Pragmatic randomised controlled trial of guided self-help versus individual cognitive behavioural therapy with a trauma focus for post-traumatic stress disorder (RAPID)

Face-to-face CBT-TF

CBT-TF is one of the primary treatments for PTSD adopted by IAPT in England and psychological therapy services in Scotland and Wales. Cognitive therapy for PTSD (CT-PTSD),25 one of the CBT-TF implemented by IAPT, was adopted for RAPID. Participants received up to 12 face-to-face, individual sessions, each lasting 60–90 minutes. In-session treatment was augmented by assignments which participants were required to complete between sessions.

CT-PTSD involves identifying the relevant appraisals, memory characteristics, triggers and behavioural and cognitive strategies that maintain PTSD symptoms. These are addressed by: (1) modifying excessively negative appraisals of the trauma and/or its sequelae; (2) reducing re-experiencing by elaboration of the trauma memories through imaginal exposure or narrative-based memory updating with less threatening meanings and discrimination of triggers; (3) dropping dysfunctional behaviours and cognitive strategies, particularly those related to avoidance of triggers for intrusive symptoms; and (4) when possible, visiting the site of the trauma with the therapist to update the trauma memory.

GSH using Spring

Spring is an eight-step online GSH programme based on CBT-TF (see Table 1 for description of steps); it uses the same principles as CBT-TF but aims to reduce contact time with the therapist by providing some of the therapy content and activities in an online format. The therapist initially meets with the participant for an hour to develop a rapport, learn about the participant’s trauma, provide log-in details and describe and demonstrate the programme, which the participant then completes online in their own time. There are four subsequent fortnightly meetings of 30 minutes, normally undertaken face-to-face, but also deliverable via the internet or telephone, according to participant preference. The participant also receives four brief telephone calls or e-mail contacts between sessions to discuss progress, identify any problems that have arisen and agree new goals. The programme was designed to be accessible through a variety of devices including PC, laptop, tablet and smartphone (via a Spring App).

The eight Spring steps are accompanied by between-session work. At each session, the therapist reviews progress by logging into a clinician dashboard and guides the participant through the programme. The aim of the guidance is to offer continued support, monitoring, motivation and problem-solving. The eight online steps are usually completed in turn with some later steps relying on mastery of techniques taught in earlier steps. Each step provides psychoeducation and the rationale for specific components of treatment, and they also activate a tool that becomes live in the Toolkit area of the website and aims to reduce traumatic stress symptoms. Specific activities become visible (with the participant’s knowledge) to the therapist via the dashboard to facilitate discussions during guidance. The programme can be accessed online via a web browser or through an app.

Primary outcome

CAPS-5 (PTSD symptoms): The CAPS-525 is a 29-item structured interview for assessing PTSD diagnostic status and symptom severity. The CAPS-5 is the gold standard in PTSD assessment and can be used to make a current (past month) or lifetime diagnosis of PTSD or to assess symptoms over the past week. Items correspond to the DSM-5 criteria for PTSD. The CAPS-5 has excellent reliability and convergent and discriminant validity, diagnostic utility and sensitivity to clinical change. 25 Twenty of the twenty-nine items are used to create the score.

Secondary outcomes

IES-R (PTSD symptoms): The IES-R26 is a brief PTSD self-report measure and has been used in many international studies. The IES-R is the outcome measure of choice for evaluating improvement in PTSD symptoms in IAPT services in England.

EuroQol-5 Dimensions, five-level version (EQ-5D-5L) (quality of life): The EQ-5D-5L27 is a widely used instrument in health economic (HE) analysis and recognised by NICE as an appropriate measure for health-related quality of life. The questionnaire provides a simple descriptive profile, which translates to a single utility score for health status. The first part of the instrument identifies the extent of perceived problems – across five levels – in each of five life dimensions: mobility; self-care; usual activities; pain and discomfort; and anxiety and depression. The responses to each of the five questions are used to generate a value set score for self-rated health status between −0.594 and 1, where −0.594 represents the worst possible health state and 1 the best possible health state. Given the NICE Position Statement,28 this value set is a crosswalk of EQ-5D-5L to EQ-5D-3L. 29 The second part is a visual analogue scale, which allows the responder to indicate their current health status on a 0–100 scale.

Work and Social Adjustment Scale (WSAS) (functional impairment): The WSAS30 is a self-report measure, which assesses the impact of a person’s mental health difficulties on their ability to function in terms of work, home management, social leisure, private leisure and personal or family relationships. The WSAS is the outcome measure of choice for evaluating improvement in functioning in IAPT services. The WSAS has been demonstrated to show good reliability and validity and is sensitive to change.

Patient Health Questionnaire (PHQ-9) (depression symptoms): The PHQ-931 is a widely used reliable and well-validated brief self-report measure of depression. It is the outcome measure of choice for evaluating improvement in depressive symptoms in IAPT services.

Generalized Anxiety Disorder-7 (GAD-7) (anxiety symptoms): The GAD-732 is a widely used reliable and well-validated brief self-report measure of anxiety. It is the outcome measure of choice for evaluating improvement in anxiety symptoms in IAPT services.

Alcohol Use Disorders Test (AUDIT-O) (alcohol symptoms): The AUDIT-O33 contains 10 multiple-choice questions on quantity and frequency of alcohol consumption, drinking behaviour and alcohol-related problems or reactions over the preceding 3 months.

Multidimensional Scale for Perceived Social Support (MSPSS) (social support): The MSPSS34 is a widely used 12-item Likert scale measuring the subjective assessment of adequacy of social support from family, friends and partners. 35 The reliability, validity and factor structure of the MSPSS have been demonstrated with a number of populations. 34–37

Insomnia Severity Index (ISI) (insomnia): The ISI38 is a widely used seven-item self-report questionnaire assessing the nature, severity and impact of insomnia. It has been shown to be reliable and valid in terms of detecting insomnia and in measuring treatment response in clinical patients.

Post-Traumatic Cognitions Inventory (PTCI) (post-traumatic cognitions): The PTCI39 was developed as a 33-item scale, which is rated on a Likert scale ranging from 1 (totally disagree) to 7 (totally agree); a shortened, 22-item form has also been developed. Scale scores are formed for three subscales: negative cognitions about self, negative cognitions about the world and self-blame. The PTCI shows good internal consistency, high test-retest reliability and good convergent validity with other measures of trauma-related cognitions. The PTCI also shows promise in being able to differentiate individuals with and without PTSD.

General Self-Efficacy Scale (GSES) (self-efficacy): The GSES is a ten-item, 4-point Likert scale that is used to measure self-efficacy. It has been used in more than 1000 studies and is reliable and well-validated. 40,41

Client Satisfaction Questionnaire-8 (CSQ-8) (treatment satisfaction): The CSQ-842 is a widely used eight-item, Likert Scale which was developed through literature review and expert ranking, pretested on 248 mental health clients in five settings. It is a self-report statement of satisfaction with a high degree of internal consistency, good concurrent validity and reliability and is brief and easy to complete. 43

Agnew Relationship Measure-5 (ARM-5) (therapeutic alliance): The ARM-544 is a validated short 5-item version of the 28-item ARM-5, comprising client and therapist versions containing parallel items. There are versions for both the participant (shown first) and the therapist (shown second).

Analysis of the primary outcome

The primary analysis was performed using analysis of covariance (ANCOVA), modelling 16-week follow-up CAPS-5 score, controlling for baseline CAPS-5 score, research centre and the following patient characteristics: gender, comorbid depression (baseline PHQ-9) and time since trauma. Reflecting the sample size calculation, analyses were undertaken with two-level hierarchical models with patients clustered within therapists.

The primary analysis utilised multiple imputation with interim collected IES-R scores as auxiliary variables to the imputation. The number of imputations datasets created from which the analysis was averaged over was 50, which was greater than the percentage of incomplete cases (defined as a case missing the primary outcome) out of all those randomised. Given that IES-R was collected four to five times for GSH arm patients and eight to twelve times for CBT-TF arm patients, there was potential bias created by undertaking any multiple imputation model. For this analysis, we then applied a different imputation model to each arm: both containing the relevant number of auxiliary variables [along with baseline CAPS-5 score, research centre, gender, comorbid depression (baseline PHQ-9) and time since trauma]. Imputed datasets were then combined for the final analyses. Full details of the imputation procedure are given below. 49

The results were summarised using point estimates, and one-sided 95% confidence intervals (CIs) and NI p-values (in line with the sample size calculation). Since this is a NI design, we checked whether the CI for the difference between arms lay entirely within the 5-point NI margin for the primary outcome. Where the treatment effect and one-sided 95% CI were entirely > 0, then superiority was assessed with a two-sided 90% CI and relevant p-value.

The secondary outcome CAPS-5 at 52-week follow-up was analysed in the same manner as the primary analysis. For the other secondary outcomes the notional NI margin was set as 0.5 times the pooled SD of the baseline values in each of the CBT-TF and GSH groups of the outcome.

Covariate adjustment

All analyses contained the following covariates:

• treatment arm (categorical; GSH or CBT-TF)

• gender (categorical; male or female; minimisation variable)

• research centre (categorical; Cardiff and South Wales, Pennine, London, NHS Lothian; Coventry, South-West Yorkshire; stratification variable)

• comorbid depression (baseline PHQ-9)

• time since trauma in months.

In all cases, except those of CSQ-8 and ARM-5, the covariate of the relevant baseline version of the same measure was included in the model such that an ANCOVA model was formed. For PHQ-9, this covariate was only included in the model once.

Sensitivity analyses

For the primary outcome, the complete case intention-to-treat (ITT) analysis and per-protocol analysis were conducted and reported under a NI framework. Results are presented using point estimates, and one-sided 95% CIs and p-values (in line with the sample size calculation). Since this is a NI design, we checked whether the CI for the difference between arms lay entirely within the 5-point NI margin.

A further sensitivity analysis (SA) of the primary outcome under a NI framework implemented a different multiple imputation model: IES-R scores taken from five clinic visits for the CBT-TF arm participants that aligned similarly in time to those of the GSH arm participants were used as auxiliary variables in an imputation model (this one with both arms combined) along with baseline CAPS-5 score, research centre, gender, comorbid depression (baseline PHQ-9) and time since trauma. The results were summarised using point estimates, and one-sided 95% CIs and NI p-values (in line with the sample size calculation). Since this was a NI design, we checked whether the CI for the difference between arms lay entirely within the 5-point NI margin.

To explore the impact of departures from randomised treatment on our primary analysis, we estimated the complier average causal effect (CACE), with the following definition of ‘complier’ considered:

• Participants who attend the necessary number of therapy sessions to be described as having adhered (as defined above).

In this case, ‘compliers’ form a principal stratification strategy as defined in the ICH E9 Draft Addendum. 50 That is, by defining compliers as the stratum in the trial population that do not experience the post-randomisation intercurrent event of non-compliance.

We used instrumental variable methods to conduct these analyses, using randomisation as an instrument. 51–53 The models were fit using two-stage least squares instrumental variables regression, including those covariates used in the primary analysis model. The results were summarised using point estimates, and one-sided 95% CIs and NI p-values (in line with the sample size calculation). Since this is a NI design, we checked whether the CI for the difference between arms lay entirely within the 5-point NI margin.

In addition, for the 52-week outcome measures, we explored the impact of the COVID-19 pandemic by examining the impacts on the primary analysis by further stratifying by two time periods, namely, before the initial national lockdown date of 23 March 2020, and after that date. Since the trial was not powered to explore the interaction effect of the two time periods and the intervention, these analyses were primarily exploratory in nature. Further analyses were also performed examining the effect of changes in the mode of data collection from face-to-face to remote collection via video and telephone calls. This analysis was conducted on the primary outcome at 16 weeks as by the 52-week outcome the choice of method of data collection was confounded by the switch to telephone/video call due to the COVID-19 pandemic.

Analysis of secondary outcomes

Secondary outcomes IES-R, EQ-5D-5L (Value Set and Visual Analogue Scale), WSAS, PHQ-9, GAD-7, AUDIT-O, MSPSS, ISI, PTCI, GSES, CSQ-8 and ARM-5 were also analysed using multiple imputation to account for missing data under a NI framework. The number of imputations datasets created from which the analysis was averaged over was 50, greater than the percentage of incomplete cases (defined as a case missing that secondary outcome) out of all those randomised. The imputation model used the variables defined above. The results are presented as point estimates, one-sided 95% CIs and NI p-values. To allow comparison to be made across all outcomes the results are standardised to Cohen’s d effects size and the relevant point estimates, one-sided 95% CIs and non-inferiority p-values are expressed likewise.

The Cohen’s d effect size was calculated as the estimated mean difference from the regression model divided by the pooled SD calculated by pooling the SD measured in each trial arm at baseline for the relevant secondary outcome. The NI margin for secondary outcomes was set as 0.5 times the pooled SD. The value of 0.5 was chosen as this approximates the effect size used in the sample size calculation for the primary outcome of CAPS-5, that is, a 5-point margin divided by an assumed SD of 10.3. This gives an effect size 0.48 which was rounded to 0.5.

Exploratory analyses

A pre-specified subgroup analysis considered the differences in treatment effects by gender, including an interaction term between treatment arm and gender. Estimates from the statistical models (stratum-specific mean differences) are presented alongside two-sided 95% CIs and p-values.

Missing data

Individual questionnaires were inspected for missing values. If specific items were missing, they were imputed using the guidance for that questionnaire or if that information was not available then by mean imputation. This was a minor issue as most questionnaires were complete if the participant had been assessed.

Dealing with missing data for the primary outcome is listed above. For the multilevel multiple imputation, a joint modelling approach was chosen as implemented in the JOMO R package. 54 This uses a Bayesian approach utilising Monte Carlo Markov Chain to fit a joint two-level model. For the level 1 model this was fit as a random intercept and slope mixed regression model of the IES-R total scores with time. In addition, the following covariates were added to the model: site, baseline CAPS-5 score, gender, baseline depression score (PHQ-9) and time since traumatic event. The level 2 model was the primary analysis model consisting of a regression of the 16-week CAPS-5 score on the baseline CAPS-5 score, gender, baseline depression score (PHQ-9) and time since traumatic event. As described above, for the primary analysis each treatment group was imputed separately due to the varying number of therapist contacts.

For the SA, the same imputation model framework was used except the IES-R scores and other model variables were combined into a single imputation model. For the 52-week outcomes it was not possible to also include the site variable in the imputation model for each group due to data sparseness in the smallest sites. However, when compared to the sensitivity and complete case analyses the results were consistent.

For each multiple imputation run we used a burn-in period of 10,000 iterations, 500 iterations ‘in-between’ to account for serial correlation; and for all outcome variables 50 multiply imputed datasets were generated. The robustness of the imputation models was checked using a range of diagnostic statistics and convergence plots available in the JOMO package and described there. Based on these diagnostic measures there was no evidence of any issues with the imputation framework.

Additional analyses

IES-R scores over time were analysed using a hierarchical modelling fitting a random slope and intercept model, also allowing for clustering by therapist as in the primary analysis. We modelled the time dimension as a linear spline with a knot at 26 weeks. Both elements of the spline were included as random effects in the form of random slopes. We fitted IES-R trajectories over time (since randomisation) interacted with intervention arm, while also controlling for the same covariates as the primary analysis. Note that these were likely collected four to five times for GSH arm patients and 8 to 12 times for CBT-TF arm patients.

An analysis explored the impact of the COVID-19 pandemic by examining the treatment effect stratified by the period before and after lockdown, defined as any contact that took place after 23 March 2020. For participants with missing outcome data, we calculated the notional 16- or 52-week date of the expected contact with their therapist. This notional date was then used to define the before or after lockdown period.

Statistical software

All analyses were performed using Stata version 17 (StataCorp LP, College Station, TX, USA)55 and the JOMO R package54 software used for multilevel multiple imputation as described above.

Screening and randomisation

In all, 196 participants were recruited into the trial, 99 in the CBT-TF arm and 97 in the GSH arm. This exceeded the planned sample size of 192 participants. Figure 1 shows the CONSORT patient flow diagram. A more detailed breakdown of the reasons for inclusion and exclusion is shown in Table 2.

FIGURE 1.

Participant flow and CONSORT diagram for the RAPID trial.

Overall, 726 referrals were made to the RAPID team and 422 were telephone screened (58.1%). Of the 422 participants telephone screened, 311/422 (73.7%) were eligible for enrolment and 303/311 (97.4%) attended a baseline assessment. Of these, 196/303 (64.7%) were randomised to the trial. Of those not randomised, 58/303 (19.1%) participants chose not to continue the baseline assessment or did not attend the assessment. A further 48/303 (15.8%) patients did not meet the eligibility criteria for entry to the trial with the reasons being PTSD was attributed to more than one event or was a complex PTSD diagnosis (23/303, 7.6%); not having PTSD as a primary diagnosis (11/303, 3.6%); and 7/303 (2.3%) having a CAPS-5 score above the eligibility threshold of 49. In addition, three patients refused consent at baseline, three participants showed a preference for one trial arm and one patient had a medication change close to the baseline assessment.

Withdrawals during the trial

During the trial, 18/196 (9.2%) participants withdrew from the trial. Of the 18 patients, 12/18 (66.7%) were in the GSH group and 6/18 (33.3%) in the CBT-TF group. 2/18 (11.1%) participants were withdrawn at the request of the therapist and 16/18 (88.9%) withdrew at the request of the participant. Of the 18 participants who withdrew, 14/18 (77.8%) withdrew from the trial intervention but allowed themselves to be followed up, 2/18 (11.1%) withdrew from further data collection and 2/18 (11.1%) withdrew from both the intervention and further data collection.

In terms of timing of withdrawals, 6/18 (33.3%) withdrew prior to randomisation, 11/18 (61.1%) withdrew after randomisation but before the 16-week assessment and only 1/18 (5.6%) withdrew between the 16-week and 52-week assessments.

There was a difference in the number of withdrawals between the CBT-TF group and GSH group. There was one participant withdrawal at the request of the therapist in each group. The participant in the CBT-TF group had an ill family member, and no reason was recorded for the GSH participant. The remaining 16 participants (11 in the GSH group and 5 in the CBT-TF group) who withdrew at their own request gave a variety of reasons for withdrawing. Examining the free text comments revealed that three participants (one in the CBT-TF group and two in the GSH group) had no reason recorded. In the four participants who gave reasons in the CBT-TF group, one patient felt the therapy was not helping and the other three participants had difficulty attending the sessions for work-related or family illness reasons. For the GSH group, where a reason was recorded, three participants indicated a desire for more intensive therapy, one participant felt the intervention was too difficult to use, one participant preferred to do individual research, one participant was taking medication to treat their PTSD symptoms, one participant did not want to revisit their PTSD event, one participant could not commit the time and finally one participant had a cancer diagnosis and did not wish to juggle two health issues. Participant withdrawal information is presented in Table 3.

Lost to follow-up

There were 36/196 (18.4%) patients lost to follow-up at the 16-week assessment, 20/97 (20.6%) in the GSH group and 16/99 (16.2%) in the CBT-TF group. This includes those patients who withdrew permission for future data collection, one in the CBT-TF group and three in the GSH group. As expected, loss to follow-up was higher at the 52-week assessment with 57/196 (29.0%) not reporting data collection overall with 29/99 (29.3%) in the CBT-TF group and 28/97 (28.9%) in the GSH group. There was no strong evidence of a differential loss to follow-up between the CBT-TF or GSH groups.

Demographics

The average age of participants was 36.5 years (SD 13.4). Female participants represented 63.8% (125/196) of the total trial population. The overwhelming self-reported ethnicity was white (all groups) which comprised 91.8% of the trial population (180/196). In terms of education and qualification, 64/196 (32.7%) participants reported having a higher education qualification and 46/196 (23.5%) reported having two or more A levels. Only 8/196 (4.1%) reported having no qualifications or education achievements.

Most participants reported having salary or wages as their main source of income (123/128, 96.1%). However, just under 35% of participants declined to report their main source of income (68/196). Participants were more inclined to report their income category (only eight declined) and 110/188 (58.5%) participants reported income below £20,000 per annum. This is lower than the UK median income as reported by the Office for National Statistics (ONS) for 2019 which was £29,900. Of participants’ current occupations the three most common responses were customer service occupations (24/96, 12.2%), teaching and education profession (17/196, 8.7%) and healthcare professionals (16/196, 8.2%). While customer service occupations remained the most common response in terms of lifetime vocation (62/196, 31.6%) the next two categories were sales occupations (31/196, 15.8%) and administrative roles (30/196, 15.3%). Very few participants reported managerial or scientific roles currently or over their lifetime.

There were no major imbalances reported among the two groups indicating the randomisation process had been successful. There were a few smaller imbalances noted. More CBT-TF participants reported having achieved a higher education qualification (37/99, 37.4%) than GSH participants (27/97, 27.8%). Slightly more CBT-TF participants reported being employed than the GSH group, 63/99 (63.6%) versus 56/97 (57.7%). This was reflected in the proportion of participants not receiving benefits which was 69/99 (69.7%) for the CBT-TF group and 57/97 (58.8%) for the GSH group. There were a larger number of participants not able to work in the GSH group (12/97, 12.4%) compared to the CBT-TF group (6/99, 6.1%). Full details of the demographic profile of the trial participants are shown in Table 4.

Experience of trauma

Participants reported previous experience of trauma using the Life Events Checklist (LEC) tool. 56 Figure 2 shows the prevalence of the worst experience reported and Table 5 reports the full details of the LEC questionnaire. Participants reported on average being 32.7 (SD 13.9, range 5–69) years old at the time of their worst trauma experience and that the event had occurred a median of 16 [interquartile range (IQR) 6–32, range 2–720] months ago. Most participants responded that the event lasted a median of 1 hour (IQR 0.2–4); however, some participants reported much longer durations, range 0–720 hours.

FIGURE 2.

Prevalence of most traumatic event reported.

The most prevalent event reported as the worst experience was a transportation accident (33/196, 16.8%). The next most common category was serious accident (not transportation) (23/196, 11.7%) and then the sudden, unexpected death of someone close to the participant (22/196, 11.2%).

There were some small imbalances observed, for example, physical assault where 15/99 (15.2%) reported this in the CBT-TF arm compared to 6/97 (6.2%) in the GSH arm. Similarly, in the life-threatening illness or injury category only 5/99 (5.1%) participants in the CBT-TF arm reported this as their worst event experience versus 12/97 (12.4%) in the GSH arm. These are relatively small numbers and so some imbalances are to be expected by chance alone even under perfect execution of the randomisation method. There was also a small difference in the ‘Any other stressful event or experience’ but the version of the LEC questions used did not ask the participant to specify the exact event or experience.

Mental health

During the baseline assessment, participants were asked to report if they had ever been told by a health professional that they had a particular mental health diagnosis. Figure 3 shows the prevalence of the reported mental health conditions. The most prevalent condition was PTSD reported by 147/195 (75.4%) participants. Generalised anxiety was the next most prevalent, reported by 129/196 (65.8%) participants, and depressive disorder the third most prevalent with 112/196 (57.1%) reporting having been told they had this condition. These three conditions were the most prevalent by a long margin. The next most prevalent condition was panic disorder which was reported by 21/196 (10.7%) participants with the remaining conditions reported by 0.0–7.7% of participants. Many conditions which were asked about had no participants reporting as being diagnosed.

FIGURE 3.

Prevalence of mental health diagnoses reported at baseline assessment.

There were no serious imbalances observed between the two intervention arms although a small imbalance was observed in participants reporting depressive disorder with 60/99 (60.6%) in the CBT-TF group compared to 52/97 (53.6%) in the GSH arm. Table 6 shows the complete details of the participant responses.

Family history of mental health

Participants were asked to give details on their family history of mental health conditions based on whether a relative was a first-degree relative, defined as mother, father, son, daughter, brother or sister; or a second-degree relative which was defined as grandparents, grandchildren, half-brother or sister, aunts or uncles. Specifically, the relative must have been told by a health professional that they had one of the listed conditions.

Figure 4 shows the prevalence of each mental health condition separately for first- and second-degree relatives. In both cases the predominant condition among relatives was depressive disorder with 91/195 (46.7%) among first-degree relatives and 46/194 (23.7%) second-degree relatives. In the case of first-degree relatives, 26/195 (13.3%) reported having two or more relatives with depressive disorder (detailed results shown in Table 7). The second most prevalent mental health condition reported was anxiety, reported for 68/195 (34.9%) first-degree relatives and 33/193 (17.1%) second-degree relatives. In the case of first-degree relatives, participants reported 22/193 (11.4%) having two or more relatives with anxiety-related disorders.

FIGURE 4.

Prevalence of mental health conditions among first- and second-degree relatives of participants.

In the case of first-degree relatives, the next two most prevalent conditions were alcohol and drug dependence (25/195, 12.8%) and post-natal depression (24/195, 12.3%). For second-degree relatives the third and fourth most prevalent conditions were Alzheimer’s disease (27/196, 13.8%) and other types of dementia (25/196, 12.8%). This reflects the older age profile of the second-degree relatives which included grandparents and uncles and aunts.

There were no obvious serious imbalances observed between the GSH and CBT-TF groups as can be seen from Table 7.

Medical history

Participants were asked about their physical health, specifically if they had been diagnosed with particular conditions. The prevalence of these conditions is shown in Figure 5. The most prevalent condition reported was migraine headaches in 65/196 (33.2%) participants. A small imbalance was present between the CBT-TF arm (29/99, 29.3%) compared to the GSH arm (36/97, 37.1%). The next most common category was ‘Any other physical health problem’ reported by 52/196 (26.5%). This was collected as free text and an examination of these comments tabulated by intervention group revealed no patterns. Again, a small imbalance was observed between the groups with 22/99 (22.2%) in the CBT-TF arm compared to 30/97 (30.9%) in the GSH arm. However, a difference of seven or eight participants could arise from chance even under randomisation. Two conditions had prevalences larger than 20%, namely asthma (52/196, 26.5%) and head injury (40/196, 20.4%). Finally, three conditions were reported in more than 10% of patients: chronic pain 29/196 (14.8%); hypertension and high blood pressure 25/196 (12.8%); and inflammatory bowel disease 20/196 (10.2%). Table 8 lists the detailed prevalence results.

FIGURE 5.

Prevalence of physical health conditions reported at baseline.

Contact with the criminal justice system

Five participants reported contact with the criminal justice system in the past 3 months, three in the CBT-TF group and two in the GSH group. These contacts consisted of police interviews or short stays at the police station. There were no overnight stays in a police cell or court appearances. No psychiatric assessments were conducted during these contacts with the criminal justice system.

Quality of life (EQ-5D)

At baseline participants reported their quality of life using the EQ-5D instrument. The participants reported on their quality of life in five domains: mobility, self-care, ability to perform daily tasks, pain and anxiety. Participants are also asked to rate their overall quality of life on a scale of 0–100. For mobility, 140/195 (71.8%) participants reported no problems walking about. 45/195 (23.1%) reported slight or moderate difficulty walking about. In terms of taking care of themselves, 157/195 (80.5%) reported no problems, with 18.5% of participants reporting slight or moderate problems taking care of themselves. When asked about performing their usual activities, more participants reported issues, with only 67/196 (34.2%) reporting no problems with doing their usual activities. Having slight or moderate difficulties with usual activities were reported by just half of the participants (94/196, 48%). A further 14/196 (7.1%) indicated they could not perform their usual activities. A similar pattern was seen on the pain dimension with 88/196 (44.9%) reporting no pain issues and 75/196 (38.3%) reporting moderate to extreme problems with pain. Finally, practically all participants reported issues with anxiety or depression, with only 9/196 (4.6%) reporting not being anxious or depressed. Of the other participants, 113/196 (57.7%) reported being slightly or moderately anxious or depressed and 74/196 (37.8%) indicated a severe or extreme feeling of anxiety or depression.

When asked about ‘Your health today’ on a scale from 0 to 100, participants reported a mean score of 58.1 (SD 20.0). Scores were reported in the range 5–98 with a median of 60 (IQR 50–70). There were no notable imbalances observed between the CBT-TF and GSH groups in terms of quality of life. Table 9 gives a complete breakdown of the baseline EQ-5D.

Primary analysis

The primary outcome was the difference in CAPS-5 score at 16 weeks between the GSH group and CBT-TF group. At 16 weeks this included data on 160/196 (81.6%) participants. Note this exceeded the assumption of 80% retention used in the sample size calculation. Full data were available on all baseline variables used in the analysis models. The results of the primary analysis and SAs are shown in Figure 7.

FIGURE 7.

Results of the primary analysis of CAPS-5 score at 16 weeks.

Analysis adjusted for the following variables at baseline: CAPS-5 score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

The primary ITT analysis resulted in an estimated mean difference in CAPS-5 score at 16 weeks between the CBT-TF group and GSH group of 1.0 points with a one-sided 95% CI of (−∞, 3.9). Since the mean difference represented the adjusted CAPS-5 mean difference in the GSH group minus the CBT-TF group, the observed difference of 1.0 points can be interpreted as the GSH patients had on average, after adjusting for the pre-specified baseline covariates, a 1.0 point higher CAPS-5 than the CBT-TF group. Since higher CAPS-5 scores are considered a worse outcome, this means that GSH had a very small overall lower impact than CBT-TF. However, the one-sided 95% CI contained 0 and it was consistent with the null hypothesis of no difference in mean CAPS-5 scores at 16 weeks. The primary outcome was assessed on a NI basis with NI attained if the upper limit of the one-sided 95% CI was < 5, indicating a mean CAPS-5 outcome at most 5 points higher in the GSH group. Since the upper limit of the one-sided 95% CI here is 3.9, it excluded five, and so non-inferiority was attained, NI p = 0.012.

Sensitivity analyses

A series of SAs were performed to check the robustness of the primary analysis using only complete cases where the estimated mean difference was 0.8 points with one-sided 95% CI (−∞, 3.5). The complete-case analysis was also consistent with the achievement of NI. A per-protocol analysis also showed NI was achieved with an estimated mean difference of −2.1 and one-sided 95% CI (−∞, 0.9). Finally, the CACE analysis did not quite achieve NI with an estimated mean difference of 1.3 and a one-sided 9% CI (−∞, 5.6). However, the CACE analysis here was highly variable compared to the other analyses although having a consistent estimated mean difference. The two-stage estimation process appears to have inflated the underlying variability for this analysis. Finally, a SA that used a different imputation model also showed evidence of NI with an estimated mean difference of 0.6 and a one-sided 95% CI (−∞, 3.5).

Prevalence of PTSD

At each assessment, participants were assessed, based on the CAPS-5 score, if they met the criteria for a DSM-5 PTSD diagnosis. At baseline all participants had this diagnosis as a function of the inclusion criteria. Overall, assessed PTSD diagnoses declined over the course of the trial with 26/160 (16.3%) participants having a PTSD diagnosis at 16 weeks and 16/139 (11.5%) participants were assessed as having PTSD at 52 weeks. At both time points over 80% of participants were not assessed as having a CAPS-5-based assessment of PTSD.

At the 16-week assessment 12/83 (14.5%) of the CBT-TF group and 14/77 (18.2%) of the GSH group met the criteria for the PTSD diagnosis. However, there was little evidence that the proportions in each group differed (p = 0.5). Similarly, at 52 weeks there was little to suggest a difference between the CBT-TF and GSH groups (p = 0.3), with 6/70 (8.6%) in the CBT-TF group and 10/69 (14.5%) in the GSH group being assessed as having PTSD.

Interpretation

Considering the primary analysis and the additional SAs there is clear evidence that GSH is non-inferior to CBT-TF at 16 weeks and in all analyses the estimated mean difference was around 1 point on the CAPS-5 score. Considering the CAPS-5 has a range from 0 to 80 a 1-point difference is for all intents and purposes a trivial difference and this was shown in all analyses where the data were consistent with the null hypothesis of no difference between the two groups. In terms of prevalence of PTSD assessed by the CAPS-5 instrument there were substantial declines in both groups with little evidence to suggest GSH had significantly higher prevalence than CBT-TF.

Primary outcome at 52 weeks

The analyses of the primary outcome were repeated at 52 weeks and the results are shown in Figure 8. The ITT analysis showed a larger reduction in CAPS-5 score for the CBT-TF group with an estimated mean difference between the GSH group and CBT-TF group of 3.2 with one-sided 95% CI (−∞, 6.0). So, at the 52-week assessment the GSH group narrowly failed to show NI. The complete case analysis was similar with an estimated mean difference of 3.5 and one-sided 95% CI (−∞, 6.3). The multiple imputation SA narrowly showed NI with an estimated mean difference of 2.3 and one-sided 95% CI (−∞, 4.9). Similar results were also observed for the per-protocol and CACE analyses.

FIGURE 8.

Results of the primary analysis of CAPS-5 score at 52 weeks.

Analysis adjusted for the following variables at baseline: CAPS-5 score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

The interpretation at the 52-week assessment was that the CBT-TF scores were roughly 3 points lower on average than the GSH scores and that NI was not achieved although only very narrowly with the ITT analysis showing a one-sided 95% CI with an upper limit of 6 a single point above the NI margin of 5. An examination of the mean scores for each group (see Figure 11) showed that the GSH group sustained their improved CAPS-5 scores while the CBT-TF group continued to improve very slightly between the 16- and 52-week assessments.

Secondary outcomes at 16 weeks

Secondary outcomes at 16 weeks were analysed using the same statistical modelling approach as the primary ITT analysis. To ensure comparability the adjusted mean differences between the GSH and CBT-TF groups were converted to Cohen’s d effect size by dividing the adjusted mean difference estimates by the regression models by the pooled SD of the baseline outcome measures. These results are shown in Figure 9 with full details in Tables 12 and 13.

FIGURE 9.

Analysis results of secondary outcomes at 16 weeks.

Analysis adjusted for the following variables at baseline: baseline score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

As can be seen from Figure 9 most of the secondary outcomes showed a modest effect size with most outcomes exhibiting observed NI. The GSES was borderline over the NI margin of 0.5 SD [one-sided 95% CI (−∞, 0.51)]. Only the client satisfaction questionnaire (CSQ-8) showed clear lack of NI at 16 weeks with an effect size of 0.6, one-sided 95% CI (−∞, 0.87), NI p = 0.27 with the upper limit exceeding the NI margin.

Secondary outcomes at 52 weeks

Figure 10 shows results for the analysis of secondary outcomes at the 52-week assessment. In general, the results show a more positive outcome for the CBT-TF group compared to the GSH group.

FIGURE 10.

Analysis results of secondary outcomes at 52 weeks.

Analysis adjusted for the following variables at baseline: baseline score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

Two outcomes, the AUDIT-O scale and the MSPSS, were within the NI margin with effect sizes of 0.1, one-sided 95% CI (−∞, 0.3), NI p = 0.003; and −0.2, one-sided 95% CI (−∞, 0.1), non-inferiority p < 0.001, respectively. Two additional outcomes, EQ-5D-5L (quality of life) and the GSES, achieved borderline NI with an effect size of 0.2 m one-sided 95% CI (−∞, 0.497), non-inferiority p = 0.048; and 0.2, one-sided 95% CI (−∞, 0.49), NI p = 0.046, respectively. There was slightly less evidence for borderline NI for the WSAS, effect size of 0.2, one-sided 95% CI (−∞, 0.53), NI p = 0.07.

While none of the other secondary outcomes showed evidence of NI, most were inconclusive in terms of superiority, that is, consistent with the null hypothesis of no difference, between the CBT-TF and GSH groups at 52 weeks. Only two scales, the IES-R, p < 0.001, and GAD-7, p = 0.013, showed evidence of a difference in favour of CBT-TF over GSH at 52 weeks.

Change in mean CAPS-5 scores

Figure 11 shows the change in mean CAPS-5 score over time. In the CBT-TF group the mean CAPS-5 score reduced from 35.6 at baseline to 13.2, 95% CI (10.7, 15.7) at 16 weeks and 10.2, 95% CI (7.9, 12.6) at 52 weeks. For participants in the GSH group their CAPS-5 score reduced from 34.6 at baseline to 14.2, 95% CI (11.4, 17.0) at 16 weeks and a slight further reduction to 13.8, 95% CI (11.4, 16.1) at 52 weeks. This analysis needs to be interpreted slightly carefully as being a within-participant comparison; it may be subject to bias from regression to the mean. However, given that proviso it suggests that both the CBT-TF and GSH groups saw a substantial reduction in their CAPS-5 scores between baseline and the 16-week assessment. Furthermore, the CBT-TF group participants continued to have slightly improved CAPS-5 scores from the week 16 assessment to the week 52 assessment while the GSH group sustained their improvement in CAPS-5 score over the same period.

FIGURE 11.

Change in CAPS-5 mean scores over time.

Means at 16 and 52 weeks adjusted for the following variables at baseline: CAPS-5 score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

Subgroup analyses

There was only one pre-specified subgroup analysis which was to use the primary analysis to compare the difference in the CBT-TF and GSH groups between females and males. An interaction term between the gender and intervention variables was fit to the primary analysis model and the effect of CBT-TF versus GSH was determined separately for males and females. The results are shown in Figure 12. There was no evidence of a difference in the intervention effect for males versus females [difference 0.5, 95% CI (−6.7, 7.6), p interaction = 0.9] with the difference in CAPS-5 score at 16 weeks 0.7, 95% CI (−5.1, 6.5), for females and 1.2, 95% CI (−3.1, 5.5) for males. It was a similar result looking at the 52-week CAPS-5 score with difference of −1.2, 95% CI (−8.1, 5.6), p interaction = 0.7 and females having a difference in CAPS-5 score of 4.0, 95% CI (−1.5, 9.4), and males having a difference of 2.8, 95% CI (−1.5, 7.0). Although this subgroup analysis was pre-specified the trial was not powered to detect any interaction effects. The subgroup results are shown in Figure 12.

FIGURE 12.

Results of planned and exploratory subgroups.

Analysis adjusted for the following variables at baseline: CAPS-5 score, gender, research centre, comorbid depression (baseline PHQ-9) and time since trauma (months).

Analysis of mode of data collection

The trial data were collected from participants either in a face-to-face session or via a telephone or video call. There was interest in exploring if there were differences in the primary outcome when stratified by mode of data collection. It is important to note that only the 16-week assessment was analysed since at 52 weeks the model of data collection had moved to telephone or online due to the COVID-19 pandemic. In addition, there was insufficient information to reliably ‘impute’ the missing mode of data collection and so the analysis was confined to those participants with a 16-week CAPS-5 assessment. The results of this analysis are shown in Figure 12. There was little evidence that the outcome differed based on data collection mode, p interaction = 0.3. The GSH group had a slightly lower (more favourable) CAPS-5 score with a mean difference of −1.5, 95% CI (−7.0, 3.9). It was the reverse for the telephone/video call with the CBT-TF group having lower CAPS-5 scores with a mean difference of 1.8, 95% CI (−2.5, 6.0). However, for both modes of data collection the results are also consistent with there being no difference between the CBT-TF and GSH groups.

Impact of COVID-19 lockdown

Since the trial was partially conducted during the COVID-19 pandemic an exploratory analysis was conducted to examine the potential impact on the primary outcome of having an assessment performed after the start of the lockdown period in the first wave of COVID-19 infections. The lockdown date was 23 March 2020 and assessments conducted after that time were considered made in the ‘lockdown’ period and so potentially impacted by COVID-19. The analysis was conducted at the 52-week assessment as only 16 participants had 16-week assessments after 23 March 2020. The full imputed dataset was used for the analysis and participants with missing 52-week assessments were allocated to the ‘before’ or ‘after’ periods based on their estimated assessment date using their baseline assessment CAPS-5 assessment as the starting point.

Of the 99 CBT-TF participants, 46 (46.5%) either had or were scheduled to have a 52-week assessment after 23 March 2020. For the GSH group, 40/97 (41.2%) had or were scheduled to have a 52-week assessment after 23 March 2020. The results of the analysis are shown in Figure 12. There was little evidence of a difference in the effect of the intervention depending on the period it was assessed, p interaction = 0.2. The estimated mean difference between the 52-week CAPS-5 score between the GSH and CBT-TF group was 4.7, 95% CI (0.4, 9.0) higher in the GSH group compared to the CBT-TF group in the pre-lockdown period and 0.9, 95% CI (−4.0, 5.9) higher in the post-lockdown period. As previously mentioned, this was a post hoc analysis and so was not powered to detect any interaction effects.

Analysis of IES-R trajectories

During the trial, participants were asked to complete the IES-R at each therapist contact. These interim IES-R scores have been used to inform the multiple imputation model as described in Introduction. In addition, it was possible to use this longitudinal data to examine the change over time in the IES-R scores. A multilevel mode was fit using piecewise linear splines (see Introduction) to quantify these changes over time. Figure 13 shows the trajectories of the IES-R scores for the GSH and CBT-TF groups.

FIGURE 13.

Longitudinal IES-R scores.

As can be seen from Figure 13 there was a rapid decline over the first 16 to 20 weeks coinciding with participants’ involvement in the GSH or CBT-TF interventions. The multilevel model fits two lines; one over the period of 0 to 16 weeks and a second one from 16 weeks onwards. Figure 13 shows this gives a good fit but slightly overestimates the scores around weeks 12–16. More complex models did not resolve this completely and it was decided to stay with the two piecewise linear lines for interpretability. In the period 0 to 16 weeks on average participants in the CBT-TF group showed an estimated mean reduction (improvement) in IES-R scores of 2.2 points, 95% CI (2.0, 2.4), p < 0.001 per week. The GSH group had a slightly lower rate of improvement of 2.0 points, 95% CI (1.8, 2.3), p < 0.001 per week. However, there was little evidence of a difference between the GSH and CBT-TF groups during this period (p = 0.4). For the period after 16 weeks the CBT-TF group’s reduction slowed to 0.2 points, 95% CI (0.1, 0.3), p < 0.001 per week. For the GSH group the reduction flattened with a small uptick of 0.02 points, 95% CI (−0.1, 0.1), p = 0.7 per week. However, this was consistent with no increase in IES-R score and so for the GSH group they maintained rather than improved their IES-R score reduction. For the CBT-TF group they continued to improve their IES-R by a slower reduction after 16 weeks. There was evidence of a difference after 16 weeks between the CBT-TF and GSH groups’ rate of reduction of 0.2 points, 95% CI (0.1, 0.4), p = 0.002 in favour of the CBT-TF group. This finding is consistent with what was observed for the CAPS-5 score as shown in Figure 11.

Costs included in the health economic analysis

The health economic HE analysis considered the following:

• costs to the NHS of GSH implementation (including staff time for training and supervision)

• costs to the NHS of CBT-TF (including staff time for training and supervision)

• cost of additional therapy (psychiatrist, psychologist and counsellor appointments)

• cost of primary and community health and social care

• costs of secondary care (inpatient admissions and outpatient appointments, A&E visits).

Costs were reported for all available cases (for the most complete overview) with ITT population (using multiple imputation) for the CEA and CUA. All costs were expressed as 2020 Great British pounds (£), inflated and converted appropriately where required. 62 Costs were calculated at a participant level, multiplying the unit cost by the resource used at each time point. The cumulative implementation and health and social care costs at baseline, 16 weeks and 52 weeks were calculated for all available cases for both the GSH and CBT-TF arms (including 95% CIs, median and IQR). The intervention and health and social care costs were summated and mean difference per patient in costs (including 95% CIs) was calculated at 16 and 52 weeks.

Intervention costs

The time spent by therapists on face-to-face therapy (in person and online), phone calls, note taking and other administration were collected in the trial. Staff costs were estimated using published unit costs (see Appendix 1, Table 32).

The therapy reports were compared to final numbers of therapy sessions reported in the CRFs, with the greater number of sessions reported from either source assumed to be correct. If therapy reports were missing, then the mean time for face-to-face contacts from those reported for that therapy arm was applied. The number of calls included was again based on the therapy reports of the CRFs and the final numbers reported. If calls were recorded in either source it was assumed these were accurate and no additional calls were added. If no calls were reported for GSH therapy, but face-to-face sessions were reported, then one fewer phone call per face-to-face session was estimated to have occurred. Where no therapy reports were submitted, and final number of therapy sessions had not been reported in the CRFs, it was assumed no therapy took place.

GSH therapy was anticipated to take 8 to 10 weeks, self-help material was provided through the intervention website or app, with face-to-face appointments with a therapist, a 1-hour initial session and four further 30-minute sessions, with catch-up calls between these sessions. CBT-TF was delivered over 10–12 weeks as face-to-face therapy sessions lasting 60–90 minutes. In addition, some therapy sessions involve a trauma site visit.

CBT-TF was delivered by mental health professionals (nurses, occupational therapists, psychologists and social workers). Therapy staff were generally NHS Agenda for Change band 7, with some band 6 and band 8a giving CBT-TF therapy. For the base case analysis, it was assumed that all therapy staff were band 7. However, it is anticipated that GSH therapy will be provided in routine practice in primary care with lower-intensity therapy training at band 4/5 and this was tested in a SA.

Training and supervision costs

Resource use resulting from GSH and CBT-TF training (including materials, consumables and staff time for trainers and attendees) was estimated through interviews and direct communications with the clinical staff involved in the trial.

Website costs for Spring

The Spring GSH website and app were developed prior to the RAPID trial; therefore, development costs have not been included in the analysis. The website and app require regular maintenance, and administration support to add users; an estimate of the costs provided by the trial team was included in the analysis. No equipment was provided as participation in the trial required access to a computer, tablet or smartphone.

Cost of health and social care resource use

Healthcare resource use (including primary care consultations, out-of-hours care, outpatient appointments, day-cases, inpatient stays and A&E attendances) and social care (social worker appointments and home help visits) were collected using data from the CSRI to assess the differences in profile of health and social care use as a result of the intervention compared to control.

If one or more items in the CSRI were completed (values of ‘0’ or greater), the CSRI was assumed to have been fully completed and any missing items were imputed with zeroes. If items were missing, for example, if the number of appointments was completed but not the time, then the median time for those appointments recorded in either treatment arm was applied. If the CRF was marked as ‘not done’ or no data were recorded, data were considered missing and multiple imputation was used to account for missing data in the analyses.

The Personal Social Services Research Unit (PSSRU) unit costs for health and social care were applied to therapist time, primary and community care including social care by number of appointments and staff member visited (see Appendix 1, Tables 32–36). NHS reference costs were applied to outpatient appointments, day cases, diagnostic tests and imaging. Outpatient visits and day cases were costed individually according to the number of appointments, reasons for healthcare contact and speciality/department recorded in the trial CRFs (see Appendix 1, Tables 37 and 40). The unit costs for imaging and blood tests were applied, and where a diagnostic test was recorded but no details given, the average cost of pathology tests in the 2018/19 NHS reference costs was applied (uplifted to 2020 prices) (see Appendix 1, Table 39). The most up-to-date NHS reference costs published at the time of analysis were for 2018/19 and so were uplifted to 2020 using the PSSRU inflation index. 63 Inpatient attendances were costed according to number of admissions and length of stay on an acute medical ward, general ward, acute psychiatric ward, long-stay rehabilitation ward or a psychiatric rehabilitation ward, and additionally if a patient was admitted via A&E (see Appendix 1, Tables 38 and 41). A cost per bed day was applied to the length of stay for the general ward and acute medical ward; as bed days were not reported in the 2018/19 NHS reference costs, those from 2017/18 were used and uplifted. The cost per day for a long-term care home for mental health support for adults was identified for the long-stay rehabilitation. The cost per bed day for a secure mental health ward was identified for the acute psychiatric ward in unit costs for 2019 which were uplifted to 2020.

Cost of medication

Current medications were recorded at baseline. Unit costs were taken from the NHS Electronic Drug Tariff64 and British National Formulary65 (see Appendix 1, Table 31). Prescriptions were costed individually based on dose, treatment duration and frequency of use. As medications were only collected at baseline these costs have not been included in the analysis and are reported separately.

Missing data

The problems concerning missing data are particularly relevant to HE health economic analysis as the main outcomes are cumulative measures collected over the trial period. Missing items relating to health-care service use may underestimate the total costs, while missing outcome data may be correlated to effects as those individuals without information may be systematically different from those for whom all information is observed. 66 As such, using complete-case assessments and available case analysis only could result in meaningful data being excluded. We therefore adopted a multiple imputation approach within the incremental economic analysis as the appropriate technique to provide a comprehensive investigation of the impact of missing data on the estimations of cost-effectiveness. 67 MI was performed using chained equations and predictive mean matching (PMM). A total of 46 imputations were used based on the maximum percentage of missingness across imputed variables (see Appendix 1, Table 42).

Impact of COVID-19 on the health economic analysis

The trial follow-up was completed in December 2020. The first UK national lockdown due to COVID-19 started on 23 March 2020 with restrictions eased in June. The majority of trial therapy was completed before the national lockdown, however, for some participants the follow-up period from 16 weeks to 52 weeks would have included full lockdown, and also any subsequent delays to treatment due to the disruption caused by lockdown. Given the randomisation of participants, the impact on face-to-face treatment was assumed to be balanced across arms and would not affect the overall results.

Cost-effectiveness analysis and cost–utility analysis

The CEA expressed the incremental cost of achieving a percentage reduction in symptoms of PTSD at 16 and 52 weeks post-randomisation measured by CAPS-5, with additional analyses expressing the cost of achieving a percentage reduction in distress caused by the traumatic event measured by the IES-R, and achieving a percentage improvement in functioning measured by WSAS score. Total costs at 16 and 52 weeks (including baseline) for the ITT population were considered in the incremental analysis.

A CUA was undertaken to assess the incremental costs per QALY gained as a result of the use of GSH compared to CBT-TF. QALYs for each patient were calculated based on the utility scores at baseline, 16 weeks and 52 weeks using the area-under-the-curve approach and linear interpolation. Total costs at 16 and 52 weeks (including baseline) for the EQ-5D ITT population were used to calculate the incremental cost.

Adjusted mean costs and outcomes, including QALYs, and the differences in adjusted mean costs and outcomes (and associated 95% CIs) at 52 weeks were estimated using seemingly unrelated regressions which accounts for the correlation between costs and outcomes. Costs and outcomes were adjusted for study site, age and time to event. Costs were also adjusted for baseline costs and outcomes for baseline utility.

The incremental cost-effectiveness ratio (ICER) resulting from the CUA was compared to the willingness to pay threshold of £20,000 and £30,000 per QALY gained as standardised by NICE. 68

No conditions for NI were applied in the analysis. Results were reported as net monetary benefit (NMB) presenting the incremental value of the intervention in monetary terms by applying the willingness to pay threshold to the change in QALYs:69

Sensitivity analyses

Deterministic SAs were undertaken to test the robustness of the results of the CUA considering the uncertainty in input parameters such as costs of therapy, and in different scenarios more representative of real-world application of GSH. The health and social care resource use delivered in private and non-NHS institutions was included in an additional analysis to consider wider costs. It is anticipated that GSH will be provided by band 4/5 healthcare professionals rather than band 7 therapists in routine clinical practice, thus this was tested in the SA. Analysis was carried out for the 16-week treatment end point. Subgroup analysis was carried out to determine if cost–utility was improved for specific groups based on demographic and clinical features. The following subgroups were tested: age groups (35 years and under, and over 35 years); level of education of the participant (people with a degree or higher qualification, and all other levels of education); months since the traumatic event (> 18 ≤ 18 months); comorbidity was taken as level 4/5 in the EQ-5D pain or anxiety score at baseline.

Probabilistic sensitivity analysis (PSA) used non-parametric bootstrapping to address joint parameter uncertainty and assess the impact on the ICER during 1000 simulations which were undertaken using random sampling of the distributions of costs and outcomes with results presented as cost-effectiveness acceptability curves (CEAC). A CEAC describes the probability of the intervention being cost-effective at different willingness-to-pay thresholds based on the PSA.

Budget impact analysis

A trial-based budget impact analysis (BIA) was undertaken for the UK population to estimate the likely impact of the use of GSH on NHS budgets through implementation costs and changes in healthcare usage. The BIA was informed by trial data supplemented by the best available published evidence where required and conducted according to recommended good practice. 70

A simple budget impact model was constructed in Microsoft Excel 365. The number of eligible patients was calculated using prevalence and incidence figures. Prevalence and incidence data for PTSD in the UK were obtained from published sources. 50 Age-specific general population mortality was not applied as a 1-year time horizon was used. Approximately 31.4% of adults in England reported experiencing at least one traumatic event according to the Adult Psychiatric Morbidity Survey in 2014. Of the participants of the survey, 4.4% screened positive for PTSD in the previous month, and half were receiving mental health treatment, although the majority were taking medication (43.6%), and only 24.0% were having psychological therapy, which is currently recommended by NICE. 50

Training and supervision costs

Training was delivered face-to-face over 2.5 days, 1 day was for GSH training, and 1.5 days for CBT-TF training, by two band 8a–c staff running the sessions for 20 participants (see Table 15). Each trainer required 2 hours of preparation time (e.g. preparation of materials) for the training sessions. There was an additional 10 hours for developing the course materials for GSH training (£823.33). No cost was applied for room rental as training can be given within NHS facilities.

In addition, for each therapy, one case was supervised as part of the training programme. For 4 out of 30 (13%) therapists, competency was not achieved in GSH from the first supervised case and they required an additional supervised case. For CBT-TF, one session was recorded and sent to the trainers for critique. After completion of training, monthly supervision was given via Skype in groups of three or four therapists.

Supervision for band 4/5 staff will be fortnightly for 4 months after training, with hourly group sessions (with six participants), followed by monthly supervision.

Website costs

It was estimated by the trial team that it will cost £10,000 per year to maintain the website and app, and provide administrative support for 1000 users.

Cost of therapy

The mean number of face-to-face appointments in the GSH arm was 3.93 (95% CIs 3.60, 4.25; median 5, IQR 3–5) and 8.63 (95% CI 7.94, 9.32; median 9, IQR 6–12) in the CBT-TF. The amount of face-to-face therapy time was 208.4 minutes (SD 69.3) for GSH and 767.0 minutes (SD 278.2) for CBT-TF.

In the GSH arm there were check-in phone calls between the face-to-face sessions; the median number of calls was 3 (IQR 2–4). This was based on data from 193 people, two people in the GSH arm and one person in the CBT-TF arm had no data reported for the therapy sessions.

The cost of therapy was calculated as time in face-to-face sessions, phone calls and non-contact time for note-taking. The therapist cost per working hour was applied as time on direct and indirect tasks was collected (see Appendix 1, Table 32). It was not possible to determine if any CBT-TF appointments were site visits which may incur extra costs due to travel time.

The cost of therapy was lower for the GSH arm as there were fewer face-to-face appointments, mean cost £277.31 (95% CI £253.27, £301.34; median £286.15, IQR £210.88–351.53) compared to £729.49 (95% CI £670.76, £788.22; median £753.71, IQR £554.37–950.60) in the CBT-TF arm.

It is anticipated that the GSH therapy will be given by band 4/5 low-intensity therapists if it is used more widely in the NHS. Applying a lower unit cost for staff in the GSH arm led to the mean cost of therapy being £160.09 (95% CI £146.22, £173.97; median £165.20, IQR £121.74–202.94).

Cost of medication at baseline

In the 3 months before baseline, the mean number of prescriptions per person was 1.40 (95% CI 1.08, 1.72) in GSH arm, and 1.35 (95% CI 1.06, 1.65) in the CBT-TF arm (see Table 16).

Cost of healthcare resource use

Other primary and community care costs

Primary and community care costs in the 3 months prior to baseline were similar in both arms, the mean cost in the GSH arm was £226.95 (95% CIs £162.78, £291.16) and £225.47 (95% CI £145.01, £305.93) in the CBT-TF arm (see Table 19). This included GP appointments, calls and out-of-hour appointments; practice, district and community psychiatric nurse appointments; social worker and home help visits, NHS111 calls, occupational therapist and other community care appointments.