Analytic approach

The current analysis was undertaken within a broader research program aiming to use economic evidence to inform mental health service delivery within Australia. The program adapted the technical methods of the assessing cost-effectiveness (ACE) Prevention study to improve comparability across cost-effectiveness evaluations [Reference Carter, Vos, Moodie, Haby, Magnus and Mihalopoulos6–Reference Voigt and King8].

A cost-utility framework was used whereby outcomes were expressed as disability-adjusted life years (DALYs) averted. DALYs are a composite measure of premature death (years of life lost, YLL) and years of life lived with a disability (YLD) [Reference Tan-Torres Edejer, Baltussen, Adam, Hutubessy, Acharya and Evans9] used in economic evaluations as a measure of population health loss, to assess the incremental benefit of one intervention over another, and to optimize the mix of services provided [10].

This economic evaluation was conducted using an Australian health-system perspective, with a focus on government (Commonwealth and states/territories) as a third-party payer. The evaluation excluded costs to the private sector, nongovernment organizations, out of pocket costs to patients and carers, as rTMS is not currently part of routine clinical practice, and limited data were available to conduct a full health-sector perspective analysis.

While there is no explicit threshold adopted in Australia for decision making, a cost-effectiveness threshold of $50,000/ DALY averted is adopted in this evaluation [Reference Carter, Vos, Moodie, Haby, Magnus and Mihalopoulos6].

Population

The target population for receiving rTMS in clinical practice is Australian adults aged 25–65 years with schizophrenia treated in a hospital inpatient or community setting who continue to experience symptoms of disease despite treatment with antipsychotic medicines at an optimized dose. rTMS has been evaluated as add-on treatment to standard care comprised of one or more antipsychotic medicines and a range of psychosocial interventions, therefore, standard care is considered the appropriate comparator.

Table 1 summarizes the input parameters and distributions for uncertainty assumptions. The eligible patient population entering the economic model was estimated by taking the whole Australian population (by age and gender) in 2019 from the Australian Bureau of Statistics [11] and applying the prevalence of schizophrenia in Australia from the global burden of disease (GBD) [Reference Morgan, Waterreus, Jablensky, Mackinnon, McGrath and Carr12]. Of these patients, an estimated 37.5% will experience hallucinations in a 12-month period [Reference Németh, Fasseeh, Molnár, Bitter, Horváth and Kóczián13], and 25% of those patients were estimated to have hallucinations refractory to medications [Reference Shergill, Murray and McGuire4].

Table 1.

Input parameters and uncertainty ranges.

^a Refer to Table S1 (Supplementary Materials) for age and sex specific mortality rates.

^b Refer to Table S2 (Supplementary Materials) for age and sex specific schizophrenia prevalence.

^c Disability weights quantify societal preferences for different health states. They range from 0 representing no disability to 1 representing death. This is an inverse scale to utility weights used to calculate Quality Adjusted Life Years (QALYs).

Model structure

A Markov cohort model was developed in Excel to undertake this cost-utility analysis (CUA). Markov cohort models are the most commonly used economic models in schizophrenia [Reference Heeg, Damen, Buskens, Caleo, de Charro and van Hout14] and synthesize a range of clinical and epidemiological inputs based on the probability of a population of people moving between health states [15]. The health states in the economic model were aligned with the 2019 GBD health state descriptors, with the corresponding disability weights applied to patients in each modeled health state to calculate total DALYs averted with treatment compared to standard care (Figure 1). These health states are a simplistic reflection of the clinical disease course, whereby patients have periods of relative stability with intermittent relapses of acute episodes [15]. As treatment with rTMS is most appropriate for patients who have continuing hallucinations but otherwise good disease insight, all eligible patients enter the model in the residual health state. In each 3-monthly cycle, patients can remain in the residual health state, transition to the acute health state, or die. Similarly, patients in the acute health state can recover and return to the residual health state, remain in the acute health state, or die. Patients in the acute health state stop receiving rTMS treatment since it is assumed that these patients would lack the required disease insight and stability in mental state to be able to tolerate and adhere to the treatment course. The 3-month Markov cycles were based on the longest follow-up data on rTMS in schizophrenia [Reference Kennedy, Lee and Frangou5]. The base case analysis covers a 12-month time horizon with no discounting applied. The sensitivity analyses with extended time horizons beyond 12 months use an annual discount rate of 3% for costs and outcomes.

Figure 1.

State transition diagram used to estimate the health outcomes associated with rTMS for people with schizophrenia.

Intervention description and effectiveness

A range of treatment schedules can be used for rTMS including treatment once or twice daily for 4–30 sessions. In clinical practice, it is unlikely that patients will only receive an intensive treatment course followed by no treatment, particularly if the patient is responding. Instead, it is anticipated that most patients commence a tapering treatment course, for example, once weekly sessions for the first 6 months, followed by once fortnightly sessions. We therefore used a regimen of once daily rTMS for 10 days since this was the most commonly reported regimen from randomized controlled trials of rTMS [Reference Kennedy, Lee and Frangou5]. This was followed by a tapering treatment course in the base case of the current analysis (Table 2). In the absence of clinical practice guidelines on longer term rTMS treatment in schizophrenia, the tapering regimen was based on expert opinion from a psychiatrist experienced in the use of brain stimulation in the treatment of psychiatric disorders including schizophrenia.

Table 2.

Repetitive transcranial magnetic stimulation treatment schedule and costs applied in the model.

To determine the effectiveness of rTMS compared to standard care in the treatment of hallucinations in schizophrenia, a recent high quality meta-analysis of rTMS was identified [Reference Kennedy, Lee and Frangou5]. This meta-analysis comprised studies evaluating rTMS as an add-on to treatment as usual (TAU) and compared to sham stimulation added to TAU. The main outcome measured was the composite hallucination score, and a statistically significant treatment effect for reduction in hallucinations of −0.51 (p = 0.0001), favoring rTMS over sham was observed.

Resource use and costs

The cost of $160 per rTMS treatment was based on an application to the Medical Services Advisory Committee (MSAC) approved in 2019 to list rTMS on the Medicare Benefits Schedule (MBS) for the treatment of major depression [16]. This is the same as the current private cost per rTMS treatment session [Reference Andrews17].

The resource use for standard care of people with schizophrenia was identified by model health state and measured from descriptions of optimal treatment recommendations for schizophrenia from Andrews et al. [Reference Morgan, Waterreus, Jablensky, Mackinnon, McGrath and Carr18] based on an expert panel’s opinions (Table 3). The panel estimated the proportions likely to receive specific treatments (i.e., hospital admission) and the number of services by remission status. Length of inpatient stay was derived from the 2010 Survey of High Impact Psychosis [19]. Unit costs for medicines were sourced from the Pharmaceutical Benefits Scheme (PBS) [20], unit costs for GP consultations, psychiatrist consultations, CBT and family education were sourced from the MBS [21], and unit costs for community mental health and inpatient admissions were sourced from the second Australian National Survey of Psychosis [Reference Neil, Carr, Mihalopoulos, Mackinnon and Morgan2]. All unit costs were inflated to 2021 prices using the health price index [22]. Health state costs were calculated as a weighted average based on the proportion of the population using each service (Table 3) and applied to both rTMS and standard care arms of the model.

Table 3.

Resource use and cost of standard care by model health state.

Health outcomes

The 2019 GBD study [Reference Green and Higgins23] describes two severity levels for schizophrenia: acute with a disability weight of 0.778 (95%CI 0.606–0.9) and residual with a disability weight of 0.588 (95%CI 0.411–0.754).

While rTMS may impact positive and negative symptoms of schizophrenia, a conservative assumption was made that there would only be a benefit on hallucinations (positive symptoms).

Transition probabilities

The transition probability of moving from the residual to acute health state varies by whether the cohort receives rTMS or standard care. This required the effect sizes to be converted into relative risk using the Cochrane conversion method [Reference Ceraso, Lin, Schneider-Thoma, Siafis, Tardy and Komossa24]. This first required an estimation of the background risk of relapse, which was not included as part of the reporting in Kennedy et al. [Reference Kennedy, Lee and Frangou5]. Instead, a background risk of relapse of 230 per 1,000 was incorporated into the model from a Cochrane review of maintenance treatment with antipsychotic drugs [25]. This was reflective of the standard care comparator arm. Using the conversion method outlined in the Cochrane handbook resulted in a relative risk of 0.549 (95%CI: 0.39–0.76) for treatment with rTMS. This was then applied to the 3-monthly rate of relapse for the standard care arm and converted to a probability, resulting in a 3-month probability of relapse for the rTMS arm of 6.9%.

Uncertainty and sensitivity analyses

Probabilistic sensitivity analysis (PSA) evaluated the effect of uncertainty in parameters on cost-effectiveness results. As shown in Table 1, the PSA parameters included were 3-month probability of hallucinations, proportion of the cohort estimated to be treatment resistant, 3-month probability of relapse with standard care, 3-month probability of relapse with rTMS, the proportion of cohort admitted to hospital, proportions receiving intensive case management or routine case management when in an acute state, and disability weights. Ersatz (version 1.35 Sunrise Beach, Australia) was used to run 5,000 iterations and calculate 95% uncertainty intervals (UI) for total costs, DALYs averted and incremental cost-effectiveness ratios (ICERs).

Additional univariate sensitivity analyses determined the effect of variables on cost-effectiveness. Due to the high upfront cost of rTMS sessions, the model is likely to be particularly sensitive to time horizon used and was extended to 5 years. In the extrapolated models, treatment is assumed to continue every fortnight with associated costs accrued, and treatment benefits continuing as observed in the Kennedy et al. [Reference Kennedy, Lee and Frangou5] meta-analysis.

Inpatient hospital costs contribute the major cost difference between the residual and acute health states. More recent Australian hospital data estimates a 8.5-day average length of hospital stay for schizophrenia with major complexity (AR-DRG U61A) at an average cost of $14,745 in 2021 $A [Reference Olsen26]. This cost was used in a univariate sensitivity analysis.

In its consideration of MBS funding for rTMS for use in treatment resistant depression, MSAC noted a lack of information available on the chosen cost of treatment ($160 per administration) [16]. The fee was varied in one-way sensitivity analyses between $100 and $160 in increments of $10.

Additional one-way sensitivity analysis, varying the effect size associated with rTMS treatment, was undertaken to identify a threshold where rTMS would be cost-effective.