Evaluation of the Cost-effectiveness of Services for Schizophrenia in the UK Across the Entire Care Pathway in a Single Whole-Disease Model

Key Points Question Which interventions are cost-effective for the prevention and treatment of schizophrenia? Findings In this decision analytical model using a simulated cohort of 200 000 individuals, the following interventions were found to be cost-effective: practice as usual plus cognitive behavioral therapy for individuals at clinical high risk of psychosis; a mix of hospital admission and crisis resolution and home treatment team for individuals with acute psychosis; receipt of amisulpride, risperidone, or olanzapine combined with family intervention for individuals with first-episode psychosis; and receipt of clozapine for individuals with treatment-resistant schizophrenia. Meaning The results of this study suggest that cost savings and/or additional quality-adjusted life years may be gained by replacing current interventions with more cost-effective interventions.


eAppendix 1. Key Consequences of Interventions Considered in the Model
The key impacts of interventions considered in the model are summarised in eTable 1. It should be noted that not all impacts of interventions were included; common reasons for exclusion are: (1) not expected to affect results; (2) lack of evidence.

Module C -Psychosis pathway
Patients can enter the psychosis pathway through different routes depending on their disease status, and the acceptability of and access to interventions:  Patients who are assessed as acute FEP move to Part 10 'Manage acute psychosis', where they will receive antipsychotic medications and other acute treatments without delay. Patients at Part 10 can experience different events: due for the next family intervention session (if the patients are receiving family intervention), receive a formal diagnosis, and achieve remission. Depending on which of the above events happen first, patients move to different parts. Patients who are due for the next family intervention session move to Part 13 'Psychosis on family intervention'. Once a family intervention session is finished, patients return to Part 10 'Manage acute psychosis'. Patients who receive a formal diagnosis move to Part 15 'Diagnosis': if the diagnosis result is schizophrenia, patients return to Part 10 'Manage acute psychosis' to continue their current treatments; if the diagnosis result is not schizophrenia, patients move to Part 3 'Out of scope'. Patients who achieve remission move to Part 12 'Non-acute psychosis on antipsychotics' to receive treatment for seTable patients.  Patients who are assessed as seTable FEP, and who have access to and agree to receive interventions move to Part 11 'FEP wait for treatment'. Those patients are assumed to experience a delay before receiving treatment. When the waiting time finishes, patients who accept antipsychotic medication with or without family intervention move to Part 12 'Non-acute psychosis on antipsychotic medication', while patients who only accept family intervention move to Part 13 'Psychosis on family intervention'. Patients at Part 12 'Non-acute psychosis on antipsychotic medication' can experience different events: due for the next family intervention session, receive a formal diagnosis, and discontinue antipsychotic due to inefficacy, intolerability or non-adherence. Depending on which of the above events happen first, patients move to different parts of the model as follows:  Patients who are due for the next family session move to Part 13 'Psychosis on family intervention'. Once a family intervention session is finished, patients return to Part 12 'Non-acute psychosis on antipsychotic medication'.  Patients who receive a formal diagnosis move to Part 15 'Diagnosis': if the diagnosis result is schizophrenia, patients return to Part 11 'Non-acute psychosis on antipsychotics' to continue their current treatments; if the diagnosis result is not schizophrenia, patients move to Part 3 'Out of scope'.  Patients who discontinue their antipsychotic medication due to inefficacy are assumed to experience a relapse and move to Part 10 'Manage acute psychosis'.  Patients who discontinue their antipsychotic medication due to intolerability are assumed to be switched to another antipsychotic medication and stay at Part 12 'Nonacute psychosis on antipsychotic medication'.  Patients who discontinue their antipsychotics due to non-adherence move to Part 14 'Psychosis stop treatment'.  Patients who are assessed as seTable FEP and who choose not to have antipsychotic medication or family intervention due to refusal or local unavailability move to Part 14 'Psychosis stop treatment'. Those patients are assumed to receive no intervention and stay at the current part until they experience a relapse, then they will move to Part 10 'Manage acute psychosis'.
Patients at any part of the psychosis pathway are at risk of death. Patients who die move to Module D 'Out of scope and death pathway'.

Module D -Out of scope and death pathway
Patients who are not at risk of psychosis or receive a non-schizophrenia diagnosis move to Part 3 'Out of scope'. For the sake of simplicity, it is assumed that patients on Part 3 cannot go on to develop psychosis. Patients who die move to Part 16 'Death', where their life-time costs and QALYs are calculated separately by aggregating adding together the cost and QALYs that each patient has accrued at each part in the model that they passed through prior to death. Patients' costs at each part in the model were determined by the interventions that they received, including cost of providing the interventions, and cost of treating adverse events (weight gain, EPS, diabetes and neutropenia). Patients' utility at different parts in the model were determined based on their age, gender, disease status (not at risk of psychosis, CHR or psychosis) and presence of adverse events (weight gain, EPS and diabetes). Patients' life expectancies were determined based on their age, gender, disease status and intervention received (e.g. whether clozapine was used for people with TRS). eAppendix 3. List of Assumptions and Simplifications of the Model 3.1 Key assumptions of the model  A DUP longer than 6 months is associated with poor prognosis, compared with a DUP less than 6 months  Patients can achieve clinical recovery without antipsychotic medications  Psychotic patients who have achieved clinical recovery while on antipsychotic medication will adhere to their current antipsychotic medication 3.2 Key simplifications of the model  Impacts of treatment for co-existing conditions was not explicitly modelled  The mental health status for people with psychosis or schizophrenia were grouped into two categories: relapse and remission  The sensitivity and specificity of a specialist assessment is assumed to be 100%  Patients who are assessed as not at risk of psychosis during mental health specialist assessment cannot go on to develop psychosis  CHRs who have access to and accept CBT will finish a full course of CBT (16 sessions), unless they die or convert to psychosis before that  Recovered CHRs cannot relapse or go on to develop psychosis  Services for patients with psychosis or schizophrenia were grouped into two categories: services for acute patients and services for non-acute patients  Reasons for discontinuing antipsychotic medication were grouped into three mutually exclusive categories: inefficacy, intolerance and non-adherence  Dosing effects of antipsychotic medication were not modelled  Patients with psychosis or schizophrenia who achieved clinical recovery are no longer at risk of relapse Take up of antipsychotic medication for patients with relapsed psychosis/ schizophrenia It was assumed that all patients with relapsed psychosis will be prescribed with antipsychotic medication and will take the medication (although they may discontinue their antipsychotic medication due to non-adherence)  Check the behaviour of individual workcentres. For example, the values of relevant attributes (e.g. patient's age and accumulated cost) were recorded when the patient entered and left a particular workcentre and checked against the patient's disease status, prognosis and treatment plans. For workcentres which involve complex calculations, the values of relevant attributes were recalculated in Excel and compared with the values produced by SIMUL8.  Check the behaviour of multiple workcentres. For example, the interventions that patients received were recorded and compared against patient's attributes when entering the model, such as patient's eligibility for interventions, acceptability of interventions and availability of interventions, in order to ensure the interventions that they received were compatible with their attributes. To illustrate, for psychotic patients on antipsychotic medication, their history of antipsychotic medication use was recorded, including types of antipsychotic medication tried, duration of use, reasons for discontinuation, and whether the patients has developed certain adverse events while on the antipsychotic medication. This information was then compared against the algorithm for switching antipsychotic medication designed for the WDM to ensure any switch made sense given the data available.  Check the behaviour of the entire model. For example, a unique numerical code was assigned to each workcentre in the model, allowing the complete pathway for individual patients to be recorded by saving the numerical codes of those workcentres that the patient passed through. The complete pathway was checked for the first 20 patients at CHR, the first 20 patients with psychosis and the first 20 patients who are not at risk.
A detailed record of all white box tests conducted is reported in eTable 8. Any errors or inconsistencies identified were checked and fixed.
© 2020 Jin H et al. JAMA Network Open.  A break-down of the lifetime LYGs and QALYs by health states is reported in eTable 11. Within each part of the pathway, the order of magnitude of LYGs appear intuitively sensible, for example:  Within the CHR pathway, the largest contributor comes from recovered CHR. This is consistent with previous findings that CHR is a transitional period which normally lasts less than 2 years and the majority of patients at CHR will not convert to psychosis 5 The results suggest that the largest contributors to overall undiscounted cost are interventions for people with psychosis/ schizophrenia (98.8%), followed by end of life interventions (0.9%), mental health specialist assessments (0.1%) and the CHR pathway (0.1%). The reason why the cost of interventions for CHR is less than the cost of mental health specialist assessments and end of life interventions is likely to be that all patients in the model have used mental health specialist assessments (at least once) and end of life interventions, but only one third of those patients were assessed as CHR and thus eligible for interventions of CHR. And of those people at CHR, only 71.1% of them have access to and accept monitoring and only 20.9% have access to and accept CBT. A break-down of the lifetime costs are reported in eTable 13 according to the broad groups of resources used within the service. Due to a lack of evidence about the lifetime cost of managing people at CHR or those people with a diagnosis of psychosis/schizophrenia, it is impossible to compare my results with published literature. However, the magnitude of costs within each part of the pathway appear to be intuitively sensible, according to the stakeholders recruited for this PhD. For example, within the psychosis pathway, the largest contributor to the total cost is treatment for acute psychosis (43.8%), followed by treatment for non-acute psychosis (42.8%). This is because although the daily cost of treating acute patients is much higher compared to the daily cost of treating stable patients, the average time that psychotic patients spend in acute status is only 10% of the time that patients spend in non-acute status.  44 Our rapid review identified one recently published clinical trial carried out in the Netherlands which compared the clinical and economic outcomes for people at CHR who received routine care only with people who received routine care plus CBT. 45 The economic findings were similar to the WDM results presented: over a 4-year follow-up period, use of CBT was associated with lower costs (mean difference=-US $5,777 per person at CHR), and a marginal non-significant difference in QALY gains (mean difference=0.12, 95% CI: 0.00-0.25) compared with routine care.

Topic B -Interventions for people with acute psychosis
No existing models were identified for Topic B. 44 Our rapid review identified one trial-based economic evaluation conducted by McCrone and colleagues in London which shows similar findings to our analysis: use of CRHT results in lower costs compared with hospital admission alone. 37 McCrone et al. did not report the impact of receiving alternative services on patients' HRQoL or mortality.

Topic C -First-line oral antipsychotic for people with FEP
Fifty-three models assessing antipsychotics were identified, 44 of which only one specifically assessed first-line antipsychotics for preventing relapse for people with early schizophrenia -this is the model developed by the NICE schizophrenia GDG. 16 Both the results of the NICE model and the schizophrenia WDM found that no antipsychotic can be considered to be clearly more cost-effective compared with the other options, although the results from both models suggest that olanzapine is one of the most cost-effective antipsychotics. However, the results of the NICE model and the schizophrenia WDM differ for amisulpride. The NICE model suggests amisulpride was the least cost-effective option whilst the schizophrenia WDM found it to be the most costeffective option. This is likely to be due to differences in the input data used. The clinical effectiveness data used in the NICE model was obtained from a network meta-analysis (NMA) conducted by the NICE schizophrenia GDG. 16 According to the results of this NMA, amisulpride is associated with the second highest probability of relapse (second only to haloperidol). However, the studies included in the NICE NMA were obtained from a systematic review conducted in 2008 and since then, two further NMAs which included additional trials have been published, and both analyses showed amisulpride to be associated with one of the lowest probabilities of all-cause discontinuation and relapse rate. 17,46 7.4 Topic D -Family intervention for people with FEP Three models were identified which compared the cost-effectiveness of antipsychotic alone versus antipsychotic plus family intervention and CBT or plus family intervention alone. [47][48][49] None of these models assessed the option of using family intervention alone. All three models suggested that antipsychotic medication plus family intervention is cost-effective compared with antipsychotic alone, which is consistent with the results of WDM.

Topic E -First-line oral antipsychotics for people with TRS
Four models assessing antipsychotics for people with TRS were identified. [50][51][52][53] Of these four models, one did not include clozapine, therefore its findings are not comparable to our study. 50 The other three models compared clozapine with chlorpromazine and haloperidol, and all of them suggested that clozapine dominated chlorpromazine and haloperidol. [51][52][53] This is consistent with the findings of the schizophrenia WDM, which indicated that clozapine is the most cost-effective antipsychotic for patients with TRS.