A pragmatic randomised controlled trial and economic evaluation of family therapy vs. treatment as usual for young people seen after second or subsequent episodes of self-harm: the Self-Harm Intervention - Family Therapy (SHIFT) trial

Inclusion criteria

1. Aged 11–17 years.

2. Self-harmed prior to assessment by the CAMHS team.

3. Engaged in at least one previous episode of self-harm prior to the index presentation.

4. Assessed in hospital following current episode, or referred directly to CAMHS from primary care with recent self-harm as a key feature of presentation.

5. When the presenting episode was because of alcohol or recreational drugs, the young person had explicitly stated that he or she was intending self-harm by use of these substances.

6. The clinical intention was to offer CAMHS follow-up for self-harm.

7. Lived with primary caregiver.

Exclusion criteria

1. At serious risk of suicide (clinical judgement) and so requiring more than TAU.

2. An ongoing child protection investigation within the family, which would have made treatment difficult to deliver.

3. Would ordinarily have been treated not in generic CAMHS but rather by a specific service, and so would not normally have received TAU.

4. Pregnant at time of trial entry and thus likely to have to interrupt treatment in either arm.

5. Actively being treated in CAMHS (as the possibility of randomisation might disrupt ongoing therapy).

6. In a children’s home or short-term foster placement and so no continuity of treatment possible in either arm.

7. Moderate to severe learning disability or lacked capacity to comply with trial requirements.

8. Involved in another research project – at the time of trial entry or within the last 6 months so as not to overburden participants and to avoid possibility of conflicting requirements across two studies.

9. Sibling had been randomised to the SHIFT trial or was receiving FT within CAMHS. As the intervention studied is a family intervention, if a sibling was already in the trial, randomisation would not have been possible.

10. The young person and one main caregiver had insufficient proficiency in English to contribute to the data collection.

Family therapy

The FT intervention was based on a modified version of the Leeds Family Therapy & Research Centre Systemic FT Manual, the development and validation of which was funded by the Medical Research Council to support trials of FT. 62 This manual was updated by the FT expert members from the Trial Management Group (TMG) to ensure that it was appropriate for work with families following self-harm. The SHIFT FT Manual can be requested from the trial team via http://medhealth.leeds.ac.uk/SHIFTManual.

Our systemic orientation to self-harm assumes that there is an interplay between a number of factors that can lead to the development of individual difficulties and problems such as self-harm. However, rather than focus on the causal explanation of individual problems, systemic FT is concerned with the way in which these problems will have become embedded in the matrix of family and wider social relationships, the felt experiences and the meanings and narratives that have become attached to and that shape these difficulties. 63 Looking for a causal account in the individual case is seen as unhelpful, partly because it inevitably oversimplifies the complexity of the individual aetiological pathway and partly because it tends to reinforce disabling, blaming narratives. 64

Self-harm in an individual case will, in part, have been a response and have meaning in relation to particular issues (e.g. problems in emotional regulation, sense of hopelessness, low self-esteem, developmental and life-cycle issues, school and peer issues, attachment difficulties, relationship problems, parental situations or problems in family communication), but it also highlights that, for some reason, alternative, more constructive solutions were unavailable. The self-harming response will have been influenced by and in turn will influence subsequent patterns of interaction and will give new meaning to them. 65 It becomes a means of communicating and constrains alternative responses. By considering these behavioural and linguistic patterns in greater detail, facilitating understanding and noticing and supporting alternatives that encourage more positive narratives to emerge, the therapist can help the client and family create a different response pattern to the issues that showed themselves in self-harm.

Family therapists [those eligible for registration with the UK Council for Psychotherapy (UKCP)] were appointed specifically to work on the trial, following the formal advertisement of a clear job description and subsequent interview. All were qualified systemic family therapists who had trained in the UK to the level of a master’s degree, which included 2 years of supervised practice. The therapists had a primary professional mental health background and qualification (generally social work or nursing), and most of them had experience in CAMHS and post-qualification experience in FT practice.

The therapists joining SHIFT at the outset attended a 2-day group training event prior to the start of the trial and then 1-day training events annually thereafter. The authors of the manual provided the training, which included review and discussion of the manual, reflections on their own experiences with/as young people, their relationship to clinical risk and team building. Additional material on adolescent self-harm was presented by a specialist on the trial. Role play and experiential exercises were included. Each family therapist worked with a pilot case (CAMHS non-trial case) with supervision. After each therapist had completed a number of sessions, they were deemed ready to commence with trial participants. The orientation of replacement family therapists for departing original trial therapists involved one-to-one training with one of the trial’s senior family therapists, a period of observation of the remaining team members’ therapy and a one-to-one session with the supervisor to review the material.

Family therapists worked in teams of three or four and provided trial FT as a team for a cluster of CAMHS. Within each team, each family therapist would be the lead for a subset of participants (the other family therapists in the team would act as observers and make only a small face-to-face contribution for those participants).

Monthly group supervision of 2 hours was provided for each team by senior family therapists (two of whom were authors of the treatment manual). Supervision included discussion of cases, adherence to the manual, broader trial issues and team dynamics. Generally, the focus of the supervision was determined by the family therapists at each meeting.

Young people and their families were offered FT sessions of approximately 1.25 hours’ duration each, delivered over 6 months at approximately monthly intervals, but with more frequent initial appointments. This equated to approximately eight sessions, but there was the expectation that some participants would receive fewer sessions because of dropout or mutually agreed termination of treatment. Equally, it was anticipated that some participants might receive more sessions if this was deemed clinically appropriate.

Wherever possible, and when consent was provided, the sessions were video-recorded, as this is part of good FT practice and facilitates supervision.

Originally, it had been intended that supervision would include a review of recordings of current cases. This proved difficult for technical reasons and so most supervision was based on therapist presentation. The adherence measure was developed at a later stage in the trial and applied retrospectively.

A selection of video-recorded sessions was centrally reviewed, whereby the first and a subsequent therapy session were reviewed for each therapist to monitor their adherence to the manual and to allow the reporting of this. Adherence was assessed through use of a tool developed specifically for the trial. 66 In addition, administrative processes that were required in the manual were also monitored, for example number of sessions offered and attended and whether or not formulation letters (summarising the therapist’s understanding of the family situation) were sent to families following attendance at their second therapy session, as expected in the manual.

Treatment as usual

Treatment as usual was the care offered by local CAMHS teams to young people referred following self-harm. It was expected that this treatment would be diverse and involve individual and/or family-orientated work, delivered by a range of practitioners with various theoretical orientations. As SHIFT was a pragmatic trial involving a number of collaborating CAMHS teams, it was not deemed possible or appropriate to specify TAU, although it was expected that CAMHS practitioners would be working in line with best practice as set out in several National Institute for Health and Care Excellence (NICE) guidelines (e.g. guidance on self-harm and depression in childhood). 67,68 Generic TAU was not delivered to the FT group as part of their clinical intervention, unless specific additional assessment or treatment was indicated during or after FT, for example referral for formal mental state assessment or for medication.

Contamination

The possibility of cross-arm contamination was considered during the design stage of the trial, with the following points mandated whenever possible and monitoring processes implemented:

• Different teams of therapists delivered the two interventions in each CAMHS site and SHIFT family therapists were prohibited from treating participants in the TAU arm for the duration of the trial. There were processes in place to record contamination if this occurred.

• Owing to the nature of appointment scheduling and the fact that this was family-specific therapy (i.e. not a group intervention), there was little opportunity for participants to meet and discuss treatment, so contamination at the participant level was very unlikely.

• Any family-orientated clinical interventions in the TAU group were likely to be different from the trial FT intervention, which required adherence to the Leeds Family Therapy & Research Centre manual, fully-trained family therapists eligible for UKCP registration, therapy delivered in a team context and regular supervision. Thus, use of family interventions in the TAU arm was not prohibited, but details of all treatment received by participants in both arms were recorded.

Screening

Data were recorded anonymously for young people who did not meet the eligibility criteria, and for those who were eligible but not willing to provide consent, in order to monitor trial uptake and representativeness of the trial population. Screening data for randomised participants were linked to the main trial data, and also provided details of the outline ‘index’ event that had prompted their referral to CAMHS and eligibility for the trial. Screening data, eligibility and consent for researcher contact were assessed and recorded by the screening clinician. During the baseline visit, the researcher confirmed the participant’s eligibility, obtained and recorded consent for trial participation, and obtained further details, including current physical and mental health comorbidities, psychotropic medications and history of abuse.

Participant assessments and follow-up

At the baseline assessment the researcher administered questionnaires to the young person, and self-reported questionnaires were also completed by the young person and caregiver.

Postal questionnaires were administered by CTRU at 3 and 6 months post randomisation. These were preceded by a telephone call from researchers, when it was possible to make contact. If questionnaires were not returned, postal reminders were sent 2 weeks after the initial mailing and then again a further 2 weeks later.

Further follow-up to administer questionnaires took place at 12 and 18 months following randomisation via face-to-face researcher interviews. When it was not possible to arrange face-to-face interviews and when participants agreed, self-reported questionnaires were posted to participants for completion. The offer of incentives to young people for participating in the 12- and 18-month follow-up interviews was introduced in August 2013 (see amendments in Appendix 2) in an effort to improve compliance.

Clinical follow-up

The primary outcome measure was obtained from accident and emergency (A&E) and inpatient Hospital Episode Statistics (HES) data downloads from NHS Digital. This method of primary outcome data collection was augmented by directed hospital record searches, which were undertaken by researchers at frequent intervals throughout the trial. Researchers searched acute trust records for further details of all attendances that were unclear from the central HES data (i.e. where self-harm relatedness or method of self-harm could not be determined from HES data alone) and for any hospital attendances for those participants who had not consented to data sharing with NHS Digital or could not be linked to HES. Given that HES data sets are England-wide, this maximised the collection of hospital attendance data, allowing for participants moving out of the catchment area (within England). The collection of these routine data also minimised bias by eliminating the possibility of preferential researcher data collection at certain acute trusts. Further details of the comparison of HES records and researcher-collected hospital records and the success of using HES data for the trial will be published separately.

Data relating to treatment received in CAMHS following trial entry were provided by the treating CAMHS clinicians or family therapists, by clinical studies officers (CSOs) employed by the Research Networks in each locality and by the researchers when blinding was no longer required (i.e. after participant-reported follow-up data had been collected). Data included initial session attendance (FT and TAU), therapeutic approach (for TAU), referrals within CAMHS for additional or alternative sessions, requirements for psychotropic medications, liaison and referrals to other agencies outside CAMHS and re-referral to CAMHS following discharge. In addition, treating clinicians and participants completed the SOFTA44 at the end of the third treatment session. Supervision sessions for family therapists and routinely provided clinical supervision for CAMHS clinicians were also recorded. Details of the scoring of questionnaires are in Appendix 3. These appendices provide further detail concerning the collection of clinical follow-up data.

Safety

Non-serious adverse events (AEs) were operationally defined in SHIFT as treatment on an emergency outpatient basis and re-referral to CAMHS, as these events were expected to occur within the adolescent study population. Deaths and hospital admissions were defined as serious adverse events (SAEs).

Adverse event data were obtained through researcher collection of data from CAMHS and acute trusts and via data transfer from NHS Digital.

Assessment instruments

Assessment instruments were incorporated into participant assessment packs for both the young person and their caregiver. The standardised questionnaires used were selected following our literature review because they related to factors thought to be important secondary outcomes in their own right or possible mediators or moderators of treatment effects. In selecting individual measures, we also tried to use those measures that would enable comparisons with other studies in the field. In one case [the Inventory of Callous–Unemotional Traits (ICU)] we were influenced by our knowledge of a large randomised controlled trial about to commence at the same time as SHIFT with a similar age group that we wanted to be able to compare with our sample. All measures were appropriate to the age range of young people and have been widely used in a range of ethnic communities. Appendix 3 contains details of scoring.

Data quality and monitoring

Data were monitored for quality and completeness by the CTRU using established verification, validation and checking processes. Missing data, except individual data items collected via the postal questionnaires, were chased for until they were received or confirmed as not available or until the trial was at analysis.

Protocol adherence, trial uptake, loss to follow-up, withdrawal, participant safety, data quality and data/questionnaire return rates were monitored, as appropriate to each group, at each TMG, Trial Steering Committee (TSC) [which included a patient and public involvement (PPI) member] and Data Monitoring and Ethics Committee (DMEC) meeting.

The DMEC reviewed, on an annual basis and by trial arm, the number and frequency of hospitalisations and deaths as a consequence of self-harm. Processes were in place to escalate any concerns raised by the DMEC or to escalate concerns identified by the trial team to the DMEC in the interim periods between meetings. The DMEC reviewed interim primary outcome data after half the events had occurred.

Sample size

The power calculation was based on a minimally important reduction in 18-month repetition rates of self-harm (leading to hospital attendance) from 29% in participants receiving TAU6 to 18.8% in participants receiving FT, that is, a reduction of 35%, providing a constant hazard ratio (HR) of 1.64. Using a 5% significance level log-rank test for equality of survival curves, 374 participants per arm were required, with 172 total events, to give 90%. Assuming, at most, 10% loss to follow-up by 18 months, the total sample size required was 416 per arm, 832 in total.

Although SHIFT was individually randomised, the inherent clustering of participants nested within therapists is known to have an impact on power and is related to the level of the intracluster correlation coefficient (ICC) and the cluster size. It was expected that the level of clustering would be low, possibly around 0.01 but no higher than 0.05 (owing to the use of therapy manuals, therapist selection, training, supervision and monitoring). It was also expected that the number of participants per therapist would be small. It was estimated that between 8 and 15 therapists would be available in the TAU team at any one centre, so across the anticipated 15 participating trusts there would be 120–225 therapists available to treat 416 participants; thus, each therapist would treat between two and four participants (maximum cluster size of four). In FT, we estimated that there would be approximately 35 therapists available across all sites to treat 416 participants; thus, each lead therapist would have direct contact with approximately 12 participants (maximum cluster size).

The design effect, describing the extent to which the sample size must be increased to obtain the same power as an individually randomised study without clustering, was therefore assumed likely to be no greater than 1.55 (ICC of 0.05), effectively reducing the sample size from 416 per group to 270 per group and the power from 90% to around 75%. If the ICC were as low as 0.01, then the design effect would be 1.11, reducing the sample size to 374 per group and the power to around 85%. We anticipated that the ICC would be towards the lower end of the possible range and therefore the trial would still be adequately powered with the sample size planned.

Participant populations

The intention-to-treat (ITT) population consisted of all randomised young people who had consented, regardless of non-compliance with the intervention, whether they were found to be ineligible post randomisation and/or remained in the trial and were analysed according to the treatment they were randomised to receive.

The per-protocol population consisted of all randomised, consented young people, excluding those who were found not to satisfy the eligibility criteria.

Statistical analysis

A single formal interim analysis was planned for the primary end point, repetition of self-harm leading to hospital attendance within 18 months of randomisation, when at least half the required number of events were reached (86 events). It was planned that the DMEC, in light of interim data, would if necessary report to the TSC with a recommendation of trial adaptation or early closure if, compared with TAU, the effect of FT was significantly inferior (p < 0.005). Final analyses were planned after all participants had reached the end of the 18-month follow-up period, when it was expected that at least 172 events would have occurred.

Predictive and process measures

Moderator analysis

Moderator analysis explored whether or not the treatment effect in the primary analysis Cox proportional hazards model depended upon participants’ baseline characteristics via inclusion of the proposed moderator alongside the interaction effect of treatment × moderator in the primary analysis model (including covariates). Covariates and responses to all baseline questionnaires were explored for moderation. Questionnaire responses were also categorised for the young person Beck Scale to indicate whether or not suicidal ideation was present; the young person CDRS-R for whether or not depression was present; and the young person and caregiver McMaster FAD subscales to indicate whether family functioning was healthy or unhealthy. A 5% significance level was used to identify moderation through the interaction of the potential moderator with treatment, irrespective of the main effect of the potential moderator, although it is acknowledged that interaction testing is not a powerful test for moderation. Analysis was of the ITT population to availability of data (complete case) for each proposed moderator.

Mediator analysis

Complier average causal effect analysis

A complier average causal effect (CACE) analysis was conducted to model the causal effect of FT receipt (as opposed to randomisation) on the primary outcome (Figure 1). An instrumental variable model was used to adjust for potential selection bias occurring because of participants who did not receive their allocated treatment, while addressing potential confounds. Randomisation allocation was used as the instrumental variable, creating additional variance unaccounted for by FT receipt alone to obtain an unbiased estimate of the effect of FT receipt, in a two-stage instrumental variable analysis using the qualitative and limited dependent model (QLIM) procedure in SAS,89 to simultaneously estimate the first equation of treatment selection (stage 1) and the second equation of the unbiased effect of treatment on outcome (stage 2). Covariates were included in each stage of the model. The primary outcome was considered as a binary variable and a probit transformation was used. The primary ITT analysis was repeated on this transformed scale for comparison with the CACE estimates, alongside a further ‘as treated’ analysis (using the QLIM procedure). TAU participants with missing treatment data were assumed to have not received FT in the analysis. Kaplan–Meier curves were also constructed for each group by receipt of FT and receipt of FT overall.

FIGURE 1.

Schematic representation of CACE analysis.

Mediation

Process variables identified as potential mediators included the overall number of sessions attended, the use of psychotropic medications and therapist characteristics.

Responses to the 3- and 6-month Family Questionnaire (subscores) and 12-month questionnaires (total scores and subscales found to have a significant treatment effect at 12 months in the secondary end-point analysis or moderator analysis) were also explored for mediation. For questionnaire responses, mediation was explored in relation to the time to event, self-harm resulting in hospital attendance, after measurement of the potential mediator. Therefore, events after 3, 6 and 12 months post randomisation were included for 3-, 6- and 12-month questionnaires, respectively. Responses at 18 months were not considered in the analysis, as outcomes were collected up to 18 months only and responses could not mediate an outcome observed before this date.

The Baron and Kenny90 steps were employed to explore mediation (Figure 2):

• Step 1 – establish an effect of randomisation on outcome that may be mediated, that is, Cox regression with Y as the dependent variable and X as predictor, estimate and test path c the total effect:(1)Y=h0(t)exp⁡(β10X).

• Step 2 – establish that there is an effect of randomisation on the mediator, that is, linear or logistic regression with M as the dependent variable and X as predictor, estimate and test path a:(2)Me=β20+β21X.

• Step 3 – establish that there is an effect of mediator on outcome while controlling for randomisation, that is, Cox regression with Y as the dependent variable and X and M as predictors, estimate and test paths b and c’:(3)Y=h0(t)exp⁡(β31X+β32Me).

FIGURE 2.

Baron and Kenny mediation model.

Following these steps, complete mediation is the case in which variable X (randomisation) no longer affects Y (time to self-harm) after M (mediator) has been controlled and so path c’ is zero. Partial mediation is the case in which the path from X (randomisation) to Y (time to self-harm) is reduced in absolute size but is still different from zero when the mediator is introduced.

Steps 1 and 3 were fitted using a Cox proportional hazards model adjusted for all covariates. For continuous mediators, Step 2 was fitted using a linear regression model containing randomised treatment and all covariates; and for binary outcomes (Beck suicide ideation, psychotropic medications, therapist characteristics) logistic regression was used containing randomised treatment and all covariates apart from NHS trust (owing to lack of convergence). For mediators based on questionnaire responses, the baseline score was also included as a covariate in each step.

Adherence and alliance to family therapy

Summaries of therapist baseline characteristics, training, therapy session recordings, initial and specialist adherence review, formulation letters to families and therapist supervision were produced. In addition, mean scores and 95% CIs are presented for the total score and for each subscale of the SOFTA.

Outcomes

Young people’s health-related quality of life was assessed using the EQ-5D,91,92 which was included at baseline and along with the young person’s resource use questionnaire at 6, 12 and 18 months. Changes in EQ-5D scores across time points were evaluated using two-sample t-tests to explore any important differences in these end points within the time frame of the trial. These tests are useful in understanding the impact of the length to follow-up on the health outcome measures, as insufficiently long follow-up periods may introduce biases in the subsequent cost-effectiveness analysis. In line with the NICE reference case93 the primary outcome for the economic evaluation was quality-adjusted life-years (QALYs). Young people’s responses to the EQ-5D were converted into health state utility values and multiplied by the proportion of 1 year the time period represented (baseline to 6 months = 0.5; 6 to 12 months = 0.5; 12 to 18 months = 0.5) to calculate QALYs. 94 Average QALYs between adjacent time points were calculated to generate smoothed estimates between time points. Therefore, QALYs were calculated using the area under the curve approach as shown below:

Quality-adjusted life-years represent a quality-of-life-weighted survival value in which 1 QALY is equivalent to 1 year of full health.

The number of self-harm events avoided at 18 months was considered as a secondary outcome. Additionally, the sensitivity analysis used the aggregated QALY gain for both the young person and his/her main carer as the health outcome. The caregiver’s health-related quality of life was assessed using the HUI,95,96 which was included along with the carer resource use questionnaire. The carer’s responses to the HUI were converted into health state utility values76 and the carer’s QALYs were calculated in a similar way to the young person’s QALYs.

Resource use

Resource use data were collected at baseline and at 3, 6, 12 and 18 months from the young person and from his/her main carer. Resource usage was converted into costs using unit cost figures from the British National Formulary (BNF), Personal Social Services Research Unit (PSSRU) and the Department of Health’s National Schedule of Reference Costs. 97–99 The base-case economic evaluation focused on health and social care costs. The currency used was the pound sterling (£) and 2014 was used as the reference financial year.

Costs and benefits were discounted at 3.5% per annum.

Missing data

Missing data were imputed using multiple imputations via chained equations86,100,101 as recommended for economic analyses alongside clinical trials. 102 For consistency with the statistical analysis and to ensure best fit, imputations were based on predictor variables including treatment allocation, gender, age, centre, total number of self-harm episodes, type of index episode, source of referral from hospital and derived scores for a number of assessment instruments (BSS, CDRS-R, Hopelessness Scale for Children and PQ-LES-Q).

Within-trial cost-effectiveness

The within-trial analysis aimed to determine the intervention that maximised health outcomes at 18 months. It used an ITT perspective.

Base-case analysis 1 was a cost–utility analysis over 18 months examining the cost per QALY gained for all participants. Descriptive statistics of costs and EQ-5D scores and parametric tests to evaluate any important differences between the two trial arms at each time point in the trial were produced. Thereafter, incremental cost-effectiveness ratios (ICERs) were calculated by dividing the average difference in costs between the two arms by the average difference in QALYs between the two trial arms:

The ICER represents the additional cost per unit of outcome gained. This indicates the trade-off between total cost and effectiveness when choosing between FT and TAU. When compared against the cost-effectiveness threshold, this gives an indication of whether or not spending additional money on FT appears efficient. As a rule, the NICE implicit cost per QALY threshold of £20,000–30,000 per QALY was used. An intervention is cost-effective so long as its ICER is within or below the £20,000–30,000 per QALY range.

Base-case analysis 2 calculated the incremental cost per self-harm event avoided because of FT at 18 months compared with TAU and was carried out in a similar way; however, there is not a clear decision rule on the cost-effectiveness threshold per self-harm event avoided and so the results were presented for a range of cost-effectiveness thresholds.

Cost-effectiveness over a longer time horizon

The NICE reference case recommends exploring cost-effectiveness over a time horizon that is long enough to reflect any important differences in costs or consequences between treatments being compared. The NHS might be interested to understand the cost and the consequences of FT and TAU beyond the trial follow-up, and decision-analysis modelling is key in this context. As far as we know, there are no prior studies in which a decision-analysis model was built to study the long-term cost-effectiveness of FT for adolescents. Considering that a cohort of young people will face such a large number of events over their adult life, we considered that building a decision-analysis model that reflects a lifetime horizon would require a larger amount of assumptions and probabilities to estimate. Therefore the decision-analysis model evaluated the cost-effectiveness of FT compared with TAU up to 5 years after randomisation. It considered health and social care costs and used young people’s QALY gain as the outcome. The results from the within-trial analysis were mainly used to make reliable assumptions for the model as appropriate literature references were not identified from which health states or probabilities beyond an 18-month follow-up could be extracted.

A Markov model with three health states was used (Figure 3). The model included three possible health states: self-harm, defined as self-harming at least once in a period of 6 months; no self-harming; and death. Markov models describe participant progression over time through a pathway of health states, with movement between the health states being triggered by events; in this case, self-harm events or death. Resource use and costs were associated with each health state and participants accumulated costs and health benefits in each state over 6-month cycles. Participant cost and utility data were available at 6, 12 and 18 months from the trial data and were directly included into the model to estimate longer-term costs and health benefits.

FIGURE 3.

Three-state Markov model. SH, self-harm; noSH, no self-harm. Reproduced from Cottrell et al. 70 © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

The model inputs were derived from the trial data. Specifically, the proportion of the participants beginning in each health state in the model was derived directly from the proportion of participants in the trial who remained in the self-harm state or moved to no self-harming. In the first cycle of the model, all participants started from self-harm in both arms and this was informed by the inclusion criteria for the trial. No participants died over the 18-month trial follow-up, but to account for possible death it was assumed that the probability of a participant moving from self-harm to death or from no self-harm to death was minimal and equal to 0.0001, regardless of the trial arm. Derivation of the post-18-month transition rates between the different states required extrapolation beyond the follow-up period of the trial.

Regarding costs and utility related to each model state, death was assumed to be associated with zero utility and zero cost. For self-harm and no self-harming states, the associated costs and utility values at each follow-up time point (6 months, 12 months and 18 months) were derived directly from the trial data.

Any intervention costs were assumed to occur equally over the first 12 months for each arm based on the trial data. Given that it was not possible to distinguish if these costs were incurred only by those in the self-harm state, in the no self-harming state or both, it was assumed to be the same for any of the states.

Presentation of results

The outputs of both the within-trial analysis and the decision-analysis model were presented as expected ICERs of FT compared with TAU, scatterplots on the cost-effectiveness plane and cost-effectiveness acceptability curves using non-parametric bootstrapping to determine the level of sampling uncertainty.

In order to test the robustness of the results of the within-trial analysis, a number of sensitivity analyses were conducted to explore the impact of costs and missing data on the results of the study as described below:

1. a non-parametric bootstrapping with 10,000 replications

2. an analysis in which it was assumed that only one therapist was involved in each treatment session in the FT arm (intervention costs from scenario 1)

3. an analysis in which it was assumed that the average number of therapists were involved in each treatment session in the FT arm (intervention costs from scenario 2)

4. an adjusted analysis to account for EQ-5D differences between arms at baseline, using a set of baseline characteristics such as treatment allocation, gender and age group (baseline EQ-5D was also included as a control to account for any impossible imbalance and to improve the precision of the estimates)

5. a complete-case analysis including only those participants with no missing quality-of-life and cost data at any time point before imputation

6. an analysis using an aggregate QALY, taking into consideration both the young people’s and caregivers’ QALY gains.

In order to test the robustness of the results of the decision model, deterministic one-way sensitivity analyses were conducted to assess the impact of individual parameter uncertainty and key model assumptions on the results. Probabilistic sensitivity analyses were conducted to assess the impact of parameter uncertainty on the results. Probabilistic analysis accounts for joint parameters uncertainty in non-linear models by assigning probability distributions to each of the input parameters and randomly drawing from these probabilities over a number of Monte Carlo model simulations to produce different cost and QALY estimates in each simulation of the model. In our model, we chose to do 30,000 simulations owing to the results being highly sensitive to a smaller number of simulations.

Participant flow

The numbers of young people screened for entry into SHIFT, eligible, randomised, followed up via post at 3 and 6 months, visited by a researcher at 12 and 18 months, withdrawn and analysed are presented in the CONSORT diagram in Figure 4.

FIGURE 4.

Consolidated Standards of Reporting Trials flow diagram. Adapted from Cottrell et al. 70 © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

Screening and recruitment

A total of 3554 young people were screened within CAMHS, of whom the clinician deemed 1603 (45.1%) to be eligible for the trial (see Figure 4). Of the remaining young people, 1831 (51.5%) were deemed to be ineligible and no eligibility information was provided for a further 120 (3.4%). The most common reasons for ineligibility were not having engaged in self-harm prior to the current CAMHS referral (44.8% of those ineligible), already being treated within CAMHS (25.2%), not intending to offer CAMHS follow-up for self-harm (21.7%), not living with their primary caregiver (17.6%), recent self-harm was not a key feature of presentation and referral into CAMHS (16.4%), presenting episode resulted from consumption of alcohol or drugs where specific intent to self-harm was not established (13.7%), in a children’s home or short-term foster placement (10%), would not ordinarily be treated in CAMHS (9.2%), currently undergoing a child protection investigation (7.8%) and at serious risk of suicide (7%).

A total of 993 (61.9%) eligible young people consented to researcher contact, of whom 481 (48.4%) provided verbal consent and 512 (51.6%) provided written consent; 273 (17%) young people did not consent, and details of researcher consent were missing for 337 (21%) eligible young people. Of the 273 young people not providing consent, 160 (58.6%) refused consent, 85 (31.1%) did not attend their first CAMHS follow-up appointment and 18 (6.6%) were not approached by the CAMHS clinician.

Researcher contact and baseline visits were sought for the 993 young people consenting to researcher contact. A total of 832 (83.8%) consented to trial participation and were randomised into the trial, 51.9% of those eligible. A further 127 (12.8%) young people did not provide consent for trial participation (100 refused; 11 could not be contacted in sufficient time).

Screening and recruitment took place over 4 years: the first participant was randomised on 14 April 2010 and the last on 30 December 2013 once the target of 832 participants had been reached, with 415 participants randomised to receive FT and 417 participants randomised to receive TAU. Figure 5 shows overall, monthly and cumulative recruitment of participants into the trial, and Figure 6 presents overall recruitment by CAMHS and hub and treatment arm while Table 102 in Appendix 4 provides a summary of overall recruitment activity by CAMHS centre, trust, hub and treatment arm and by CAMHS.

FIGURE 5.

Recruitment graph.

FIGURE 6.

Recruitment by CAMHS and hub.

Characteristics of the screened, eligible and randomised participants – generalisability

Tables 2 and 3 compare the characteristics of the screened, eligible and randomised young people and show similar characteristics according to gender, ethnicity and treatment received. Between 82% and 89% were female within each population; and between 79% and 84% were of white ethnicity. The mean age was 14.8 [standard deviation (SD) 1.55], 14.6 (SD 1.38) and 14.3 (1.37) in each respective population with an increasing proportion of younger adolescents (aged 11–14 years) in later stages of the screening process.

There were more marked differences according to referral source; almost 60% of those screened were referred via hospital while 25% had been referred via their general practitioner (GP). However, of those randomised, only 37% had been referred via hospital while 46% had been referred via their GP; this is discussed in relation to the primary outcome in Primary analysis: Cox proportional hazards regression. Similarly, the proportion of participants whose most recent index self-harm episode was poisoning decreased from 41% of screened to 23% of randomised while the proportion cutting increased from 40% to 63%.

Young people referred to CAMHS through hospital therefore had a far lower eligibility rate than those from the community (31.8% vs. 70.2%) and a lower rate of consent for researcher contact (54.6% vs. 67.8%), suggesting that hospital referrals were a more highly selected group.

Compared with community referrals, young people referred through hospital were more likely to be ineligible because their index episode was their first reported episode of self-harm (48.8% vs. 29.9%) or because they were already being actively treated in CAMH (28.3% vs. 14.5%). The proportion of young people who did not consent to researcher contact because they did not attend their CAMHS follow-up appointment was higher among those referred from hospital than among those referred through the community (37.9% vs. 23.8%), while the proportion of non-consent caused by refusal was far lower (47.9% in hospital referrals vs. 70% in community withdrawals).

The characteristics of young people referred from hospital differed from those of community referrals in that they were slightly older [mean age of 14.9 (SD 1.49) years vs. 14.5 (SD 1.57) years], their index episode of self-harm was more likely to be caused by self-poisoning (63.5% vs. 6.2%), and they were more likely to have received at least some treatment following the episode (31.3% required no treatment, compared with 66.7% of community referrals).

Baseline characteristics

Protocol violators

A total of 16 (1.9%) young people were identified as not fulfilling the eligibility criteria (Table 14), five in the FT arm and 11 in the TAU arm. The main criterion violated was participants already being treated within CAMHS (defined as three or more sessions of active treatment in the months preceding randomisation) for seven participants: three in the FT arm and four in the TAU arm.

Research withdrawals

Participants could withdraw from further clinical data collection, further researcher follow-up visits or postal questionnaire follow-up. In 160 (19.2%) cases, the young person, the caregiver or both withdrew from some or all of the study process: 60 (14.5%) in the FT arm and 100 (24.0%) in the TAU arm.

Table 15 presents the type of withdrawals and who withdrew (young person or caregiver). All 160 participants withdrew from researcher follow-up visits, 136 (16.3%) withdrew from postal questionnaire follow-up (FT arm, 11.6%; TAU arm, 21.1%) and 22 (2.6%) withdrew from further clinical data collection (i.e. the primary end point), nine in the FT arm and 13 in the TAU arm.

Figure 7 presents the timing of withdrawals and shows that these were largely clustered around postal and researcher follow-up time points at 3, 6, 12 and 18 months post randomisation. The main reasons for withdrawal (Table 16) included the participant not wanting to be involved any more (38 participants); other issues or events going on in the participants’ or their family life, which meant that they were too busy to continue (30 participants); the participant having improved or moved on and not wanting to revisit the past (26 participants); and dissatisfaction with the therapy offered or received (18 participants).

FIGURE 7.

Time between randomisation and withdrawal.

Questionnaire follow-up

Postal questionnaires were completed and returned by 426 (51.2%) young people and 440 (52.9%) caregivers at 3 months, and by 353 (42.4%) young people and 363 (43.6%) caregivers at 6 months (Table 17). Response rates were higher in the FT group than in the TAU group.

Questionnaire completion at the 12- and 18-month researcher follow-up visits (Table 18) was similar to postal follow-up, with 465 (55.9%) young person questionnaires completed at 12 months and 395 (47.5%) at 18 months, again with higher follow-up rates in the FT group than in the TAU group.

At the 12-month researcher visit, full follow-up, defined as completion of questionnaires by the young person, caregiver and researcher, was achieved for 411 (49.4%) participants: 237 (57.1%) in the FT arm and 174 (41.7%) in the TAU. In the case of 350 (42.1%) participants, 147 (35.4%) in the FT arm and 203 (48.7%) in the TAU arm, no questionnaires at all were completed. Of the 421 participants for whom at least one questionnaire was missing, the most frequent reasons for loss to follow-up were that the researcher was unable to contact the participant (29%), the researcher made contact but was unable to arrange the visit (24.5%), both the young person and the caregiver withdrew consent (21.1%) and that a visit was arranged but was subsequently cancelled or no one was at home when the researcher visited (8.3%).

At the 18-month researcher visit, full follow-up was achieved for 352 (42.3%) participants: 197 (47.5%) in the FT group and 155 (37.2%) in the TAU group. No questionnaires at all were completed for 412 (49.5%) participants: 186 (44.8%) in the FT group and 226 (54.2%) in the TAU group. Of the 480 participants for whom at least one questionnaire was missing, the most frequent reasons for loss to follow-up largely mirrored the reasons at 12 months; however, inability to contact the participant and withdrawal were more frequently cited as reasons for all questionnaires being missing (37.1% and 29.6% of participants, respectively).

The timing of questionnaire completion for the young person is presented in Figure 8 with completion largely well defined around required time points and a similar distribution for the caregiver and the researcher questionnaires (at 12 and 18 months).

FIGURE 8.

Time between randomisation and young person questionnaire follow-up.

Researcher unblinding

Unblinding of one or more researchers occurred on a total of 190 occasions for 161 (19.4%) participants: 109 times in 93 participants (22.4%) in the FT group and 81 times in 68 participants (16.3%) in the TAU group (Table 19 and Figure 9). Unblinding most frequently occurred around the 12- and 18-month researcher follow-up time points (45.3% and 17.4%, respectively); however, 25.8% of unblinding occurred irrespective of a specific follow-up visit, largely within the first few months post randomisation and because a clinician notified the researcher. The rates and reasons for unblinding were similar across trial arms, with the most frequent cause of unblinding arising from the researcher being informed by the caregiver (74 cases, 38.9%), a clinician (44 cases, 23.2%), the young person (34 cases, 17.9%) or both the young person and the caregiver (18 cases, 9.5%), with a smaller number resulting from the young person’s response implying a particular intervention (eight cases, 4.2%). Some unblinding occurred during researcher follow-up: 22 prior to the assessment, 56 during the assessment and 30 after the assessment.

FIGURE 9.

Time between randomisation and unblinding.

To investigate the potential impact of unblinding on the primary outcome, we identified attendances where the only source of reason for presentation (self-harm related or not) was the unblinded researcher record [due to unclassified HES data, or researcher collected only attendances (two cases)]. There were 23 such attendances, involving 16 participants. Ten participants in the FT arm accounted for 16 attendances and six participants in the TAU arm accounted for seven attendances. The researchers reported three of these attendances to be related to self-harm (two in the FT arm and one in the TAU arm) and hospital follow-up took place a mean of 7 (SD 5.3) months after the researchers were unblinded (range 0–16 months).

Intention-to-treat population

All summaries and analyses have been carried out using the ITT population, which consists of all randomised participants, excluding those who withdrew their full consent to trial participation or for whom written informed consent was not obtained.

Written informed consent was received for all randomised participants and, although some participants withdrew their consent to researcher and postal contact and further clinical data collection, no participants withdrew their full consent and asked for data already collected to be removed. Therefore, the ITT population consists of all 832 participants.

Two participants consented to all trial procedures excluding the registration of their details with NHS Digital. Primary outcome data for these participants were not obtained from NHS Digital and were collected only during visits by the researcher to acute trusts.

Complete written informed consent was obtained from the primary caregiver for all but two participants: in one case, the caregiver’s consent form was not received and in another the form had not been dated. Implied consent was, however, obtained for both caregivers following the completion of the caregiver set of baseline questionnaires.

Per-protocol population

The per-protocol population consisted of the 816 (98.1%) participants without an eligibility violation. The number of patients excluded from the per-protocol population and the reasons are summarised in Table 14. No analyses were repeated for the per-protocol population.

Process and treatment summaries

Child and Adolescent Mental Health Services treatment pathways

Following randomisation, participants attended sessions within CAMHS following one of three main pathways (Figure 10 and Tables 20 and 21).

• Initial SHIFT FT or TAU sessions as per randomisation: at least one session was attended by the young person and/or a family member for 394 (94.9%) participants allocated to FT and 338 (81.1%) participants allocated to TAU.

• Slightly more participants in the TAU arm did not attend initial therapy: 34 (8.2%) participants, compared with 21 (5.1%) participants in the FT arm. At the end of initial treatment, 233 (56.1%) FT participants and 158 (37.9%) TAU participants had completed all of the required sessions.

• In the FT arm, 173 (41.7%) participants attended SHIFT FT for more than the anticipated 6 months or had more than the intended eight sessions. This was primarily as a result of unresolved issues, difficulties with timetabling or cancellations.

• Referrals within CAMHS for additional or alternative sessions: referrals were reported for 98 (23.6%) participants allocated to FT during or following their SHIFT FT and for 96 (23.0%) TAU participants referred as part of usual care. In the FT arm, 77 (18.6%) young people and/or a family member attended at least one referred session while 68 (16.3%) participants did so in the TAU arm.

• Reasons for referral show that the proportion of participants referred for psychiatric mental state/risk assessment including medication was higher in the FT arm than in the TAU arm (43.9% vs. 28.1%), as was the proportion referred for ongoing treatment after trial treatment finished (19.4% vs. 2.1%). In contrast, the proportion of participants referred for alternative/additional therapeutic input was higher in the TAU arm than in the FT arm (47.9% vs. 17.3%).

• Sessions that took place prior to initial SHIFT FT: in the FT arm, 84 (20.2%) participants and/or a family member attended at least one session. These sessions occurred because they had been booked prior to randomisation, or were booked as holding appointments because of waiting times, with the majority of sessions (88.1%) consisting of assessment only. Only 13 (11.9%) sessions in 10 (2.4%) participants were for treatment.

FIGURE 10.

Diagram of attendance of at least one session, by anyone, within CAMHS pathways.

TABLE 20 - Attendance of sessions within initial randomised treatment in CAMHS

Other reasons for additional time/sessions were that the family was keen to have expected final session; because of concerns arising and to minimise risks between SHIFT ending and referral for assessment and treatment commencing in CAMHS; developmental history had to be taken for autistic spectrum disorder assessment referral; father joined therapy, which extended the work; family felt issues were resolved.

Attendance at sessions, n (%)	FT (N = 415)	TAU (N = 417)	Total (N = 832)
Any attendance (young person or family) during initial FT/TAU
Yes	394 (94.9)	338 (81.1)	732 (88.0)
No	21 (5.1)	34 (8.2)	55 (6.6)
Missing	0 (0.0)	45 (10.8)	45 (5.4)
Young person’s attendance summary
Did not attend at all	23 (5.5)	39 (9.4)	62 (7.5)
Dropped out with no negotiation	66 (15.9)	83 (19.9)	149 (17.9)
Negotiated ending in response to threat of dropout	92 (22.2)	57 (13.7)	149 (17.9)
Completed all required sessions	233 (56.1)	158 (37.9)	391 (47.0)
Treatment ongoing	0 (0.0)	10 (2.4)	10 (1.2)
Missing	1 (0.2)	70 (16.8)	71 (8.5)
Did the family attend SHIFT FT for more than 6 months and/or more than eight sessions?
Yes	173 (41.7)
No	242 (58.3)
Rationale for additional time/sessions
Issues not resolved, more sessions required	104 (60.1)
Difficulty timetabling/cancellations	42 (24.3)
Othera	5 (2.9)
Missing	22 (12.7)
Total – attending for > 6 months or > 8 sessions	173 (100.0)

TABLE 21 - Referrals within CAMHS and additional sessions attended within CAMHS

N/A, not applicable.

Twenty-six young persons (11 FT, 15 TAU) were reported to have been referred but were missing details of appointments for the referral and a further 23 young persons (10 FT, 13 TAU) were referred but attended no sessions. There were a total of 202 referrals in 194 young persons with 141 (72 FT, 69 TAU) referred once, 24 (14 FT, 10 TAU) referred twice, one young person (TAU) referred three times and two young persons (one FT, one TAU) referred five times each, and the remaining 26 were missing details of all referrals.

Referrals within CAMHS and additional CAMHS sessions, n (%)	FT (N = 415)	TAU (N = 417)	Total (N = 832)
Young person referred within CAMHS
Yesa	98 (23.6)	96 (23.0)	194 (23.3)
Attended referred session(s)	77 (18.6)	68 (16.3)	145 (17.4)
No	314 (75.7)	267 (64.0)	581 (69.8)
Missing	3 (0.7)	54 (12.9)	57 (6.9)
Reason for referral (non-mutually exclusive)
Assessment
Psychiatric mental state/risk assessment including medication	43 (43.9)	27 (28.1)	70 (36.1)
Treatment
For alternative/additional therapeutic input	17 (17.3)	46 (47.9)	63 (32.5)
Ongoing treatment after trial treatment finished	19 (19.4)	2 (2.1)	21 (10.8)
For more intensive treatment (e.g. day/inpatient unit)	3 (3.1)	2 (2.1)	5 (2.6)
Assessment/treatment
For input to specific comorbid diagnoses (e.g. ASD, eating disorder)	12 (12.2)	7 (7.3)	19 (9.8)
Other
Administrative reasons (e.g. change of address, local protocol, travel problems)	3 (3.1)	4 (4.2)	7 (3.6)
Missing	11 (11.2)	16 (16.7)	27 (13.9)
Total number referred	98 (100)	96 (100)	194 (100)
Young person attended sessions prior to SHIFT FT
Yes	84 (20.2)	N/A	N/A
Treatment sessions	8 (1.9)
Assessment sessions	74 (17.8)
Assessment and treatment sessions	2 (0.5)
No	302 (72.8)
Missing	29 (7.0)
Purpose of attended sessions
Treatment	13 (11.9)
Assessment	96 (88.1)
Total number of sessions prior to SHIFT	109 (100.0)

Overall, 399 (96.1%) FT participants and 339 (81.3%) TAU participants had at least one session attended within CAMHS by either the participant (young person) or family member.

Number and duration of sessions within Child and Adolescent Mental Health Services

During participants’ initial randomised treatment, the number of SHIFT FT sessions attended by the family or the young person ranged from 0 to 21, with a mean of 6.1 (SD 4.02) sessions over a mean period of 4.7 (SD 3.03) months. The number of initial sessions attended in the TAU arm was more variable, ranging from 0 to 163, with a mean of 8.2 (SD 12.61) sessions over a mean period of 5.7 (SD 5.41) months (Tables 22 and 23 and Figures 11–14). As a result of highly skewed data, although some TAU participants a large number of sessions, the overall median number of sessions and duration of attendance was lower in the TAU arm, a median of 4 sessions over 4.1 months, compared with 6 sessions over 5.1 months in the FT arm.

TABLE 22 - Number of CAMHS sessions attended during initial randomised treatment, additional referrals and sessions prior to SHIFT FT

Number of sessions attended	FT, n (%) (N = 415)				TAU, n (%) (N = 417)			Total, n (%) (N = 832)
	Initial FT	Additional TAU referral	Sessions prior to initial FT	All sessions	Initial TAU	Additional TAU referral	All sessions	All sessions
0	21 (5.1)	324 (78.1)	302 (72.8)	16 (3.9)	34 (8.2)	280 (67.1)	33 (7.9)	49 (5.9)
1	40 (9.6)	16 (3.9)	68 (16.4)	28 (6.7)	52 (12.5)	23 (5.5)	45 (10.8)	73 (8.8)
2	29 (7.0)	19 (4.6)	11 (2.7)	32 (7.7)	37 (8.9)	10 (2.4)	36 (8.6)	68 (8.2)
3	37 (8.9)	2 (0.5)	2 (0.5)	36 (8.7)	45 (10.8)	8 (1.9)	43 (10.3)	79 (9.5)
4	32 (7.7)	5 (1.2)	2 (0.5)	29 (7.0)	23 (5.5)	4 (1.0)	23 (5.5)	52 (6.3)
5	29 (7.0)	6 (1.4)	1 (0.2)	31 (7.5)	28 (6.7)	4 (1.0)	30 (7.2)	61 (7.3)
6	43 (10.4)	3 (0.7)		41 (9.9)	13 (3.1)	1 (0.2)	15 (3.6)	56 (6.7)
7	29 (7.0)	2 (0.5)		32 (7.7)	15 (3.6)	2 (0.5)	10 (2.4)	42 (5.0)
8	50 (12.0)	4 (1.0)		38 (9.2)	11 (2.6)	1 (0.2)	12 (2.9)	50 (6.0)
9	30 (7.2)	3 (0.7)		21 (5.1)	12 (2.9)	2 (0.5)	11 (2.6)	32 (3.8)
10	24 (5.8)	2 (0.5)		23 (5.5)	10 (2.4)	4 (1.0)	9 (2.2)	32 (3.8)
11–15	38 (9.2)	7 (1.7)		48 (11.6)	29 (7.0)	2 (0.5)	32 (7.7)	80 (9.6)
16–20	12 (2.9)	3 (0.7)		25 (6.0)	27 (6.5)	3 (0.7)	26 (6.2)	51 (6.1)
21–25	1 (0.2)	0 (0.0)		6 (1.4)	17 (4.1)	2 (0.5)	19 (4.6)	25 (3.0)
> 26	0 (0.0)	5 (1.2)		9 (2.2)	19 (4.6)	2 (0.5)	28 (6.7)	37 (4.4)
Missing	0 (0.0)	14 (3.4)	29 (7.0)	0 (0.0)	45 (10.8)	69 (16.5)	45 (10.8)	45 (5.4)
Number of sessions attended
N	415	401	386	415	372	348	372	787
Missing	0	14	29	0	45	69	45	45
Mean (SD)	6.1 (4.02)	1.4 (5.39)	0.3 (0.64)	7.7 (7.23)	8.2 (12.61)	1.1 (4.44)	9.3 (13.85)	8.5 (10.90)
Median (range)	6 (0–21)	0 (0–57)	0 (0–5)	6 (0–70)	4 (0–163)	0 (0–51)	5 (0–163)	6 (0–163)

TABLE 23 - Timing and duration of treatment

Timing and duration of treatment	FT			TAU			Total
	Initial FT (N = 415)	Additional TAU referral (N = 98)	All sessions (N = 415)	Initial TAU (N = 417)	Additional TAU referral (N = 96)	All sessions (N = 417)	All sessions (N = 832)
Time to first attendance (days/months)	Days	Months	Days	Days	Months	Days	Days
n	394	77	399	338	67	339	738
Missing	0	11	0	45	16	45	45
Did not attend sessions	21	10	16	34	13	33	49
Mean (SD)	24.5 (21.49)	4.8 (3.75)	19.9 (18.81)	37.0 (44.87)	6.2 (4.27)	36.8 (44.77)	27.6 (34.36)
Median (range)	18.0 (1–137)	4.1 (0.1–12.9)	15.0 (0–137)	22.0 (0–380)	5.3 (0.2–17.7)	22.0 (0–380)	17.0 (0–380)
Time to last attendance (months)
n	394	77	399	338	67	339	738
Missing	0	11	0	45	16	45	45
Did not attend sessions	21	10	16	34	13	33	49
Mean (SD)	5.5 (3.06)	9.5 (5.83)	6.2 (4.03)	6.9 (5.40)	9.3 (5.31)	7.4 (5.59)	6.7 (4.84)
Median (range)	5.8 (0.3–17.4)	8.7 (0.7–18.0)	5.8 (0.3–18.0)	5.0 (0.2–18.0)	9.0 (1.1–18.0)	5.3 (0.2–18.0)	5.7 (0.2–18.0)
Duration first to last attendance (months)
n	394	77	399	338	67	339	738
Missing	21	21	16	79	29	78	94
Did not attend sessions	21	10	16	34	13	33	49
Mean (SD)	4.7 (3.03)	4.7 (4.99)	5.5 (4.06)	5.7 (5.41)	3.1 (3.83)	6.2 (5.63)	5.8 (4.86)
Median (range)	5.1 (0.0–15.7)	3.0 (0.0–17.7)	5.3 (0.0–17.9)	4.1 (0.0–17.7)	1.4 (0.0–15.4)	4.4 (0.0–17.9)	5.1 (0.0–17.9)

FIGURE 11.

Number of initial sessions attended by anyone. Not displayed are three further participants in the TAU arm who had 75, 90 and 163 initial and overall sessions attended by anyone and one participant in the FT arm who had 70 overall sessions attended by anyone. (a) FT and (b) TAU.

FIGURE 12.

Number of overall sessions attended by anyone. Not displayed are three further participants in the TAU arm with attendance of 75, 90 and 163 initial and overall sessions and one participant in the FT arm with attendance of 70 overall sessions. (a) FT and (b) TAU.

FIGURE 13.

Duration of initial randomised treatment (excluding referrals and sessions prior to SHIFT FT). (a) FT and (b) TAU.

FIGURE 14.

Duration of overall treatment (including referrals and sessions prior to SHIFT FT). (a) FT and (b) TAU.

The inclusion of additional referred sessions attended by 96 (23.0%) participants in the TAU arm referred within CAMHS to additional sessions or sessions alternative to their initial treatment provides the overall summary of all TAU received. Combined with initial TAU sessions, the overall number of all sessions attended ranged from 0 to 163, with a mean of 9.3 (SD 13.85) sessions over a mean period of 6.2 (SD 5.63) months and a median of 5 sessions over a period of 4.4 months.

In the FT arm, 98 (23.6%) participants received additional TAU sessions as part of a referral within CAMHS. The number of referred sessions attended ranged from 0 to 57, with a mean of 1.4 (SD 5.39) sessions over a mean period of 4.7 (4.99) months and a median of 0 sessions over 3 months. Combined with initial SHIFT FT sessions and sessions prior to starting FT, the overall number of all sessions attended in the FT arm ranged from 0 to 70, with a mean of 7.7 (SD 7.23) sessions over a mean period of 5.5 months (SD 4.06 months) and a median of 6 sessions over a median period of 5.3 months.

Overall therapeutic orientation and therapy pathways

Overall, 398 (95.9%) FT participants and 301 (72.2%) TAU participants attended at least one therapeutically orientated session (excluding assessment sessions), and one FT participant and 32 TAU participants attended assessment sessions only (Table 25). A further six TAU participants attended only sessions for which the content was unknown, while 16 (3.9%) FT participants and 33 (7.9%) TAU participants attended no sessions at all.

TABLE 25 - Treatment mix of sessions: therapy and assessment

Treatment mix of sessions, n (%)	FT (N = 415)	TAU (N = 417)	Total (N = 832)
Any therapy sessions attended?
Yes	398 (95.9)	301 (72.2)	699 (84.0)
No	17 (4.1)	71 (17.0)	88 (10.6)
Missing	0 (0.0)	45 (10.8)	45 (5.4)
Any assessment sessions attended?
Yes	117 (28.2)	197 (47.2)	314 (37.7)
No	298 (71.8)	175 (42.0)	473 (56.9)
Missing	0 (0.0)	45 (10.8)	45 (5.4)
Overall session mix
No sessions attended	16 (3.9)	33 (7.9)	49 (5.9)
Assessment only	1 (0.2)	32 (7.7)	33 (4.0)
Therapy only	282 (68.0)	136 (32.6)	418 (50.2)
Assessment and therapy	116 (28.0)	165 (39.6)	281 (33.8)
Unknown only	0 (0.0)	6 (1.4)	6 (0.7)
Missing	0 (0.0)	45 (10.8)	45 (5.4)

Therapeutic orientation of sessions in TAU varied considerably (Table 26). One-quarter of all sessions were for supportive therapy/counselling, 17.4% were for cognitive–behavioural therapy (CBT), 11.5% were for family work, 10.7% were for formal systemic FT and < 5% were for each of psychodynamic therapy, communication skills/problem-solving, interpersonal therapy, dialectical behaviour therapy (DBT), psychoeducational therapies and other therapy. The most frequently attended sessions on a per-participant basis were supportive therapy/counselling, family work, CBT and formal systemic FT. A substantial proportion of sessions were for assessment rather than therapy, with 12.5% of all sessions in 161 (38.6%) participants for other assessment/review.

TABLE 26 - Therapeutic orientation of all sessions and per participant

DBT, dialectical behaviour therapy.

For example, introductory meetings/administration.

Reproduced from Cottrell et al. 70 © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

Therapeutic orientation of sessions, n (%)	Therapeutic orientation per session		Therapeutic orientation per participant (non-mutually exclusive)
	FT (N = 3207)	TAU (N = 3466)	FT (N = 415)	TAU (N = 417)
Therapy
SHIFT FT	2532 (79.0)	0 (0.0)	394 (94.9)	0 (0.0)
Supportive therapy/counselling	142 (4.4)	871 (25.1)	21 (5.1)	158 (37.9)
CBT	79 (2.5)	602 (17.4)	10 (2.4)	88 (21.1)
Family work	35 (1.1)	397 (11.5)	9 (2.2)	116 (27.8)
Formal systemic FT	14 (0.4)	371 (10.7)	5 (1.2)	87 (20.9)
Communication skills/problem-solving	0 (0.0)	62 (1.8)	0 (0.0)	29 (7.0)
Psychoeducational	22 (0.7)	29 (0.8)	6 (1.4)	18 (4.3)
Interpersonal therapy	5 (0.2)	38 (1.1)	1 (0.2)	10 (2.4)
DBT	0 (0.0)	30 (0.9)	0 (0.0)	8 (1.9)
Psychodynamic	1 (0.0)	131 (3.8)	1 (0.2)	5 (1.2)
Other therapy	78 (2.4)	168 (4.8)	17 (4.1)	28 (6.7)
Assessment
Mental state/risk assessment	110 (3.4)	137 (4.0)	47 (11.3)	58 (13.9)
Other assessment/review	153 (4.8)	434 (12.5)	89 (21.4)	161 (38.6)
Medication review	26 (0.8)	80 (2.3)	13 (3.1)	28 (6.7)
Non-therapya	2 (0.1)	27 (0.8)	1 (0.2)	23 (5.5)
Unknown: per session/all sessions	8 (0.2)	89 (2.6)	5 (1.2)	27 (6.5)
No sessions attended	–	–	16 (3.9)	33 (7.9)
Missing all treatment data	–	–	0 (0.0)	45 (10.8)

Seventy-nine per cent of all sessions attended in the FT arm were SHIFT sessions, with other types of sessions accounting for < 5% each. Eighty-nine participants (21.4%) attended an assessment/review and 47 participants (11.3%) attended a mental state/risk assessment.

The main therapeutic orientation(s), making up one-third or more of all therapy sessions per participant (excluding assessment and unknown sessions), varied from one to a maximum of three main therapies per participant (Table 27).

TABLE 27 - Main therapeutic orientation(s) of sessions (> 33% of therapy sessions)

Note

Up to three main therapeutic orientations were derived per young person. Note that four TAU young persons had evenly spread sessions across three orientations, with each slightly below the 33% threshold, and these have been included.

FT (N = 415)		TAU (N = 417)
Main therapy (not mutually exclusive)	n (%)	Main therapy (not mutually exclusive)	n (%)
No sessions attended	16 (3.9)	No sessions attended	33 (7.9)
Assessment only	1 (0.2)	Assessment only	32 (7.7)
SHIFT FT	389 (93.7)	Supportive therapy/counselling	115 (27.6)
Supportive therapy/counselling	14 (3.4)	CBT	73 (17.5)
CBT	6 (1.4)	Family work	66 (15.8)
Other therapy	6 (1.4)	Formal systemic FT	57 (13.7)
Psychoeducational	3 (0.7)	Communication skills/problem-solving	15 (3.6)
Family work	2 (0.5)	Other therapy	15 (3.6)
Interpersonal therapy	1 (0.2)	Psychoeducational	5 (1.2)
		Interpersonal therapy	4 (1.0)
		Psychodynamic	4 (1.0)
		DBT	3 (0.7)
		Unknown session orientation	6 (1.4)
		Missing all treatment data	45 (10.8)

In the TAU arm, supportive therapy/counselling was again one of the main therapies for 115 (27.6%) participants, CBT was the main therapy for 73 (17.5%), family work was the main therapy for 66 (15.8%) and formal systemic FT was the main therapy for 57 (13.7%). As the main therapy comprised up to three different therapies, these summaries are not mutually exclusive and full details of combinations of therapies can be found in Table 28. In the TAU arm, 256 (61.4%) participants attended only one type of therapy in over one-third of all attended sessions, 46 (11.0%) participants attended two main therapies and five (1.2%) participants attended three main therapies.

TABLE 28 - Main therapeutic orientation of sessions (> 33% of therapy sessions) for participants randomised to receive TAU (mutually exclusive)

Main therapy, n (%)	TAU (N = 417)
Single main therapy	256 (61.4)
Supportive therapy/counselling	85 (20.4)
CBT	51 (12.2)
Family work	44 (10.6)
Formal systemic FT	43 (10.3)
Other therapy	9 (2.2)
Communication skills/problem-solving	8 (1.9)
Unknown only	6 (1.4)
Interpersonal therapy	4 (1.0)
Psychodynamic	4 (1.0)
Psychoeducational	2 (0.5)
Two main therapies	46 (11.0)
Supportive therapy/counselling, family work	14 (3.4)
CBT, formal systemic FT	7 (1.7)
CBT, supportive therapy/counselling	6 (1.4)
CBT, family work	2 (0.5)
Communication skills/problem-solving, family work	2 (0.5)
DBT, supportive therapy/counselling	2 (0.5)
Formal systemic FT, other therapy	2 (0.5)
Supportive therapy/counselling, communication skills/problem-solving	2 (0.5)
Supportive therapy/counselling, other therapy	2 (0.5)
CBT, DBT	1 (0.2)
CBT, other therapy	1 (0.2)
CBT, psychoeducational	1 (0.2)
Communication skills/problem solving, formal systemic FT	1 (0.2)
Family work, formal systemic FT	1 (0.2)
Psycho-educational, supportive therapy/counselling	1 (0.2)
Supportive therapy/counselling, formal systemic FT	1 (0.2)
Three main therapies	5 (1.2)
CBT, family work, formal systemic FT	1 (0.2)
CBT, formal systemic FT, other therapy	1 (0.2)
CBT, supportive therapy/counselling, communication skills/problem-solving	1 (0.2)
CBT, supportive therapy/counselling, family work	1 (0.2)
Psychoeducational, communication skills/problem-solving, family work	1 (0.2)
Assessment only	32 (7.7)
No sessions attended	33 (7.9)
Missing	45 (10.8)

In the FT arm, the allocated therapy, SHIFT, was the main therapy attended by the vast majority (94%) of participants; however, the main therapies attended in addition to, or instead of, SHIFT are detailed in Table 29.

TABLE 29 - Main therapeutic orientation of sessions (> 33% of therapy sessions) attended by participants randomised to receive FT (mutually exclusive)

Main therapy, n (%)	FT (N = 415)
Single main therapy	375 (90.4)
SHIFT FT	367 (88.4)
Supportive therapy/counselling	4 (1.0)
Other therapy	2 (0.5)
Family work	1 (0.2)
Psychoeducational	1 (0.2)
Two main therapies	23 (5.5)
SHIFT FT, supportive therapy/counselling	10 (2.4)
SHIFT FT, CBT	5 (1.2)
SHIFT FT, other therapy	4 (1.0)
SHIFT FT, psychoeducational	2 (0.5)
SHIFT FT, interpersonal therapy	1 (0.2)
CBT, family work	1 (0.2)
Assessment only	1 (0.2)
No sessions attended	16 (3.9)

Eighty-seven (20.9%) participants in the TAU arm were reported to have attended at least one formal systemic FT session, albeit of a type different from SHIFT, which is a specific manualised form of formal systemic FT (Table 30 and Figure 15). Furthermore, five (1.2%) participants in the FT arm, in addition to attending 7–16 SHIFT sessions, also attended 1–5 sessions of formal systemic FT as part of a referral within CAMHS for ongoing treatment after trial treatment finished.

TABLE 30 - Summary of FT sessions received

N/A, not applicable.

For three young persons, the TAU Formal systemic FT was delivered by non-SHIFT clinicians. For one young person, a trained SHIFT clinician was involved; however, this clinician had not been part of the assigned team initially treating the young person for SHIFT, whereas for the remaining young person, two additional non-SHIFT formal systemic FT sessions were received immediately following the end of SHIFT FT involving the same clinicians, albeit with the addition of a non-SHIFT clinician who continued to then see the young person for family work and other therapy.

FT sessions received, n (%)	FT (N = 415)	TAU (N = 417)	Total (N = 832)
Any SHIFT FT?
Yes	394 (94.9)	0 (0.0)	394 (47.4)
No	21 (5.1)	372 (89.2)	393 (47.2)
Missing	0 (0.0)	45 (10.8)	45 (5.4)
Any TAU formal systemic FT?
Yes	5 (1.2)a	87 (20.9)	92 (11.1)
No	410 (98.8)	285 (68.3)	695 (83.5)
Missing	0 (0.0)	45 (10.8)	45 (5.4)
Any FT?
Yes	394 (94.9)	87 (20.9)	481 (57.8)
SHIFT FT	389 (93.7)	0 (0.0)	389 (46.8)
SHIFT FT and TAU FT	5 (1.2)	N/A	5 (0.6)
TAU FT	0 (0.0)	87 (20.9)	87 (10.5)
No	21 (5.1)	285 (68.3)	306 (36.8)
Missing	0 (0.0)	45 (10.8)	45 (5.4)

FIGURE 15.

Diagram of FT attendance across trial arms.

Therapist characteristics and training

Thirty-one family therapists received training in SHIFT FT. Training was conducted in formal group sessions for the majority of family therapists and via individual training and observation for those few therapists who joined the trial late following staff changes. Family therapists recruited early to the trial had the opportunity to deliver therapy to at least one pilot case prior to delivering trial FT; therapists recruited following staff changes always observed existing team members prior to commencing trial FT.

Baseline characteristics were received for 189 (65.6%) of the 288 treating therapists: 29 (93.5%) SHIFT family therapists and 160 (62.3%) TAU clinicians. The mean age of family therapists was 48.9 (SD 6.95) years and of TAU clinicians was 43.6 (SD 10.58) years. There were more female therapists than males: 21 family therapists (72.4%) and 117 (73.1%) TAU clinicians were women. Job titles varied for TAU clinicians, with over half listed as practitioners (31, 19.4%), clinical psychologists (28, 17.5%) or nurses (27, 16.9%). Eighteen (11.3%) TAU clinicians had family therapist as their job title. As the use of job titles changes from service to service, TAU clinicians were classified into ‘senior’ (doctors at consultant level and all other non-medical staff on pay grade 8C or above) and ‘non-senior’ (all other clinicians). All SHIFT family therapists were classified as ‘senior’ whereas only five (3.1%) TAU clinicians were. The majority of family therapists (20, 60.6%) reported that their basic qualification was in social work, while TAU clinicians were trained in nursing (51, 25.2%), social work (30, 16.9%) and clinical psychology (29, 16.4%). Of the additional training received, 45.9% of training received by the SHIFT family therapists was in formal systemic therapy, compared with 21.4% for the TAU clinicians. TAU clinicians received more additional training in CBT and behaviour therapy and communication skills/problem-solving techniques.

Adherence and supervision

One thousand six hundred and fifty-four (65.3%) initial FT sessions were recorded and, of those, the digital versatile disc (DVD) was returned for 1433 (86.6%). Fifty-two (3.6%) DVDs underwent initial review and, of these, the young person was present in all but one session. Twenty-three (44.2%) reviewed recordings related to the first therapy session and 29 (55.8%) related to later sessions. Twenty-six (83.8%) family therapists had therapy sessions reviewed; two sessions were reviewed for each therapist.

Overall adherence for the 52 reviewed sessions was scored between 0 and 5, with 5 indicating greater adherence. Overall competence was scored between 0 and 6, with 6 indicating greater competence. Initial reviews scored overall mean adherence as 4.6 (SD 0.72) and overall competence as 4.4 (1.03).

Specialist reviews were conducted when the overall competence score was 0, 1 or 2 or when requested by the initial reviewer. Specialist reviews were conducted for 7 (13.5%) of the initially reviewed sessions and the young person was present for all of the sessions. Three (42.9%) related to the first session and four (57.1%) related to later therapy sessions. Five family therapists had specialist reviews; two sessions were reviewed for two therapists and one session was reviewed for three therapists.

Tables 33 and 34 present a comparison of overall adherence scores and competence ratings between the initial and specialist reviews.

Two hundred and thirty-eight (67.2%) families who attended at least two sessions of SHIFT FT received a formulation letter from their family therapist. Formulation letters were sent by 25 (80.6%) of the 31 SHIFT family therapists. Compliance by therapist, for those who sent letters, ranged between 22.0% and 100.0%, with a median of 80.0%.

All but one of the family therapists had at least one session of supervision; the family therapist who did not receive supervision was not a main therapist and attended only one session of FT, as part of the reflecting team. Two hundred and twenty-five supervision sessions were conducted in total: 98 in Yorkshire, 38 in Manchester and 89 in London. The number of sessions received by individual therapist is presented by hub in Table 35. Family therapists in Yorkshire received the most supervision followed by London and Manchester. Two hundred and sixty-six young people (119 in Yorkshire, 56 in Manchester and 97 in London) were discussed 584 times in supervision; overall, each young person was discussed a mean of 2.2 (SD 1.45) times, and this was similar for both Yorkshire and London; however, young people in Manchester were discussed less often (1.5 times, SD 0.76 times). The mean length of supervision sessions overall was 96.3 (SD 38.97) minutes and this was similar between hubs.

Family therapy alliance

The strength of therapeutic alliance as reported by the young person, caregiver and SHIFT family therapist as captured in the SOFTA questionnaire at the participant’s third treatment session is summarised in Table 36 and Figure 16, with higher scores representing greater alliance. Caregivers consistently reported the highest levels of alliance across all four dimensions, while the therapist generally reported the lowest, with the exception of ‘engagement in the therapeutic process’, for which the young person reported lower alliance. Young person and therapist alliance were most closely aligned, with overlapping CIs, for the two dimensions reflecting the uniqueness of conjoint treatment ‘shared sense of purpose’ and ‘safety within the therapeutic system’, while there was no overlap for engagement or ‘emotional connection’, nor for any of the caregiver scores. Alliance is further explored by the participant’s primary outcome in the mediation results section.

FIGURE 16.

The SOFTA subscale scores with 95% CIs.

Psychotropic medications

During treatment, clinicians and CAMHS notes reported that 104 (12.5%) young people had been prescribed a psychotropic medication (Table 37): 44 (10.6%) young people in the FT arm and 60 (14.4%) young people in the TAU arm. The most frequently prescribed psychotropic medications were antidepressants, prescribed for 80 (76.9%) of these young people (reported by clinician, CSO or family), which in the majority of cases (all but five prescriptions) were selective serotonin reuptake inhibitors as opposed to tricyclic antidepressants.

A total of 118 (14.2%) young people were taking or had been prescribed a psychotropic medication at baseline or during the 18-month trial period: 48 (11.6%) in the FT arm and 70 (16.8%) in the TAU arm.

Re-referrals to Child and Adolescent Mental Health Services

Re-referrals to CAMHS following discharge (or inactivity for 3 months) were reported from clinical notes within CAMHS by the participant’s CAMHS clinician or a CSO, and also by the family during the 18-month researcher visit (Table 38).

Based on clinical data, 108 (13%) participants were re-referred to CAMHS within 18 months of randomisation, with similar rates between the trial arms; however, this information was missing for 19% of participants. Of all re-referred participants, 12 (11.1%) were re-referred on more than one occasion and the referrals of 83 (76.9%) were related to self-harm: 44 (84.5%) in the FT arm and 39 (69.6%) in the TAU arm.

Details of family-reported re-referrals were missing for over three-quarters of participants (because no researcher had visited or family had not been asked); however, re-referrals were reported for 39 (4.7%) participants.

Referrals to psychiatric day or inpatient units

Referrals to psychiatric inpatient units within 18 months of randomisation were reported from clinical notes within CAMHS by participants’ CAMHS clinician or CSO or via HES data, and also by the family during the 18-month researcher visit (Table 39).

Based on clinical data, a total of 25 (3%) participants were referred to a psychiatric inpatient unit during the 18-month follow-up period, of whom 21 (84.0%) had a referral related to self-harm, with similar rates between the trial arms. However, information was missing for 149 (17.9%) participants. The total length of inpatient stays over all referrals ranged from 1 to 185 days, with a mean of 72.2 days in the FT arm and 43.3 days in the TAU arm.

A total of seven referrals to psychiatric inpatient units were reported for five (0.6%) participants by the family, of which five had also been reported via the clinical data, with one referral in each arm reported by the family but not from the clinical data.

Referrals to other agencies

Referrals to adult mental health services or other agencies for the participating young person, their caregiver(s) or sibling(s) within 18 months of randomisation were reported from clinical notes within CAMHS by the participant’s CAMHS clinician or a CSO, and also by the family during the 18-month researcher visit (Table 40).

Based on clinical data, a total of 18 (2.2%) young people or family members were referred to adult mental health services: eight young people only referred, nine family members only referred, and one young person and their family member referred. Information was missing for 155 (18.6%) young people. Referrals were reported within CAMHS notes only if the referral had come from CAMHS or if the young person had been in contact with CAMHS at the time. A higher proportion of adult mental health referrals were reported by the family: 25 (3%) participants (nine young people, 11 caregivers and five siblings).

Conversely, referrals to other agencies were reported for 143 (17.2%) young people, caregivers or siblings via clinical data and for 38 (4.6%) young people, caregivers or siblings by the family. Based on clinical report, 68 referrals were to social services, 24 referrals were to special education services, 10 referrals were to the voluntary sector, six referrals were to youth offending team(s) and 118 referrals were to a wide range of other agencies or services.

Overall safety

A total of 1036 AEs were reported, defined as treatment on an emergency outpatient basis through an A&E attendance, attendance at a MIU or WIC and re-referral to CAMHS. A&E attendances constituted the largest portion of AEs (Table 41). AEs were reported in 443 (53.2%) participants, with a similar number of events and participants between the trial arms, and a mean of 1.2 AEs, ranging from 0 to 26 per participant.

No deaths were reported. All 598 SAEs reported for 297 (35.7%) participants were hospital admissions and readmissions. Again, similar rates were observed in both trial arms, although slightly more SAEs were reported in fewer participants in the TAU arm than in the FT arm: 323 SAEs in 141 (33.8%) in the TAU arm compared with 275 SAEs in 156 (37.6%) participants in the TAU arm.

Further details of the total number of hospital-related events – MIU/WIC attendances, A&E attendances and hospital admissions – are presented in Table 42. Overall, 1513 attendances were reported in 511 (61.4%) participants, with similar rates in both trial arms and a mean of 1.8 attendances per participant.

Over half of all attendances were for non-mental health reasons (59.7%), while 26.7% were for self-harm, 4.2% were for other mental health reasons and 3.4% resulted from a combination of mental health and non-mental health reasons (Table 43). A total of 387 (46.5%) participants attended hospital or a MIU or WIC at least once for non-mental health reasons; 223 (26.8%) participants attended for self-harm, 85 (10.2%) participants for other mental health reasons and in eight (1.0%) participants the reasons for all attendances were unknown.

Hospital follow-up

Following the interim review of primary outcome results, it was agreed that the stopping rule had not been reached and the DMEC recommended that the trial continue.

Table 44 details the length of follow-up for hospital attendance for all participants.

A final HES data download was obtained from NHS Digital in May 2015, providing HES data to 31 January 2015. The final download provided full data for the majority of participants: 724 (87.0%) recruited prior to 31 July 2013 for whom HES data covered the entire 18 month follow-up period. A further 75 participants had partial HES coverage during their follow-up period: 71 (8.5%) recruited after 31 July 2013 for whom HES data covered at least the first 13 months’ follow-up, and four (0.5%) for whom linkage to HES had been unsuccessful for at least one prior NHS Digital download resulting in incomplete HES data from randomisation until their first successful linkage (1.3, 3.8, 11.1 and 14.3 months).

For the 75 participants with partial HES coverage and a further 12 (1.4%) participants for whom linkage to HES was not made during any data download, hospital attendance data were obtained via researcher follow-up at acute trusts within participants’ recruiting hub. A further 21 (2.5%) participants withdrew from further clinical data collection during the trial, thus preventing further linkage to HES data or researcher follow-up.

In summary, a total of 795 (95.6%) participants had complete follow-up over their entire 18 months post randomisation. Four (0.5%) participants were completely lost to follow-up and 33 (4%) participants had at least some but not complete 18-month hospital follow-up.

The duration of follow-up for all participants can be seen to be evenly balanced between the trial arms (see Table 44).

Summary of primary outcome events

A total of 221 (26.6%) young persons experienced the primary outcome event, that is, having a repeat self-harm event leading to hospital attendance within 18 months post randomisation: 118 (28.4%) young persons randomised to receive FT and 103 (24.7%) young persons randomised to receive TAU.

Table 45 details the type and outcome of all primary outcome events. The overall median time to first post-randomisation self-harm event is 5.2 months. This is earlier for young persons randomised to TAU (4.3 months, compared with 5.7 months in the case of young persons randomised to FT).

The most common reason for attendance was self-poisoning (57.5% of primary outcome events), with a further 19.9% attending as a result of cutting. The most common outcome was admission to a hospital ward, for 161 (72.9%) events, with similar rates of admission between trial arms.

For participants with a primary outcome event, Table 46 presents the type of event by the method used in the index event prompting referral into CAMHS. Among those participants whose index episode was self-poisoning, the first repeat event was also self-poisoning in the majority (38, 73.1%), whereas 11 (21.2%) self-injured and three (5.8%) used combined methods. Of those who used combined methods at the index event, similar proportions were observed, with the majority self-poisoning at first repeat (16, 66.7%). Of those in whom the index event was self-injury, a greater proportion switched method, with only 59 (40.7%) self-injuring at first repeat, 73 (50.3%) self-poisoning and 12 (8.3%) using combined methods.

Table 47 presents the number of participants with one or more event, the number without an event and censored in the analysis, the reasons for censoring and the time to censoring. The majority of censored participants (n = 583, 95.4%) were censored because no observed self-harm events resulting in hospital attendance occurred over the entire 18-month follow-up period; however, 15 (2.5%) participants were censored because of their withdrawal from further clinical data collection, as no events were observed prior to their withdrawal; and a further 13 (2.2%) participants were censored either because they were not linked to HES data and relied on researcher follow-up alone or because their HES data did not include their full 18-month follow-up period; in both cases researcher follow-up was available only for the periods not covered via HES data. Similar patterns of censoring are observed between trial arms.

Primary analysis: Cox proportional hazards regression

The results of the Cox proportional hazards regression modelling, adjusted for covariates, are displayed in Table 48. A HR of > 1 indicates an increased rate of self-harm in the variable listed before ‘vs.’ in the table (e.g. participants with three or more previous self-harm episodes repeat self-harm at 1.22 times the rate of participants with just two previous self-harm episodes).

There was no evidence to suggest a statistically significant difference in self-harm repetition rates between treatment groups. The HR for FT compared with TAU was 1.14 (95% CI 0.87 to 1.49; p = 0.3349). Figure 17 presents the Kaplan–Meier curve for time to self-harm by randomised treatment arm. There is crossover between treatment groups until around month 11, after which the curves diverge and the lower curve represents slightly more FT participants with a primary outcome self-harm event. Further estimates of the self-harm repetition rate, with 95% CIs, at 1, 3, 6, 12 and 18 months post randomisation, are presented in Table 49 by arm and for the difference between arms, with a positive difference indicating a higher repetition rate in the FT arm. As there were few events, it was not possible to obtain the Kaplan–Meier estimates of the median time to self-harm.

FIGURE 17.

Kaplan–Meier plot of time to self-harm by randomised treatment group with 95% CIs. Reproduced from Cottrell et al. 70 © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

Covariates age group (p = 0.0106) and the type of index self-harm episode (p = 0.0333) were found to have a significant effect on the risk of self-harm at the 5% level. The rate of self-harm was lower in older participants (aged 15–17 years) than in those aged 11–14 years (HR 0.7, 95% CI 0.53 to 0.92) and higher in participants whose index episode combined self-injury and poisoning than in those with self-injury only (HR 1.83, 95% CI 1.14 to 2.96). There was also weak evidence (at the 10% level) of an effect for gender and trust, with a higher rate of self-harm in females than in males (HR 1.6, 95% CI 0.98 to 2.61; p = 0.0589) and substantial variation across centres (p = 0.0938). The occurrence of primary outcome events, overall and within each treatment arm, for each covariate is further summarised in Tables 50 and 51.

In addition to the primary analysis model, further analysis excluding centre as a fixed effect (see model checking, Table 52) finds an increased risk of self-harm in participants referred to CAMHS from hospital, and summary statistics suggest a differential treatment effect based on participants’ source of referral (see Table 50). The rate of self-harm among participants referred via hospital was higher in the FT arm than in the TAU arm (37.8% vs. 25.7%), an absolute increase in risk of 12.1% and a relative increase of 47.3% (see Table 46), while there was little difference between groups in the case of participants referred to CAMHS through the community, 22.8% in the FT arm compared with 24.2% in the TAU arm, an absolute decrease in risk of 1.4% and a relative decrease of 5.7%. The presence of a moderating effect by source of referral was investigated in the moderator analysis; however, although there was an indication of an increased hazard in the FT arm in participants referred via hospital, the interaction between treatment and referral source was not found to be statistically significant (see Moderator variables).

Model checking

Graphical and statistical tests of the adequacy of the Cox proportional hazards regression model for treatment and covariates were generally satisfactory and are presented in Appendix 5, along with Kaplan–Meier curves for time to self-harm for each covariate.

There was, however, evidence of non-proportional hazards for a number of trusts, and further Cox proportional hazards regression models were fitted to investigate the effect on the treatment estimate (Table 52). In all three alternative models, the estimated treatment effect matched very closely with that of the primary analysis, confirming the robustness of the treatment effect to variations in modelling and assumptions around trust.

Unlike the primary analysis model, there was evidence of an increased risk of self-harm in participants referred into CAMHS via hospital, with estimates from the frailty model suggesting a HR of 1.39 (95% CI 0.99 to 1.95) with p = 0.0598. A chi-squared test detected a highly significant (p < 0.0001) association between trust and referral source (see Appendix 5); thus, referral source is not identified as a significant effect in the primary model because of a lack of independence and collinearity with trust, with each providing ‘overlapping’ information.

Differences were also found for gender, which was no longer statistically significant when centre was excluded, with only weak evidence of an effect in the model, including hub and the shared frailty model. There was no evidence of an overall association between trust and gender (see Appendix 5); however, in the two trusts with the lowest risk of self-harm the case mix also included the highest proportion of females. Thus, in excluding centre, the reduced self-harm observed in these trusts is no longer explained by trust and is instead incorporated into the other covariates for these participants, including gender, thus diluting the effect of gender overall.

Figure 18 presents the estimated random centre effect in the primary Cox proportional hazards model with shared frailty (random centre effect). The impact of these fluctuations across trusts does not, however, have any bearing on the estimated treatment effect. The model including trust is, therefore, retained for the primary analysis purposes; however, although referral via hospital does not appear to be statistically significant, this is an important factor in the repetition of self-harm and retained in the model.

FIGURE 18.

Estimated random centre effect in the primary Cox proportional hazards model with shared frailty (random centre effect). Estimated random effects > 0 represent higher frailty, that is, shorter time to and higher rate of self-harm, while estimates < 0 represent lower frailty, that is, longer time to and lower rate of self-harm.

Sensitivity analysis for missing presenting details