Tackling obesity in areas of high social deprivation: clinical effectiveness and cost-effectiveness of a task-based weight management group programme a randomised controlled trial and economic evaluation

Weight management strategies

In 2011, the Department of Health published a policy paper that called for efforts to reduce the proportion of adults with excess weight by 2020. 11 A range of strategies will be required if the UK is to reduce its obesity rates. 9 These strategies include personal health-care interventions, such as weight management programmes and weight loss medicines, as well as education and environmental changes.

The menu of evidence-based interventions currently available for people unable to lose weight on their own is relatively limited. A stepped-care approach is currently the recommended approach for weight management depending on the severity of the patient’s obesity. Current pharmacological treatments have modest effects that can be beneficial but are likely to be lost once the medication is stopped. 12 Surgical interventions are more successful but are currently expensive and unsuitable for large-scale use, and are usually indicated for the morbidly obese or those with coexisting conditions. 13,14 Dietary interventions on their own have only modest effects15 and brief routine interventions within primary care have generally reported disappointing results. 16

More intensive behavioural interventions generate a small but sustainable weight loss,17 which can engender significant and clinically worthwhile long-term health benefits. 18 Despite the fact that some of the initial weight lost is regained, interventions that lead to at least a 5% reduction in body weight can lead to health improvements (e.g. a decreased risk of type 2 diabetes mellitus). 19–22

Obesity is a chronic condition that requires lifelong management, as weight is often regained, but achieving changes in behaviour is challenging. 23 Weight management in overweight individuals who seek help normally requires changes to their habitual lifestyle, which are difficult to implement and maintain without specialist input, structure and support. 24,25 The National Institute for Health and Care Excellence (NICE) guidance on Managing Overweight and Obesity in Adults recommends multicomponent interventions as the treatment of choice (Box 2). 26 These interventions should include behaviour change techniques to increase people’s physical activity levels or decrease inactivity, improve eating behaviour and the quality of the person’s diet, and reduce energy intake. Several systematic reviews have demonstrated that the combination of diet, exercise and behavioural approaches are effective management strategies. 27 However, few studies have specifically targeted primary care patients.

Intensive weight management programmes can make considerable demands on staff expertise and budgets, and they also face the challenge of participant retention. The programmes usually include, as one of their core active ingredients, assignment and monitoring of tasks. These are difficult to implement for most participants and the participant dropout is usually large. 28

Primary care interventions

Similar to smoking cessation, primary care has the potential to play a key role in helping overweight and obese people to achieve a healthy weight because of its unique role in the health-care system.

General practice is potentially an ideal location for running weight management services. People trust the advice of their general practitioner (GP) team, their GP practice is often local and convenient, and large GP practices now have multidisciplinary teams and physical space to operate weight management services. Some patients, especially those from ethnic minority groups, may be less likely to use commercial providers. 29–31 GPs have also been incentivised via the Quality and Outcomes Framework to maintain a register of patients (aged ≥ 16 years) with a BMI of ≥ 30 kg/m² as part of routine care. 32

Current guidelines recommend that primary care physicians in England should identify people with obesity and offer clinical management, although few options for treatment exist. In 2013, the Royal College of Physicians published a report on how the NHS should adapt to deal with the rising rates of obesity. 33 This report highlighted the role of GPs and the practice team, recommending that GPs should deal with excess weight and obesity as an important risk factor for non-communicable diseases. Although most obesity management in the UK takes place in primary care, the approach is not co-ordinated or consistent. 34

Despite the advantages of targeting obesity in primary care, the effectiveness of interventions in this setting has not been widely evaluated. Where studies have been conducted, the results can be disappointing (Box 3). 34–36 If research evaluations showed little impact then real-life impact is likely to be worse.

Several recent systematic reviews have suggested that weight loss interventions in primary care yield small reductions in weight that are not likely to be clinically significant. 42,43 In the USA, a recent review concluded that obesity treatment delivered in primary care has limited effectiveness. 44

One strategy that has shown modest effectiveness is primary care referral to evidence-based commercial programmes for weight loss treatment. 36

The NICE guidelines recommend that GP practices raise the issue of weight loss with overweight patients and refer them to weight management services, where these exist. 26 These guidelines also recommend referring people to a group rather than an individual programme if they express no preference because, on average, group programmes tend to be more cost-effective.

In the field of health behaviour modification, group approaches can dramatically reduce the costs of treatments and increase their reach. 45 They may also have potential to improve participant retention. Social support has been associated with positive change in a number of areas, including weight management. 46 Some potentially useful pointers can be derived from the field of smoking cessation, which shares a number of key features with weight management. Interaction-oriented groups have been shown to improve attendance and participant retention,47 mutual linking of individual tasks improves treatment compliance and short-term outcome,48 and on a national scale group treatments seem to be yielding results superior to individual treatment. 49 Current group weight management programmes usually have a strongly didactic focus, with limited efforts to utilise social support and to link the progress of individual participants. It is likely that the mutual support-oriented group approach, which has proved useful in smoking cessation, can be used here as well.

Several types of such programmes have been commissioned by local councils, but their efficacy is generally not known. With previous evidence suggesting that obesity interventions in primary care have had little impact, there is a need for evidence-based public domain weight management programmes that are clinically effective and cost-effective, and readily accessible and attractive for patients from diverse ethnic and socioeconomic backgrounds.

Primary objective

To determine whether or not the WAP can generate better-sustained weight loss over 12 months in overweight adults than a best practice intervention delivered by nurses in general practice.

Secondary objective

To determine the cost-effectiveness [in terms of costs of interventions and quality-adjusted life-years (QALYs) derived from the European Quality of Life-5 Dimensions-5 Levels (EQ-5D-5L) questionnaire] of the two interventions.

Inclusion criteria

Participants were eligible if they were aged ≥ 18 years, wanted to lose weight and had a BMI of ≥ 30 kg/m² or ≥ 28 kg/m² plus comorbidities.

Exclusion criteria

Those who were unable to read/write/understand English, had a BMI of > 45 kg/m², had lost > 5% of their body weight in the previous 6 months, were pregnant, were taking psychiatric medications, were not registered with a GP in the local areas or were involved in a current research project were excluded.

The decision to exclude participants on psychiatric medication, including antidepressants, was based on the fact that these medications can have a significant effect on weight, and that psychiatric illness often makes follow-up and adherence to long-term programmes more difficult. We did not exclude people with a history of psychiatric illness if they were no longer taking psychiatric medication.

We did not exclude any other comorbidities to ensure that the study addressed NHS needs and that the results are generalisable. Clients who were unable to exercise were not excluded as both the nurse and the WAP interventions are multimodal and do not rely solely on exercise.

Recruitment

Box 4 lists the strategies that we used to recruit participants.

FIGURE 1.

Leaflets (a) and posters (b) used to promote the study.

FIGURE 2.

General practitioner fax referral form.

Recruitment commenced in September 2012 with the first participant enrolled at the Barkantine Practice, London, UK, on 24 September 2012. The first participant was screened at the Lawson Practice, London, UK, on 18 October 2012.

We encountered some difficulties in recruiting our sample in the short time frame that we had set. These are discussed in Chapter 5, Recruitment barriers and facilitators.

All publicity (except GP fax referrals) invited potential participants to contact the study team by telephone. A researcher would explain the study, assess interest and eligibility, and invite the potential participant to attend the initial screening session.

Screening procedures

Participants were either invited to telephone the study team if recruited by posters or leaflets, or telephoned by the study team if referred by GP fax referral.

At the initial screening telephone call, a good clinical practice (GCP)-trained member of staff provided the participant with information on the study. If interested, participants were screened for eligibility over the telephone. Eligible participants were booked onto the next available screening session and were posted or e-mailed the participant information sheet, baseline questionnaire and letter of invitation in advance of this. Participants who were not eligible but still interested in losing weight were either offered the option to attend the standard care clinic or advised to visit their GP for further advice on weight management.

Informed consent procedures

Participants were provided with detailed trial information and allowed sufficient time (at least 24 hours) to consider whether or not they wanted to participate in the trial. All participants were provided with a participant information sheet with more details of the study.

All participants provided written informed consent at the baseline (first) screening session, prior to being randomised to the study arms.

Written informed consent was obtained by an appropriately GCP-trained member of staff delegated by the investigator as documented on the site delegation log, prior to any participation/study-specific procedures.

Randomisation procedures

If eligible, participants were invited to attend the randomisation session a few days later. At this session, participants completed further questionnaires and had their weight, waist circumference and blood pressure recorded. They were then randomised (see Randomisation for more detail of randomisation procedure) to the WAP or nurse (weight loss intervention from a trained GP practice nurse) arms. The first session of the WAP and the nurse intervention were provided within 7 and 14 days of the randomisation session. Table 2 summarises the main purpose of the study visits. All visits were held face to face.

Practice nurse intervention

We standardised the nurse intervention to ensure that participants received a consistent standard of care. The nurse intervention was modelled on the best practice intervention in primary care, derived from discussions with GPs and practice nurses, and incorporating national guidelines at the time51 and NHS materials (Your Weight, Your Health: Raising the Issue of Weight in Adults52).

In 2011, when we were designing this project, we conducted a survey of weight management interventions in a range of general practice surgeries. GPs typically provided brief advice followed by referral to a practice nurse. A minority of practices used dietitians but this was slowly being phased out. Some practice nurses had received 1-day training in weight management and provided one-off sessions or sessions with a degree of follow-up, either optional or scheduled, over 2–8 weeks. In about half of the practices, the nurses also referred patients to local community-based physical activity programmes.

We modelled the nurse intervention on the more intensive end of the spectrum, which is still routinely practicable across GP surgeries. Participants received weight management intervention from a practice nurse who had been given training in the study procedures by the research team. The nurses provided the intervention in four sessions delivered over 8 weeks.

The intervention included advice on (1) diet (instructions on understanding food groups, food labels and calories; eat at least five portions of a variety of fruit and vegetables each day in place of foods higher in fat and calories; eat breakfast; watch the portion size of meals and snacks; and replace high-calorie food with healthier options); and (2) activity (make enjoyable physical activities part of everyday life; minimise sedentary activities; build activity into the working day; and take up one of the local exercise opportunities). Table 3 shows a summary of the control intervention. Each session lasted up to 30 minutes. Participants received information about local exercise provision and ‘exercise on prescription’, and received relevant vouchers and referrals. This advice was supported with written materials. Participants received a:

• Drink Swap: How to Cut Down on Calories in Drinks without Having to Say ‘No’ leaflet53

• Portion Swap: How Smaller Plates and Portions Help Prevent us Eating too Many Calories leaflet54

• Snack Swap: How to stay Healthy Without Giving Up all Snacks leaflet55

• Walk 4 Life: Tips to Get Walking Every Day leaflet56

• The Eatwell Plate leaflet57

• 5 a Day: What Counts? leaflet58

• Food Labels leaflet (created by the Health and Lifestyle Unit, Wolfson Institute of Preventive Medicine, London, UK – available on request)

• calorie guide (created by the Health and Lifestyle Unit, Wolfson Institute of Preventive Medicine, London, UK – available on request)

• exercise guide (created by the Health and Lifestyle Unit, Wolfson Institute of Preventive Medicine, London, UK – available on request).

Where appropriate, participants were given an information sheet about orlistat (based on the information provided on the NHS Choices website59) and advised to see their GP if they wished to use it as part of their weight loss programme.

Participants’ weight was recorded at all treatment and follow-up visits. Participants were not restricted from using any other weight loss intervention (including pharmacological treatment if their GP agreed it was appropriate) during the study. They were, however, asked to report on the use of such interventions during the study period.

Weight Action Programme group intervention

The WAP is a multimodal health behaviour modification intervention developed at the Wolfson Institute of Preventive Medicine via extensive client feedback and piloting with underprivileged groups since 2002. The programme is a multicomponent service utilising evidence-based behaviour change techniques in the context of group support targeted to individual needs that aims to provide participants with tools to lose weight and to maintain a long-term healthy lifestyle.

The evidence-based strategies and contents include:

1. self-regulation through the use of (1) food diaries to monitor caloric intake; (2) self-monitoring of weight; and (3) goal-setting and contingent reinforcement

2. motivational components incorporating the standard elements of cognitive–behavioural interventions aimed at encouraging and improving self-efficacy, facilitated by a range of concrete and verifiable tasks agreed individually with each participant (e.g. participants agree incremental pedometer targets)

3. fostering a non-judgemental support network strengthened by shared experience, outcome expectations, positive reinforcement and information on coping with lapses and long-term support

4. dietary advice, information on healthy eating and caloric content of food, cue management, provision of opportunities for exercise and close monitoring of exercise levels.

Participants commit to implementing each of a series of concrete and verifiable tasks for at least 1 week (see Box 8 for a full description). They can drop the task after that if they find it unhelpful.

Another innovative feature of the programme consists of the use of group-oriented interventions aiming to increase participant retention, involvement and adherence to weekly tasks. For example, ‘buddy pairs’ of participants are made responsible for each other’s completion of the weekly task and weight loss of 1 lb (0.45 kg) between the pair. The group format also makes the programme more cost-efficient. Facilitator-led group support creates an environment in which participants can discuss their progress, identify patterns of behaviour and develop coping strategies to facilitate weight loss and maintenance.

The programme was initially implemented within NHS Tower Hamlets, and then modified in the light of participants’ feedback to make it suitable for underprivileged groups, including ethnic minorities. Where information is imparted, it is mostly in a pictorial and easily understandable format.

The WAP has been evaluated in two pilot studies of 162 overweight adults (mean BMI of 35 kg/m²) from multiethnic areas of high deprivation. 60 The average weight loss was 2.8 kg at the end of treatment and 4.5 kg at the 3-month follow-up (with 24% of participants attending follow-up losing ≥ 5% of their body weight). Limited promotion via GP practices and local adverts generated a large volume of interest. The client retention was at least as good as in comparable programmes conducted in research settings with more traditional clients (59% completed the 6-week treatment) and the programme received very high approval ratings. Clients also demonstrated significant improvements in knowledge of healthy eating, and in their exercise levels, as measured by pedometer monitoring. Clients considered the group support essential in helping them to stick to their tasks and to lose weight. 60 In its current form, the WAP also includes information on orlistat.

The version of WAP used in the trial comprised eight weekly sessions, followed by monthly follow-up visits lasting up to 1 hour each. The content of the programme is summarised in Table 4. The target weight loss was 1 lb (0.45 kg) per week. Two advisors conducted the WAP sessions in groups of 10–20 participants.

Participants were provided with an Oregon pedometer PE980 (Oregon Scientific, Tualatin, OR, USA).

As in the control intervention, participants were not barred from using any other weight loss intervention (including pharmacological treatment from their GP). They also received information about local exercise provision and where ‘exercise on prescription’ was available, they received relevant vouchers and referrals. Participants were asked to report on the use of such interventions during the study period.

To help improve replication and further evidence synthesis, Table 5 summarises the content of the WAP according to the CALO-RE (Coventry, Aberdeen, London – Refined) taxonomy of behaviour change techniques for changing physical activity and healthy eating behaviours. 61

Primary outcome

The primary outcome is the change in weight (in kg) at 12 months post randomisation.

Secondary outcomes

We recorded the following secondary outcomes:

• change in weight (in kg) at 1, 2 and 6 months post randomisation

• change in BMI at 1, 2, 6 and 12 months post randomisation [BMI is calculated as weight (in kg) divided by the square of height (in metres); the height measured at screening was used for each follow-up assessment]

• change in waist circumference (in cm) at 2, 6 and 12 months post randomisation

• change in systolic blood pressure (in mmHg) at 2, 6 and 12 months post randomisation

• change in diastolic blood pressure (in mmHg) at 2, 6 and 12 months post randomisation

• change in the Food Craving Inventory score (frequency domain) at 1, 2, 6 and 12 months post randomisation

• change in the Food Craving Inventory score (strength domain) at 1, 2, 6 and 12 months post randomisation

• change in Food Knowledge Assessment Questionnaire score at 2, 6 and 12 months post randomisation

• change in the Three-Factor Eating Questionnaire score (cognitive restraint domain) at 2, 6 and 12 months post randomisation

• change in the Three-Factor Eating Questionnaire score (uncontrolled eating domain) at 2, 6 and 12 months post randomisation

• change in the Three-Factor Eating Questionnaire score (emotional eating domain) at 2, 6 and 12 months post randomisation

• change in the International Physical Activity Questionnaire score [metabolic-equivalent (MET) minutes/week domain] at 2, 6 and 12 months post randomisation

• change in the International Physical Activity Questionnaire score (sitting domain) at 2, 6 and 12 months post randomisation

• proportion of participants losing 5% of body weight at 2, 6 and 12 months post randomisation

• proportion of participants losing 10% of body weight at 2, 6 and 12 months post randomisation.

Baseline

The following variables were collected at baseline:

• Demographics: includes age, sex, ethnicity, employment and level of education.

• Health and lifestyle: includes smoking status, alcohol consumption and general health.

• Weight loss history: includes number of past weight loss attempts, methods used, most weight ever lost and regular monitoring of weight.

• Concurrent medications: all current medications are recorded.

• Height and weight: measured in centimetres and kilograms respectively, BMI was calculated from these. Height was measured, without shoes, on a Seca 2013 portable stadiometer (Seca, Birmingham, UK). Weight was measured on an Omron HBF 400 Body Fat Monitor and Scale (Omron Healthcare UK Ltd, Milton Keynes, UK), with participants wearing light clothing and no shoes. Accuracy was ensured by calibration against standard weights.

• Waist circumference: measured in centimetres.

• Blood pressure: resting blood pressure recorded using an Omron 705IT BP monitor (Omron Healthcare UK Ltd, Milton Keynes, UK) using an appropriately sized cuff.

The following questionnaires were also administered at baseline:

• International Physical Activity Questionnaire62

• Food Craving Inventory63

• Three-Factor Eating Questionnaire 64

• EQ-5D-5L questionnaire65

• Use of Health Services Questionnaire (see Appendix 1).

We also administered a picture-based food knowledge assessment at baseline and at follow-up. This was developed by the Health and Lifestyle Unit, Queen Mary University of London, to measure basic knowledge of caloric content of different food groups.

Scoring details for the Food Craving Inventory, the Food Knowledge Assessment Questionnaire, the Three-Factor Eating Questionnaire and the International Physical Activity Questionnaire are available in Appendix 2.

Timing of measurements

The study sessions are summarised in measurement schedule shown in Table 6.

Adverse events

We used the sponsor’s definition of an adverse event (AE), defined as any untoward medical occurrence in a subject to whom the intervention has been administered, including occurrences that are not necessarily caused by or related to the intervention.

At every visit all participants were asked whether or not they had experienced any AEs since their last contact with the research team.

All AEs were categorised by a member of the research team, blinded to treatment group, according to their severity and whether or not they were related to participation in the Peer-Support Weight Action Programme (SWAP). When possible, serious adverse events (SAEs) and any AEs for which relatedness to participation in SWAP was not clear were followed up by a telephone call to the participant. AEs that occurred before the baseline measurement period were not recorded.

Serious adverse events

A SAE was defined as an adverse event meeting at least one of the following criteria:

• fatal

• life-threatening

• necessitating inpatient hospitalisation or prolongation of existing hospitalisation

• resulting in persistent or significant disability/incapacity

• a congenital anomaly/birth defect

• otherwise considered medically significant by the investigator.

Any SAEs were reported immediately to the chief investigator, the sponsor and the Research Ethics Committee. A report of all SAEs was provided at every Trial Management Committee and Trial Steering Committee meeting.

Follow up

The following variables were collected during follow-up visits: weight, waist circumference, blood pressure, International Physical Activity Questionnaire,62 Food Knowledge Assessment, Food Craving Inventory, Three-Factor Eating Questionnaire,64 EQ-5D-5L,65 Use of Health Services Questionnaire (see Appendix 1), AEs, participant feedback (see Appendix 3) and use of any concomitant weight loss treatment.

In the intervention arm, the following were collected during the 8-week intervention phase: pedometer use, food diary use and adherence to weekly tasks (e.g. increase in fruit and vegetable intake, increase in exercise, monitoring television and computer use).

Data collection

All data were collected in the paper clinical record form, questionnaires, and on participant diaries and task cards. All data were kept in accordance with GCP and data protection requirements. 66

Data entry

Data were entered into Oracle Database version 11 (Oracle Corporation UK Ltd, London, UK), an online database hosted at the Barts Cancer Centre. The electronic data capture forms are web based and built using Java, with data validation in JavaScript (Java framework Struts 2; Oracle Corporation UK Ltd, London, UK).

Data quality

When recruitment and follow-up were completed, the study team cleaned the data. Source data verification was also conducted by taking a random sample of 10% of case report forms. A member of the quality assurance (QA) team (based at the Pragmatic Clinical Trials Unit; PCTU) compared all written entries with those entered onto the main study database. The prespecified data quality target was a ≤ 2% discrepancy rate between entries in the case report form and the electronic database, which was met.

Process measures

The process measures included attendance throughout the programme, duration of involvement in the programme (time to dropout), results of knowledge tests, participant feedback on components of treatment at 2, 6 and 12 months (e.g. weekly tasks, new information, group discussion, buddy system), and use of concomitant treatments. Some of these process measures were available only in the WAP arm.

Change from planned analysis

In version 1.0 of the trial protocol we specified that we would use a baseline observation carried forward (BOCF) approach for dealing with patients with missing weight data during follow-up. This approach assumes that all those who were lost to follow-up returned to their exact baseline weight. Although this approach has been commonly used in other RCTs, it is problematic because it will provide biased estimates of the treatment effect when this assumption is incorrect (i.e. when participants do not return to their exact baseline weight when they fail to show up to their 6- or 12-month appointment). 68 In addition, BOCF will often lead to an inflated type I error (false-positive) rate as it tends to underestimate the standard error for the treatment effect (as a result of ignoring the within-patient variability in weight when imputing using BOCF). 68 This is particularly problematic in the SWAP trial, as it is unlikely that all participants who are lost to follow-up will return to their baseline weight; in many cases, we would expect them to gain weight. Cross-sectional and prospective cohort studies show that individuals gain weight over time, with an average weight gain per year of 0.5–1 kg. 69

We therefore decided to use a mixed-effects linear regression model for the primary analysis. This analysis method provides unbiased estimates of treatment effect and correct type I error rates, provided that the data are missing at random. That is, the probability that a participant is lost to follow-up depends on their previous weight measurements (e.g. their weight at baseline and 6 months if they are lost to follow-up at 12 months) and baseline patient characteristics. 70

This strategy of analysis has been widely recommended in the presence of missing outcome data. 68 The decision to change analysis methods was made before we had any access to the trial data or ongoing trial results and, therefore, there was no risk of bias associated with this decision.

The statistical analysis plan is provided in Appendix 4.

General analysis principles

All analyses were performed using intention-to-treat principles, meaning that all participants with at least one recorded outcome during follow-up were included in the analysis, and participants were analysed according to the treatment group to which they were randomised. 71 More information on which participants were included in each analysis is available in Missing data for outcomes. All p-values are two-sided, and the significance level was set at 5%.

All analyses accounted for clustering by group in the WAP arm and clustering by nurse in the nurse arm. 72,73 Each participant has been defined as belonging to a cluster, by which group they belonged to if they were in the intervention arm and by which nurse they were treated by if they were in the control arm. This variable has been included as a random intercept in a mixed-effects regression model. This analysis assumes that the intraclass correlation coefficient is the same between groups in the intervention arm as it is between nurses in the control arm. The Kenward–Roger degree of freedom correction was used for all linear mixed-effects models. 74

All analyses were adjusted for baseline weight, age, sex, ethnicity (white British, white other, black, Asian, mixed or other), smoking status (smoker vs. non-smoker) and GP practice (Lawson vs. Barkantine) as covariates in a regression model. 75–77 Outcomes that were measured at baseline were also adjusted for the value of the outcome at baseline (this includes weight, BMI, waist circumference, systolic and diastolic blood pressure, Food Craving Inventory, Food Knowledge Assessment, Three-Factor Eating Questionnaire, and International Physical Activity Questionnaire). Continuous covariates (baseline weight and age) were assumed to have a linear association with outcome. Binary and categorical covariates (sex, ethnicity, smoking status and GP practice) were included in the regression model using indicator (dummy) variables. Missing baseline data were accounted for using mean imputation. 78

Missing data for outcomes

For outcomes that are measured at multiple time points during follow-up, we based our analysis strategy on that proposed by White et al. 71 To deal with incomplete data (i.e. when patients have missing data at one of the follow-up time points) we:

• attempted to follow up all randomised patients even if they withdrew from the study

• performed a main analysis of all observed data that are valid under a plausible assumption about the missing data

• performed sensitivity analyses to explore the effect of departures from the assumptions made in the main analysis

• accounted for all randomised participants, at least in the sensitivity analyses.

In the analyses we:

• Included all patients with at least one post-randomisation assessment (i.e. if they have recorded data for at least one follow-up time point) in the analysis. This allows data from patients who dropped out before 12 months to contribute to the treatment effect estimate at 12 months (e.g. patients with recorded data at 1 month but who dropped out after that would still contribute towards the 12-month analysis).

• Used a mixed-effects model adjusted for baseline covariates, which assumes that the data are missing at random (i.e. they are missing based on their observed outcome at other time points, and other patient characteristics, Box 6).

• Performed sensitivity analyses under other missing data assumptions (e.g. that patients who were lost to follow-up gained more weight than patients who remained in the trial).

Analysis of primary outcome

The primary outcome (change in weight at 12 months post randomisation) was analysed using a mixed-effects linear regression model. The model included change in weight at 1, 2, 6 and 12 months as outcomes.

The model included a random intercept for ‘cluster’ (group or nurse, depending on treatment arm). The correlation between observations at different time points from the same patient (1, 2, 6 and 12 months) was modelled using an unstructured correlation structure. The model was estimated using restricted maximum likelihood. Treatment arm, time point (month 1, 2, 6 or 12) and the interaction between treatment arm and time point were included in the model as fixed factors. Time point was included as an indicator variable. The covariates listed in General analysis principles were also included in the model as fixed factors.

This analysis approach meant that any participant who had a recorded weight for at least one follow-up session (at either 1, 2, 6 or 12 months) was included in the analysis for the primary outcome. So, for example, a participant who lost 0.5 kg at 1 month but had no further weight measurements at 2, 6 or 12 months would still be included in the primary analysis, and would contribute towards the estimated treatment effect at 12 months. Their 12-month weight would be estimated based on their weight at 1 month, their treatment group and their baseline factors, such as baseline weight, age, sex, ethnicity, smoking status and GP practice. This analysis approach provides unbiased estimates of treatment effect provided the reason the participant’s weight data at 2, 6 and 12 months are missing is based upon their observed weight at 1 month or their baseline characteristics (e.g. participants with lower weight loss at 1 month are more likely to be lost to follow-up at 12 months). 79

Sensitivity analyses for primary outcome

Analyses of secondary outcomes

Change in weight at 1, 2 and 6 months were included as outcomes in the same analysis model as change in weight at 12 months.

The analyses for change in BMI, waist circumference, systolic and diastolic blood pressure, and the Food Craving Inventory, Food Knowledge Assessment and Three-Factor Eating Questionnaires all used the same method of analysis as the primary outcome, with the exception of which baseline covariates were included in the analysis. These differences are summarised in Table 7.

The analysis of the proportion of patients losing 5% of their body weight used a mixed-effects logistic regression model. The model included as outcomes whether or not participants had lost 5% of their body weight at 2, 6 and 12 months. The model included three levels. The top level included a random intercept for ‘cluster’ (group or nurse, depending on treatment arm). The second level included a random intercept for patient and a random slope for time point. The third level included patient’s visit (i.e. whether it was the patient’s 2-, 6- or 12-month visit). Treatment arm, time point and the interaction between treatment arm and time point were included in the model as fixed factors. The fixed effect for time point was included as an indicator variable. This analysis adjusted for the same baseline covariates as that of the primary outcome.

The proportion of patients losing 10% of their body weight was analysed separately at 6 and 12 months (the analysis at 2 months was not performed because of the small number of events at this time point). The analysis at 12 months used a mixed-effects logistic regression model, with a random intercept for ‘cluster’ and adjusted for the same baseline covariates as in the analysis of the primary outcome. The analysis at 6 months also used a mixed-effects logistic regression model, but adjusted only for baseline weight (as the model did not converge when the other covariates were included).

Other data summaries

Data summaries are also provided for:

1. number of participants on both treatment arms who used orlistat

2. weight change at 12 months in participants who used orlistat versus those who did not

3. participant feedback (mean and SD, number and per cent) in both treatment arms at 2, 6 and 12 months.

Statistical software

All analyses were implemented in Stata version 14 (StataCorp LP, College Station, TX, USA).

Departures from the statistical analysis plan

• Sensitivity analysis: patients who became pregnant or had bariatric surgery.

  • The statistical analysis plan stated that for this sensitivity analysis patients who became pregnant or had gastric surgery would be excluded from the point at which they had surgery or became pregnant (i.e. their follow-up data from before pregnancy or surgery would be included in the analysis). However, the date of pregnancy was unavailable for a number of patients. Therefore, this analysis completely excluded patients who became pregnant or had gastric surgery.

• Secondary outcome: proportion of patients losing ≥ 10% of their body weight.

  • The statistical analysis plan stated that this outcome would be analysed in the same way as the proportion of patients losing ≥ 5% of their body weight (i.e. using a three-level mixed-effects logistic regression model). However, this analysis model did not reach convergence. We therefore tried to refit the model after removing the random slope for time point, but the model still did not reach convergence. We therefore analysed this outcome separately at 6 and 12 months (i.e. separate logistic regression models were used at each time point); we did not analyse this outcome at 2 months because of the small number of events. The analysis at 12 months was performed using a mixed-effects logistic regression model with a random intercept for cluster and adjusted for the same baseline covariates as the analysis of the primary outcome. However, this model did not converge for the 6-month analysis; we therefore removed all baseline covariates from the model except for baseline weight.

Sensitivity analyses on primary outcome

We performed two sensitivity analyses to assess the robustness of our primary analysis to different assumptions regarding the missing data.

Figure 6 shows the results for different missing not at random assumptions. These scenarios assume that participants were lost to follow-up because of their weight, and that those lost to follow-up (non-responders) had different weight values at 12 months than those who provide data (responders). The values along the x-axis represent the mean change in weight from baseline for non-responders in the nurse arm (usual care). Then, each colour (black, green or blue) represents the mean change in weight for non-responders in the WAP arm (black indicates that non-responders in the WAP arm lost, on average, 5 kg more at 12 months than non-responders in the nurse arm; green indicates that they lost the same amount; and blue indicates that WAP non-responders gained 5 kg more, on average, than the nurse arm non-responders). The points and bars in the graph indicate the treatment effect and 95% CI for each scenario.

FIGURE 6.

Treatment effect and 95% CI for change in weight at 12 months under the assumption that non-responders are missing not at random. UC, usual care.

The outcome here is that results are not materially affected unless we assume that non-responders (patients who were lost to follow-up at 12 months) in the WAP arm gained more weight than non-responders in the nurse arm, or if we assume that non-responders in both arms gained a large amount of weight (i.e. 10 kg). Under the assumption that, on average, non-responders in both arms had no change from their baseline weight, the results are unaffected (difference –2.4 kg, 95% CI –4.3 to –0.5 kg).

Table 19 shows the results of the two other sensitivity analyses. One is the complete-case analysis, where only patients with recorded data at 12 months are included, and the other excludes women who fell pregnant and participants who had bariatric surgery, after randomisation. The treatment effect is not substantially altered.

Changes in body mass index

As expected, change in BMI followed change in weight, with participants in the WAP arm showing a greater reduction in BMI than those in the nurse arm (Table 20).

Changes in waist circumference

Table 21 shows that participants in the nurse arm had a greater reduction in waist circumference than those in the WAP arm at the end of treatment (–7.7 vs. –3.9 cm; p = 0.001), but this was reversed at 6-month follow-up (–1.5 vs. –5.0 cm; p = 0.004). By 12 months, the difference had narrowed (–2.0 vs. –4.1 cm) and the difference was no longer significant (p = 0.07).

Proportion of participants losing at least 5% or 10% of their baseline body weight

At the 12-month follow-up the proportion of participants who had lost at least 5% of their body weight was significantly greater in the WAP arm than in the nurse arm (41% vs. 27%; p = 0.004). Similarly, the proportion of participants who lost 10% of their baseline body weight was higher in the WAP arm than in the nurse arm (participants in the WAP arm were twice as likely as those in the nurse arm to have lost 10% of their weight), but the difference was not significant (Table 22).

Changes in blood pressure

The only significant change in blood pressure between the arms was a greater drop in systolic blood pressure in the nurse arm at the end of treatment (–9.6 mmHg vs. –2.1 mmHg; p = 0.02). At the 12-month follow-up there was no significant difference in blood pressure between participants in the nurse and WAP arms (Table 23).

Changes in food knowledge

Participants in the WAP arm showed a significant increase in their knowledge of the calorie content of foods compared with participants in the nurse arm at the end of treatment and at the 6-month follow-up (Table 24). By the 12-month follow-up this effect had disappeared.

Changes in food craving

All participants showed a decrease in the frequency and strength of food craving at the 1-, 2-, 6- and 12-month follow-up points (see Table 24). There were no significant differences between the groups.

Changes in Three-Factor Eating Questionnaires

Changes in the scores of the domains of the Three-Factor Eating Questionnaire were minimal (see Table 24). Participants in both arms showed a small increase in cognitive restraint scores, but the increase was greater in the WAP arm than in the nurse arm at the end of treatment (0.4 vs. 0.2; p = 0.05). Participants, on average, showed small decreases in uncontrolled and emotional eating scores. There were no significant differences in the changes between study arms.

Changes in levels of physical activity

Participants in both arms increased their levels of physical activity above baseline across the duration of the study to the same extent (Table 25) (818 MET-minutes/week vs. 264 MET-minutes/week; p = 0.09).

Participants reported reducing their sitting time by 1 hour at the end of treatment, but no significant differences between groups was observed.

Resource utilisation and costs

Nurse intervention (control)

The control group intervention was based on best usual care, consisting of four sessions lasting 1 hour each delivered over 8 weeks by a practice nurse. The cost of the control intervention also includes equipment costs (materials, digital scales, blood pressure monitors, batteries, measuring tape, stationery and venue). The cost per participant of the nurse intervention is £180.90. For further details of costs related to nurse intervention, see Appendix 2, Table 42.

In addition to the direct cost of providing weight management interventions, the economic analysis considers consequences of intervention on wider NHS resources. Table 27 provides observations of the resource utilisation (by study group), as measured using the resource-use questionnaire in four categories. The first category indicated contacts with general practice and community nursing services. The second widens the perspectives to consider contacts with social services. The third considers contact with psychiatric services (both hospital and community). The fourth, and final, category considers other in-hospital services, NHS Direct, paramedics and prescriptions. All statistics on resource use are reported as mean and SD, based on the final available sample for the complete-case cost-effectiveness analysis (WAP arm, n = 116; nurse arm, n = 63).

TABLE 27 - Resource utilisation. Available sample to inform complete-case analysis (WAP arm, n = 116; nurse arm, n = 63)

A&E, accident and emergency.

Full details of other services, as specified by participants, are available on request.

Type of service	Time point
	Baseline				6 months				12 months
	WAP arm		Nurse arm		WAP arm		Nurse arm		WAP arm		Nurse arm
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
General practice and community nursing services
GP (surgery)	2.696	2.381	3.097	3.308	1.904	2.675	1.887	2.057	1.53	2.019	1.871	1.806
GP (home)	0.009	0.093	0.081	0.635	0.009	0.093	0	0	0.035	0.263	0	0
GP (telephone)	0.583	1.026	1.097	2.133	0.513	1.165	0.581	0.915	0.383	0.96	0.645	1.319
Nurse (surgery)	1.078	1.434	1.113	1.45	0.565	1.01	0.774	1.453	0.591	1.059	0.532	1.004
Nurse (home)	0	0	0.661	5.08	0	0	0	0	0	0	0.016	0.127
Counsellor (surgery)	0.148	1.086	0.629	4.331	0.052	0.292	0.129	0.64	0.348	2.347	0.097	0.469
Other practice contactsa	0.104	0.502	0.033	0.256	0.052	0.346	0.048	0.216	0.035	0.227	0.177	1.274
Social services
Social worker	0.052	0.475	0	0	0.07	0.588	0	0	0.052	0.346	0.016	0.127
Home help	1.67	15.806	1.968	15.494	0	0	0	0	0.209	2.238	0	0
Care assistant	0	0	0	0	0.73	7.833	0	0	0.887	9.324	0.032	0.254
Day centre (visit)	0	0	0	0	0	0	0	0	0	0	0	0
Other social service contactsa	0.017	0.187	0.161	1.27	0.07	0.434	0	0	0.017	0.187	0.194	1.524
Psychiatric hospital and community services
Psychiatrist (hospital)	0.035	0.227	0.129	1.016	0.026	0.208	0.129	1.016	0.009	0.093	0.048	0.381
Psychiatrist (home)	0	0	0	0	0	0	0	0	0	0	0	0
Psychologist	0.009	0.093	0.161	1.27	0.009	0.093	0.161	1.27	0.009	0.093	0.403	2.036
Psychiatric nurse	0.017	0.131	0	0	0.009	0.093	0	0	0.009	0.093	0	0
Other psychiatric service contactsa	0	0	0	0	0	0	0.5	2.702	0.209	1.899	0	0
Other services
Hospital (day case)	0.043	0.205	0.952	6.109	0.017	0.131	0.048	0.216	0.061	0.305	0.032	0.254
Hospital (A&E)	0.165	0.494	0.323	0.647	0.104	0.36	0.129	0.527	0.13	0.45	0.065	0.248
Hospital (outpatient)	0.443	1.371	0.71	1.712	0.522	1.629	0.419	0.967	0.487	1.334	0.419	0.95
Hospital (inpatient)	0.096	0.577	0.113	0.483	0.026	0.208	0.081	0.635	0.017	0.131	0.016	0.127
Other hospital contactsa	0.165	0.794	0.129	0.461	0.061	0.404	0.032	0.178	0.07	0.508	0.113	0.483
NHS Direct	0.096	0.418	0.081	0.329	0.009	0.093	0.048	0.216	0.061	0.483	0	0
Ambulance or paramedic	0.043	0.244	0.097	0.433	0.026	0.16	0.081	0.417	0.07	0.472	0.016	0.127
Prescriptions received (6 months)	2.991	5.905	4.048	6.759	2.757	4.26	2.29	3.659	2.252	3.541	2.613	4.451

Examining the profile of resource use at baseline, there are small differences evident between the two study groups across the four categories. In the category General practice and community nursing services (see Table 27), across all subcategories rates of service contact was higher in the nurse arm. Social work and psychiatric services others displays a pattern of large magnitudes of service use in the nurse arm compared with the WAP arm. These patterns may indicate potential selection bias. Differences in service use are most commonly explained by individuals’ health status (at baseline) or by differences in the age distribution between groups. Baseline health-state utilities are presented later in this chapter (see Table 30). Differences in age distribution by study group are presented in Figure 7. Observations would suggest there exists some potentially influential difference in the age distributions between groups and, therefore, is it important that cost-effectiveness analysis controls for age when explaining both costs and outcomes.

FIGURE 7.

Age distribution by study arm.

Follow-up data were collected at 6 and 12 months and these visits represent NHS contacts following the intervention. The most frequently reported service contact in the post-intervention period was contact with a GP in the surgery. The mean number of contacts by group at 6 months was 1.90 (SD 2.675) in the WAP arm and 1.89 (SD 2.057) in the nurse arm. At 12 months, the mean number of contact in the WAP arm was 1.53 (SD 2.019) and in the nurse arm was 1.87 (SD 1.806). Observing this pattern over time may suggest that the nurse arm shows reduced service use at 6 months (compared with the WAP arm), but by 12 months the mean number of GP surgery visits was lowest in the WAP group (potentially attributable to an effect of the available maintenance phase for up to 12 months with the WAP arm).

At baseline, the mean number of nurse contacts within the surgery setting was comparable between the WAP (1.08, SD 1.434) and nurse arms (1.11, SD 1.45), and the magnitude of variance was the same in both trial arms. At the 6-month follow-up, the mean number of service contacts in the WAP arm (0.56, SD 1.01) was lower than in the nurse arm (0.77, SD 1.453), and variance was also reduced. This trend seems to have been sustained to 12 months in the WAP arm (0.59, SD 1.059); however, by this time, the mean number of nurse visits in the nurse arm (0.53, SD 1.004) was comparable to that in the WAP arm.

The home help variable demonstrates a significant amount of variance from baseline in both the WAP (1.67, SD 15.806) and nurse arms (1.97, SD 15.494). A small number (n = 3) of influential outliers reported very high levels of home help (ranging between 24 and 168 visits from home help during the 6-month period).

The mean number of calls to GP at baseline was lower in the WAP arm (0.58, SD 1.026) than in the nurse arm (1.10, SD 2.133). By 12 months, the average number of calls decreased in both the nurse arm (0.64, SD 1.319) and the WAP arm (0.38, SD 0.96); similar to surgery contact, this may be attributed to the maintenance phase within the WAP.

To assign a monetary value to the above resource consequences, Table 28 presents the identified units per item of resource use. Table 29 multiplies the quantity of resource use by the unit cost to obtain cost consequences to the NHS.

TABLE 28 - Unit costs

A&E, accident and emergency; HRG, Healthcare Resource Group; PSSRU, Personal Social Services Research Unit.

Item	Unit cost (£)	References	Price year
Interventions
WAP session	10	PSSRU 2013:83 two hospital dietitians (band 5). Attendance: 15 participants per session. Indirect time per session: 3 hours	2012–13
Practice weight loss	44	PSSRU 2013:83 nurse (GP practice), per hour of face-to-face contact,	2012–13
General practice and community nursing services
GP (surgery)	45	PSSRU 2013:83 ‘per-patient contact lasting 11.7 minutes’	2012–13
GP (home)	114	PSSRU 2013:83 ‘per out-of-surgery visit lasting 23.4 minutes’	2012–13
GP (telephone)	27	PSSRU 2013:83 ‘per telephone consultation lasting 7.1 minutes’	2012–13
Nurse (surgery)	13	PSSRU 2013:83 £52 per hour. Face-to-face contact, duration of contact 15.5 minutes	2012–13
Nurse (home)	60	PSSRU 2013:83 community nurse (includes district nursing sister, district nurse). Cost per hour of home visiting	2012–13
Counsellor (surgery)	63	PSSRU 2013:83 counselling services in primary medical care. Cost per hour of client contact	2012–13
Social services
Social worker	159	PSSRU 2013:83 social worker (adult services). Cost per hour of face-to-face contact	2012–13
Home help	24	PSSRU 2013:83 home care worker. Cost per hour (weekday)	2012–13
Care assistant	30	PSSRU 2013:83 senior ‘home care worker’. Cost per hour (weekday)	2012–13
Day centre (visit)	38	PSSRU 2013:83 local authority social services day care for people with mental health problems. Cost per user session	2012–13
Psychiatric services
Psychiatrist (hospital)	261	PSSRU 2013:83 consultant – psychiatric. Cost per hour (face-to-face contact)	2012–13
Psychiatrist (home)	261	PSSRU 2013:83 consultant – psychiatric. Cost per hour (face-to-face contact)	2012–13
Psychologist	134	PSSRU 2013:83 clinical psychologist. Cost per hour of client contact	2012–13
Psychiatric nurse	65	PSSRU 2013:83 nurse (mental health). Cost per hour of face-to-face contact	2012–13
Other services
Hospital (day case)	697	PSSRU 2013:83 Day cases HRG data (weighted average of all stays)	2012–13
Hospital (A&E)	956	PSSRU 2013:83 ‘outpatient and A&E (all service users)’	2012–13
Hospital (outpatient)	135	PSSRU 2013:83 outpatient procedures (weighted average of all outpatient procedures)	2012–13
Hospital (inpatient)	598	PSSRU 2013:83 non-elective inpatient stays (short stays)	2012–13
NHS Direct	52	PSSRU 2013:83 £52 per hour of telephone contact with nurse specialist	2012–13
Ambulance/paramedic	221	National Audit Office 201189 (£213.5 mid-point of cost per incidence in range £176–251). Applied annual rate of 3.5%	2012–13
Prescriptions	44.64	PSSRU 2013:83 prescription costs per consultation (net ingredient cost)	2012–13

TABLE 29 - Cost consequences. Available sample to inform complete-case analysis (intervention, n = 116; control, n = 63)

A&E, accident and emergency.

Type of service	Time point
	Baseline				6 months				12 months
	WAP		Nurse		WAP		Nurse		WAP		Nurse
	Mean (£)	SD (£)	Mean (£)	SD (£)	Mean (£)	SD (£)	Mean (£)	SD (£)	Mean (£)	SD (£)	Mean (£)	SD (£)
General practice and community nursing services
GP (surgery)	121	107	139	148	86	120	84	92	69	90	83	81
GP (home)	1	11	9	72	1	11	0	0	4	30	0	0
GP (telephone)	16	28	29	57	14	31	15	25	10	26	17	35
Nurse (surgery)	15	19	15	19	8	15	10	19	8	14	7	13
Nurse (home)	0	0	7	50	0	0	0	0	0	0	0	1
Counsellor (surgery)	8	56	39	271	6	48	7	39	22	148	12	73
Other practice contacts	1	8	0	0	1	9	0	0	1	7	1	5
Social services
Social worker	3	27	0	0	10	74	0	0	8	55	0	3
Home help	79	755	46	369	0	0	0	0	15	160	0	0
Care assistant	0	0	0	0	43	468	0	0	104	1114	0	4
Day centre (visit)	0	0	0	0	0	0	0	0	0	0	0	0
Other social service contacts	1	8	0	0	1	9	0	0	1	7	1	5
Psychiatric hospital and community services
Psychiatrist (hospital)	9	59	33	263	7	54	33	263	2	24	12	99
Psychiatrist (home)	0	0	0	0	0	0	0	0	0	0	0	0
Psychologist	1	12	21	169	1	12	21	169	1	12	51	270
Psychiatric nurse	1	7	0	0	1	6	0	0	1	6	0	0
Other psychiatric service contacts	2	14	0	0	2	14	0	0	1	8	1	8
Other Services
Hospital (day case)	30	142	653	4225	12	91	33	150	42	211	22	176
Hospital (A&E)	157	471	303	615	99	342	121	500	124	428	61	235
Hospital (outpatient)	64	190	94	230	76	226	56	130	69	182	56	127
Hospital (inpatient)	57	344	66	286	15	124	47	377	10	78	9	75
Other hospital contacts	9	89	0	0	9	89	0	0	1	4	2	12
NHS Direct	1	4	1	7	0	2	1	7	1	12	0	0
Ambulance or paramedic	10	54	21	95	6	35	18	91	15	104	4	28
Prescriptions received (6 months)	134	264	178	300	127	193	101	163	104	162	117	197
Total cost (at time point)	716	1219	1681	4411	523	957	548	933	613	1485	455	576

The main variable to inform the relative cost consequences of intervention and to inform cost-effectiveness analysis is the total cost. First, it should be noted that, at baseline, the mean total cost in the nurse arm (£1681, SD £4411) is effectively double the cost in the WAP arm; hence there may exist significant differences in the overall resource requirement between the two study groups.

To conduct the cost-effectiveness analysis, only cost consequence after intervention (i.e. at the 6- and 12-month follow-ups) would be incorporated within a base-case cost-effectiveness. However, the assumption that retrospective differences in total costs may not be of relevance raises important questions regarding the validity of the base-case estimate on value for money; implications of potential differences (as indicated by baseline total costs) are explored using a sensitivity analysis (see Sensitivity analysis 1).

Outcomes

Health-related quality of life using the EQ-5D-5L is measured to enable the estimation of QALYs. Table 30 presents the EQ-5D-5L mean (SD) utility scores at each time point and the group QALY.

The difference in baseline utility scores between groups is 0.011, indicating that we should control for baseline utility. The unadjusted QALY for the WAP arm is 0.404 (SD 0.079) and in the nurse arm is 0.389 (SD 0.072), indicting that the unadjusted incremental QALY (i.e. the difference between the WAP and nurse arm) is 0.015.

Cost-effectiveness analysis: base-case analysis

To inform the base-case estimation of cost-effectiveness, a seemingly unrelated regression is utilised to jointly explain the change in total cost and QALYs (Table 31). As outlined above, group differences were observed in both group baseline utility and age distributions, and hence it is necessary to control for age and baseline utility in the regression analysis. This provides adjusted estimates of the effect of the WAP on the incremental QALY and incremental total cost.

Adjusting for age, the WAP is associated with a mean incremental total cost of £80 (95% CI –£505 to £667). This suggests no significant difference in cost between the two arms of the trial. The level of significance on the age coefficient confirms the importance to control for group differences in age distribution.

Controlling for age and baseline, intervention with the WAP with an adjusted mean incremental QALY is 0.0104 (95% CI –0.0015 to 0.0224; p = 0.060). The estimated magnitude of the change in QALY indicates that there remains some uncertainty in the health gains from the WAP.

The following formula shows the calculation of the adjusted ICER based on the estimated mean incremental total cost and QALY is:

An ICER of £7742 per QALY would suggest that the WAP would represent a cost-effective intervention with respect to the cost per QALY threshold of £20,000, as used by NICE. However, assuming that a decision-maker may be risk averse, it is also important to consider uncertainty in the ICER.

We can demonstrate the uncertainty surrounding the ICER (and to address skewness resulting from the count nature of cost data) by initially bootstrapping the cost and outcome data using 10,000 bootstrap replications. Figure 8 presents the results of the bootstraps on both a cost-effectiveness plane and cost-effectiveness acceptability curve. Presenting the result of the bootstrap on the cost-effectiveness plane helps to illustrate uncertainty in probabilistic terms as they relate to decisions related to each quadrant. The first observation is the uncertainty in whether the WAP will cost more or less than best usual care, and bootstrap results suggest a 38% probability that the WAP will cost less than usual care, of which 36% also have a positive health gain. Despite the low level of significance surrounding the adjusted mean incremental QALY, the results suggest a 96% probability of a positive health gain from the WAP. However, the most probable scenario is that the WAP will be more effective and will, overall, cost more to the NHS (60%).

FIGURE 8.

Cost-effectiveness plane (a) and cost-effectiveness acceptability curve (b).

To further explore the decision in the context of uncertainty, bootstrap results are utilised to form a cost-effectiveness acceptability curve. This plot demonstrates how the probability that an intervention is cost-effective increases as the decision-makers willingness to pay increases. First, it can be observed from the cost-effectiveness acceptability curve that should a decision-maker not be willing to pay any more than usual care, there is a 0.38 probability that the WAP will be cost-effective. Decision-makers must make decisions in the context of uncertainty. The probabilities that the WAP falls within the NICE willingness-to-pay threshold of £20,000–30,000 are p(ICER < 20,000/QALY) = 0.6826 and p(ICER < 30,000/QALY) = 0.7746.

Given the available information (in addition to base-case required assumptions), the base-case cost-effectiveness analysis suggests that the WAP may represent value for money to the NHS.

To explore the implication of underlying assumptions, the following three sensitivity analyses aim to explore the implications of assumptions of the base-case findings.

Study set-up

Recruitment

Our primary avenue for recruitment was via GP practices. However, posters were displayed in a small range of other community venues and workplaces.

We asked GP practices to query their patient database for potentially eligible participants. Owing to data protection issues, practices had to perform all database queries and printing of letters in-house. The processing of the letters (i.e. folding, stuffing envelopes and franking) took considerable time and effort. Later, practices offered to send text messages instead to their patients who were potentially eligible for participation. This method was significantly easier and less costly, although it relies on people having an up-to-date mobile phone number. We sent out approximately 3800 letters and 6500 text messages.

The most frequently reported route into the study was via GP text and mailshots to potential participants (Figure 9).

FIGURE 9.

Number of volunteers contacting the study by source.

Recruitment barriers and facilitators

It was originally planned to recruit approximately 30 participants per month over a 12-month period starting in October 2011. However because of delays in R&D approvals and contracting, the project plan was revised to start recruitment in July 2012, recruiting 40 participants per month over a 9-month period.

Despite this revised timetable the start of recruitment was delayed until September 2012 and was slower than anticipated (see Chapter 3, Recruitment).

One of the barriers to recruitment was our reliance on the GP team completing referral forms that we asked the practice to fax to us. To try to increase the profile of the study, the research team attended GP team meetings on a fortnightly basis to remind GP staff of how to refer. We had participants from our early waves who wanted to share their success with their GP practice and so attended some of these meetings. However, although staff were interested and hugely supportive, we did not see an increase in GP referrals.

Plans for improving recruitment were discussed during the Trial Management Committee and Trial Steering Committee meetings. The recruitment targets were revised in February 2013 (see Chapter 3, Recruitment), extending the recruitment period to January 2014, adding an additional 10 months to the recruitment period and extending the 12-month follow-up to January 2015. These changes were presented at a Health Technology Assessment monitoring meeting in July 2013 and a 6-month no-cost extension was approved to enable these changes.

The recruitment strategy was redrafted in November 2012 to include mobile phone text mailshots, advertising on websites and boards in practices, producing newsletters on study progress to staff and GP patients, and holding stalls to advertise the study to potential participants.

It was also agreed that participants could be recruited from neighbouring practices. This, however, was not straightforward. Practice managers proved difficult to get hold of and often did not return calls. The process was extremely time-consuming, requiring study staff to explain the study and answer any questions posed. In the end, 15 surgeries were contacted via telephone, letters and e-mails, nine replied expressing an interest and four participated, referring 290 participants to the study. We also needed to find a process by which participants from outside the practice could attend for their weight management appointments without mistakenly being turned away because they were not registered with the practice.

Screening sessions

Our experience in clinical practice with running smoking cessation and weight management clinics is that approximately half of people invited to attend the first session do not attend. We therefore double-booked all appointments for the first session. We did, however, have a lower did-not-attend rate (35%) than expected, which meant that some screening sessions were busy and participants were required to wait for up to 30 minutes longer than usual.

Weight Action Programme groups

Over the course of the study a total of 15 groups were run. The size of the groups ranged from 10 to 21 participants. Table 36 shows the times and days of the week that these groups were run. A greater number of evening clinics (17.30–18.30 hours) were offered, as this tended to be when most people were able to attend. We tried running one clinic between 14.00 and 15.00 hours, but this was not repeated as it clashed with collecting children from school.

Participant attendance at treatment sessions

More than two-thirds of participants in both study arms completed at least half of all treatments sessions (Table 37). Session attendance generally declined over time (Table 38 and Figure 10).

FIGURE 10.

Proportion of participants in the nurse arm attending each session. Sessions 1–4: fortnightly treatment sessions over 8 weeks.

Not everyone who dropped out of the WAP provided a reason for doing so. Among those who did, inability to attend because of inconvenient times was the most common reason (Figure 11). Although participants were informed of the clinic times at the very first contact, it was often because life circumstances had changed (e.g. changes in work rota or child care).

FIGURE 11.

Reasons given for dropping out of the WAP treatment (n = 19).

Participant attendance at maintenance sessions

Attendance at each session is shown in Figure 12. Participation declined over time, with only around one in five participants attending maintenance sessions in the last 6 months. The maintenance sessions were held as ‘open sessions’, with participants at different stages of the intervention attending the same maintenance session.

FIGURE 12.

Proportion of participants in the WAP arm attending each session. Sessions 1–8: weekly treatment sessions; sessions 9–18: monthly maintenance sessions.

Follow-up rates

We implemented a number of strategies for minimising loss to follow-up (Box 9). Follow-up rates were higher than predicted (70% vs. 50% predicted at 12 months). There was no difference in follow-up rate by study site, but the proportion of participants followed up at 12 months was greater in the nurse arm (76%) than in the WAP arm (67%).

Participant adherence to Weight Action Programme tasks

Of all tasks, pedometer use was the most likely to be adhered to, with close to 90% of participants who handed in task cards reporting to use these devices daily. Approximately half of participants weighed themselves regularly (50–57%) and around 60% adhered to the 5-a-day task. Removal of triggers to eat was the least used tool (6%).

Use of orlistat

During the planning stage we were alerted to the fact that people were offered orlistat as part of standard care. We therefore included information about orlistat at the third WAP session. All participants were given an information sheet about orlistat. Among those who expressed an interest in using orlistat, eligibility was checked by study staff, to prevent participants making an unnecessary visit to their GP. Those who were eligible were advised to make an appointment with their GP to obtain a prescription and bring the medication back to the session the following week, during which participants were given a recap on how to use orlistat.

All eligible participants were prescribed a 1-month course of orlistat in the first instance, with continued prescriptions contingent on weight loss, as per NICE guidelines. At subsequent sessions, participants were asked about their orlistat usage.

At total of 75 (23%) participants opted to use orlistat as part of their weight loss attempt. Participants in the WAP arm were significantly more likely to use orlistat than participants in the nurse arm (31% vs. 6%; odds ratio 6.50, 95% CI 2.78 to 15.59; p < 0.001). Weight loss at 12 months was greater in those who used orlistat (mean –5.4 kg, SD 8.1 kg) than in those who did not (mean –2.9 kg, SD 6.6 kg), with the difference (–2.5 kg) being statistically significant (95% CI –4.5 to –0.4 kg; p = 0.02).

Participant feedback

Participants in both arms provided feedback on the helpfulness of the weight loss intervention they received at the end of treatment (Table 39) and at the 12-month follow-up (Table 40). Ratings of helpfulness in losing weight were high in both arms, but significantly greater from participants who used the WAP (9.1 vs. 8.0; p < 0.001). The WAP participants were also more likely to recommend the programme to others (9.3 vs. 8.1; p < 0.001). Ratings were only slightly lower at the 12-month follow-up, but remained significantly greater in the WAP arm (see Table 40).

Participants also ranked their top three most useful aspects from each of the treatment programmes (nurse and WAP). Few participants (n = 5) in the nurse arm responded, but at the end of treatment all highly ranked advice from the nurse.

Of the respondents in the WAP arm, the component most frequently ranked in the ‘top 3′ at the end of treatment was ‘monitoring with a pedometer’ (30%; Figure 13). This was followed by ‘having weight regularly monitored’ (14%) and ‘coming to group sessions’ (13%). ‘Avoiding temptation’, ‘leaflets provided’, ‘exercise programmes’ and ‘buddy system’ were not seen as particularly helpful.

FIGURE 13.

Percentage of participants rating each component of the WAP treatment in their top three most useful.

In their feedback, some participants provided written comments about what they liked about the WAP. These comments were collapsed into three themes.

Metabolic-equivalent minutes/week domain

This score represents the total MET-minutes/week, and is expressed on a continuous scale with a minimum score of 0. It is calculated as:

Sitting domain

This score represents the number of minutes per day spent sitting. It is calculated directly from question 4.