Treatment of Major Depressive Disorder in Patients Failing Initial Therapy: An Excel-Based Pharmacoeconomic Model

BaCkgRounD: Treatment-resistant depression (TRD) presents a unique challenge in managed care, requiring review of both the clinical and economic components of care. oBjeCTive: To review the TRD disease state as well as data supporting the various therapeutic options available for the treatment of persistent depression in managed care. SuMMaRy: While there is no consensus on the definition of TRD, persistent disease can generally be defined as depression that fails to respond to adequate treatment. When initial treatment is not effective or tolerable after 6 to 8 weeks of therapy, the american Psychiatric association (aPa) treatment guidelines recommend dose titration, augmentation, or switching. in the case of a therapy switch, the body of evidence suggests that selection of an agent with a different mechanism of action than the initial agent may be the most effective treatment. Furthermore, when patients maintain continuous therapy for the recommended treatment duration, outcomes are improved compared with patients who discontinue therapy early. as a result, the most effective treatment strategies promote improved patient compliance as well as the use of agents associated with a reduced incidence of premature discontinuation and therapy change early in the treatment program. While data supporting these clinically effective components of therapy exist, few data are available demonstrating the most cost-effective therapeutic options for TRD. ConCluSion: This analysis suggests that managed care providers could benefit from a model that they can customize to evaluate the overall cost-effectiveness of different strategies in the management of depression.

M ajor depressive disorder (MDD) is the second most common psychiatric disorder in the United States. 1 The lifetime prevalence for major depression appears to be between 6% and 16%. 2,3 While the etiology of depression is not fully understood, evidence suggests that depression is the result of a complex interaction among biological, genetic, psychosocial, and environmental factors. 4,5 The Diagnostic and Statistical Manual, Fourth Edition, Text Revision (DSM-IV-TR) outlines common signs and symptoms of a major depressive episode. 6 While not a definitive list, the following signs and symptoms are derived from DSM-IV-TR criteria: 1. Loss of interest, satisfaction, or pleasure in almost all activities, lasting at least 2 weeks 2. Appetite and sleep disturbance (early morning awakening is "classic") 3. Decreased energy, concentration, or libido 4. Low self-esteem or excessive guilt 5. Recurrent thoughts of death or suicide 6. Psychomotor agitation or retardation 7. Occasional psychotic features (delusions, hallucinations) 8. Atypical features may be present in elderly, children/ adolescents The risk of depression may be related to the same combination of factors that produce depression. 5 The highest rates of depression occur in individuals between the ages of 25 and 44 years. Females are almost twice as likely (10%-25%) as males (5%-12%) to experience depression. Genetic predisposition appears to be a significant risk factor. Individuals with firstgeneration relatives with major depression have a 1.5 to 3 times greater chance of experiencing depression compared with individuals without a similar family history. 4,7 Individuals who have been victims of trauma or abuse are also at increased risk of depression. 8,9 In addition to the risk factors described, some medications can cause depression-like symptoms, including sedatives, narcotics, and pain relievers. 10 Untreated depression has significant economic, social, physical, and psychological consequences. Several factors contribute to the economic burden of depression, including the prevalence of the disease, treatment rate, and rate and degree of impairment. 11 Studies conducted in 1990 estimated that depressed workers in the United States lost an average of 5.6 productive hours per week. The same studies estimated that depression-related costs of direct treatment, lost earnings, and indirect workplace costs translated into an economic burden of between $44 and $53 billion per year. These estimates did not include labor costs associated with short-and long-term disability. 12,13 Between 1990 and 2000, the total economic burden of depression remained relatively stable. While treatment rates increased substantially over that period, indirect workplace costs remained the largest single cost component. 11 The characteristics of depression, including fatigue, reduced concentration, and difficulty performing routine tasks, all contribute to reduced productivity and increased absenteeism. 11 Patients with depression also have increased medical morbidity and mortality, including higher rates of premature death related to cardiovascular disease and myocardial infarction. 14,15 In addition, 15% of people diagnosed with major depression will commit suicide, and two thirds of all suicides are related to depression. 16 nn Treatment Options Seven different pharmacologic classes of medications can be used to treat depression (Table 1). 17 The primary targets of most major antidepressant drug classes are the neurotransmitters serotonin and norepinephrine. The oldest agents are the tricyclic antidepressants (TCAs) and monoamine oxidase inhibitors (MAOIs). TCAs inhibit the reuptake of serotonin and norepinephrine. MAOIs block the activity of enzymes (MAO-A, MAO-B) that are involved with the breakdown of serotonin, norepinephrine, and dopamine. Although both TCAs and MAOIs are effective, their use is limited, primarily because of side effects. TCAs are associated with cardiac, anticholinergic, and hypotensive side effects, as well as the potential for severe toxicity with overdose. Oral MAOIs require adherence to dietary restrictions, except for the newer transdermal systems at entry-level dosages. Newer agents are as effective as TCAs and MAOIs but have been shown to be safer and more tolerable. 17 Among the newer antidepressant agents, selective serotonin reuptake inhibitors (SSRIs) and serotonin and norepinephrine reuptake inhibitors (SNRIs) have been commonly used. Studies have shown that SSRIs and SNRIs are effective for MDD when patients remain on therapy at the minimum recommended dose and duration of time set forth in clinical practice guidelines (at least 4-9 continuous months). [18][19][20] Although many antidepressants have similar efficacy as first-line agents, few studies have compared them as second-line treatments following initial treatment failure. 21 nn Treatment Issues In the managed care environment, initial and subsequent treatment efficacy, tolerability, and adherence influence clinical outcomes and pharmacoeconomic aspects of care. In the treatment of depression, outcomes of particular concern that are affected by these factors include rates of response and remission. While a response is defined by a partial improvement in depressive symptoms, remission is characterized by a full recovery from depressive symptoms and a return to normal functioning. 22 Measures of economic burden (probability of paid employment, time lost from work, and total health care costs) correlate significantly to these clinical outcomes. 23 In a study of 290 primary care patients with MDD, patients who achieved remission at 12 months had a 16% higher probability of paid employment and missed 10 fewer days of work per year compared with patients with persistent depression (mean, 16.8 days). 23 At year 2, patients who achieved remission had 49% lower total health care costs compared with those with persistent depression after adjusting for baseline costs and  other covariates. 23 It should be noted, however, that only 1 of the economic endpoints in this study reached statistical significance.
Even when an agent is effective, lack of compliance can be a major barrier to successful treatment. The ability of the patient to tolerate a drug' s side effects strongly influences their compliance with therapy. Developing effective pharmacotherapy strategies that improve adherence and increase remission rates can potentially lower costs, reduce the risk of relapse, and improve psychosocial functioning and productivity. 24,25 nn

Treatment-Resistant Depression
While there is no consensus on the definition of treatmentresistant depression (TRD), certain guidelines based on accepted clinical outcomes measures, such as the Hamilton Rating Scale for Depression (HAM-D), can be used to identify TRD. Importantly, most published definitions of TRD imply that the patient has had either no response or inadequate response to adequate treatment. Nierenberg and DeCecco suggested that TRD in patients who received adequate treatment could be defined based on any of 3 criteria: failure to achieve a minimum response (e.g., less than a 25% decrease from baseline HAM-D score), failure to achieve a response (e.g., less than a 50% decrease from baseline HAM-D score), or failure to achieve remission (e.g., a final HAM-D score of at least 7). 26 Patients who are treatment resistant use a disproportionately larger share of health care resources, have significantly more claims for comorbid conditions, and cost employers more in lost productivity compared with patients with major depression who respond to treatment. 27 Many depressed patients fail to achieve a response or remission after being placed on initial therapy with an SSRI. Second-line pharmacologic treatment options include titrating the dose of the initial antidepressant, augmenting therapy with a second agent, or switching to another SSRI or an agent with a different mechanism of action, such as an SNRI. [28][29][30][31] The ARGOS study evaluated an SNRI, venlafaxine extended release XR, in patients who had failed to respond to or could not tolerate conventional antidepressants, primarily SSRIs, in a psychiatric outpatient setting. Those treated with venlafaxine XR had significantly higher remission rates (59.3%) at 24 weeks compared with those treated with conventional antidepressants (i.e., paroxetine, citalopram, sertraline, fluoxetine, mirtazapine, or other treatments) (51.5%; P <0.001). 21 Recent results from the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study showed that among patients who did not achieve remission with initial therapy, approximately one third achieved remission by augmentation with a second agent and about one fourth achieved remission by switching to a different antidepressant. 32 Current American Psychiatric Association (APA) practice guidelines suggest that if at least moderate improvement is not observed following 6 to 8 weeks of initial pharmacotherapy, the treatment regimen should be reevaluated. 33 nn Managed Care Strategies to Improve Depression Treatment Outcomes When depressed patients maintain continuous therapy for the recommended treatment duration, health care resource costs are reduced compared with patients who discontinue therapy early. 20,34 It then follows that overall costs should decrease when patient compliance is improved and agents associated with a reduced incidence of early discontinuation and therapy change are utilized early in the treatment program. 35 Managed care organizations (MCOs) are challenged to identify optimal, cost-effective strategies for the treatment of depression and improve patient adherence to antidepressant therapy. 36 MCOs have a limited role in direct patient care; therefore, they must be creative in developing programs or identifying treatment strategies that have the potential to influence patient adherence. Compared with standard care models, patient support programs (collaborative care model) that educate patients about the value of medication adherence, increase awareness of potential adverse events associated with antidepressant medications, and provide follow-up to ensure continued compliance were found to improve efficacy in the treatment of depression. [37][38][39][40] Mail-based educational intervention has also been shown to positively impact patient adherence to therapy. 36 When initial treatment is not effective or tolerable, APA treatment guidelines recommend that the clinician should consider treatment with another agent. While there is not a solid body of data to guide clinicians in decisions concerning secondline treatment options, the STAR*D trial evaluated 4 levels of treatment in patients who had not responded adequately to an initial standard antidepressant trial: level 1 (identify treatment-resistant patients), level 2 (switch and/or augment antidepressants), level 3 (switch to an agent with a different mechanism of action), and level 4 (treat with either an MAOI or venlafaxine XR plus mirtazapine). 32,[41][42][43] The trial results may help to define which subsequent treatment strategies, in what sequence, and in what combination(s) produce the best clinical results with the least side effects.
Data on switching and related resource utilization in managed care patients are limited. However, some evidence exists that an earlier switch (before 6 weeks of initial treatment) to an agent with an alternative mechanism of action may prevent unnecessary cycling. 44 Direct medical costs associated with switching antidepressants were recently reported in a poster presented at the Academy of Managed Care Pharmacy Educational Conference, October 5-8, 2006. Analysis of data from a national database (PharMetrics) of medical and pharmacy claims suggested that, overall, costs declined when patients switched antidepressant classes. Greater cost reductions due to reduced medical costs were realized when patients switched to an SNRI (venlafaxine) from an SSRI (citalopram, fluoxetine, paroxetine, sertraline) compared with switching to an SSRI from an SNRI (Figures 1  and 2). 44 In addition, higher costs were associated with patients who switched among multiple SSRIs before switching to an SNRI. 44 However, it is important to note that these findings may have been influenced by differences in the baseline characteristics of the patients involved.
Recently, an economic model was developed to explore the results of using generic SSRIs, such as escitalopram, paroxetine controlled release, sertraline, and venlafaxine XR as second-line agents for the treatment of unresolved depression following a treatment failure. Efficacy parameters used to develop the model were derived from clinical trial results based on the HAM-D and Montgomery-Asberg Depression Rating Scale (MADRS). 45,46 While managed care experts express a preference for the self-administered Patient Health Questionnaire. 47 or the Quality Improvement of Depression Scale) 48 as evaluation tools, clinical studies do not use those instruments to evaluate primary endpoints, hence the reliance on HAM-D and MADRS scores in the model.
The economic model, which was constructed in Microsoft Excel, is a budget-impact and decision analysis tool that allows the user to input managed care-specific information. Results of the analyses and a full description of the model are addressed in the next article of this publication.

nn Conclusion
When patients with depression fail initial therapy, the result is often increased costs to health plans and a poorer quality of life for patients. Because no clear data support the use of one agent over another as second-line therapy in patients with TRD, managed care providers could benefit from a model that they can customize with their own variables, such as acquisition costs and outcome parameters, to evaluate the overall costeffectiveness of different strategies.

Acknowledgments
The author would like to acknowledge Wendy Gloffke, PhD, medical writer, and the staff writer, Communications Impact, LLC, for their assistance with this article.

disclosures
The author serves as a consultant and/or on the speakers bureau for Wyeth Pharmaceuticals Inc., Cephalon, and Bristol-Myers Squibb. He received an honorarium for his participation in this study.   A Budget-Impact and Cost-Effectiveness Model for Second-Line Treatment of Major Depression

Daniel C. Malone, PhD
A s addressed in the preceding article, major depressive disorder (MDD) is a significant health problem with economic implications for patients, health insurance companies and other health care payers, and employers. Use of second-generation antidepressants has become the de facto standard for the treatment of MDD. Expenditures for antidepressants represented the third-largest drug therapy class in 2000. 1 Estimates of the economic burden of depression range from $52 billion in 1990 to $83 billion in 2000. 2,3 The burden of depression includes not only the direct medical costs of treatment, but also the social and intangible costs of missed worked, lower productivity, and decreased quality of life. [4][5][6][7] The risk of death is also much higher in patients with depression. 8 nn Antidepressant Therapy Assessing the Cost-Effectiveness of Antidepressant Therapy A number of issues arise when assessing the cost-effectiveness of antidepressant therapy. Among these issues are which agents to assess, how to measure their effectiveness, the duration of clinical trials versus the duration of analysis, which costs to consider, the perspective of the analysis, consideration of tolerability issues, and a multitude of clinical trial design options. Although it is beyond the scope of this article to discuss all of these issues in detail, it is important to consider some of them in the context of cost-effectiveness analysis.
Several depression rating scales have been developed, such as the Hamilton Rating Scale for Depression (HAM-D), the Montgomery-Asberg Depression Rating Scale (MADRS), the Beck Depression Inventory, and the Clinical Global Impression-Severity (CGI-S), that assess the clinical efficacy of various treatments for MDD. Although many rating scales exist for the purpose of evaluating antidepressant therapy, the vast majority of clinical trials have incorporated either the HAM-D or MADRS instruments. The HAM-D was first published in 1960 to assess the effectiveness of first-generation antidepressants. 9 A published review of the HAM-D concluded that the instrument has psychometric properties that are adequate and consistently meet established criteria for validity and reliability. 10 The MADRS instrument was first published in 1979 and is a commonly used instrument to assess the efficacy of antidepressants. 11 Both instruments (HAM-D and MADRS) have been widely used in clinical trials to evaluate the efficacy of antidepressant therapy.

Defining Response and Remission
The goal of depression treatment is remission-the virtual absence of depressive symptoms. Remission is associated with improved function and a more optimistic prognosis than is a response without remission. 12 Further, achieving remission BACkgrounD: Depressed patients who initially fail to achieve remission when placed on a selective serotonin reuptake inhibitor (SSrI) may require a second treatment.
oBjECTIvE: The purpose of this study was to evaluate the effectiveness, cost, cost-effectiveness, and budget impact of second-line pharmacologic treatment for major depressive disorder (MDD).
METhoDS: A cost-effectiveness analysis was conducted to evaluate secondline therapies (citalopram, escitalopram, fluoxetine, paroxetine, paroxetine controlled release [Cr], sertraline, and venlafaxine extended release [Xr]) for the treatment of depression. Effectiveness data were obtained from published clinical studies. The primary outcome was remission defined as a score of 7 or less on the hamilton rating Scale for Depression (hAM-D) or a score of 10 or less on the Montgomery-Asberg Depression rating Scale (MADrS) depression rating scales. The wholesale acquisition cost (WAC) for medications and medical treatment costs for depression were included. The perspective was derived from a managed care organization (MCo) with 500,000 members, a 1.9% annual incidence of depression, and treatment duration of 6 months. Assumptions included: second-line treatment is not as effective as first-line treatment, WAC price reflects MCo costs, and side effects were identical. Sensitivity analyses were conducted to determine variables that influenced the results.
rESuLTS: Second-line remission rates were 20.4% for venlafaxine Xr, 16.9% for sertraline, 16.4% for escitalopram, 15.1% for generic SSrIs (weighted average), and 13.6% for paroxetine Cr. Pharmacy costs ranged from $163 for generic SSrIs to $319 for venlafaxine Xr. Total cost per patient achieving remission was $14,275 for venlafaxine Xr, followed by $16,100 for escitalopram. The incremental cost-effectiveness ratio (ICEr) for venlafaxine Xr compared with generic SSrIs was $2,073 per patient achieving remission, followed by escitalopram with an ICEr of $3,566. The model was most sensitive to nonpharmacy costs.

Supplement to Journal of Managed Care Pharmacy S9
has been associated with ar eduction in the rate of relapse and improved psychological functioning compared with patients with partial response. [13][14][15][16] Studies indicate that compared with patients who achieve remission, those with incomplete resolution of symptoms are3to5times morelikely to relapse. 15,17 Consensus has suggested that values of 7o rl ess on the HAM-D arei ndicative of clinical remission. 18,19 For the MADRS instrument, cut-offv alues of 9o rl ess have been found to be approximately 1s tandardd eviation from the mean of nondepressed patients, 20 with various other cut-offv alues (9-12) having been used. 21 Consequently,many clinicians have come to accept that values of 10 or less arelikely to indicate remission.

Evaluating Antidepressant Therapy Effectiveness
Several meta-analyses and evidence reviews have been conducted to evaluate the effectiveness of selective serotonin reuptake inhibitor (SSRI) and serotonin and norepinephrine reuptake inhibitor (SNRI) products. 22,23 Most recently,RTI International conducted an evaluation of second-generation antidepressants and their comparative effectiveness for the treatment of depression. 24,25 The overall conclusion of the report was that therewas little evidence to indicate that SSRIs and SNRIs are clinically different from one another despite evidence of statistical differences. 25 The results suggested that escitalopram improved response over citalopram (relative risk [RR], 1.14; 95% confidence interval [CI], 1.04-1.26), sertraline improved response over fluoxetine (RR, 1.11; 95% CI, 1.01-1.21), and venlafaxine extended release (XR) improved response over fluoxetine (RR, 1.12; 95% CI, 1.01-1.24) as well as having higher rates of remission (odds ratio [OR], 1.42; 95% CI, 1.17-1.73). 24 The authors commented that these differences represented a relatively small difference on the HAM-D instrument and were likely to be clinically insignificant. Another aspect of the RTI report was that no overall analysis was reported, which is unfortunate given the substantial amount of data collected. In addition, the report focused on studies that used identical instruments, such as the HAM-D, to evaluate remission rates. This analytical decision prevents indirect comparisons between agents that might provide valuable information about their effectiveness.
One of the major difficulties of conducting such meta-analyses is that published reports sometimes fail to contain sufficient information to properly evaluate the study.F or example, ap roper meta-analysis will requiret he mean and standardd eviation of the treatment groups both at baseline and at the end of the study.I ti sn ot uncommon for authors to report only the standarddeviation at baseline but not at the end of the study.Also, many authors will report only the change from baseline scoreand not the scoreatthe end of the study.Another complicating issue when comparing studies is that some analyses will not report means, but only percentages of patients achieving some threshold. As mentioned earlier,s ome consensus exists on values of the HAM-D and MADRS that constitute remission, but not all studies use identical cut-off points. Therefore, studies that fail to report an end-of-study mean and as tandardd eviation cannot be used if ad ifferent threshold for remission is used.

Previous Economic Studies of Antidepressant Therapy
Ar ecent review of the literaturee valuated the economics of antidepressants and various other therapies for the treatment of MDD. 26 This review identified 18 manuscripts that evaluated the cost-effectiveness of antidepressant therapies, most of which werem odels. None of the analyses based on clinical trials compared second-generation antidepressants, although some economic models have compared SSRIs as ag roup to tricyclic antidepressants (TCAs). Results from the models evaluating these classes of antidepressants werem ixed, to some extent depending on how side effects and adverse effects weret aken into consideration. For example, aCanadian study reported no advantage of SSRIs versus TCAs with respect to cost-effectiveness. 27 Anumber of other studies have suggested that SSRIs are cost-effective relative to TCAs. [28][29] Other models have compared specific SSRIs and SNRIs. All of these models either stated explicitly or assumed that treatment was for first-line therapy.M any models included options for af ailed response to the first-line agent, but there was heterogeneity in the structureo ft he models with respect to second-line treatment options. Another important issue to highlight with economic models is the outcome of interest. While all studies have used some measureo fp atient-reported symptoms, the operationalization of these severity scales is not necessarily consistent. Some models have used response, while others have used remission as the outcome of interest. Other studies have converted performance on the HAM-D or MADRS instruments into symptom-free days (SFDs) or depression-free days (DFDs), which weree ssentially equivalent. Many of the economic studies reported to date derive estimates of clinical efficacy from as ingle study,w hich can be problematic if the clinical trial population does not reflect the overall population being treated. Abrief summaryofthe models and their findings is provided in some detail in the following section.
nn Summary of Economic Models Wade et al. compared citalopram and escitalopram using a 6-month probabilistic model from aU nited Kingdom (U.K.) perspective. 30 The study used a6-month time horizon, and the measureo fe ffectiveness was remission, defined as aM ADRS scoreo f1 2o rl ess without switching medications. The study found that escitalopram was morecost-effective than citalopram, with the overall remission rate for escitalopram being approximately 10% greater than citalopram while costs wereabout 16% lower for escitalopram. Using the Monte Carlo sensitivity analysis of 10,000 iterations, moret han 99% of the cases were situations in which escitalopram was ad ominant therapy (less cost and moreeffective). It should be noted that the prices of escitalopram and citalopram in the U.K. werei dentical; therefore, cost differences wered riven by decreased utilization of nondrug health carer esources for escitalopram relative to citalopram. In another study,Demyttenaeyreetal. reported on the costeffectiveness of citalopram, escitalopram, and venlafaxine using a2-stage model with a6-month time horizon. 31 The study was evaluated from the Belgium Insurance Scheme as well as from a societal perspective. The first stage of the model assumed that patients who did not have as uicide attempt weret reated with antidepressants and either achieved remission or could be switched to another agent or have the initial antidepressant dose titrated. Patients who had asuicide attempt wereentered into a different model whereby treatment options included switching therapies, augmentation, or titration. Clinical evidence for the model was based on 3c linical trials comparing 2o f3a gents directly (citalopram to escitalopram, escitalopram to venlafaxine XR). It was found that escitalopram was moreeffective and less costly than citalopram. Remission rates were52.8% for escitalopram versus 43.5% for citalopram. The societal cost of generic citalopram was € 0.85 (euros) for 20 mg per day; escitalopram was € 1.14 for 10 mg per day.R esults from the Monte Carlo sensitivity analysis found that in 93.5% of the cases, escitalopram was dominant when compared with citalopram. In contrast, the comparison with venlafaxine XR was equivocal. Clinical efficacy rates werenearly identical between the 2agents (escitalopram =6 9.9%, venlafaxine XR =6 9.7%). Results from the Monte Carlo sensitivity analysis suggested that escitalopram was morec ost-effective than venlafaxine XR in approximately 61% of the cases. 31 This result appears to be due to differences in medication costs, with venlafaxine being slightly more expensive than escitalopram.
Another study comparing citalopram with escitalopram was conducted by Hemels et al. 32 This analysis was conducted from an Austrian societal and Social HealthcareI nsurance System perspective using a6 -month time frame. Similar to previous models, the primaryoutcome of interest was remission of depression symptoms, and response was defined as a5 0% reduction in depression symptoms from baseline. Clinical efficacy was derived from clinical trials including both agents. Because the costs of the medications weren early identical, results from the model wered riven by statistical differences in clinical efficacy.Again, escitalopram was found to be adominant strategy regardless of the perspective taken. One-way sensitivity analyses suggested that the results werec onsistent across the range of parameters used in the model. Kulp et al. used aM arkov model to comparee scitalopram with venlafaxine XR for first-line treatment of depression from a German health insurance perspective. 33 The model included 3o utcomes: response (greater than 50% reduction from baseline), partial response (25% to 50% reduction from baseline), or no response. Clinical efficacy was derived from asingle randomized clinical trial comparing the 2a gents. For escitalopram, 77.4% of subjects werec onsidered responders to therapy compared with 79.6% for venlafaxine XR. Cost-effectiveness of the respective therapies was expressed in terms of proportion of patients responding to therapy.D rug costs wereb ased on defined daily doses as supplied by IMS Health Inc. and were € 1.30 for escitalopram and € 1.81 for venlafaxine XR. The duration of the study was 70 days. The results of the model suggested that the incremental cost-effectiveness of using venlafaxine XR was € 7,446 if general practitioners treated patients.
Similarly,Fernandez et al. reported on the cost-effectiveness of escitalopram relative to venlafaxine XR based on the multicenter and multicountryc linical trial used in the Kulp et al. analysis. 34 However,t his analysis was slightly different because the outcome of interest was quality of life as measured by the EuroQOL (quality of life) questionnaire( EQ-5D). Health care resource use was also captured, and health carecosts wereused from 6o ft he 8c ountries represented in the study,w hich was conducted from asocietal perspective. No statistically significant difference between the 2t reatment groups was found with respect to EQ-5D scores. Health careresource use varied by the type of health careprovider evaluated, but therewas no clearly consistent trend or difference between escitalopram or venlafaxine XR. Likewise, therew as no significant difference in total health carec osts between the 2g roups. To simulate cost-effectiveness, ab ootstrap sampling program was run 10,000 times. The results found that escitalopram was less costly in the majority of cases, but therewas no clear difference in quality of life.
In another model evaluating escitalopram, Sullivan et al. constructed a2-stage model of the treatment of depression that incorporated side effects of the various SSRI products on the market in the United States. 35 The model included patented and generic SSRI products with data on adverse events derived from the literatureorproduct package inserts. The time frame of the model was 6m onths, and data wered erived from am anaged careo rganization (MCO) perspective. The authors valued the costs of adverse events at moret han $5,000 based on ap reviously reported study. 36 The model used quality-adjusted life-years (QALYs) as the primaryoutcome of interest, adjusting QALYv alues downwardi nt he presence of adverse events. In this model, escitalopram was deemed to be the least costly ($3,891) and most effective (0.34) compared with citalopram, generic fluoxetine, venlafaxine XR, sertraline, generic paroxetine, paroxetine controlled release (CR), or venlafaxine. However,C Is for cost and effect wereo verlapping across all agents. The authors reported that in the analysis comparing escitalopram with fluoxetine, moret han 99% of the cases from a simulation of 10,000 werebelow the $50,000-per-QALYthreshold that is commonly used to define health technologies as cost-effective.
Using ac linical trial framework, Revicki  cost-effectiveness of pharmacotherapy (paroxetine or bupropion) or cognitive behavior therapy with community referral for major depression in women receiving carea tp ublic health or social service facilities over a1 2-month time frame in the Washington, DC, metropolitan area. 37 The primaryo utcome of interest was QALYs, which weree stimated from DFDs. To determine the number of DFDs over the course of the study,the following algorithm was used: HAM-D scoreo f7o rl ess =1 DFD; HAM-D scoreo f8t o2 1=p roportional weight to DFD; HAM-D of 22 or greater =0DFD. DFDs werethen converted to QALYs. Health carer esource use was obtained from patient charts for behavioral therapy and patient self-reports using structured telephone interviews. Costs of medical services were based on aM aryland Medicaid program fee schedule, and medication costs wereb ased on the lowest published average wholesale price (AWP). The study found that pharmacotherapy was moree ffective than community referral after 12 months. The cost per QALYw as $30,023 for pharmacotherapy and $37,568 for cognitive behavior therapy.R esults weren ot presented separately for paroxetine and bupropion.
Another model compared 3g roups of antidepressants with respect to cost-effectiveness. 38 In this model, SSRIs consisting of fluoxetine, paroxetine, and fluvoxamine werec ompared with SNRIs (venlafaxine) and TCAs (amitriptyline). The perspective of the model was derived from the U.K. National Health System with a6 -month time horizon. An expert panel of 3g eneral practitioners and 2psychiatrists assisted in the development of the decision model and provided estimates of probabilities for decision nodes wherec linical evidence was not available. Clinical efficacy was obtained from am eta-analysis of clinical trials for SSRIs and venlafaxine, as well as from asingle study for amitriptyline that was compared with am eta-analysis of TCA studies. The clinical parameters of interest werer emission, which was defined as ascoreof7orless on the HAM-D instrument. Remission rates werea lso converted into SFDs, but the exact algorithm was not provided. Overall remission rates were 45% for venlafaxine, 35% for SSRIs, and 24% for amitriptyline. The low response rate for amitriptyline was due to am uch higher drop-out rate than experienced with the other agents and to an intention-to-treat analysis that was used to calculate efficacy.T he number of SFDs over the 6-month study period was estimated to be 61 for venlafaxine, 52 for SSRIs, and 44 for amitriptyline. The cost per SFD was £21.20 (British pounds) for venlafaxine, £26.12 for SSRIs, and £31.80 for amitriptyline.
In another analysis involving venlafaxine, Trivedi et al. developed ac ost-effectiveness model based on am eta-analysis of 8c linical trials involving venlafaxine and SSRIs (fluoxetine, paroxetine, and fluvoxamine). 39 The meta-analysis was the same one used by Lenox-Smith et al. 38 and was conducted from the perspective of aU.S. MCO with an 8-week time horizon. As with the previous economic model, this analysis evaluated the percentage of patients achieving remission and converted this percentage into DFDs. The calculation of DFDs in this study was similar to the calculation of SFDs in other studies, whereby HAM-D scores of 7o rl ess =1D FD, values greater than 15 =0DFD, and scores between 8and 14 werelinearly weighted to equal values between 0a nd 1D FD. This model evaluated efficacy at 2-week intervals. Also, quality-adjusted days (QADs) werec omputed by assigning utility values to DFDs based on a previously published study. 40 Medication prices wereb ased on 2002 AWPs. Whereg eneric medications werea vailable, the lowest-priced product was used and weighted by an estimate of the generic to brand use of the product. Nonpharmacy costs werel imited to physician visits and laboratoryc osts. Results of the analysis found that venlafaxine had ah igher percentage of patients achieving remission (44.9%) than did SSRIs (34.7%); venlafaxine also had moreDFDs (22.82) than did SSRIs (18.61). The cost per patient achieving remission was $1,304 for venlafaxine and $1,515 for SSRIs. The cost per DFD was $25.66 for venlafaxine and $28.25 for SSRIs. AM onte Carlo simulation found the incremental cost-effectiveness ratio for SSRIs to have a95% CI that ranged from $326 to $1,176, with amean value of $586.
In summary, based on the studies conducted to date, SSRIs arec ost-effective relative to TCAs for treatment of MDD. Also, many studies conclude that treatment with products producing fewer side effects or those morerecently marketed appear to be cost-effective. However,am ajor limitation of these analyses is that they arelargely based on select clinical trials that tended to enroll naïve patients with MDD. No known economic models have examined treatment-resistant MDD.
nn Budget Impact of Antidepressant Therapies As mentioned previously,a ntidepressants represent the thirdlargest class of pharmaceuticals in the world in terms of dollar sales. 1 MCOs have recognized the need to appropriately manage these agents. Due to the availability of generic SSRI products, it makes sense for MCOs to encourage their first-line use because of their relatively similar efficacy but substantially lower costs. However,i fp atients fail initial therapy,c hoice of the second agent should incorporate such factors as reasons for failure, including side effects, nonresponse, and patient adherence. It is also important to keep in mind that the budget for antidepressants will include expenditures for the initial and treatmentresistant episodes. Because most health plans contain minimal data on reasons for failureo ft he initial agent, it becomes difficult to actively manage patients receiving antidepressant therapies. Health plans have limited alternatives for obtaining this data other than examining the pharmacy budget and estimating patient persistence to antidepressants by using pharmacy claims.
The Academy of Managed CareP harmacy (AMCP) and the Foundation for Managed CareP harmacy (FMCP) have written guidelines for formularym anagement explicitly stating that health plans should evaluate pharmaceuticals from ac osteffectiveness and budget-impact perspective. 41 Budget-impact analysis is an ecessarycomponent because it gives health plans the ability to anticipate futureexpenditures for pharmaceuticals that area dded to the formularya nd to explores trategies that might assist in controlling expenditures and improving health outcomes of the enrolled population. For these reasons, dossiers submitted to health plans should incorporate ab udget-impact assessment.
In an attempt to meet the real-world issues regarding the use of antidepressants, ac ost-effectiveness and budget-impact model was constructed and is outlined in the following section. Mores pecifically,t he purpose of this model was to examine the cost-effectiveness of pharmacologic treatment options for treatment-resistant MDD and to assess the overall budget implications for an MCO.
nn Methods

Model Description
Ab udget-impact and decision-analysis model was constructed in Microsoft Excel. Products included in the model were generic SSRIs consisting of citalopram, fluoxetine, and paroxetine; escitalopram (Lexapro); paroxetine CR (Paxil CR); sertraline (Zoloft); and venlafaxine XR (Effexor XR). Even though it is off-patent, fluvoxamine was not included in the model due to lack of clinical studies (see the following section) and its infrequent use. Venlafaxine immediate release was also not included in the model because it is moree xpensive that venlafaxine XR and provides no additional advantages over the extended-release formulation. Duloxetine was also not included in the study because, at the time the study was conducted, no evidence concerning its efficacy in treatment-resistant depression had been published. The model used an MCO perspective with a 6-month time horizon.

Model Structure
The model first takes into account the proportion of patients who failed first-line therapy based on estimates from clinical trials. The model assumes that if patients failed first-line treatment, they continued on to second-line therapy.F or those patients with an improvement (defined as response), they continued and an assessment of remission was determined.
For patients responding to therapy but not achieving remission, they entered 1o f3b ranches: switch, titrate, or augment therapy.T he use of generic SSRIs was assumed for patients who switched or augmented therapies. For titration, it was assumed that dose escalation was conducted, whereby the cost of treatment was increased but the clinical probabilities remained constant. Patients who did not have an improvement werec onsidered failures due to either al ack of efficacy or an adverse drug reaction (ADR). If patients had al ack of efficacy, treatment was switched, titrated, or augmented. For those who failed due to an ADR, it was assumed that they switched therapies (to another generic SSRI).

Cost and Market Share
The model begins with information on the health plan organization and prevalence of treated depression. The base model assumed ah ealth plan of 500,000 members with a prevalence rate of 1.9%. The model also assumed that the duration of treatment was 180 days and that medications would be supplied in 30-day increments. Consumer cost sharing was initially set at $10, $20, and $40 for first (generic), second (preferred branded SSRIs), and thirdtier (nonpreferred branded SSRIs), respectively.I nt his analysis, no agents werea ssigned into the thirdtier.
The model included the ability to alter various aspects of product use. Input parameters included the daily average consumption (commonly referred to as DACON); wholesale acquisition price (WAC); rebates, if any,e xpressed in terms of percentage offW AC; market shareb yp roduct strength; and copayment tier.T he model assumed that DACON for all products was constant and had av alue of 1, and therew as no rebate for any of the products. Medication prices wereobtained from aMedispan data file (September 2005). The market share by strength was taken from data obtained from IMS Health Inc. (December 2004). Aw eighted price for each product was calculated to reflect the market share. For example, citalopram is marketed in 10-mg, 20-mg, and 40-mg strengths. From IMS Health, Inc. data, it was determined that the 10-mg strength comprised 6.5% of all citalopram sales, whereas the 20-mg and 40-mg strengths comprised 53.2% and 40.3% of the market, respectively.T he WACp rice for each strength of citalopram was $0.64, $0.62, and $0.67 for the 10-mg, 20-mg, and 40-mg strengths, respectively.C onsequently,t he price of citalopram was estimated to be $0.64. The net plan cost for each agent was calculated by multiplying the weight price by the days of treatment, but subtracting out patient copayments. Distributions of market shareand price areshown in Table 1.
Generic SSRIs werea ssumed to be first-line treatment for MDD. The market sharefor first-line treatment was 20%, 40%, and 40% for citalopram, fluoxetine, and paroxetine, respectively. These market shares werea lso based on data from IMS Health Inc. If ap atient failed initial treatment, the model assumed another agent such as ab randed SSRI, ag eneric SSRI, or venlafaxine XR, which was then used. The respective market sharef or second-line use was 15% for citalopram, 15% for fluoxetine, 15% for escitalopram, 15% for paroxetine, 5% for paroxetine CR, 20% for sertraline, and 15% for venlafaxine XR. Thus, generic SSRIs accounted for 45% of second-line agents.
The base analysis was conducted by increasing the market shareofvenlafaxine XR from 15% to 20%; generic SSRIs would decrease by 3% and sertraline would decrease by 2%. The other products' market sharewould remain the same. Rather than use expert panels to estimate nondrug medical carecosts, such as physician visits, laboratorytests, and inpatient mental health care, the costs associated with remission, response, and treatment failurew ereb ased on an analysis of 1,814 persons enrolled in 10 antidepressant studies conducted by Group Health Cooperative in westernWashington state. 42 Clinical response to antidepressant therapy was categorized in an identical manner as the economic model. Medical carec osts weret hen estimated based on the endof-study status with respect to clinical outcomes.

Clinical Evidence
Clinical evidence for this study came from published trials of the agents of interest. The following criteria wereu sed to identify relevant studies: (1) randomized, comparative, double-blind, controlled clinical trials of SSRIs and SNRIs in the treatment of MDD that evaluated any of the following: citalopram, escitalo-pram, fluoxetine, sertraline, paroxetine, and venlafaxine XR; (2) depression was assessed with either the HAM-D or MADRS instruments; (3) aduration of at least 8weeks of therapy; (4) a sample size of greater than 30 patients per group; and (5) reported baseline mean and standardd eviation as well as end-of-study mean and standarddeviation. Studies wereidentified through searches of MEDLINE and PsycINFO databases using the following search strategies. First, studies involving the treatment of depression weres elected and werel imited to double-blind trials. Next, further restrictions wereimplemented to the medications of interest for this study (citalopram, escitalopram, fluoxetine, paroxetine, sertraline, and venlafaxine XR). For the PsycINFO database, searches werec onducted using terms related to depression or major depression. Studies were limited to clinical trials and to the medications of interest. Finally,M EDLINE and PsycINFO results werec ompared and duplicates removed. Clinical efficacy was determined by examining data for response and remission. Response was defined as a5 0% or greater improvement from the baseline depression rating score. Remission was defined as ascoreof7orless for the HAM-D or 10 or less for the MADRS instruments. For generic medications, remission and response rates werew eighted by market share, similar to medication pricing. The clinical evidence for paroxetine CR was based on studies including paroxetine and paroxetine CR. Studies of second-line treatment (treatmentresistant) for depression werel ess common, leading to relaxation of the inclusion and exclusion criteria. Treatmentresistant evaluations included both randomized-blinded and open-label studies.

Methods to Combine Clinical Studies
For each product of interest, data weres ummarized across all studies evaluating that specific agent by weighting the included studies by sample size. For those studies not reporting the percentage of patients who achieved remission or response, an ormal distribution was used to estimate the proportion achieving these thresholds. Because of the lack of data for parox-etine CR, it was assumed that estimates of clinical efficacy were similar to nonextended-release paroxetine.
Second-line response rates weree stimated by adjusting downwardr esponse and remission rates from initial treatment trials. The amount of the adjustment was obtained from clinical studies that examined the use of each agent in treatmentresistant depression.
Discontinuation rates werealso estimated from clinical trials. Similar to clinical efficacy rates, discontinuation rates were weighted by sample size to obtain ap ooled estimate for each agent.
Estimates of patients switching or titrating medications were obtained from ar etrospective analysis of pharmacy claims conducted by Verispan. 43 However,due to the time frame of that study,paroxetine CR and escitalopram werenot included. Therefore, it was assumed that titration and switch rates for paroxetine CR weres imilar to paroxetine, and the same rates for escitalopram werebased on citalopram.
nn Results Previous studies have found that the treated prevalence of MDD is 1.9%. Taking this into account, the proportion of patients who would undergo treatment in ah ealth plan with 500,000 individuals was estimated to be 9,500. This estimated population serves as the basis for our budget-impact model. The remission rate for first-line generic SSRIs was 35.5%, representing 3,371 enrollees with depression. Therefore, atotal of 6,129 enrollees with depression would be unsuccessfully treated and moved to asecond-line agent in this analysis.
Studies examining remission and response rates for treatment-resistant patients arefewer in number.For the SSRI agents, these products werei ncluded in an open-label study that examined second-line treatment of depression in an aturalistic manner compared with venlafaxine. 62 Several other studies werea lso available for citalopram and escitalopram, 63 fluoxetine and sertraline, 64 and venlafaxine. [65][66][67][68][69] Because these studies did not necessarily report sufficient information to adjust both remission and response rates, an assumption was made that the results werec onsistent for both remission and response. The estimated remission and response rates for second-line treatment of depression ares hown in Table 3. Relative to first-line treatment, the remission and response rates areall considerably lower than what was observed in studies evaluating treatment-naïve patients.

Model Outcomes
As mentioned earlier,the model projects that acohort of 6,129 persons would fail initial therapy and move to second-line treatment. Because market shared istribution for the various agents was unknown, it was assumed that all agents would have a15% market shareexcept for sertraline (20%) and paroxetine CR (5%). The distribution of patients and outcomes assuming this market shared istribution ares hown in Table 4. In this scenario, atotal of 1,187 patients would achieve remission, with an additional 1,185 having aresponse. Comparatively speaking, as shown in Table 5, morepatients receiving venlafaxine achieved remission (22.2%) than did generic SSRIs (18.5%), escitalopram (19.4%), paroxetine CR (17.7%), or sertraline (19.5%). Data in Table 5p resent results from the cost-effectiveness analysis, showing overall total costs and effectiveness for each agent and cost-effectiveness ratios. Therapy with the lowest total cost per patient was generic SSRIs with an estimated cost of $3,095. The next-lowest cost product was escitalopram with an estimated cost of $3,127, followed by venlafaxine XR ($3,172), sertraline ($3,178), and paroxetine CR ($3,206). In terms of cost per patient achieving remission, the agent with the most favorable ratio was venlafaxine XR ($14,275), followed by escitalopram ($16,100). Following the assumption that generic SSRIs arem ost commonly used for second-line treatment and maintain the lowest available cost, incremental cost-effectiveness ratios (ICERs) werecalculated using generic SSRI agents as the referent group (see Table 5). The lowest ICER was for venlafaxine ($2,073 per additional patient achieving remission), followed by escitalopram ($3,566) and sertraline ($8,613). Paroxetine CR dominated this analysis, meaning that it had higher cost and lower effectiveness than generic SSRIs.

Budget Impact Analysis
The AMCP Format for Formulary Submissions has promoted the use of both an economic model and abudget-impact analysis to accompany requests for formularyl isting. 41 Consistent with these guidelines, we conducted ab udget-impact analysis to determine the financial impact on ahealth plan by encouraging the use of cost-effective antidepressants for second-line treatment of depression. Due to alack of data about the market share for each of the respective antidepressants used in second-line treatment, as stated earlier,t he model assumed a1 5% market sharef or most therapies (citalopram, escitalopram, fluoxetine, paroxetine generic, and venlafaxine XR), with exceptions being a2 0% market sharef or sertraline and a5 %m arket sharef or paroxetine CR. As stated previously,anestimated 6,129 persons SSRIs=selective serotonin reuptake inhibitors. in a managed care plan with 500,000 enrollees would fail first-line treatment. For this cohort, approximately 1,187 would achieve remission after switching to another agent. The estimated pharmacy cost was estimated to be $1,341,246. Total costs, including medical and pharmacy for the entire cohort, was estimated to be $19,205,737. If there was a 5% increase in the market share of venlafaxine XR, the most cost-effective therapy, and a 3% decrease in generic SSRI use as well as a 2% decrease in sertraline use, the resulting pharmacy costs would increase by $19,000, but total costs would decline by $2,000 mainly due to lower costs associated with a greater number of persons who achieved remission (1,195). Similar findings occurred if the venlafaxine XR market share was increased and other brandname antidepressants had reductions in their market share.

Sensitivity Analysis
A series of 1-way and multiple-way sensitivity analyses were conducted to determine the impact of changing model parameters on the results. Using combined clinical data, 95% CIs were constructed for remission, response, and discontinuation rates.
Using the lower bound of the 95% CI for remission and response and the upper limit for discontinuation resulted in cost per remission increasing to $16,213 for venlafaxine, $18,352 for escitalopram, $18,663 for sertraline, and $18,799 for generic SSRIs. The ICER for venlafaxine XR relative to generic SSRIs increased from $2,073 to $2,605. If the lower limits for venlafaxine XR were changed, holding constant all of the other product response and remission rates, the resulting cost per patient achieving remission increased from $14,275 to $15,176, which is still lower than the other products evaluated in the model.

nn Discussion
This study examines the cost-effectiveness of SSRIs and venlafaxine XR for second-line treatment of MDD. Results suggest that the lowest cost option was generic SSRIs, but the agent with the highest clinical success was venlafaxine XR. The ICER for venlafaxine relative to generic SSRIs was $2,072 per patient achieving remission. This result is consistent with another cost-effectiveness analysis conducted by Trivedi et al. that compared SSRI agents as a group with venlafaxine XR. 39 The Trivedi et al. model used data from 8 double-blind clinical trials involving 2,045 patients. 39 In that study, the measure of success was a DFD, defined as having a HAM-D score of 7 or less. DFDs were then converted to QADs. The results found that persons receiving venlafaxine had a 44.9% remission rate compared with 34.7% for SSRIs. Economic analysis based on an 8-week period found that the cost for venlafaxine was $585 and for SSRIs was $526. The ICER for venlafaxine versus SSRIs was $586.08 per patient achieving remission, $14.20 per DFD, and $34.55 per QAD.
Numerous cost-effectiveness analyses have compared 2 or 3 select agents, with many involving venlafaxine. 28,[30][31][32][33][34]70,71 These analyses do not consider the role of generic SSRIs, which are becoming more prominent with products going off patent. Those economic analyses that have included generic SSRIs assume first-line treatment of MDD. 34,37 As more generic SSRIs enter the market, it will become increasingly important to consider these market changes when conducting costeffectiveness analyses in MDD.
Several limitations to this study should be noted. Clinical trial results were not drawn from head-to-head studies, although every clinical trial included in the study was assessed against an active comparator to minimize selection bias. Unfortunately, the published literature contains only a few studies for treatmentresistant MDD. The majority of clinical information for treatment-resistant effectiveness is derived from large open-label studies that trade greater external validity for lower internal validity. The literature is much richer concerning treatment of initial depression due to the design of clinical studies for purposes of registration and regulatory approval. More studies of treatment-resistant depression in real-world settings are needed. In addition, lack of inclusion of all potential treatments in existing studies of treatment-resistant depression limits the present study, especially with the exclusion of duloxetine. It is important to note that in this study, it was assumed that patients who failed initial therapy were transitioned into a second pharmacologic treatment. It is logical to question this assumption, but no known data could be used to estimate this transition probability.
Another limitation of this study is that the analysis did not explicitly consider differences in side effect profiles between the various agents. The results may change if substantive differences in side effect rates result in premature discontinuation or additional consumption of health care services. Another limitation is the lack of data concerning the market share for each of the respective agents as second-line therapy. We estimated a 5% increase in venlafaxine XR use for illustrative purposes in the budget-impact analysis. Actual changes in market share will vary between health plans depending on local market conditions. Another limitation is the use of WAC pricing, which may not reflect actual costs to MCOs. Finally, the analysis does not take other treatments into account, such as counseling, which may be effective for treating depressed patients.

nn Conclusion
Patients with depression often fail initial therapy, resulting in higher costs to health plans and a poorer quality of life. This analysis examines the issue of second-line treatment for depression by comparing brand-name and generically available products and taking persistency into account. Using a budgetimpact and cost-effectiveness model, this study found that increasing the market share of venlafaxine XR is likely to result in overall savings to a health plan because more patients achieve