Randomised controlled trial
The results of the RCT have been published.55 To date, there has been a dearth of research developing and evaluating mental health interventions for prison leavers.27 Trials within prison settings are particularly challenging to undertake because of a range of factors, such as organisational difficulties, maintaining masking, poor follow-up rates and selecting appropriate outcomes measures.1,17,28–30
Design
The individually randomised superiority trial of the Engager intervention was set in three prisons (south-west of England, n = 2; north-west of England, n = 1) and nearby communities between 2016 and 2019.
To detect a change of 3.5 units for the primary outcome measure (i.e. CORE-OM43) with 90% power and a significance threshold of 0.05, assuming a standard deviation (SD) of 7.556 and allowing for 30% attrition, 280 participants were recruited and randomised [1 : 1 allocation; stratified by site (i.e. Devon or Manchester)] to either the Engager intervention plus usual care (n = 140) or usual care only (n = 140).
Intervention
The Engager intervention was a manualised person-centred intervention that aimed to address mental health needs, as well as to support wider issues, such as accommodation, education, social relationships and money management. Experienced support workers and a supervisor with experience of psychological therapy delivered the Engager intervention. A mentalisation-informed approach underpinned all elements of the intervention. At the pre-release stage, the practitioner and participant developed a shared understanding of the participant’s needs and goals, recognising the links between emotion, thinking, behaviour and social outcomes. A goal attainment plan was developed and followed, including liaison with relevant agencies and the participant’s social networks.
Engagement was maintained throughout the pre-release period and, when required, all-day support was given on the release day. Following release, the practitioner provided support for the participant to re-enter the community and engage with services. The practitioner continued to work with the participant and any relevant organisations to help them achieve their goals, while encouraging the participant to take responsibility for self-care. The practitioner also prepared the participant for the end of the intervention, while liaising with relevant community organisations regarding continuity of care.
Participants
Individuals recruited into the trial needed to be released to defined geographical areas and were required to score ≥ 10 points on the Patient Health Questionnaire-949 to indicate depression, ≥ 10 points on the General Anxiety Disorder-750 to indicate anxiety or score ≥ 3 points on a bespoke questionnaire to indicate likelihood of difficulties on release. Those individuals with psychosis under the care of prison in reach teams or part of the personality disorder service were excluded. All participants were assessed at six time points, that is, at baseline (before randomisation), during the week before release from prison, and then at 1, 3, 6 and 12 months post release from prison.
A total of 3102 prisoners received assessment for eligibility. Of these prisoners, 673 were identified as potentially eligible for the trial and approached for consent. One hundred and eighty-seven of the 673 prisoners declined involvement and a further 90 prisoners were found not to be eligible or were excluded for other reasons, leaving 396 prisoners to be screened for common mental health problems. Of these prisoners, 78% were found to be eligible and, therefore, selected to take part. After a small number of withdrawals, the 280 prisoners who remained were randomised into two groups of 140. Figure 4 depicts the CONSORT (Consolidated Standards of Reporting Trials) diagram.
FIGURE 4 A CONSORT flow diagram. Reproduced with permission from Byng et al.55 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)
Analysis
Demographic data and participant characteristics were reported descriptively at baseline, using mean and SD, with median and minimum and maximum, as appropriate, for continuous measures, and percentages for categorical measures.
For the primary analyses of the outcome variables, the intention-to-treat principle was used, with participants being analysed according to their randomised allocation irrespective of treatment actually received. Only participants with observed data were included for the primary analyses. Primary analyses used linear regression models for continuous outcomes and logistic models for binary outcomes.
Sensitivity analyses were performed for comparison with the primary analyses, that is, per-protocol analysis, complier-average causal effect analysis and inclusion of only those participants whose follow-up assessments took place within the specified ‘window’. The definition of per protocol for the Engager intervention plus usual-care group was receipt of at least two contacts prior to initial release and at least eight contacts following initial release. As a secondary analysis, mixed-effects models (random effect on participant intercept) were performed including participants with data available at any of the four follow-up time points. Results of all inferential analyses were reported as between-group differences with 95% confidence intervals (CIs). Global p-values were also provided with regard to categorical explanatory variables. The threshold for determining significant effects was set at p < 0.05. No adjustment of p-values was made in the light of multiple testing.
All analyses were adjusted for baseline values of continuous outcomes and study site. Baseline participant characteristics were to be compared by treatment group, with any covariates found to be unbalanced included in the analyses if thought to be predictive of outcome.
For the CORE-OM at 6- and 12-month follow-up, we investigated interactions (i.e. moderation effects) between intervention and specific participant characteristics, including study site, personality disorder (based on the Standardised Assessment of Personality – Abbreviated Scale), previous trauma, pre-prison accommodation and alcohol/substance abuse. Further sensitivity analyses were performed to address missing data, based on the assumption that missing data were missing at random. The level of missingness for the CORE-OM was evaluated, with 55% of participants found to have missing data at 12 months. Sensitivity analyses were performed using combined observed and imputed data.
Results
There were some imbalances between the two groups at baseline. The Engager intervention group had a higher proportion of people with unstable accommodation [n = 76 (54%) vs. n = 58 (41%)], a higher proportion of people who were unemployed [n = 104 (75%) vs. n = 85 (61%)] and a higher proportion of people with physical health problems [n = 64 (45%) vs. n = 48 (34%)]. A higher proportion of people in the usual-care group had experienced relational trauma [n = 89 (64%) vs. n = 74 (53%)]. At 6 months, 184 (66%) participants were followed up, with 45% of participants followed up at 12 months.
There were between-group imbalances in participant characteristics at baseline for type of accommodation (people in the usual-care group had more stable accommodation before prison), employment (people in the usual-care group were more likely to be employed), previous trauma (people in the usual-care group had experienced more relational trauma) and physical health problems (people in the usual-care group had fewer problems). Therefore, these characteristics were included as covariates in all analyses.
Of 140 participants allocated to the intervention, 92% received at least one session in prison [mean 5.7 (SD 3.9)] and 108 (77%) participants received at least one post-release community session [mean 12.9 (SD 11.6)].
There were no significant differences in mean CORE-OM (primary outcome) scores between the groups at 6 months [Engager intervention group mean 12.6 (SD 6.9), n = 92; usual-care group mean 11.9 (SD = 7.7), n = 90; between-group difference 1.1 (95% CI –1.1 to 3.2); p-value 0.325]. Figure 5 depicts the change in CORE-OM score over the four time points.
FIGURE 5 Graph of CORE-OM score over time. Reproduced with permission from Byng et al.55 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)
There were a small number of statistically significant differences between the two groups on the secondary outcomes [including CAN-FOR, Intermediate Outcomes Measurement Instrument, EuroQol-5 Dimensions, five-level version (EQ-5D-5L) and ICEpop CAPability measure for Adults (ICECAP-A)] for a variety of sensitivity analyses (i.e. complier-average causal effect and per protocol), but this is most likely due to the effects of multiple testing. At the time of writing, it had still not been possible to obtain Police National Computer data on reoffending.
Further exploratory analysis
After our neutral finding was revealed and the baseline imbalances were identified, we carried out some further exploratory analyses to rule out or put forward any contributing factors or possible explanations for the neutral finding. The details of this exploratory analysis are found in Report Supplementary Material 9. We found imbalances in CORE-OM scores at baseline, with the Engager intervention group scoring as less mentally unwell (mean 15.2 vs. 16.9). It is possible that the people in the usual-care group were more willing to acknowledge their common mental health problems or may have struggled with being in prison but may have had less social disadvantage on release into the community. Conversely, people in the Engager intervention group may be less likely to report their difficulties in the prison environment, as they may not wish to show their vulnerabilities.
We identified that the mean scores for the CORE-OM dropped differentially over time, depending on which trial arm participants were in and whether or not the participants were serving a new prison sentence at time of follow-up.
Health economic analyses
We conducted a trial-based economic evaluation to identify the mean incremental cost per quality-adjusted life-year (QALY) gained with the Engager intervention compared with current practice. Report Supplementary Material 10 describes in detail the health economics modelling, and the results of the health economic analyses are published.57
The cost of the Engager intervention included the cost of training and supervision based on practitioner-reported activities. As a conservative estimate, the cost per participant of training and supervision was calculated as the total cost of training and supervision for the Engager trial divided by the number of participants in the intervention arm of the trial.
Unit costs for accommodation-, education-, training-, employment-, money-, relationship- and criminal justice-related service use are reported in Hunter et al.57 table 1. All costs are reported in 2017/18 GBP (i.e. the most recent year costing data were available), with costs for earlier years adjusted to 2017/18 values.
The trial primary outcome CORE-OM was converted to the Clinical Outcomes in Routine Evaluation – 6D (CORE-6D) utility to calculate QALYs. QALYs were calculated as the area under the curve using the CORE-6D responses at baseline and at 1, 3, 6 and 12 months post release. QALYs were also calculated and reported in a similar manner using responses to the EQ-5D-5L at baseline and at 3, 6 and 12 months post release and (1) the mapping algorithm and (2) the EQ-5D-5L value set. Years of full capability (YFCs) (equivalent) were calculated for the Engager intervention compared with usual care using patient-level responses to the ICECAP-A at baseline and at 3, 6 and 12 months post release.
We assumed that data missing at follow-up were missing at random. No predictors of ‘missingness’ were found. Costs, utility scores and the ICECAP-A tariff were imputed for the number of 30 data sets using chained equations [i.e. multiple imputation using chained equations (MICE)] and predictive mean matching.
For the incremental cost-effectiveness ratio, we used regression to account for the correlation between costs and outcomes to calculate the incremental mean cost per QALY or per YFC gained of the Engager intervention compared with usual care. The primary analysis was calculated using the multiple imputation data set and was bootstrapped. The incremental mean cost per QALY and per YFC were used to calculate cost-effectiveness acceptability curves and cost-effectiveness planes. A range of sensitivity analyses and a cost–consequences analysis were also conducted.
The total cost of training and supervision for the duration of the Engager trial was £59,303. If the total cost of the training and supervision is divided by the 140 participants randomised to the intervention, then the training and supervision cost was £424 per participant. The total average cost per participant of delivering all intervention sessions (i.e. prison, meeting at the gate and community) was £467 (SD £475). When the cost of training and supervision is added, then this results in an average cost per participant in the Engager arm of £891.
Descriptive statistics for CORE-6D, EQ-5D-5L (cross-walk and time trade-off tariff) and ICECAP-A tariff are reported.57 There were no significant differences between trial arms.
The bias-corrected bootstrapped difference in health-care costs, adjusting for baseline, of the Engager intervention compared with usual care is £751 (95% CI –£1593 to £3096). When the cost of the Engager intervention is added (including the cost of training and supervision), the difference in costs is £1795 (95% CI –£575 to £4164). The cost difference for CJS costs including the cost of the intervention is £8166 (95% CI £1692 to £14,640). If productivity gains/losses are included, then the cost difference is £5344 (95% CI –£1228 to £11,916).
For the primary economic evaluation (i.e. a health and social care cost perspective, QALYs calculated from the CORE-6D and using MICE for missing data), the mean cost difference was £2133 (95% of iterations between £997 to £3374) and mean QALY was a difference of –0.017 (95% of iterations between –0.042 to 0.007) and, therefore, the Engager intervention is less effective and more costly than usual care. The cost-effectiveness plane is shown in Figure 6. There is a 0% probability that the intervention is cost-effective at the £20,000–30,000 willingness-to-pay threshold for a QALY gained (i.e. the standard willingness-to-pay threshold for recommending treatments for the NHS).
FIGURE 6 Cost-effectiveness plane of the Engager intervention compared with usual care from a health and social care cost perspective over 12 months. (Figure not reproduced here; see the original article.)
The Engager intervention had a very low probability of being cost-effective and this was the case across all secondary and sensitivity analyses. The Engager intervention group cost significantly more from a health service perspective, and had higher CJS costs. On the other hand, the Engager intervention group also showed greater productivity gains, although caution should be taken in interpreting these results, particularly given the number of analyses run. Very few cost-effectiveness analyses are carried out in CJS settings, with self-reported outcomes in this group being particularly hard to collect because of this population’s transient nature. This trial demonstrates the complexity of economic evaluations in this area and represents a significant contribution to the health economic evidence base for this population group.