An evidence-based approach to the use of telehealth in long-term health conditions: development of an intervention and evaluation through pragmatic randomised controlled trials in patients with depression or raised cardiovascular risk

Importance of the research and its relevance to the priorities and needs of the NHS

Over 15 million people in England have a LTC and treatment of LTCs accounts for 70% of total health-care expenditure. Improvements in LTC management could have major benefits in terms of patient health, quality of life and use of NHS resources. 5

In this programme we decided to test a new telehealth approach, delivered by NHS Direct, within two exemplar conditions. These were (1) patients with depression and (2) those at high risk of cardiovascular disease (CVD). We chose these examples for several reasons. They are very different types of conditions, one affecting mental health and one physical health. However, both conditions are very common and have a major impact on quality of life. They also account for a high proportion of health-care resources and there was good reason to suggest that the potential need for care is considerably greater than the capacity of the NHS to deliver care. Furthermore, in both cases there was some evidence of effective telehealth interventions, but this evidence was inconclusive. Further discussion of these points in relation to the exemplar conditions is provided in the following sections.

Studying two very different LTCs in parallel, and developing management programmes for them based on a common theory-driven approach, would enhance generalisability. The vision was that, if these telehealth programmes were both effective, this would provide a framework and a justification for the future development of further programmes for other LTCs.

Need for research in this area

Postscript: closure of NHS Direct

This research programme ran from 2009 to 2015, with the trial of the intervention being conducted between 2013 and 2014.

NHS Direct was closed in March 2014 in favour of the less expensive ‘111’ telephone helpline and the NHS Choices website. The closure of NHS Direct had been announced in October 2013 but there had been widespread rumours of closure ever since the BBC leaked in August 2010 that the incoming coalition government planned to replace NHS Direct. 55

Although the potential of NHS Direct had provided the impetus for this research programme, the underlying research questions about the role of a comprehensive management programme for patients with LTCs based on telehealth remain entirely relevant. The closure of NHS Direct inevitably led to some difficulties in conducting the research programme and in particular meant some delays and pauses in the delivery of the intervention; however, we were able to complete the trial thanks to Solent NHS Trust community trust replacing NHS Direct as the host for the intervention and the research programme.

NHS Direct offered advantages because of its national reach and infrastructure. However, in practice, the NHS Direct intervention staff worked from one site (Nottingham) and the intervention that was developed through this programme could be delivered by other provider organisations and made available throughout the UK (as was demonstrated by our ability to successfully relocate the service to Solent NHS Trust). Therefore, we do not believe that the closure of NHS Direct has had an important impact on the relevance or findings of this research programme.

Aim

The aim of this programme of research was to develop, implement and evaluate new programmes of care delivered via NHS Direct for patients with LTCs and to provide evidence about the benefits and costs of these initiatives.

The intended benefits of these new forms of provision were to improve health outcomes for patients, facilitate self-management, improve the patient experience and improve the cost-effectiveness of care provision.

The programme focused on two exemplar conditions: depression and high CVD risk.

Objectives

1. To review evidence about telehealth interventions designed to improve health care for patients with LTCs in order to develop a theory about which types of interventions potentially delivered by NHS Direct are most likely to be effective.

2. Using qualitative methods, to critically examine how NHS Direct resources could best be incorporated into the LTC management of patients and integrated with current primary care professional practice.

3. To explore patient factors and access factors that are associated with unmet need and with willingness to use NHS Direct and specifically types of telehealth interventions that are most likely to be acceptable to different patient groups.

4. Based on the first three activities, to develop and optimise interventions offered by NHS Direct that are likely to be acceptable, effective and efficient.

5. To determine the clinical effectiveness and cost-effectiveness of LTC management provided by NHS Direct in the two exemplar conditions: depression and CVD risk.

Introduction

The evidence base for home-based telehealth for LTCs is large, with numerous systematic reviews of a variety of telehealth interventions. Before commencing our evidence synthesis, there were already numerous systematic reviews of a variety of telehealth interventions and three relevant meta-reviews published. Meta-reviews synthesise the evidence from systematic reviews. The first meta-review focused on internet interventions only (including virtual communities, internet-based educational and behavioural programmes and internet-based CBT) and showed improved outcomes, particularly for patients with depression. 62 The second was a wider review encompassing all types of real-time telehealth and conditions, which included, but was not specific to, LTCs or home-based telehealth. 23 This concluded that the quality of the evidence base was poor, but indicated benefits of telehealth for the management of chronic diseases. These benefits included improving health outcomes (and mortality in heart failure), but there was no benefit for resource utilisation or processes of care. The third meta-review focused on telemedicine and included e-health, information communication technologies, internet-based interventions and telehealth for both diagnosis and management of a wide range of conditions, concluding that 21 of 80 reviews identified telemedicine as effective and 18 reviews identified telemedicine as promising. 63 Although these reviews offered important conclusions regarding the effectiveness of telehealth, they did not focus exclusively on home telehealth or the management of LTCs. Therefore, there was a need for a further meta-review of home-based telehealth for managing LTCs to inform the Healthlines study intervention.

Methods

Results

The evidence base

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram shows the stages that we went through to identify 36 reviews (Figure 1). Of these, 11 were excluded because they provided no outcomes, were not systematic or focused on acute care, a single condition or ‘smart home technology’. Twenty-one reviews satisfied the first three mandatory quality criteria and 16 of these met at least four criteria. Our meta-review includes these 16 high-quality systematic reviews,8,11,16,17,20,40,68–77 which cover 662 individual studies (Table 1). Six reviews included a meta-analysis. The reviews were undertaken by authors from Canada (n = 6), the USA (n = 5) and Europe (n = 5) and covered a range of telephone, mobile, internet-based and computer interventions. Details around data extraction are reported in Appendix 1.

FIGURE 1.

The PRISMA flow diagram for the meta-review of systematic reviews. CINAHL, Cumulative Index to Nursing and Allied Health Literature; HTA, Health Technology Assessment.

TABLE 1 - Quality assessment of the 16 reviews included in the meta-review

N, no, the study does not meet the relevant criterion; U, unsure, insufficient information to determine if the study meets the relevant criterion; Y, yes, the study meets the relevant criterion.

Review	Inclusion/exclusion criteria reported?	Search adequate?	Included studies synthesised?	Validity of studies assessed?	Sufficient details about studies presented?
Barlow 200720	Y	Y	Y	Y	U
Botsis 200870	Y	Y	Y	Y	Y
Bowles 200717	Y	Y	Y	N	Y
Cole-Lewis 201071	Y	Y	Y	Y	Y
Cuijpers 200840	Y	Y	Y	Y	Y
Dellifraine 200816	Y	Y	Y	Y	Y
García-Lizana 20078	Y	Y	Y	Y	Y
Hersh 200672	Y	Y	Y	Y	U
Krishna 200973	Y	Y	Y	N	Y
Murray 200511	Y	Y	Y	Y	Y
Oake 200968	Y	Y	Y	Y	Y
Paré 201069	Y	Y	Y	N	Y
Polisena 200974	Y	Y	Y	Y	Y
Rains 200975	Y	Y	Y	Y	Y
Stinson 200976	Y	Y	Y	Y	Y
Tran 200877	Y	Y	Y	Y	Y

Conclusions

The evidence base addressing the effectiveness of home-based telehealth for LTCs was extremely large and generally positive. However, a number of systematic reviews recommended caution when using this evidence base because of the poor quality of studies, citing small sample sizes, weak study design and lack of adequate comparators. There was also very limited evidence on cost-effectiveness. This was supported by a more recent evidence synthesis of the value of telemedicine in the management of five common LTCs: asthma, COPD, diabetes, heart failure and hypertension. 54 This concluded that most studies have reported positive effects but have measured outcomes in the short term only and that the evidence base is ‘on the whole weak and contradictory’ (p. 219). 54

The conclusion from this part of the evidence synthesis was that rigorous evaluation of telehealth interventions for LTCs, including their cost-effectiveness, is needed. The implications for the Healthlines study were that the evidence base was too diffuse to make a significant contribution to the development of the study intervention, although there was sufficient indication of positive effects from telehealth to make it worthwhile to develop a new intervention as proposed.

Introduction

After reading the meta-review, the Healthlines study team was interested in the evidence specific to telehealth interventions for the two exemplar LTCs in the study: depression and CVD risk factors. The review of CVD risk factors was not pursued at this stage because of resource constraints. Few trials or reviews had selected patients at risk of CVD and it was not feasible within the time available to conduct individual systematic reviews for the evidence in relation to each of the large number of factors that constitute raised CVD risk (hypertension, obesity, smoking, hyperlipidaemia, etc.). As we explain later (see Systematic review of telehealth interventions for primary prevention of cardiovascular disease), an opportunity did arise to carry out the CVD risk review at a later stage. However, we were able to review the evidence base for depression in the first phase of this programme, which is summarised here.

Methods

The focus of this review was on identifying systematic reviews of telehealth for depression and other mental health problems. We searched six databases for relevant systematic reviews that were published between January 2005 and March 2010. With regard to depression, we identified nine systematic reviews of telehealth and/or web-based interventions for depression,78–86 three of which provided a meta-analysis. 79,80,86 We also identified 11 reviews about a range of mental health problems or anxiety disorders15,40,87–95 and, of these, five provided a meta-analysis. 40,87–89,95 Most of these reviews included a small number of trials and small sample sizes, raising concerns about the quality of the evidence base.

Findings

When conducting this review we focused on five key questions that we sought to answer from the evidence base. We approached the evidence in this way because the main purpose of this work was to inform the development of the subsequent intervention to be tested later in the research programme. Each of the questions addressed by the review, as well as the available evidence derived from the depression review, is discussed in turn in the following sections.

Conclusions

The evidence base for telehealth for depression and other mental health problems was similar to the overall evidence base: although systematic reviews were available, these were based on a small number of trials with small sample sizes, creating an essentially weak evidence base. Nonetheless, the findings about effectiveness were generally positive. There was support for telephone-based and computerised- or internet-based interventions for depression. In fact, CBT interventions appeared to be very effective, with evidence that professional support could further improve the effectiveness of computerised CBT. There was little evidence of cost-effectiveness.

The implications of this part of the evidence synthesis for the Healthlines study were that further rigorous evaluation of interventions would be welcome, given the quality of the evidence base, but that there is support for a range of interventions, including computerised CBT. In addition, there was some indication that the provision of professional support for computerised interventions might be beneficial.

Introduction

In a third piece of work, we conducted a review of qualitative research into home-based telehealth for LTCs to develop an understanding of patient and organisational perspectives of telehealth.

Methods

Findings

Conclusions

Qualitative literature on patients’ views of telehealth suggests that it is generally accepted and appreciated because of perceptions that it increases access, particularly for remote and hard-to-reach patients or older patients, improves self-management if monitoring is involved and improves care or support, particularly in depression and mental health. However, the technology can be a barrier for some older patients, some patients with LTCs that involve physical disabilities and patients with low literacy levels. There were also some concerns about dependence on technology and health professional contact may be important for mental health interventions.

Professionals were less accepting of telehealth than patients, voicing concerns over loss of role, confidentiality, loss of face-to-face contact or the therapeutic relationship and loss of non-verbal communication. However, they also perceived some benefits of telehealth, such as increasing access and patient contact, improving communication and monitoring and facilitating self-management.

The implications for the Healthlines study were that there would be few problems with acceptability of interventions to patients with LTCs, there was potential to impact on self-management when monitoring was used, attention would need to be paid to directing computer-based interventions at people who were used to computers or technical support should be offered (particularly for older people) and health professional input might be of benefit for the intervention for depression.

Introduction

This part of the evidence synthesis used a realist synthesis to focus on the mechanisms that might be at play within telehealth interventions.

Methods

We followed Pawson’s seven stages for a realist synthesis: identify the question and clarify the purpose of the review; theory elicitation; search the evidence; appraisal; extract the results; synthesise findings; and draw conclusions and make recommendations. 126 This was an iterative rather than a linear process and in our study we had searched for much of our evidence prior to theory elicitation.

Findings

Conclusions

In terms of ‘relationships’, the evidence suggested two underpinning mechanisms by which relationships enable successful telehealth (or indeed any) interventions: relationships provide support (professional, peer, clinical and social) for behaviour change and relationships provide opportunities for professional feedback, which reinforces positive or required behaviour change. This suggests that interventions that enable connections and contact, notably between patients and professionals, can facilitate support and reinforcement necessary for behaviour change. When telehealth interventions limit or remove the relationship between patients and professionals, other opportunities to support and reinforce behaviour may be necessary. In terms of ‘fit’, the literature pointed to the importance of acceptability and ease of use of telehealth interventions for patients and professionals. Telehealth can increase accessibility of care for some populations. Simple technologies appeared to work as well or better than more complex ones and there are some patient groups who are less able to use some technologies (notably, the web). The third proposition centred on ‘visibility’. This had both positive and negative dimensions in that visible monitoring and surveillance appeared to be beneficial, but anonymity was valued by some patients. The evidence suggested that visibility operated in the following ways: by enabling feedback, which reinforces positive or required behaviour change; by providing incentives, reminders and behaviour prompts for action; and by inducing negative feelings (fear) regarding surveillance, stigma and punishment.

There were three implications for developing the Healthlines study intervention. First, relationships may be a necessary component of care interventions. For this reason, the research team needed to consider whether and how the telehealth intervention enables or limits the possibility for relationships with professionals and/or peers. If the intervention removes or replaces relationships, then other mechanisms for support and reinforcement may be necessary to effect behaviour change. Given the apparent importance of relationships, it seems likely that patients and professionals might resist or reject interventions that threaten or limit these. Second, successful telehealth interventions are well integrated into everyday life and health-care routines. Interventions that enhance or improve access to care (by enabling access or making it timelier) are more likely to be acceptable to patients. Ease of use is important for adoption of technologies. Our synthesis suggests that the intervention should be based on a comparatively simple technology (e.g. telephone), which is easy to access and use. The intervention should be designed so that it offers minimal disruption to patient lives and professional routines. Third, the design of the telehealth intervention should address the issue of visibility. How it does this may depend on the condition and patient group involved. Monitoring systems can offer opportunities for visible feedback and prompts to actions that serve as reinforcements of behaviour change. This may be especially important if relationships are not fostered by the intervention (i.e. monitoring may be used to mitigate the loss of a relationship with a HCP). This strategy is likely to work best for some physical conditions and diseases such as diabetes and heart failure. For some conditions, notably mental health, visibility may have negative connotations, as patients may wish to remain anonymous when using the system. The design of the telehealth intervention should consider if and how symptoms and signs are made visible by the system and how these are responded to by the technology, the patient and the HCP.

Introduction

In this fast-moving field, the horizon-scanning element considered new developments worldwide that might inform the Healthlines study intervention development.

Methods

A team member, Simon Brownsell, drew on his wide knowledge of developments in this field. This was complemented by systematic internet searches. Some documents were available in confidence and a detailed report of findings was circulated to the research team in June 2010. A summary is presented here because the horizon-scanning exercise was relevant only for intervention development. The summary draws on the World Health Organization survey of e-health activity, analysis of the Veterans Association home telehealth service model and the extensive knowledge of Dr Simon Brownsell, an expert in the field [Research Fellow in Health Services Research at the School of Health and Related Research (ScHARR) at University of Sheffield].

Results

Conclusions

The conclusions were that few mainstream telehealth services exist. Therefore, the Healthlines study intervention would be important because it was designed in partnership with a NHS organisation (NHS Direct) and would be delivered by NHS staff, with the aim of ensuring that it would be easier to implement within the NHS context. It was also the case that NHS Direct was a national service with the potential to provide telehealth routinely throughout England.

Introduction

During 2013, an opportunity arose to explore in more depth the evidence about the effectiveness of telehealth interventions in the primary prevention of CVD in adult patients in community settings, as the subject of a master’s thesis undertaken by Dr Sam Merriel under the supervision of Professor Chris Salisbury, Chief Investigator for The Healthlines Study. This was too late to influence the intervention development, but we have included it here to offer a more complete picture of the evidence base that we will draw on in our discussion chapter.

There are a number of systematic reviews assessing the effect of telehealth interventions on individual CVD risk factors, particularly in the areas of hypertension,9,143,150,157 tobacco use158–161 and obesity. 162–165 However, risk factors for CVD should not be considered in isolation, since it is the combination of factors that determine a person’s risk of CVD. 166 Similarly, since many people with high CVD risk have multiple risk factors, it seems appropriate to design interventions addressing all of their modifiable risks, since that is how patients are managed in primary care, rather than focusing on only one of their risk factors. Neubeck et al. 167 conducted a systematic review of the effectiveness of telehealth interventions in the secondary prevention of CHD and showed some evidence of lower all-cause mortality and reduction in multiple CVD risk factors. However, at the time of this search in 2013, there appeared to be no previous reviews of the evidence for telehealth interventions to reduce overall CVD risk in disease-free individuals (primary prevention).

Methods

Data extraction and quality assessment

One researcher (SM) and a second reviewer assessed search hits against the eligibility criteria independently. The Cochrane Risk of Bias tool was selected to assess the methodological quality of the included papers. The quality assessment was performed at the individual study level only. Data were extracted by the two reviewers using a standardised form that was developed based on two guides for systematic reviews. 176,177 The form was trialled with a sample of selected papers to allow iterative refinement. Data were extracted from the published articles in the following areas: study characteristics, participants, intervention, primary outcome and secondary outcome(s). Results of the independent data extraction were directly compared to ensure accuracy of data collection. When insufficient data were reported to perform the meta-analysis, study authors were contacted to request the relevant data. Figure 2 shows the PRISMA diagram.

FIGURE 2.

The PRISMA diagram for CVD risk review study selection. This figure is reprinted from Preventive Medicine, vol. 64, Merriel SW, Andrews V, Salisbury C, Telehealth interventions for primary prevention of cardiovascular disease: a systematic review and meta-analysis, pp. 88–95, copyright 2015, with permission from Elsevier. 61

Results

Impact on overall cardiovascular disease risk

Four of the included studies measured overall CVD risk. Bove et al. 181 compared a nurse management CVD risk reduction programme augmented with telemedicine communication with nurse management alone, in medically underserved urban and rural communities. The mean baseline Framingham CVD risk was relatively high in both study groups (17.50 vs. 17.80), and after 12 months follow-up the intervention (–2.50, 95% CI –3.38 to –1.61; p < 0.05) and control (–2.70, 95% CI –3.74 to –1.68; p < 0.05) groups had a similar reduction in overall risk.

The study by Nolan et al. 182 assessed the effects of a telehealth protocol using motivational interviewing on CVD risk factors for patients with existing diabetes or CHD, a Framingham 10-year absolute risk for CHD > 20%, or two or more modifiable CVD risk factors. Subgroup analysis of the participants without a history of CHD showed significant small reductions in Framingham 10-year CVD event risk for both the exposed (–1.12 95% CI –0.36 to –1.88) and active control (–1.77 95% CI –0.89 to –2.65) groups, with no statistically significant difference between the groups in adjusted analyses (0.65 95% CI –0.53, 1.83 p = 0.28).

Wister et al. 48 also utilised a telehealth counselling approach to deliver a ‘Heart health report card system’. This study was the only one that showed a significantly larger reduction in Framingham risk score in the intervention arm when compared with the control arm (difference = –1.97, 95% CI –2.85 to –1.09 p = 0.002). Claes et al. 180 utilised the European Society of Cardiology SCORE charts to assess 10-year risk of CVD mortality. This study evaluated the effect of a Medical + Lifestyle Programme, consisting of a CVD risk profile and personalised lifestyle follow-up via a website, e-mail and telephone, to a Medical Programme comprising just the CVD risk profile. Both groups had a small reduction in mean overall risk (–0.002 and –0.004 respectively) and the difference between groups was not statistically significant (p = 0.33, CIs not reported).

Combining the data from the three studies that used the Framingham 10-year CVD risk score in a random-effects model showed no evidence for a reduction in CVD risk (SMD –0.35, 95% CI –1.97 to 1.27) in the telehealth intervention groups compared with controls. Random-effects modelling was chosen given the high level of study heterogeneity (Figure 3).

FIGURE 3.

Random-effects modelling of differences in Framingham 10-year CVD risk between intervention and control groups. IV, inverse variance.

Impact on modifiable CVD risk factors

Eight studies measured systolic blood pressure at baseline (Figure 4). Four studies found a significantly larger reduction in systolic blood pressure in the intervention groups when compared to the control groups. 179,181–183 The trial by Dekkers et al. 179 featured two telehealth treatment arms, one delivered via telephone and the other via the internet. Summarising all included studies using a random-effects model suggests that multi-focal telehealth interventions have a small effect on reducing systolic blood pressure (SMD –1.22 mmHg, 95% CI –2.80 to 0.35 mmHg).

FIGURE 4.

Random-effects modelling of differences in systolic blood pressure between intervention and control groups. IV, inverse variance. This figure is reprinted from Preventive Medicine, vol. 64, Merriel SW, Andrews V, Salisbury C, Telehealth interventions for primary prevention of cardiovascular disease: a systematic review and meta-analysis, pp. 88–95, copyright 2015, with permission from Elsevier. 61

Six studies assessed total cholesterol. Three of the trials demonstrated a small but significant reduction in total cholesterol when the telehealth intervention groups were compared with control groups. 48,179,183 Overall, random-effects modelling showed that these telehealth interventions reduced total cholesterol by 0.07 mmol/l (95% CI –0.19 to 0.06). Figure 5 gives the full meta-analysis results.

FIGURE 5.

Random-effects modelling of differences in total cholesterol between intervention and control groups. IV, inverse variance. This figure is reprinted from Preventive Medicine, vol. 64, Merriel SW, Andrews V, Salisbury C, Telehealth interventions for primary prevention of cardiovascular disease: a systematic review and meta-analysis, pp. 88–95, copyright 2015, with permission from Elsevier. 61

Four studies measured high-density lipoprotein (HDL) cholesterol. None of the individual studies demonstrated a larger change in the participants receiving telehealth interventions compared with their controls. Consequently, random-effects modelling (see Figure 6) calculated that these interventions did not affect HDL cholesterol levels (SMD –0.01 mmol/l, 95% CI –0.03 to 0.02).

FIGURE 6.

Random-effects modelling of differences in HDL cholesterol between intervention and control groups. IV, inverse variance. This figure is reprinted from Preventive Medicine, vol. 64, Merriel SW, Andrews V, Salisbury C, Telehealth interventions for primary prevention of cardiovascular disease: a systematic review and meta-analysis, pp. 88–95, copyright 2015, with permission from Elsevier. 61

Four trials measured participants’ smoking status at baseline. There was no significant reduction in the number of current smokers in any of the relevant included trials. Study heterogeneity was very low for smoking, so a fixed-effects model was employed (Figure 7). The OR for smoking after receiving a telehealth intervention compared with no intervention was 1.09 (95% CI 0.82 to 1.44).

FIGURE 7.

Fixed-effects modelling of odds of smoking in intervention and control groups. IV, inverse variance. This figure is reprinted from Preventive Medicine, vol. 64, Merriel SW, Andrews V, Salisbury C, Telehealth interventions for primary prevention of cardiovascular disease: a systematic review and meta-analysis, pp. 88–95, copyright 2015, with permission from Elsevier. 61

Meta-analyses and a summary of other modifiable CVD risk factors measured in the included trials was not performed due to lack of consistent inclusion in overall CVD risk score calculations, or the inability to convert reported outcomes into single measures for comparison and analysis.

Conclusions

We identified 13 trials that measured the impact of telehealth interventions on overall CVD risk and/or multiple CVD risk factors. Meta-analyses showed that these types of interventions have no effect on reducing overall CVD risk or cigarette smoking, or increasing HDL cholesterol. There was weak evidence of a small reduction in systolic blood pressure and total cholesterol. There are a number of possible reasons why this review failed to find any major effect for primary prevention telehealth interventions, given that some reviews focused on individual CVD risk factors (e.g. blood pressure) have shown more positive findings. It may be because of the quality of the included trials or the use of more appropriate methods when focusing on individual risk factors, or it could be that the effect of an intervention is diluted when addressing multiple CVD risk factors. The vast majority of studies included in this review had follow-up periods of 12 months or less, possibly affecting their ability to detect a true difference. Finally, it could be that small impacts on one risk factor intervention (e.g. to reduce blood pressure) do not translate to meaningful improvements in overall risk if other risk factors are not also improved.

This review had some limitations. The exclusion of non-randomised trials, unpublished studies, and articles not published in English could have introduced publication bias. There was no set minimum standard for study quality in assessing eligibility, resulting in the inclusion of some potentially low-quality studies. Some of these were inadequately analysed or reported, for example, not reporting 95% CIs or making between-group comparisons.

The implications for The Healthlines Study are that it is possible that the trial of an intervention may find evidence of an effect in reducing specific individual risk factors for CVD, but not overall CVD risk. The plan to assess cost-effectiveness will fill a current gap in the evidence.

Design

This was a cross-sectional postal survey.

Sampling and recruitment

General practices in two geographical areas of the UK, the south-west and the north-east, were invited to take part in the study. General practices were purposefully selected to represent a wide mix in terms of the socioeconomic characteristics of their patients. Between August 2010 and May 2011, a query was run on practice records to identify patients with either depression (aged ≥ 18 years, had consulted their doctor about a mental health issue and were prescribed an antidepressant medication within the last year) or raised risk of CVD (aged 40–74 years, QRISK2195 or Framingham196 10-year risk ≥ 20% and at least one modifiable risk factor, including hypertension, obesity or smoking). We calculated the QRISK2 score to assess CVD risk when possible but, as this score was not available through all general practice computer systems, we used Framingham risk scores in some practices. Patients were excluded if they were terminally ill, had cognitive impairment or had a severe mental health condition such as psychosis.

Fifty-four patients per practice from each of the two groups of eligible patients were selected using stratified random sampling. However, three practices had < 54 eligible patients with depression and so all eligible patients were selected from these practices. We sampled female and male patients in proportion to the number of eligible patients in each general practice. The CVD risk group was further stratified by age, such that equal proportions of young (aged 40–59 years) and older (60–74 years) participants were selected. This was because a CVD risk of ≥ 20% is more prevalent among older individuals, whereas access to technology is inversely associated with age. 190

Prior to invitation, GPs reviewed the patient lists and excluded any patients for whom it would be inappropriate to send a questionnaire (e.g. because of recent bereavement). The remaining patients were then mailed a letter by their general practice inviting them to take part in a study looking at new ways that the NHS could help people to improve their health, as well as a participant information leaflet and a questionnaire. The questionnaire could be completed and returned in the supplied prepaid envelope or there was the option to complete the questionnaire online or by telephone. A multi-language insert was included with instructions in 10 different languages on how to complete the questionnaire via telephone using a translator. Patients were asked to return a blank questionnaire if they did not want to take part. Those who did not respond were sent up to two postal reminders at approximately 2-weekly intervals. In an attempt to boost the rate of response, a ‘thank-you’ note with a teabag affixed to it was included in the first invitation questionnaire packages for the last 19 practices recruiting patients, representing half of the participating practices. All correspondence was sent by staff to the patients’ general practice and the researchers did not have access to patient-identifiable data at any point. Ethical approval was granted by the Southmead Research Ethics Committee (REC).

Sample size

Assuming an approximate 60% response rate, inviting 54 patients from each of 32 practices would provide around 960 respondents for each LTC group. This would provide 80% power to detect an absolute difference of ≤ 9.2% points in interest in using telehealth (binary outcome, equivalent OR ≤ 1.45), with a two-sided 5% alpha.

Measures

The questionnaire included questions about the key constructs that we hypothesised would predict interest in telehealth, namely sociodemographic variables, health needs, difficulties accessing health care, availability of and attitudes to technology and prior use of telehealth. To ensure the coherence of the questions included to assess these constructs and to reduce the questionnaire items to a smaller number of factors for data analysis, principal components analyses (PCAs) with orthogonal (varimax) rotation were carried out using Stata 11.2 (StataCorp LP, College Station, TX, USA) on constructed items. Decisions regarding the number of factors to extract were based on Kaiser’s criterion (eigenvalues > 1.0) by examining the scree plot and the subjective coherence of the factors. For each factor, items with an association of ≥ 0.3 were retained. 197 Next, the reliability of each factor was examined with Cronbach’s alpha, with coefficients > 0.70 indicating adequate reliability. Finally, mean summary scores for each reliable factor were calculated for individuals providing ratings for ≥ 50% of the relevant items. We treated each factor as a scale and labelled it according to the questions that it included.

Outcome variable

Interest in telehealth was assessed using questions about the participants’ interest in using a range of technologies. The item reduction techniques described above resulted in three summary scores for interest in telehealth, which related to interest in three types of technology: telephone (α = 0.82; landline or mobile telephone), e-mail/internet (α = 0.94; using e-mail or carrying out searches on the internet) and social media (α = 0.85; using chat rooms and social networking sites). These ‘interest’ summary scores were equal to the averaged sum of responses to three question items each (range 1–3), such that each corresponding summary score ranged from 1.0 (‘not at all interested’/’I don’t know what this is’) to 3.0 (‘very interested’), with scores of 2.0 equivalent to ‘fairly interested’.

Explanatory variables

Questions about sociodemographic characteristics of the respondents included those on sex, ethnicity,198 age group, employment status,199 educational qualifications200 and home ownership. 201 These questions were based on those used in previous validated surveys when possible.

Health needs were assessed using the Short Form questionnaire-12 items version 2 (SF-12v2). 202 Physical component summary (PCS) and mental component summary (MCS) scores were derived from the 12 items using proprietary scoring software (QualityMetric, Incorporated, Lincoln, RI, USA). These indexes of physical and mental health functioning are standardised with a mean of 50 and standard deviation (SD) of 10, such that lower scores indicate poorer health or greater needs.

The remainder of the questionnaire contained items constructed for the purposes of this research, although guided and informed by relevant literature and piloted with service users in advance. These are described below, with the specific questions that constituted each scale provided in Appendix 6.

Difficulty accessing health care was assessed using a series of questions that were based on themes identified through previous research. 203,204 Two ‘access difficulty’ summary scores resulted. Service delivery difficulties (seven items, α = 0.87) included questions about the convenience of accessing health care, as well as the nature or quality of the care itself (e.g. getting the right amount of care), whereas physical access difficulties (four items, α = 0.78) included questions about difficulties getting to appointments because of physical, psychological and transport problems, including costs. These summary scores ranged between 1.0 (‘no difficulty’) and 3.0 (‘lots of difficulty’).

Technology-related factors were assessed using questions on the availability of technologies and attitudes towards telehealth. Technology availability was assessed by asking respondents which of a range of technologies was easily available for them to use. Telephone availability (two items: landline, mobile) and e-mail/internet availability (two items: have e-mail address, have internet access) scores were formed by summing tallies (0 = absent, 1 = present) for these technologies.

Questions about attitudes towards telehealth were based on the theory of planned behaviour. 205 This theory suggests that perceived behavioural control – a concept capturing the extent to which one believes one is able to perform a behaviour – directly influences one’s intention to carry out a behaviour and may predict behaviour itself. Beliefs about one’s capability, which should be reflected in confidence levels, affect perceived behavioural control. Therefore, questions about confidence using different types of technology were devised. After item reduction, there were three clusters representing confidence in using telephone-based technologies (three items, α = 0.74), e-mail-/internet-based technologies (three items, α = 0.96) and social media-based technologies (three items, α = 0.88). Again, larger scores indicated greater technology confidence [range from 1.0 (‘not at all confident’/‘I have never tried this’/‘I don’t know what this is’) to 3.0 (‘extremely confident’)].

The theory of planned behaviour also states that positive or negative attitudes towards a behaviour predict one’s intention to perform that behaviour and are influenced by beliefs about the advantages and disadvantages of that behaviour. 205 Hence, items about the potential advantages and disadvantages of telehealth were generated based on previous qualitative research. 206 Summary scores for telehealth advantages (seven items, α = 0.87) and disadvantages (seven items, α = 0.90) were similarly formed [range from 1.0 (‘strongly disagree’) to 5.0 (‘strongly agree’)], with higher scores reflecting greater perceived advantages and disadvantages.

Finally, satisfaction with previous use of telehealth that was delivered by NHS Direct was evaluated by a single item. Respondents rated how satisfied they were with previous use of NHS Direct services on a scale from 1 (‘not at all’) to 5 (‘extremely’). At the time of the patient survey, NHS Direct provided health information and advice by telephone and through its interactive website throughout England. The rationale was that NHS Direct and other similar services could act as a provider of a wider range of telehealth services.

Patient and public involvement

To ensure that the questionnaire was as user-friendly as possible, we sought feedback from two service user groups. In both cases we asked volunteers to provide us with comments on the layout of the questionnaire and the wording used throughout, as well as whether or not the questions were clear and made sense. Service users could complete the questionnaire online or using a paper copy and could provide feedback by e-mail or telephone or as written comments. First, we contacted volunteers through the Mental Health Research Network, part of the National Institute for Health Research (NIHR), using a postal invitation letter. The other service user group was contacted via a general e-mail sent out from the service evaluation manager at NHS Direct. We received comments from a few members of both service user groups, which were collated and discussed among the research team. The questionnaire was then modified in accordance with this feedback and so the service user groups importantly contributed to the final version of the questionnaire used in this study.

Statistical analysis

We sought to examine two main questions in our analyses. First, we wanted to establish whether or not greater health needs were associated with greater difficulty accessing traditional health care. Then, the primary analysis investigated the extent to which interest in the use of telehealth was related to five key constructs: sociodemographic factors, health needs (including physical and mental health), access difficulties (including service delivery and physical access), technology-related factors (availability of technology and attitudes towards telehealth) and satisfaction with previous use of telehealth. We first used appropriate descriptive statistics (mean and SD or n and %) to summarise the sociodemographic characteristics of respondents and their needs, access difficulties, technology factors and interest in using telehealth. This included an exploration of how needs, health-care access and technology factors varied by age and LTC group. We then used multivariable regression models to examine (1) the relationship between health needs and access difficulties and (2) the associations between these variables and interest in telehealth, adjusting for the other variables in the model and taking into account the stratified survey design.

Response rate

Thirty-four general practices took part in the survey. GPs excluded 11.2% (201/1790) of patients with depression and 5.2% (96/1836) of the CVD risk group prior to mailing questionnaires. Of the 3329 patients sent a study questionnaire, 1478 (44.4%) returned it. The majority of questionnaires, regardless of study location or patient group, were returned by post (89%, 1310/1478), with 11% (164/1478) completed online and four questionnaires completed by telephone (< 1.00%, 4/1478, all English language). None of the patients made use of the translator service. The response rate was higher for patients with a high risk of CVD (50.1%, 872/1740) than for patients with depression (38.1%, 606/1589). Separate logistic regression analyses for the two patient groups revealed that response rates for both depression and CVD risk were higher in older people, whereas the likelihood of responding did not differ by respondent sex or location (Table 2). The inclusion of the thank-you note and accompanying teabag had no significant effect on rate of response for patients with depression (OR 1.14, 95% CI 0.87 to 1.48) or CVD risk patients (OR 1.31, 95% CI 0.97 to 1.77) and so this variable was excluded from further analyses.

Sample characteristics

Patients with CVD risk were older than those with depression [mean (SD) 61.9 (7.8) years vs. 49.1 (15.9) years], reflecting the inclusion criteria. Three-quarters of the depression group were female (452/606, 74.6%), whereas three-quarters of the CVD risk group were male (654/872, 75.0%). Both patient groups were predominantly white (depression 575/594, 96.8%; CVD risk 825/851, 96.9%), most patients were not currently employed (unemployed, studying, retired, etc.) (depression 317/597, 53.1%; CVD risk 498/861, 57.8%), only a minority had received higher education (depression 222/606, 36.6%; CVD risk 212/872, 24.3%) and the majority were home owners (depression 410/595, 68.9%; CVD risk 647/859, 75.3%).

Overview of health needs, access difficulties, technology-related factors and satisfaction with previous telehealth use

As expected, patients with CVD risk reported poorer physical than mental health whereas the reverse was true for patients with depression (Table 3). Whereas the reported physical health of patients with CVD risk was 0.5 SDs below the national average [UK mean (SD) 50.9 (9.4)], the reported mental health of patients with depression was more than 1.5 SDs below the national average [UK mean (SD) 50.9 (9.4)]. 207

Few patients reported access difficulties, with all summary scores approximating the ‘no difficulty’ response category (see Table 3). Despite these low mean summary scores, an important minority of participants indicated some difficulty in accessing health care and both patient groups were more likely to report having service delivery than physical access difficulties. For example, 27.9% (399/1432) of patients reported difficulties getting care when they need it most (service delivery) and 14.2% (206/1448) reported difficulties travelling to appointments because of their physical health (physical access).

Technology availability was high across both patient groups (see Table 3). Telephone technologies were more prevalent than computer-based technologies and markedly so for the patients with CVD risk. In fact, nearly all patients had access to telephone technologies. Across patient groups, age was associated only with the availability of computer-based technologies: 89.8% (115/128) of the two youngest age groups (18–44 years), 78.1% (400/512) of those aged 45–59 years and 60.5% (393/650) of those aged 60–74 years reported that they have these technologies readily available to use. It was only among the oldest, and proportionally smallest, age group (n = 49) that less than half of the respondents (n = 13, 26.5%) reported having easy access to computer technologies.

Technology confidence ratings were similar between patient groups but they varied somewhat across the technology types (see Table 3). In general, patients reported greatest confidence using telephone technologies, with mean summary scores approaching the ‘extremely confident’ response category and least confidence using social media technologies, with mean summary scores close to the ‘not at all confident’ response category. Respondents were ‘quite confident’ with regard to e-mail- and internet-based technologies.

Figure 8 shows the proportion of depression respondents reporting that they were confident using the various technologies across the different age groups. The pattern of findings was similar among the CVD risk group (Figure 9). Least associated with age were the telephone technologies, which received high confidence ratings by all age groups. The one exception was low confidence in text messaging by the oldest age group. Although confidence using e-mail/internet and social media technologies consistently decreased with age, more than half of the respondents in all age groups (except those aged > 75 years) reported confidence using e-mail/internet technologies. Conversely, confidence using social media technologies was strongly related to age, with only the younger age groups expressing high levels of confidence.

FIGURE 8.

Proportion of depression respondents within each age group reporting confidence using individual technologies. Source: This figure was adapted from Edwards et al. 185 © Louisa Edwards, Clare Thomas, Alison Gregory, Lucy Yardley, Alicia O’Cathain, Alan A Montgomery, Chris Salisbury. Originally published in the Journal of Medical Internet Research (www.jmir.org), 8 May 2014. This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on www.jmir.org/, as well as this copyright and license information must be included.

FIGURE 9.

Proportion of CVD risk respondents within each age group reporting confidence using individual technologies. Source: This figure was adapted from Edwards et al. 185 © Louisa Edwards, Clare Thomas, Alison Gregory, Lucy Yardley, Alicia O’Cathain, Alan A Montgomery, Chris Salisbury. Originally published in the Journal of Medical Internet Research (www.jmir.org), 8 May 2014. This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on www.jmir.org/, as well as this copyright and license information must be included.

Summary scores indicate similar levels of perceived advantages and disadvantages of using telehealth across patient groups (see Table 3). The most highly endorsed advantages were convenience and ability of telehealth to be delivered when and where one desires (Table 4). Dislike of non-face-to-face care and concerns over security issues emerged as the top disadvantages of telehealth (see Table 4). Of those respondents who had ever used NHS Direct (see Table 3), the majority were satisfied with that experience: 26.9% (157/583) were ‘moderately’ satisfied, 33.1% (193/583) were ‘quite a bit’ satisfied and 18.9% (110/583) were ‘extremely’ satisfied.

Health needs and health-care access difficulties

The results clearly showed support for the first examined relationship, with greater physical (PCS) and mental (MCS) health needs reliably associated with greater service delivery and physical access difficulties for both patient groups (Table 5) after adjusting for sociodemographic factors (sex, ethnicity, age group, employment status, higher education status and study location). Although sociodemographic factors were mostly non-influential, two out of seven characteristics emerged as consistently related to access difficulties for both groups: those who were unemployed (depression: β = – 0.107, 95% CI –0.204 to –0.009, p = 0.03; CVD risk: β = –0.041, 95% CI –0.066 to –0.016, p = 0.002) and those not owning a house (depression: β = –0.133, 95% CI –0.210 to –0.055, p = 0.001; CVD risk: β = –0.050, 95% CI –0.098 to –0.003, p = 0.04) reported greater physical access difficulties.

Overview of interest in using telehealth

Regardless of patient group, there was moderate interest in telephone technologies [depression mean (SD) 1.9 (0.7); CVD risk mean (SD) 1.7 (0.6)] and e-mail/internet technologies [depression mean (SD) 1.9 (0.7); CVD risk mean (SD) 1.7 (0.7)]; these mean summary scores approximate the ‘fairly interested’ response category. In contrast, there was very little interest in social media technologies [depression mean (SD) 1.3 (0.5); CVD risk mean (SD) 1.2 (0.4)].

Figure 10 shows which of the individual technologies the respondents were more or less interested in using, with ratings of interest dichotomised into ‘some’ and ‘no interest’ for ease of interpretation. This shows that patients with depression were more interested than those with CVD risk in nearly every form of technology for telehealth. There was a clear preference for the landline telephone (1072/1428, 75.1% overall), followed by finding information on the internet (876/1427, 61.4% overall). Again, there was hardly any interest in the social media technologies. Averaging across the technology types and both patient groups (depression and CVD risk), there was moderate interest in using telephone-based (854/1423, 60.0%) and e-mail-/internet-based (816/1425, 57.3%) telehealth, but very little interest in social media (243/1430, 17.0%).

FIGURE 10.

Proportion of respondents interested in individual telehealth technologies by patient group. This figure was adapted from Edwards et al. 185 © Louisa Edwards, Clare Thomas, Alison Gregory, Lucy Yardley, Alicia O’Cathain, Alan A Montgomery, Chris Salisbury. Originally published in the Journal of Medical Internet Research (www.jmir.org), 8 May 2014. This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on www.jmir.org/, as well as this copyright and license information must be included.

What factors are associated with interest in telehealth?

To address the main research question, sociodemographic factors, health needs, access difficulties, technology-related factors and satisfaction with past telehealth use were simultaneously regressed onto interest for each of the three telehealth mediums – telephone-, e-mail-/internet- and social media-based telehealth – separately for each patient group (depression and CVD risk). From these multivariable linear regression analyses, three variables were reliably related to interest in telehealth: greater technology confidence and perceiving both greater advantages and fewer disadvantages of telehealth (Tables 6–8). Moreover, these factors were consistently related to interest in each of the three telehealth mediums for both patient groups. Importantly, however, the technology confidence finding was modality specific. This means that greater telephone confidence was associated with greater telephone-based telehealth interest, greater e-mail/internet confidence was associated with greater e-mail-/internet-based telehealth interest and greater confidence using social media technologies was associated with greater interest in social media-based telehealth. Therefore, although confidence utilising a particular type of technology discriminated between interest in different modes of telehealth, perceiving greater benefits of and fewer drawbacks to using telehealth was uniformly related to greater interest. To aid interpretation of these multivariable regression results, Table 9 shows the unadjusted means and SDs for the level of interest in different forms of telehealth in relation to the explanatory variables (sociodemographic characteristics, health needs, access difficulties, technology-related factors, patient satisfaction).

Three other consistent findings emerged. First, for patients with depression but not those with CVD risk, greater difficulties with getting convenient, high-quality care (service delivery aspects of access) were related to more interest in telephone- and e-mail-/internet-based telehealth technologies. Second, as anticipated, greater satisfaction with previous use of NHS Direct was associated with heightened interest in future use of telephone-based telehealth and this was consistent across both patient groups. Third, there was more interest in e-mail-/internet-based and social media-based telehealth among those with CVD risk who were not home owners. Apart from these findings, the remaining variables in the model were unimportant to telehealth interest. After adjusting for the other variables in the model, health needs, access difficulties, technology availability and even sociodemographic factors did not reliably and consistently have an independent effect on interest in telehealth, either across patient groups or across more than one telehealth type within a patient group.

Summary of the main results

The results from this patient survey show that patients with two very different LTCs are interested in using telephone- and e-mail- or internet-based telehealth, but not telehealth accessed through social media websites. Although greater health needs were associated with greater difficulties accessing traditional health care in these patient groups, neither factor was consistently related to interest in telehealth. Interest in all three forms of telehealth appears to stem from the perceived advantages and disadvantages of telehealth, as well as confidence in using the relevant technology. This is significant because beliefs and levels of confidence are far more malleable than most of the other constructs included in this study, such as technology availability or socioeconomic status. It is also noteworthy that these other constructs were not consistently and independently associated with interest in telehealth. First, interest in telehealth was not reliably related to health needs. This suggests that willingness to use telehealth spans across those with good and poor health. Furthermore, it is not only those who have difficulty accessing traditional health care who are motivated to use telehealth; by and large, those with and without access difficulties were interested. Sociodemographic factors were generally not, in themselves, systematically related to telehealth interest. Therefore, older people were just as interested as their younger counterparts after adjusting for other factors, such as confidence in using the technology. Although availability of technology was quite high, this factor also did not consistently relate to interest in telehealth. The ramifications of these findings are important for policy makers, researchers, health professionals and patients alike.

Results in the context of existing research

We examined whether those with greater health needs or greater difficulties accessing traditional health care were more interested in using telehealth but found only weak evidence for this in patients with depression. Our results revealed an association between greater service delivery access difficulties (getting the right amount of care, from the right health professional, at the right time) and heightened interest in both telephone- and e-mail-/internet-based telehealth among patients with depression. This finding aligns well with one aim of telehealth treatments for depression, namely, overcoming barriers to care. Given the results of systematic reviews which showed that telephone-administered psychotherapy attrition rates80 are far lower than those of face-to-face care,208 this level of engagement may suggest that telehealth meets this aim to some extent. It is important to note, however, that respondents in our survey reported very few difficulties with health-care access and did not report especially great health needs, except for the mental health needs of those with depression. Overall, this restriction in range may have made it difficult to detect an effect of need or a more widespread effect of access difficulties on telehealth interest.

Sociodemographic factors were found to be relatively unimportant after adjusting for attitudes towards telehealth and availability of technology, which suggests that telehealth appeals to a broader demographic than young, educated and affluent patients. Although this runs contrary to some previous literature,190,191 it might be explained by the fact that more proximal variables, such as technology confidence and beliefs about telehealth, had not been included in previous research. Indeed, when similar behavioural or motivational factors are assessed, other research is consistent with our findings in demonstrating the integral role,209 or even superiority,210 of these constructs over demographic variables, albeit in terms of using the internet alone. Furthermore, a systematic review concluded that focusing on patient factors alone, as the majority of research in this area has done, is probably not comprehensive enough to understand patient interest in using telehealth. 189 We agree with this review that future research must cut across a broader spectrum of factors, especially those at the level of human–technology interaction, the health-care system and other social or normative influences.

Technology confidence is an example of a human–technology interaction variable and a key finding of this and other189 research is that confidence is consistently associated with interest in using telehealth. Whereas we asked respondents about technology confidence, it is interesting to note that other studies have used negative framing and asked about technology anxiety211,212 or difficulty using the internet. 210 Nonetheless, these studies also showed the equivalent association between lower technology anxiety and heightened interest in using telehealth.

We drew on the theory of planned behaviour205 when devising some of the items in the survey instrument, in particular technology confidence (akin to perceived behavioural control), perceived benefits and drawbacks of using telehealth (akin to attitudes towards the behaviour) and interest in using telehealth (akin to behavioural intention). Our findings do show that these predictors of behavioural intention – technology confidence and perceived advantages and disadvantages of telehealth – were the only two sets of variables that consistently related to interest in using the three mediums of telehealth, lending some support for the applicability of the theory of planned behaviour in assessing the potential uptake of telehealth. It is important to note, however, that our intention was not to assess the suitability of this theory and we did not include other constructs from the theory of planned behaviour (namely, subjective norm) in our analysis. Nonetheless, a meta-review of applications of the theory of planned behaviour to a variety of health-related behaviours (e.g. exercising, smoking, health screening checks) found that both attitudes and perceived behavioural control were the most important predictors of intention to perform a health-related behaviour. 213 More recently, a systematic review and meta-analysis of the effectiveness of theoretically based internet interventions in promoting health behaviour change found that larger effect sizes were obtained in studies that made greater use of theory in intervention development. 214 Of the social cognitive theories examined, the authors note that only those interventions that drew on the theory of planned behaviour resulted in considerably larger effects. Therefore, it is not surprising that a measure of patient acceptance of telehealth that incorporates the constructs from the theory of planned behaviour has now been devised. 215 This measure was completed by a group of patients with LTCs (heart failure or COPD) who were current users of telehealth and the results showed that the theory of planned behaviour constructs were associated with greater motivation to continue using telehealth. 216 Taken together, our results agree well with other research which shows that greater perceived behavioural control and more positive attitudes are related to greater patient interest in telehealth use.

Strengths and weaknesses

The main strength of this research is that it is a broad exploration of patient interest in several general forms of telehealth. The findings, therefore, are not limited to a specific intervention, but highlight some of the key elements that we must pay particular attention to when designing and implementing future telehealth initiatives. This is also the only study to our knowledge that has gathered ratings about interest in using a variety of forms of telehealth from patients who are the most likely recipients of this type of care in the future – those with LTCs. If telehealth is going to have an important role in effectively supporting patients with LTCs, then knowing what interests patients in taking up different forms of telehealth is an important first step.

The primary limitation of this study is the response rate of 44.4%, although this compares favourably with response rates in other community surveys. 217 It is possible that those who did not respond had different characteristics from those who chose to respond, which could have implications for the findings. With respect to telehealth, non-response bias by age is important given the widely held perception that older people do not like or use technology much. In this survey, responders were actually older than non-responders and yet a fair amount of interest in telehealth was reported. There was also considerable variation in patient health and sociodemographics, as well as the other key variables of interest. Nevertheless, these findings may not be generalisable to other LTC patient groups. A second limitation is that the telehealth interest ratings are based on questions about hypothetical and general technologies, rather than existing or specifically named telehealth services. Although this approach was directly in line with the purpose of the research – to inquire about future interest in services that could be delivered by existing health-care providers – it is difficult to know what types of applications respondents were thinking about when they gauged their interest in telehealth delivered through social networking websites, for instance. Moreover, it is likely that there is a relationship between how frequently technology is used and technology confidence210,212 and this relationship should be controlled for in future research. Finally, the large number of variables analysed raises the possibility of type 1 error because of multiple comparisons. Therefore, we have conservatively drawn attention only to findings that were consistent across several analyses.

Implications of the results

There are several interesting implications of the finding that telehealth interest is most strongly associated with technology confidence and perceived advantages and disadvantages of telehealth. First, it suggests that telehealth interest is likely to increase over time as the population, as a whole, becomes more familiar with and comfortable using different forms of technology. This may be particularly true of social media-based telehealth,218 the newest type of technology included in the survey and also the technology that respondents reported least confidence and interest in using. Following on from this, as our results revealed that technology confidence was modality specific, whereby confidence using one type of technology was related only to interest in using that same form of telehealth, it suggests that willingness to use telehealth is not restricted to patients who are confident using technology in general. Third, if patients were provided with adequate training and support in using telehealth equipment, they might be more interested in using telehealth. Finally, as some of telehealth’s advantages are realised and other disadvantages are dispelled through telehealth use, the strength of this effect may increase. There is good reason to expect such positive experiences of telehealth as the majority of telehealth research that asks about patient satisfaction does report fairly high levels of satisfaction. 219 Our study is no exception: high levels of satisfaction with past NHS Direct use – a form of telehealth – were reported. Positive experiences of telehealth may stimulate interest in using additional forms of telehealth, in an upwards spiralling effect.

Patients

Health-care professionals

Views from the practices

Trust and confidence in proposed new systems

It became apparent that trust, as the basis of engagement, was important on a number of levels. THS care managers emphasised that their work was based on a foundation of trust. This was built over time as a consequence of careful nurturing of the relationships between care managers and their patients. THS patients appreciated the informal nature of the communications between them and their care manager and care managers were often placed in the role of confidante.

The telephone and internet are mediums of communication that are both associated with suspicion rather than trust. Telephone subscribers have become accustomed to cold calling and telesales and, likewise, the internet is associated not only with advertising but also with fraud, identity theft and the dissemination of misinformation. Any new telehealth initiative will need to allay anxieties and suspicions associated with its means of delivery. The idea that a new intervention should be personally endorsed by GPs and practice nurses seems to be a simple and practical suggestion in this regard.

It was very clear from the interviews that patients trust the NHS, NHS Direct, NHS Choices and internet sites that are linked to NHS sites. Patients felt reassured that NHS services would not be commercial and would be neutral sources of information. The idea of NHS Direct supporting the management of LTCs was positively viewed by patients. In contrast, HCPs’ views of NHS Direct tended to be negative. This clearly has implications for the branding of any new service. For patients, the inclusion of the name ‘NHS’ will be positive and will inspire confidence, whereas for many HCPs the association of a new service with NHS Direct is likely to be a deterrent.

Linked to trust is confidence that a proposed new service will actually work. Patients may need the endorsement of primary HCPs to begin the process of engagement with new services. HCPs need to be convinced that (1) there is evidence of efficacy of telehealth interventions and that telehealth will add something to patient care; (2) telehealth is cost-effective and will be compatible with the existing QOF scoring system; and (3) telehealth will not simply duplicate services that are already available. On this last point, it is important to note that practice nurses undertake much of the work associated with support of patients with LTCs. Telehealth interventions should therefore be orientated to the existing work of practice nurses. There is a potential for practice nurses to feel that areas of their work will be duplicated or replaced by a new telehealth service. In the changing landscape of the NHS, GPs will be the commissioners of telehealth services, illuminating questions about efficacy, cost-effectiveness and duplication.

An element for building trust, and one that was lacking in the THS operation, involves fostering clear channels of communication between staff in general practices and those running a telehealth intervention. At THS, the lack of such communication led to frustration on the part of care managers and to a mutual lack of appreciation and co-ordination of the work being carried out to support patients with LTCs. This was true both in the practices and at THS. The aims and contents of a potential telehealth intervention need to be apparent to those working in general practices, alongside clear channels of communication to avoid duplication and the giving of mixed messages to patients.

The exemplar conditions and areas of need

The clear message from both HCPs and patients was that support for patients with depression is a pressing need. Patients voiced anxiety about being on drug treatment for depression and felt that they needed time and non-drug therapy. There are long waiting lists for talking therapies and variability of access in different areas. Many interviewees were positive about the idea of delivering psychological therapies by telephone or via the internet, with the caveat that some personalised contact be included. It was clear in the case of the delivery of a package of care over the internet that both patients and HCPs thought that some kind of human interaction was necessary, whether this be face-to-face, over the telephone, using a webcam, using webchat or by e-mail.

The picture was less clear for those at risk of CVD and this was connected to the complex nature of understanding what it means to be ‘at risk’. Nevertheless, telehealth has the potential to promote lifestyle change. Diet and weight loss were identified by some patients as areas where support would be welcomed. HCPs also mentioned this and pointed to the lack of support for alcohol dependency, in contrast to a plethora of initiatives in the area of smoking cessation.

The promotion of lifestyle change risks encroaching on areas of life that many regard as being personal, such as dietary habits, daily routine and alcohol consumption. The idea of being ‘nagged’, ‘lectured’ or ‘nannied’ was not welcomed, but it was also identified as the basis of approaches and information currently on offer. THS patients pointed out that their care managers did not nag them and that they appreciated this. THS care managers were trained in motivational interviewing techniques and consciously avoided a lecturing style in their attempts to promote lifestyle change.

The importance of context and targeting

Telehealth has the potential to reach and support patients whose access to conventional health services may be difficult or limited. According to the Office for National Statistics, most households in the UK (88%) have a telephone and most (73%) have an internet connection (the most recent figures at the time of writing refer to 2010). 249 However, at the time of the study there was a disparity between the highest decile income group, with 97% accessing the internet, and the lowest decile income group, in which only 69% have an internet connection. 249 The Office for National Statistics reported in 2010 that 60% of those aged > 65 years had never accessed the internet, although this is rapidly changing. 249 These data correspond with the interview material cited here, with older patients and those living in less affluent areas not routinely using the internet. Thus, accessibility to the medium through which telehealth is to be delivered has to be an important consideration when developing a new intervention.

In addition to thinking about the means of delivering telehealth, it is also relevant to consider that the burden of LTCs and care and support for LTCs is not experienced uniformly. The patients and HCPs interviewed for this study raised a variety of context-specific issues. For those living far from metropolitan centres, telehealth offers the possibility of increased access to specialist services, such as psychological therapy. Likewise, telehealth can also increase access to care and support for isolated frail and elderly patients and the current THS service appeared to be particularly well received by older patients. However, the health-care challenges facing residents within metropolitan areas can vary dramatically from one residential area to the next. HCPs in some of the THS practices described problems of drug and alcohol dependency in areas where there was also a high dependency on income support. Incentives for maintaining claims to illness are shaped and reinforced by the current benefits system, thus providing a particular challenge for health-care interventions aimed at engaging people with an approach that runs counter to one demanding illness behaviour.

Within the Midlands and north-east of England, there are pockets of the population that are moderately and extremely affluent and THS care managers observed that many of their patients living in these areas had engaged particularly well with the service. They also observed a much lower level of uptake of the service and engagement with the service among the large South Asian populations in the same areas. Language and cultural sensitivities are relevant factors to be considered in terms of engaging ethnic minority patients with LTCs in new telehealth interventions.

A ‘one-size-fits-all’ telehealth intervention is unlikely to suit the needs of all patients when considering the complex and context specificity of lifestyle change. If the aim of a potential intervention is to address health inequalities, it needs to be sufficiently flexible to be tailored and targeted at particular areas of need.

Evidence of benefits

In Chapter 1, we highlighted that, although the potential benefits of telehealth in chronic condition or LTC management have been rehearsed for at least 20 years, evidence to support these arguments is limited. 54,251,252 Systematic reviews have been conducted in specific chronic conditions, along with overviews that have combined findings from a range of conditions, and these have concluded that the evidence in favour of telehealth is weak and inconsistent. 11,20,23,54,69,74,251,253 Evidence of effectiveness is stronger for some conditions (e.g. heart failure) than it is for others (e.g. diabetes). Some studies report positive findings whereas others do not and it has been difficult to identify a pattern in terms of disease, type of technology or patient characteristics to explain these inconsistencies. There is a lack of evidence about mechanisms of action and about wider impacts of telehealth on the utilisation of other health-care services. 251 There is inconsistent reporting of outcomes, suggesting a lack of clarity about the intended benefits of telehealth and making it difficult to compare studies. Evidence about cost-effectiveness, or of successful large-scale implementation, is particularly limited.

The need for a conceptual model

Telehealth is a complex intervention252,254 involving a number of interacting components, such as the type of technology, the infrastructure, the human support available and the capabilities of the patient in relation to the technology. For any individual, telehealth is likely to be only one facet of the health care that they receive, so telehealth cannot be understood in isolation from the health-care system in which it is provided.

Over the last 15 years there has been increasing awareness of the importance of theory both in the development and in the evaluation of complex interventions. 56 Theory is needed to understand the relationship between context, mechanism of action and intended outcomes, but this has largely been neglected in the field of telehealth. 255–257 Although there are well-recognised theories in related topics, such as behaviour change (e.g. the theory of planned behaviour,205 the behaviour change wheel,258 the model of Ritterbrand et al. 259) and why technologies get used (e.g. the technology acceptance model260), there is no overarching theory that connects these and other elements (such as co-ordination between service providers) essential to LTC management in the context of telehealth.

What is needed is a clear conceptual model for how and why a telehealth intervention for patients with LTCs is intended to have specified beneficial effects. Making explicit the theoretical chain of causation by which an intervention is intended to lead to its effects focuses attention on the most important features of the intervention that need to be delivered for it to be effective. A conceptual model also provides a framework for evaluation by identifying the contextual factors, steps in the causal chain and most important outcomes that need to be assessed.

A useful theoretical model is one that is simple, clear and based on meaningful concepts, demonstrating lines of causation (including whether variables are proposed to have moderating or mediating effects), and which is sufficiently general to be tested in a range of settings. A model that applies to only one context is of little practical value as it has little explanatory value – the power of a model depends on the extent to which the relationships it proposes are demonstrable in a range of contexts and patient groups. In the context of telehealth for LTCs, this means that the model should ideally be applicable to a range of conditions, types of interventions and health-care settings.

This chapter describes the development of a conceptual model for the role of telehealth in the management of LTCs. This was developed to inform the design of an intervention to support people with two exemplar conditions: depression or raised CVD risk (because of risk factors such as hypertension, smoking, obesity and hyperlipidaemia). These exemplars were chosen to represent very different types of conditions, which would test the generalisability of the model but which are both common and for which there is existing evidence that some forms of telehealth could be effective. 16,69 By taking into account the views of patients and providers, and considerations about cost as well as evidence of effectiveness, the intention was to develop a model for interventions that are likely to be suitable for implementation at wide scale, acceptable to stakeholders and cost-effective.

Evidence review

The model was based on several sources of evidence. The methods and results for each strand of evidence are summarised below, but are described in more detail elsewhere in this report.

• A meta-review and realist synthesis of existing quantitative and qualitative evidence on telehealth for LTCs. 253 This consisted of an overview of existing systematic reviews of telehealth interventions. We focused on reviews of LTCs generally, rather than in relation to specific conditions. Details of the methods are provided in Chapter 2. We supplemented the meta-review with a review of depression, as well as a new systematic review to look in more detail at studies of telehealth interventions focused on those for the prevention of CVD. 61 In addition, we identified and reviewed published qualitative studies of patients’ experiences of using telehealth interventions. In total, we included 16 systematic reviews (representing 662 quantitative studies) and 29 qualitative studies. We combined these sources of data in a realist synthesis in which we sought to identify mechanisms of action for telehealth in LTCs. Realist synthesis is an approach to reviewing research evidence on complex interventions in order to provide an explanatory analysis for how and why they work (or do not work) in particular contexts or settings. 261

• A survey of patients to assess relationships between patient characteristics, health needs, difficulties with access to health care, attitudes towards and availability of various technologies and interest in using different types of telehealth (see Chapter 3). 185

• A qualitative study of the potential role of telehealth in LTCs. 220 This involved interviews and observation with patients, doctors and nurses providing primary care for patients with LTCs and health information advisors (HIAs) who provided an existing telephone-based health coaching and care management service for patients with LTCs such as heart failure or diabetes. 156 Further details are provided in Chapter 4.

• Comparison with other models of chronic disease management. To take account of and compare our emerging conceptual model with existing models and frameworks, we familiarised ourselves with other widely used models of chronic condition management, particularly (but not exclusively) those relating to the use of telehealth. We wanted to identify common factors in these models that appeared to be associated with improved care and benefits for patients.

• Analysis of national guidelines. To apply the model to our exemplar conditions, we identified the main recommendations and priorities for treatment from the current UK guidelines and compared these with guidelines from the USA and Europe. We cross-referenced these recommendations with our meta-review to identify evidence for the effectiveness of telehealth interventions (e.g. the use of online programmes to deliver CBT for depression; the use of home monitoring of blood pressure in patients with hypertension).

Synthesis

We synthesised the findings from our evidence review in two stages. First, it was clear from the meta-review and the qualitative study that engagement from both patients and professionals appeared to be key to the success of a telehealth intervention. We therefore used a modified PRECEDE–PROCEED262 approach to intervention development in which we used the insights from our evidence sources to map the predisposing, enabling and reinforcing factors that determine engagement with telehealth, creating separate ‘maps’ for patients and health professionals. Predisposing factors provide the motivation to act in some way; enabling factors are those that make it possible to carry out the action; and reinforcing factors influence the likelihood that one will perform the behaviour in the future, based on positive or negative feedback. Through discussion within the research team, we listed and grouped themes from the literature reviews, qualitative research and patient survey, cross-referenced to the sources of evidence. Next, commonalities across these three sources of evidence were highlighted and key themes relating to engagement with telehealth were identified. These key themes were then independently organised into predisposing, enabling and reinforcing factors by members of the research team familiar with the PRECEDE–PROCEED262 definitions. As it is possible that the same information can serve first as a predisposing factor and then later as a reinforcing factor, differences in classification, although rare, were resolved through discussion. Nonetheless, the real importance of classifying information into these types of causal factors was to devise temporally appropriate strategies to enhance motivators of and mitigate barriers to the target behaviour.

Second, we developed a draft model for the use of telehealth to support the management of LTCs that encapsulated the main findings from the evidence review. We discussed the findings from the various studies within the research team, seeking to describe hypothesised relationships between different constructs in a schematic manner. Several different layouts and versions of the model were discussed iteratively in meetings of the research team as we critiqued and sought to improve the model. Finally, we convened an intensive 1-day workshop for a wide range of stakeholders (n = 38), including patients, care providers, managers, commissioners of services, independent academics and the research team. We presented the findings of the evidence review and the draft model to the stakeholders, who discussed it in small groups and provided feedback. We used this to refine the final model, which we labelled the TECH model. The term ‘chronic disease’ rather than ‘long-term condition’ was used as this is more widely used internationally and would enhance the international applicability of the model.

Using the model to design an intervention

The research team used the TECH conceptual model to design a telehealth intervention, which we called the Healthlines Service. In this chapter we describe the Healthlines Service model to support the management of patients with raised CVD risk, but a similar approach was used to develop an intervention for patients with depression. This was designed to be delivered by NHS Direct, which, at the time that the intervention was designed, provided health information and advice throughout England based on a network of telephone call centres and an associated website.

The intention was to design an intervention that would be likely to be cost-effective by maximising patient benefit at minimum cost and that could feasibly be rolled out quickly on a national scale if it proved to be effective. For these reasons, the design of the intervention sought to incorporate technologies that were already available and approaches for which there was already some evidence of effectiveness. We avoided cutting-edge technologies that were not already developed or tested and high-cost solutions that would be unlikely to be widely available or deliverable to large numbers of patients. To maximise population benefit, the aim was to focus on the large number of patients at moderate risk of health problems (e.g. patients with hypertension and other CVD risk factors), rather than the small number of patients at high risk (e.g. patients who have already had a stroke).

The research team used the patient and health professional ‘maps’ generated through the PRECEDE–PROCEED method to develop strategies to promote engagement with the telehealth intervention by addressing each of the predisposing, enabling and reinforcing factors previously identified.

The model as a framework for evaluation

The TECH conceptual model was used to provide a framework for evaluation by describing the extent to which each element of the model was successfully delivered and the intended outcomes were achieved. The Healthlines Service was evaluated within two pragmatic parallel RCTs and accompanying process and economic evaluations. Details of the methods for these trials are provided in Chapter 7. In brief, we recruited 43 general practices providing primary health care in three areas of England. Adult patients from these practices with either (1) depression or (2) a raised risk of a first cardiovascular event (10-year risk ≥ 20%) were recruited to take part and were individually randomised to receive either usual primary care plus extra support from the Healthlines Service or usual primary care alone.

Evidence review

Engagement of patients and primary care providers

The literature meta-review highlighted that many telehealth interventions have been unsuccessful because of low uptake by patients and high rates of dropout. Both the patient survey and our qualitative research illustrated the range of factors that act as motivators or barriers to patients using telehealth, summarised in Box 4, based on our PRECEDE–PROCEED map of predisposing, enabling and reinforcing factors for patients.

With regard to HCPs, our qualitative research indicated that many were unenthusiastic and, in some cases, resistant towards telehealth interventions. Our PRECEDE–PROCEED map for professionals identified several factors that were likely to influence engagement in telehealth. These included the belief that medicine should be evidence based and scepticism about the evidence for telehealth (predisposing factor), concerns about duplication of care (predisposing factor), the need for technology to be simple and reliable (enabling factor) and the importance of clarity of roles for conventional and telehealth providers and good communication between them (reinforcing factor).

The engagement of patients and professionals is likely to be symbiotic – doctors are unlikely to be supportive if their patients are not interested in using telehealth and patients are less likely to be interested if their doctors are dismissive of it. Therefore, a key priority for a telehealth intervention is to develop strategies to maximise the likelihood that patients will want and be able to use it.

Effective long-term disease management

Management provided through telehealth must be effective, with clear evidence that it is associated with improved patient outcomes. Our evidence synthesis and review of existing models of LTC management suggested that strategies that contribute to effective care and which could be delivered using telehealth can be summarised under three headings: promoting self-management, optimising treatment and care co-ordination. The various strategies that make up each of these headings are shown in Box 5, along with citations for specific studies or reviews that provide evidence of effectiveness for each element (not necessarily in the field of telehealth).

Partnership

Our qualitative research highlighted that a telehealth intervention is just one aspect of the health care provided to a patient with a LTC. These patients are likely to continue to get the majority of their care from their family practitioner, with whom they may have had a long-term relationship and whom they will continue to consult for reasons apart from their LTC. In addition, many patients with LTCs are likely to be receiving help from hospital specialists and other health- and social-care agencies.

However, our evidence review suggested that many previous telehealth interventions appear to have failed because they were designed in isolation from the rest of the health-care system, leading to duplication of effort, lack of co-ordination between providers, inefficiency and confusion for patients. This is likely to reinforce the resistance expressed by other health-care providers. Our qualitative research showed that these other providers may perceive the telehealth intervention to be an unnecessary interference in their area of responsibility, possibly representing a threat to their future role.

Therefore, it is important for a model for telehealth interventions to emphasise that telehealth should be delivered in partnership, identifying the role that telehealth can play to support, rather than compete with, patients’ main primary health-care providers.

Context: characteristics of patients and wider social and health system

The patient survey and the literature review both indicated that the characteristics of patients are likely to have an impact on how telehealth affects outcomes. These include sociodemographic characteristics, particularly age, the nature of their LTC and the severity of their condition. The design of a telehealth intervention must also take account of the wider social and health system context. 222,282 For example, a programme designed to work within a health system context with a strong primary care foundation may need different features from one designed for a system in which patients consult different hospital specialists for each of their LTCs. Similarly, a system which assumes that patients have access to a fast and reliable internet connection will not work when this does not apply. Context is not only important in designing interventions, but also in interpreting evaluation. An intervention may have different effects in different contexts, but this does not necessarily undermine the theoretical basis for the intervention. Therefore, the telehealth LTCs model may be generalisable even if a specific intervention is not. Finally, different funding models for health care create different financial incentives for providers and patients, which may have a major influence on how telehealth systems are implemented.

Principal findings

This chapter describes the development of the TECH conceptual model for the effective use of telehealth among patients with LTCs and illustrates how it has been used to develop telehealth interventions for patients with either depression or raised risk of CVD, as well as design the evaluation of those interventions. If these evaluations in different LTCs are positive, this will provide support for the model with regard to how this type of telehealth intervention works, suggesting that it can then be applied to other LTCs.

Alternatively, if the intervention is unsuccessful, it will be possible to assess each of the processes in the hypothesised causal chain to determine if the intervention was not delivered as intended or if the assumed causal relationships were incorrect. For example, the model posits that one way in which telehealth works is by allowing people to monitor their own health, which will lead to changes in their behaviour and a positive impact on their health. Having a model highlights the need to assess the extent to which participants actually did self-monitor as intended, whether or not this was associated with behaviour change and whether or not this led to improved health outcomes. This kind of approach provides a framework for correction and adaptation of an intervention through understanding which intervention components are more or less effective at impacting proximal outcomes in the causal chain. 284

Strengths and limitations

The strength of this research is that we have used diverse sources of evidence to develop a conceptual model that creates a framework for intervention development and evaluation. The model suggests that, if patients engage with the intervention, and if the intervention promotes self-management that leads to behaviour change, along with optimised treatment in line with evidence-based guidelines, this will lead to improved health both now and in the future. Better co-ordination of services will reduce duplication and inefficiency, which may lead to more cost-effective care and may improve access to care. Each of the components of the model can be justified from our own research and evidence from previous literature.

Although it is arguable that the TECH model could be applicable not only to telehealth but also to all long-term disease management programmes, the model draws attention to topics that are particularly important for telehealth (such as the need for partnership with primary care providers and attention to patient engagement) but which have been neglected in many previous telehealth interventions.

Recognising that the simplest models have the greatest utility, we sought to provide a simple graphical depiction of the hypothesised causal chain in a successful telehealth intervention. However, we recognise that the model diagram oversimplifies the multiple potential mechanisms by which a telehealth intervention may have its effect. There are likely to be associations and interactions between different elements of the model and both recognised and unrecognised confounding factors. However, to indicate all of these potential relationships in the model would, in our view, reduce its usefulness in providing a framework.

A further limitation is that the strength of underlying evidence to support each of the components of the model is variable. For example, evidence of the benefit of patient self-monitoring is strong for some LTCs but not all and, although providing patient information and shared decision making are viewed as important aspects of LTC management in the CCM and other similar models, the evidence that these strategies lead to improved patient outcomes is limited. Nevertheless, we have sought to include components in the model for which the overall weight of evidence supports their value.

Relationship to previous studies

There are several existing models of behaviour change based on psychological theory that have been applied to, or are relevant to, telehealth. 205,258,259 However, behaviour change is only one aspect of the TECH model and this is not its main purpose. The TECH model is intended to provide a framework for the design and evaluation of telehealth services at scale within health-care systems, taking into account a much wider range of factors, such as the potential efficiencies gained through better co-ordination of services.

Several previous authors have argued for the importance of theory in designing telehealth interventions from a range of perspectives255,280,285 and there are also existing frameworks for the assessment (rather than the design) of telehealth for chronic conditions, such as the model for assessment of telemedicine (MAST). 286

The intervention that is most relevant to our study and well described in terms of its underlying theoretical basis is the Comprehensive Health Enhancement Support System (CHESS), an umbrella term for several e-health programmes combining information, adherence strategies, decision-making tools and support services. 287 Pingree et al. 284 used the example of CHESS to demonstrate how telehealth interventions are likely to be multifaceted and to depend on a complex interplay between patients, clinicians and the health-care system. Like the Healthlines intervention described here, CHESS was developed by combining several intervention features, each of which had some theoretical justification. However, CHESS was developed without any clear theory about how the programme features related to each other284 and the TECH model underpinning the Healthlines intervention is intended to address this limitation.

Catwell and Sheikh280 claimed that a telehealth programme should be viewed as an intervention in the same way as a new drug and that it should be evaluated in a similar way, while recognising the complexities of achieving this. Baker et al. 285 have highlighted the challenges of evaluating e-health because of the rapid state of change, but emphasise the need to focus on timeless universal principles and to develop models to elucidate treatment mechanisms. Kaplan255 has argued that the evaluation of telehealth interventions requires methodological pluralism to understand the range of social and organisational factors that determine how telehealth is implemented, used and has effects.

Greenhalgh et al. 288 have taken a more radical stance and argued against the quasi-experimental approach advocated by previous authors in favour of in-depth case studies, viewing programme evaluation not as experimentation but as social practice. They claim that there is a need to recognise the complex political dynamics and language games practised by different stakeholders and to question rationalist assumptions about ‘what works’. We need to ‘transcend . . . the linear and deterministic metaphors (“research into practice”) implying that the core task is determining what works, after which implementing an eHealth technology will be a straightforward exercise in project management’ (p. 558). 288

We recognise the importance of these political considerations in terms of how telehealth programmes are implemented and evaluated and in terms of how the findings from such evaluations are sometimes interpreted to fulfil a previous agenda. However, this does not undermine the need to develop interventions based on a understanding of how and in what ways telehealth programmes might be effective; indeed, a clear theoretical basis for interventions and clarity about intended outcomes might provide the most robust defence against selective use of findings and may allow a more nuanced understanding about why interventions are more or less effective in different contexts.

Implications for clinicians and policy makers

This chapter describes a clear conceptual model, based on several sources of evidence, that helps to articulate the theoretical basis for how, why and under what circumstances telehealth could provide specified benefits for patients with LTCs. Because it is based on evidence-based components and the views of stakeholders, the TECH model provides the basis for the design of telehealth interventions that are likely to be effective, cost-effective and acceptable to patients and health-care providers. Importantly, it also provides a framework for the evaluation of these interventions.

Depression trial

In the depression trial, the key resource that participants were offered was access to the Living Life to the Full Interactive (LLTTFi) programme, supported by regular telephone contact. LLTTFi is an online interactive multimedia programme that delivers CBT-based treatment for depression (or comorbid depression and anxiety; see Appendix 11). It was developed by Professor Chris Williams of the University of Glasgow, UK. 290,291 It involves six self-directed sessions, completed approximately every 2 weeks. The programme was offered to Healthlines participants either as an online interactive programme provided by Media Innovations Ltd [see www.llttfi.com/ (accessed 27 September 2016)] or, for those who preferred it, by giving them the corresponding Overcoming Depression and Low Mood: a Five Areas Approach workbook to use at home. 292 As part of the CBT package, participants completed the PHQ-9 (a nine-item questionnaire that assesses depression severity and response to treatment)289 at the start of the programme, midway through and then after the final session. This provided clinical data to monitor participants’ well-being, track progress through the CBT programme and assess whether or not participants were experiencing suicidal ideation. Moreover, these PHQ-9 scores were shared with the participants’ GPs, to keep GPs updated about their patients’ progress and health (see Appendix 12).

It is important to note that the intervention was not solely an online CBT programme, but also included a number of other intervention components in line with the TECH model. First, the self-directed CBT package was crucially supplemented with telephone support, which included motivational interviewing. Our research in the first phase of this research programme (see Chapters 2 and 4) indicated that online CBT alone would be less effective, less engaging and less valued by patients with depression if it did not also contain supportive, human contact. Therefore, the core of the depression intervention was regular telephone calls from a HIA, which were designed to motivate and support participants in making use of the mainly online resources and materials. Participants with depression received seven telephone calls at approximately fortnightly intervals over the course of 4 months and then two further contacts were made once every 3 months over the course of 6 months (see Appendix 13). Although the length of the calls and the topics discussed varied somewhat across the encounters, they each lasted around 15–20 minutes, with a mean duration of 18.5 minutes. The telephone scripts used during each contact were written by the NHS Direct Clinical Content Team and Professor Chris Salisbury and incorporated protocols for providing support to participants in using and applying LLTTFi or the book equivalent. In addition, they included modules covering monitoring of depression symptoms, optimising and stepping up treatment in cases of inadequate response, medication adherence, exercise and alcohol use (see Appendix 13).

Participants in the depression trial had access to the Healthlines web portal. Other than information about depression and links to support services, organisations and trusted online information resources, the web portal also included a link to the LLTTFi programme. Another resource offered through the web portal was Big White Wall (BWW) [see www.bigwhitewall.com (accessed 12 September 2016)]. BWW is a UK-based digital mental health network that is equipped with a 24/7 clinically moderated online forum involving other BWW members, a facility for posting ‘bricks’ on the wall (a form of creative self-expression using graphics or text displayed within a rectangular brick shape), information resources and several validated self-assessment tools [e.g. Generalised Anxiety Disorder-7 items (GAD-7)293 and PHQ-9289] that track scores and offer more information and pragmatic advice (see Appendix 14).

Cardiovascular disease risk trial

In the CVD risk trial, the software and telephone scripts were adapted from effective interventions developed by Professor Hayden Bosworth and colleagues at Duke University (Durham, NC, USA). 294,295 The content of these scripts was based on recognised behaviour change principles such as stimulus control, problem solving, cognitive restructuring and goal-setting. 271,272 The content and language of the original interventions were modified when appropriate to make them consistent with UK treatment guidelines and the NHS environment and to incorporate additional components written by the NHS Direct Clinical Content Team and Professor Chris Salisbury. The software guided the delivery of a series of modules during 12 telephone contacts approximately 4 weeks apart over a 12-month period, with each call lasting around 18 minutes. The modules covered a wide variety of topics associated with the management of raised CVD risk. The software was designed so that the intervention staff covered topics that were relevant to the participants during each telephone contact and the content of later encounters reflected the topics that were covered in earlier calls (see Appendix 15). Each telephone encounter included discussion of several topics (modules within the software) and some modules were covered repeatedly (e.g. weight loss was revised regularly in people trying to lose weight). The modules provided were:

• knowledge about cardiovascular risk and healthy lifestyles

• medication and side effects review

• blood pressure medication optimisation

• home blood pressure monitoring

• statin medication review

• support for medication adherence

• smoking and nicotine replacement therapy

• diet

• weight loss and orlistat use

• alcohol use

• exercise.

The complete set of modules was not necessarily discussed with each participant. The intervention was tailored to each participant, such that the total number of modules covered partly depended on the particular CVD modifiable risk factors (high blood pressure, obesity or currently smoking) that a participant had, as well as the particular lifestyle changes the participant was interested in working on with his or her HIA. As will be covered in more detail in Chapter 7, and among other criteria, CVD risk participants were required to have one or more of these modifiable risk factors to be eligible to join the CVD risk trial. Therefore, if a participant did not smoke or was not obese, the smoking module or weight loss module, respectively, would not apply and would be skipped. Likewise, if a participant smoked but clearly indicated that he or she was not ready or willing to give up smoking, the HIA would not cover the smoking module. In that case, the HIA would not persist with the issue at that time, but return to it gently at a later date to see if the participant was now ready to contemplate quitting smoking. The rationale was that participants were likely to be at different stages of readiness to change,296 but through participating in the Healthlines Service – a lifestyle and behaviour change telehealth intervention with motivational interviewing – and developing rapport with their HIA, they might move into a state of readiness over time.

A key aspect of the CVD risk intervention was support for home blood pressure monitoring. Participants with a systolic blood pressure of ≥ 140 mmHg who did not have atrial fibrillation were offered an OMRON M3 home blood pressure monitor (OMRON Healthcare UK Ltd, Milton Keynes, UK) to take home and use for the duration of the 12-month study, if they wished to do so. This device is a basic arm cuff monitor, validated for home use by the European Society of Hypertension International Protocol. 297 Participants were given instructions on how and when to record their blood pressure and how to enter their readings on the Healthlines web portal. Participants were requested to take blood pressure readings twice daily for the first week and weekly thereafter. The web portal calculated average readings over the previous 6 days initially (if the participant had entered at least four readings) and then over the previous 6 weeks thereafter. After entering a set of readings, participants were automatically advised by the web portal on whether or not their blood pressure was above their target, when to take their blood pressure again and what to do if their blood pressure was too high or too low. Target blood pressure was based on UK NICE guidelines,268 but there was the option for a patient’s GP to modify the target if desired. The target blood pressure levels for different groups of participants were as follows: < 135/85 mmHg for participants aged < 80 years and without diabetes, < 125/75 mmHg for participants with type 1 diabetes (or for those with chronic kidney disease and type 1 or 2 diabetes) and < 135/75 mmHg for participants with type 2 diabetes. Regardless, the HIA reviewed each participant’s blood pressure readings during each monthly call. If the readings were above target, the HIA advised the participant to see his or her GP and also e-mailed the GP a letter asking the GP to review the participant’s treatment (see Appendix 16). Reinforcing the evidence-based nature of the intervention, the letter to the GP included a copy of the relevant NICE guidelines268 about recommended steps for intensifying treatment. A copy of this letter was available to the participant via the web portal.

Additional theoretical components incorporated into the intervention

In addition to the regular telephone calls and support for using resources available online, the intervention incorporated several other features designed to address components of the underlying theoretical model. For example, one recognised problem with previous telehealth interventions is low patient engagement and so intervention mapping was used to identify features that would act as barriers to or facilitators of patient engagement (see Chapter 5, Engagement of patients and primary care providers). As a result, the intervention was designed to address perceived barriers and highlight factors that would facilitate and reinforce engagement. One factor discovered during the developmental work that could enhance patient engagement was the importance of receiving consistent support from a named person, rather than from an impersonal call centre. Therefore, participants were offered follow-up calls from the same HIA as much as possible, knew the name of their HIA and had access to colour photographs of the individual HIAs.

Previous telehealth interventions have also had problems related to engagement of primary care clinicians. 220 The present research aimed to address this problem by framing the intervention as evidence based and complementary to primary care and by working with primary care clinicians, rather than duplicating their efforts or competing with them. Accordingly, letters were regularly e-mailed to the GPs about each participant’s progress during the course of the intervention, such as PHQ-9 scores (depression) or average monthly blood pressure readings (CVD risk). Letters making recommendations about treatment were also shared with participants via the web portal, which further reinforced participant engagement and could make it more likely that recommendations were acted on. Another problem with participant engagement experienced in other telehealth interventions has been lack of ability to use the software. Accordingly, the HIAs spent time in initial encounters helping participants to log in to the Healthlines portal and to gain confidence in using the software.

Intervention staff

As already mentioned, the Healthlines intervention was delivered by a team of NHS Direct HIAs. These staff were existing NHS Direct employees based within an existing NHS Direct call centre in Nottingham, UK. All of the HIAs had a minimum of a diploma-level qualification or equivalent experience in a health-care or social-care setting. They were also experienced with working for NHS Direct as HIAs, providing expert health and medication information. It is important to note that the HIAs were not nurses.

The HIAs received 3 weeks of training prior to delivering the Healthlines intervention. This training included classroom sessions, hands-on training and side-by-side call reviews with expert training staff. In addition, it covered health coaching skills, training on how to introduce and support the online or workbook CBT package for depression (delivered by Professor Chris Williams, who developed the CBT package) and how to use the CVD risk intervention software, as well as condition-specific knowledge and medicines training, with this being delivered by expert pharmacists. Therefore, the same HIAs were trained to deliver both the depression and the CVD risk content and they had access to clinical advice and management support as required throughout. Staff performance was also monitored using a rigorous call review process.

Initially, the intervention team consisted of six HIAs (two male, four female). However, about 6 months after implementing the Healthlines Service, which was an intensive period of participant recruitment, an additional six HIAs (one male, five female) were trained to deliver the intervention and allocated to the service. At this time, the service was operating under a ‘blended’ delivery model whereby the HIAs partly worked on the Healthlines Service and partly retained their role as call handlers within NHS Direct. With 12 HIAs on board who all worked part-time, it became challenging for one HIA to regularly maintain contact with the same participants throughout. Therefore, the decision was made to switch to a more ‘dedicated’ model whereby the intervention team was reduced to four HIAs (all female) who worked solely on the Healthlines Service and maintained regular contact with the same participants as much as possible. This change was fully implemented by August 2013, approximately 1 year after the service was first deployed.

Hours and mode of operation

To improve accessibility of care to participants, the Healthlines Service was available for extended hours: Monday to Friday from 1000 to 2000 and on Saturday from 1000 to 1400. At the outset, the service provided an initial assessment with participants, which was then followed by regular telephone contacts at prearranged appointment times that were convenient for the participants. When possible, all contacts with an individual participant were made by the same staff member to promote continuity of care and build rapport. The work of intervention staff was supported by computer software, which was used to guide the content of the telephone calls and create a record of what was discussed. In addition, participants were able to request a call back between their scheduled appointments via the intervention website or by leaving a telephone voicemail message.

To address early problems with participants forgetting their prearranged appointment with their HIA, automatic e-mail reminders were introduced and proved successful. Nevertheless, if a participant did not answer the HIA’s scheduled call, the HIA would reattempt telephone contact on six more occasions. This was later reduced to a more manageable three telephone attempts, followed by an e-mail requesting that the participant re-establish contact with the Healthlines Service if he or she wished. If the participant did not contact the service within approximately the next month after being sent this e-mail request, he or she was considered no longer interested in receiving support and classified as having discontinued the intervention.

Intervention completion

As detailed above, participants could remain in receipt of the intervention for up to 12 months. The final participant contact included a review of progress, information about other ongoing sources of support and creation of a final summary to be sent to their GP. On completion of the intervention, participants returned to the sole care of their GP. This transition was clearly communicated to both the participants and their GPs.

Closure of NHS Direct and transfer of the Healthlines Service to Solent NHS Trust

In October 2013, the government announced its intention to close NHS Direct on 31 March 2014. Following this announcement, negotiations were initiated between NHS Direct and Solent NHS Trust regarding the transfer of the study. Solent NHS Trust was the most suitable potential host because it was near to the NHS Direct site in Southampton where most of the NHS Direct staff involved in managing the research were based. Although the transfer of the research side of the study was agreed quite quickly, it was more difficult to finalise the transfer of the intervention to Solent NHS Trust. The complicating factors were the transfer of four intervention delivery staff (based in Nottingham) and a manager (based in Southampton), as well as making arrangements for ongoing clinical supervision, support for the IT infrastructure used to deliver the intervention and ensuring that contracts and finances were quickly organised. Further, as the building that the HIAs worked out of was no longer available with the closure of NHS Direct, the HIAs agreed to work from home for the remainder of the intervention period. Therefore, laptops had to be configured and mobile phones arranged, at the same time as transferring the IT platform from NHS Direct to NHS Solent Trust’s network. To allow for this IT work to be carried out and tested, it was necessary to institute a temporary break in intervention delivery between 1 April and 6 May 2014. Further details around the transfer of the service to Solent NHS Trust, as well as the impact on the intervention delivery schedule and participants, are described in Chapter 7 (see Transfer of the service from NHS Direct to Solent NHS Trust and Impact on intervention encounters).

Outline of the trial design

This phase of the programme involved two linked, multicentre, parallel-group, two-arm randomised superiority trials, with individual participants allocated on a 1 : 1 basis. The trials involved two different patient groups: those with depression and those with a raised CVD risk. The study infrastructure, including participating general practices, research staff and intervention staff, and the underlying theoretical basis for the intervention were common across both trials. However, the specific content of the intervention packages, data analysis and reporting were distinct. Once again, we refer to these trials as the depression trial and the CVD risk trial.

Figure 12 provides a flow chart summarising participant recruitment and follow-up procedures for each of the trials. The first two general practices recruited into the study formed part of a run-in phase, during which study procedures were tested and refined before patients were recruited from the remaining practices.

FIGURE 12.

Flow chart of participant recruitment procedures. CIS-R, Clinical Interview Schedule – Revised; DEP, depression. Source: Adapted from Thomas et al. 283 © Thomas et al. ; licensee BioMed Central Ltd. 2014. This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Aim and objectives of the trials

The aim of the trials was to determine the clinical effectiveness and cost-effectiveness of a NHS-delivered telehealth intervention to support patients with two exemplar LTCs: depression and raised CVD risk. Specifically, the study addressed the following research questions:

1. Does the intervention, in addition to usual primary care, improve condition-specific clinical outcomes compared with usual care alone?

2. Does the intervention have any effect on other patient outcomes, including quality of life and satisfaction with care?

3. What is the cost-effectiveness of the intervention in each condition?

4. What is the acceptability of and compliance with the intervention and what are the facilitators of and barriers to its delivery?

Settings

Forty-three general practices in the environs of Bristol (n = 20), Sheffield (n = 17) and Southampton (n = 6) took part in the study. Because we reached and surpassed our recruitment target for the CVD risk trial before invitations were sent to patients in the last practice (in the Southampton area), we invited patients to take part only in the depression trial in that practice and only 42 general practices were involved in the CVD risk trial. The 43 general practices taking part in the trials were recruited to represent a mix of sociodemographic characteristics, including urban and rural patient populations. Two practices participated in the run-in phase and the remainder were involved in the main trials. The two practices in the run-in phase were treated as an internal pilot and their patients were included in the main trial analysis as there were no changes in procedures between the run-in and the main trial phases.

Inclusion criteria

In both trials participants were required to have access to a telephone (landline or mobile), access to the internet and an e-mail address for personal use. It was not necessary for participants to have the internet in their own home but they had to have easy and available access to it. Therefore, it was permissible for participants to use the internet in a public place, such as the library or an internet café, or to use it at a friend’s or family member’s house, should they agree to this.

Additional inclusion criteria for the depression trial were:

• aged ≥ 18 years on the date of invitation to participate

• confirmed diagnosis of depression using the Clinical Interview Schedule – Revised (CIS-R) scale298

• PHQ-9 score of ≥ 10. 289

Additional inclusion criteria for the CVD risk trial were:

• aged between 40 and 74 years on the date of invitation to participate (the age range for which CVD risk scores have been validated and the same age range used for the NHS Health Checks vascular screening programme)

• 10-year risk of a cardiovascular event of ≥ 20%, calculated using the QRISK2 score195

• at least one of the following modifiable risk factors: (1) current systolic blood pressure ≥ 140 mmHg (and suitability for home blood pressure monitoring, if this was the patient’s only modifiable risk factor); (2) BMI ≥ 30 kg/m²; and (3) currently smoking.

Patients who met the inclusion criteria for both the depression and the CVD risk trials were invited to take part in the depression trial only. Satisfying all inclusion criteria was essential to meet eligibility for the study.

Exclusion criteria

Exclusion criteria applying to both trials were:

• bipolar disorder

• psychotic illness

• dementia or substantial cognitive impairment

• severe learning disability

• substance dependency (including alcohol)

• receiving palliative care

• significant suicide risk

• GP’s determination that participation would cause distress (e.g. because of recent bereavement)

• inability to communicate verbally in English sufficiently to receive telephone-based support delivered in English; patients who could communicate verbally in English but who were unable to read English were deemed eligible if they had a family member or friend who was willing and able to translate written materials, such as information sheets, consent forms and online material, for them.

Additional exclusion criteria for the depression trial were:

• currently receiving case management from a specialist mental health worker

• currently receiving face-to-face, telephone or computerised CBT or similar psychotherapy

• given birth in the previous 12 months.

Patients were excluded from the study if any of the exclusion criteria were met.

Additional exclusion criteria for the CVD risk trial were:

• established diagnosis of CVD, defined as history of heart attack, angina, heart failure, stroke or transient ischaemic attack (TIA)

• currently pregnant or planning to become pregnant within the next 12 months

• patients who will be invited to participate in the NHS Health Checks Programme during the period of the trial (only relevant in the Bristol recruitment area, in which the local PCT requested this)

• patients with atrial fibrillation for whom high blood pressure was their only modifiable risk factor, as these patients were not suitable for home blood pressure monitoring.

Recruitment procedures

Trial run-in phase

In the run-in phase of the trial recruitment commenced 1 month ahead of that in the main trials, with the patient searches starting in May 2012 and the first participants randomised into the trial in July 2012 (depression trial) and December 2012 (CVD risk trial). Two general practices were involved and each aimed to recruit 20 patients to the depression trial and 20 to the CVD risk trial. The purposes of the run-in phase were to test the study recruitment and follow-up procedures and to allow any required adjustments to be made before the main trials. These steps also allowed intervention staff the opportunity to develop their skills in using the intervention software and treatment protocols. To provide intervention staff with a larger number of participants to work with during the run-in phase, allocation in this phase was at a ratio of 3 : 1 in favour of the intervention group. Data collection, intervention schedules and intervention content were the same for run-in phase participants as in the main trials and so data provided by these participants were included in the final trial analysis. No substantial changes were made between the run-in phase and the main trials, other than increasing the number of patients sent an initial study invitation, as described earlier.

Commencement of the depression trial and delay to the cardiovascular disease risk trial

As covered in more detail in Chapter 6, after participants were randomised into the Healthlines Service group by the research team, it was anticipated that within the next 1–2 weeks the HIAs would have completed the initial assessment encounter with them. This was usually the case for the depression participants, but there was an unexpected delay to the start of the CVD risk trial. Although the first randomised depression participants began receiving the intervention by August 2012, the corresponding group of CVD risk participants was not randomised into the study until early December 2012. The delay was partly due to the IT infrastructure around the CVD patient management software. This required more work to adapt it to the NHS context and there were difficulties integrating the Duke software with the software used normally used within NHS Direct.

We believed that the IT infrastructure was in place by early December 2012 and began to randomise the CVD risk participants, but it then became apparent that there were several problems with the telephone scripts to support patients with high blood pressure, particularly patients who were not already taking blood pressure medication or who had not yet picked up their blood pressure monitors. To remedy this, revision to the scripts was required, which meant altering the patient management software.

The IT changes were not fully integrated until March 2013. In the meantime, the intervention progressed as planned from December 2012 for all other CVD risk patients not affected by the two blood pressure issues described above. Likewise, as soon as those patients already taking blood pressure medication had collected their home blood pressure monitor, they could be referred to the intervention team for their initial assessment telephone call. Nonetheless, some CVD risk patients were randomised and then experienced a significant delay of 2–3 months before they began receiving the intervention. We describe the effect of this on intervention delivery in the final subsection of this chapter.

Transfer of the service from NHS Direct to Solent NHS Trust

As covered in Chapter 6, we negotiated to transfer the study and intervention from NHS Direct to NHS Solent Trust following the closure of NHS Direct on 31 March 2014. To summarise, the main challenges in transferring the intervention across NHS organisations involved the reassignment of intervention staff and their manager in two different locations, making arrangements for ongoing clinical supervision, organising support for the IT infrastructure used to deliver the intervention and ensuring that contracts and finances were quickly organised. In terms of the main IT issues, laptops had to be configured and mobile telephones arranged for the HIAs who had agreed to work from home for the remainder of the study, at the same time as transferring the IT platform from NHS Direct to NHS Solent Trust’s network. To allow for this IT work to be carried out and tested, it was necessary to institute a temporary break in the intervention delivery between 1 April and 6 May 2014.

In preparation for the break in service delivery, all ongoing intervention participants and participating general practices were notified by e-mail that the NHS provider of the intervention was changing, but that other aspects of the intervention, including the staff delivering it, would remain unchanged. The possibility of a delay in some scheduled appointments and the new web portal address and telephone call-back number were all included in this notification. The web portal and resources available to participants otherwise remained the same. Ongoing control group participants were also notified by e-mail of the change of lead NHS organisation for the research.

At most, we estimate that the temporary break affected 39% (120/307) of depression and 67% (218/325) of CVD risk participants, the maximum number of participants remaining in the service at the time, but some of these participants would not have had a scheduled call during the service closure period. For those participants who did already have a scheduled telephone appointment during the temporary closure period, these appointments had to be rescheduled for after the break. As a result, some patients were expected to have a slightly longer period between some appointments for a time and then a slightly shortened period between other appointments thereafter. This adjustment to the usual call schedule was required to fit in their remaining appointments in the time available. Nevertheless, we did not anticipate that this would have a major effect on the total number of appointments that participants were due to receive over the entire study period, except for some patients whose 12-month follow-up period ended during the break.

As planned, the HIAs resumed patient telephone calls from 6 May 2014. However, because of technical difficulties surrounding the transfer of the IT infrastructure, the patient management software was not fully functional until 9 June 2014. As the patient management software included access to the telephone scripts and the facility for recording details around the telephone calls, including updating notes around the patients’ health, the intervention could not be delivered as planned until 9 June 2014. Likewise, CVD risk patients were unable to enter blood pressure readings on the web portal between 1 April and 16 June 2014. As these patients had also been supplied with a blood pressure recording booklet, they were able to keep track of their weekly blood pressure readings and enter these retrospectively when the web portal regained functionality after 16 June.

To carry out the scheduled telephone appointments between 6 May and 9 June 2014, and to do so without access to the patient management software, the research team developed a paper-based call log to record the details around calls and devised a ‘holding script’ for the HIAs to deliver. The ‘holding script’ explained about the unfortunate IT difficulties and asked how patients were getting on with the various resources (e.g. CBT package, blood pressure monitoring) they were using as part of the intervention, how they were doing in terms of their health needs and goals and whether or not they had any particular issues that they wanted to raise. Unless this was the final intervention call, the HIA then booked participants in for a further call, which would cover the next encounter that they were due to complete. If this appointment was the final intervention call at the end of the 12-month intervention period, the HIA thanked the participants for taking part and explained that they would return to the sole care of their GP from now on. Overall, the feedback from participants was that they liked the ‘holding script’ and it was a positive experience to have this kind of informal review. The details recorded in the paper-based call logs were later entered into the patient management software.

Impact on intervention encounters

The final research assessment for both depression and CVD risk participants was due 12 months after participants were randomised into the study. It was intended that participants’ intervention sessions would be delivered over the same 12-month period. However, as outlined in the previous two sections above, there were some delays between recruitment and the initial intervention session and delays in scheduling regular intervention calls, for example because of holidays or because participants had been difficult to contact or because of the necessary IT ‘down time’ as the service transferred from NHS Direct to Solent NHS Trust. This meant that many participants did not receive the full number of intervention sessions by the time that they were due for their final research assessment.

To remedy the situation and attempt to maximise the number of sessions that participants received ahead of their 12-month assessment, the frequency between intervention calls was reduced for all remaining participants from January 2014 as follows:

• For depression participants, the fortnightly call schedule continued for encounters 1–8, but encounters 9 and 10 were scheduled at 6-weekly intervals instead of the 2- to 3-monthly intervals that were originally planned.

• For CVD risk participants, encounters were scheduled every 3 weeks instead of monthly. If the date that occurred 3 weeks ahead was not possible (e.g. because of holiday), HIAs were encouraged to first try further reducing the appointment interval to 2 weeks, before considering dates ≥ 4 weeks ahead.

In addition to those procedures described above, the research team arranged one final scheduled session with the HIA for participants who had not received all of the intervention sessions by the 12-month research assessment, if participants wished. GPs were informed by e-mail when their patients completed the intervention.

Subsidiary studies

Within the Healthlines study, four subsidiary studies were embedded within the two RCTs. The first was an attempt to increase study uptake among invited participants in both trials by enhancing the participant information sheet, with the remaining three studies attempting to increase response rates to the follow-up questionnaire among participants in the depression trial alone.

First, the MRC Systematic Techniques for Assisting Recruitment to Trials (MRC START) study is funded by the MRC Methodology Research Programme. The purpose of the study is to develop the conceptual, methodological and logistical framework for nested studies and to assess their feasibility. 303 The Healthlines study acted as a host trial for one of the MRC START recruitment interventions. In particular, the aim was to test the impact of an enhanced patient information sheet and invitation letter on recruitment rates. 304 Through reduced use of jargon, simpler sentences and an altered layout of the materials, the MRC START materials sought to improve readability and ease of comprehension compared with the standard versions. The substudy was carried out in the final three general practices at the Bristol site only, with participants in both the depression and the CVD risk trials randomly allocated on a 1 : 1 basis to receive the optimised or standard study materials. After generating the list of participants to invite to the Healthlines study, it was sorted by a randomly generated number and the first half of the list was assigned the optimised study materials. There were no other differences in the information provided to these participants or the research procedures. The results of this MRC START substudy have now been published. 305

Next, as difficulties with motivation and concentration are at the heart of depression, this is a difficult patient group to keep engaged in research, especially over the course of a 12-month study. Therefore, we attempted three different methods to increase response rates with these participants at follow-up. This consisted of three individually nested studies within the depression trial alone. These studies aimed to investigate the impact on follow-up questionnaire response rates of pre-calling participants ahead of sending out the follow-up questionnaire or not (study 1), including a colour photograph of the research team with the questionnaire cover letter/e-mail or not (study 2) and using an urgent action subject line (‘ACTION REQUIRED’) in the e-mail reminder compared with one that simply states that this was a reminder (study 3). These three ideas were generated following discussion within the research team and in light of previous studies306 which suggested that such changes may have an effect on follow-up rates. Study 1 occurred partway through the 8-month follow-up time point, with participants in the Bristol location alternately allocated to receive the intervention or not. Based on an early and informal indication that there was a benefit from pre-calling these participants at 8 months’ follow-up, this procedure was adopted universally for all final 12-month follow-up appointments. Therefore, studies 2 and 3 included the manipulation from study 1 as an adopted standard procedure but examined the possibility of further boosting the 12-month follow-up response rates with two separate manipulations with participants from the different study locations, with the photo manipulation in study 2 conducted with all participants from the Bristol study centre only and the e-mail subject line manipulation in study 3 including respondents from the Sheffield and Southampton centres. Based on a random number generated using Microsoft Excel, participants in studies 2 and 3 were randomised on a 1 : 1 basis to receive the intervention using simple randomisation. Details of the methods and results of these studies have been published. 307

Outcome measurement

Serious adverse events

Finally, serious adverse events (SAEs) in participants were monitored throughout the study. A SAE was defined as an untoward occurrence that:

• resulted in death

• was life-threatening

• required hospitalisation or prolongation of existing hospitalisation

• resulted in persistent or significant disability or incapacity

• consisted of a congenital anomaly or birth defect or

• was otherwise considered medically significant by the investigator.

Under the terms of the Standard Operating Procedures for RECs, any SAE that was considered ‘related’ and ‘unexpected’ had to be reported to the REC. 319 ‘Related’ was defined as resulting from the administration of any research procedures (e.g. as a result of the Healthlines Service intervention, completion of follow-up questionnaires, participation in the process evaluation) and ‘unexpected’ was defined as a type of event not listed in the protocol as an expected occurrence.

Data collection and follow-up

Data were collected from self-report questionnaires, primary care medical records, the Healthlines Service records and clinical measurement. Information was collected at baseline from all participants but the timing of follow-up data collection differed between the two trials (as detailed below). To maximise response rates, participants were offered the choice of completing the questionnaires on a secure survey website or on a paper copy sent in the post. The PHQ-9, which has been validated for telephone use, could also be completed by telephone with one of the researchers. The health economics quality of life measure (EQ-5D-5L) was not available for online completion because of licensing and technical restrictions and thus was completed on paper by all participants at each of the time points. If participants had opted to complete the questionnaire online, it was explained that the EQ-5D-5L would also be posted to them for completion. Up to five reminders, which were communicated by e-mail, telephone (twice) and post (twice in the depression trial, once in the CVD risk trial), were used if participants did not return their questionnaires. To maximise collection of the primary outcome, the second postal reminder in the depression trial contained the PHQ-9 alone. Likewise, participants in the depression trial were provided with the option to complete the PHQ-9 over the telephone in the second telephone reminder.

In the depression trial, self-report questionnaires were completed at four time points: at baseline and at 4, 8 and 12 months after randomisation. The PHQ-9 questionnaire was completed at each of these time points. At baseline it was completed as part of an initial screening questionnaire by post; at all other time points it was included with the follow-up questionnaire and therefore could be completed online or by post.

In the CVD risk trial, self-report questionnaires were completed at three time points: at baseline and at 6 and 12 months after randomisation. At each of these time points, participants were telephoned to ask them to attend an appointment with a nurse or HCA at their own general practice. As described in Primary outcomes, practice staff collected CVD risk factor information, including carrying out a blood test to measure cholesterol (at baseline and 12 months only), blood pressure, weight and smoking status (all time points). This information was used to calculate each participant’s QRISK2 score.

Statistical considerations

Research governance

Management, operational and academic aspects of the trials were the responsibility of the Trial Management Group, which met approximately every 6 weeks from the start of the trial until the main analyses were completed. Conduct of the trials was overseen by an independent Trial Steering Committee and an independent Data Monitoring Committee. The whole research programme was also overseen by the Programme Management Group, which included all of the co-applicants on the programme grant and two patient and public representatives. The University of Bristol was the study sponsor and the study was approved by the National Research Ethics Service Committee South West–Frenchay (reference no. 12/SW/0009). The trials were registered under Current Controlled Trials and allocated an ISRCTN as follows: depression trial ISRCTN14172341 and CVD risk trial ISRCTN27508731.

Patient and public involvement

Two patient and public representatives attended, and were active contributors at, the Programme Management Group and the independent Trial Steering Committee meetings throughout this phase of the research programme. Both representatives were recruited from a pool of NHS Direct volunteers. In addition to directly providing valuable feedback on the study questionnaires and recruiting additional members of the lay public to review these documents, the representatives provided feedback on study methods and procedures and the development of the dissemination plan and strategy for the trials.

Introduction to methods used for economic evaluation

The aim of the economic evaluation was to determine the cost-effectiveness of a telehealth intervention provided within an existing NHS organisation for LTC management in the two exemplar conditions of depression and raised CVD risk. The economic evaluation was conducted alongside the two RCTs.

Perspective

Estimates of direct and indirect costs incurred were calculated from the perspectives of (1) health- and social-care providers [the NHS and Personal Social Services (PSS)], (2) personal costs incurred by participants and (3) productivity losses because of time off work. Each type of cost was recorded and valued separately and costs were measured from randomisation until the completion of 12 months of follow-up.

Form of analysis

Costs were related to outcomes using the following forms of economic evaluation. A cost–consequences analysis was used to compare costs with a range of outcomes. In the depression trial these were response to treatment (defined as a PHQ-9 score of < 10 and a reduction of at least 5 points 4 months after randomisation); change in PHQ-9 depression score; EQ-5D-5L327 score at the 12-month follow-up; and quality-adjusted life-years (QALYs). In the CVD risk trial the outcomes included response to treatment (defined as maintenance of or reduction in 10-year risk at 12 months after randomisation); change in QRISK2 score; EQ-5D-5L score at the 12-month follow-up; and QALYs.

Two forms of cost-effectiveness analysis were undertaken for each trial. The first compared incremental NHS and PSS costs with the proportion of participants responding to treatment. The second analysis related incremental NHS and PSS costs to incremental QALYs, which permitted the cost-effectiveness of the interventions to be compared with estimates of society’s cost-effectiveness threshold (i.e. the opportunity costs that would result from health foregone elsewhere in the NHS if the Healthlines interventions were funded). This analysis used values for the cost-effectiveness threshold of £20,000–30,000, as suggested by NICE. 328

As the primary outcome of the CVD risk trial (risk of a CVD event) is not a clinical outcome, and the 12-month period of trial follow-up may not capture all relevant impacts of the intervention on clinical outcomes, we used cohort simulation modelling techniques to estimate the longer-term effect of the intervention on cost per QALY over the lifetime of CVD risk trial participants. This modelling is described in Introduction to long-term cardiovascular disease risk modelling.

Sources of data: outcomes

The sources for the clinical outcomes used in the cost–consequences analyses have been described previously in this chapter (see Primary outcomes). In the depression trial, participants reported responses on the EQ-5D-5L instrument at baseline and 4, 8 and 12 months after randomisation. In the CVD risk trial, responses were reported at baseline and 6 and 12 months. In the absence of a valuation set specific to the EQ-5D-5L at the time that this analysis was conducted, these responses were valued using the EuroQol-5 Dimensions, three-level version (EQ-5D-3L) value sets using the ‘cross-walk’ values available from EuroQol. 329

Quality-adjusted life-years were estimated using the area under the curve method. Imbalance between the intervention and usual care group values of the EQ-5D-5L at baseline were adjusted for in the cost-effectiveness analysis using the methods described in Manca et al. 330

Sources of data: resource use, personal costs and productivity

Valuation of resources

The principal sources used in the valuation of resources were national reference cost sources and questionnaire responses. When information was not available from these sources, estimates were taken from published estimates and other literature. These sources are detailed in the text below and in Appendices 22 and 23. Costs are reported in 2012/13 UK pounds.

Costs of the Healthlines interventions

The elements of resource use that contributed to the costs of the interventions are summarised in Table 15. The principal cost involved in providing the interventions was the salary of the HIAs. All HIAs were remunerated at band 4 of the NHS salary scale. 331 Salary levels and related costs (such as on-costs and capital-related costs) were estimated using the methods of Curtis;332 these were adjusted for a 40-hour working week, to reflect the actual working hours of HIAs, and an appropriate estimate of non-contact time, to estimate the cost per hour of participant contact time.

Non-contact time was estimated from anonymised HIA work diaries relating to the period 1 May to 30 June 2013. These 2 months were considered to be representative of workload and case mix during the period for which the intervention was operating. The diary entries classified the working days of each HIA on duty into the following categories: encounter completed, encounter incomplete/DNA (did not attend, meaning that the participant did not answer the telephone), paid break, unpaid lunch, appointment booked, meeting, auditing, administration, mentoring and training and sickness. The categories of encounter completed and encounter incomplete/DNA were taken to be participant contact time and the remainder were assumed to be non-contact time. The ratio of contact to non-contact time was estimated to be 2 : 1 based on an analysis of the diary data. The HIAs indicated that this ratio did not differ between the two participant groups (depression and CVD risk) and hence the basic cost per hour of participant time was estimated to be the same in both trials.

In addition to the basic cost per hour of participant time, differences between the costs of serving the two groups arose when considering the training of the HIAs, licences purchased and consumables. HIAs received initial and ongoing training from a nurse-grade trainer, the cost of which was incorporated into the cost per hour of HIA time. The initial training was assumed to last for 3 years. HIAs were trained in both the depression and the CVD risk software together. The CVD risk software was more complex and training for this took longer, so this was reflected in our cost estimates. Extra training was provided by a consultant psychiatrist in the use of the LLTTFi package, which was available to depression participants, and this was included in the analysis in the same way as was the initial nurse training. The purchase of a licence for LLTTFi was included by apportioning the cost across all depression encounters and the cost of the CBT book and DVD was allocated to those participants who chose to use them. Use of the BWW website was provided at no cost in the trial, but sources at the organisation provided a realistic cost, which was included in the analysis. Some CVD risk participants were provided with a blood pressure monitor to use at home, the costs of which were included on a per-patient basis as an upfront cost.

Primary care and community health and community services

The source for primary care consultation costs and community health services costs was Curtis,332 except when stated otherwise. The details of these services and associated costs are provided in Appendix 22.

Hospital-related services

The costs of hospital-related services, including accident and emergency and ambulance services, were based on NHS reference costs 2012/13333 and Curtis332 (see Appendix 23). Costs were reviewed by a clinician (CS) prior to analysis to ensure that only costs relevant to either mental health or CVD risk were included. For example, a patient in the CVD risk trial reported day care in a NHS hospital for eye complications associated with diabetes and blood pressure following a retinal scan and this resource use was retained in the analysis and costed as a non-admitted ophthalmology face-to-face consultation.

Personal costs of participants and carers

Productivity

Productivity costs were calculated by valuing participant reports of work absence as a result of their condition at median hourly earnings (£11.59) using the Annual Survey of Hours and Earnings 2013 from the Office for National Statistics. 334

Prescribed medication

Medication costs were based on prescriptions issued during the year from randomisation to the end of follow-up. Prescription costs were related to the study data using Prescription Cost Analysis, England (PCAE). 335 Relevant drugs (those issued for treating and managing depression and CVD risk) were identified by the primary investigator (CS). The generic name for each drug was used, if available, as this was likely to reflect what would be prescribed by general practices. Brand name versions were used otherwise.

For each drug listed in the medication records, the net ingredient cost per quantity from PCAE was used to uniquely identify each entry. The net ingredient cost per quantity provided a single unit cost (e.g. for one tablet, one capsule or 1 ml) for each prescribed drug. This was multiplied by the unit quantity (e.g. 28 tablets, 150 ml) for each listed prescription to give a total cost for each prescription. All drug preparations in the form of sachets and liquids were checked against the British National Formulary (BNF). 336

Analysis

Per-participant costs and QALYs were estimated from the sources described in the previous sections. A cost–consequences matrix was constructed for each trial. The cost-effectiveness analysis was conducted from the perspective of the NHS and PSS.

The sensitivity of the results to the inclusion of certain costs was also tested in each trial. The base-case results used data generated through multiple imputation (see Missing data and imputation), but cost-effectiveness in complete cases was also calculated as a sensitivity analysis for the main results in each trial.

In both trials, the main cost-effectiveness analyses were rerun excluding NHS costs collected from participant questionnaires as a sensitivity check on the main results. These costs relate to resource use, such as district nurse home visits, overnight hospital stays and ambulance use. The rationale for the exclusion of these costs was to assess whether or not resource use based on patient recall had a material influence on the conclusions of the economic analysis. Additionally, elements of the cost of the intervention in the depression trial were subject to deterministic sensitivity analysis to account for potentially lower costs if the intervention was deployed throughout the NHS. All analyses were conducted using Stata 13.1.

Missing data and imputation

Missing data are important for cost-effectiveness analysis because cost and QALY data are constructed using cumulative sums of responses to cost and quality of life questionnaires and from sums of resource use from other data sources across all follow-up time points. Missingness in data relevant to the cost-effectiveness analysis arose in the trial for a number of reasons. Medical records, obtained from general practices, were incomplete in a very small number of cases (< 0.5% across both trials). Questionnaires were sometimes unavailable (because participants had not returned them) or were only partially completed.

Missing data were addressed in two ways. The first method used mean imputation for variables that had very low levels of missingness and were unlikely to be major cost drivers for the analysis overall. Mean imputation was used for the very small fraction of participants with incomplete primary care data (< 0.5% of all randomised participants in both trials, representing 10 patients overall). This incompleteness arose because patients declined consent for their records to be accessed or moved away from the practice with which they were associated at the time of randomisation.

Available cases, as reported in Chapter 10, were defined as cases for which it was possible to sum across all data sources, including questionnaires, for all follow-up time points for the resource use, cost or quality of life measure concerned. This meant that the number of available cases for a particular resource varied by question. Available cases were not used for cost-effectiveness inference, although they are referred to descriptively in cost–consequence matrices and elsewhere in summary tables in Chapter 10.

Complete cases for the purposes of economic analysis were defined as a subset of available cases. Complete cases, after the pre-imputation data cleaning described above, had complete data on all cost items and all health-related quality of life items, to enable full cost-effectiveness inference to be conducted. However, using only complete cases in analysis is equivalent to an assumption that data are missing completely at random and may lead to bias, whereas ignoring a proportion of available data is likely to be inefficient. Data from multiple imputation were therefore used as the second means to address missing data to inform the base-case cost-effectiveness results in each trial.

The same imputation models were used for the main trial analyses and for the economic analyses. Costs were imputed at the level of aggregate cost categories (primary care costs, medication costs, intervention costs, PSS costs and remaining NHS costs), rather than as individual resource use items. Health-related quality of life utility measures at each time point were imputed, which were then used to generate QALY measures.

Cost-effectiveness analysis was performed on the imputed data set in a way that reflected the variation within and between the imputed data sets in accordance with Rubin’s rules. 323–325 The economic analysis of cost-effectiveness on the imputed data was based on the methods and discussion in Faria et al. 324 – both seemingly unrelated regression analysis and bootstrapping (based on 5000 model iterations) were used to produce incremental cost-effectiveness ratios (ICERs) and cost-effectiveness acceptability curves (CEACs). The regression analysis was used as the basis for the main cost-effectiveness results.

Uncertainty

Both bootstrapping and regression analysis were used to characterise the level of uncertainty around point estimates of the ICER. Five thousand bootstrap replicates of the ICER were constructed by sampling, with replacement, from the original data and were used to generate cost-effectiveness planes. Cost-effectiveness statistics, such as net monetary benefit (NMB) mean estimates and associated CIs, were calculated parametrically from the regression output.

Discounting

Costs and outcomes in the trial-based analyses reporting at time horizons of up to 12 months were not discounted. Discounting for the long-term CVD risk modelling is described in Model perspective and discount rate.

Introduction to long-term cardiovascular disease risk modelling

The primary clinical outcome of the CVD risk trial was 10-year risk of a cardiovascular event, estimated using the QRISK2 prediction algorithm. A 10-year period is substantially longer than the follow-up period of the trial, but may be shorter than the actual duration of the effect of the intervention on patient health. The principles of economic evaluation337 and NICE guidance328 require that studies adopt a time horizon that is sufficient to reflect costs and effects that may differ between intervention and usual care groups. Data from the trial and from other sources were therefore used to develop and populate a state transition cohort simulation model to estimate the cost per QALY of the intervention over periods up to and including the lifetime of the CVD risk trial population. A similar model was not attempted for the depression trial in the absence of a comparably comprehensive set of economic and epidemiological data needed to populate a lifetime economic model.

The state transition cohort simulation model

The cohort simulated was a hypothetical population of patients with risk profiles that reflect those of the Healthlines CVD risk trial population. The extent to which the intervention was successful in altering modifiable risk factors between trial arms over the period of follow-up was used as a basis for simulating the costs, utility and mortality pathways experienced over the lifetime of the simulated cohort. The model is based on modified and updated versions of models used and described in previous health technology assessments in the area of CVD risk. 338,339

The model consists of a number of discrete and mutually exclusive health states, each of which is associated with the presence or absence of a particular CVD-related health condition or with death. Events, such as a stroke, establish the state to which an individual patient in the cohort is assigned at any moment in time. The probability of a primary, or initial, event was determined from incidence data, whereas the probability of moving from one state to another was determined by transition matrices calculated primarily from epidemiological data.

Sequences of particular types of health experienced by members of the cohort established the simulated clinical pathway of patients. Different sequences of events, averaged over the entire cohort modelled in each arm, give rise to differences in cost, utility and mortality. This probability-weighted averaging involves identifying how many patients are in a particular health state at a moment of time, for how long patients have been in particular states and the consequences of membership of each health state. This information permits the calculation of QALYs and costs for each arm and thus ICERs and net benefit statistics.

A total of 1000 iterations of the model were run in a Monte Carlo simulation in which values of input parameters were varied simultaneously. Each iteration took a value from the probability distributions assigned to the input parameters of the model. Probabilistic sensitivity analysis was used to quantify the uncertainty around the post-estimates of cost-effectiveness measured by the net benefit statistic. The model was implemented in Microsoft Excel 2013 and was based on an earlier Microsoft Excel model used in Ara et al. 338

Cohort size and characteristics

A cohort of 1000 patients was simulated, including one cohort for each arm of the trial and one for each sex. Baseline characteristics of the cohorts reflected those in the CVD risk trial participants. For each arm, the weighted average of male and female results was calculated to obtain an overall estimate of cost-effectiveness.

Definition of cardiovascular disease events and model states

Each patient enters the model in an ‘event-free’ state, meaning that no patient had a diagnosis of any CVD condition. This is consistent with the inclusion criteria of the trial and reflects the basis on which the QRISK2 algorithm was constructed. The output of the QRISK2 algorithm used in the trial is the risk of the first instance of CVD in a 10-year period. In practice, specific episodes of CVD differ in their impact on mortality, quality of life and cost, and may recur following an initial event, and these differences needed to be reflected in the simulation model.

It was necessary, therefore, to convert the 10-year predicted risk of any event into the risk for a specific first event in a single year. As a first step, 10-year risk was converted into an annual risk. The second step converted this risk into the risks of specific events as follows: CVD was defined for the purposes of constructing the QRISK2 algorithm as CHD (angina and myocardial infarction), stroke or TIA. These instances of CVD were used to define events in the simulation model, except that stable and unstable angina were distinguished as separate events. Data on the incidence of all of these conditions, described in Incidence rates for primary events and annual growth of QRISK2 scores, were used to define the probability of a first specific event by age and sex by distributing the overall age-adjusted risk of any first event outputted by the QRISK2 algorithm for each arm of the trial into its component parts.

The probability of moving between different states was initially determined by the interaction of baseline risk characteristics and age-dependent, time-dependent transition matrices. Data on the transition from a first (event-free) health state to subsequent health states were taken from the different sources described below. Together, the initial event-free state, the primary states (i.e. the first state after the event-free state) and subsequent or secondary health states (i.e. all states after a primary state) define the complete set of possible states into which patients may transition over the course of their simulated ‘life’. The set of states, and the permitted transitions between them, are informed by a combination of clinical feasibility and data availability (Figure 13).

FIGURE 13.

Cardiovascular disease risk simulated model states. AMI, acute myocardial infarction. This figure is based on Figure 27 in Ward et al. 340 Contains information licensed under the Non-Commercial Government Licence v1.0.

Transitions differed by age and sex. The list of permitted transitions between states is described in Appendix 24 (see Box 6). Patients remained in each ‘state’ until a different state occurred.

Each patient was constrained to experience no more than three CVD ‘events’ over the course of their lifetime. This assumption was partly guided by model tractability and the consideration that, in practice, few of the 1000 patients simulated approached three events. Therefore, the assumption is not likely to influence cost-effectiveness conclusions. A similar assumption was used in the study by Ara et al. 338

Cycle length

The cycle length of the model was 1 year. A cycle distributes members of the simulated cohort to mutually exclusive states, each of which was defined by the events described above. Patients cycled at annual intervals through a finite number of health states until they died or reached age 100 years, at which point they were assumed to die. A half-cycle correction was applied to both costs and outcomes. 341

Model perspective and discount rate

A UK NHS perspective was adopted. The personal costs borne by patients in managing their condition or receiving treatment were not calculated, nor were wider impacts on society such as productivity effects. Modelled costs and outcomes were both discounted at 3.5%, in line with NICE recommendations. 328

Duration of the effect of the intervention

In the absence of information about the duration of effect of the intervention beyond the end of the 12-month follow-up, the difference in lifetime CVD risk between trial arms, estimated by the QRISK2 algorithm, was modelled in a number of scenarios and the implications for cost-effectiveness compared. The scenarios were defined by assumed durations of intervention effect of 1, 2 and 5 years and for the remaining life of the modelled patients up to and including age 100 years. For example, if the assumed duration of the intervention effect was 2 years, then after 2 years any benefit of the intervention would have ‘disappeared’ – in practical terms, this would mean that the intervention arm would have a lower risk for 2 years if indeed the intervention was shown to reduce the QRISK2 score, but the risk would be indistinguishable between arms (controlling for age and sex) once this 2-year period had expired. The same logic applies to different durations of effect: once the duration of the effect had expired, however long this duration of effect was modelled as persisting, it was assumed that there would be no difference in the risk faced by individuals in each arm of the trial, with intervention patients facing the same risk as usual care patients. The exception to this assumption was a permanent duration of effect persisting for the remaining lifetime of patients. To avoid sudden jumps at the point at which the intervention ends, the model used a smoothing adjustment that calculated the average score of the intervention risk and usual care risk for the year after which the duration of effect expired – this allowed for the intervention risk to more slowly adjust to the usual care arm risk, rather than jumping immediately to a higher risk. This approach to modelling different durations of effect is comparable to that taken by Mistry et al. ,342 who adopted a similarly ‘agnostic’ approach to the extrapolation of effect beyond the end of follow-up in a CVD-related intervention modelled using a similar type of simulation model.

Annual growth in QRISK2 score

Patients enter the model with a known age and QRISK2 score, corresponding to the average values of these parameters in each arm of the trial. These scores are incremented for each subsequent year of life (corresponding to the annual cycle length of the model) by a background growth rate in QRISK2 score. This background growth rate was obtained from the growth rates in annual risk in males obtained from the QRISK2 algorithm when all input factors other than age were held constant. Input values used for this exercise assumed that patients did not have any of the clinical conditions assessed during calculation of risk (such as atrial fibrillation) and were non-smokers and used values imputed by the algorithm itself for information on patient characteristics, such as cholesterol-to-HDL ratios, systolic blood pressure, BMI and ethnicity.

The growth risk in men was used for both men and women. This is because the 10-year or annual CVD risk is higher for men than for women, holding all other variables constant, at all ages. The growth rate in female risk at younger ages is higher than that for males at younger ages because the level of the former risk is lower than the level of the latter risk, but these differences narrow with age. It was observed in model development that risk levels in women exceeded those in men when extrapolating from the levels of risk estimated in the trial. The solution adopted to this issue was to use male growth rates for both men and women, which constrained female risk levels to below male risk levels at all ages. It is conceivable that risk grows more slowly when the level of risk is already high, or it may continue to grow more quickly than in the general population. It is also plausible that risk may change with age and with changes in the underlying risk behaviours that determine overall CVD risk. On balance, the approach adopted appeared to offer a reasonable and conservative compromise between these different considerations.

The QRISK2 algorithm is not validated for ages beyond 84 years. As a proportion of patients in the simulation will reach ages beyond 84 years, it was necessary to make an assumption about how risk would change beyond this age. A ‘healthy survivors’ effect would suggest that risk may have plateaued; alternatively, risk may continue to increase given how important age is to the level of risk at ages younger than 85 years. For the base-case analysis, risk was increased for ages beyond 84 years by extrapolating the 10-year risk by 1% per annum. This 1% figure was obtained from an ordinary least squares (OLS) regression of 10-year QRISK2 scores on ages between 45 and 84 years.

Incidence rates for primary events and annual growth of QRISK2 scores

Incidence rates for primary events were used in the model to calculate the probability of a primary CVD event – stable angina, unstable angina, acute myocardial infarction (AMI), stroke or TIA. They were used to disaggregate the risk of any such event – calculated for each trial arm by the QRISK2 algorithm – into the risk of one of the component events.

Transitions and secondary events

Incidence rates for primary events describe the probability of transitioning from the event-free state, in which all patients enter the model, to a first CVD event. Subsequent transitions require estimates of the probability of experiencing secondary events after a primary event and after other secondary events.

The model used the probabilities from the study by Ara et al. ,338 one of the studies on which the model itself was based. The probabilities were obtained by Ara et al. 338 using a two-stage regression analysis. In the first stage, logistic regression was used to estimate the probability of experiencing any type of secondary event. In the second stage, this probability was disaggregated using multivariate regression to give a distribution across event types, conditional on an event occurring. The same methodology was used for all subsequent events. The transition probabilities differ by age and sex. The values of these probabilities, and the data sources used in their construction, are described in detail in the study by Ara et al. 338

Utility

Each health state in the model is associated with a specific level of utility. For the ‘event-free’ state, this utility equals the age- and sex-adjusted utility of the general population. 349 For other health states, this baseline utility is modified by the utility impact of the specific condition or event that defines one of the model’s mutually exclusive states. QALYs were adjusted to account for the difference between arms observed at the end of trial follow-up.

The work of Ara and Brazier350 was the principal source for most of the utility values used. Ara and Brazier350 used data from the Health Survey for England (HSE), an annual survey that uses random sampling of every private household address in England. The 2003 and 2006 surveys asked questions about history of CVD, with a random sample of respondents asked to complete the EQ-5D. Baseline utility was estimated by Ara and Brazier350 from individuals who reported no history of angina, heart attack or stroke (n = 25,080). An OLS regression was estimated to relate EQ-5D scores to age and sex. Results from the model so obtained (EQ-5D = 0.9454933 + 0.0256466 × male-0.0002213 × age + 0.0000294 × age²) are summarised in Figure 14.

FIGURE 14.

Baseline utility of health states in the CVD risk simulation model by age and sex.

This baseline utility corresponds to the utility experienced by individuals in the ‘event-free’ health state. For example, a man aged 67 years is estimated to have a utility of 0.82 (out of a maximum possible utility of 1) in the event-free state, whereas a woman aged 69 years is estimated to have a utility of 0.80 in the event-free state.

Ara and Brazier350 calculated state utility values for angina, heart attack and stroke as follows. Data from individuals who reported a history of one of these CVD conditions was assumed to be representative of individuals who had a first ever primary CVD event. Individuals who reported more than one such condition were assumed to be representative of individuals who had more than one CVD event. Individuals who reported CVD events within the 12 months preceding the survey date were assumed to be representative of patients who recently experienced that event, whereas individuals who reported experiencing the event in question > 12 months before the survey were assumed to be representative of patients in the ‘post-event’ state. For example, patients who reported a stroke in the last 12 months were estimated to have a utility of 0.626 whereas those who experienced a stroke > 12 months before the survey date were estimated to have a utility of 0.668. The impact of histories of multiple CVD conditions was calculated in a similar way. In the absence of evidence to the contrary, the utility impact of a first secondary event was assumed to be the same as that of a subsequent event of the same type, for example a first secondary stroke was assumed to have the same utility impact as that of a subsequent stroke.

Ara and Brazier350 provided utility values for angina, heart attack, stroke and the baseline ‘event-free’ health state. Other sources were used to distinguish between stable and unstable angina and to provide estimates for TIA utilities. It was assumed that the angina utility value reported in Ara and Brazier350 related to stable angina. An estimate for unstable angina was obtained by assuming that the ratio of 90% for stable to unstable angina utilities was the same as that reported in Ara et al. 338 Data on TIA utilities were taken from the study by Luengo-Fernandez et al. 351 This study was based on the Oxford Vascular Study, a UK study that followed up stroke and TIA patients for 5 years and calculated EQ-5D utilities using the UK valuation. A table of health state utility values is provided in Appendix 24 (see Table 135).

Utility calculations were based on multiplicative adjustments to baseline utility. This assumes a constant proportional effect of specific conditions on baseline utility. This has been shown to be similar to additive models (which assume a constant additive effect of conditions on baseline utility) and appears to be more suitable than minimum models for the analysis of CVD-related health states. 350

To account for uncertainty in mean utility values, the condition-specific utility values (the numerator in the ratio used to create a multiplier) were drawn from univariate normal distributions in line with the suggestions of Ara and Wailoo. 349 Information on the variance structures associated with the baseline utility equations was not available. The utility of the health states was constrained to be no greater than that of the associated ‘post’ health state. For example, this meant that having an AMI in one model cycle was associated with a greater utility decrement than the utility decrement of the ‘post-AMI state’. Further information on utility sources and assumptions is provided in Appendix 24.

Mortality

Patients modelled in the simulation face an annual risk of death from non-CVD causes, as well as the CVD causes explicitly modelled. The risk of death from non-CVD causes was calculated for men and women from Office for National Statistics interim life tables. 352 The most recently available year for which data were available was 2012. A standardised mortality ratio was calculated by age and sex excluding CVD-related ICD-10 codes I20–I25 (ischaemic heart disease) and I60–I69 (cerebrovascular diseases). All-cause mortality data used in the model were based on draws from a univariate normal distribution.

Costs

A number of costs of treatments and health states were used in the cohort simulation model (see Appendix 26), as distinct from the costs described above concerning intervention-related costs collected as part of the Healthlines CVD risk trial. Many of the costs were based on estimates contained in the studies by Ward et al. 340 and Ara et al. 339 These technology assessments undertook systematic reviews to identify cost information. For the present study, more recent cost estimates were investigated, used and updated when appropriate and noted in Appendix 26.

Summary of health state costs

Health state costs were taken from a number of different sources and entered the model as random draws from the distributions described in Appendix 26. Mean costs are described in Table 16. To account for uncertainty, mean cost, information on variance and published CIs were used to parameterise a gamma distribution for each cost, from which the cost used in a single iteration of the model was a random draw.

The proportion of patients prescribed particular types of medication (see Appendix 26) were subject to random draws from a beta distribution for each iteration of the model, with the alpha scale parameter set equal to the proportion of patients described in each case.

Trial cost data measured during the 12 months of follow-up were used for the first year in which the simulation model was run. The model did not account for costs that fall outside the health system. This could be important for conditions such as stroke, for which the consequences of long-term disability can have material cost implications.

Sensitivity of the simulation model results to trial costs and quality-adjusted life-years

An alternative specification of the simulation model considered only the intervention cost from the trial, excluded other trial follow-up costs and did not include any adjustment for the QALY difference observed between arms in the trial. This sensitivity analysis was intended to investigate the effects of the intervention in modifying QRISK2 alone, at the cost of the intervention. This assumes no difference in costs between arms other than differences in intervention costs and disregards utility data from the trial in the initial year of the simulation. The analysis was intended to isolate the effect of the intervention on QRISK2.

Qualitative interview study

Binary PHQ-9 response and repeated measures analysis

The proportion of participants who responded to treatment increased in both arms at 8 and 12 months (Table 23). Although the between-group difference appeared to diminish with increasing duration of follow-up, there was no statistical evidence of an interaction between treatment arm and time point (interaction p = 0.402). Therefore, we estimated an overall treatment effect using all follow-up data at 4, 8 and 12 months in a repeated measures analysis. This provided some evidence that the intervention had an average effect over 12 months in increasing the proportion of responders compared with usual care (see Table 23).

Continuous PHQ-9 score

Mean PHQ-9 scores decreased over time in both groups (Table 24), with the adjusted between-group difference being greatest at 8 months. Although the between-group difference appeared to change over the duration of follow-up, there was no statistical evidence of an interaction between treatment arm and time point (interaction p = 0.345). Therefore, we estimated an overall treatment effect using all follow-up data at 4, 8 and 12 months in a repeated measures analysis, which provided some evidence that the intervention was effective in decreasing the PHQ-9 score compared with usual care (see Table 24).

Continuous GAD-7 score

Mean GAD-7 scores decreased over time for both groups and, as for the PHQ-9 mean scores, the greatest between-group difference was at 8 months (see Table 24).

Patient satisfaction

There was not much evidence that satisfaction with primary HCPs differed between the groups. However, satisfaction scores for NHS Direct and overall treatment were higher in the intervention arm (Table 25). Nonetheless, these satisfaction scores should be interpreted with caution. This is because participants in the intervention group may not have discriminated between NHS Direct as a public service and the Healthlines Service, which was delivered by NHS Direct staff, although they were asked to provide satisfaction with the two services separately. Satisfaction with GPs and nurses was not collected at baseline and therefore these analyses were not adjusted for any potential baseline differences between the two groups.

Intervention satisfaction

Satisfaction with aspects of the Healthlines Service among participants in the intervention group increased slightly over the follow-up period (Table 26). Although comparisons with the usual care group are not possible, these questions about intervention satisfaction used the same response scale as those questions about primary care and treatment satisfaction (see Table 25). Therefore, it is possible to make general comparisons about the level of satisfaction between the Healthlines Service and other forms of care among the intervention group. Participants in the intervention group appeared to be slightly more satisfied with the care that they received from the HIAs than with the care that they received from GPs and nurses in their own practice. They were also satisfied with the Healthlines website and CBT programme, LLTTFi.

Access to care

There is moderate to strong evidence that participants in the intervention group had fewer difficulties in accessing aspects of care related to service delivery (getting support and advice at times that suit the patient, when they need it most and from the HCPs they want to see) and they also reported greater satisfaction with the amount of support that they received (Table 27).

Amount of intervention received

Number of Healthlines encounters

Figure 16 shows the number of Healthlines encounters started by intervention participants out of a possible total of 10 encounters. Three participants did not receive the intervention because of an administrative error and another participant received only non-encounter calls and so these four participants were categorised as having no encounters. Data were also missing from the intervention database for 12 participants and so it was not possible to tell how many encounters they received. Out of the remaining 295 participants (307 randomised), 26% of participants received none or little of the intervention (started two or fewer encounters), 44% received some of the intervention sessions planned (three to eight encounters) and 29% received all or almost all of the planned encounters (nine or ten encounters). The median number of encounters started was five [interquartile range (IQR) 2 to 9, n = 295], the mean encounter duration was 18.5 minutes (SD 12.7 minutes) and there was a total of 1972 encounter calls. Figure 16 shows that one group of participants received two or three encounters before dropping out, whereas the other group continued for most or all of the course.

FIGURE 16.

Distribution of Healthlines encounters started in the depression trial.

Use of cognitive–behavioural therapy

As described in Chapter 6 (see Depression trial), a central aspect of the Healthlines Service was CBT. Participants could access CBT through a book, Overcoming Depression and Low Mood: a Five Areas Approach,292 or online, using the LLTTFi programme. Table 30 presents the numbers of participants who chose to use the two different CBT formats.

TABLE 30 - Use of CBT in book form or via the online programme

ODLM, Overcoming Depression and Low Mood: a Five Areas Approach.

Note

Three participants did not receive the intervention because of an administrative error and 19 participants had missing data. Data in this table pertain to the 285 remaining intervention group participants.

Form of CBT provided	n	%
Given the ODLM book	45	16
Given access to LLTTFi online	229	80
CBT declined	3	1
Given both LLTTFi access and the ODLM book	8	3
Total	285	100

Living Life to the Full Interactive is operated by Media Innovations Ltd. In total, 204 participants logged onto the LLTTFi website and completed a median of two (IQR 0 to 5) out of the seven CBT modules each. Sixty-six participants who logged on did not complete the first module (Figure 17).

FIGURE 17.

Distribution of LLTTFi modules completed.

Patient engagement

According to the TECH model, patient engagement with telehealth partly determines the success and effectiveness of such services. Few participants in either arm made use of the publicly available NHS Direct telephone services or an online forum or group for physical or mental health during the trial (Table 32). Around one-quarter of participants reported using online searching for health information, although there was no evidence that use differed between arms at 12 months.

General practitioner engagement

Based on the medication-related letters that the HIAs sent to GPs, a form of GP engagement in this trial was the number of medication changes, which would be expected to be greater in the intervention group. We defined a medication change as a change in drug name or dosage from baseline, which was captured from participants’ medical records. There was no difference in the number of changes in antidepressant medications between the two groups, with both groups experiencing a median of zero changes (IQR 0 to 1) (Figure 18).

FIGURE 18.

Distribution of changes in antidepressant medications: (a) control; and (b) intervention.

Self-management

Participants completed a number of the subscales from the Health Education Impact Questionnaire (heiQ),313 which measure different aspects of self-management. First, participants in the intervention group undertook slightly less physical activity than those in the usual care group at 8 months, but slightly more than those in the usual care group at 12 months (Table 34). However, there was no evidence of a difference between the groups at 12 months after adjusting for baseline physical activity. Next, Table 34 shows that the between-group unadjusted mean differences for the four subscales that tap aspects of self-management skills and self-efficacy were small. Nonetheless, there was a slight increase in these mean scores over the trial for both groups.

Treatment optimisation

The proportions of participants who reported taking (Table 35) and adhering to (Table 36) antidepressants, as well as the proportions identified from practice records as being prescribed an antidepressant at least once during the trial (Table 37), were very similar in both arms. The adjusted difference in mean Morisky score315 for antidepressant medication adherence at 12 months’ follow-up was –0.1 (95% CI –0.2 to 0.1; p = 0.501). There was no evidence of a difference between groups in the proportions taking antidepressant medication at either 4 or 12 months (see Table 35).

Care co-ordination

There were no meaningful differences in any of the scales that were intended to capture participants’ experiences of co-ordination of care (Table 38).

Patient context