Trial design and methods

Patient and public involvement

We incorporated patient and public involvement (PPI) throughout the course of the study, with the aim of including PPI at the stages of study design, management and dissemination.

When we were developing the project in response to the HTA commissioned call, we presented our research questions to and consulted with the local NIHR Biomedical Research Centre (BRC) Service User Research Enterprise Advisory Group (SUAG) and SUs with DSs from our clinical service at the South London and Maudsley NHS Foundation Trust. Information was also sought via a postal/online survey, for which we contacted SUs from the SUAG and our clinical service. SUs confirmed the importance of this research.

Suggestions made by SUs with DSs were used to guide how we explained the study to potential recruits and the methods used to facilitate retention in the study. These included, for example, regular reminders to record seizure frequency; updates on study progress; a project team member to contact to discuss attendance difficulties; reimbursing transport and parking costs; flexible use of paper/electronic methods to collect seizure data; limiting the length of outcome measure completion time; offering a voucher for follow-up data completion; and providing feedback for participants at the study end as part of dissemination. SUs commented on and informed our choice of outcome measures. We consulted further with SUs with DSs prior to the submission of a full application and to guide our responses to feedback/queries from the HTA Board. SUs provided feedback on the information leaflets to be provided in the neurology and psychiatry clinics.

Subsequently, we identified four people (one from the SUAG and three SU representatives), three of whom had a diagnosis of DSs. Two became members of our Trial Management Group (TMG) and two joined our Trial Steering Committee (TSC). We provided a training session in June 2014 for all four people as the study commenced. This was facilitated by a former staff member from Epilepsy Action (Leeds, UK), which is a user-led organisation that had an active training programme and had committed to offering such training.

The chief investigator and the trial manager also held an interim face-to-face meeting with one SU representative from each committee in September 2017 to review their experiences in the study committees and to consider the challenges that they felt in participating in the study and how we might address these. The PPI representatives on each committee received all relevant trial paperwork and were given their own standing agenda item where they could comment on trial matters if they had not already done so as part of the meeting. They provided feedback on relevant paperwork and, early in the project, on our trial website. The trial manager acted as the main point of contact and would discuss with them the agenda and arising issues before or after the meetings, as appropriate.

Our SU representatives have taken an active role in advising on study progress and on means to improve follow-up rates, and supported our earlier need to extend recruitment and follow-up periods. They have made contributions to dissemination and commented on our communications with participants about study progress. They have advised on the wording of study outputs (e.g. papers and feedback to study participants) and two are co-authors on this report. We have been very fortunate to maintain the active input of all four individuals who initially joined our TMG and TSC (allowing for any times when their own circumstances made this difficult), and we feel that their willingness to give clear opinions on what we have been doing has benefited the study considerably.

We have also shared our experience of PPI within our institution. One of the SU members of the TSC worked with the CODES trial manager and a research associate in the NIHR Maudsley BRC to develop new guidelines for SU involvement on steering and advisory committees for clinical trials and other research projects. These are now available for researchers on the NIHR Maudsley BRC website (www.maudsleybrc.nihr.ac.uk/patients-public/support-for-researchers/involvement-guidelines/; accessed 10 January 2021). This SU member also gave a presentation to the NIHR BRC’s SUAG about her experience on the CODES TSC to encourage other SU representatives to become involved in committees for future clinical trials.

Study design

The CODES study consisted of two phases: an observational screening phase that lasted approximately 3 months after patients had initially been recruited in neurology/specialist epilepsy clinics and then, for eligible and willing patients, an intervention phase, namely a multicentre, pragmatic, two-arm RCT with assessor (research worker and statistician) blinding. The observational phase was to allow for patients who experienced rapid spontaneous remission following diagnosis, as well as to allow time for psychiatric assessments to be scheduled. Patients with DSs can show early remission after communication of the diagnosis alone,93 and we did not want to confound our evaluations by including people whose DSs remitted quickly. Participants were randomised at the beginning of the intervention phase using a 1 : 1 ratio, stratified by site, into two treatment arms. One treatment arm consisted of CBT plus SMC (CBT + SMC) and the other treatment arm consisted of SMC alone. SMC was standardised across the trial and was not simply the treatment that would usually be delivered at a particular centre. Although some demographic data were collected at the initial recruitment into the screening phase, further measures were collected at baseline (prior to randomisation) and at two follow-up assessments (at 6 and 12 months post randomisation).

Trial approval and monitoring

The trial was approved by London – Camberwell St Giles Research Ethics Committee (REC) (REC reference 13/LO/1595). It was also approved by local research and development departments at each NHS trust. The trial was monitored by the TSC and the Data Monitoring and Ethics Committee (DMEC). Both committees met at least once per year and at most twice per year during trial set-up, recruitment and follow-up. The DMEC monitored all serious adverse events (SAEs) within the RCT as they were reported. The TMG (comprising the chief investigator, co-investigators, PPI representatives and the junior statistician and junior health economist) met at regular intervals during the study to review ongoing progress and other relevant issues such as funding and dissemination.

Study settings and care pathway adopted in the study

The trial was run from the Institute of Psychiatry, Psychology and Neuroscience, King’s College London, with the chief investigator, trial manager and some research workers based at this location. Research workers were also located at the University of Edinburgh and the University of Sheffield.

Our study design was based on a care pathway incorporating neurology and liaison/neuropsychiatry settings. After an initial assessment with a neurologist who established the diagnosis of DSs and provided the patient with an explanation for their DSs, the patient was referred to a psychiatrist, who carried out a detailed clinical assessment. The purpose of this was to review the diagnosis, establish a formulation of the patient’s problems, determine the existence of any psychiatric comorbidities and establish eligibility for the RCT. In some cases, pharmacological treatment of anxiety and/or depression was considered. We decided that the psychiatrists would be best placed to provide SMC follow-up sessions given that, in many settings, neurologists discharge patients following DS diagnosis, although SMC provision by neurologists was not proscribed. Most sites that were involved provided only neurology or only psychiatry input, but a small number of sites had both services within the same NHS trust (Table 1). Where sites comprised neurology services only, they referred patients on to designated psychiatry services, either following normal commissioning routes or following agreement for the study.

Screening and recruitment

Eligibility for the trial was ascertained at two stages. Participants were initially consented into the screening phase on the basis that the RCT was to be conducted on patients whose seizures persisted beyond diagnosis, and we judged this in relation to receipt of diagnosis by the neurologists. At the time of trial set-up, research indicated that only ≈ 14% of DS patients were likely to achieve seizure freedom 3 months after diagnosis.93 Sufficient time before entering the RCT was, therefore, built into the protocol to ensure that patients continued to experience DSs before being randomised to a treatment arm within the RCT, as well as to allow sufficient time for appointments to be arranged with psychiatrists.

Screening phase

Participants were initially identified in neurology/specialist epilepsy outpatient clinics. Neurologists would make and explain the DS diagnosis (see Neurologists’ delivery of standardised medical care) and give the patient a booklet with further information about their condition (downloadable from www.codestrial.org/information-booklets/4579871164; accessed 10 January 2021). A protocol was developed for this process, as follows. If patients met the eligibility criteria and were interested in participating, the patient consented to the neurologist forwarding their contact details to the CODES team. A research worker would then contact the participant, explain the trial in more depth and cover the material in the participant information sheet (see Report Supplementary Material 1). If the patient was interested in proceeding, the research worker confirmed eligibility and obtained informed consent. This was carried out mostly in person but occasionally by post. Basic demographic information and a very brief medical history pertaining to the patient’s condition were obtained, and the patient was instructed how to keep seizure diaries. Patients were referred to the designated liaison/neuropsychiatry service by the neurologist. During this intervening period, a research worker contacted the participant fortnightly by telephone, text message or e-mail (based on participants’ preferences) to obtain seizure diary data, comprising the number of seizures experienced per week and how many were severe.

Intervention phase

A liaison or neuropsychiatrist assessed the patient approximately 3 months after their initial neurology diagnosis. This appointment was used in part to undertake further screening for eligibility for the RCT. The appointments included a reiteration of diagnostic points, provision of a more in-depth booklet about DSs (downloadable from www.codestrial.org/information-booklets/4579871164; accessed 10 January 2021) and a detailed clinical psychiatric assessment. A guide was written for the psychiatrist to facilitate consistency and good communication. If the patient met the eligibility criteria and was interested in participating in the RCT, with the patient’s consent, the psychiatrist informed the research worker, who then contacted the participant to explain the RCT further. If the participant still wished to participate, the research worker met with them and covered the material in the participant information sheet for the RCT (see Report Supplementary Material 2). The research worker confirmed eligibility, obtained informed consent and completed the baseline assessments with the patient. Participants were then randomised to one of the two treatment conditions. Enrolled patients were asked to consent to the research team contacting a carer/informant who could provide their own perspective on the participant’s DS frequency at the follow-up stages. If they agreed to this, carers/informants received a participant information sheet (see Report Supplementary Material 3) and, if they agreed to participate, they completed a consent form either at a face-to-face meeting or by post.

Eligibility criteria

Inclusion criteria (screening phase)

• Adults aged at least 18 years old who had experienced DSs within the previous 8-week period and whose diagnosis was corroborated by either video EEG or if not available, clinical consensus (see Neurologists’ delivery of standardised medical care).
• No recorded history of intellectual disability.
• Had the ability to keep seizure diaries and fill out questionnaires.
• Showed readiness to keep seizure diaries on a regular basis and attend a psychiatric assessment 3 months following receipt of their DS diagnosis in the study.
• Were able to provide informed consent.

Exclusion criteria (screening phase)

• A diagnosis of currently occurring epileptic seizures in addition to DSs (‘current’ is characterised as an epileptic seizure occurring in the prior year).
• Lacking the ability to independently maintain seizure records or fill out questionnaires.
• Met criteria for current alcohol or drug dependency in line with the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV)94 criteria since this might, among other things, make attendance and symptom reporting less reliable.
• Insufficient fluency in English to complete questionnaires or later undergo CBT without an interpreter.
• Currently undergoing CBT for another diagnosis, if this intervention would still be ongoing by the time the psychiatry assessment takes place.
• Having previously had a CBT-based treatment for DSs at one of the trial participating centres.

Inclusion criteria for randomised controlled trial (intervention phase)

• Adults aged at least 18 years old who had been recruited into the study in the screening phase following their diagnosis.
• Indicated willingness to continue filling out seizure diaries and complete questionnaires.
• Had given the research team data about their seizure occurrence on a regular basis since receiving their diagnosis of DSs in the screening phase.
• Indicated that if they were allocated to CBT, they would be willing to attend weekly or biweekly therapy sessions.
• Both the participant and their clinician believed that randomisation was acceptable.
• Ability to provide written informed consent.

Exclusion criteria for randomised controlled trial (intervention phase)

• Was currently experiencing epileptic seizures in addition to DSs.
• DS had not been experienced during the 8-week period leading up to the psychiatry assessment.
• Had previously had a CBT-based intervention for DSs at one of the centres taking part in the RCT.
• Was currently undergoing a CBT intervention for another condition.
• Was experiencing active psychosis.
• Met criteria for current alcohol or drug dependence according to DSM-IV criteria since, in addition to making symptom recording and session attendance less reliable, it might be used to reduce anxiety and would reduce the impact of exposure during CBT, and have possible impact on patients’ memory for sessions.
• Evidence of current use of benzodiazepines that exceeded the equivalent dose of 10 mg of diazepam per day, for reasons similar to those for alcohol or drug dependence.
• Was at high risk of imminent self-harm, following the psychiatry assessment or according to the results of the structured psychiatric assessment [the Mini-International Neuropsychiatric Interview (M.I.N.I.)] administered by the research worker, subsequently followed up by a discussion with the relevant psychiatrist.
• Had received a diagnosis of factitious disorder.

Randomisation

Randomisation took place following participants’ consent and completion of baseline assessments. Participants were randomised in a 1 : 1 ratio between the SMC-alone treatment arm and the CBT + SMC treatment arm, using randomly varying block sizes and stratified by site. This was intended to ensure a 1 : 1 allocation in each location in which patients were recruited. Participants were enrolled by research workers and randomised using the online randomisation system at the King’s Clinical Trials Unit (CTU) at the Institute of Psychiatry, Psychology and Neuroscience. For each participant, the relevant research worker entered participant information into the randomisation system that then generated confirmation of randomisation e-mails. Research workers received blinded confirmation and the trial manager and chief investigator received unblinded confirmation with the participants’ treatment allocation. The chief investigator was unblinded for practical and administrative reasons.

Blinding and protection from bias

To protect from bias, research workers collecting outcome data and the trial statisticians were kept blind to participant treatment allocation throughout the trial. In addition to the chief investigator, the trial manager was unblinded so that they could contact participants to inform them of their treatment allocation and inform therapists about which participants to contact to arrange CBT appointments. Participants were asked not to inform research workers of their treatment allocation when completing follow-up assessments. If participants had a treatment-related question, they would contact the trial manager. To evaluate whether or not the research workers remained blind, they completed a ‘treatment guess’ form at the 12-month follow-up or point of withdrawal. If for any reason a research worker became unblinded to a participant’s treatment allocation, they reported this to the trial manager so that the outcome assessments could be completed by a blinded research worker. Participating clinicians and patients were not blinded to treatment allocations.

Interventions

The intervention was DS-specific CBT plus SMC, or SMC alone. SMC is described below from initial diagnosis with the neurologist to the end of the trial and, as both trial arms received this, it is described first.

Control intervention

Across the UK, medical and psychological care for DSs is variable, with different specialties contributing in specific ways.71 To produce a broadly consistent treatment environment across the trial, key approaches were employed to create what we termed ‘standardised medical care’ for patients with DSs. This included providing briefing sessions to the clinicians (e.g. at site initiation visits), a detailed leaflet about how they might explain the diagnosis to patients, crib sheets containing the essential information that they should provide to patients during sessions (see, for example, Appendix 2 for the crib sheet for psychiatrists) and sets of frequently asked questions for clinicians providing SMC (see Report Supplementary Material 4–8 for the remainder of these materials). These are techniques that have been found to be acceptable in other studies.93 It was intended that SMC would be provided by both neurologists and psychiatrists, and would start from the initial meeting with the neurologist at diagnosis. There was no mandatory number of SMC sessions; however, after the initial neurology and psychiatry assessment, we estimated that there would be up to two SMC sessions with the neurologist and three to four sessions with the psychiatrist. However, owing to local service procedures and clinical need, we could not be prescriptive about this.

One important component of SMC was the provision of information. We created two information booklets about DSs to be given at different stages of SMC to supplement the information given to patients by their medical clinicians. These booklets were devised by the clinical members of the project team, with input from SUs with DSs and a hospital information officer. The two booklets were:

1. Dissociative seizures factsheet (neurology). This was to be given to patients by their neurologists when the diagnosis of DSs was first communicated. It was also downloadable from www.codestrial.org/information-booklets/4579871164 (accessed 10 January 2021).
2. Dissociative seizures factsheet (psychiatry). This included content from the neurology leaflet but also provided more extensive information that could help provide further details relevant to psychiatric assessment and treatment. This was to be given to patients when they attended their psychiatric assessment. It has subsequently been made available at www.codestrial.org/information-booklets/4579871164 (accessed 10 January 2021).

When research workers contacted patients about potential participation in the study phases, they asked whether or not these factsheets had been provided; if not, they ensured that participants received the relevant factsheet.

Neurologists’ delivery of standardised medical care

The key elements of SMC provided by neurologists included making a firm diagnosis and explaining it, giving the patient the factsheet on DSs and referring the patient to the study psychiatrist. When making the diagnosis of DSs, neurologists were asked to undertake their usual assessments to clarify the nature of their patient’s seizure disorder to establish the diagnosis. We acknowledged that neurologists may sometimes make the patient’s diagnosis based on clinical history, physical assessment and information provided by informants. In some cases, this had been supplemented by mobile phone recordings of seizures but also by EEG or video encephalographic data. Although video EEG is the ‘gold standard’ for diagnosing DSs, yielding least diagnostic uncertainty,17 we acknowledged that this was not always readily available to a clinical service or deemed cost-effective when other clinical information led to a high level of diagnostic certainty. Thus, video EEG was not essential in the study. If video EEG was not undertaken, a diagnostic consensus was required either between two neurologists in the patient’s clinical service pathway or between the patient’s neurologist and a neurologist in the study team, who reviewed the patient’s clinical records and all relevant investigations. We anticipated that neuroimaging would be conducted only when clinically necessary.

Neurologists were asked to explain the disorder to the patient using the guidelines provided in a crib sheet (see Report Supplementary Material 4 and 6) and explained in detail elsewhere.84 A key aspect of the neurologist’s role at this stage was to explain the referral to psychiatry and why this was appropriate, including the potential benefits of being seen by a psychiatrist.84 Additional information given to patients by the neurologists that was likely to be tailored to the individual included explaining (1) that anti-epileptic drugs (AEDs) are not effective in treating DSs; (2) that talking therapies might be helpful, although there is currently insufficient evidence and this is why the trial was being undertaken; (3) the disorder to significant others, and how to best respond to the DSs; and (4) driving regulations. They might also possibly discuss distraction techniques in general. Although neurologists were not expected to undertake the equivalent of a psychiatric assessment, if risks relating to self-harm, harm of others or psychosis were identified they were expected to refer patients to relevant services or instruct the patient’s general practitioner (GP) to do so if necessary.

After the initial diagnosis session, it was recommended that neurologists offer a minimum of one follow-up appointment at which any of the following might be included:

• assessment of patient progress
• reviewing the patient’s understanding of their diagnosis and, if necessary, going through this again
• if appropriate, supervising the withdrawal of AEDs
• managing comorbid physical disorders that required interventions
• re-evaluating major psychiatric risks that might require interventions
• consideration of the value of prescribing antidepressant or anti-anxiety medication where this was indicated on clinical grounds
• completing any forms required by government departments if required.

However, given service and clinical limitations, this follow-up did not always occur.

Psychiatrists’ delivery of standardised medical care

The psychiatrists’ delivery of SMC was scheduled to begin with a clinical psychiatric assessment approximately 3 months following the neurological assessment and diagnosis. Where patients did not attend the first scheduled appointment, attempts were made to reschedule appoints as often as possible, allowing for service regulations regarding non-attendances and discharge.

This pre-randomisation assessment was intended to perform a partly educational function and cover several important aspects. Psychiatrists were asked to follow specific communication guidelines (see Report Supplementary Material 5 and Appendix 2); restate the points covered by the neurologist to reinforce and further explain the diagnosis; provide patients with a more detailed booklet on DSs (as indicated above); and acknowledge any fears that patients might have about being given a psychiatric diagnosis. The assessment would include a clinical assessment of relevant Axis I and Axis II psychiatric diagnoses and an assessment for risks related to self-harm and suicide; active suicidality would require exclusion from the trial and urgent treatment. It was anticipated that psychiatrists would explain and treat any other psychiatric or other functional somatic symptoms (using psychopharmacological approaches or referral to physiotherapy where relevant), and discuss any factors of possible aetiological significance that were elicited from the clinical history. Other components could include:

• providing information concerning DS warning symptoms and the possibility of distraction without providing specific interventional techniques
• liaising with other mental health clinicians involved in the provision of the patient’s care without referring for psychotherapy, instead focusing on psychoeducation and management of comorbid psychiatric conditions, as would normally be undertaken
• encouraging the person to engage in or resume social activities and/or return to college/work where relevant, and liaising with the appropriate settings to facilitate this
• involving family or friends in these areas as appropriate
• completing forms for government departments as necessary.

Further follow-up appointments with the psychiatrist were intended to include general review and support, considering any psychiatric comorbidities and pharmacological treatment according to clinical need. Psychiatrists were instructed that no additional CBT techniques should be employed during the delivery of SMC.

Cognitive–behavioural therapy for dissociative seizures

Cognitive–behavioural interventions have generally now been shown to have positive benefits for a range of medically unexplained symptoms, as reported in a number of systematic reviews;78,95–99 however, these are not specifically studies of DSs and have not adopted specific models to underpin treatment of DSs. Although mechanisms of change have been studied in adults with medically unexplained symptoms,100 this work has not included adults with DSs. Recent conceptualisations of factors giving rise to DSs have, nonetheless, been developed and include an integrative cognitive model that seeks to explain DSs as automatic activations of a central representation of seizures (referred to as a seizure scaffold) in the context of dysfunction of inhibitory processing.101 The seizure representation may be shaped by different factors that are relevant to seizures in the person’s life and yet further factors, such as chronic stress and arousal, may compromise adequate inhibitory processing. This and other conceptualisations of DSs (e.g. ‘panic without panic’43) offer the possibility of employing cognitive–behavioural interventions to address DS occurrence.

However, although acknowledging the richness of relatively recent models,101 the development of our CBT approach, which predates such models, stems from a single case study81 and was further refined in an open-label study82 and then in our pilot RCT.1 A further description of the approach is given elsewhere.83 A finding of increased symptoms of autonomic arousal as relevant to the concept of ‘panic without panic’43 is not in conflict with the approach taken in these studies.

Our cognitive–behavioural model incorporated the fear escape–avoidance model.102,103 We considered DSs to represent a dissociative response to heightened arousal accompanying cognitive/emotional/physiological or environmental cues that may or may not be associated with previous/current distressing or life-threatening experiences. Alternatively, DSs may have occurred after events, such as panic attacks or syncope. All of these events may previously have led to unbearable feelings of distress and/or fear.

The treatment broadly consisted of engagement and rationale giving; helping the patient develop and use seizure control techniques; helping the person reduce avoidance behaviours via exposure; helping the person tackle maladaptive cognitions associated with seizure occurrence and facilitate emotional processing; dealing with trauma; and planning for relapse prevention. The key elements of the conceptualisation of the disorder and the treatment elements incorporated in the approach used here are illustrated in Figure 1.

FIGURE 1 Model showing responses associated with DSs and the CBT techniques used to target these in our DS-specific CBT. ABC, antecedents, behaviours, consequences. (Figure not reproduced here; see the original article.)

Cognitive–behavioural therapy was intended to be delivered in 12 sessions (plus one booster session) informed by a treatment manual. This number of sessions was based on our previous work1,82 that we were seeking to extend here; in our 2010 pilot RCT1 that provided preliminary evidence for efficacy, not all patients attended both booster sessions so, for easier implementation for the therapists, we limited this to one booster session. The DS-specific CBT was designed to help facilitate the patient to:

• develop an understanding of their seizure disorder
• develop an understanding of how cognitive, emotional, physiological and behavioural aspects of their DSs are related
• understand factors that led to the persistence of their DSs
• learn how to prevent DSs by interrupting behaviours, cognitions or physiological responses occurring before or at the start of their DSs
• improve their lifestyle by undertaking previously avoided activities
• address thought patterns and attributions about their disorder that act to maintain their DSs
• address and deal with previous traumas, poor mood, anxiety or reduced self-esteem, where relevant
• increase their levels of independence and help them to comprehend the contribution of significant others to their disorder.

The CBT sessions included usual CBT components, such as session agendas and planning and reviewing homework activities, but also required patients to complete seizure diaries and undertake activities designed to help with seizure control. Although treatment was manualised, there was room for flexibility, allowing individual formulations. Participants receiving CBT were given a booklet (the ‘Manual for Patients Attending CBT’) that contained written material to supplement the therapy sessions and included pages for making notes. The titles of the individual topics covered in this manual were Introduction to cognitive behaviour therapy and dissociative seizures; A guide for other people; Distraction and re-focusing techniques; Progressive muscle relaxation exercises; Breathing exercises; Graded exposure; Trauma in the context of dissociative seizures; Identifying negative automatic thoughts; Alternatives to negative thoughts; Preparing for the future; and Discharge plan.

Intervention training

Before treating any patients, CBT therapists who had been identified as potentially delivering therapy to trial patients attended a 3-day workshop. The therapists were provided with information relating to the administration and running of the trial, including reporting any SAEs. We asked therapists to report suicidal ideation, suicide attempts and new deliberate self-harm. The workshop also included teaching on DSs and dissociation; a cognitive–behavioural model of DSs; eliciting information about DS occurrence/triggers/perpetuating factors; coping behaviours and avoidance; how to convey the rationale for treatment and engagement of patients; how to deal with DSs occurring in sessions; developing seizure control techniques; using graded exposure to deal with avoided activities; challenging unhelpful thoughts; dealing with trauma and facilitating emotional processing; and ending therapy.

Teaching was supplemented by Microsoft PowerPoint^® (Microsoft Corporation, Redmond, WA, USA) slides and academic papers relating to the content of the treatment and our group’s prior work in this area. Skills specific to DSs were role-played. Three sets of workshops were held: 5–7 November 2014, 14–16 January 2015 and 14–16 October 2015. A total of 59 therapists attended these workshops. The first workshop was video-recorded in its entirety; the videos and accompanying Microsoft PowerPoint slides were edited to be viewable in manageable sections and were made available to all therapists via secure internet links as a means of reviewing the course content. This material was also made available to four additional therapists who joined later at different times during the study and for whom it was not possible to arrange full workshops. They watched the videos and had discussion sessions with the chief investigator, the lead for the intervention (Trudie Chalder) and the trial manager.

During the trial, therapists received group (and occasionally individual) supervision every 4–6 weeks. The supervisors were three senior therapists who were experienced in delivering the treatment model to patients with DSs (median experience 10 years; range 10–30 years). We expected that therapists would receive general service-related supervision in their workplace and supervision specifically related to trial patients from trial supervisors. Part of this supervision involved using recordings of therapy sessions to provide feedback to therapists and to ensure that they were adhering to the treatment model. A therapist rating scale based on the University College London CBT Competency Framework104 was used to rate one session from the initial patient for each therapist (focusing on the treatment rationale). The supervisors scored the therapists and fed back whether or not they were within the predetermined competency levels on the scale. Regular supervision allowed therapist competence and adherence to the manualised therapy to be monitored throughout the trial.

Intervention delivery

The CBT was planned to be delivered as an outpatient service at clinical centres, although occasional telephone sessions were utilised where necessary; we did not set out to evaluate the impact of telephone sessions on outcome, but telephone delivery has been shown to be feasible and effective in other disorders.105,106 Session delivery was recorded in therapy logs. It was intended that the 12 sessions of CBT would be scheduled to occur over 4–5 months, with a further booster session being offered at approximately 9 months after randomisation. Therapists recorded and monitored attendance; reasons for rescheduling and non-attendances; disruption of therapy or injuries owing to DSs; and participants’ completion of homework and adherence to the therapeutic model on a session-by-session basis in a therapy log. We collected demographic details of those therapists who delivered the therapy. Each therapist was allocated a therapist identification number via the MACRO randomisation system (MACRO electronic data capture system, version 4, Elsevier) that was used on the therapy logs to ensure anonymity.

The treatment manual given to therapists was used as a guide, as treatment needed to be individually tailored to recognise that not all elements may be applicable to all participants and different participants may progress through the stages at different speeds or, if necessary, in a different order. The manual outlined what could potentially be covered within each session.

Completion of follow-ups

Follow-up collections of measures were conducted at 6 and 12 months post randomisation. The 6-month follow-up, undertaken to maintain participant involvement in the study and to improve data modelling, was usually conducted by post, and the 12-month follow-up was generally completed in person; a flexible approach was employed to ensure retention and ease for the participant. Efforts were made to minimise dropout rates. A large part of this was the regular contact attempted with participants throughout the trial; we also used procedures that have improved response rates when posting questionnaires, such as enclosing a personalised letter, providing a Freepost envelope for its return and using colour printing for the 6-month questionnaire packs.107 Participants were contacted before the 6-month follow-up to inform them that packs would be arriving by post. The 12-month follow-ups were scheduled well in advance of the required time point wherever possible. Researchers also phoned participants to confirm receipt of posted questionnaire packs and to offer assistance completing the questionnaires. Participants received a £10 ‘Thank you’ shopping voucher for completing the 6-month follow-up questionnaires and a £15 shopping voucher for completing the 12-month follow-up questionnaires.

Remuneration

In addition to the ‘thank you’ shopping vouchers for completing the follow-up measures, we offered participants a maximum of £25 towards the cost of travel to the initial psychiatric assessment, towards attendance at each CBT session and towards any travel incurred in the follow-up (i.e. data collection) assessments.

Adverse event reporting and serious deterioration in health

Information regarding adverse events (AEs) that may have arisen during the intervention was collected over the entire 12 months following randomisation. If a participant indicated a change to health status, the research worker, research nurse or clinical studies officer would ask for further information to determine if the event met the criteria for an AE. Although we may have been made aware of AEs throughout participants’ time in the study, given the frequent contact with participants owing to seizure diary collection, research workers specifically asked at the 6- and 12-month follow-ups about changes in health and these were all self-reported by the participant.

An AE was defined as any health event reported by a participant that was a change from baseline but did not fulfil the criteria for a SAE. This included events where the participant consulted their GP or another medical advisor or took medication. Any AEs that were fatal, life-threatening or disabling, required hospitalisation or prolongation of hospitalisation, jeopardised the participant in a way that may result in one of the above outcomes without medical or surgical intervention or were new episodes of deliberate self-harm or suicidal ideation or a suicide attempt were classed as SAEs.

Seizures were specifically excluded from AE reporting, except where other criteria, such as hospitalisation, were met. Clinicians also reported AEs. During the treatment phase of the trial, any event reported in CBT sessions or during a SMC session was ultimately reported by the SMC doctor. This was to ensure accurate reporting, but also to maintain the blinding of the research workers and the statisticians by removing mentions of CBT. All AEs and SAEs were reported to the trial manager. They were reviewed by the chief investigator and the DMEC. The latter reviewed all SAEs in the treatment phase individually, but remotely, and they were unblinded to treatment arm after reviewing the incident.

To provide an independent assessment of AEs/SAEs in the study, we recruited three clinicians who were experienced in working with patients with DSs to review the AEs; they were initially blinded to treatment allocation. The three clinicians (two consultant neurologists and one consultant psychiatrist who had not been involved in recruiting or treating any study patients) were sent spreadsheets of accounts of entirely anonymised AEs, including information about events, the phase of the study in which these events had occurred, the body system involved and whether or not these events were DS related. They were asked to judge whether or not these met the criteria for SAEs and to rate the severity (mild, moderate or severe) of each event. For both AEs and SAEs, the raters were sent information about the treatment allocation and the initial assessment of relatedness to the intervention. They were then asked to rate the AEs/SAEs for relatedness to the treatment intervention. In undertaking the ratings, the independent clinicians were asked to consider whether or not the event met the protocol definition of ‘serious’ because, if not, the event would be classified as an AE. They were also asked to consider whether or not any SAEs needed to be upgraded to serious adverse reactions (SARs)/suspected unexpected serious adverse reactions (SUSARs) and whether or not any SARs/SUSARs needed to be downgraded to SAEs. Majority decisions were adopted with discussion if there was disagreement over the ratings of severity and relatedness.

We defined a serious deterioration in health as the occurrence of any of the following outcomes: (1) a decrease of 20 points [i.e. a drop by two standard deviations (SDs)] in the Short Form questionnaire-12 items, version 2 (SF-12v2), Physical Component Summary score between baseline and both the 6-month and the 12-month follow-up assessments; (2) participant-rated scores of ‘much worse’ or ‘very much worse’ on the CGI scale; or (3) a SAR.

Evaluation of treatment fidelity

With participants’ consent, CBT sessions were recorded using high-quality digital voice recorders and uploaded remotely onto a secure password-controlled audio-upload system housed by King’s CTU. Recordings were then deleted from the voice recorders following successful upload.

We had initially anticipated that around 15 therapists would treat the study patients.84 Eventually, given local service differences in how therapists participated in the RCT and the overall increase in the number of participants (see Summary of changes to the project protocol), 39 CBT therapists were allocated patients in the RCT. Thus, the independent raters were asked to evaluate one session from each therapist from whom we had usable recordings (not all patients gave their consent to record sessions and some recordings were not uploaded owing to recording errors). We finally had usable recordings from 36 therapists. For each therapist, we chose one patient at random (where the therapist had treated more than one patient) and then chose, on a pseudorandom basis, one of two randomly selected sessions (the third or seventh session) to be rated by each independent rater, so that an equal number of the two sessions were assessed. Sessions were stratified according to whether they had occurred earlier or later in the trial’s progression. If the specific session was missing and at least one other patient’s recordings existed for that therapist, we looked to rate the same session from either the other patient or another randomly selected patient seen by that therapist.

Two independent and experienced CBT practitioners undertook ratings of the integrity of treatment delivery from a sample of these recordings. They were asked to rate the extent to which specific CBT skills were used in the context of working with patients with DSs; whether or not therapists adhered to the therapy, as described in the treatment manual, and were seen to be delivering CBT; and the quality of therapeutic alliance. The individual items rated for each selected therapy session are shown in Appendix 3.

An initial training session was held with the two raters, the chief investigator, the trial manager and Trudie Chalder. The raters listened to and rated a randomly selected session (not included in the main rating exercise), discussed the items and clarified their meanings with the other team members. The ratings were then piloted on four randomly selected sessions and a discussion of the ratings was held to further identify difficulties in ratings and further clarify the meaning of individual items to improve the clarity of coding rules, and achieve rating scores from each rater that fell within one scale point of each other. Further discussions were held after sets of 10 recordings to enable recalibration and to prevent rater drift; discussions were held when item scores differed by more than one scale point.

Raters were blind to the identity of the patient, the treating HCP and the trial outcome. Ratings were made independently; raters were asked to avoid using mid scores as far as possible. Scores were then converted to standardised scores out of 100. For the single-item subscales (i.e. overall therapist adherence, therapeutic alliance and overall CBT delivery), scores were calculated by dividing the score by 7 and multiplying by 100. For the specific DS skills scale (i.e. DS skill items 3–6 in Appendix 3), scores were totalled and then divided by [7 × number of relevant (‘yes’ rated) items scored] × 100 to generate standardised scores.

Summary of changes to the project protocol

At an early stage, we sought ethics approval to complete initial consents into the screening phase of the study by post or telephone if it was difficult to arrange face-to-face visits. At the 6-month follow-up assessments occurring early in the study, it became apparent that not all of the participants were willing to complete the questionnaire pack and return it by post or complete it by telephone; thus, we obtained approval to offer face-to-face data completion. Similarly, we gained approval for completion of 12-month follow-ups by post or telephone where it proved difficult to arrange a face-to-face appointment with the participant.

As the study progressed, we found that our clinical colleagues requested clarification of certain inclusion and exclusion criteria.

In particular, clinicians in the different services asked for further clarity over the length of DS freedom before patients became ineligible for the first phase of the study. We therefore changed our first inclusion criterion to indicate that the person should have been having DSs in the previous 8 weeks. In addition, it was considered more appropriate to exclude patients from the first phase of the study if it was already known that they had previously experienced a CBT-based treatment for DSs at a trial participating centre, rather than applying this criterion immediately prior to consenting to the second phase of the study. Similarly, it was considered more appropriate to exclude patients from the first phase of the study if they were known to be undergoing CBT for another condition (unless this would have finished by the time of the psychiatric assessment, when eligibility for the RCT would be considered).

Finally, we amended a previous exclusion criterion concerning the patient being thought to be at ‘imminent risk of self-harm, after (neuro)psychiatric assessment or structured psychiatric assessment by the research worker with the M.I.N.I.’; in practice, if the psychiatrist felt that the patient was at imminent risk of self-harm the patient would not be considered to be eligible and so would not be assessed on the M.I.N.I. by the research worker, so both conditions would not occur. Instead, we changed this criterion to read ‘the patient is thought to be at imminent risk of self-harm, after (neuro)psychiatric assessment or structured psychiatric assessment by the research worker with the M.I.N.I., followed by consultation with the psychiatrist’. In this way, if the patient reported a high risk of self-harm on the M.I.N.I. to the research worker, the research worker would then consult the psychiatrist about the patient’s suitability for the study.

In the early stages of the 6-month follow-up data collection, it appeared that follow-up rates might be lower than estimated. We, therefore, sought HTA and ethics approval to extend our sample sizes in both stages of the study. This was also necessary because, although we had anticipated that approximately 60% of patients in the screening phase would subsequently enter the RCT, the figure consistently hovered around 51%. We gained permission to recruit 698 (rather than 501) participants into the screening phase to take into account a lower than expected rate of participants progressing from phase 1 to phase 2 (≈51% instead of ≈60%) and to randomise 356 (rather than 298) into the RCT to allow for the initially larger loss to follow-up at 6 months than expected (a conservative estimate of ≈30% rather than the expected ≈17%).

We initially intended that our nested qualitative study would focus on participants in the RCT only. With ethics approval, we extended this qualitative work to include a sample of CBT therapists delivering therapy in the RCT as well as a sample of liaison/neuropsychiatrists delivering SMC. We also obtained approval to undertake an online survey of the neurologists participating in the study.

Although we had considered obtaining Hospital Episode Statistics (HES) data for not only baseline but the full post-randomisation period84 (and as suggested in our initial trial registration on ClinicalTrials.gov), we subsequently obtained ethics approval to obtain these data for baseline and for only the last 6 months of the follow-up period for practical reasons (these data are reported in Chapter 4).

Measures

Clinical and demographic information

We collected the following clinical and demographic information: date of birth, gender, ethnicity, living arrangements, marital status, dependants, attained qualifications, employment status, receipt of disability benefit, previous epilepsy diagnosis, prescription of AEDs, age at first seizure and whether or not the person had previously sought medical help for a mental health concern. Postcodes were collected from all participants to derive an Index of Multiple Deprivation (IMD), which provides a score that is indicative of the level of deprivation in a specific area. The separate databases for England (IMD), Scotland (Scottish IMD) and Wales (Welsh IMD) were published in different years and were based on slightly differing numbers of domains used to derive the scores. We used the versions in use at the time of the first recruitment into the study108–110 and, for consistency, we used the same versions throughout the study. We allocated IMD scores to quintiles ordered across the three databases so that the lowest quintile reflected the least deprivation.

At recruitment to the screening phase, participants rated how strongly they believed that they had been given the correct diagnosis (0 = not at all, 10 = extremely strongly). At recruitment into the intervention phase, self-report information was collected on other medical conditions with which participants were currently diagnosed, along with their treatment preference for CBT + SMC or SMC alone or whether they had no preference. In addition, expectation of the outcome of treatment was assessed by questions on how logical treatment seemed and how confident they were that the treatment would help them when considering CBT, being seen by a neurologist and being seen by a psychiatrist (see Appendix 4).

As well as demographic data, at recruitment into the intervention stage psychiatric comorbidities were assessed using a structured screening instrument (the Mini-International Neuropsychiatric Interview; M.I.N.I. v6.0)111 and a screening measure of personality [the Standardised Assessment of Personality Abbreviated Scale, Self-Report (SAPAS-SR)].112 We included a measure of personality in the light of accounts of personality disorder or personality clusters in people with DSs49,57 and to allow potential future examination of the effect of personality in moderating outcome.

The M.I.N.I. is a commonly used, structured, psychiatric diagnostic interview and is divided into modules that correspond to diagnostic categories including major depressive episode; suicidality; manic and hypomanic episodes; panic disorder; agoraphobia; social phobia; obsessive–compulsive disorder; PTSD; alcohol dependence/abuse; substance dependence/abuse; psychotic disorders and mood disorder with psychotic features; anorexia nervosa; bulimia nervosa; generalised anxiety disorder; and antisocial personality disorder. Research staff who administered the M.I.N.I. attended a 1-day training event that was led by one of the psychiatrists in the project team (Nick Medford). In addition to didactic teaching, role plays were held. Follow-up consultations with Nick Medford were arranged to address administration/scoring difficulties.

The eight-item SAPAS-SR poses questions about how the person sees themselves. ‘Yes’ responses are scored as 1 and ‘no’ responses are scored as 0, giving a final score between 0 and 8 with a cut-off score of 4 for the presence of personality disorder. It has high test–retest reliability; the cut-off score of 4 demonstrates a sensitivity of 0.83 and specificity of 0.8112 when identifying the presence of personality disorder. As we did not then undertake formal assessment of personality disorder for those with scores of ≥ 4, we interpret these scores as indicative of maladaptive personality traits rather than frank personality disorder.

Primary outcome measure

The primary outcome measure used in the trial was self-reported monthly DS frequency at 12 months post randomisation. This enabled us to include all participants’ outcomes irrespective of their improvement or otherwise during the trial. DSs were recorded by patients in a seizure diary that was collected by research workers every 2 weeks, using the format most acceptable to the participant (e.g. via paper diary, text or e-mail). In addition, participants were asked about seizure occurrence at baseline, 6 months and 12 months by a single-item measure for which they were asked to write down how many seizures they had experienced in the previous 4 weeks. The gold standard measure for the primary outcome was to calculate it as the sum of four seizure diary entries for weeks 49–52 (post randomisation). Where these data were not available, full details about constructing the primary outcome can be found in Appendix 5.

Secondary outcome measures

A number of measures were used to assess secondary outcomes. These are grouped together with measures of the same general psychological or other process in the following sections.

Seizure experiences

Secondary outcomes that were directly related to DSs measured seizure severity, seizure reduction and seizure freedom. Two items on the Seizure Severity Scale113 measured subjective severity of DSs and how bothersome DSs were judged to be. Seizure reduction was measured from seizure diaries and self-report questions at baseline, 6 months and 12 months, as appropriate. Seizure freedom was measured in two ways. First, we determined the presence of seizure freedom in the penultimate 3 months of the study. Second, we examined the longest duration of seizure freedom between the 6-month and 12-month follow-up points. We also determined whether or not there was a > 50% reduction in DS frequency at 12 months. To obtain a further measure of seizure reduction, participants were asked to consent to our asking an informant to rate the difference in the participant’s seizures since the beginning of the trial.

Health-related quality of life

Health-related quality of life was assessed by the SF-12v2 Health Survey114 that provides two overall physical and mental health summary scores: the Physical Component Summary and the Mental Component Summary. Both have high internal consistency reliability.115 Higher scores indicate better HRQoL. The scores are norm-based T-scores and are constructed to have a mean value of 50 points and a SD of 10 points. Scoring was undertaken using QualityMetric Health Outcomes™ Scoring Software 4.5 (QualityMetric Incorporated, Lincoln, RI, USA). We also measured HRQoL using the visual analogue scale (VAS), measuring current health from the EuroQol-5 Dimensions, five-level version (EQ-5D-5L),116 on a scale of 0–100 (where higher scores represent better health). While the five domains measured by the EQ-5D-5L were used for the health economics analysis (see Chapter 4), we adopted the VAS as a clinical secondary outcome as it was readily quantifiable and meaningful.

Psychosocial functioning

Psychosocial function was measured by the Work and Social Adjustment Scale (WSAS),117 a five-item self-report scale that was used here to measure patients’ perceptions of the functional impact of DSs on their lives in terms of work; home management; social leisure and private leisure activities; family; and other relationships.117 Scores of > 20 are considered to represent moderately severe impairment or greater; scores between 10 and 20 represent significant functional impairment, but less severe clinical symptomatology; and scores of < 10 are thought to represent subclinical populations.117

Psychiatric symptoms, psychological distress and somatic symptom burden

Four self-report questionnaires were used to investigate psychiatric symptoms, psychological distress and somatic symptom burden: Generalised Anxiety Disorder-7 (GAD-7),118 Patient Health Questionnaire-9 (PHQ-9),119 the Clinical Outcomes in Routine Evaluation-10 (CORE-10)120 and the modified Patient Health Questionnaire-15 (PHQ-15).121,122

The GAD-7 is a seven-item scale that is used for screening generalised anxiety and assessing its severity. It has good reliability,118 as well as criterion, construct, factorial and procedural validity. Higher scores indicate higher levels of anxiety. Scores of ≥ 10 indicate cases of generalised anxiety disorder.118

The PHQ-9 is a nine-item scale that is used to measure depression based on DSM-IV criteria. It can be used to make criteria-based diagnosis of depressive disorders and measure depression severity. It has high internal reliability and test–retest reliability.119 Scores of ≥ 10 indicate a diagnosis of depressive disorders.

The CORE-10 is a 10-item general measure of psychological distress.120 Scores of ≥ 11 are in the clinical range for distress. The CORE-10 has good internal consistency and correlates well with other measures of depression, anxiety and overall mental health.120

A modified version of the PHQ-15 was used to measure other somatic symptoms experienced by participants.121 This version measures whether or not patients had ‘been bothered a lot’ by a list of symptoms over the previous month. Symptoms included the 15 most common physical symptoms of patients presenting to primary care (excluding upper respiratory tract infections),123 the 10 most common ‘neurological’ symptoms124 and the five psychological symptoms taken from the PRIME-MD Questionnaire,123 including worrying about a lot of different things; feeling down, depressed or hopeless; and ‘nerves’ or feeling anxious or on edge. Total scores were measured.

Clinical impression of improvement

A scale that was derived from the CGI125 was used as a self-rated global measure of change for participants; it asked them to rate how much they felt that their health had changed since the start of the study, using a seven-point scale from 0 = very much worse to 6 = very much better. This was also completed by clinicians (mainly psychiatrists) at the 12-month follow-up and by the CBT therapists at the end of the 12th treatment session, at which they were asked to rate how much the participant had changed since the start of the study on the same seven-point scale. Because the therapists administered this measure only to the arm receiving CBT + SMC, the therapist rating cannot be evaluated formally as a secondary outcome.

Satisfaction with treatment

Patients rated their satisfaction with treatment via a single-item measure on a seven-point scale from 0 = very dissatisfied to 6 = very satisfied.

Mediators and moderators

A number of other measures [Beliefs about Emotions Scale,126 a locally devised measure of avoidance behaviour (Avoidance of People, Places and Situations) and a measure of belief in the diagnosis of DSs and belief in having been given the correct treatment] were identified as potential mediators or moderators of outcome rather than main secondary outcomes. However, our current analysis was designed around our identified secondary outcomes, so these other variables will instead be dealt with in secondary analyses to be documented elsewhere.

The timing of the administration of the measures above is summarised in Table 2.

Health economics

These measures are described in Chapter 4.

Nested qualitative studies and survey of neurologists

As part of attempting to gain further insights into reasons for treatment outcomes and to identify issues that may be relevant to the roll-out of interventions, we undertook a mixed-methods approach and conducted three nested qualitative studies to understand patient participants’ experiences of CBT and SMC, the experiences of psychiatrists delivering SMC and the experiences of CBT therapists delivering the psychological intervention in the study, with a view to triangulating the findings. The study methods and results from these nested studies are presented in Chapter 5. We also undertook an online survey of neurologists participating in the study to understand further the impact of CODES on their practice and their evaluation of the care pathway and treatment elements employed in the study. Methods and results are similarly presented in Chapter 5.

Data management

Data were collected on paper source data worksheets and were transferred to an online data collection system for clinical trials (MACRO), which was hosted on a dedicated server at King’s College London and managed by the King’s CTU. Source data worksheets were kept in a research office with restricted access in locked filing cabinets. Research workers entered data and the trial manager performed data entry checks on a minimum of 30% of all data and on up to 50% of specific measures.

The trial statistician received blinded data extracts from MACRO and the randomisation system throughout the trial (e.g. to compile DMEC reports) and ran systematic data queries (such as flagging up discrepancies and missing values) that were sent to the trial manager, who resolved them by either referring to source data or contacting sites. This data checking process was repeated prior to the database lock in an iterative process until the data were deemed ‘clean’ and ready for analysis. Once the data were locked, they were sent to the trial statistician and were imported, labelled, scored and reshaped in Stata^® (StataCorp LP, College Station, TX, USA). While the trial statistician undertook this process for the clinical outcome data, checking of the health economics data was undertaken by the junior health economist in the team who then analysed those data.

Participants’ contact information was kept on a secure central network server with access granted to study staff only. All computers used were password protected and held in an office with restricted access. Audio-recordings of CBT sessions were uploaded from digital voice-recorders at individual sites to a bespoke audio-upload system that was devised and housed by King’s CTU; although recordings could be listened to by specific people via password-controlled access, downloading recordings was not possible.

Power calculations and sample size

The sample size calculation was based on the results from our pilot RCT,1 which was the largest study at the time comparing CBT and standard medical care in a comparable patient population. We reported a large, standardised effect size of Cohen’s d = 0.75 for reduction in seizure frequency in the arm that received CBT [plus standard medical (i.e. neuropsychiatric) care] compared with standard medical care alone at the time point corresponding to the end of CBT treatment, after controlling for pre-randomisation seizure frequency. We also reported a moderate effect size after a further 6 months (Cohen’s d = 0.42).

In other studies of CBT-based psychotherapy for functional symptoms, a moderate effect size is common.127,128 For this reason, it was decided to base the power calculation on detecting a more conservative moderate effect size, comparable to those found in other CBT-based interventions with patients with functional symptoms.

Initial calculations suggested that to detect an effect size of d = 0.42 with 90% power at the 5% significance level, using a two-sided t-test for logarithmic frequencies, 121 participants per trial arm would be required. However, adjustments had to be made owing to therapist effects, using pre-randomisation seizures as a covariate and rates of attrition. First, calculations of potential therapist effects were based on the assumption of around 15 therapists delivering the CBT, so an intraclass correlation coefficient (ICC) of 0.002 was applied, based on a typical therapist ICC.129 This increased the sample size to 149 participants per treatment arm to achieve 92.6% power (using the cluspower command in Stata, allowing for clustering in only one trial arm). Second, because the use of pre-randomisation seizures as a covariate increases the precision of intervention effect estimates, a deflation factor of 0.83 was applied. This was based on a correlation between pre-randomisation DS frequency and follow-up DS frequency of r = 0.42.1 Finally, we needed to account for attrition. We reported an 11% loss to follow-up in the pilot RCT;1 however, a more conservative attrition rate of 17% at the 12-month follow-up time point was decided on for the trial. The final sample size was, therefore, 149 participants per arm: 298 participants in total.

To obtain the target sample of 298 randomised participants, a larger number of patients was required from which to recruit into the RCT. To calculate the number of participants to be recruited into the screening phase, several factors had to be considered. One factor was that eligible participants might not wish to be randomised; from data collected in the pilot RCT,1 this was predicted to be approximately 30%. This meant that 426 participants who were still having DSs would need to be seen by psychiatrists. This number represents 85% of those initially diagnosed with DSs, as some participants may be seizure free at 3 months post diagnosis (approximately 15%93). An estimation of approximately 25% of newly diagnosed eligible patients with DSs was made to account for those who might decline to participate in the study at the point of diagnosis. This led to a final target sample size of 698 eligible patients diagnosed with DSs from whom to recruit and identifying a total pool of 1108 newly diagnosed DS patients. As noted earlier, these values were subsequently revised when the trial was under way.

Trial outcome analysis

The analyses followed the SAP that was agreed by the TSC and was published before database lock.92

We stated in our protocol paper84 (page 8) and in the update in which we published our SAP92 (page 2) that both the primary outcome measure (seizure frequency) and the secondary outcome measures would be evaluated at 12 months post randomisation. In addition, our ethics-approved protocol made it clear that our outcome variables were to be evaluated at 12 months post randomisation. Although we appreciate that there may have been some lack of clarity in the initial completion of the trial registration, which we have subsequently tried to rectify, and in our protocol paper,84 in terms of distinguishing outcome variable measurements from trial end points, this was not meant to indicate evaluation of 6-month variables as trial outcomes. Our current analysis is clearly based on the pre-trial intention to evaluate outcome variables only at 12 months and not also at 6 months post randomisation.

Significance testing or construction of CIs for the difference between the trial arms at baseline were not carried out, as per the SAP. This is because randomisation of participants to intervention arms should have ensured that any imbalance over all measured and unmeasured baseline characteristics was because of chance.130 The formal statistical analyses estimated the differences in relevant summaries (incidence rates, means and proportions) between patients randomised to CBT + SMC and patients randomised to SMC alone with an intention-to-treat (ITT) approach, that is all those with data were analysed in groups as randomised irrespective of treatment received.

Generalisability of the trial sample

To assess whether or not our trial sample of participants (n = 368) was similar to those who were eligible for the trial but did not go on to be randomised (n = 58), we compared key characteristics measured after consent to screening. Variables were chosen for this comparison if they might predict outcome in patients with DSs, that is to check that our randomised sample of 368 participants did not differ from the 58 participants in factors that might later be relevant to the outcome in the RCT. Factors that may predict outcome in patients with DSs more generally include symptom duration, receipt of social security benefits, having previous psychiatric diagnoses, employment status, educational achievement and gender (see Chapter 1). Although we tested a wider range of variables to encompass age, gender, social relationships, diagnostic method, predominant DS semiology, previous diagnosis of epilepsy, current prescription of AEDs, belief in diagnosis of DS and previous medical help-seeking for a mental health problem, to compare these subgroups more widely we did not compare the two subgroups on all of our demographic variables to avoid excessive testing. We employed Wilcoxon’s rank-sum test and Fisher’s exact tests for continuous and categorical variables, respectively. The variables incorporated in such comparisons are indicated in Chapter 3, Tables 6–8.

Descriptive statistics

In Chapter 3, descriptive statistics for all baseline, primary and secondary outcomes are reported by trial arm and overall for all of the time points at which data were collected (baseline, 6 months and 12 months); a formal analysis of a treatment difference was conducted for outcomes at 12 months only. Medians [interquartile range (IQR)] are used to describe count outcomes (seizure frequency or seizure freedom) owing to potential skewness, means (SD) are used to describe all other continuous outcomes, and frequencies (%) are used to report categorical and binary outcomes. All totals (n) are reported. One of the secondary outcomes (informants’ rating of patient’s seizures at 12 months post randomisation) could not be formally analysed because it was completed by only a small proportion (7.3%) of participants.

Predictors of loss to follow-up and the multivariate imputation by chained equations procedure

Prior to unblinding the trial statisticians, a binary variable was created to indicate whether or not participants provided complete 12-month follow-up data for the primary outcome (1 = provided primary outcome data, 0 = did not provide primary outcome data). This binary variable was sent to an independent King’s CTU-affiliated statistician along with the CODES therapy database; the independent statistician then created a second binary variable to indicate treatment compliance. As per the SAP, this was defined as attending at least nine sessions of CBT if allocated to the CBT + SMC arm (1 = nine or more CBT sessions attended, 0 = fewer than nine CBT sessions attended). This number of sessions was chosen to be consistent with our pilot RCT.1 The independent statistician then ran a chi-squared test to assess whether or not treatment compliance within the intervention arm was predictive of missing 12-month primary outcome data. The chi-squared test confirmed this association to be statistically significant (p < 0.001): 94% (131/140) of participants in the intervention arm who were compliant with CBT provided primary outcome data at 12 months, compared with 54% (25/46) of those who were non-compliant. Therefore, as described in the TSC-agreed and published SAP,92 multivariate imputation via chained equations (MICE) was deemed appropriate and necessary for the main analysis to produce inferences valid under the detected missing-at-random (MAR) data-generating process. This meant that missing outcome variables at all time points were imputed instead of using the mixed-modelling approach originally suggested in the brief (pre SAP) analysis section of the protocol.84 Multiple imputation (MI) consists of an imputation step and an analysis step. First, missing values in specified variables are multiply imputed. Second, an analysis model is fitted to each imputed data set and the analysis results are combined using Rubin’s rules.131 This two-stage procedure provides inferences that are valid under a MAR missing data mechanism. Importantly, as defined by the imputation model, the observed variables are allowed to drive missingness at 12 months.

To inform the imputation model, logistic regression methods were used to detect which baseline variables were associated with missing follow-up at 12 months in each randomisation stratum (sites). Almost all baseline variables were included in this process, with just a few exceptions for the following reasons: (1) if measured only in small subgroups (IMD quintile Wales, type of dependant, type of carer and status of previous epilepsy diagnosis); (2) if small subgroups were merged to reduce the chance of perfect prediction (ethnicity was grouped into white, black or other; relationship status was dichotomised into married or living with partner vs. single or other; binary variables were used for ‘any current M.I.N.I. diagnosis’ and ‘any previous M.I.N.I. diagnosis’); or (3) if measured more than once (continuous score of SAPAS-SR was used, so the binary version was not included).

First, unadjusted logistic regression was implemented, with the binary follow-up variable as the outcome and the baseline measures were each tested separately. Seven variables were found to be significantly associated at the α < 0.05 level: modified PHQ-15 score (p = 0.002), at least one current M.I.N.I. diagnosis (p = 0.003), number of days seizure free in the last 6 months (p = 0.005), having a carer (p = 0.007), relationship status (p = 0.023), previously sought help for a mental health problem (p = 0.035) and SF-12v2 Mental Component Summary score (p = 0.047).

Second, manual forward stepwise regression was used, with a liberal inclusion threshold of α < 0.1. The process was as follows: a logistic regression on binary follow-up was run fixing psychiatrist site as a covariate because randomisation stratified on this; each of the seven variables above, found to be univariately associated with missing primary outcome data, was then included one by one in a step-wise fashion; all of the p-values were compared and the variable with the lowest p-value (if p < 0.1) was added. This process was repeated until no further variables could be added (i.e. p > 0.1).

The variables that were found to be associated with loss to follow-up (within sites) were number of somatic symptoms on the modified PHQ-15 – those with fewer symptoms were more likely to provide primary outcome data (17 vs. 19.7 median symptoms; p = 0.0030); carer – those with a carer were more likely than those without a carer to provide primary outcome data (91.3% vs. 80.8%; p = 0.0021); relationship status – those who were single were more likely than those married/living with partner to provide primary outcome data (89.6% vs. 81.0%; p = 0.0078); number of days seizure free in the last 6 months in the study – those with longer periods of time without a seizure were more likely to provide primary outcome data (9.5 vs. 7 median days; p = 0.0078); and previously sought help for a mental health problem – those who had never sought help were more likely than those who had to provide primary outcome data (90.6% vs. 82.2%; p = 0.037).

Once the baseline predictors of missingness had been established, the MICE procedure132 to impute missing values for each outcome could continue. We generated 100 imputed data sets for each outcome and combined the analysis results according to Rubin’s rules,131 where each imputed data set was analysed according to the corresponding analysis model.

The primary outcome imputation model included dummy variables for treatment compliance in the CBT + SMC arm and included the five baseline predictors of dropout, as above. All dummy variables were coded 0 and 1. The imputation model also included all variables of the analysis model because MI theory requires all variables of the analysis model to be included in the imputation model. Any previous (or ‘auxiliary’) measures of the primary outcome (e.g. baseline and 6 months) were included to ensure that the observed values of the same outcome measured at other time points could contribute to the prediction of missing values in the MICE procedure. Unless complete, all auxiliary variables were imputed. For the primary outcome, incomplete auxiliary variables were 6-month seizure frequency, longest period of time seizure free in the last 6 months and baseline modified PHQ-15 score.

The three count variables (baseline and 6-month seizure frequency, and days of seizure freedom) were log-transformed prior to the MI step (plus 1 to avoid logging zero). The reason for modelling the longest period of seizure freedom (days) in the last 6 months on the log-scale was because of its inverse relation to the primary outcome and, although it refers to a 6-month period, there is no maximum value owing to its self-reported nature. Baseline and 6-month seizure frequency were also log-transformed because the relationships between them and 12-month seizure frequency are multiplicative, translating into an additive effect of log-counts on the linear prediction scale. Predictive mean matching was used to impute both incomplete count variables because of zero-inflated distributions.

Continuous outcomes were imputed using linear regression. Psychiatry sites were included as random intercepts in the analysis model; therefore, we included dummy variables for fixed effects of psychiatry site in the imputation model. This was acceptable because a fixed-effects model is more general than a random-effects model. To avoid perfect prediction and overparameterisation, some small categories were merged within dummy variables (e.g. one site had only one participant and another had three participants).

Both the imputation and the analysis model assumed a negative binomial distribution because monthly seizure frequency (count) was very overdispersed (i.e. the variance was much greater than the mean). Otherwise, a Poisson model may have been appropriate. Count outcomes are constructed using the number of events (count) over an exposure period (time): for us, this corresponded to the number of seizures over number of days (7, 14, 21 or 28 days from weekly diaries).

For the primary outcome, the analysis model hence included trial arm and baseline monthly seizure frequency as the independent variables for the following reasons, respectively: to formally assess treatment arm difference and because baseline values of the outcome variable are known to be predictive of the post-randomisation outcome. Psychiatry site needed to be conditioned on because it was the randomisation stratification factor, and we chose to model the effects by site-varying random intercepts rather than by fixed effects. This approach allowed us to estimate the treatment effect in the population from which the trial sites are a sample, which provides better generalisability than if we had used a fixed-effects approach in this multicentre trial.

We also considered the random effects for SMC doctors and CBT therapists (in the CBT + SMC arm). SMC doctors (generally psychiatrists) were nested within psychiatry sites, which meant that there was commonly only one or two doctors per site. Therefore, SMC doctor effects were not distinguishable from site effects and were not included. Therapist effects were assessed empirically and, again, no evidence for their existence was found. This was evaluated using complete cases (CCs): a likelihood ratio (LR) test was run for each outcome that compared the log-likelihoods of the models including therapist effects (in the intervention arm) with the models that did not include therapist effects. None of the LR tests was significant at the α < 0.1 level and half of the tests reported a p-value equal to 1.0000 [LR χ²(1) = 0]. For some outcomes, the LR test could not even be run [i.e. p-value = not applicable (N/A)] because the model with the therapist effects could not converge; our interpretation here was that therapist effects were not helping to explain the variability in the outcome. Most therapists were also nested within psychiatry sites (in the CBT + SMC arm only), so this may have influenced the LR tests.

Given the random effects for site, the primary outcome analysis model was a mixed-effects negative binomial model that was fitted using the Stata command ‘menbreg’. The dependent variable (monthly seizure frequency at 12 months) was constructed as a count and exposure period, and the two independent variables were constructed as one dummy variable (CBT + SMC vs. SMC) and one log-transformed continuous variable (baseline monthly seizure frequency). As explained above, random intercepts for psychiatry site were included to account for common site experiences. The inbuilt Stata option ‘cmdok’ was used to allow estimation of the mixed-effects negative binomial model with multiply imputed data. Incidence rate ratios (IRRs) were reported.

All secondary outcomes were analysed following the same approach: using MI to allow for the detected MAR process. For all outcomes, the imputation model included the five predictors of missingness at 12 months; a treatment compliance dummy variable in the CBT + SMC arm; a trial arm dummy variable; psychiatry site dummy variables, and any previous measures of the same variable. A few secondary outcomes were strongly correlated and had different missing patterns, so were imputed simultaneously (i.e. imputations for both outcome variables were generated from a single, larger, imputation model). This was the case for seizure severity and bothersomeness, and self-report and doctor-rated CGI scale score change.

For the one other count outcome, the longest period of time (consecutive days) seizure free in the last 6 months, a mixed-effects negative binomial model (with site-varying random intercepts) was also used. For the secondary outcomes that were treated as continuous, mixed-effects linear regression (with site-varying random intercepts) was used: seizure severity, seizure bothersomeness, Physical Component Summary score (SF-12v2), Mental Component Summary score (SF-12v2), health today (EQ-5D-5L VAS), impact on functioning (WSAS), anxiety (GAD-7), depression (PHQ-9), distress (CORE-10), other somatic symptoms (modified PHQ-15), CGI scale change self-report, CGI scale change doctor rated and satisfaction with treatment. Finally, mixed-effects logistic regression (with site-varying random intercepts) was used for the secondary binary outcomes: seizure freedom in the last 3 months of the study and > 50% reduction in seizure frequency.

Owing to the different scales of the secondary outcomes, all estimated treatment effects were standardised to aid comparisons. For outcomes with a baseline measure, this was calculated by dividing the estimated difference between arms on the original scale by the baseline SD or by the pooled SD of the outcome. None of the p-values was adjusted for multiple testing; therefore, interpretation of statistically significant (p < 0.05) secondary outcomes should be undertaken with caution.

Checking of regression assumptions and multiply imputed data

As written in the SAP,92 regression assumptions were checked for all 17 outcomes to ensure that imputed values looked sensible. First, to check the multiply imputed data, the 100 MI data sets were saved for each outcome and summary statistics were used to compare the imputed values with the observed data; namely, the minimum, mean and maximum values of m = 1 to m = 100 were compared with these summary statistics from the observed sample to check that they were reasonably similar.

Second, a subsample of MI data sets were used to perform MI diagnostics of all count or continuous outcomes: this consisted of every 10th MI data set (m = 10, 20, . . ., 90, 100). Kernel density estimates were plotted for these 10 MI data sets against the observed and completed data sets. This meant that the distributions of the observed, imputed and completed values could be compared graphically. Kernel density plots are similar to histograms in that the y-axis depicts the density function but, compared with a histogram that uses bins (or bars), they plot the distribution using smooth lines [similar to a Locally Weighted Scatterplot Smoothing (LOWESS) curve].

For the primary outcome and other count variables, the reason for using negative binomial models instead of Poisson models was the violation of the assumption that the variance is equal to the mean; for example, for our primary outcome data the mean was 37.6 and the variance was 8611.7. It was not possible to test whether or not each seizure was independent of each other, but it was reasonable to make this assumption.

For all continuous variables, the following regression assumptions were checked:

1. Homoscedasticity of residuals – the constant variance of residuals across all data points on regression line. This is checked by inspecting scatterplots of fitted values versus standardised residuals.
2. Linearity of independent variables – the linear relationship between dependent and independent variables. For those outcomes with a corresponding baseline variable that was adjusted for in the model, this was checked by inspecting scatterplots using Stata’s inbuilt ‘avplot’ command.
3. Normality of residuals – error terms display a normal distribution. This was checked by inspecting box plots of the residuals.

Agreement between auxiliary measures and sensitivity analysis

Given that there were two different methods of recording the primary outcome measure, the diary (which allowed a 1- to 4-week exposure period) and the single measure (which specifically asked about a 4-week period), we tested whether or not there was sufficient agreement between the two variable types at the primary outcome time point (12 months post randomisation). Both versions were log-transformed (after adding 1 to avoid logging 0) so that the agreement could be measured using log-frequencies (i.e. on the log-scale). An ICC was calculated by treating the log-transformed seizure diary (pro rata 4-weekly seizure frequency) as one ‘rater’ and the log-transformed single measure (seizure freedom question) as a second ‘rater’. The Stata command ‘kappaetc’ was used to measure inter-rater reliability.

In addition to calculating the ICC, a sensitivity analysis was run to check whether the size, direction or level of significance of treatment effect was ‘sensitive’ to how the primary outcome was constructed. To do this, the gold standard measure was treated as the only accepted measure for monthly seizure frequency: all participants who had not provided seizure diary data at 12 months but had provided the single measure were treated as missing follow-up. The same MICE procedure was then followed.

Complete-case and complier-average causal effect analysis

As well as the main ITT analysis and the sensitivity analysis, each outcome was analysed with just the CCs, that is without any MI and without adjusting for baseline predictors of missingness. The same analyses models as for the MI were used. This was carried out as a ‘reality check’ to see whether or not the results differed in comparison with those derived from the imputed data; if a difference was found in terms of the statistical significance, direction of effect or size of effect, then it may highlight the bias that has been corrected for by the MICE procedure.

Similarly, to assess the efficacy of the CODES intervention (CBT) in the presence of non-compliance, we ran a complier-average causal effect (CACE) analysis for the primary outcome, as stipulated by the SAP. This analysis models the effect of receiving the intervention (at least nine sessions of CBT) and aims to estimate the efficacy of the trial, rather than the effectiveness, which is estimated by ITT analysis. Treatment receipt regardless of trial arm allocation was used as the explanatory variable of interest. The same imputation step was used as for the main ITT analysis, but this time the log-transformed primary outcome was modelled and an instrumental-variables regression was run using the two-stage least squares estimator.