Cost and quality-of-life impacts of community treatment orders (CTOs) for patients with psychosis: economic evaluation of the OCTET trial

OCTET was a multi-site, non-blinded, parallel-arm randomised controlled trial, with recruitment taking place across England between 2008 and 2011 (for further details on the clinical trial, see [4]). The clinical trial was registered with the International Standard Randomised Controlled Trial Number Register (reference: ISRCTN73110773), ethical approval was given by the Staffordshire National Health Service (NHS) Research Ethics Committee (30/10/2008, reference no. 08/H1204/131), and data were reposted in https://​www.​reshare.​ukdataservice.​ac.​uk/​852414/​. Patient inclusion criteria were (1) aged between 18 and 65 years (i.e. the standard age range for UK adult mental health services), (2) diagnosed with psychosis, (3) currently admitted under a treatment section (Section 3 or 37) of the Mental Health Act, (4) being considered for a CTO by their clinicians (psychiatrist and Approved Mental Health Professional), (5) being able to give written and informed consent, and (6) not already participating in the study. Consenting patients were randomly assigned to be discharged from hospital on CTO (CTO group) or to voluntary status via brief Section 17 leave (non-CTO group), with the latter corresponding to standard care. In England and Wales, patients on CTO are not part of separate programmes, and patients in both arms received case-management follow-up by specialised community mental health teams.

The clinical analysis of OCTET was based on a total sample of 333 patients. Of these, five participants died during the 12-month follow-up period of reasons unrelated to treatment (CTO: n = 3, non-CTO: n = 2) and were excluded from the health economic analysis. Table 1 presents the sociodemographic and clinical characteristics of the remaining 328 patients (CTO: 163, non-CTO: 165) included in the current economic evaluation. Resource use and outcome data were collected based on face-to-face interviews with the participants at baseline, 6 months and 12 months supplemented with information from patient records.

The primary economic analysis was a cost-utility analysis. This was based on patient-rated measures, which were collected in face-to-face interviews with trained research assistants at baseline, 6 and 12 months. Quality-adjusted life years (QALYs) were used to measure outcomes calculated based on utilities measured by the EQ-5D-3L index on a scale where 0 represents death and 1 represents full health [13]. EQ-5D-3L is a standardised, non-disease-specific instrument designed for describing and valuing health-related quality of life (HRQoL), i.e. self-perceived health status [14]. EQ-5D-3L is based on five dimensions (mobility; self-care; usual activities; pain/discomfort; anxiety/depression) [15], and is the currently preferred measure in economic evaluations by the National Institute for Health and Care Excellence [16]. EQ-5D-3L responses were valued based on the UK tariff [17]. Changes in HRQoL over the 12 month-period were reported in QALYs gained in comparison to the baseline values [18]. Quality-of-life changes between the baseline, 6-month and 12-month measurement points were assumed to have occurred linearly. In addition, HRQoL was also assessed based on the EQ-5D Visual Analogue Scale (VAS) values as indicated on a feeling thermometer between 0 (worst imaginable health state) and 100 (best imaginable health state). In a secondary outcome analysis, capability-weighted life years (CWLY) gained were calculated using the OxCAP-MH capability wellbeing measure. The OxCAP-MH is a multi-dimensional, non-preference based, self-reported instrument based on Sen’s capability approach developed for broader quality of life measurement in mental health research [19]. Raw OxCAP-MH scores range from 16 to 80 with standardised scores ranging between 0 and 100 where 0 represents ‘no capabilities’ and 100 represents ‘full capabilities’ [2]. For the CWLYs calculation, standardised OxCAP-MH scores were transferred to a 0–1 scale.

Resource use data collection was based on an amended version of the Client Service Receipt Inventory (CSRI) instrument [20], a widely used and validated resource use measurement instrument designed for use in mental health research. In line with the adopted analytical perspectives, collected resource use information included all hospital and community health and social services, psychiatric medication, productivity losses and the cost of informal family caregiving (see Supplementary Table 1 for detail, [21‐29]). Additionally, data on the number of manager hearings and tribunals were included from the trial Case Report Forms (CRFs) as a measure of use of legal resources. All resource and cost data were extracted from patient records by trained research assistants. Due to the lack of relevant tariffs, costs associated with these legal services were not included in the cost-effectiveness analysis.

UK national-level unit costs were used to value all resource use items. To match the final year of trial follow-up, all unit costs used in the analysis refer to year 2012/13 and are expressed in British pounds (£). These together with the sources of information are reported in Supplementary Table 1. For medication costs, daily dose information was multiplied with the average (proprietary and non-proprietary) unit price for each compound reported in the British National Formulary [21]. The human capital approach was adopted to estimate lost productivity costs [30]. For study participants in employment, absent work days were multiplied by the average daily UK national salary [31, 32]. Informal care was valued based on average UK hourly salary multiplied by the number of hours family and friends spent on supporting participants as a result of their illness. Costs were assessed over 12 months, and so no discount rate was necessary.

In line with current NICE guidance [16], the main analysis took the perspective of the health and social care system. Additional analyses were conducted from a broader societal perspective including lost productivity and informal care.

Self-reported resource use and outcome (EQ-5D-3L) data were missing at 12 months for 42% and 58% of the participants in the CTO group and 45% and 55% of the participants in the non-CTO group, respectively (Table 2, Supplementary Table 3). Missing data were supplemented from patient records or dealt with using multiple imputation (chained predictive mean matching). In the latter case, missing values were replaced with values predicted based on randomisation group, age, gender, main clinical diagnosis, illness duration and length of inpatient stay as covariates [33]. The number of imputations sets was 30 for costs and EQ-5D and 50 for the OxCAP-MH index to match the percentage of incomplete cases [34]. Results of the health economic analysis are presented separately for the full dataset (n = 328) as main analysis and the complete cases datasets (health and social care perspective: n = 121; broader societal perspective: n = 102) as secondary analyses. Further sensitivity analyses were carried out to assess the impact of the original linear assumption on quality-of-life changes.

Results are reported as means with standard deviations (SD) or as mean differences with 95% confidence intervals. A regression framework was used for comparing the differences in mean costs and effects with p < 5% considered as statistically significant. Non-parametric bootstrapping [35] of the cost and effectiveness data was applied to generate a joint distribution of the mean incremental costs and effects and to calculate the 95% confidence intervals of the incremental cost-effectiveness ratio (ICER). Uncertainty around the main cost-effectiveness estimates was represented by cost-effectiveness acceptability curves (CEACs) based on the net benefit approach [36, 37]. CEACs indicate the probability that each option is cost-effective across a range of different maximum costs per QALY gained ceiling ratios that a decision-maker might be willing to pay for an additional unit of improvement in outcomes.

All analyses were conducted according to the intention-to-treat principle and carried out in Microsoft^® Excel and Stata^® (StataCorp, College Station, TX, USA).

The mean values of the EQ-5D-3L utility, EQ-5D VAS and the OxCAP-MH capability index at baseline, 6 months and 12 months are summarised in Table 2. Table 2 also reports changes in EQ-5D levels from baseline in the form of QALYs gained and changes in OxCAP-MH levels from baseline in the form of CWLYs gained during the 12-month follow-up. No significant difference could be detected between the CTO and non-CTO groups with regards to QALYs gained or CWLYs gained over the 12 months (Table 2).

Mean observed resource use information on health and social care service utilisation is presented in Supplementary Table 2. Regarding employment, one participant in the CTO group and five participants in the non-CTO group had any period of employment/self-employment during the 12-month follow-ups with only one of these participants reporting some absence due to sick leave. There were significantly more manager hearings and tribunals in the CTO group in comparison with the non-CTO group (Table 3).

Table 4 presents the mean health and social care costs and productivity and informal care costs per participant over the 12-month follow-up based on all cases with multiple imputation for missing data. Mean per patient total health and social care costs were estimated at £35,959 (SD: £44,886) in the CTO group and £36,003 (SD: £41,406) in the non-CTO group. With more than 85% of annual health and social care costs, mental health hospitalisations were the biggest cost driver. Results of the secondary cost analyses based on complete cases are presented in Supplementary Table 3. No significant differences between the CTO and non-CTO group could be found in any of the included health and social care cost categories, either in the main analysis or in the complete case analyses.

Participants in the CTO group had significantly higher mean in terms of the cost of informal family caregiving than patients in the non-CTO group (£6138 vs. £2993) (Table 4).

There were no significant changes over time in any of the outcome measures, nor could a significant effect difference be detected between the CTO and the non-CTO groups in terms of QALYs gained (full imputed dataset: 0.006; 95% CI − 0.04 to 0.05; complete cases: 0.003, 95% CI − 0.07 to 0.07) or CWLYs gained (full imputed dataset: 0.008, 95% CI − 0.01 to 0.03; complete cases: 0.013, 95% CI − 0.03 to 0.06). Differences in mean total health and social care costs (full imputed dataset: £− 44, 95% CI − 9393 to 9305; complete cases: £5388, 95% CI − 7107 to 17,883) or mean total societal costs (full imputed dataset: £3102, 95% CI − 6378 to 12,582; complete cases: £7067, 95% CI − 7219 to 21,353) between the groups were also non-significant.

Figure 1 shows the scatterplots of the bootstrapped incremental cost and effectiveness pairs between the CTO and non-CTO groups for the different analytical scenarios. As the points in the scatterplot fall within all four quadrants of the cost-effectiveness plane, major uncertainties remain around the cost-effectiveness of CTO vs. non-CTO, and no clear conclusion can be drawn in this respect. Supplementary Fig. 1 illustrates the uncertainty around the ICER in the form of CEACs. These clearly demonstrate that in any of the analytical scenarios the probability of CTO being cost-effective remains at around 50% or below at the currently considered maximum UK cost-effectiveness threshold value of £20,000–£30,000/QALY [16].

In the sensitivity analysis, changing the assumption that quality-of-life changes between follow-up points occurred linearly did not change any of the above results or conclusions. The QALY differences between CTO and non-CTO groups for the full imputed dataset were 0.008 (95% CI − 0.10, 0.08; p = 0.87) and 0.017 (95% CI − 0.03, 0.07; p = 0.49), when assuming changes occurred at the beginning or at the end, respectively.