-up. 9–11 It should be noted that if the reduction of 0.5% that was achieved in this study was maintained over a longer period, e.g. of several years, this significantly reduces the risk of diabetes microvascular complications. A longer term follow-up is needed for this, but these findings do hold promise in suggesting that an initial outlay of setting up psychotherapy services in diabetes could lead to longer term health benefits than those measured here.

Second, with appropriate manual skills training, diabetes nurses can make a significant contribution in improving glycaemic control.

Third, the effectiveness of combining MET with CBT in improving glycaemic control may be related to its focus on diabetes-specific problems rather than on general psychological distress.

Fourth, we observed in subgroup analysis that those with the worst glycaemic control appeared to make the largest gains in reducing their glycaemic control. This finding should be interpreted with caution as it was underpowered, but suggests potential clinical subgroups into which this intervention could be translated.

Fifth, the economic evaluation found greater health and social care costs in the intervention groups than in the usual care group. This appeared to be a result of the additional costs of the interventions, rather than increased use of other services. Therefore, although a component of our complex intervention was paradoxically to increase the use of diabetes resources through improved and greater responsibility for self-care, there was no evidence of this occurring.

The average reduction in glycaemic control was small in absolute numbers and it could be argued that this was not of sufficient clinical significance. We consider this reduction as clinically significant based on the evidence that there is a continuous association between glycaemic control and diabetes complications such that any reduction in the former is associated with a reduction in the risk of the latter. It also needs to be emphasised that the study population was a difficult to treat group of patients who despite attending clinics are still not achieving optimal control. Further, if the focus is on the average HbA_1c, we risk ignoring the possibilities of change suggested by the CI which includes the possibility of nearly 1% reduction. The strength of this study is that psychological treatments clearly seem to have a positive effect and it is likely that further refinement of the intervention may lead to larger effects.

Half the sample completed all 12 sessions in the MET + CBT arm and over two-thirds completed at least half the sessions. This is similar to rates reported in other RCTs of CBT. 212 This raises several questions; if the completion rate for all sessions had been higher, would this have translated to a larger effect size? This is difficult to answer as people drop out of psychotherapy for many different reasons. We did not measure this qualitatively or quantitatively, but in translating into the clinical setting it would be important to evaluate this and develop techniques to reduce attrition to therapy, such as better preparation for therapy and better identification of patients who are more likely to benefit. This intervention asked people with diabetes to take time out of their week when they are already taking time out of their many social roles and responsibilities for various medical appointments. There needs to be a culture shift in both the medical teams and the patients to accepting the need for psychological care as a dimension of diabetes care. Incorporating explicit well-defined psychological care and psychological outcomes as objectives in national guidelines would be a way forward in improving the acceptability of psychological treatments in diabetes.

Research significance of findings

We developed a training programme within a research context to train general medical nurses to deliver the interventions. With input from experts in those fields of clinical practice and research we developed treatment manuals, patient workbooks and a training programme that, in a short period of time, gave nurses basic psychotherapy skills as well as MET- and CBT-specific skills. The training had active learning components and a treatment fidelity assessment component to ensure nurses were competent before the onset of the study recruitment. The syllabus is relatively easy to replicate and adapt and indeed we have already translated the MET component into a postgraduate/Masters course design at King’s College London (www.kcl.ac.uk/nursing) which can be taught and can certify health professionals and with supervision can be used in clinical settings.

This was a definitive phase III intervention, but there are aspects of it that probably need further study in order to understand what are the key components of the intervention. The two main, not mutually exclusive, interpretations, whether it was treatment type or treatment duration, cannot be fully answered in this study. Our opinion is that it is more likely to be treatment type as we found that there was adequate treatment fidelity and that the two therapies could be distinguished. The underlying philosophy of our study was that MET would bring about change in motivation and that once the person with diabetes was ready to change, this was when CBT would be most effective as the person was in a better position to use the therapeutic alliance to develop new psychological skills in managing diabetes. This can be partially addressed if a process evaluation of the sessions that were taped was conducted.

Given that the improvements in glycaemic control were small and there were no improvements in psychological outcomes, the question is raised as to how the intervention could be improved. We adapted standard CBT techniques to diabetes and accordingly this raises the question as to whether there should be a more flexible approach by including psychological problems, such as depression and anxiety, in the assessment and through problem formulation such that the intervention incorporates treatment of key mental health problems. In the original protocol, including common mental disorders would have intensified and prolonged the training and therefore the costs. However, considering the increasing recognition of depression as a comorbidity in diabetes and other chronic metabolic conditions and that primary care and (non-mental) health professionals are increasingly expected to have basic skills in psychological assessment and psychological care, translation of the intervention should perhaps include components to deal with the assessment and management of common mental disorders.

We could conduct qualitative analyses of all the therapy sessions to throw more light onto the content of the sessions and identify recurrent themes. It would also be interesting to investigate the role of ‘sudden gains’ or rapid improvements which usually occur early in therapy and have been found to be associated with better long-term outcomes in depression. 213,214 For the development and evaluation of brief and time-limited psychological interventions, the sudden gain effect may prove to be particularly useful in terms of cost-effectiveness, the introduction of such interventions to clinical settings, and improvement in attendance rates. We could improve our baseline assessment to include better measures of social support and other psychological measures, ensure we have less missing data on the diabetes-related complications, introduce techniques to improve therapy attendance and have longer follow-up. We could also refine our recruitment strategies and improve the follow-up assessments even more.

The fact that participants in the intervention arms received their intervention from different therapists potentially complicated the analysis and conclusions in two ways. First, differences in effectiveness between the intervention groups could be due to imbalance in the distribution of MET and MET + CBT patients between the therapists. For example, if all the MET + CBT patients received therapy from the first three therapists while all the MET patients received therapy from the other three therapists, differences between the groups could merely reflect the difference in ability of therapists. In this study all six nurses delivered therapy to both MET and MET + CBT patients and most therapists had a good balance of MET and MET + CBT patients and there was no statistical evidence that 12-month HbA_1c varied according to therapist. The second potential complication is that if outcomes did vary according to therapist, this should be accounted for in regression models, to allow for the clustering of patients within therapists. However, as there was no evidence of differences between outcomes by therapist, a model with a random therapist effect could not be estimated.

Future research

Transcribing of all the available audio-taped sessions would help analyse their contents to obtain a more in-depth idea of recurring themes and the process of behavioural, cognitive change and change in effect as it occurs within and across sessions. This thorough investigation might also shed more light on the nature and timing of ‘sudden gains’ and the possible reasons for therapy non-attendance. There is an existing significant body of research that has looked at such processes that we could use to form and test hypotheses in diabetes-specific MET and CBT and assess whether these factors are associated with outcome. 181,195,214,215

User feedback of the therapeutic process and participating in the study could be collected to help inform how to improve identifying potentially eligible participants, to reduce the follow-up rates and to improve the delivery of any future intervention.

There is more scope to study the type of training needed for optimal level of competency to deliver the maximum effect on glycaemic control. We also need to understand in more detail why younger people and those with worse control made bigger changes than those with better control.

The technologies evaluated here focused on glycaemic control rather than psychiatric morbidity. Previous RCTs in diabetes have tended to focus on improving psychological well-being, and perhaps it is better to include improving psychological outcomes as a primary outcome and not just glycaemic control as they are more likely to be ‘patient important’ outcomes. 216 It is possible that the values and preferences of patients are different to the medical team’s values of achieving optimal clinical targets.

In designing future RCTs, attention needs to be given to the components of usual care. Diabetes education is part of usual care and should be incorporated into future studies that aim to compare true usual care with additional psychological interventions.

The patterns of change in mean glycated haemoglobin at 3, 6 and 9 months showed that MET was effective up to a certain time point. It is possible that the effect of brief MET could be maintained up to 12 months and longer if ‘booster’ MET sessions after a few months from the end of therapy were added instead of CBT.

There is also scope for exploring the utility and effectiveness of self-help materials and for web-based therapies versus face-to-face interventions. A recent study evaluated an interactive CD-ROM for depression and although only a quarter of those who were given the option to use it actually did so, those who did had a clinically and statistically significant reduction in their symptoms of depression. 217 It is possible that a CD-ROM with diabetes-specific components could have a similar impact. In the same line of thought, it would be helpful in terms of service availability and use to explore the potential of replacing some or all face-to-face sessions with telephone ones. Sacco and colleagues’218 small pilot study (n = 10) showed some encouraging results. They evaluated a brief, regular, semi-structured, CBT-based telephone ‘coaching’ intervention designed to be delivered by trained psychology undergraduates. Patients found the intervention acceptable and effective in reducing the percentage of glycated haemoglobin. Such an intervention could potentially increase the response rate particularly to extended therapy interventions.

Quality-adjusted life-years estimated from the SF-36 and EQ-5D resulted in similar CEACs, but different ICERs because of differences in the size of utility estimates generated by the two instruments. Further work is needed to explore the appropriateness and reliability of each approach for different contexts and what the implications for decision-making are when faced with comparing evidence based on a mixture of such approaches.

Conclusion

This Health Technology Assessment report evaluated the effectiveness of improving glycaemic control of two types of diabetes-specific psychological interventions compared with usual care for adults with type 1 diabetes and persistent sub-optimal glycaemic control. Patients were randomly allocated to one of the three groups: usual diabetes care, MET and MET + CBT. Both interventions were delivered competently by nurses. Compared with usual care, glycaemic control as measured with glycated haemoglobin at 12 months significantly improved in the MET + CBT group, but not in the MET group. It was not possible to fully distinguish the independent effects of treatment duration and treatment type on the main outcome.

Neither intervention fell within a notional policy-making threshold of cost-effectiveness (i.e. £20,000 per QALY gain) in a 1-year evaluation. In comparisons against usual care, MET + CBT achieved HbA_1c improvements at a lower cost (£1756 per additional point improvement) than MET. Probabilities of cost-effectiveness were higher based on HbA_1c outcomes than on QALY outcomes. Therefore, decisions regarding the provision of such interventions depend on the relative importance of these two outcomes.

This study substantially adds to the evidence that psychological treatments can have as effective a role to play as adjunct therapies to help improve diabetes control. This study also raises a number of important questions about how to translate these findings, such as whether to broaden the training of nurse therapists to include psychological techniques for the assessment and management of psychiatric comorbidity. There are a number of research opportunities, such as modifications of the intervention and focusing on specific subgroups in diabetes, that arise from this study that may be associated with better outcomes.

Contribution of authors

Dr Khalida Ismail (Diabetes and Psychiatry) developed the study hypotheses and was the principal investigator and guarantor. Dr Ismail, Professor Janet Treasure (Pyschiatry), Professor Ulrike Schmidt (Psychiatry), Professor Trudie Chalder (Cognitive Behaviour Therapy), Dr Stephen Thomas (Diabetologist) and Dr Anita Patel (Health Economics) contributed to the design and supervision of the study protocol. Professor Chalder led the design of the CBT intervention and the training of the nurses. Professor Treasure and Professor Schmidt led the design of the MET intervention and the training of the nurses. Ms Esther Maissi was the trial co-ordinator for the whole study and carried out the recruiting and the follow-ups and data collection in the London sites. Dr Chris Dickens (Liaison Psychiatry) and Professor Francis Creed (Liaison Psychiatry) were the lead investigators for the Manchester sites. Mr Jonathan Bartlett developed the statistical plan and performed the statistical analyses. Dr Ismail drafted the manuscript, and all authors contributed to the writing and approval of the final manuscript. The corresponding author had full access to all the data in this study and had final responsibility for the decision to submit it for publication.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.