Synopsis

Background

Headaches are second only to low back pain as a global cause of years lived with disability, accounting for 6.4% of the total years lived with disability.1 Only dental caries are more common than migraine and tension-type headaches.1 Headaches are the most common neurological disorder treated in primary care, accounting for around 3% of general practitioner (GP) consultations; however, 70% of these patients do not get a formal diagnosis.2,3 Two-thirds (64%) of people seen in a specialist headache clinic have not had a headache diagnosis from their GP. Of those with migraine, fewer than half have been offered specific migraine treatment.4

For many people with headache disorders their symptoms are intermittent and they can be managed with symptomatic treatment as required. However, for a substantial group of people, headaches become a chronic disabling disorder. This group contributes disproportionally to the health burden, and economic cost, of headache disorders.

Treatment guidelines for headache disorders are typically formulated in a biomedical framework, with the main focus being on drug treatments for those with a diagnosed headache disorder. For example, the only non-pharmacological intervention recommended for people with chronic tension-type headache or chronic migraine in the 2012 National Institute for Health and Care Excellence (NICE) guidelines5 published was acupuncture. There is a need for more non-pharmacological treatments to help those living with headache disorders.

In 2015, when we started this programme of work, very little was known about how best to support those who have developed a chronic problem consequential to a primary headache disorder. We anticipated that any supportive self-management programme would need to include helping more people to get a headache diagnosis, avoiding medication overuse headaches, providing medication appropriate for the headache type and supportive self-management for a chronic painful disorder.

Headache researchers usually expect those individuals studied to have an established headache diagnosis prior to study entry.6,7 This work, however, started from the premise that most people with frequent headaches do not have an established diagnosis, and was developed from the perspective of headaches as a chronic disorder. There is not, however, a recognised clinical entity of chronic headaches. In epidemiological studies researchers use a definition of chronic headaches based on the definition of chronic migraine or chronic tension-type headache, specifically headaches on ≥15 days per month for the previous 3 months.8–11 This matches neither the conventional definition for chronicity of pain used by International Association for the Study of Pain (IASP), of pain persisting for greater than 3 months,12 nor the definitions of chronicity used for other headache disorders, for example chronic cluster headache.9 In addition, the chronicity of headaches defined in this manner is labile.9,12 In a community study of migraine,13 three-quarters [386/526 (73%)] of people with chronic migraine had one or more 3-month period in the following year when their headache frequency was consistent with episodic (<15 headache days per month) rather than chronic migraine.13 Nevertheless, this epidemiological definition is a useful shorthand to describe our population of interest: people living with headaches affecting them on most days. Our overarching aim was to develop and test a supportive education and self-management group intervention implementable in primary care for people with chronic headaches. This is an area with little previous methodological work and an absence of substantial previous UK experience of recruiting people from primary care for studies of headache interventions. A broad programme of research was needed to set the scene for our randomised controlled trial (RCT).

Key areas of uncertainty were as follows:

• Was it possible to recruit people from primary care who met our definition of chronic headache?
• How easy would it be to identify people who might have other headache types not suitable for our intervention package?
• What is the patient experience of living with chronic headaches (including chronic migraine)?
• What should be the content of the self-management support intervention?
• What format should be used for the delivery of the self-management support intervention?
• How acceptable would our intervention package be to people living with chronic headaches?
• How should we measure outcomes?

These were all addressed in the phase 1 feasibility study. In phase 2 we ran a full RCT, with a cost-effectiveness analysis and an embedded process evaluation.

Phase 1: trial feasibility

The trial feasibility phase was made up of four core areas of work, each of which mapped onto our work packages (WPs):

• WP1 – developing a strategy to identify people with chronic headaches from primary care.
• WP2 – developing and evaluating a telephone headache classification interview.
• WP3 – developing and evaluating an education and self-management support intervention for people living with chronic headaches.
• WP4 – selecting the most appropriate patient-reported outcome measure (PROM).

The trial feasibility phase aimed to answer the questions of what can be done, what should be done, and how best it can be done for a future RCT. Here we describe these packages of work and how these informed the trial feasibility (see Figure 1).

FIGURE 1 Components of the trial feasibility phase. PPI, patient and public involvement. Reproduced with permission from White et al.14 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The figure includes minor additions and formatting changes to the original. (Figure not reproduced here; see the original article.)

Recruiting feasibility sample (work package 1)

Objective: to develop and test strategies for recruiting people living with chronic headaches from primary care.

To determine recruitment feasibility, we needed to:

• Develop a strategy to identify people with chronic headaches from primary care. Our scoping work identified that the standard clinical terminology system (Read codes) for coding chronic headache disorders in general practice was rarely used. We developed a search strategy incorporating age, consultation for headaches and prescription of headache-specific medication to identify our target population.
• For any future trial, we needed to be sure we could recruit practices. To test this, during the feasibility phase we aimed to recruit practices in the West Midlands. With the help of our local Clinical Research Network (CRN) we recruited 14 practices, giving us a total practice population of 128,634. A detailed account of the recruitment process is given in our published paper (see Appendix 1).14

From the practice population of 128,634 we obtained informed consent from 131 participants (see Figure 2).14 Participants’ mean age was 49 years [interquartile range (IQR) 38.5–58 years, standard deviation (SD) 13.3 years], 108 (82%) were female and 125 (95%) were white. These participants consented to:

FIGURE 2 Practice and participant recruitment for the feasibility study. Reproduced with permission from White et al.14 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The figure includes minor additions and formatting changes to the original. (Figure not reproduced here; see the original article.)

• completing an electronic (smartphone/web-based) headache diary (a paper version was also available) for 3 months
• filling out a baseline, 2-week and 3-month questionnaire
• taking part in two telephone interviews
• the research team sharing the headache assessments with the participants’ GPs.

They also agreed to potentially being invited to two further studies:

• an interview study to explore the experience of living with frequent headaches, how we might refine our proposed support programme, and what outcomes are important to people living with frequent headaches
• a pilot study in which we invite people to take part in our chronic headache education and self-management support programme.

The sample size for the feasibility study was driven by the need for sufficient data to measure the level of agreement of the classification interviews in WP2. The original sample size was 170; this was revised to 153 paired interviews owing to a change in planned analyses approved by the Programme Steering Committee (PSC) and the funder in March 2016. Later, the PSC agreed that recruitment could stop at 131 participants after reviewing the agreement analysis for the first 100 paired classification interviews. This cohort of 131 participants provided us with a sampling frame for the feasibility work; full details are in our published paper (see Appendix 1).14

Developing and evaluating a structured telephone interview for diagnosing common chronic headache disorders (work package 2)

Objective: to develop and evaluate a brief diagnostic interview to support diagnosis for people with chronic headaches, focusing on the diagnosis of the three common chronic headache disorders – migraine, chronic tension-type headache and medication overuse headache.

Within this WP it was important not only to be able to identify the population of interest for the main trial, but also to feel confident that those who have other headache types not suitable for our trial were appropriately identified, and referred for relevant support. We first reviewed the literature on diagnostic tools.

Systematic review of diagnostic tools for chronic headache

We searched for studies aiming to validate tools for diagnosis and/or classification of headaches. We searched the published literature between January 1988 and June 2016 using MEDLINE^® (National Library of Medicine, Bethesda, MD, USA), Applied Social Sciences Index and Abstracts (ASSIA), Embase^® (Elsevier, Amsterdam, the Netherlands), Web of Knowledge™ (Clarivate™, Philadelphia, PA, USA) and PsycInfo^® (American Psychological Association, Washington, DC, USA). Methodological quality was assessed using items from the Quality of Diagnostic Accuracy Studies (QUADAS-2) tool. We identified 4348 titles and removed 2459 duplicates; after screening the remaining titles we obtained full-text results for 195 papers. Thirty-eight papers met our inclusion criteria validating 30 tools designed to diagnose, classify or screen for headache disorders. Of these, 21 tools were for classification of one headache type and nine were for multiple headache types. Full details are given in our published paper (see Appendix 2).15

We did not find any tools that we could use for our proposed trial and felt that it was important to develop our classification tool based on the evidence as well as a collaboration with clinicians and patients. We therefore organised a consensus conference. Full details are in our published paper (see Appendix 3).16

Consensus conference

The aim of the day was to reflect on the evidence from the review and draw on the expertise of the delegates to help inform the content of the Chronic Headache Education and Self-management Study (CHESS) nurse telephone classification interview.

The purpose of the meeting was to develop a classification interview that would:

• confirm study eligibility – participants aged ≥18 years with chronic headache, defined as a headache on ≥15 days per month for at least 3 months
• classify the participant’s headache type as part of the active intervention to inform treatment and advice (this was done during the face-to-face part of the intervention).

Twenty-six delegates attended (10 neurologists with specialist interest in headaches, three general neurologists, five headache specialist nurses, one GP with a special interest in headaches and seven people living with chronic headaches) attended our ‘Classification Consensus Conference’ at the University of Warwick in August 2015.

They were split into four multidisciplinary groups. Using facilitated discussions each group was asked to address the following four questions:

• What do we need to know from a person to exclude secondary headaches?
• What do we need to know from a person to exclude primary headaches other than chronic migraine and tension-type headache?
• What do we need to know from a person to distinguish between chronic tension-type headache and chronic migraine?
• What do we need to know from a person to identify medication overuse headache?

The facilitators aimed to get consensus on discussed items; when there was uncertainty, items were taken to a plenary session in which they were further discussed followed by voting to gain consensus. Further work by the research team developed and refined the final classification logic model underpinning the classification interview (see Figure 3).

FIGURE 3 Classification logic model. a, Tension-type headache, has mild nausea. NSAID, non-steroidal anti-inflammatory drug. Reproduced with permission from Potter et al.16 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The figure includes minor additions and formatting changes to the original. (Figure not reproduced here; see the original article.)

The classification logic model is presented as a linear process. However, it was developed to be completed non-sequentially as information was obtained during the interview and it is not a diagnostic algorithm.

Having developed the classification tool, the next phase was to validate the tool by training nurses to use the tool.

Training and validation

On 22 February 2016 six research nurses who would complete the telephone classification interviews attended a training day. Nurses were provided with a detailed manual and further one-to-one support to ensure that they felt confident in conducting the interviews (see Report Supplementary Material 1).

To validate the telephone classification tool, we did paired interviews whereby participants received an interview with the nurse and soon after a second interview with a headache specialist doctor from the National Migraine Centre (NMC).17 These doctors used their standard approach to telephone assessment and did not use the CHESS logic model.16

Level of agreement was measured using proportion of concordance, the kappa statistic and prevalence-adjusted bias-adjusted kappa.18 The sample size calculation was based on the kappa statistic of the level of agreement between the nurse interview and the specialist doctor. Nurses carried out 111 classification interviews and the doctors carried out 108 interviews. We obtained paired data on 107 participants.

There was generally good agreement between nurse and doctor interviews (proportion of concordance > 0.75).16 We reviewed cases in which both parties disagreed on the classification and those in which both classified the headaches as ‘other’ (non-chronic migraine or chronic tension-type headache).16 Typically, the disagreements were around whether the headaches were episodic or were chronic migraine. Four people had an excluded headache type: two people had cluster headaches, one had a hemicrania continua and one had a primary stabbing (ice pick) headache. This confirmed that, although ineligible primary headache types are uncommon, they were sufficiently common to justify identifying them prior to randomisation.

A striking, and unexpected, observation from this work was that only a very small proportion of those we assessed had chronic tension-type headaches: only 6 out of 107 (6%). This had consequences for our approach to the primary analysis for the RCT described in phase 2.

Developing and evaluating the intervention (work package 3)

Objective: to develop and pilot an education and self-management support intervention for the management of common chronic headache disorders (the CHESS intervention) that is both theoretically informed and based on best evidence.

Details of our intervention development are in our published paper (see Appendix 4).19

We first reviewed the existing literature to understand the experience of chronic headaches from the patient perspective, what content and approaches might be effective for this population and what modifiable prognostic factors exist to be targeted in future interventions.

Systematic reviews

Details of each review are in our published papers (see Appendices 5–7).20–22

Lived experience of chronic headache (see Appendix 5)20

We systematically reviewed and appraised the qualitative literature on the lived experiences of those living with chronic headaches. We included qualitative studies of adults (aged ≥18 years) with chronic headaches. We searched MEDLINE, Embase, ASSIA, PsycInfo, Scopus^® (Elsevier, Amsterdam, the Netherlands) and Web of Science™ (Clarivate) between January 1988 and July 2016. We included studies that used qualitative methodology, or mixed methodology if the qualitative findings were reported separately. We excluded studies that did not have a patient’s perspective, theses, dissertations and conference papers. We appraised the included studies for risk of bias. We used a thematic analysis across the studies followed by a meta-ethnographic approach.

Four studies met our inclusion criteria. Analysis identified three overarching themes:

• ‘headache as a driver of behaviour’ – forcing patients to stop activities or take increasing medication to function
• ‘the spectre of headache’ – the worries, fear and guilt that patients carry
• ‘strained relationships’ – the effect their headaches and behaviour have on those around them.

Although chronic tension-type headaches were represented in the data, they may have been overshadowed by chronic migraine features.

Prognostic factors in chronic headache (see Appendix 6)21

We included prospective cohort studies and RCTs of chronic headaches, published in English. We included adults with chronic migraine, or chronic tension-type headache, with or without medication overuse headache disorders. We excluded studies with participants < 18 years old, dissertations and conference proceedings. We searched Cochrane, MEDLINE/PubMed^® (National Library of Medicine), Embase, PsycInfo, Web of Science and ASSIA, from January 1980 to June 2016. Two reviewers independently extracted data and assessed the methodological quality. RCTs were included only if a subgroup analysis was reported or enough data to perform subgroup analysis were presented. We assessed the adequacy of any moderator analyses.

Twenty-seven studies met our inclusion criteria: 17 prospective cohort studies and 10 RCTs with subgroup analyses. There was moderate evidence for depression and anxiety, poor sleep, stress, medication overuse and poor self-efficacy predicting a poor outcome. There was inconclusive evidence for treatment expectations, age and age at onset, body mass index, employment and headache features predicting a poor outcome.

Broadly speaking, the factors identified were consistent with prognostic factors seen in people with chronic painful musculoskeletal disorders, supporting the notion that adapting approaches used to help people live better with other chronic disorders can be applied to people living with chronic headaches.23,24

Style and content of intervention programmes (see Appendix 7)22

Our aim in this review was to identify the components and method of delivery used in non-pharmacological educational and self-management interventions for headache disorders. We included RCTs comparing a relevant educational and/or self-management intervention for headache disorders with usual care. We excluded studies with participants aged < 18 years old, invasive treatments such as acupuncture, interventions purely focusing on physical exercise, dissertations and conference proceedings. We searched the Cochrane Library, MEDLINE, Embase, Web of Science and PsycInfo from January 1980 to June 2016.

We included 16 studies in the review. We found positive overall effects of self-management interventions over usual care for pain intensity, headache-related disability and quality of life. A moderate effect was seen on mood. A greater effect on mood was observed in interventions that included a cognitive–behavioural therapy (CBT) component than those without, and for group interventions when compared with one-to one delivery.

Interview study

We conducted the interview study to build on our understanding from the systematic reviews to aid the development of the intervention. More details are included in our published paper (see Appendix 4).19

We had planned to conduct interviews on the sampling frame developed in WP1. At the time we needed to do these interviews to inform intervention development, this sampling frame was not established. Therefore, our sample was obtained through Migraine Action, one of our charity partners (Migraine Action merged with The Migraine Trust in 2018) and approved by the PSC and funder in October 2015. We sent 100 invitations leading to 21 responses. Of these, seven met our inclusion criterion of headaches on ≥15 days per month for at least 3 months. A topic guide was informed by the literature review. The guide allowed the exploration of perceptions of helpful and unhelpful treatment strategies. All interviews were audio-recorded for transcription.

The results suggested that participants had tried a range of therapies and interventions, some of which were helpful while others were not. Access to education and peer support was deemed positive, as was learning new skills such as relaxation, mindfulness and stress management.

Developing the intervention package

The reviews and the interview study were summarised and presented at a multidisciplinary intervention development day held in November 2015 at the Royal College of General Practitioners in London. The aim of the day was to start to scope out what the CHESS intervention should look like. Eighteen people attended, bringing together clinical, academic and lay expertise. The facilitated discussions were factored around four core areas (see Box 1).

• Tailored headache education i.How can the classification interview be used for the intervention for supporting optimisation of drug treatment?ii.What written information is needed (for GP and patient)?iii.What should be the structure of this consultation?iv.What should it be the content of this consultation?v.How long should it last?vi.Where should it be conducted and by whom?
• Generic chronic headache self-management i.What format should the self-management intervention take?ii.What should the content be?iii.How should it be delivered (format, length)?iv.Who should it be delivered by?v.Where should it be delivered?vi.What material do we need to develop for the intervention arm?vii.Do we need any material for the GP?
• Control group i.What would be deemed an acceptable control arm?ii.What material do we need to develop for the control arm?iii.Do we need any material for the GP?
• Ongoing support i.How should any ongoing post intervention support be provided?ii.How should this be standardised?iii.How should this be recorded?

There was overall agreement that the intervention should be a group education and self-management intervention with an integrated one-to-one consultation. The group intervention would be for 8–10 participants and be modular, but participants should attend all the sessions. Suggestions were to run the programme during school hours in community settings when possible. It was agreed that the intervention should be delivered by a nurse and a layperson living with chronic headaches and delivered in a non-didactic manner. The content should include educational material, self-management material, medication advice, plus a digital versatile disc (DVD) suitable to share with friends and family. Providing a DVD for family and friends was a suggestion from patient partners that the academic team had not previously considered. Ongoing support was agreed as up to 8 weeks of telephone support by the nurse, individually negotiated with participants.

The group felt strongly that there should be a comparator control group arm and not just usual care. As the literature22 suggested beneficial use of relaxation, this was deemed a good control intervention. It was agreed that a relaxation compact disc (CD), adapted from a previous study,25 would be developed for CHESS. Control participants were also to be provided with customised information on their headache type after a classification interview.

In the feasibility phase the intervention was designed to be delivered, by a nurse and a layperson, over 2.5 days with a one-to-one nurse consultation and individualised follow-up.

Testing and refining the intervention package

Having developed the intervention package and a facilitator manual to accompany this (see Report Supplementary Materials 2 and 3), we delivered three groups using the 2.5-day intervention format described. Groups were run between July and September 2016. We approached 79 participants from our sampling frame (see Appendix 1).14 Thirteen participants attended: six attended the first group, three the second group and four the third group. Difficulty with attendance was as a result of commitment of time over 2 consecutive days.

As part of the formative evaluation the process evaluation team observed group two and concluded that the intervention had been delivered as per protocol. They interviewed 12 out of the 13 participants who attended these feasibility groups. Some participants had also completed a feedback form at the end of each day.

Overall, everyone had appreciated connecting with others with the same condition, and this was a driver to return. The course content and pace were well received. Group discussions were appreciated, as was the lay facilitation. The nurse one-to-one sessions were highly valued by most, with the majority wanting telephone follow-ups. The majority said that they would recommend the course to others. Only two had some reservations: one felt that it would be useful for people newly diagnosed and one wanted more individual tailoring of advice.

We interviewed three nurses and two lay facilitators to ask about their experiences of running the groups. This feedback resulted in the removal of the half-day follow-up because participants found it difficult to get the time off to attend. The final and the third feasibility group piloted the 2-day revised format intervention, which was subsequently adopted for the main RCT (see Figure 4). For the main study we also provided additional facilitator training on medication use.

FIGURE 4 Intervention structure. (a) Feasibility phase and (b) main study. (Figure not reproduced here; see the original article.)

During the feasibility phase it became clear that recruiting laypeople with chronic headaches as facilitators would be challenging because of the unpredictable nature of the condition. This was discussed at a trial management meeting in June 2016, at which there was agreement to recruit allied health professionals to co-facilitate with the nurse.

The theoretical underpinnings and behaviour change rational and techniques for the CHESS intervention, and course content, are described elsewhere (see Appendix 4).19 The CHESS intervention materials are available from http://wrap.warwick.ac.uk/171671.

Choice of clinical effectiveness and cost-effectiveness outcome measures (work package 4)

Objective: to select most appropriate outcome measures for the RCT of the CHESS intervention package.

This work is described in our published papers and appendices (see Appendices 8–10).26,27

Systematic review of patient-reported outcome measures (see Appendix 8)26

We wanted to assess the quality and acceptability of outcome measures for chronic and episodic headache. We searched for multi-item PROMs evaluated following completion by adults aged ≥ 18 with episodic or chronic headache. We searched published literature between January 1980 and December 2016 using MEDLINE and Embase. We assessed study methodological quality using the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) checklist, and PROM measurement quality and acceptability by reference to accepted international standards.28–32

We included 46 papers providing evidence for 23 PROMs. Six measures looked at the impact of headaches overall and five were specific to the impact of migraines. Six assessed responses to, and/or satisfaction with, migraine-specific drug treatments. A further six generic measures had been assessed in headache populations. Assessment of reliability was generally limited with acceptable evidence for the six-item Headache Impact Test-6 (HIT-6).33 Assessment of responsiveness was rare and patient involvement was limited and poorly reported. Overall, the HIT-6, the Migraine-Specific Quality of Life Questionnaire (MSQ) version 2.134 and the Patient Perception of Migraine Questionnaire – Revised (PPMQ-R)35 had acceptable evidence of reliability and validity, although that was still limited. For a generic ‘headache’ population only the HIT-6 had acceptable evidence of validity and reliability.

Based on the review, and cognitive interviews, detailed below, the HIT-6 was selected as the primary outcome measure for the RCT. Many of the assessed measures had a migraine focus, making them challenging to apply to a generic chronic headache population. However, among the team, patient and public involvement (PPI) research partners and the lay advisory group, there was a sense that a lot of the questions in the 14-item MSQ v2.1 had relevance to our population.

Outcomes for the trial (see Appendix 9)27

With the permission of GSK plc (formerly GlaxoSmithKline plc; Brentford, UK), the developers of the MSQ v2.1, we modified this measure, changing the focus of each item from ‘migraine’ to ‘headache’. The adapted measurement was renamed as the Chronic Headache Quality of Life Questionnaire (CHQLQ).

We did a mixed-methods comparative evaluation of the CHQLQ and HIT-6. Feasibility study participants completed the postal questionnaires at three time points: baseline and at 2 and 12 weeks. This provided the raw data necessary to inform the psychometric evaluation. A range of analyses informed the determination of data quality, reliability, validity, responsiveness to important change in health and score interpretation. The questionnaire included headache-specific, generic and several domain-specific measures. In addition, we carried out semistructured cognitive interviews with 14 participants within 24 hours of them completing the 2-week questionnaire to explore the relevance, acceptability, clarity and comprehensiveness of the headache-specific (and generic) measures.36–39 We wanted to explore what participants felt was missing and how individuals determined any improvement in their headache. Interviews were audio-recorded and transcribed.

Both the CHQLQ and HIT-6 were well completed, had good psychometric properties and were relevant to the experience of headaches. The CHQLQ captured the wide-ranging impact of chronic headache, in particular the emotional impact, to a greater extent than HIT-6.

Our original intention was to make a final decision on the primary outcome for the trial informed by our quantitative study. It was not possible to complete this work before starting the main trial. For this reason, we set HIT-6 as the primary outcome for the trial and the CHQLQ as a secondary outcome.

Electronic data capture (see Appendix 10)

We wanted to explore electronic means of capturing data on the frequency, severity and duration of headaches. The application (app) was developed by Clinvivo Ltd, a University of Warwick spin–out. We developed three questions to capture headache frequency, duration and severity. The developers drafted a version that was initially tested by the research team and members of the lay advisory group.

We agreed a secure data management process to enable data captured from the app to be tracked against each participant’s trial number. A system of flagging participants who had not responded for more than 3 weeks was also implemented. Eight feasibility participants were asked to complete the app over an 11-week period. The overall feedback was positive and completion rates reasonable: the team agreed to include the smartphone app as part of the main trial.

Professor Martin Underwood, the chief investigator, is a director and shareholder of Clinvivo Ltd. The use of this company was suggested in the original application for funding. Professor Underwood subsequently recused himself from all contracting decisions, which followed University of Warwick standard financial procedures. He had no involvement in this aspect of the work from either a University of Warwick or a Clinvivo Ltd perspective during the lifetime of the study. He is not an author on the draft paper describing this work. He has edited this report with respect to the use of the smartphone app.

Mapping study of health outcomes in people living with chronic headaches (see Appendix 11)40

A piece of work mapping between health-specific outcomes and health utility measures was included in the original proposal embedded in our existing data collection. During the lifetime of the programme we concluded that it would be better to collect data for this outside the main trial. Recruiting an external cohort of chronic headache patients (separate from the main trial participants) meant that externally generated mapping coefficients could be obtained to inform the economic evaluation of the CHESS intervention. We set up a separate substudy to collect data for headache clinics. This additional work was approved by the funder in September 2019 but was considerably delayed because of the COVID-19 pandemic. In the mapping substudy, mapping or crosswalk algorithms were developed to estimate EuroQol-5 Dimensions, three-level version (EQ-5D-3L), and Short Form questionnaire-6 Dimensions (SF-6D) health utilities from responses to the HIT-6 and the CHQLQ. Data from cross-sectional cohort of 349 people living with chronic headaches in England were used to develop the mapping functions while baseline data from CHESS participants served as a validation sample. Appendix 11 presents further details of the methods, analyses, and results. Overall, censored least absolute deviations models generated the best performance in terms of accuracy of predictions. EQ-5D-3L and SF-6D utilities were best predicted from the HIT-6 without the need for additional patient-level information, whereas predictions for the CHQLQ required age and gender in addition to the summary score.

Other related work

We describe here other activities done as part of the CHESS programme that are outside the main narrative.

Core Outcome Set in Migraine (see Appendix 12)39

In WP2 we identified inconsistencies in outcome reporting alongside often poorly defined outcomes. We recognised the need and opportunity to develop a core outcome set for migraine trials. We decided that this would have wider relevance than focusing on the needs of just our trial. We made the decision to focus on a core outcome set for migraine trials, rather than for headache trials, informed by the overwhelming proportion of those we recruited in WP1 having migraines.

In a two-step process, we defined the core domain set (what to measure), followed by the core measurement set (how to measure specified domains). We identified >50 domains from our systematic reviews and our qualitative work. These data were presented in two questionnaires, one for episodic migraine, the other for chronic migraine. We did a modified, three-round electronic-Delphi study with patients and professionals.

Professor Underwood, the chief investigator for this study, is a director and shareholder of Clinvivo Ltd, who provided the Delphi platform. He recused himself from any discussions related to the choice of Delphi platform for this study.

The results of this Delphi study (see Report Supplementary Material 4). were discussed at a consensus day, at which the aim was to ratify the core domains, agree on the core measurement set and recommend the core outcome set. Through group facilitation and discussion, a two-domain core outcome set was agreed for chronic and episodic migraine:

• migraine-specific pain – to be assessed with an 11-point numerical pain rating scale, and frequency as the number of headache/migraine days over a specified period
• migraine-specific quality of life to be assessed with the Migraine Functional Impact Questionnaire.41

Although the Migraine Functional Impact Questionnaire is a new PROM, it has strong evidence of face and content validity and essential measurement properties (when compared with existing measures of headache-specific quality of life). Participants in the consensus meeting felt that it better represented the important elements of headache-specific quality of life that were identified during the Delphi process. The Migraine Functional Impact Questionnaire had not been published when we started the CHESS RCT and so we were not able to use this as an outcome.

Relationship between chronic headaches and chronic low back pain (see Appendix 13)42

Our approach to seeking to help people with chronic headaches draws on approaches used to treat people with chronic musculoskeletal pain. As an additional piece of work, we did a systematic review of studies looking at the association between chronic headache and chronic back pain, full details published elsewhere.42 We identified 14 studies reporting on our primary outcome: the association between chronic headache disorders and persistent low back pain (LBP). Different papers found odds ratios ranging between 1.55 [95% confidence interval (CI) 1.13 to 2.11] and 8.00 (95% CI 5.3 to 12.1). The strength of these findings was constrained by the variable approaches used by the original authors to define both chronic headaches and back pain. This supports our decision to use a biopsychosocial approach, grounded in previous work on chronic musculoskeletal pain, to inform our headache intervention.

Multicentre trial (work package 5)

Objective: to run a multicentre RCT, including an economic evaluation, of the CHESS intervention package.

Details of the trial have been published elsewhere (see Appendices 14 and 15).43,44 Further information is available in our original application, final protocol, data management plan, statistics analysis plan and health economics analysis plan (see Report Supplementary Material 5–8). Here we address the research question:

• Is the CHESS intervention package clinically effective and cost-effective when compared with a usual-care control?

Clinical methods

Practices identified people who had consulted with headaches or who had been prescribed a migraine-specific drug (triptans/pizotifen) in the previous 2 years.14 The list of people was screened by a GP in the practice to identify people it would not be appropriate to approach, for example people with a severe uncontrolled mental health problem or a terminal illness. Table 1 provides a list of the full inclusion/exclusion criteria. The practice then sent out packs inviting people to express an interest in the trial. Those who were interested returned an expression of interest form to the research team. The research team telephoned the potential participant to confirm their eligibility and obtain verbal consent to start the smartphone app, or paper headache symptom diary. A study pack with the participant information leaflet, consent form and baseline questionnaire was sent to the potential participant. Once the consent form and baseline questionnaire were received, a classification interview call was arranged (see Figure 5).

FIGURE 5 Study flow chart. NMC, National Migraine Centre. Adapted with permission from Patel et al.43 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)

Nurses used the classification tool (see Figure 3) to confirm study eligibility and flagged any people with suspected non-eligible headaches for a second telephone interview with a doctor from the NMC. Participants classified as having an eligible headache type were then eligible for randomisation. If there was not a group an individual could attend, they were not randomised.

Our population of interest was people meeting an epidemiological definition of chronic headaches, that is, people with headaches for ≥ 15 days per month for at least 3 months.8,10,11 At our classification day, it was decided to focus the trial just on people with migraine or tension-type headaches. People suspected of having other chronic headache types were directed to their GPs. For reporting we present three primary headache phenotypes:

• definite chronic migraine – people meeting International Classification of Headache Disorder, third edition (ICHD-3) criteria for chronic migraine, that is at least 8 days per month with a migraine attack with or without aura9
• chronic tension type headache and episodic migraine – people meeting ICHD-3 chronic tension-type headache and ICHD-3 criteria for episodic migraine
• chronic tension-type headache.

Each group included those with and without medication overuse headache. Because not all migraine attacks meet the strict criteria for a migraine attack either because of early treatment or because they are mild,9 we report chronic migraine and chronic tension-type headache with episodic migraine together for our primary analysis. People meeting our definition of probable chronic migraine are typically managed in the same manner as people with definite chronic migraine and received the same advice within the trial.

In the feasibility study, 97 out of 103 (94%) of those assessed as having an eligible headache type had ‘chronic migraine’.16 With the agreement of the Trial Steering Committee, the Data Monitoring Committee and the funder, we specified that, if at least 85% of our participants had ‘migraine’, that our primary analysis would just be on the population with ‘migraine’, with sample size inflated, if necessary, to ensure adequate statistical power for this analysis.

Between April 2016 and March 2018, we trained 30 facilitators. Quality assurance for the intervention was monitored in several ways, including observations of sessions, audio-recordings of sessions, participant feedback and facilitators’ personal reflections (see Report Supplementary Material 9).

The control intervention quality was assessed by keeping record of when packs had been sent to participants, any contact with control participants, attrition rates, when letters had been sent to the GP and date of classification interview.

Randomisation and masking

The unit of randomisation for this parallel-group study was the individual participant. The randomisation allocation ratio was 1 : 1.07 in favour of the intervention group to account for clustering in one arm. Randomisation was done using minimisation, stratifying by geographical locality (Midlands and Greater London) and headache type (chronic migraine, chronic migraine with episodic migraine or chronic tension-type headache only, with or without medication overuse headache). To maintain allocation concealment all baseline data were collected prior to randomisation. Randomisation was done using Warwick Clinical Trials Unit’s randomisation programme by a person independent of the research team. It was not possible to mask the study team, facilitators or participants from the treatment allocation.

Groups of four or five geographically proximate practices were clustered with the aim of starting recruitment around the same time in several practices. Participants were randomised in batches, with a target size of around 20 to ensure sufficient participants to populate a group and reduce any delay between randomisation and starting the intervention group. Participants unable to attend the group that they were originally allocated to were offered attendance at another group if available.

Participants were informed of their allocation by the study team via a telephone call. Written confirmation of the randomisation and headache classification was also sent to the participant and to their GP.

Post-randomisation withdrawals and exclusions

All participants were followed up when possible, and data were collected in accordance with the trial protocol until the end of the trial. No further data were collected for participants who explicitly withdrew their consent, and only the data collected up the point of withdrawal were used in the final analysis.

Outcome measures

Primary outcome measure

Our primary outcome was the HIT-6 score at 1 year.33 It consists of six questions with five responses (never to always: score 6, 8, 10, 11 and 13 points). The score ranges from 36 to 78 points, with higher scores indicating greater headache severity (see Report Supplementary Material 10).

Secondary outcome measures

We used the CHQLQ as a secondary headache disability outcome. Our other secondary outcome measures were headache days in the preceding 28 days; typical headache duration and severity in previous 28 days; EuroQoL-5 Dimensions, five-level version (EQ-5D-5L);45 SF-12 version 2,46 reported as physical and mental component scores; Hospital Anxiety and Depression Scale (HADS);47 Pain Self-Efficacy Questionnaire (PSEQ);48 and Social activity: Social Integration Subscale (SIS) of the Health Education Impact Questionnaire (heiQ).49

Baseline data collected included basic demographic data and data on the troublesomeness of other bodily pains.50

We collected data on total headache days, average duration of headache and headache severity from participants weekly for 6 months and then monthly, starting from the initial eligibility call. These outcomes were collected according to patient preference using a smartphone app or paper diary records (not both).

We sent postal questionnaires at 4, 8 and 12 months. HIT-6 scores, headache days and EQ-5D-5L scores were, if needed, collected by telephone. To maximise follow-up rates, we used several strategies, including sending high street vouchers with each initial questionnaire and study pens with reminder questionnaires and a shorter questionnaire being sent as a second reminder (see Report Supplementary Material 11).

Sample size

We estimated the sample size using Moerbeek and Wong’s51 method, which accounts for clustering in one arm. Based on similar trials52 we used an intracluster correlation coefficient of 0.01.

The sample size (n = 689: relaxation arm, n = 333; self-management arm, n = 356) was estimated to assess the clinical effectiveness in the migraine population, providing 90% power to detect a target (worthwhile) between-group difference of 2 points (SD 6.87, from the feasibility study) in the HIT-6 outcome at 12 months using a two-sided test and a 5% significance level with 20% loss to follow-up.14 Some support for this being a plausible effect size came from a pilot study53 of a similar intervention for migraine. Some overrun on sample size was expected to allow all groups to be adequately populated.

Primary and secondary analyses

The primary analysis approach was intention to treat. Data were summarised and reported in accordance with Consolidated Standards of Reporting Trials (CONSORT) guidelines for RCTs.54 Our statistical analysis plan is available in Report Supplementary Material 7.

Participant characteristics and outcomes were summarised as mean and SD for continuous data or frequency and percentage for categorical data, summarised by treatment arm. The median and IQR were presented if data deviated substantially from a normal distribution.

The primary end point was 12 months. For the primary and secondary analyses, treatment effects were estimated using linear mixed-effects models with partial clustering to account for the trial design with clustering in the self-management arm [command ‘mixed’ from Stata^® (StataCorp LP, College Station, TX, USA)]. Analyses were adjusted for age, gender, the baseline value of the dependent variable and baseline stratification factors (type of headache and geographical locality). The adjusted treatment effect estimates and associated 95% CI s were presented for all analyses. All statistical tests were two-sided at the 5% significance level. Analyses were conducted using the statistical software package Stata 15 and R (The R Foundation for Statistical Computing, Vienna, Austria) version 4.0.3.

Complier-average causal effect analyses

We carried out complier-average causal effect (CACE) analyses for both levels of adherence for the primary outcome only. Minimal adherence was defined as attending day 1 of the intervention plus the one-to-one session. Full adherence was defined as the participant attending both days, plus individualised contact with the nurse.

Subgroup analyses

We carried out prespecified subgroup analyses using formal statistical tests for interaction to examine whether baseline anxiety (HADS anxiety score, ≤ 10 and > 10 points), depression (HADS depression score, ≤10 and >10 points) and severity (HIT-6 score, ≤ 64 and > 64 points) moderated treatment effect.55

Symptom diary (total headache days, average duration of headache and headache severity)

These data were analysed using longitudinal analyses adjusting for the same variables as those used in the primary analyses (fixed effects) and participant as random effects.

Additional analyses

We assessed treatment effects in terms of the primary outcome for the whole population, including those with tension-type headache only, for chronic migraine and chronic tension-type headache with episodic migraine separately, and for those with or without medication overuse headache. These results will contribute to future meta-analyses and inform future guidelines.

Sensitivity analyses

We performed two sensitivity analyses: one excluding those interviewed for the process evaluation and another excluding those who reported <15 headache days on their baseline questionnaire.

Adverse events and serious adverse events

The frequency and percentage of adverse events (AEs) and serious adverse events (SAEs) in the trial are reported.

Clinical results

The results are in Appendix 15.44 A detailed statistical report is available as Report Supplementary Material 12. A separate statistical report is also available for all 736 randomised participants, including those without migraine, as Report Supplementary Material 13.

Screening and recruitment

Between April 2017 and March 2019 we recruited from 164 general practices with a combined patient population of 1,529,684. Of the 32,998 potential participants identified from screening we approached 31,020 (94%). We received 2179 expressions of interest (including 41 self-referrals). We contacted 1871 (86%); of these, 1159 (62%) were eligible. Of these, 92 (8%) did not proceed because there were no suitable groups for them to attend and 785 of the remaining 1067 (74%) returned consent forms. We did classification calls with 751 (96%) of these. Nine people (1%) withdrew at this time. Six (<1%) were excluded because of a non-eligible headache (two cluster headache, two new daily persistent headache, one cervicogenic headache and one hemicrania continua; Figure 6). The final number of recruited participants was therefore 736 (including 27 self-referrals).

FIGURE 6 Consolidated Standards of Reporting Trials flow chart. a, Participants withdrew from intervention only and continued to be on follow-up; b, Complete withdrawal. One withdrawn on the day of randomisation, and one 2 days after randomisation; c, cumulative number of complete withdrawals. Adapted with permission from Underwood et al.44 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)

Of these, 727 (99%) had chronic migraine or chronic tension type headache and episodic migraine. Henceforth, unless otherwise specified, all the results refer to this group. Two participants were withdrawn from the trial by the study team soon after randomisation; one because they were living with someone already randomised to the trial, the other because the person made known that they had recently started in a trial of a calcitonin gene-related peptide (CGRP) monoclonal antibody for their headaches.

Participant baseline characteristics

Baseline characteristics were well matched between treatment arms. Most participants were white (586/727; 81%) and the majority were female (604/727; 83%) and the mean age was 48 years (SD 15 years). Just over half (396/727; 55%) had definite chronic migraine, 46% (331/727) had chronic migraine or chronic tension type headache with episodic migraine and 407 out of 727 (56%) had medication overuse headache. The median number of headache/migraine days over the last 4 weeks reported at baseline was 16 (IQR 11–20) days. Thirty-eight per cent (274/721) reported <15 headache days in the preceding 4 weeks.

We reached a diverse population that was representative of national averages in terms of ethnic mix and levels of deprivation, with a good mix of rural and urban areas.

Medication use was similar across both groups. At baseline, 662 out of 727 (91%) had used acute treatments, and 235 out of 727 (32%) had used prophylactic medications.

The overall mean HIT-6 score (primary outcome) at baseline was 64.5 points (scale range 36–78 points; SD 5.5 points), suggesting that most participants had severe symptoms. People with chronic migraine had greater headache severity, lower quality of life, less self-efficacy and less social interaction than those with chronic tension type headache and episodic migraine. Many participants had chronic pain other than their headaches: 375 out of 727 (52%) had at least moderately troublesome neck and 277 out of 727 (38%) had at least moderately troublesome back pain (see Report Supplementary Material 12).

Participant follow-up

Follow-up rates for the primary outcome were 76% (551/727) at 4 months, 75% (546/727) at 8 months and 80% (582/727) at 12 months. Three participants had missing baseline data. The primary analysis included data from 579 out of 727 (80%) participants. Thirty-two (4%) participants withdrew completely from the study including follow-up.

Adverse events

There were two SAEs, both deaths unrelated to the trial. There were seven AEs: five in the self-management arm that occurred during the intervention sessions and were related to developing a migraine or becoming upset during a session. Two were in the usual-care arm: one related to the content of the relaxation CD, and one participant became upset during a process evaluation interview.

Intervention data

We had 42 intervention groups, run by 20 facilitators, at 35 venues in the Midlands and London. Median group size at randomisation was nine (IQR 7–12) and median attendance on day 1 was 6.5 (IQR 5–9). The first session was attended by 286 out of 376 (76%) of those randomised, 259 out of 376 (69%) achieved partial adherence and 216 out of 376 (57%) full adherence (see Report Supplementary Material 12).

Effect of COVID-19

One-year follow-up was due to be completed soon after the UK national lockdown on 23 March 2020. Inability to access the office during this time meant that we were unable to manage reminder questionnaires. For this reason, more 12-month core outcomes were collected by telephone. The COVID-19 pandemic also made it impossible for the study team to visit general practices to collect data for the health economics analyses. We developed new processes to allow practice teams to extract these data remotely on our behalf.

Primary outcome: six-item Headache Impact Test

We found no evidence of a positive effect at 12 months, the primary end point (mean difference –0.3 points, 95% CI –1.23 to 0.67 points), or at 8 months (mean difference 0.07 points, 95% CI –0.95 to 1.09 points) (see Table 2). At 4 months participants in the self-management support group had statistically significantly lower HIT-6 scores (better headache-related quality of life: mean difference of –1.0 point; 95% CI –1.91 to –0.006 points) than participants in the standard-care group. The intracluster correlation coefficients were very small (< 0.0001) in all our analyses. The findings from our CACE analyses, sensitivity analyses and additional analyses were not materially different. There was no evidence of treatment effect modification in our prespecified subgroup analyses.

Secondary outcomes

There was no evidence of a difference in headache days between the groups at the 8-month and 12-month follow-ups. However, at 4 months participants in the self-management support group reported 1.5 (95% CI 0.48 to 2.56) more headache days in the previous 28 days than participants in the standard-care group. The mean difference of EQ-5D visual analogue scale score at 12 months was 3.9 points (95% CI 0.90 to 6.88 points), favouring the self-management support group, but there were no statistically significant differences at 4 or 8 months. At 4 and 12 months, but not at 8 months, those in the self-management group had stronger self-efficacy beliefs as measured by the PSEQ. The mean differences were 2.3 points (95% CI, 0.51 to 4.00 points) and 2.1 points (95% CI 0.17 to 3.96 points) at 4 and 12 months, respectively. We did not find any differences in the role restrictions, limitations or emotional impact domains of the CHQLQ, EQ-5D-5L, the SF-12 mental and physical component scores, the HADS, or the SIS of the heiQ at any time point (see Figure 7).

FIGURE 7 Treatment differences and 95% CI s for secondary outcomes adjusted for age, gender, baseline value of the dependent variable, headache type and geographical locality at the 4-, 8- and 12-month follow-ups. MCS, mental component score; PCS, physical component score; SF-12/36, Short Form questionnaire-12/36 items. Adapted with permission from Underwood et al.44 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)

At 8 months the mean headache duration in the self-management group was 9.2 hours (SD 7.3 hours), whereas the mean duration in the standard-care group was 8.0 hours (SD 6.9 hours) (difference, 2.0 hours; 95% CI 0.55 to 3.42 hours). There were no statistically significant differences at 4 or at 12 months (see Report Supplementary Material 12).

Estimates and 95% CI rescaled to range from 0 to 100 for graphical representation purposes only (see Figure 7). To obtain the estimated difference and its 95% CI in its original scale, the value from graph is multiplied by (maximum value/100). For example, the estimated difference for the HADS anxiety score at the 4-month follow-up was (−0.801 × 21/100) −0.16821. Full details of results are available from Appendix 15 and Report Supplementary Material 12).

Medication use

At baseline, 91% of participants were using acute treatments (painkillers/triptans) and 32% were using preventative medications. There was no change over time or any between-group differences in proportions using acute prescribed and over-the-counter acute medications or prophylactic medications. Neither were there any statistically significant between-group differences in the defined daily doses used by those using acute and prophylactic drug treatments.

There were two statistically significant differences in drug use in those reporting use of that group of drugs; more defined daily doses of beta-blockers (p = 0.005) at 4 months and fewer defined daily doses of opioids (p = 0.02) were used at 8 months in the self-management support group (see Appendix 15 and Report Supplementary Material 12).

There were no differences in proportions using acute medications for ≥10 or ≥15 days in the last 28 days at any time point. Overall, at 12 months, 43%, and 21% participants, respectively, used painkillers/triptan for headaches on ≥10 or ≥15 days out of the last 28 days. This compares with 63% and 38%, respectively, at baseline (see Report Supplementary Material 12).

Second-line prophylactic drugs [Botox^® (AbbVie Inc., North Chicago, IL, USA) and CGRP monoclonal antibodies] were used by five participants. Four received Botox injection (self-management support group, n = 2; standard-care group, n =2) and two people from the self-management support group were prescribed erenumab. One participant received both Botox and erenumab.

Headache symptom diary

No statistically significant differences were observed between the two groups from the longitudinal analyses for any of these outcomes. The estimated between-group difference for the number of headache days over 1 year was 0.2 (95% CI –0.11 to 0.46) days, for the duration of headache the estimated difference was 0.4 (95% CI –0.47 to 1.28) hours and for headache severity the estimated difference was 0.2 (95% CI –0.08 to 0.46) points on a 0–11 scale.

Health economics methods

Health economic analyses

Our health economic analyses are reported in more detail in Appendix 16. For a full report of our economic analyses, see Report Supplementary Material 14.

We did a prospective within-trial economic evaluation to estimate the cost-effectiveness of the CHESS intervention. For costs we used 2019 Great British pounds and for outcomes we used quality-adjusted life-years (QALYs). Our base-case analysis was conducted from the perspective of the NHS and Personal Social Services (PSS).56

We estimated resource use using the intervention costs, calculated in a micro-costing exercise, and NHS health and social care costs estimated from participant questionnaires and general practice records. We derived the unit costs of community health and social services from the Unit Costs of Health and Social Care 2019,57 published by the PSS Research Unit. Drug costs were estimated from the Prescription Cost Analysis58 and the British National Formulary.59

For our analyses we converted the EQ-5D-5L into health utilities based on the UK tariff for the EQ-5D-3L using the van Hout et al.60 and Hernandaez-Alarva and Pudney61 crosswalk algorithms.60,62,63 We used the van Hout et al.60 crosswalk method for our base-case analysis. For our sensitivity analyses, we used the Hernandaz-Alava and Pudney61 method to estimate QALYs from the ED-5D-5L and Brazier and Roberts’64 algorithm to generate these from the Short Form questionnaire-12 items (SF-12).

To account for missing data we used multiple imputation by chain equations implemented through the R package MICEmic65 assuming that data were missing at random. We imputed missing costs and health utility values at the level of resource category and health-related quality of life assessment, stratified by intervention arm.66 We pooled parameter estimates across 50 imputed data sets using Rubin’s rules.67

Base-case cost-effectiveness

Our base-case cost-effectiveness analysis estimated the cost–utility of the CHESS intervention compared with usual care from the perspective of the NHS and PSS. We calculated economic costs and QALYs for each patient over a 12-month post-randomisation time horizon. We calculated total costs by summing costs associated with the delivery of the intervention and utilisation of broader hospital- and community-based health and social care services.

We fitted bivariate generalised linear mixed-effects regressions assuming a gamma distributed error structure and logarithmic link function to imputed data in R using methods for cost-effectiveness analyses of cluster-randomised and multinational trial data. The models account for the within- and between-cluster correlation between costs and effects measured from the same individuals.

We calculated the incremental cost-effectiveness ratio (ICER) for the CHESS intervention compared with standard care by dividing the between-group difference in adjusted mean total costs by the between-group difference in adjusted mean QALYs. We calculated the incremental net (monetary) benefit of the intervention compared with usual care for cost-effectiveness thresholds ranging from £15,000 to £200,000 per QALY gained.

We estimated the uncertainty of our cost-effectiveness estimates using the Monte Carlo method using 2000 bootstrapped replications.68 We did the following sensitivity analyses:

• QALYs generated from EQ-5D-5L utilities using the Hernandez-Alava and Pudney61,62 crosswalk function
• utilities generated from via the SF-12/SF-6D tariff for the UK64
• costs calculated from a societal perspective
• unadjusted analysis of the multiple imputation data
• adjusted complete-case analysis.

We did the following subgroup analyses:

• medication overuse (yes vs. no)
• geographical locality (London vs. Midlands)
• gender (female vs. male)
• age (<40 vs. ≥40 years).

Health economics results

The acquisition cost of the CHESS intervention was £266.95 per participant. In our base-case analysis the ICER was £8617. There was an 83% probability that the CHESS analysis was cost-effective at a willingness-to-pay threshold of £20,000 per QALY gained. This finding was robust to either of the EQ-5D-5L algorithms. However, using data from the SF-12 the ICER was £32,083. From a societal perspective the ICER was just £765 (see Table 3).

In our subgroup analyses we found lower ICERs for those aged ≥40 years, females, those with medication overuse headache, and those living in the Midlands (see Appendix 16, table 3).