Adding web-based behavioural support to exercise referral schemes for inactive adults with chronic health conditions: the e-coachER RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 13/25/20. The contractual start date was in January 2015. The draft report began editorial review in January 2019 and was accepted for publication in November 2019. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Taylor et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2020 Queen’s Printer and Controller of HMSO

Scientific background

As described elsewhere,1 evidence-based guidelines recommend both aerobic training and strength training for improving health markers and quality of life among those people with common chronic metabolic conditions1–5 and musculoskeletal conditions. 6 To prevent or reduce depression, mostly aerobic exercise is recommended. 7 Public health guidelines of 150 minutes of moderate activity per week (accumulated in bouts of ≥ 10 minutes) or 75 minutes of vigorous activity per week are met by only 66% of men and 58% of women aged ≥ 19 years in England,8 which is similar to results from the Scottish Health Survey 20149 (68% men; 59% women), based on self-report data from a national representative survey. The guidelines also highlight the importance of reducing sedentary behaviour and regularly doing bouts of resistance exercise. Physical inactivity, based on data from 2013–14, collected by Clinical Commissioning Groups in the UK, costs the NHS £455M. 10

Even small increases in physical activity (PA) and reduced sedentary time, especially among the least active,10 are likely to provide health benefits. 11,12 Patients with obesity, hypertension, type 2 diabetes, osteoarthritis and depression do less PA than the general population2 and report greater barriers to increased PA.

Exercise referral schemes in the UK

A variety of approaches have been explored to promote PA within primary care, such as referring patients to ‘exercise on prescription’ [i.e. an exercise referral scheme (ERS)]. As described previously,1 in the UK, ERSs have been popular, with an estimated 600 schemes involving up to 100,000 patients per year in 2008. 13 There is currently no single model for ERSs in the UK, but they mainly involve referral to a programme (e.g. 10–12 weeks) of structured, supervised exercise at an exercise facility (e.g. gym or leisure centre) or a counselling (and signposting) approach to support patients to engage in a variety of types of PA. 13 ERSs operate diversely to accommodate patient choice and local availability of facilities, the common goal being to reduce the risk of long-term metabolic, musculoskeletal and mental health conditions due to physical inactivity.

Evidence from a meta-analysis of eight randomised controlled trials (RCTs) involving 5190 participants eligible for ERSs14 indicated only a small increase in the proportion of participants who achieved 90–150 minutes of PA of at least moderate intensity per week, compared with no exercise control, at the 6- to 12-month follow-up among at-risk individuals. However, uncertainty remains regarding the effects for patients with specific medical conditions, no study assessed long-term PA objectively and many of the eight studies reviewed had relatively small sample sizes.

A review15 reported that the average ERS uptake (attendance at the first ERS session) ranged from 66% in observational studies to 81% in RCTs, and average levels of ‘adherence’ ranged from 49% in observational studies to 43% in RCTs. Predictors of uptake and adherence have been explored; women were more likely than men to begin an ERS, but less likely to adhere to it, and older people were more likely to begin and adhere to an ERS. 15 As an example of a large observational retrospective study,16 of 6894 participants who had attended an ERS over 6 years, 37.8% (n = 2608) dropped out within 6 weeks and 50.03% (n = 3449) dropped out by the (final) 12th week, and males (p < 0.001) and older people (p < 0.001) were more likely to adhere than females and younger people, respectively. ERSs may help patients to become familiar with medical conditions17 and target key processes of behaviour change. However, the following features of an ERS may reduce uptake and adherence: inconvenience, cost, limited sustainable PA support (e.g. for 10 weeks) and low appeal for structured exercise and/or the medical model (i.e. ‘exercise on prescription’), which may in some schemes do little to provide autonomous support or empower patients to develop self-determined behaviour to manage chronic medical conditions. 13,18,19

It therefore appears that additional support may be needed that is accessible and is low cost, can be tailored to support a wide range of individual needs and empowers patients to develop and use self-regulatory skills (e.g. self-monitoring, goal-setting) to self-manage their chronic conditions. In one study, training for ERS staff to foster self-determined behaviour increased PA more than when the ERS was delivered by untrained staff at 3 and 6 months. 20 Similarly, training of ERS staff in behaviour change techniques and motivational interviewing led to small additional changes in self-reported PA after 12 months17 compared with no ERS. Challenges in training staff across a wide geographical area across Wales and monitoring intervention fidelity were noted.

Intervention technologies to promote physical activity

To address the challenges with face-to-face promotion of PA noted above and to encompass the growing use and availability of new technologies, a wide variety of online and mobile support has been developed and used to promote PA.

As described previously,1 there is growing evidence on the effectiveness of technology-based interventions for promotion of PA. 21,22 Studies include a wide range of interventions (from quite simple self-monitoring to interventions with multiple complex behaviour change components), targeted at different clinical groups with different baseline levels of PA, with various PA outcomes reported (very few using objective measures), and with mostly short-term follow-ups. In addition, some comparisons are between intervention versus no intervention and others are versus human contact, although none reports on the effects of adding web-based support to complement face-to-face support provided by ERSs. The impact of web-based and technology interventions on increasing PA is small to moderate (effect size ≤ 0.4). However, there is evidence from more rigorous studies that interventions with more behaviour change components and ones targeting less active populations are more effective. 21,22 A systematic review23 highlighted the importance of maximising sustained engagement in web-based interventions for enhancing change in the target behaviour. A recent study24 highlighted that self-monitoring of PA and tailored feedback were important to increase engagement, and periodic communications helped to maintain participant engagement.

The LifeGuide© (version 1.0.7.30, University of Southampton, Southampton, UK; www.LifeGuideonline.org/; accessed 25 February 2020) platform has been extensively used to develop and evaluate the acceptability and impact of online behaviour change and self-management interventions with a variety of clinical groups, including in primary care. 25–27 As an example, when adding online LifeGuide support to face-to-face support there was a greater lasting reduction in obesity than with face-to-face dietetic advice alone. 28 The LifeGuide platform provides a researcher-led tool to develop theory-driven interventions and evidence of the effectiveness of techniques. 29,30 It also provides the opportunity to capture intervention engagement and assess the utility of different behaviour change components.

The potential for e-coachER

Following iterative development work and user group testing and involvement, drawing on some online modules used in other LifeGuide interventions, for example in secondary prevention of coronary heart disease,25 we developed a bespoke intervention. We called the intervention ‘e-coachER’ and it was designed to support patients with chronic physical and mental health conditions who have been referred from primary care to an ERS to receive face-to-face support. 1 The overarching aim of the e-coachER intervention was to facilitate long-term PA by promotion of evidence-based self-regulatory skills and to encourage interaction with others (including the ERS professional, family and friends), and founded on self-determination theory31 to build a sense of competence in managing PA, autonomy or control over PA, and connection or relatedness with others. We wanted to encourage uptake and adherence to the ERS but if that was not acceptable or feasible for participants then we offered support to find alternative ways to be physically active in a way that may support their needs as someone who was inactive. It was also important that the intervention could be scaled up to promote PA for patients with obesity, hypertension, type 2 diabetes, osteoarthritis and risk of depression at probably low cost32,33 and also potentially make it available for patients with other chronic medical conditions (e.g. low back pain, heart disease and cancer).

Developing the e-coachER intervention

The intervention development included the piloting of the welcome pack and development of an initial version of e-coachER, built on wide-ranging experiences from the development of other self-management interventions using the LifeGuide platform34 and beta testing over 7 months. Co-applicants and researchers then provided feedback on a time-truncated version, and ERS users provided feedback on a real-time version for 5 months before the website was locked for the RCT.

The development of all components of the e-coachER intervention followed a logic model as shown in Figure 1. 1

FIGURE 1.

Logic model for the e-coachER intervention. BCT, behaviour change technique; MVPA, moderate and vigorous physical activity. Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text.

Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text.

The key components of e-coachER include the following:

• On allocation to the intervention, all participants received a ‘welcome pack’ (Figure 2) that contained a user guide and the participant’s unique user log-in and registration details to access the e-coachER website, a simple pedometer (step counter) with instruction sheet and a fridge magnet with tear-off sheets to record daily PA (complete with trial branding and e-coachER helpline number). Participants were encouraged to use the pedometer and the PA record sheets for self-monitoring and goal-setting in conjunction with the website.

• The e-coachER support system, hosted on the LifeGuide platform, provided support through seven ‘steps to health’ lasting approximately 5–10 minutes each, as shown in Table 1.

• The steps were designed to do the following: encourage participants to think about the benefits of PA (motivation); seek support from an ERS practitioner, friends/family and the internet (support/relatedness); set progressive goals; self-monitor PA with a pedometer and upload step counts or minutes of moderate and vigorous physical activity (MVPA) (self-regulation, building confidence/autonomy); and find ways to increase PA more sustainably in the context of day-to-day life and deal with setbacks (building confidence). The sequential content, objectives and how this was implemented were mapped against a taxonomy for behaviour change techniques shown in Table 1. 35 The website content is illustrated in Appendix 1.

• Participants were encouraged to use e-coachER support as an interactive tool by using pre-set or personally set reminders to upload step counts or minutes of MVPA, and messages of encouragement. A lack of engagement (e.g. not reviewing a goal by entering step counts 1 week later, or not signing into the website for 1, 2 and 4 weeks) triggered reminder e-mails. Participants were reminded by prompts to review goals the next day.

• An avatar was used throughout the content to avoid having to represent a range of individual characteristics, such as age, gender and ethnicity. The avatar delivered brief narratives to normalise and support behaviour change and encourage the use of e-coachER. Automatic and user-defined e-mails generated by the LifeGuide system promoted ongoing use of functions such as recording weekly PA and goal-setting. Participants were provided with links to reputable generic websites for further information about the chronic conditions of interest and lifestyle, links to other websites and apps (applications) for self-monitoring health behaviour and health, as well as modifiable listings of local opportunities to engage in PA.

• The only webpages that were not ‘locked’ after the initial participant began the intervention were pages for the respective recruitment sites that displayed links to the following websites: (1) local community opportunities for engaging in PA, (2) disease-specific (e.g. Diabetes UK, Royal College of Psychiatrists) informational sites about the role of PA in preventing and managing the condition, and (3) sites about other methods to optimise ways to be physically active (e.g. more sophisticated technologies to self-monitor).

• Our aim was to maximise accessibility and use. Therefore, a local researcher provided technical support if requested, but did not support behaviour change directly. If participants did not register on the e-coachER website within the first few weeks, they were followed up with a telephone call to offer support to use e-coachER. If technical support to resolve any user operational issues with the website (e.g. re-issuing passwords) was needed, participants were referred to a centralised technician within the LifeGuide team for further support.

Trial aim and objectives

The overarching aim was to determine if e-coachER online support combined with usual ERS provided a clinically effective and cost-effective approach to supporting increases in PA in people referred to an ERS with a range of chronic conditions.

The specific objectives were to:

• determine whether or not there is an increase in the total weekly minutes of MVPA at 12 months post randomisation in the intervention group compared with the control group

• determine whether or not there is an increase in the proportion of participants in the intervention group compared with the control group who:

  • take up the opportunity to attend an initial consultation with an exercise practitioner

  • maintain objectively assessed PA at 4 and 12 months post randomisation

  • maintain self-reported PA at 4 and 12 months post randomisation

  • have improved health-related quality of life at 4 and 12 months post randomisation

• quantify the additional costs of delivering the intervention, and determine the differences in health utilisation and costs between the intervention and control groups at 12 months post randomisation

• assess the cost-effectiveness of the intervention compared with control at 12 months post randomisation [incremental cost per quality-adjusted life-year (QALY)] using a previously developed decision model to estimate future costs and benefits

• quantitatively and qualitatively explore whether or not the impact of the intervention is moderated by medical condition, age, gender, socioeconomic status, information technology (IT) literacy or ERS characteristics

• quantitatively and qualitatively explore the mechanisms through which the intervention may have an impact on the outcomes, through rigorous mixed-methods process evaluation and mediation analyses (if appropriate).

Study design

This was a multicentre, parallel, two-arm RCT with participant allocation to usual ERS alone (control) or usual ERS plus web-based behavioural support (intervention) with parallel economic and mixed-methods process evaluations. 1 The trial design is summarised in Figure 3, from Ingram et al. 1

FIGURE 3.

Participant pathway. Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text.

Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text.

Recruitment to the trial took place over a 21-month period (July 2015 to March 2017) in three areas in the UK: Greater Glasgow, West Midlands and South West England (including Plymouth, Cornwall and Mid Devon). The majority of participants lived in urban locations. Further information about the characteristics of the cities involved and the respective ERSs to which participants were referred is given in Ingram et al. 1

Ethics approval and research governance

Ethics approval for the study was granted by North West Preston NHS Research Ethics Committee (REC) in May 2015 (reference 15/NW/0347). Approval for activity at non-NHS sites was obtained from the same REC for the following ERSs: Everyone Active (Plymouth), Teignbridge District Council (Cornwall), Tempus Leisure (Cornwall), Be Active Plus (West Midlands) and Live Active (Glasgow) in December 2015, and Docspot (West Midlands) in November 2016.

NHS Research and Development approval was granted by the Royal Devon and Exeter NHS Foundation Trust Shared Management Service for the Plymouth site (July 2015); NIHR Clinical Research Network (CRN) West Midlands for the Birmingham site (July 2015); and the NHS Greater Glasgow and Clyde Health Board for the Glasgow site (January 2016).

Prior to commencing recruitment, the trial was registered with the International Standard Randomised Controlled Trial Number Register (ISRCTN) under reference number 16900744 and NIHR CRN Portfolio 19047. A summary of the changes made to the original protocol is given later in this chapter.

Patient and public involvement

Participants

The study population was composed of patients who had been referred to an ERS administrator, or were about to be referred, mostly by a general practitioner (GP), and occasionally by a nurse to a local ERS for a programme of support to increase PA.

Patients were eligible for the study if they were aged between 16 and 74 years, inclusive, were contactable via e-mail, had some experience of using the internet and had one or more of the following conditions:

• obesity [i.e. a body mass index (BMI) of 30–40 kg/m²]

• a diagnosis of hypertension

• prediabetes

• type 2 diabetes

• lower limb osteoarthritis

• current or recent history of treatment for depression

• categorised as ‘inactive’ (i.e. 0 hours per week of physical exercise and in a sedentary occupation) or ‘moderately inactive’ (i.e. some activity but < 1 hour per week and in a sedentary occupation, or 0 hours per week of physical exercise and in a standing occupation) according to the General Practice Physical Activity Questionnaire (GPPAQ). 36

Patients were excluded for the following reasons:

• They did not meet the eligibility criteria for their local ERS.

• They had an unstable, severe and enduring mental health problem.

• They were being treated for an alcohol or drug addiction that may have limited their involvement with the study.

• They were unable to use written materials in English, unless they had a designated family member or friend to act as translator.

Intervention

All participants had been offered usual ERS. Participants allocated to the intervention group were additionally offered access the e-coachER web-based support package. Development and delivery of the intervention is fully described in Chapter 1, Developing the e-coachER intervention.

Usual care

There is currently no single model for ERSs in the UK, but the predominant mode of delivery involves referral to a programme (e.g. 10–12 weeks) of structured, supervised exercise at an exercise facility (e.g. gym or leisure centre) or a counselling approach to support patients to engage in a variety of types of PA. 13 ERSs operate diversely to accommodate patient choice and local availability of facilities, the common goal being to reduce the risk of long-term metabolic, musculoskeletal and mental health conditions as a result of physical inactivity. The three participating sites were selected from different regions of the UK (different ERS providers) to provide diversity of approach; the schemes are described in Ingram et al. 1

Recruitment procedures

Patients were identified as potentially eligible for the trial in a number of ways.

1. By health-care professionals in primary care at the point of being actively referred to an ERS or having been opportunistically found to be eligible for an ERS at a consultation with the primary care practitioner.

2. Via a search of patient databases at the participating GP practices (conducted by the local NIHR Primary Care Research Network team).

3. Via patient self-referral to the GP arising from community-based publicity for the trial.

4. By the ERS programme administrator on receipt of an ERS referral form from a GP practice.

5. By exercise advisors at the ERS service at enrolment on the ERS. With the patient’s consent, the exercise advisor provided the local researcher with the patient’s contact details for the purposes of the trial.

Potentially eligible patients were approached by the primary care practitioner or the local researcher, depending on how the patient had been identified. Some patients self-referred to the local researcher in response to publicity campaigns. These various means of identification and approach were designed to accommodate the variation in usual-care referral pathways to ERSs across the participating sites and individual GP practices.

Amenable patients were offered a study-specific participant information sheet (see Report Supplementary Material 1) by post, by e-mail or by hand (the route used largely depended on the preference of the participating GP practice or ERS service). Interested patients were asked to communicate their expression of interest to the local researcher via a prepaid reply slip, by telephone or by e-mail. On receipt of an expression of interest, the local researcher contacted the potential participant by telephone to discuss the trial, confirm eligibility and seek informed consent.

Informed consent

Informed, written consent (see Report Supplementary Material 2) was obtained prior to undertaking the baseline assessment, either over the telephone or at a face-to-face visit, depending largely on the patient’s preference but also on the availability of suitable venues, such as the GP practice. The original completed informed consent forms were held securely at the site and a copy was given to the participant.

In the case of telephone consent, the researcher was required to sign and date the informed consent form, but the participant was not required to sign or date their copy.

Patients who were deemed to be ineligible for inclusion in the study were informed of this outcome.

Randomisation, concealment and blinding

On receipt of the baseline accelerometer at the Peninsula Clinical Trials Unit (PenCTU) (after 1 week of wear), participants were randomised to usual ERS or usual ERS plus access to e-coachER in a 1 : 1 ratio, stratified by site with minimisation by participant’s perception of main medical referral reason (i.e. weight loss, diabetes control, reduce blood pressure, manage lower limb osteoarthritis symptoms, manage low mood/depression) and by self-reported IT literacy level on a visual analogue scale (i.e. lower or higher confidence).

To maintain concealment, the minimisation algorithm retained a stochastic element.

Randomisation was conducted by means of a secure, password-protected web-based system created and managed by the clinical trials unit (CTU).

Blinding of participants was not possible, given the nature of the intervention. Given that the primary outcome was an objective measure of PA recorded by accelerometer, and the secondary outcomes were assessed by a participant self-completed questionnaire, the risk of assessor bias was considered to be negligible in this study. However, to minimise any potential bias, the statistical analysis was kept blinded and the code for group allocation was not broken until the primary and secondary analyses had been completed. The ERS practitioners would not have been aware of trial participants’ treatment allocations.

Process evaluation: qualitative interviews

Participants who had engaged with the e-coachER website (i.e. logged on to the website, as a minimum) were invited to participate in one or more qualitative interviews to inform the process evaluation (see Chapter 4, Qualitative process evaluation).

Data collection and management

Data were collected and stored in accordance with the Data Protection Act 199837/General Data Protection Regulation 2016. 38

Participant numbering

Following receipt of an expression of interest, each patient was allocated a unique identification number and was then identified in all study-related documentation by this number and their initials. A record of names, addresses, telephone numbers and e-mail addresses linked to participants’ identification numbers was stored securely on the study database at the CTU for administrative purposes only.

Data collection

Data were recorded on study-specific paper-based case report forms (CRFs) by the local researcher, and participants completed a paper-based questionnaire booklet comprising validated and non-validated self-report outcome measures.

Accelerometers [GENEActiv™ Original accelerometer (version 3.0_09.02.2015), Active Insights Kimbolton, UK; www.geneactiv.org/ (accessed 26 February 2020)] were configured for use prior to being issued to participants by the local researcher at baseline and the CTU thereafter, using GENEActiv software. A recording window of 10 days, starting at midnight of the day of issue and recording at 75 Hz, was pre-set, thus accounting for transits in the post while optimising the battery life of the device.

Accelerometers received by the CTU following 1 week of wear by the participant were physically cleaned with liquid detergent in accordance with the manufacturer’s instructions before data were downloaded and linked to the participant identification number (see Accelerometer data processing). Accelerometers were then issued to other participants in the trial as required.

Data on participants’ uptake of the ERS were collected via participant self-report at 4 weeks post baseline and 4 months post randomisation, and via the ERS service provider.

Recording and reporting of non-serious adverse events (AEs) in this study were not required. Serious adverse events (SAEs) were reported to the CTU via the self-report questionnaire booklet administered at 4 and 12 months, but were also reported to the CTU or local researcher by the participant (or relative) outside these time points, until the end of follow-up. SAEs were categorised using the Medical Dictionary for Regulatory Activities (MedDRA) Terminology Systems Organ Classification List Internationally Agreed Order Version 19.0, March 2016 (www.meddra.org/sites/default/files/guidance/file/intguide_19_0_english.pdf; accessed 26 February 2020).

All persons authorised to collect and record study data at each site were listed on the study site delegation logs, which were signed by the principal investigator.

Data processing

Baseline assessment

Consented participants attended a baseline assessment with the local researcher. This assessment was conducted over the telephone or in person at a suitable venue in the community.

Demographic data were collected (i.e. age, gender, BMI, blood pressure, ethnic group, relationship status, domestic residence status, smoking status, employment status, education status, GPPAQ score, internet use capability, requirement for translator for trial purposes and clinical condition).

Information technology literacy level was determined using a visual analogue scale for self-reported ‘confidence using the internet’, where 1 = not at all confident and 10 = totally confident. Arbitrarily, scores of 1–5 were set to indicate a low literacy level and scores of 6–10 indicated a high literacy level, for the purposes of stratification.

‘Clinical condition’ was the participant’s perception of the reason for referral to the ERS; where more than one medical condition was stated, the participant ranked these in order of importance and the most important reason was used as a stratification variable.

The participant was issued with the wrist-worn waterproof accelerometer to wear constantly for 1 whole week (day and night), and a self-report questionnaire booklet to complete at the beginning of the week-long period. At the face-to-face baseline assessments, the accelerometer was fitted by the local researcher; at telephone visits, the accelerometer (and self-report questionnaire booklet) were posted to participants. All participants were provided with a bespoke guidance sheet for wearing the accelerometer (see Report Supplementary Material 3), including instructions to wear the accelerometer on the wrist of the non-dominant hand (i.e. the hand not favoured for writing). After 1 week of wear, participants were required to post the accelerometer and completed questionnaire to the CTU in a preaddressed padded envelope provided. A prepaid postal service was used so as not to incur costs to the participant.

The measures collected at baseline are summarised in Table 2.

Follow-up assessments

The measures collected at follow-up are summarised in Table 2.

At 4 weeks post baseline, a short survey on initial uptake of the ERS was administered via e-mail.

At 4 and 12 months post randomisation, participants were sent an accelerometer (along with the guidance sheet on how to wear it), a questionnaire booklet, and a pre-addressed, prepaid envelope to return the items to the CTU.

To maximise data completeness at follow-up assessments, participants were sent standard letters from the CTU: (1) 7 days before delivery of the accelerometer, (2) 3 days into the 10-day recording window as a prompt to begin wearing the accelerometer (if not already doing so) and (3) at 3 and 5 weeks after issue as a reminder to post the accelerometer to the CTU (for those participants who had not already done so). If a participant had not sent the accelerometer back to the CTU after 6 weeks, the local researcher telephoned the participant to remind them to return the device. Participants who returned the accelerometer to the CTU were provided with a £20 voucher for an online store as a token ‘thank you’ to maximise response rates.

Once a participant’s involvement in the ERS had concluded, the ERS service providers completed a simple pro forma to confirm whether or not the participant attended an appointment with the exercise specialist and, if so, how many appointments were available to the participant.

Measures

Sample size

In the absence of a published minimally important difference for MVPA, assuming a ‘small’ to ‘moderate’ standardised effect size of 0.35, we estimated that 413 participants were required, with 88% power and a two-sided alpha of 5%, assuming 20% attrition, or 90% power at a two-sided alpha of 5% allowing for 16% attrition [using ‘sampsi’ in Stata^® version 14 (StataCorp LP, College Station, TX, USA)]. Given that the intervention was delivered at the level of the individual participant, clustering was not factored into the sample size calculation. Based on the baseline SD for MVPA total weekly minutes in ≥ 10-minute bouts of 104–113,50 an effect size of 0.35 corresponds to a between-group difference of 36–39 minutes of MVPA per week.

Statistical analysis

Analyses were in accordance with International Conference on Harmonisation guideline 9 (ICH-9) statistical guidelines51 for clinical trials and updated Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines52,53 for non-drug trials. All primary and secondary analyses were undertaken and reported in accordance with a prespecified detailed statistical analysis plan that was agreed with the Project Management Group, TSC and Data Monitoring Committee (DMC).

Following data lock by the CTU data manager, the statistician undertook primary analyses blinded to group (i.e. randomised groups were coded ‘A’ or ‘B’). Following the blinded presentation of the primary results to the Project Management Group and agreed interpretation of the results, the groups were unblinded.

Changes to the project protocol

Participant recruitment

A total of 450 participants were recruited (randomised) to the trial over a 20-month period (3 September 2016 to 10 April 2017).

Table 4 shows the number (%) of participants who entered the trial by referral source (i.e. by mailout from the GP or opportunistically in primary care, at the point of initial contact with the ERS providers, or by word of mouth or community advertisement) across the different sites.

Flow of participants in the trial

Figure 4 shows the flow of participants through the trial. Of those expressing an interest in participating, 477 (63%) individuals were eligible. The main reasons for ineligibility were BMI being too high (n = 104, 14%), being too active according to the GPPAQ (n = 47, 6%) and clinical condition of interest not being present (n = 26, 3%). An additional 27 individuals (4%) could not be contacted following their expression of interest. A detailed CONSORT flow diagram is given in Appendix 2.

FIGURE 4.

Participant flow.

Baseline comparability

The baseline characteristics of the whole sample (n = 450) and those who were included in the primary analysis (n = 232) are shown in Tables 5 and 6, respectively. The groups were well balanced.

Study attrition

Accelerometer and questionnaire booklet return rates

Accelerometer and questionnaire booklet return rates are given in Table 7. Receipt of the baseline accelerometer at the CTU was a prerequisite for randomisation; therefore, there is a 100% return rate for this accelerometer.

At 12 months, 329 participants returned the accelerometer (return rate of 92%). The wear-time criteria were met by 243 participants, this being 54% of those randomised (Table 8). At this time point, 325 participants returned the questionnaire booklet (return rate of 91%), that is 72% of those randomised.

In Table 8, purple shading denotes the number of participants meeting the wear-time criteria. A day was ‘valid’ when the accelerometer was worn for ≥ 16 hours on that day. Participants who did not meet the ‘valid’ criterion (n = 181) were 94 participants lost to follow-up prior to the 12-month time point, 60 participants who returned an accelerometer at 12 months but failed to meet the wear-time criteria and 27 participants who remained in follow-up but did not return the accelerometer at 12 months.

Losses to follow-up and non-compliance in returning the accelerometer were balanced across the two trial groups. Failure to meet the wear-time threshold was less consistent across the two trial groups.

Intervention engagement

Table 9 shows the level of engagement in the various steps offered online.

The sample was evenly split, with 36% of participants not registering and logging in to e-coachER and 36% progressing through the support to record at least one goal review (i.e. having set a PA goal review, the participant logged back in about 1 week later to record PA against the goal and obtain feedback on the PA achieved against the goal set). Participants were routinely ‘locked out’ of accessing the web-based support to ensure that they did not complete it in one or two visits to the website, and then reminded by e-mail after 1 week to log in and continue with the steps and later record their PA in minutes, set goals and review them. Reaching step 5 involved > 4 weeks of intervention engagement.

Among all participants allocated to the intervention group, the mean number of goal reviews was 2.5 (SD 4.5), with a range of 0–52. Among the 144 participants who registered for e-coachER, they logged in for a mean and median number of times of 14.1 (SD 16.7) and 6, respectively, with a range of 1–101. Of these participants, 81 (36%) completed a goal review; the mean and median number of reviews was 14.4 (SD 13.8) and 4.5, respectively, with a range of 1–52.

Table 10 shows the mean, SD and median time spent during the respective stage (session) for the 144 participants who registered or 81 participants who completed at least one goal review. These data come with the limitation that participants may have left their browser open after some sessions rather than logging off, which leads to an overestimation of time spent. ‘n’ refers to the number of visits used to estimate the descriptive data for the time in sessions 1–5, the number of visits when a goal was first set and the number of sessions when a goal was reviewed.

On the basis that participants spent approximately 6 minutes logged in for steps (sessions) 1–5, and for 3 minutes for each goal review, for those 144 participants who registered, the total mean and median time that participants spent on steps 1–5 was 24.1 (SD 5.9) minutes and 30 minutes respectively.

For those 81 participants who completed at least one goal review, the overall mean and median time that participants spent doing goal reviews was 43.3 (SD 37.3) minutes and 21 minutes respectively. The 144 participants who registered spent a total mean and median time of 48.4 (SD 41.9) minutes and 36 minutes respectively. The range was 6–186 minutes.

Descriptive data for the primary and secondary outcomes by group and time

The descriptive statistics for the primary and secondary outcomes at baseline and 4- and 12-month follow-up are shown in Table 11 for all participants who provided data.

The groups were well balanced at baseline. Only 4% of participants achieved at least 150 minutes of accelerometer-recorded MVPA (in ≥ 10-minute bouts) over 1 week at baseline and the average weekly minutes of MVPA (not in 10-minute bouts) was 49 minutes, which reflects our success in recruiting inactive or moderately inactive participants with chronic conditions. In contrast, 80% achieved 150 minutes of accelerometer-recorded MVPA without regard for ≥ 10-minute bouts at baseline. Cassidy et al. 55 have also shown lower levels of accelerometer-recorded MVPA minutes when data are processed using ≥ 10-minute bouts compared with bouts of at least 1 minute. This proportion drops to 36% for self-reported MVPA, reflecting the way that the 7-day Physical Activity Recall questionnaire (7-D PAR) measure focuses on only discrete bouts of memorable MVPA. There were also no baseline differences between groups for the EQ-5D-5L and the two HADS scales.

Descriptive data for PA and participant-reported outcomes are also shown in Table 11 for those providing data at baseline, 4 and 12 months by trial arm, without controlling for baseline or other covariates. Qualitatively, the intervention group was more active than the control group at 4 and 12 months for all primary and secondary outcomes. The intervention group also qualitatively had higher well-being (EQ-5D-5L) and lower depression and anxiety scores than the control group at 4 and 12 months.

Primary outcomes

The primary outcome, ITT complete-case analysis showed a weak indicative effect in favour of the intervention group in the primary outcome at 12 months (mean difference 11.8 weekly minutes of MVPA, 95% CI –2.1 to 26.0 minutes; p = 0.10) (Table 12). A plot of the repeated-measures model estimates for the primary outcome over time for the intervention and control groups is shown in Appendix 3. Although the alternative model p-values varied somewhat, a similar pattern of results was seen across alternative post hoc models. In interpreting these results, it is important to recognise the limitations of all these models: lack of fit of the predefined primary and post hoc models, and the need to assume data as counts with both negative binomial and zero-inflated binomial models.

The results of the CACE analysis for the primary outcome were consistent with the ITT results (Table 13). In other words, when controlling for whether or not intervention participants completed a prespecified ‘dose’ of the intervention [i.e. they completed a goal review (reached step 5)], this made no difference to the findings, but qualitatively the difference between the intervention and control groups did appear to be larger (in favour of the intervention).

As Table 14 shows, there was no evidence of any interactions between stratification variables and age and gender with the intervention effect for the primary outcome at 12 months. The ITT complete-case model was adjusted for stratification variables age, gender and baseline scores, and random effects for site and fulfil the criteria of includable PA data.

Table 15 shows the descriptive data for participants included in the primary analysis (n = 232). Data shown for the 4-month assessment are from participants who were included in the primary analysis and also had complete data at 4 months. The intervention group had qualitatively greater mean weekly minutes of MVPA than the control group at baseline.

An analysis of complete data available at both baseline and 4 months (Table 16) shows that the control group significantly increased MVPA up to 4 months, whereas the intervention group did not. Between baseline and 12 months, there were no changes in either group, although qualitatively the control group had reduced their MVPA.

Secondary outcomes

Results of the complete-case imputed ITT and repeated-measures analyses (including both 4- and 12-month follow-ups) for the PA secondary outcomes appeared consistent with the analysis for the primary outcome (see Table 12). As Table 17 shows, there were no significant between-group differences in ITT complete-case analyses for any of the secondary outcomes at 12 months.

The ITT imputed comparison at 12 months showed that the intervention group had a greater proportion of participants achieving ≥ 150 minutes of self-reported MVPA (p = 0.05) than the control group. In the repeated-measures analyses (including both 4- and 12-month follow-ups), a greater proportion of intervention participants achieved ≥ 150 minutes of accelerometer-recorded MVPA (in ≥ 10-minute bouts) than control group participants. As Table 11 shows, 2% and 6% of the participants achieved this at 4 months in the control and intervention groups, respectively, and 2% and 5% at 12 months, respectively.

There were no significant between-group differences in the ITT complete-case or imputed comparison analyses for EQ-5D-5L scores or for the HADS anxiety and depression scores at 12 months. In complete-case repeated-measures analyses (including both 4- and 12-month follow-ups), the intervention participants reported lower depression (p < 0.05) and anxiety (p = 0.05) scores than the control group.

Exercise referral scheme uptake was derived from records held by the ERS service provider, with imputed participant-reported attendance at 4 weeks and/or 4 months where the ERS service data were missing. Data were not available via any of the three sources for four participants (control group, n = 3; intervention group, n = 1), resulting in 223 participants in each group with ERS uptake data. A total of 173 participants (78%) in the control group attended the ERS at least once, compared with 167 participants (75%) in the intervention group.

Adverse events

In total, 42 SAEs were reported among 35 participants (see Appendix 4), which were all deemed to be either ‘not related’ or ‘unlikely to be related’ to the trial. In the control group there were 26 SAEs among 21 participants and in the intervention group there were 16 SAEs among 14 participants. One SAE was reported as a life-threatening event (asthma attack) and all other SAEs were hospitalisations. SAEs were consistent with the patient population.

Introduction

The overarching aim of the e-coachER trial was to determine whether or not adding support to usual ERS would be more clinically effective and cost-effective than usual ERS alone for supporting increases in PA in inactive patients referred to an ERS with a range of chronic conditions. The qualitative part of this chapter will explore participants’ and e-coachER researchers’ views and experiences of the support package and how it did or did not contribute to changes in PA. The quantitative part of the chapter will seek to understand if specific survey process measures were changed by the intervention, compared with usual ERS, as predicted from our logic model (Figure 5). Finally, we explore whether or not changes in the process outcomes mediated intervention effects on the primary outcome.

FIGURE 5.

The e-coachER logic model, adapted to show relationship with qualitative research objectives. Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original.

Reproduced with permission from Ingram et al. 1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original.

Previous research has quantitatively and qualitatively explored the barriers to and facilitators19,56 of engaging in ERS, and the moderators of ERS engagement. 15,57,58 The e-coachER intervention was designed to help overcome many of the reported barriers to, and enhance the use of facilitators of, attending ERSs, becoming physically active in other ways, or both, through this primary care-based intervention. 59

The e-coachER intervention is theoretically underpinned by self-determination theory,31 which asserts that all humans possess three innate basic psychological needs (i.e. autonomy, competence and relatedness). When met, these needs foster intrinsic motivation resulting in personal growth and satisfaction. A range of behaviour change techniques35 based on self-determination theory were employed within e-coachER’s seven ‘steps to health’ (see Table 1), so this process evaluation seeks to establish whether or not the intervention influenced basic psychological needs and behaviour change processes.

A mixed-methods process evaluation seeks to best understand how participants engaged in the intervention (and trial methods) and what the consequences were with respect to our logic model.

Qualitative process evaluation

The logic model shown in Figure 1 has been adapted to show the causal pathways proposed to contribute to behaviour change and intervention outcomes and the objectives for the qualitative work (see Figure 5).

Quantitative process evaluation

Methods

The survey used to capture the process outcomes was described in Chapter 2. Briefly, items were derived from extensive reviews of the literature to ensure that they matched our expected changes within our logic model but also were fit for purpose within a RCT. In other words, they had to make sense for a responder whether or not they did any PA and they had to have some sensitivity to identify change. In response to PPI input, we also had to maximise the participant completion rates so we ensured that the questions were easy to interpret and the response format was clear. The items and respective scales are supplied in Report Supplementary Material 4.

Questions about importance and confidence are single items using an 11-point scale. The remaining items were chosen to represent specific constructs and create composite scores. Confirmatory factor analysis revealed support for adding survey items to assess: perceived competence in being regularly physically active (four items), autonomous in decisions about PA (four items), availability of support (three items), frequency of support (three items), action-planning (five items) and self-monitoring (two items). The Cronbach’s alpha coefficient of all factors were found to be in excess of 0.77, indicating good internal consistency of each. Composite scores were calculated and used in the analysis.

Using a model adjusted for age, gender, stratification variables and baseline scores, and random effects for centre, with only participants included in the primary analysis (i.e. based on complete accelerometer data and had completed the respective survey items), we compared each of the eight process outcomes by group at both 4 and 12 months. We restricted the analysis to outcome data deemed valid, with no more than 200 minutes spent in 10-minute bouts of MVPA, as per the primary analysis.

The size and significance of any mediating effects were evaluated through the product of coefficients method. 63 This was performed irrespective of the results from the main analysis, as mediation may still be possible without having detected a significant effect of the intervention on the primary outcome. 64 Referring to the causal diagram in Figure 6, the coefficient, a, for the intervention effect on process measures in path A was derived from the mixed model of changes in process measures regressed on the intervention, adjusted for age, gender, stratification variables and random effects for centre. Utilising the same adjustment variables, the coefficient, b, for the change in process measures on the primary outcome in path B was obtained by modelling the outcome on the process measure change, also adjusting for the effect of the intervention. The coefficient of the mediating effect was, therefore, calculated as the product a × b. The CIs were calculated using the Sobel test,65 dividing the coefficient product by the estimated standard error used:

⁽¹⁾ SE ab = √ ( a 2 × SE b 2 + b 2 × SE a 2 ) .

FIGURE 6.

A priori path model for testing mediation effects.

Summary

Our mixed-methods approach to understanding if and how the intervention worked for some participants provided interesting insights into engagement with a novel technology-based support system and how that complemented the available support from usual ERSs. The qualitative and quantitative approaches were conducted independently. The interviews highlighted the role of developing self-monitoring and SMART goal-setting skills to increase confidence, which featured strongly in being physically active on a regular basis. The survey process outcomes confirmed that the intervention increased importance and confidence up to 4 months but the effects (compared with usual ERSs) had dissipated by 12 months.

The qualitative interviews were mostly not able to follow intervention participants for long. The lack of intervention effects on changes in other process outcomes (e.g. action-planning and self-monitoring) was surprising given that 36% of the participants reached step 5 (completed a goal review). Some participants continued to complete goal reviews for up to 12 months, but it would appear that an insufficient number of participants did so, and most had stopped completing goal reviews long before completing the 4-month follow-up assessment. This was also despite periodic reminders from the online system to log in.

The interviews provided a little information on how e-coachER had prompted intervention participants to find and use social support to increase PA (e.g. support to get active – step 2). Although designed to be specific to the participant’s geographical area, some felt that the e-coachER links to other PA opportunities were not relevant to their needs. For other participants, however, e-coachER did help to facilitate an open discussion about health within the participants’ social networks. It is possible that simply talking about being in the e-coachER study (i.e. an opportunity to connect) would have been done by participants in both of the trial groups and few actually got around to joining groups or setting plans to exercise with others. Hence, this may explain why the intervention had no effects on the process outcomes, namely identifying and using social support.

In this pragmatic trial, the aim was to determine if there were intervention effects in addition to usual ERSs. An alternative explanation to null intervention effects on some of the process outcomes is that usual ERSs have sufficient positive effects on PA beliefs to make additional effects unlikely. The Glasgow ERS differed from the other ERSs involved in the trial, in that the exercise professional supported participants with behaviour change counselling and signposting to different PA options. In this sense, the Glasgow ERS may have been expected to build a sense of autonomy and the e-coachER support may not have added much. Further analysis is needed to explore differences in intervention effects on process outcomes by site, although site did not interact with the intervention effects on the primary outcome. Health interventions, including ERSs, are notorious for having short-term effects on health behaviour that dissipate with time. Given that ERSs can provide barriers to engagement and sometimes may not promote a range of sustainable PA options, we designed e-coachER to have more lasting appeal and to develop self-regulatory skills and the intrinsic motivation to be physically active for managing chronic conditions, above and beyond usual ERSs. Some key processes did change as a result of e-coachER engagement in the short term but these were not sustained, which is consistent with the absence of any mediating effects of change in these measures on intervention effects on the objective primary outcome of MVPA.

Introduction

This chapter reports the economic analysis of an augmented ERS with web-based behavioural support (e-coachER) versus ERS alone. Comparing the costs and consequences of alternative approaches to promoting health care is key to facilitating efficient allocation of resources. 66

Although economic evidence on ERSs exists (mostly compared with usual care), it is unclear;17 little is known about the value for money of an augmented ERS with online behavioural support. A review of other reviews (n = 3) conducted by Pavey et al. 54 found one study67 showing ERSs to be cost-effective, another one68 reporting mixed findings and a third69 that found limited evidence of effectiveness but higher cost. In an analysis of 21 economic evaluation studies published by NICE from 2006 to 2010, Owen et al. 70 found ERSs to be cost-effective at the NICE threshold of £20,000 to £30,000 per QALY. In a systematic review of economic evaluations, Vijay et al. 71 reported three studies showing a cost per disability-adjusted life-year (DALY)/QALY estimate below £10,000 for exercise prescriptions.

The most recent NICE guideline development on ERSs identified two key gaps in knowledge. 72 First, there is paucity of cost-effectiveness evidence on alternative models of ERS. Second, information on the cost-effectiveness of ERSs for people with comorbidities is lacking. We are aware of an ongoing multicountry, multicentre RCT examining the cost-effectiveness of enhanced ERSs plus self-management strategies compared with ERS alone among inactive patients. 73

This chapter addresses the gaps in the literature by estimating the cost-effectiveness of e-coachER compared with ERSs alone, in adults with range of chronic conditions. The analysis uses a 1-year time horizon (from baseline to 12 months post randomisation) and is conducted from the viewpoint of the NHS, Personal Social Services and patients. The base-case analysis covered NHS and Personal Social Services perspectives.

Methods

The intervention and control population are as defined in Chapter 2. In line with the clinical effectiveness analysis, the samples for base-case analyses are participants who provided valid accelerometry data. Sensitivity analysis explored the impact of using the whole sample.

Measurement and valuation of cost

Total costs were expressed per participant and calculated by multiplying resource use with each relevant unit cost and summing across the range of resource use possibilities across the 12 months of the study.

Resource use, including those ‘in kind’, associated with both intervention and control was identified through discussions with the management team of the trial. Following this, resources were measured and valued without research-driven resource use. The range of resource use covered (1) set-up and design of the intervention, (2) delivery of the intervention including handbooks, pedometers, guide for using the LifeGuide platform, technical support and maintenance of the website, (3) consultation provided by an exercise specialist and staff support to participants, (4) primary and secondary health service use: GP, nurse, social worker, care worker, physiotherapist consultations (both home and practice visits), prescriptions, hospital admissions, accident and emergency (A&E) visits, and other (e.g. podiatrist visits); and (5) time and money expenses borne by participants in relation to participation in the intervention (e.g. time spent on web platform), visit to exercise specialist and PA (e.g. membership fees for gym or sports club). Data on resource use were collected using the trial administrative records, key informant interviews (e.g. trial manager), review of trial management records and participants’ questionnaires at baseline, 4 and 12 months.

Resources were valued using national tariffs (e.g. Unit Costs of Health and Social Care 2017,74 NHS Reference Costs 2015 to 2016,75 Annual Survey of Hours and Earnings76) to increase generalisability. In the absence of available national costs, unit costs came from trial administrative records. Appendix 6 provides details of unit costs. The unit cost of capital costs (i.e. pedometer) was calculated pro rata (costs were spread over their expected lifetime) because use can occur beyond the time period of this analysis. 77 Costs were expressed in 2017/18 Great British pounds, using the Hospital and Community Health Service (HCHS) inflation index where appropriate (PSSRU 2017). 74 No discounting was used as the time horizon of the analysis is 1 year.

Measurement and valuation of outcomes

Two types of outcomes were used for estimating cost-effectiveness: physical units (PA indicator) and QALYs. The PA measure is the primary outcome measure of the trial (as described in Chapter 2): total weekly minutes of MVPA in ≥ 10-minute bouts, recorded objectively by accelerometer. QALYs were estimated by converting EQ-5D-5L utilities using the area under the curve method. EQ-5D-5L questionnaires were completed by trial participants at baseline and 4 and 12 months. In line with the 2018 NICE recommendation,66 utility weight based on the crosswalk function78 was used to assign utility weights. Sensitivity analysis explored the impact of using other valuation sets.

Methods of analysis

Results

Summary statistics (unadjusted for baseline differences) on costs to both providers and participants, and quality of life are provided in Table 23 (see Appendices 7–9). At 12 months, the cost to providers per participant was higher in the intervention group (£1730) than in the control group (£1385). The biggest component of the cost borne by providers was health and social service use cost (91%) and the least was cost associated with the provision of support to participants (< 1%, £0.60 per participant). In terms of cost to participants, the pattern was different as participants in the control group incurred an average cost of £298, compared with £255 in the intervention group. The vast majority of this cost was because of cost related to participation in PA (69%). Fees paid for child/dependant care during the exercise specialist visit were the smallest part (< 1%, £0.20 per participant). The intervention group experienced higher quality of life at 12 months with a QALY of 0.662 (Table 23).

Table 24 shows the main results based on regression estimates that adjusted for baseline differences. Consistent with the pattern observed in the unadjusted estimates, the average cost per participant was £1355 (95% CI £701 to £2008) and £1793 (95% CI £1635 to £1952) in the control and intervention groups respectively. This represents an additional cost of £439 (95% CI –£182 to £1060) in the intervention group, although the difference is not statistically significant. The intervention led to a weak indicative effect on total weekly minutes of MVPA in bouts of ≥ 10 minutes (mean difference 11.8 minutes, 95% CI –2.1 to 26.0 minutes), compared with the control group. In terms of quality-of-life outcome, the intervention group (mean 0.663, 95% CI 0.625 to 0.701) had more QALYs than the control group (mean 0.637, 95% CI 0.585 to 0.688) (see Table 24). The difference in QALYs (0.026, 95% CI 0.013 to 0.040) between the two groups was statistically significant. The cost–utility ratio shows that, compared with the control group, the intervention cost an additional £16,885 per QALY. This is below the NICE threshold of £20,000–30,000 per QALY.

Table 25 shows the results of the deterministic sensitivity analyses. The base-case finding was robust to deterministic sensitivity analyses with a few exceptions. The intervention was found to be more expensive but produced more QALYs. Excluding health and social service use costs, or changing the perspective of analysis, improved the cost-effectiveness of e-coachER, with incremental cost-effectiveness ratios (ICERs) ranging between £9000 and £15,885. Using estimates based on the whole sample (all participants who were randomised) was decisionally significant. It produced a worse cost-effectiveness ratio, with the intervention becoming not cost-effective (£43,900 per QALY).

Compared with the base-case findings, subgroup analysis showed the intervention to be more cost-effective in groups that reported that hypertension (dominates control) or osteoarthritis (ICER £12,389 per QALY) or type 2 diabetes (ICER £14,886 per QALY) was the primary reason for referral. Among individuals who reported that being overweight was the primary reason for referral, e-coachER was still cost-effective but at a higher ICER value (£18,421 per QALY). In the group that reported that low mood was the primary reason for referral, e-coachER was more expensive (additional cost of £989 per participant) and less beneficial (fewer QALYs of –0.056 per participant) than the control.

Figure 7 shows the cost-effectiveness plane for the intervention compared with the control. The majority of the cloud of points (representing mean differences in costs and QALYs) are located in the north-east quadrant of the plane. This indicates that the intervention has high likelihood of generating more QALYs but at higher costs. The probability of the intervention being cost-effective (compared with control) at multiple willingness to pay per QALY values is presented in Figure 8. At £10,000 per QALY, the intervention has about 30% chance of being cost-effective compared with the control. The likelihood of cost-effectiveness nearly doubles (51%) at the £20,000-per-QALY threshold and increases further to 63% at the £30,000 threshold.

FIGURE 7.

Cost-effectiveness plane for intervention vs. control at 12 months.

FIGURE 8.

Cost-effectiveness acceptability curve showing the probability of within-trial cost-effectiveness for intervention vs. control at different willingness to pay per QALY thresholds.

Discussion

This study shows that providing online behavioural support for participants of ERSs cost providers £1793 per person. Although the intervention costs £439 more than offering the ERS alone, it leads to better quality-of-life outcomes (0.026 more QALYs per person) and increased participation in PA (12 more weekly minutes of MVPA in ≥ 10-minute bouts per person). Although the differences were mostly not statistically significant, it is important to note that this is not sufficient proof of no significant effect, as the clinical trial was not powered to identify the changes in the economic outcomes (cost and QALY), particularly given the small number of observations in the subgroup analysis. Compared with the control, the intervention cost an additional £37 to gain a 1-minute increase in weekly MVPA (in ≥ 10-minute bouts) and £16,885 per QALY gain per person. Based on the NICE threshold of £20,000–30,000 per QALY, e-coachER could be considered more cost-effective than ERS alone. If decision-makers are willing to pay £30,000 for 1 QALY, e-coachER has a 63% probability of being cost-effective. The findings were robust to sensitivity analysis, including when the cost of participants was added to total costs of the programmes. Exploratory subgroup analysis found that the intervention was cost-effective among participants who reported that hypertension, osteoarthritis, type 2 diabetes or weight problems (i.e. being overweight) was the primary reason for referral but was not cost-effective among those who reported that low mood was the primary reason.

The novelty of e-coachER, its comparison with usual ERSs and the disease-specific population makes it complicated to relate the findings here meaningfully to the existing economic literature on ERSs. We identified one comparable study. Murphy et al. ,17 a pragmatic RCT, assessed the cost-effectiveness of the Wales National Exercise Referral Scheme (NERS), a 16-week programme including motivational interviewing, goal-setting and relapse prevention compared with usual care. Participants were inactive people with depression and/or coronary heart disease risk. The NERS intervention was shown to be cost-effective at 12 months, with a cost-per-QALY estimate of £12,111 and 89% likelihood at the £30,000-per-QALY threshold.

Our analysis found that costs associated with health and social services constitute the largest proportion of total costs associated with the intervention. Although the impact on cost-effectiveness was not decisionally insignificant, an important consideration is why the intervention group had higher health service use cost (mean cost of £135) than the control group. Further exploration shows that in the first 4 months of the trial, on average, the intervention group had a health service cost of £819 per participant and the control group had a health service cost of £639 per participant (see Appendix 9). In terms of the subcomponents of health service use cost collected in this study (n = 13), the intervention had higher costs in all except two [A&E visits and hospital visits (outpatients)]. The reverse pattern was, however, observed from 5 to 12 months – the control group had higher health service costs than the intervention group (£713 vs. £668). Similarly, the control group had higher costs for all subcost components apart from prescriptions and care worker visits. A systematic review on injury consequences of PA found that, although relatively minor, increased activity could lead to adverse health consequences. 86 However, for older adults, improved participation reduces the risk of fall-related injuries. 87 Future studies are required to further investigate the impact of adverse effects on the cost-effectiveness of PA programmes.

This study feeds into a limited evidence base around the efficiency of different models of ERS. To the best of our knowledge, this is the first study to demonstrate the cost-effectiveness of online support for an ERS and in a population with common chronic conditions. The strengths of this study include the use of robust clinical effectiveness data from a large multicentre trial, rigorous analyses to determine the impact variant measures of quality of life and participant perspective, subgroup analysis among different disease groups, comprehensive coverage of health service and social care use cost. A key limitation, however, is the short-term perspective of the analysis herein. Modelling of the long-term costs and effects is important where benefits and costs extend beyond the end of a trial. It is of particular relevance to this trial, where the benefits or costs could be experienced in the future too (as resource savings from reduced disease and, therefore, benefits in terms of increased quality of life). We expect that the impact of the long-term effects may have underestimated the cost-effectiveness of e-coachER.

There is currently limited national public health guidance on the implementation of ERSs enhanced with online support for participants. The results herein show that based on the NICE threshold of £20,000–30,000 per QALY, e-coachER could be considered more cost-effective than ERS alone. To strengthen the economic case, future studies are recommended to examine the long-term cost-effectiveness of the e-coachER and ascertain the impact of the trajectory of activity levels on future health service and future quality of life.

Summary of findings

The 450 trial participants were 64% female (n = 290), had an average age of 50 years and had an average BMI of 32.6 kg/m²; 65% and 35% were classified as inactive (n = 293) and moderately inactive (n = 157), respectively, and 50% reported that weight loss was the primary reason for referral (n = 225), followed by low mood (n = 85, 19%), osteoarthritis (n = 54, 12%), diabetes or prediabetes (n = 49, 11%) and high blood pressure (n = 36, 8%). Participants also noted that weight loss (n = 364, 81%), low mood (n = 243, 54%), high blood pressure (n = 149, 33%), diabetes or prediabetes (n = 117, 26%) and osteoarthritis (n = 108, 24%) may have been one of the reasons for referral.

The e-coachER support package was developed, which included mailing participants a pedometer, a fridge magnet with attached tear-off strips to record daily PA and a bespoke website to overcome barriers to increase daily PA for some people who are not willing or able to engage in an ERS. The e-coachER intervention demonstrated a mixed level of engagement. About one-third of participants did not register online and about one-third completed what we thought would be an adequate ‘dose’ of at least five of the seven steps, involving at least setting a PA goal and reviewing it 1 week later.

The results show that, compared with usual ERSs, the e-coachER intervention group had slightly more, but not significantly more, accelerometer-recorded minutes of MVPA (recorded in bouts of ≥ 10 minutes) at 12 months. The pattern was similar when considering the level of intervention engagement, with a slightly larger difference in favour of the intervention group. Applying the same approach as in the primary analysis, there were no between-group differences at 12 months in any of the other accelerometer-derived or self-reported MVPA outcomes, with one exception. The intervention group had significantly more daytime sedentary time (accumulated in blocks of ≥ 5 minutes) at 12 months. In ITT imputed comparison at 12 months, the intervention group was more likely than the control group to self-report that they had achieved 150 minutes of weekly MVPA (odds ratio 1.55, 95% CI 0.99 to 2.42; p = 0.05). The intervention also had no effect on ERS attendance (78% vs. 75% in control and intervention, respectively), or EQ-5D-5L or HADS scores at 12 months, compared with the control group. In ITT imputed comparison at 12 months, the intervention group had lower HADS depression and anxiety scores than the control group. The proportion of participants attending the ERS was comparable to findings from a review15 that reported that the average ERS uptake was 81% in RCTs.

Economic evaluation

Over the 12-month follow-up, the average cost per participant was £1355 (95% CI £701 to £2008) and £1793 (95% CI £1635 to £1952) in the control and intervention groups respectively. Compared with the control group, the intervention group incurred an additional mean cost of £439 (95% CI –£182 to £1060) but generated more mean QALYs (0.026, 95% CI 0.013 to 0.040), with an incremental cost-effectiveness ratio of an additional £16,885 per QALY.

Although insignificant, an important consideration is why the intervention group had higher health service use cost (mean cost of £135) than the control group given the intervention was effective and led to increases in both PA and health-related quality of life. The difference in the cost, which was not statistically significant, was observed irrespective of the central measure of tendency, mean or median (see Appendix 9). Further exploration shows that in the 4 months of the trial, on average, the intervention group had a health service cost of £819 and the control group had a cost of £639. In terms of the subcomponents of health service use cost collected in this study (n = 13), the intervention group had higher costs in all except two components [A&E visits and hospital visits (outpatients)]. The reverse pattern was, however, observed from 5 to 12 months, with the control group qualitatively having greater health service costs (£713 vs. £668). Similarly, the control group had higher costs for all subcost components apart from prescriptions and care worker visits.

Effectiveness of and engagement in the intervention

One of the criticisms of e- and m-health interventions is that engagement is not appealing to enough people and can be rather short-lived. 23 The 9-month work in developing the intervention, building on other effective LifeGuide interventions and with public and patient involvement, aimed to maximise engagement to theoretically have the greatest impact on MVPA outcomes at 12-month follow-up. Automated periodic e-mails were sent to intervention participants up to 12 months. The intervention included evidence-based and theory-driven components as well as a pragmatic approach to enhance engagement. Pedometers and recording sheets attached to fridge magnets were provided in the initial introductory pack as basic tools for self-monitoring PA and setting and reviewing SMART goals. These may have been sufficient to get some of the 36% of intervention participants who never registered to think about behavioural self-regulatory processes to increase PA for managing their chronic condition(s). Attending the ERS may also have provided sufficient support to become more physically active, without the use of the pedometer or web-based support. Other participants may also have regarded the pedometer as rather basic and decided to use a more sophisticated app on their smartphone.

The fact that 64% of participants did register online and completed their first step, with 36% going on for over 4 weeks to complete a goal review, provided us with some assurance that the e-coachER intervention was acceptable for a reasonable proportion of participants. There is evidence from previous research with LifeGuide interventions that even limited online engagement can be effective and there is no clear ‘adequate dose’ to optimise behavioural change. Our CACE analysis confirmed that completing our prespecified intervention engagement level (step 5) did not lead to significantly greater 12-month objectively recorded minutes of MVPA, recorded in ≥ 10-minute bouts, compared with the control group, although the between-group differences were qualitatively greater (22.9 vs. 11.8 minutes).

Our logic model predicted that e-coachER engagement would strengthen various beliefs that would in turn translate into increases in MVPA, compared with usual ERS support. Among only the participants included in the primary analysis, the intervention did increase the following compared with the control group: perceived importance of doing at least 30 minutes of moderate-intensity PA (e.g. brisk walk) on at least 5 days per week, confidence in achieving at least 30 minutes of moderate-intensity PA (e.g. brisk walk) on at least 5 days per week and perceived competence in being regularly physically active at 4 months (but not 12 months). Changes (from baseline to 4 months) in these process outcomes did not mediate changes in the primary outcome at 12 months.

In the qualitative part of our process evaluation, 12% of intervention participants were interviewed. Overall, e-coachER was acceptable and positively experienced and did ‘do what it said on the tin’ in terms of enhancing autonomy, competence and relatedness for many participants. Inevitably, because the web-based support was available to support participants with a range of levels of IT literacy and chronic conditions, some of the content did not appeal to everyone. That said, the idea that the support aimed to facilitate engagement with the self-monitoring process and then encourage them to move on to using more sophisticated devices to self-monitor PA for example, was overlooked by some participants, but not all. Similarly, we tried to help participants to find any source of social support to support increases in PA and a preferred form of PA that they could enjoy, whether or not that involved the ERS, and that overarching intent ‘may have been lost’ by some of those interviewed. The user guide sent to intervention participants initially could have spelled out some aims of the support more explicitly and if this had been understood then some comments from participant interviews may not have been made. That said, some interviewees did note that they had a strong IT background and acknowledged that their comments were personal and acknowledged the need to appeal to those with a lower level of IT literacy than themselves.

Strengths and limitations

Implications for health care

Offering e-coachER support had only small but non-significant effects on objectively recorded MVPA compared with usual ERSs at 12 months. The cost–utility ratio shows that when compared with the control, the intervention cost an additional £16,885 per QALY and has a 63% probability of being cost-effective for increasing MVPA, compared with usual ERS.

As a result of usual ERS alone (i.e. in the control group), across the sites there was only weak evidence of a change in accelerometer-recorded MVPA at 4 months, but none at 12 months, albeit without comparison with no ERS. If anything, there were small but non-significant reductions in MVPA among only those engaging in usual ERS. The rationale for conducting the present study was that we offered an additional ‘package’ of support (e-coachER) aimed at developing self-determined PA alongside usual ERSs or instead of usual ERSs. If shown to add additional benefit, our intervention would be available to primary care professionals to offer to inactive patients, with a range of physical and mental health conditions, at the time of making a referral to a local ERS.

Providing web-based behavioural support to participants of ERSs offers an additional strategy to augment usual ERSs to promote PA. There are a few small aspects of the intervention that we would change based on participant feedback, but the intervention could also be extended to be suitable for use by patients with other chronic conditions (e.g. cancer). There was reasonable engagement in e-coachER support for participants with a range of confidence in using IT, indicating that, if implemented, it would have low costs and moderate value in promoting PA in addition to usual ERSs. Our process evaluation indicated that there were changes in some of the measures that we collected to assess key components of the logic model. For example, the intervention led to improvements up to 4 months in confidence and competence in doing PA, perceiving the importance of PA, a sense of availability of support, action-planning and self-monitoring compared with the control group. But these intervention effects were sustained at 12 months only for perceived importance of PA. In our exploration of whether or not any of the changes between 0 and 4 months mediated any intervention effects on the primary outcome (accelerometer-recorded minutes of MVPA in bouts of ≥ 10 minutes) at 12 months, there was no evidence that this was the case. Our ability to detect these mediation effects may have been limited by the overall intervention effects on MVPA at 12 months.

The present study found that 36% of participants did not log into the online e-coachER support, but did receive a pedometer and a fridge magnet to record MVPA. We will further analyse the data to determine if this was sufficient to increase MVPA compared with usual ERSs. Other evidence suggests that providing primary care patients with a pedometer to self-monitor PA is effective in changing objectively recorded MVPA. 61 It would be relatively easy to provide patients with a pedometer at the same time as referring them to an ERS.

The LifeGuide platform has been used to deliver evidence-based and theory-driven interventions online to support change in a wide range of health behaviours. One of its strengths is the ability to capture intervention engagement to help understand fidelity issues and add to the literature on how people change as a result of e-health interventions. Although LifeGuide-delivered interventions do these things well and have been shown to be effective in supporting change in a range of health behaviours and weight loss, other more sophisticated technological innovations are rapidly taking over. Indeed, feedback captured within our process evaluation noted that the e-coachER support was rather unsophisticated, and typing step counts or minutes of MVPA accumulated in the past week into a website and getting feedback on whether or not goals had been achieved can instead undoubtedly be done on a range of more sophisticated devices that are embracing digital technology and artificial intelligence. For these reasons, the LifeGuide platform will no longer support digital interventions from 2021.

We hoped that the intervention would encourage participants to go to the local ERS that they had been referred to but this did not happen. Almost 25% of referred participants did not attend any sessions.

Future research implications

Previous research has compared usual ERSs with an enhanced ERS, involving additional exercise practitioner training, but showed no additional effect20 on PA. The present trial provides no clear support for adding a web-based support package to usual ERSs, to increase long-term MVPA. Our process evaluation revealed that some improvements to the web support could be made, such as mobile options with smartphone apps for self-monitoring and goal-setting.

The modest engagement in the online e-coachER support suggests that work is needed to understand which factors influenced intervention engagement and how best to further develop low-cost and scalable support to increase ERS uptake and maintenance of PA. Once this has been done, further research could examine the effects of a modified e-coachER-type intervention for participants with chronic conditions involved in the present study and others (e.g. with cancer, back pain and in cardiac rehabilitation).

The e-coachER study has provided a rich data set, which offers the chance to explore additional questions including the following:

• What were the characteristics of participants that predicted changes in 4- and 12-month PA?

• How did different measures of MVPA (i.e. self-report and accelerometer derived) influence the findings, beyond what we present here?

• What other aspects of intervention engagement (derived from the LifeGuide platform) were used, and did any influence changes in process and behavioural outcomes?

• Among subsets of the sample (e.g. those with low mood), what changes in quality of life, depression and anxiety occurred as a result of the intervention versus usual ERSs?

Conclusions

With modest engagement in the evidence-based and theory-driven e-coachER intervention, which was captured by the web-based system, the intervention effects on a rigorously defined, objectively assessed, PA primary outcome at 12 months were only small and not significant, and, because of a smaller sample size than intended, should be treated with caution.

In the cost-effectiveness analyses, the cost–utility ratio shows that, compared with an ERS alone, ERS plus the e-coachER intervention cost an additional £16,885 per QALY and has a 63% probability of being cost-effective based on the UK threshold of £30,000 per QALY.

Contributions of authors

Professor Adrian H Taylor (https://orcid.org/0000-0003-2701-9468) (Professor of Health Services Research) conceived the idea for the study with co-applicants: Professor Rod S Taylor (https://orcid.org/0000-0002-3043-6011) (Professor of Health Services Research, Statistician), Dr Nana Anokye (https://orcid.org/0000-0003-3615-344X) (Senior Lecturer in Health Economics), Professor Sarah Dean (https://orcid.org/0000-0002-3682-5149) (Professor in Psychology Applied to Rehabilitation and Health), Professor Kate Jolly (https://orcid.org/0000-0002-6224-2115) (Professor of Public Health and Primary Care), Professor Nanette Mutrie (https://orcid.org/0000-0002-5018-6398) (Director of Physical Activity for Health Research Centre), Professor Lucy Yardley (https://orcid.org/0000-0002-3853-883X) (Professor of Health Psychology), Professor Colin Greaves (https://orcid.org/0000-0003-4425-2691) (Professor of Psychology Applied to Health), Professor John Campbell (https://orcid.org/0000-0002-6752-3493) (Professor of General Practice and Primary Care), Mr Ben Jane (https://orcid.org/0000-0002-0774-2942) (Senior Lecturer, Exercise and Public Health), Dr Jo Erwin (https://orcid.org/0000-0002-9648-4825) (Research Associate), Professor Paul Little (https://orcid.org/0000-0003-3664-1873) (Professor of Primary Care Research within Medicine) and Professor Anthony Woolf (https://orcid.org/0000-0001-8482-8056) (Consultant Rheumatologist, Honorary Professor of Rheumatology).

Professor Adrian H Taylor, all co-applicants listed above, Dr Wendy M Ingram (https://orcid.org/0000-0003-0182-9462) (Clinical Trials Manager) and Mr Douglas Webb (https://orcid.org/0000-0001-6818-1838) (Assistant Clinical Trials Manager) contributed to the final study design and development of the protocol.

Professor Adrian H Taylor, Dr Jeffrey Lambert (https://orcid.org/0000-0003-4774-9054) (Research Associate in Primary Care) and Dr Mary Steele (https://orcid.org/0000-0003-2595-3855) (Research Fellow) developed the web support using LifeGuide and, with Ms Jennie King (https://orcid.org/0000-0001-7571-6829) (Research Assistant), led the patient and public input into testing and feedback.

Professor Sarah Dean developed the process evaluation plan with Dr Jeffrey Lambert, Professor Colin Greaves, Mr Nigel Charles (Associate Research Fellow) and Dr Rohini Terry (https://orcid.org/0000-0003-4188-8504) (Research Fellow). Professor Sarah Dean led the analysis and reporting of the qualitative analysis for the process evaluation, in collaboration with Professor Colin Greaves, Mr Nigel Charles, Dr Rohini Terry and Professor John Campbell (https://orcid.org/0000-0002-6752-3493).

Dr Nana Anokye provided the health economics evaluation plan and conducted and reported the economic evaluation in accordance with the health economics evaluation plan.

Professor Rod S Taylor provided the statistical analysis plan and conducted and reported the statistical analysis in accordance with the statistical analysis plan.

Dr Adam Streeter (https://orcid.org/0000-0001-9921-2369) (Research Fellow in Medical Statistics) provided additional support to the statistical analysis (primarily the quantitative process evaluation).

Mr Ben Jones (https://orcid.org/0000-0001-7374-1107) (Research Assistant in Medical Statistics) assisted with statistical data analysis.

Dr Lisa Price (https://orcid.org/0000-0001-9478-3488) (Lecturer in Physical Activity and Health) was responsible for accelerometer data processing, advising and reporting on accelerometer-derived measures.

Dr Ben Ainsworth (https://orcid.org/0000-0002-5098-1092) (Senior Research Fellow) developed the web support and processed intervention engagement data on the LifeGuide platform.

Professor Adrian H Taylor, Professor Kate Jolly and Professor Nanette Mutrie were principal investigators, assisted by Dr Chloe McAdam (https://orcid.org/0000-0002-7574-7489) (Research and Knowledge Exchange Co-ordinator at the Physical Activity Health Research Centre) at the Glasgow site.

Ms Jennie King was the lead research assistant and Ms Lucy Hughes (https://orcid.org/0000-0002-6015-0356) was the research assistant at the Birmingham site.

Dr Wendy Ingram and Mr Douglas Webb (https://orcid.org/0000-0001-6818-1838) were the trial managers.

Mr Chris Cavanagh was the PPI representative.

All authors critically revised successive drafts of the manuscript and approved the final version.

Publications

Ingram W, Webb D, Taylor RS, Anokye N, Yardley L, Jolly K, et al. Multicentred randomised controlled trial of an augmented exercise referral scheme using web-based behavioural support in individuals with metabolic, musculoskeletal and mental health conditions: protocol for the e-coachER trial. BMJ Open 2018;8:e022382.

Taylor AH, Taylor RS, Ingram W, Dean S, Jolly K, Mutrie N, et al. A randomised controlled trial of an augmented exercise referral scheme using web-based behavioural support for inactive adults with chronic health conditions: the e-coachER trial. Br J Sports Med 2020; in press.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to available anonymised data may be granted following review and appropriate agreements being in place.

Disclaimers

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

Serious adverse events reported in the control group

Serious AEs were hospitalisations, with the exception of SAE010029/001 (Morton’s neuroma), which was categorised as ‘persistent/significant disability/incapacity’.

Serious adverse events reported in the intervention group

Serious AEs were hospitalisations, with the exception of SAE010011/001 (asthma attack), which was categorised as a life-threatening event.

Participant telephone interview schedule

Interview topic guide for e-coachER study research assistants

Thank you very much for taking part in an interview to talk about your experiences of working on e-coachER. The aim of the interview is hear your views and experiences of what worked well and what could be done differently if a full trial were to be carried out. Additionally, it is hoped that your experiences and views can help to develop a clearer understanding of ‘general’ issues related to the research process, particularly recruitment of patients to a trial.