Research design
The study is a parallel-group, superiority, RCT of dCBT (+ treatment as usual, TAU) versus SHE (+TAU). The trial design is summarised in Fig. 1. The study will be carried out completely online. Participants will access the screening assessment, participant information statement, informed consent form, allocation to condition and intervention, and baseline and follow-up assessments via web or mobile platforms (see also Additional files 1 and 2). Participants will be able to print a copy of the participant information statement and will have the opportunity to contact the researchers via phone or email before consenting. The study has received ethical approval from the University of Oxford Medical Sciences Inter-Divisional Ethics Committee (ref MS-IDREC-C2-2015-024) and is registered at ISCRTN (ISRCTN60530898).
Participants
We will recruit 1000 community participants. Our inclusion criteria comprise: (1) a positive screen for probable DSM-5 insomnia disorder, (2) a test score of ≤ 16 on the Sleep Condition Indicator (SCI) [35], (3) being aged 18 or older (no upper age limit), (4) having reliable Internet access at home or at work, and (5) being able to read and understand English. We will screen for co-morbid conditions and medication use at baseline but exclude only those people whose health may be considered to be unstable such as significant current symptoms of (a) an additional sleep disorder (e.g. excessively sleepy and possible obstructive sleep apnoea), (b) psychosis or mania, (c) serious physical health concerns necessitating surgery or with a prognosis less than 6 months, (d) those undergoing a psychological treatment programme for insomnia with a health professional, and (e) habitual night shift, evening, or rotating shift-workers. Information covering (a) to (e) will be gathered using online self-report questions, there will be no medical examinations or access to medical records. We will not omit participants who take medication for sleep problems, or for any other physical or mental health problems providing they report their health to be stable. The study will recruit through several channels. These may include online, print and broadcast media announcements or advertisements and the use of contact lists where adults who have volunteered to be involved in research will be re-contacted (See Additional file 3). For example, following completion of open access sleep surveys such as the Great British Sleep Survey (GBSS: www.greatbritishsleepsurvey.com or World Sleep Survey, WSS: www.worldsleepsurvey.com, which generate data on tens of thousands of respondents). Potential participants will also be alerted to the study by information placed on the Sleepio® website (www.sleepio.com) and on the Sleepio® App site (which generate data on hundreds of thousands of respondents). Participants will not receive reimbursement for their participation, but will be provided with access to the dCBT programme free of charge, regardless of allocation to experimental condition. Informed consent will be obtained from all participants.
Randomisation and allocation concealment
This study will use simple randomisation with an allocation ratio of 1:1, as recommended for large clinical trials [36]. It will be carried out by the automated online system. Hence the research team will be unable to influence randomisation, and will have no access to future allocations.
Blinding
This is a single-blind trial. Self-report assessments will be completed online and hence the research team will be blind to outcomes during the trial. Participants will be informed of their randomisation outcome by an automatic email, and so they will not be blind to treatment allocation. The research team will have limited contact with research participants and will be unable to bias the allocation or influence the assessments. If participants do contact the team and reveal the allocation, the assessments will remain blinded. Statistical analyses will be conducted by an independent researcher (RE; University of Manchester) who will have access to all data.
Assessment points
Assessments will take place at weeks 0 (baseline), 4 (mid-treatment), 8 (post-treatment), and 24 (follow-up). In consideration of ethical matters, at week 25 all participants in the control group will be offered dCBT to help with their sleep problems, and so at that point the controlled element of the trial will be complete. Thereafter, there will be a naturalistic follow-up. All participants will be invited to complete further assessments at weeks 36 and 48.
Planned intervention
Digital cognitive behavioural therapy will be delivered using the Sleepio® programme [34] (www.sleepio.com and associated Sleepio® App). The programme is fully automated and its underlying algorithms feed the delivery of information, support, and advice in a personally tailored manner. Delivery is structured into six sessions, lasting an average of 20 minutes each. Certain tools (such as sleep diaries and relaxation audios) can also be accessed using the web browser of any smartphone. All of the six core sessions, sleep diaries, relaxation audios, and the scheduling tool can also be accessed using an iOS App, but this is only an option for participants who have an iPhone®. The treatment content is based on CBT for insomnia manuals [26–28] and includes a behavioural component (sleep restriction, stimulus control, and relaxation), a cognitive component (paradoxical intention, cognitive restructuring, mindfulness, positive imagery, and putting the day to rest) and an educational component (psycho-education and sleep hygiene).
The programme is highly interactive, and content is presented by an animated virtual therapist. Participants make a time for the session and are prompted via email and/or short text message (SMS) if they do not ‘attend’. Participants complete daily sleep diary information throughout the intervention, which is used by the programme to provide tailored, personalised help. Participants receive an email and/or SMS reminder each morning to prompt them to fill in their sleep diary. In addition, participants complete a short questionnaire at the beginning of therapy to set treatment goals. Throughout the course of therapy, participants have access to a moderated online community and an online library of information about sleep. Participants can view their online case file, which includes four sections: a progress review, a reminder of strategies to try out between sessions, an agreed sleep schedule, and a list of further reading. The system provides online analytics, which can be used to monitor adherence by assessing how many sessions were completed and the number of weeks to complete the course. All information gathered for the programme will be stored in encrypted form on secure servers. Passwords are stored in encrypted form and all sensitive traffic is transmitted securely via SSL by default. Participants will have access to the intervention for up to 12 weeks. Digital CBT will in effect be dCBT + TAU because there will be no requirement for participants to alter their usual care in any way. Physicians, for example, will be free to offer appointments, to prescribe, and to maintain/discontinue prescriptions as they see fit.
Sleep hygiene education (SHE) has been selected for the control arm because this is what people with insomnia are offered most typically in routine care. To ensure consistency of approach and content, SHE will be delivered on a dedicated website where materials can be viewed and downloaded. SHE will be based on recognised sleep hygiene advice [37–39] and will comprise behavioural advice concerning both lifestyle factors and environmental factors associated with sleep and sleeplessness. The latter in particular will focus on creating the optimal bedroom environment for good sleep. The content of SHE will cover the importance of limiting caffeine, nicotine, and alcohol and of carefully managing diet and exercise (lifestyle), as well as limiting noise and light, managing room temperature and body temperature, and improving air quality and bed comfort (environment). SHE will in effect be SHE + TAU because again there will be no requirement for the usual care of participants to be altered in any way.
Outcome measures
Participants will be prompted by email to complete the assessments in an online, secure environment (www.surveygizmo.eu). The order of the assessments will be consistent across all participants and all time-points. If participants do not complete measures within 2 days they will receive further email reminders. The full battery of questionnaires amounts to around 100 items in total, and takes 20–25 minutes to complete. We have evidence from a recently completed protocol that data on such measures can be successfully collected online [40]. Demographic and descriptive clinical data will be gathered at baseline only. Measurements to permit health economic evaluation will form part of the descriptive demographic data, with some aspects audited at each assessment point (e.g. medication use, visits to health professionals, other health care utilisation). Descriptive data on health behaviours such as smoking and alcohol use will also be reported.
The co-primary measures that relate to functional health and well-being will be the Patient Reported Outcome Measurement Information System: Global Health scale [41] (PROMIS-10), the Warwick-Edinburgh Mental Well-being Scale [42] (WEMWBS) and the Glasgow Sleep Impact Index [43] (GSII). The PROMIS-10 is a reliable (α ≥ .92) but brief (10 items), generic measure that has proven to be very useful in measuring outcomes in clinical trials (http://www.nihpromis.org/science/). It can also be used to estimate cost-utility using QALYs (quality-adjusted life-years) [44, 45]. The WEMWBS is a short (14 items) and psychometrically robust measure (α ≥. 91) of mental well-being and is included because it focuses entirely on positive aspects of functional mental health. A previous pilot trial suggests the WEMWBS is sensitive to changes after CBT for insomnia [46]. The GSII is a patient-reported outcome (PRO) measure that asks patients to individually generate, and then assess, three domains of sleep-related impairment unique to their own individual context. Its strength is ecological validity and the GSII has been shown to be sensitive to change following CBT [43]. This combination of PRO, generic functional health status, and positive mental state (rather than symptom reduction) matches our intention to evaluate the impact of dCBT upon quality of life domains.
Secondary outcomes relate to specific measurement of the six areas of daytime consequence that are associated with the clinical diagnosis of insomnia disorder [1, 14, 47]. These are mood (Patient Health Questionnaire (PHQ-9: 9 items [48]) and Generalised Anxiety Disorder (GAD-7: 7 items [49])); energy (Flinders Fatigue Scale (FFS: 7 items [50]); relationship satisfaction (Relationship Assessment Scale (RAS: 7 items [51]); cognitive functioning (Cognitive Failures Questionnaire (CFS: 25 items [52])]; work performance and satisfaction (Work Productivity and Activity Impairment questionnaire: Specific Health Problem (WPAI:SHP: 6 items [53]), 1 item on job satisfaction [54]); and sleepiness (Epworth Sleepiness Scale (ESS: 8 items [55]). As an exploratory measure, participants will also complete 1 item about their general life satisfaction [56].
In order to appraise the mediating effects of sleep improvement per se we will use the SCI [35] and estimates of sleep diary parameters [57]. The SCI is an internally consistent (α = .86) measure with a clinical cut off that can correctly identify 89 % of those with probable DSM-5 insomnia disorder. The SCI and sleep diary variables have proven to be sensitive to change following dCBT [34].
In addition to these formal assessments the web/mobile platform will provide online analytics for the dCBT group. These can be used, for example, to measure the process of change (sleep diary) and to monitor how many sessions were completed and the number of weeks to complete the course. These will be used in exploratory analyses. We will also gather information on the demographics of the sample, employment, work satisfaction and economic outcomes, their health characteristics, and their use of clinical services during the period of their trial participation.
Assessment of safety
The likelihood of serious adverse events occurring during this trial is low since dCBT for insomnia has not been reported to cause them. The intervention offered in the trial has previously been tested in a RCT testing change in insomnia and no adverse outcomes were reported [29, 34]. However, studies have shown that daytime sleepiness and vigilance impairment may increase during sleep restriction therapy (SRT) (one component of CBT-I), owing to restricted sleep opportunity [58]. We will record the occurrence of any serious adverse events in trial participants, defined as: (1) all deaths, (2) suicide attempts, (3) serious violent incidents, (4) admissions to secure units, and (5) formal complaints about the online intervention. Owing to the online nature of the assessments and intervention, it is unlikely that the research team will become aware of all such events. At the end of treatment we will also ask participants to complete an adapted version of a previously employed adverse effects measure [59] to assess differential rates of self-reported adverse effects.
Sample size calculation
Our planned primary intention-to-treat analyses will compare dCBT + TAU versus SHE + TAU for each of the three primary outcomes separately. Assuming a significance level of 1.667 % (adjusted from 5 % because of having three primary outcomes) and a power of 90 %, to detect a standardised effect size of 0.25 we require a minimum of 433 participants in each of the groups in the analysis. Accounting for a conservative dropout rate of 13 %, we will recruit 500 participants in each treatment group, or 1000 participants in all. This sample size will have more than 80 % power to detect a large-sized indirect effect through the sleep mediator (proportion mediated approximately 75 %) for each between-group comparison. We believe that a sample of this size is feasible to recruit for two reasons. First, online/mobile interventions are much more accessible than in person interventions; and second, published research using dCBT has already shown that large samples may be recruited over comparatively short recruitment periods.
Statistical analysis
All analyses will be carried out using Stata [60]. In accordance with Consolidated Standards of Reporting Trials (CONSORT) guidelines, we will report all participant flow. Descriptive statistics of recruitment, dropout and completeness of interventions will be provided.
The main efficacy analysis will be via intention-to-treat including all participants, with no planned interim analysis for efficacy or futility. Baseline characteristics will be presented by randomised group without formal statistical tests. We will test the primary hypothesis for between-group change in the primary outcomes at 8 weeks using analysis of covariance with baseline outcome measure and treatment assignment as fixed effects, and apply standard regression diagnostics. The analysis will use statistical techniques for handling missing outcome data under a missing-at-random assumption. The secondary outcomes will be analysed using an analogous method, as will subsequent measures of the primary outcomes at 24 weeks. Analysis of all treatment effects will be undertaken after all 24-week outcome measures are completed.
We will use modern causal inference methods to investigate the mediation hypothesis [61]. If the efficacy analysis shows significant between-group differences in the SCI at 4 and 8 weeks, then we will use parametric regression models to test for the indirect effect of SCI on outcomes, and the residual direct effect of treatment on outcomes at 8 and 24 weeks respectively. Since all the measures are continuous, the indirect effects are calculated by multiplying relevant pathways and bootstrapping is used to produce valid standard errors for the indirect effects. All analyses will adjust for baseline measures of the SCI, outcomes and putative measured confounders. Mediation analyses are potentially biased by measurement error in mediators and hidden confounding between mediators and outcomes and we will investigate the sensitivity of the estimates to these problems.
Dissemination policy
Results from this study will be published in peer-reviewed scientific papers, and presented at major national and international clinical scientific meetings. Findings, wherever possible, will be made accessible online, dependent upon journal policies.