Cost-effectiveness of Electroconvulsive Therapy vs Pharmacotherapy/Psychotherapy for Treatment-Resistant Depression in the United States

Key Points

Question  What is the cost-effectiveness of electroconvulsive therapy compared with antidepressant medications and/or psychotherapy for treatment-resistant major depressive disorder in the United States?

Findings  In this mathematical modeling analysis integrating data from multiple published sources, offering electroconvulsive therapy as third-line treatment for depression would cost an estimated $54 000 per quality-adjusted life-year gained. Over 4 years, this would reduce time with uncontrolled depression from 50% to 34% of life-years.

Meaning  Electroconvulsive therapy may be an effective and cost-effective treatment for treatment-resistant depression and should be considered after failure of 2 or more lines of pharmacotherapy and/or psychotherapy.

Importance  Electroconvulsive therapy (ECT) is a highly effective treatment for depression but is infrequently used owing to stigma, uncertainty about indications, adverse effects, and perceived high cost.

Objective  To assess the cost-effectiveness of ECT compared with pharmacotherapy/psychotherapy for treatment-resistant major depressive disorder in the United States.

Design, Setting, and Participants  A decision analytic model integrating data on clinical efficacy, costs, and quality-of-life effects of ECT compared with pharmacotherapy/psychotherapy was used to simulate depression treatment during a 4-year horizon from a US health care sector perspective. Model input data were drawn from multiple meta-analyses, randomized trials, and observational studies of patients with depression. Where possible, data sources were restricted to US-based studies of nonpsychotic major depression. Data were analyzed between June 2017 and January 2018.

Interventions  Six alternative strategies for incorporating ECT into depression treatment (after failure of 0-5 lines of pharmacotherapy/psychotherapy) compared with no ECT.

Main Outcomes and Measures  Remission, response, and nonresponse of depression; quality-adjusted life-years; costs in 2013 US dollars; and incremental cost-effectiveness ratios. Strategies with incremental cost-effectiveness ratios of $100 000 per quality-adjusted life-year or less were designated cost-effective.

Results  Based on the Sequenced Treatment Alternatives to Relieve Depression trial, we simulated a population with a mean (SD) age of 40.7 (13.2) years, and 62.2% women. Over 4 years, ECT was projected to reduce time with uncontrolled depression from 50% of life-years to 33% to 37% of life-years, with greater improvements when ECT is offered earlier. Mean health care costs were increased by $7300 to $12 000, with greater incremental costs when ECT was offered earlier. In the base case, third-line ECT was cost-effective, with an ICER of $54 000 per quality-adjusted life-year. Third-line ECT remained cost-effective in a range of univariate, scenario, and probabilistic sensitivity analyses. Incorporating all input data uncertainty, we estimate a 74% to 78% likelihood that at least 1 of the ECT strategies is cost-effective and a 56% to 58% likelihood that third-line ECT is the optimal strategy.

Conclusions and Relevance  For US patients with treatment-resistant depression, ECT may be an effective and cost-effective treatment option. Although many factors influence the decision to proceed with ECT, these data suggest that, from a health-economic standpoint, ECT should be considered after failure of 2 or more lines of pharmacotherapy/psychotherapy.

Electroconvulsive therapy (ECT) is a highly effective treatment for depression.¹ Research indicates that ECT can be significantly more effective than pharmacotherapy, with 50% to 60% of patients achieving rapid remission of depression after a course of ECT compared with 10% to 40% with pharmacotherapy/psychotherapy.^2-4 Beyond its immediate clinical effects, ECT is also associated with decreased psychiatric hospitalization rates^5,6 and reduced long-term risks of suicide and all-cause mortality.⁷

Despite these benefits, ECT use remains low. A 2017 analysis found that only 1.5% of psychiatric inpatients with severe affective disorders received ECT while hospitalized,⁵ and among individuals with depression in a large insurance claims database, fewer than 1% had ever received ECT.⁸ Numerous explanations have been proposed for the infrequent use of ECT including stigma, adverse effects, and lack of access.^1,5,9

Although ECT can be a first-line treatment for depression with life-threatening psychotic or suicidal features, it is most often used in the United States for depression that has failed to respond to pharmacotherapy and/or psychotherapy.^10-12 While treatment-resistant depression is a common indication for ECT, many clinical guidelines do not clearly specify how many pharmacotherapy/psychotherapy trials must fail before ECT is offered.^9,12-14 In practice, patients often spend months to years with uncontrolled depression before considering ECT; in 1 US-based trial, patients averaged 38 weeks and 5 prior medication trials before initiating ECT.¹⁵

Another issue that may limit use of ECT is its cost, estimated at $300 to $1000 per treatment.^16,17 With 5 to 15 treatments per initial course and 10 to 20 maintenance treatments per year, the annual cost of ECT can exceed $10 000 vs several hundred dollars for many antidepressant medications.¹⁸ Amidst ongoing efforts to improve the value of US health care, it is critical to assess whether ECT confers a clinical benefit commensurate with its higher cost. Although prior studies have suggested that ECT for depression may be cost-effective in some health care systems,^19,20 we are aware of no study that has reported the cost-effectiveness of ECT for treatment-resistant depression in the United States; furthermore, to our knowledge, no prior study has evaluated when in the course of depression treatment ECT should be offered.

To address these issues, we developed a decision analytic model to simulate the clinical and economic effects of ECT. Our analysis has 2 main objectives: (1) to quantify the cost-effectiveness of ECT for treatment-resistant depression in the United States and (2) to define criteria for offering ECT that will maximize its health-economic value.

We used a decision analytic model to assess the cost-effectiveness of varying strategies for offering ECT to patients with major depressive disorder. Strategies are characterized by the number of prior treatment lines (defined as a unique combination of antidepressant medications and/or psychotherapy)^4,21 a patient must have tried before receiving ECT. We simulated 7 strategies: 1 strategy with no ECT available and 6 strategies with 0 to 5 lines prior to ECT. These strategies specifically target treatment-resistant depression because it is the most common indication for ECT in the United States.¹² Because the decision to offer ECT to patients with life-threatening suicidality or psychosis relies on different costs, benefits, and clinical factors from those included in our model, we preferentially used data sources that do not target these indications.¹² However, because there is substantial overlap between psychosis, suicidality, and treatment-resistant depression, many of our data sources included some patients with psychotic or suicidal features.^22-24

We simulated the clinical and economic effects of these 7 strategies over a 4-year horizon. This horizon was chosen to allow adequate time for the costs and benefits of the various strategies to accrue while remaining within the range of prior modeling studies and long-term depression outcomes data.^25,26 From these simulations, we calculated overall quality-adjusted life-years (QALYs, a measure combining longevity with health-related quality of life²⁷) and costs from a health care sector perspective.²⁸

The incremental cost-effectiveness ratio (ICER) of each strategy was calculated as the ratio of its incremental cost (in 2013 US dollars) to its incremental benefit (in QALYs) relative to the next most effective nondominated strategy. A strategy is considered dominated if it is less effective and more costly than another strategy (strong dominance) or if it is less effective and would have a higher ICER than another strategy (extended dominance).²⁷ Commentators recommend using ICER thresholds ranging from $50 000 per QALY to $150 000 per QALY to define cost-effective health interventions in the United States^29,30; for simplicity, we designated strategies with an ICER of $100 000 per QALY or less as cost-effective and also provided exact ICER values. In describing our model, analysis, and results, we adhered to the 2013 Consolidated Health Economic Evaluation Reporting Standards.³¹ Because this was secondary research without use of any identifiable data, this study was not regulated by the University of Michigan Health Sciences institutional review board. For the same reason, no consent was obtained.

We developed a deterministic, state-transition model of depression treatment using Microsoft Excel 2016 (Figure 1); its mathematical details are fully described in the eAppendix in the Supplement and summarized here. The model includes 5 health states:

1. Initiation: the first month of a treatment line, before treatment benefits typically emerge³;

2. Remission: near-complete recovery of depressive symptoms (defined by score on a validated rating scale, eg, 24-item Hamilton Rating Scale of Depression <10)³²;

3. Response: partial recovery of depressive symptoms without remission (defined by at least 50% reduction in 16-item Quick Inventory of Depressive Symptomatology-Self-Report)⁴;

4. Nonresponse: initial failure to achieve remission or response; and

5. Relapse: return of depressive symptoms after initial remission or response.

A simulated cohort of patients with major depressive disorder enters the model at initiation of first-line treatment and progresses through up to 9 treatment lines. Consistent with results of the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial,⁴ we do not explicitly distinguish between specific medications or between the presence/absence of psychotherapy. Instead, each change in treatment line represents switching between or augmenting the medications/psychotherapy used.

Table 1^{2,4,16,17,22-24,28,32-45} shows base case model inputs, along with ranges and distributions used in sensitivity analysis. Where available data permit, ranges reflect 95% confidence intervals; in other cases, the source of the range is described in subsequent sections.

To reflect their current value, future costs and QALYs are discounted at an annual rate of 3%.²⁸ The model has a 1-month cycle length; this ensures patients do not switch from a failed treatment until at least 8 weeks have elapsed, consistent with clinical guidelines and STAR*D outcomes.^4,46

Based on STAR*D,⁴ we simulated a population with a mean (SD) age of 40.7 (13.2) years, and 62.2% women. We apply 2013 CDC life table data³³ to this age/sex distribution, along with a relative mortality rate of 1.58 (95% CI, 1.47-1.70) for individuals with depression compared with the general population, from a 2014 meta-analysis.³⁴ Combining these data yields an overall annual mortality probability of 0.00479.

Although some observational studies suggest that depression treatment may reduce suicide risk,⁷ randomized studies have not found a significant effect.⁴⁷ Hence, we make the conservative assumption that treatment does not affect mortality rates.

We derived probabilities of initial remission (13.0%-36.8%), response (16.3%-48.6%), and nonresponse (51.4%-83.7%) across 4 steps of antidepressant therapy from the STAR*D trial.⁴ Of note, the first-step selective serotonin reuptake inhibitor remission and response rates observed in STAR*D (36.8% and 48.6%) are markedly lower than those reported by multiple meta-analyses of selective serotonin reuptake inhibitor efficacy,^35,48 possibly reflecting preexisting treatment-resistance among STAR*D participants. For this reason, first-line remission (42.4%) and response (60.3%) rates in our base case are drawn from 1 such meta-analysis,³⁵ and the 4 steps of STAR*D are used for treatment lines 2 to 5.

Annual probabilities of relapse after initially successful treatment (47.5%-88.1%) are also derived from STAR*D.⁴ Relapse probabilities increased for later treatment lines and after response compared with remission.⁴

Annual health care costs of people with depression (including outpatient, inpatient, medication, and laboratory costs) are derived from 2 studies that used the MarketScan insurance claims database to assess costs by number of prior medication trials.^43,44 Consistent with other analyses,^8,49 annual costs increased with increasing numbers of prior failed treatments. We used the Bureau of Economic Analysis medical care expenditure indices for mental health care (eAppendix in the Supplement) to inflate costs to 2013 US dollars; this is the most recent year for which these indices are available.^50,51 This yielded annual costs ranging from $7833 to $16 816 per year for 0 to 8 prior trials.

Based on a prospective cohort of patients receiving antidepressant pharmacotherapy, we used utility values of 0.85, 0.72, and 0.58 for the remission, response, and nonresponse/relapse/initiation states.⁴² These values are consistent with clinical trial results⁵² and provide a more conservative estimate of the benefit of remission than reported in a meta-analysis that included lower-quality studies.⁴⁹

Based on a meta-analysis of observational evaluations of ECT efficacy in unipolar depression, we used an initial remission probability of 50.9%²; meta-analyses of randomized trials comparing repetitive transcranial magnetic stimulation vs ECT (53.0%)⁵³ and right-unilateral vs bitemporal ECT (52.5%)³² report similar values. Our primary source did not assess response rates, so we calculated a weighted mean response rate (66.6%) from the included studies.^{23,24,36-38,54}

Relapse rate after ECT was derived from 3 studies that are, to our knowledge, the only randomized trials of maintenance ECT plus pharmacotherapy vs pharmacotherapy alone; 6-month relapse rates in the ECT groups were 8%,³⁹ 29%,⁴⁰ and 13%.²² A weighted mean of these results yields a 6-month relapse probability of 16.8%, corresponding with an annual probability of 30.8%.

We assumed all patients offered ECT would accept it (ie, we simulated exclusively ECT-willing patients). Although real-world uptake is substantially lower, this will not affect the cost-effectiveness of ECT among those who do accept it; to confirm this, we varied uptake between 10% and 100% in sensitivity analysis.

Because recent, nationally representative cost data were lacking, we used a per-treatment ECT cost derived from a 1998 report by Bailine⁴⁵ updated to 2013 US dollars, yielding a value of $586.⁵⁰ We excluded inpatient care from the per-treatment ECT cost estimate because ECT is increasingly performed outpatient or initiated after a patient has already been hospitalized and hence would not cause increased inpatient length of stay; we also excluded the cost of psychiatric evaluations because these were already incorporated into our overall cost estimates.^5,6 In sensitivity analysis, we varied this cost between $328 (reflecting per-treatment Centers for Medicare and Medicaid Services reimbursement to inpatient psychiatric facilities^5,16) and $1000 (as estimated by an ECT specialist¹⁷). We assumed that ECT use had no direct effect on background health care costs, but that it could affect these costs as mediated by enhanced remission/response of depression.

Based on multiple ECT trials, we simulated 8 ECT sessions during 1 month during the initial treatment course.³² For maintenance treatment, we used a base-case value of 16 treatments per year, as reported in multiple maintenance ECT trials.^39,55 In sensitivity analysis, we varied this between 7.1 treatments per year,²² from a trial of adaptive maintenance ECT, to 24.9 treatments per year, for symmetry. We made the conservative assumption that the number of maintenance treatments in the first year applied to every subsequent year.

In sensitivity analysis, we assessed incorporating a direct utility decrement for patients receiving ECT (reflecting the possibility of a worse adverse event profile than background pharmacotherapy/psychotherapy⁵⁶) or a utility benefit (reflecting the faster clinical response and substantial utility increases observed in ECT trials^3,57). For maintenance ECT, we varied this utility effect between −0.04 and 0.04 per month, from an estimate of the disutility associated with antidepressant medications.⁴¹ Because adverse cognitive effects of ECT are predominantly observed in the first 15 days of treatment,⁵⁶ we tripled this range (−0.12 to 0.12) during the initial course.

To assess the robustness of our results to model input uncertainty, we performed univariate, probabilistic, and scenario sensitivity analyses. In univariate sensitivity analysis, we varied each individual model parameter over the ranges shown in Table 1, and calculated cost-effectiveness outcomes at its upper and lower bounds.

In probabilistic sensitivity analysis, the model was run using parameter values drawn at random from distributions reflecting the uncertainty in the estimation of each parameter, ie, second-order uncertainty (Table 1)⁵⁸; to estimate overall uncertainty, we repeated this process 10 000 times. For each run, the optimal strategy, defined as the strategy producing the most QALYs with ICER of $100 000 per QALY or less, is determined. Prior research suggests that modeling probabilistic inputs independently may underestimate overall uncertainty.⁵⁹ To address this, we performed 1 probabilistic sensitivity analysis with minimal parameter covariance (independent parameter variance), and 1 with maximal covariance (linked parameter variance); further details are provided in the eAppendix in the Supplement.

Finally, in scenario sensitivity analysis, we assessed alternative approaches to modeling pharmacotherapy efficacy, depression cost, and maintenance ECT use. First, we assessed using first-line efficacy data from the previously described meta-analysis,³⁵ with subsequent response and remission probabilities calculated based on a 19% reduction in the odds of remission/response with each successive line.⁶⁰ Alternatively, we assessed using only STAR*D efficacy data, without incorporating first-line efficacy data from the aforementioned meta-analysis.^4,35 Next, we used costs of depression derived from a study by Gibson et al.⁸ This study is from 2010, rather than 2002 for our base-case data source, but uses a metric of prior treatment experience that is less analogous to lines in our model; annual costs range from $6916 to $15 576 for treatment lines 1 to 9. We next assessed incorporating an absolute mortality reduction of 0.07% for patients with remission/response of depression, reflecting 1 estimate of the reduction in suicide mortality with effective treatment.⁶¹ Finally, we simulated using maintenance ECT only after 1 relapse with maintenance pharmacotherapy alone; annual relapse probability with maintenance pharmacotherapy alone is set to 51.1%.⁶²

To provide external validation of our model, we compared model-generated outcomes with independent published results. During the 24 months after initiation of first-line antidepressant treatment, the model estimated an annual per-patient cost of $9413, compared with $9287 from a 2017 study of primarily privately insured patients⁶³ and $11 263 among Medicaid patients.⁶⁴ During a 4-year horizon without ECT, the model estimated patients spend 50.2% of their life-years in nonresponse/relapse/initiation states compared with a mean fraction of time ill of 46% (95% CI, 34%-58%) from a meta-analysis of long-term unipolar depression outcomes.²⁶ Finally, the model estimated that mean utility increased by 0.159 after initial ECT course and 0.201 six months later, compared with 0.177 and 0.204 from a large ECT trial.^65,66

Base case results are presented in Table 2. Over 4 years, ECT was projected to reduce the fraction of time with nonresponse of depression from 50.2% of life-years to 37.1% to 32.9% of life-years, with greater reductions when ECT is offered earlier. These reductions translate to mean quality-adjusted survival gains of 0.12 to 0.15 QALYs, at incremental costs of $7300 to $12 000. Cost increases were attributable primarily to the cost of ECT; mean ECT-related cost increased from $9100 to $17 100 with earlier ECT offer, while the mean cost of other health care declined from $42 500 (without ECT) to $37 400 (with first-line ECT).

Using a willingness-to-pay threshold of $100 000 per QALY, third-line ECT was projected to be cost-effective, with an ICER of $54 000 per QALY. Second-line and first-line ECT were not cost-effective, with ICERs of $564 000 per QALY and $815 000 per QALY. Analyzed incrementally, fourth-, fifth-, and sixth-line ECT were all dominated (ie, they offered fewer QALYs at a worse ICER than other strategies). However, when compared with not offering ECT, each of these strategies would be cost-effective, with cost-effectiveness ratios between $60 000 to $70 000 per QALY.

Univariate sensitivity analysis results are shown in Figure 2. Third-line ECT remained cost-effective (ICER ≤$100 000 per QALY) across all individual parameter variations. The cost of ECT was an important source of uncertainty, with annual maintenance cost and initial course cost both among the 4 parameters with the most influence over the ICER of third-line ECT (ICER ranges were $20 000 per QALY to $87 000 per QALY for maintenance cost and $41 000 per QALY to $74 000 per QALY for initial course cost). Extending beyond the preestablished uncertainty ranges, third-line ECT would reach an ICER of $100 000 per QALY at an annual maintenance cost of $16 590 or an initial course cost of $12 180.

In probabilistic sensitivity analysis, we found a 74% to 78% likelihood that at least 1 ECT strategy was cost-effective (ICER ≤$100 000/QALY) depending on the degree of parameter covariance (Table 3). Third-line ECT was the single optimal strategy in 56% to 58% of PSA simulations.

Finally, the 5 scenario sensitivity analyses (with alternative approaches to pharmacotherapy efficacy, depression cost, maintenance ECT use, and suicide) did not markedly change our results, with third-line ECT remaining cost-effective (ICER ≤$100 000/QALY) in all scenarios (eTable in the Supplement).

We used a decision analytic model to evaluate the clinical and economic effects of varying strategies for using ECT to treat treatment-resistant depression in the United States. We found that ECT would substantially improve clinical outcomes, reducing time with uncontrolled depression from 50.2% to 32.9% to 37.1% of life-years over a 4-year horizon. We projected a 74% to 78% likelihood that ECT would be cost-effective in the United States given commonly accepted cost-effectiveness thresholds (ICER ≤$100 000 per QALY).²⁹ Across a range of realistic variations in model input data, we found that offering ECT after failure of 2 lines of pharmacotherapy/psychotherapy would most reliably maximize its health-economic value. For patients with 3 or more prior treatment trials, offering ECT later in the course of treatment would still be cost-effective. When compared with other depression interventions that have been evaluated, the cost-effectiveness of ECT for treatment-resistant depression is superior to pharmacogenetic testing⁴¹ and similar to a rural collaborative care intervention.⁶⁷

Based on these findings, we recommend that clinicians consider offering ECT to patients with major depressive disorder who have failed to respond to 2 trials of pharmacotherapy and/or psychotherapy. Notably, our findings align well with 2017 recommendations from Conway et al, who propose defining treatment-resistant depression by “failure of 2 adequate dose-duration antidepressants or psychotherapy from different classes”²¹ and recommend considering ECT at this point. Similarly, the 2017 Florida best practice guidelines for treatment of major depressive disorder include ECT as a level 3 treatment, to be considered after failure of levels 1 and 2.⁶⁸

Despite these recommendations, we recognize that many clinicians will regard offering ECT as third-line treatment to be overly aggressive, especially given the common view of ECT as a last-resort treatment for depression.^15,69 To understand how third-line ECT can be reasonable and cost-effective, it is crucial to recognize how challenging treatment-resistant depression is to manage. Observational and trial data indicate that fewer than one-third of patients with this condition respond to pharmacotherapy and/or psychotherapy, and those who do respond are at high risk of relapse.^4,70 Given the limited effectiveness of commonly used treatments, it may be less surprising that earlier ECT use would greatly improve clinical outcomes for patients with treatment-resistant depression.

Our analysis has limitations related to the model’s structure, the input data used, and the scope of the analysis. By necessity, a model must include some simplifying assumptions. Where possible, we have chosen these assumptions to be conservative with respect to ECT’s cost-effectiveness, such as assuming indefinite maintenance ECT use, excluding ECT’s effects on mortality and speed of remission,^3,7 and excluding the possibility of greater efficacy with earlier ECT use.¹⁰ In addition, our model assumes that transition probabilities between states (eg, relapse probabilities⁷¹) do not vary over time. Despite these simplifications, the model’s outcomes are well validated by independent data on costs, long-term clinical outcomes, and quality of life.

Next, there are several shortcomings in our model’s input data. Much of our cost data are more than a decade old and reflect primarily privately insured patients,^43-45 and there is uncertainty regarding the cost of ECT.^16,17,45 In addition, many of our estimates of relapse rates with pharmacotherapy or maintenance ECT reflect sample sizes of 100 patients or fewer.^4,22,39,40 Finally, our reliance on clinical trial data may limit the generalizability and external validity of our results.⁴ However, our main findings are robust to sensitivity analysis using alternative data sources or appropriately broad confidence intervals.

Finally, the scope of our analysis introduces several limitations. Our choice of a health care sector perspective is intended to be relevant to patients, clinicians, and payers; however, this perspective excludes indirect costs and benefits such as patient transportation and productivity gains after depression treatment.^17,72 Additionally, we did not evaluate other novel interventions for treatment-resistant depression, such as repetitive transcranial magnetic stimulation or ketamine.^13,46

In this mathematical modeling analysis, we found that ECT is a cost-effective treatment option for treatment-resistant depression in the United States. Offering ECT after 2 failed lines of pharmacotherapy/psychotherapy is most likely to maximize its health-economic value and is concordant with recommendations from some national guidelines and ECT specialists. Increasing use of ECT by offering it earlier in the course of treatment-resistant depression could greatly improve outcomes for this difficult-to-treat patient population.

Corresponding Author: Eric L. Ross, BA, University of Michigan Medical School, 1135 E Catherine St, Room 5100, Taubman Health Sciences Library, Ann Arbor, MI 48109 (ericlr@umich.edu).

Accepted for Publication: March 6, 2018.

Published Online: May 9, 2018. doi:10.1001/jamapsychiatry.2018.0768

Author Contributions: Mr Ross had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Ross, Maixner.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: All authors.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Ross.

Administrative, technical, or material support: Zivin.

Supervision: Zivin, Maixner.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by the Department of Veterans Affairs Health Services Research and Development Services grant IIR 14-324 (Dr Zivin).

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.