Cost–effectiveness of combinatorial pharmacogenomic testing for depression from the Canadian public payer perspective

Depression is a significant health and economic burden in Canada. In addition to the impact of depression on patients’ health and quality of life [1], it costs the Canadian economy more than CAD $32 billion annually [2]. The cost of depression stems directly from increased healthcare resource utilization and indirectly as a result of increased disability and absenteeism [3,4]. For example, the direct and indirect per-patient costs were 3.5- and three-times higher, respectively, among a cohort of patients with depression compared with a non-depressed control cohort in Manitoba, Canada [4].

A direct approach to reducing costs associated with depression is through improved treatment efficacy, thereby increasing the number of patients who achieve remission from the disease. The primary treatment approach for depression is antidepressant medication; however, more than half of patients do not respond to their first medication trial and even fewer achieve remission [5]. This can result in a process of medication trial-and-error, leading to additional treatment failures or medication side effects. Better treatment approaches are needed in order to increase patients’ probability of achieving remission from their depression and decrease the time to such an outcome.

Pharmacogenomic (PGx) testing can be used to improve the treatment of depression. Combinatorial PGx is a third generation of PGx testing where multiple pharmacokinetic and pharmacodynamic gene–drug interactions are weighted and counterbalanced. This approach is distinct from the first-generation single cytochrome P450 gene tests or the second-generation tests of multiple gene–drug interactions and therefore, combinatorial PGx testing merits its own assessment of efficacy and cost–effectiveness. A recent randomized controlled trial of 1,167 patients with major depression in the USA demonstrated that combinatorial PGx-guided treatment results in a 50% relative increase (15.3 vs 10.1%) in remission rate compared with treatment as usual (TAU), as well as decreased depressive symptoms and greater response rate [6]. These findings are supported by several smaller USA clinical trials [7–9] and in Canada, by the large naturalistic IMPACT study that further demonstrated the clinical utility of the combinatorial PGx testing approach [10].

In addition to improving treatment efficacy, previous studies have demonstrated that combinatorial PGx testing results in significant reimbursement savings, reduction in polypharmacy and healthcare utilization and is cost-effective among patients with depression in the USA [9,11–14]. However, as there are substantial differences between the USA and Canadian healthcare systems and corresponding costs [15] and there are no published data on the cost–effectiveness of combinatorial PGx testing to guide depression treatment in Canada, it is important to evaluate this in the Canadian setting.

The purpose of the current study was to evaluate the cost–effectiveness of the use of combinatorial PGx testing, compared with treatment as usual, to guide the treatment of patients with moderate-to-severe depression from the perspective of the Canadian public healthcare system, which is a publicly funded healthcare system, including universal coverage for medically necessary healthcare services.

Materials & methods

Overview

We have developed a model analyzing the cost–effectiveness of using combinatorial PGx testing to guide the treatment of moderate-to-severe major depression compared with using TAU. The model was based loosely on the model by Hornberger and colleagues [13], with notable modifications to include remission as the target end point for depression instead of the earlier end point, response. The model was also updated to include cycle-based transition probabilities for remission, relapse rates and a hazard ratio to allow differential rates of relapse for patients in each model arm. The modelling approach is described in detail in the sections below. This modeling approach followed guidelines outlined in the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist.

The decision diagram and state-transition probability analysis are depicted in Figure 1. Four states, assessed annually within the time horizon, have been modeled based on a patient’s survival and remission from their depression. The duration of each cycle in the model is 1 year. All base-case input parameters, including the probabilities associated with mortality and remission, are provided in Table 1. Four versions of the model were run. Model 1 refers to the base case and models 2–4 refer to scenario analyses. The base-case and three scenario analyses are described in the ‘Remission rates’ section below.

Study population, treatment & time horizon

This analysis focused on patients with moderate-to-severe depression whose antidepressant treatment was guided by combinatorial PGx testing compared with patients whose treatment was unguided, and therefore underwent TAU. The starting age of the patient in this model was 32 years based on the median age of onset for major depressive disorder reported by Kessler et al. [18]. Due to the chronic and recurrent nature of major depression and in order to capture both early and downstream costs and savings associated with the management of this disease, a lifetime time horizon would be appropriate. However, we selected a conservative time horizon of 5 years, to align with previous analyses by Health Quality Ontario [19].

Remission rates

In order to quantify remission rates associated with combinatorial PGx-guided treatment and TAU for depression, we conducted meta-analyses of the per-protocol populations of the four published prospective clinical trials of the GeneSight combinatorial PGx testing for depression (Supplementary Figure 1) [6–9]. The methodology used for the meta-analyses was based on the previous meta-analysis of GeneSight conducted by Brown et al. [20], where study selection, PRISMA guidelines and assessment of bias were described in detail. The remission rates for combinatorial PGx and TAU derived from the meta-analysis of all four clinical trials are included in model 1, the base-case model. For the scenario analyses (models 2–4), remission rates were derived as follows. A second meta-analysis (model 2) was performed to determine remission rates reported in blinded randomized controlled trials (n = 2 studies), where placebo effect has been accounted for [6,9]. A third meta-analysis of the open-label clinical trials (model 3, n = 2 studies) provided an estimate of remission rates in the real world, where placebo effect is not controlled [7,8]. Finally, for the last scenario analysis (model 4), we used the remission rates from only one study – the largest (n = 1167) blinded, randomized, controlled trial of combinatorial PGx for depression [6].

The model’s output is dependent on both the absolute difference in remission rates between treatment strategies and the relative benefit of one strategy over the other. Therefore, a random effects model was used for all meta-analyses to weight the remission rates from each study and derive the relative risk (RR) associated with combinatorial PGx testing relative to TAU, referred hereafter as the relative benefit ratio (RBR). The combinatorial PGx group remission rate was calculated from the weighted TAU remission rate and the RBR, using the following formula: remission rate with combinatorial pharmacogenomic testing = 1 - (1 - remission rate TAU) ∧ RBR. The formula ensures that the remission rate for the combinatorial PGx strategy cannot exceed 100% no matter how large the RBR gets. All meta-analyses were conducted using the ‘meta’ package in R software version 3.5.0 (R Foundation for Statistical Computing, Vienna, Austria). Aside from remission rate and RBR, all other parameters were identical for the base-case scenario of the four models.

It was assumed that the relative benefit of the combinatorial PGx testing strategy over TAU in terms of remission rate would persist for 3 years, as the pharmacological treatment effect over placebo has been reported to persist up to 36 months according to a systematic review of randomized trials of continued antidepressant treatment [16]. Therefore, in our model, after 3 years, the relative benefit of combinatorial PGx testing ended and patients in both arms experience a similar probability of achieving remission.

Relapse rates

The rate of relapse following remission was determined for the combinatorial PGx group using unpublished data from the Genomics Used to Improve DEpression Decisions (GUIDED) clinical trial (NCT02109939). Of note, the primary outcomes from this trial have been published by Greden et al. [6]. At 24 weeks following treatment guided by combinatorial PGx testing, 9.1% of patients relapsed after previously achieving remission from their depression. We were unable to determine the relapse rate at 24 weeks for the TAU arm in the same clinical trial because this group was unblinded at week 12 and therefore was able to use the combinatorial PGx test report to guide treatment for weeks 12–24 of the study. Instead, our model used the TAU relapse rate of 23.3%, reported in a systematic review of antidepressant clinical trials by Sim et al. [17]. A hazard ratio (HR) of 0.39 was calculated from the combinatorial PGx and TAU relapse rates. The relapse rate for the combinatorial PGx strategy was then adjusted using the following formula to ensure that the rate remains between 0 and 1 regardless of the HR: relapse rate with combinatorial pharmacogenomic testing = 1 - (1 - relapse rate TAU) ∧ HR. This resulted in a relapse rate of 9.9% in the combinatorial PGx group. The reduced relapse rate in the combinatorial PGx compared with TAU group was projected to last 3 years. Like remission rates, after 3 years the relapse rates in both arms of the model became equal.

Mortality rates

Mortality rates were estimated from a Canadian study of the clinical and economic burden of depression [4]. The study compared clinical outcomes, healthcare utilization, direct costs and social services costs between a depression and matched-control cohort. For the purpose of our model, the control and depression cohorts served as surrogates for remitters and nonremitters, respectively. Annual suicide mortality rates were <0.01 and 0.04% for remitters and nonremitters, respectively [4]. To estimate the impact of the treatment arm on all-cause mortality risk, the RR was calculated for remitters and nonremitters compared with the general Canadian population [4,21]. The RR of all-cause mortality for remitters and nonremitters was 1.37 and 2.29, respectively.

Costs

The direct and indirect costs included in the analysis were derived from the same Canadian study of the clinical and economic burden of depression [4]. Direct costs consisted of annual hospital, physician and prescription drug costs. Indirect, or social services, costs consisted of annual rent assistance and employment income assistance. In this model, direct and indirect costs accumulated over a patient's lifetime, within the time horizon, whereas the $2500 combinatorial PGx test price was applied as a one-time cost for that treatment strategy. All costs are reported in 2018 Canadian dollars.

Utilities

Quality-adjusted life years (QALYs), a standardized measure of the quality and quantity of life lived used to quantify the value of the intervention, were derived from utility values for patients with major depression who are in remission (0.83) compared with the baseline score for patients with major depression (0.55), representing patients who are not in remission [3]. Utility scores can range from 0 to 1, signifying the health states of death to perfect health. The QALYs were calculated as the utility-weighted sum of the probability of being in remission or not in remission for each treatment strategy, across the 5-year time horizon. A discount rate of 3% was applied to all costs and health outcomes.

QALYs and incremental costs are used to calculate the incremental cost–effectiveness ratio (ICER), which is a standardized measure used to evaluate the economic value of one intervention over another. The ICER is equal to the incremental cost divided by QALYs.

Sensitivity analyses

One-way sensitivity analyses were performed separately for all four models. All base-case parameters were varied individually to determine their influence on incremental cost difference and ICER for the combinatorial PGx strategy compared with TAU. The range was set to ±25% of the mean for all parameters except for those measured in years, where the range was ±2 full years. A probabilistic sensitivity analysis was also performed for each of the four models using PyMC3, a probabilistic programming module for Python (PyMC3 version 3.7; Python version 3.7.3). Realistic probability distributions were defined for each model parameter (Supplementary Table 2) and simulations were run based on the parameter values sampled from the underlying distribution using Markov chain Monte Carlo sampling. In this study, 10,000 simulations were performed for each of the four models and outcome metrics were assessed across the ensemble of simulations.

Results

Base-case & scenario analyses

The results of the four models are provided in Table 2. In the base-case model, the combinatorial PGx strategy was associated with a 1.53-times greater remission rate and patients were projected to gain 0.17 QALYs (2.02 months), relative to the TAU strategy. After accounting for the price of the test, the combinatorial PGx strategy saved $2431 compared with TAU. The negative ICER demonstrated that the combinatorial PGx treatment strategy was dominant over the TAU strategy, as it was both more efficacious and cost-saving in all scenarios.

The same trends were observed in the scenario analyses (models 2–4). The relative benefit of the combinatorial PGx treatment strategy ranged from 1.47–1.55, resulting in a projected gain of 0.14–0.19 QALYs (1.72–2.28 months) relative to the TAU strategy. After accounting for the price of the test, the combinatorial PGx strategy saved $1687–3056 compared with TAU. According to the ICERs for the three models, the combinatorial PGx treatment strategy remained dominant over the TAU strategy.

Sensitivity analyses

One-way sensitivity analyses were performed separately for the four models (Figure 2; Supplementary Figure 2 & Supplementary Table 1). In all models, the incremental costs and ICERs were most sensitive to changes in the remission rate in the combinatorial PGx group and second, to variation in the RBR of the combinatorial PGx treatment strategy over TAU. In models 1, 2 and 4, the combinatorial PGx strategy was dominant over the TAU strategy and the ICER remained cost-saving (negative cost per QALY gained) when varying more than 90% of the input parameters (20/22 parameters). The two scenarios where the combinatorial PGx strategy was more costly, but still more effective, than the TAU strategy were when the combinatorial PGx remission rate and the RBR took their lower limits. For model 1, this resulted in incremental costs for the combinatorial PGx strategy compared with the TAU strategy of $284 when remission rate took the lower limit and $64 when the RBR took the lower limit. For model 2, the combinatorial PGx incremental costs were $490 and 356, respectively, when remission rate and RBR took the lower limit. For model 4, the incremental costs for the same scenarios were $649 when remission rate took the lower limit and $527 when the RBR took the lower limit. In model 3, the combinatorial PGx strategy was dominant over the TAU strategy when varying more than 95% of the parameters (21/22 parameters). In the one higher cost scenario, the incremental cost of the combinatorial PGx strategy relative to TAU was $124. This occurred when remission rate in the combinatorial PGx group took the lower limit. In all four models, The ICERs remained below the willingness to pay (WTP) threshold of $50,000/QALY when varying all input parameters.

These sensitivity analyses were limited to a range of 2 to 7 years for the time horizon. However, should a lifetime time horizon of 50 years be employed in our model, the difference in QALYs, cost savings and ICERs change by up to 32, 66 and 26%, respectively, where combinatorial PGx further dominates the TAU strategy.

A probabilistic sensitivity analysis was run for the four models (Figure 3; Supplementary Figure 3 & Supplementary Table 2). The difference in total costs, or incremental cost, and the difference in QALYs are depicted using scatter plots. The northwest, northeast, southwest and southeast quadrants of the scatter plots represent the four planes of cost–effectiveness, indicating neither effective nor cost-saving, effective but not cost-saving, not effective but cost-saving and both effective and cost-saving, respectfully. In all models, the combinatorial PGx treatment strategy dominated the TAU strategy for the majority of simulations (70.3–93.0%). That is, the combinatorial PGx treatment strategy was both more effective and cost-saving (southeast quadrant) than the TAU strategy. The probability of combinatorial PGx treatment being cost-effective at the WTP threshold of $50,000 for all four models is 94.5–96.7%.

Discussion

In the base-case model and scenario analyses, combinatorial PGx dominated the TAU strategy, indicating that it is not only more efficacious but also cost-saving, supporting use in clinical practice. The ICERs derived from the base case model and scenario analyses remained well below the common cost–effectiveness threshold of $50,000 when varying any of the parameters in the one-way sensitivity analyses. The parameter that had the greatest impact on the model when varied in one-way sensitivity analyses was the remission rate associated with combinatorial PGx-guided treatment. When remission rate took the lower limit, the combinatorial PGx strategy was associated with $124–649 higher costs than the TAU strategy, while ICERs remained well below $50,000. Based on the probabilistic sensitivity analysis for all four models, there was greater than 94% probability of combinatorial PGx being cost effective within the WTP threshold of $50,000, indicating the robustness of this model. Furthermore, the relative improvement in remission rates for the combinatorial PGx treatment approach compared with TAU is consistent across the different types of studies, despite the differences in remission rates between the RCT and open-label studies. All four models demonstrated a similar 50% improvement in remission rate for combinatorial PGx compared with TAU, strengthening the validity of the base-case model and the remission rates used therein.

In order to put these findings into perspective, the cost–effectiveness of combinatorial PGx-guided treatment could be considered in relation to other treatment approaches for depression that are available in Canada. For example, Health Quality Ontario, a provincial organization that conducts health technology assessments and provides public funding recommendations to the Ontario Ministry of Health and Long-Term Care, has published cost–effectiveness analyses of repetitive transcranial magnetic stimulation, internet cognitive behavioral therapy and group and individual psychotherapy for depression, compared with standard care [19,22,23]. These models produced ICERs ranging from $3,715 to $98,242 per QALY and all three treatment approaches were given positive funding recommendations for use when patients with depression met prespecified criteria. In comparison, the combinatorial PGx-guided treatment approach yielded significantly lower ICERs ranging from -$11,861 to -$16,124 per QALY gained, with the probability of cost–effectiveness ranging from 94.5–96.7% at a WTP threshold of $50,000 per QALY gained or 89.4–96.6% at a WTP threshold of $20,000 per QALY gained. These results are one piece of evidence in which combinatorial PGx testing to guide depression treatment demonstrates cost savings for the public payer.

Furthermore, emphasis should be placed on novel treatment approaches that are efficacious in both primary care and psychiatric care settings. In Canada, the number of practicing psychiatrists is decreasing and the unmet need for psychiatric services is increasing [24]. A Canadian study of the use of combinatorial PGx to guide depression treatment among primary care providers and psychiatrists demonstrated clinical utility in both physician groups, with the greatest improvement among patients whose treatment was provided by their primary care provider [10]. Optimizing the treatment of depression in primary care could increase psychiatrists’ capacity to focus on the most severely ill patients. Therefore, given that the combinatorial PGx treatment strategy is effective in multiple treatment settings and is projected to be cost saving for provincial governments, this may be one approach to reduce wait times for specialists in Canada.

Another cost–effectiveness analysis [25] has demonstrated the cost–effectiveness of PGx testing among n = 261 patients with moderate-to-severe depression. As the current study modeled the impact of combinatorial PGx on n = 1,425 patients with depression achieving remission, the results cannot be directly compared with the model by Groessl et al., where n = 261 patients were included and the target end point was response.

There were several limitations to the modeling approach used in this cost–effectiveness analysis. First, as long-term economic data associated with combinatorial PGx testing for depression were not available in the Canadian healthcare setting, the current study used an indirect assessment of costs associated with combinatorial PGx-guided depression treatment compared with TAU. We applied remission rates from the literature to Canadian costs associated with depression from the literature in order to derive costs associated with each treatment strategy. Moreover, the depression-related costs used in the model were mainly direct costs associated with the treatment of depression (hospital, physician and prescription drug costs), whereas indirect costs were limited to select social services (rent assistance and employment income assistance). Additional indirect costs that could be considered in future cost–effectiveness modeling include disability, absenteeism and lost productivity associated with depression, which are associated with a significant burden on Canada’s economy of approximately $32 billion annually [2].

Furthermore, this study is limited by assumptions made when empirical data were not available. For example, where long-term durability data were unavailable, we projected a 3-year relative benefit of combinatorial PGx over TAU. This was based on the duration of benefit of pharmacotherapy [16] and mirrored the previously published US cost–effectiveness model of combinatorial PGx for depression [13].

This study may also be limited by the use of a 5-year time horizon, which may underestimate the long-term cost-savings associated with combinatorial PGx-guided treatment of depression. Both the National Institute for Health and Care Excellence (NICE) and the Canadian Agency for Drugs and Technologies in Health (CADTH) guidelines for economic evaluations recommend a time horizon that is long enough to capture all important differences in costs or outcomes between the two treatment strategies [26,27]. In addition, when a treatment strategy may influence mortality, the recommended time horizon is the patient’s lifetime [28]. When using a lifetime time horizon of 50 years in an exploratory analysis within the current model, the difference in QALYs, cost savings and ICERs changed by up to 32, 66 and 26%, respectively, where combinatorial PGx further dominates the TAU strategy. However, using the lifetime perspective can create analytical challenges where highly uncertain extrapolations are used. Therefore, for the purpose of this cost–effectiveness assessment, we selected a conservative 5-year time horizon for our primary analysis, as indicated in a recent Health Technology Assessment of psychotherapy for depression and anxiety by Health Quality Ontario [19].

Additionally, we conservatively selected a discount rate of 3%, instead of the 1.5% discount rate suggested by NICE and the CADTH for economic analyses [27]. However, when the model was re-run with the 1.5% discount rate, there was little impact on the model output and the incremental costs, QALYs and ICERs further favored combinatorial PGx over TAU.

At last, the relapse rates associated with each treatment strategy used in the model were derived from different studies. The relapse rate for patients using the combinatorial PGx treatment strategy was derived from unpublished data from the GUIDED clinical trial, whereas relapse for patients using the TAU strategy was reported in a systematic review of antidepressant clinical trials by Sim et al. [17]. As there will inevitably be differences in study design, the relapse rates are simply estimates for the two treatment strategies and sensitivity analyses are particularly important in this situation. When relapse rates were varied in one-way sensitivity analyses, the resulting incremental costs and the ICERs remained lower than -$10,000 (i.e., cost saving). Extending this further, when the combinatorial PGx relapse rate was increased by 100%, all ICERs remained below -$5000, signifying that even at the upper limit of relapse, the model remains cost saving.

Conclusion

The present study examined the cost–effectiveness of using combinatorial PGx testing to guide the treatment of depression compared with treatment as usual. The combinatorial PGx treatment strategy was more efficacious (gain in QALYs) and less costly (negative incremental costs) compared with the TAU strategy for depression. With the support of one-way and probabilistic sensitivity analyses, we conclude that combinatorial PGx testing is cost–effective compared with TAU for patients with moderate-to-severe depression from the perspective of the Canadian public healthcare system. Considering that approximately 1.6 million Canadians have reported unmet or only partially met mental healthcare needs [29], our approach for treating mental illness, including depression, requires improvement and greater investment. However, in Canada, mental health accounts for 7.2% of total government health expenditures, which, in the international context, is less than most developed countries [30]. According to this cost–effectiveness analysis, investing in the use of combinatorial PGx testing to guide the treatment of depression is projected to improve patient outcomes and quality of life, as well as provide cost-savings for the Canadian public healthcare system.