The trial has received ethical approval from an NHS Research Ethics Committee (South Central – Oxford B Research Ethics Committee; ref 15/SC/0508) and has been registered (Current Controlled Trials ISRCTN18705064). Informed consent will be obtained from all participants. A Data Monitoring and Ethics Committee (DMEC), Trial Steering Committee (TSC), and Patient Advisory Group (PAG) have been formed.
Randomisation and blinding
The trial assessor will be blind to group allocation, but the patients and trial therapists will not be. Patient consent and assessments will be carried out by the trial assessor. Randomisation will occur after completion of the baseline assessment. An online randomisation system has been written by the University of Oxford Primary Care Clinical Trials Unit. Randomisation using a permuted blocks algorithm, with randomly varying block size, will be stratified by therapist. Therapists will provide both interventions in order to reduce the confounding of therapist effects and increase statistical power. The trial co-ordinator will use the online system, after being provided by the trial assessor with basic patient details (date of birth, gender).
The trial coordinator will inform trial therapists who will then inform patients of the randomisation outcome, so that the research assessors remain blind to group allocation. Precautionary strategies to prevent breaks of the blind include the following: the patients being reminded by team members not to talk about treatment allocation; the assessor not looking at the patient’s clinical notes after the baseline assessment; and if an allocation is revealed between assessment sessions, then re-blinding with another assessor. We envisage concealment of treatment allocation from the trial assessor will be easier than treatment-as-usual comparison trials because all patients are receiving a psychological intervention from the same therapists.
Assessments
Basic demographic and clinical data will be collected (for example, age, gender, ethnicity, and clinical diagnosis). The primary outcome measure will be conviction in the persecutory delusion (using a 0 to 100 % scale), assessed within the Psychotic Symptoms Rating Scale-Delusions scale [
23]. Recovery is defined as the conviction in the delusional belief falling below 50 %; that is, there is greater doubt than belief in the delusion. Conviction greater than 50 % is a standard definition of the presence of a delusion (for example, [
24]), although such beliefs are typically held with much greater certainty. For example, in our Feeling Safe Programme pilot study (n = 12), the initial conviction levels in the delusions showed a mean of 90 % (SD = 17) [
20], and in a previous study with 100 patients with delusions, the mean conviction rating was 82 % (SD = 20) [
25]. In the Worry Intervention Trial, at baseline, half of the 150 patients had 100 % conviction in the persecutory delusions [
13].
Psychological well-being will be assessed by the Warwick-Edinburgh Mental Well-being Scale [
26], health status by the EQ-5D-5 L (see
http://www.euroqol.org/), quality of life by the Long Term Conditions Questionnaire (LTCQ) [
27], and patient satisfaction using an adapted version of the CHOICE, a service user-led outcome measure [
28]. Activity levels will be assessed using a step count and a time-budget measure [
29]. We will also include measures of overall paranoia (Green et al. Paranoid Thoughts Scale) [
30], suicidal ideation (Columbia-Suicide Severity Rating Scale) [
31], and depression (Beck Depression Inventory) [
32].
We will include the following as moderators: working memory [
33], illicit drug use [
34], and anger (Dimensions of Anger Reactions (DAR-5)) [
35]. For mediation, we will include the following: the Penn State Worry Questionnaire [
36], Brief Core Schema Scales [
37], Specific Psychotic Experiences Questionnaire - hallucinations subscale (SPEQ) [
38], Insomnia Severity Index [
39], jumping to conclusions [
40] and belief flexibility [
18], and the Safety Behaviours Questionnaire – Persecutory Beliefs [
41]. We will record service use and other relevant health economic data using an adapted version of the Economic Patient Questionnaire [
42] (EPQ) that includes questions from the Client Service Receipt Inventory [
43].
In collaboration with the McPin Foundation, qualitative interviews will be carried out with a small number of patients and family members about the Feeling Safe Programme to assess the acceptability of the experimental intervention.
Psychological interventions
Both treatments are provided to patients individually in approximately 20 sessions over 6 months. Treatments will be provided by the trial clinical psychologists, with weekly supervision. The number of sessions and length will be recorded, sessions will be taped when patients are agreeable, and tapes will be rated for fidelity and competence. Patient beliefs about the potential effectiveness of the intervention that he or she receives will be assessed after the first session with the Credibility/Expectancy Questionnaire [
44], and therapeutic empathy will also be assessed with a patient questionnaire [
45].
In The Feeling Safe Programme, following an assessment, the patient is offered a menu of appropriate treatment modules. Typically three to four modules are completed, based on patient preference. The range of modules that can be offered are improving sleep, reducing worry, increasing self-confidence, reducing the impact of voices, improving reasoning processes, and behavioural tests for reducing fear beliefs. Befriending, called in the trial ‘Feeling Safe and Supported’ will follow a protocol devised by one of the trial team members (DK) that has previously been used in two large clinical trials for patients with psychosis over 20 sessions [
21,
22]. Essentially, the aim is to simulate how a good friend would respond and involves a general focus on non-threatening topics (although patients are not actively dissuaded from talking about concerns), non-confrontation, empathy, and supportiveness.
Statistical and economic analysis plan
A full statistical analysis plan will be written by the trial statisticians (RE, GD) prior to any analysis being undertaken. We will report data in line with the Consolidated Standards of Reporting Trials (CONSORT) 2010 Statement (
http://www.consort-statement.org/consort-2010), showing attrition rates and loss to follow-up. All analyses will be carried out using the intention-to-treat principle with data from all participants in the analysis, including those who do not complete therapy. Every effort will be made to follow up all participants in both arms for research assessments.
Analysis will be conducted in Stata version 14 [
46]. Descriptive statistics within each randomised group will be presented for baseline values. These will include counts and percentages for binary and categorical variables and means and standard deviations, or medians with lower and upper quartiles, for continuous variables, along with minimum and maximum values and counts of missing values. There will be no tests of statistical significance or confidence intervals for differences between randomised groups on any baseline variable.
Descriptive statistics will be used to summarize assessments of feasibility and acceptability in terms of recruitment, drop-out, and completeness of therapy.
The primary hypothesis for change in the primary outcome measure, conviction in the persecutory delusion (using a 0 to 100 % scale) at 6 months, will be analysed using a linear regression model allowing for the baseline measurement of outcome, severity of delusion, therapist and treatment assignment as fixed effects. To compare rates of recovery (scores falling below 50 %), we will use logistic regression models instead of linear models. Secondary outcome measures will be analysed using the same modelling approach. This includes analysis of the primary outcome and secondary outcomes at 12 months.
The mediation analysis will investigate putative mediational factors using modern causal inference methods [
47,
48]. This involves using parametric regression models to test for mediation of the Feeling Safe Intervention on outcome through the putative mediators. Analyses will adjust for baseline measures of the mediator, outcomes, and possible measured confounders. We will include repeated measurement of mediators and outcomes to account for classical measurement error and baseline confounding, and where feasible, use instrumental variable methods (baseline covariate by randomization interactions as potential instruments) to investigate the sensitivity of the estimates to these problems and that of unmeasured confounding.
Moderators will be assessed separately by repeating the primary analysis models and including interaction terms between the randomised intervention and each moderator. The coefficient of the interaction term is a measure of whether the treatment effect differs between levels of the moderator.
Missing data on individual measures will be pro-rated if more than 90 % of the items are completed; otherwise the measure will be considered as missing. We will check for differential predictors of missing outcomes by comparing responders to non-responders on key baseline variables. Any significant predictors will be included in the analysis models. This accounts for missing outcome data under a missing at random assumption, conditional on the covariates included in the model. As a sensitivity analysis, we will assess whether treatment adherence is associated with missing data, and if it is associated, use inverse probability weights or multiple imputation to compare results.
An economic evaluation will estimate the cost per quality adjusted life year (QALY) gained from a health and social care perspective over the 1-year timeframe of the trial. An economic model will be used to explore the potential cost effectiveness of the intervention over the patient’s lifetime. A detailed analysis plan for the economic evaluation will be prepared by the trial health economist (LMD) prior to the analysis. This will be informed by exploratory analyses of the pooled baseline data and published literature.
For a recovery rate in delusions of 50 % in the Feeling Safe Programme, compared to 20 % with befriending, a study will have over 90 % power with 60 patients in each arm. The trial will, however, gain greater power by also examining change in delusion dimensional scores. If the standardised effect of the new intervention compared to befriending were smaller than 10 percentage points on the conviction scale (0 to 100 %) (d = 0.5), then we would not consider further development of the intervention to be worth pursuing. If the true effect size were this ten-point difference (SD = 20), then a two-sample t-test with a two-sided significance level of 0.05 would have 80 % power to detect a statistically significant effect with outcome data available for 64 participants per randomised arm. We aim to recruit 75 per arm. This conservatively allows for a drop-out of 15 %. Allowing for stratum membership and baseline levels of the measures in a more refined analysis of covariance will increase both statistical power and precision.