Cost effectiveness of duloxetine in the treatment of fibromyalgia in the United States

Abstract

Objective:

To evaluate the cost effectiveness of duloxetine when considered as an alternative treatment for patients in the United States (US) being treated for fibromyalgia pain.

Research design and methods:

A Markov model was used to evaluate the economic and clinical advantages of duloxetine in controlling fibromyalgia pain symptoms over a 2-year time horizon. A base-case treatment sequence was adopted from clinical guidelines, based on tricyclic antidepressants, serotonin-norepinephrine reuptake inhibitors, anticonvulsants, and opioids. Treatment response was modeled using changes from baseline in pain severity, and response thresholds: full response (at least a 50% change), response (30–49% change), and no response (less than a 30% change). Clinical efficacy and discontinuation data were taken from placebo- and active-controlled trials identified in a systematic literature review and mixed-treatment comparison. Utility data were based on EQ-5D data.

Main outcome measures:

Additional symptom-control months (SCMs), defined as the amount of time at a response level of 30% or less, and quality-adjusted life-years (QALYs) over a 2-year time horizon.

Results:

For every 1000 patients, first-line duloxetine resulted in an additional 665 SCMs and 12.3 QALYs, at a cost of $582,911 (equivalent to incremental cost-effectiveness ratios [ICERs] of $877 per SCM and $47,560 per QALY). Second-line duloxetine resulted in an additional 460 SCMs and 8.7 QALYs, at a cost of $143,752 (equivalent to ICERs of $312 per SMC and $16,565 per QALY).

Limitations:

Response data for TCAs are limited to 30% improvement levels, reported trials are small, and have low placebo response rates. The model necessarily assumes that response rates are independent of placement in the treatment sequence.

Conclusions:

The results suggest that the introduction of duloxetine into the standard treatment sequence for fibromyalgia not only provides additional patient benefits, reflected by time spent in pain control, but also is cost effective when compared with commonly adopted thresholds.

Key words::

• Cost effectiveness
• Duloxetine
• Fibromyalgia
• Model
• United States
• US

Introduction

Fibromyalgia (FM) is one of the most common causes of chronic, widespread, musculoskeletal pain, and the condition can lead to significant levels of functional disability1–4. The symptom burden for patients with FM is centered on the presence of widespread pain, but FM also can lead to much broader and variable symptoms, typically involving fatigue, cognitive problems, morning stiffness, sleep disturbance, irritable bowel, headache, and depression and/or anxiety4–6. FM also has been associated with significant reductions in formally assessed health-related quality of life (HRQoL) compared with the standard population norms, with levels of HRQoL similar or lower than those seen for rheumatoid arthritis, osteoarthritis, or osteoporosis7.

FM typically occurs in adults aged between 45 and 60 years, although it can appear in adult patients in other age groups. Further, FM is a condition that is seen predominantly in females, who represent over 90% of diagnosed cases8,9. The annual incidence of FM has increased rapidly over the last 10 years, with an estimated annual prevalence in the United States (US) of 2%8,9. This observed increase in incidence cases is thought to be due, at least in part, to the increased ability to successfully diagnose cases. Despite this, a formal clinical diagnosis of FM remains difficult to achieve and is typically based on the published set of America College of Rheumatology (ACR) symptom criteria, defined as widespread pain of at least 3 months’ duration and pain on palpation in at least 11 of 18 paired tender points5.

The published clinical trials of treatments for FM have focused primarily on the effective control of pain-related symptoms, and this has led to a move to standardize the methodologies used to assess baseline and endpoint pain severity. The Brief Pain Inventory (BPI) is a generic pain assessment instrument that has been consistently applied in recently published FM-based studies10,11. With the BPI, patients are asked to score their pain, as experienced over the last 24 hours, on a scale of 0 to 10 (i.e., over 11 discrete points). Another instrument, the Fibromyalgia Impact Questionnaire (FIQ), is a disease-specific tool that can be used to assess the overall symptom burden of patients with FM12,13. The FIQ has a specific pain-score item that is similar in construct to the BPI (rated on a scale of 0–10) but based on a patient’s self-assessment of pain during the previous 7 days. The FIQ also can be used to assess wider symptom dimensions outside of pain, providing an overall total symptom score that is typically expressed on a scale of 0 to 80. Using one of these 11-point pain severity scales (the BPI, a visual analog scale, or the FIQ), a score of at least 4 typically is used in clinical trial inclusion criteria to define patients with moderate to severe levels of pain.

Several medications, such as tricyclic antidepressants (TCAs), selective serotonin reuptake inhibitors, serotonin norepinephrine reuptake inhibitors (SNRIs), and anticonvulsants, have demonstrated efficacy in treating the symptoms related to FM1,2,4. Currently, only a limited number of internationally recognized clinical guidelines for the management of FM have been developed by health services, clinical, or professional associations such as the American Pain Society (APS), the European League Against Rheumatism, or the British Society for Rheumatology14. The APS guidelines of 2005 recommend TCAs and their analogs as the preferred primary drug treatment option for fibromyalgia, although TCAs are not specifically licensed for the treatment of fibromyalgia15. As the most frequently used medications in the treatment of fibromyalgia, TCAs are effective in providing analgesic effect, aiding sleep, and treating concomitant mood disorders16. However, TCAs also are associated with side-effects that can limit their use in some patients, including dry mouth, constipation, blurred vision, fatigue, and low blood pressure. Selective serotonin reuptake inhibitors can play a role in improvement of mood and have an improved side-effect profile, but they appear to have little impact on pain17. The following treatments recently were approved by the US Food and Drug Administration (FDA) for FM pain: pregabalin (an anticonvulsant), approved in 2007; duloxetine (an SNRI), approved in 2008; and milnacipran (an SNRI), approved in 2009.

There are only limited published economic data for treatments of FM, and no formal US-based cost-effectiveness models or economic evaluations of approved FM treatments in the current literature. A United Kingdom (UK)-based study, published since the development of our model, considered the cost effectiveness of pregabalin18 in a nonsequence treatment-based context. The study claimed pregabalin was cost effective over placebo with ICERs in the range £22,500–23,500 (approximately $36,000–37,000) per quality-adjusted life-year (QALY). Published US-based cost burden data are available from before- and after-diagnosis cohort studies19–23 but these studies were intended to explore only the cost burden and differences that may be associated with diagnosed and undiagnosed FM. Only one published study of FM, conducted in Spain, presents cost data formally stratified by recognized pain severity scoring systems24. The study demonstrated that the healthcare costs of FM are clearly differentiated by disease severity as assessed using the BPI scale, with the annual direct cost of FM increasing linearly by 115 euro for every one-step increase in BPI score.

Evaluating an economic profile is a critical step in establishing the overall value of a new health technology, and this type of evaluation should be conducted within the context of both current standard therapy and treatment alternatives. The aim of this study was to conduct an economic evaluation of the cost effectiveness of duloxetine when considered as an alternative treatment option for US patients with pain related to FM. The evaluation was performed using a decision-analytic modeling approach and reflected a typical treatment sequence for fibromyalgia. The model assessed direct healthcare resource use and cost from the perspective of a US healthcare payer. The following specific research questions were considered:

• What were the incremental costs and clinical benefits associated with the range of alternative duloxetine-based pharmaceutical treatment strategies for FM, when considered from a typical US-based treatment perspective?

• What was the overall cost effectiveness of duloxetine-based treatment sequences for FM?

• What was the optimal positioning of duloxetine (first-line, second-line, third-line, etc.) within the treatment sequence, from a clinical and cost-effectiveness perspective?

Patients and methods

A MEDLINE search (conducted in July 2009) of the published literature identified no published economic models to use as the basis for an evaluation of cost effectiveness for FM treatment alternatives. We therefore developed a decision-analytic model framework from current clinical FM guidelines1,15 and from the health technology assessment reference case technical requirements for formal economic evaluations as set out by the National Institute for Health and Clinical Excellence in the UK25. A Markov health-state transition approach was used to consider the costs and health benefits of alternative duloxetine-based treatment strategies for the management of FM-related pain symptoms over a 2-year time horizon. The 2-year period corresponded to the time needed to sequence patients through the full range of possible treatment options, with the overall treatment objective of optimizing levels of pain response and thus minimizing pain severity and associated reductions in HRQoL.

Patient definition

The population cohort in the model consisted of adult patients (average age of 50 years) who were eligible for pharmacotherapy and who had received a clinical diagnosis of FM by fulfilling the following 1990 ACR classification criteria26: at least a 3-month history of symptoms (which may include pain with additional sleep disturbance, headache, fatigue, morning stiffness, bowel or bladder symptoms, and/or depression or anxiety), the presence of at least 11 of 18 ACR-recognized tender points, and a minimal BPI or FIQ pain score of at least 4 on an 11-point scale (a moderate-to-severe level of pain).

Analysis perspective

The perspective of the economic analysis was that of a US healthcare payer and provider. Resource use and cost data were limited to direct healthcare related to FM, including drug treatments, hospitalizations, outpatient services, and physician contact for general follow-up care and supervision, throughout the modeled treatment period. The wider social impacts of FM on patients and their families (e.g., supportive care at home, adaptations to the home, or reduced productivity) were considered in the sensitivity analyses.

Form of economic analysis

The model was designed to perform a standard set of cost-effectiveness analyses for a disease-specific outcome, using an incremental cost-effectiveness ratio (ICER) of cost per additional symptom control month (SCM), where clinical benefit was defined as additional time spent in adequate pain symptom control (based on a ≥30% improvement from baseline pain severity).

The model also was designed to calculate a standard ICER using a cost per quality-adjusted life-year (QALY) as a cost-utility analysis. In this case, the model based the assessment of clinical benefit on the additional time spent by patients in pain response health states; these states were converted to an underlying pain severity score, each having an associated utility weight.

Example of ICER calculations comparing sequence A versus B:

Although a direct comparison to a current treatment strategy is necessary in all economic evaluations, relative comparisons against all possible treatment options, where multiple treatment options exist, are equally informative for decision makers. These comparisons provide a more appropriate view of the true incremental costs and benefits and allow decision makers to consider the full value of additional investments in treatments that may provide further benefits (e.g., to consider first-line vs. second-line duloxetine). Our economic model was designed to present relative treatment comparisons through the use of a cost-effectiveness frontier plot (which places each treatment comparator/sequence on one chart, with an x-axis of incremental clinical benefit and a y-axis of incremental cost over base case).

Model health-state structure

The model structure was designed to represent the movement of patients through a series of treatment levels (first line, second line, third-line, etc.) within the context of a predefined treatment sequence (Table 1). In Table 1 each row represents a treatment sequence, and each column is a position in the sequence (the first row is the base-case sequence where duloxtine is not available). This allowed the model to be used to compare duloxetine strategies with a standard sequential treatment strategy for FM from the perspective of typical US clinical practice. It was important that the evaluation consider all the subsequent treatment options because these can affect the overall cost-effectiveness profile of the new treatment.

Table 1.  Modeled treatment sequences.

Sequence identifier	Treatment sequence (covering up to 6 lines of treatment)
	Line 1	Line 2	Line 3	Line 4	Line 5	Line 6
X) Base case	TCA	SNRI	ANTI	TRAM	PRAM
(A) First line	DUL	TCA	SNRI	ANTI	TRAM	PRAM
(B) Second line	TCA	DUL	SNRI	ANTI	TRAM	PRAM
(C) Third line	TCA	SNRI	DUL	ANTI	TRAM	PRAM
(D) Fourth line	TCA	SNRI	ANTI	DUL	TRAM	PRAM
(E) Fifth line	TCA	SNRI	ANTI	TRAM	DUL	PRAM
(F) Sixth line	TCA	SNRI	ANTI	TRAM	PRAM	DUL

ANTI, anticonvulsant; DUL, duloxetine; PRAM, pramipexole; SNRI, serotonin norepinephrine reuptake inhibitor; TCA, tricyclic antidepressant; TRAM, tramadol.

A standard Markov health-state transition methodology (using standard version of Microsoft Excel 2003) was used to track patients through a series of health states, delineated by predefined levels of improvement in baseline pain severity score (Figure 1). A 3-month cycle was used for the regular transition calculations, in accordance with the primary response assessment point used in the duloxetine clinical studies, and in line with expected clinical practice. Pain response was estimated from published data for duloxetine and its comparators, which consisted of assessments using comparable 11-point pain assessment scales (the BPI, a visual analog 100-mm scale, or the FIQ). Two levels of definition of pain response were used to define and drive health outcomes in the model:

• At least a 30% improvement in baseline pain severity score (interpreted as a minimal level of clinical improvement)27,28,

• At least a 50% improvement in baseline pain severity score (a significantly higher level of pain response and improved HRQoL)27,28.

Figure 1.  Markov health states and transition routes. Where the patient fails to achieve at least a partial response, or experiences an intolerable AE, they then are switched to the next treatment in the treatment sequence. AE, adverse event.

The proportion of patients achieving the predefined ≥30% and ≥50% levels of pain improvement (assessed from baseline pain severity scores) were used to separate the cohort into three discrete pain-response health states. The possibility of experiencing intolerable treatment-related adverse events also was considered. This resulted in five clinically important health states (Figure 1): full response (≥50% improvement in pain severity), with and without adverse events; partial response (30–49% improvement in pain severity), with or without adverse events; and inadequate response (<30% improvement in pain severity).

Where patients fail to achieve at least a partial response, or experiences an intolerable AE, they are then switched to the next treatment in the treatment sequence.

The model assumed average pain improvements, applied during the first 3-month treatment period, for full responders (70%) and partial responders (38%); these assumptions were based on the duloxetine clinical trial datasets and were applied equally for all treatments. Starting with an average baseline pain score of 6.5, as seen in the duloxetine clinical studies, the model assumed the majority of improvements would be achieved within the first 2 weeks of treatment; the treatment effect reached a plateau at 3 months (as seen with duloxetine response in clinical trials). The economic analysis explores variation in the timing of response, from early to late in the initial 3-month period of each treatment.

Beyond the 3-month point, the model assumed that patients could maintain levels of pain response during the 2-year time horizon, provided they remained on active treatment. This assumption was supported by the 6-month duloxetine trial data, which suggested similar levels of maintained response at both the 3-month and the 6-month follow-up points29,30. The assumption of maintained response was applied equally for all the active treatments considered in the model.

The model assumed that patients who moved through the full treatment sequence and remained in an inadequate response health state continued to experience their baseline pain severity. The evidence to support this assumption came from the pooled data for patients classified as having an inadequate response in the duloxetine trials29–31. Patients in the trials with inadequate response (a < 30% improvement) experienced an average 2% worsening of pain severity over a 3-month treatment duration (i.e., effectively no change in pain severity).

Drug treatment comparators and sequences

In developing an economic model suitable for formal health technology assessment with applicability to a managed-care organization, the expectation was to directly compare costs and benefits against currently licensed treatments for FM that typically would be seen as representing the existing standard treatment option(s). However, current US clinical treatment practices for FM are extremely variable and are dominated by treatments that are not specifically licensed for this indication (for example, the widespread use of generic TCAs as a first-line treatment in FM). Therefore, in defining a common standard of care comparator arm it is important to recognize both licensed and commonly used off-label treatments.

The definition of the model’s base-case treatment sequence was taken from the 2005 APS guidelines and was supported by discussions with US-based clinicians (Table 1). The sequence consisted of a first-line TCA, a second-line SNRI, and a third-line anticonvulsant. The use of TCAs, SNRIs, and anticonvulsants reflected the most commonly prescribed drug categories for treating FM in the US. In the base-case sequence, first-line treatment was most typically amitriptyline (25–75 mg per day); second-line treatment was milnacipran (200 mg per day); and the third-line treatment was the anticonvulsant pregabalin (450 mg per day). Milnacipran and pregabalin are FDA-approved treatments for FM. The base-case treatment sequence also had anticonvulsant nonresponders move to a fourth-line treatment, tramadol, and a final fifth-line treatment, pramipexole.

The economic model then was used to consider a range of alternative placements of duloxetine into the predefined standard base-case treatment sequence (Table 1).

Clinical efficacy evidence

Three, pivotal, randomized clinical trials specifically evaluated the clinical efficacy of the FDA-approved, 60-mg, once-daily dose of duloxetine29–31. These trials assessed treatment response to FM-related pain over a 3–6-month follow-up period, with the 3-month response seen as the most critical assessment point. All trials used a placebo control arm29–31. We programmed a full mixed-treatment comparison (MTC) statistical model to generate direct relative response rates (for both the ≥30% and the ≥50% improvement-defined response rates) against placebo (Table 2) by pooling the common-comparator placebo response rates from the duloxetine studies29,31. The use of an MTC analysis32 to generate direct relative risks was selected because it allowed us to pool not only the relative effect size but also to consider the scale of the absolute placebo rate in each study, thus avoiding a potential over-reliance on abnormally low absolute placebo rate data.

Table 2.  Clinical response input data.

Treatment	Class	Response (30% defined)		Response (50% defined)		PSA distribution	Data source
Drug dose		Rate (%)	RR (95% CI)^a	Rate (%)	RR (95% CI)*
Placebo		34.6	Reference	23.3	Reference	Reference	Pooled from trials
Duloxetine 60 mg	SNRI	56.8	1.64 (1.32–2.04)	42.5	1.82 (1.32–2.50)	Lognormal (95% CI)	Arnold et al., 2005; Russell et al., 200829,31
Amitriptyline	TCA	40.9	1.18 (0.41–2.98)	30.6	No data	Lognormal (95% CI)	Goldenberg et al., 199635,†
Milnacipran 100 mg	SNRI	45.4	1.31 (1.09–1.59)	35.3	1.51 (0.99–2.03)	Lognormal (95% CI)	Clauw, 2008; Mease et al., 200936,37
Milnacipran 200 mg	SNRI	48.1	1.39 (1.18–1.66)	35.5	1.52 (1.07–2.14)	Lognormal (95% CI)	Clauw, 2008; Mease et al., 2009; Gendreau et al., 200536–38
Pregabalin 300 mg	ANTI	47.1	1.36 (1.14–1.64)	30.1	1.29 (0.89–1.89)	Lognormal (95% CI)	Crofford et al., 2005; Arnold et al., 2008; Mease et al., 20083,39,40
Pregabalin 450 mg	ANTI	53.7	1.55 (1.30–1.86)	37.1	1.59 (1.13–2.31)	Lognormal (95% CI)	Crofford et al., 2005; Arnold et al., 2008; Mease et al., 20083,39,40
Tramadol 300 mg	OPIOID	56.1	1.62 (1.16–2.36)	44.6	1.91 (1.23–3.07)	Lognormal (95% CI)	Bennett et al., 200341
Pramipexole	NEDA	65.0	No data	48.6	2.08 (1.14–4.00)	Lognormal (95% CI)	Holman, 200542,‡

*RR data and CI taken from the results of a mixed-treatment comparison.

†Limited data on pain response: one trial available and limited to 30% response data; no published data for 50% response; default estimated based on duloxetine 50%–30% response ratio of 42.5/56.8.

‡Limited data on pain response: 30% response estimated based on duloxetine 30–50% response ratio of 56.8/42.5.

ANTI, anticonvulsant; CI, confidence interval; NEDA, nonergot dopamine agonist; PSA, probabilistic sensitivity analysis; RR, relative risk; SNRI, serotonin norepinephrine reuptake inhibitor; TCA, tricyclic antidepressant.

Relative response rates (including 95% confidence intervals) for each treatment comparator were applied to the pooled placebo data to generate a set of adjusted ≥30% and ≥50% response rates, using a standard, adjusted, indirect analysis methodology33,34. This approach standardized the indirect response data to the duloxetine trials’ placebo control arms (Table 2). A similar approach was taken to include an assessment of drop-out rates due to adverse events (Table 3).

Table 3.  Adverse event drop-out rate input data.

Treatment	Class	AE drop-out rate		PSA distribution	Data source
Drug dose		Rate (%)	RR (95% CI)*
Placebo		11.7	Reference	Reference	Pooled from trials
Duloxetine 60 mg	SNRI	22.1	1.88 (1.30–2.67)	Lognormal (95% CI)	Russell et al., 2008; Arnold et al., 200529,31
Amitriptyline	TCA	8.7	0.74 (0.48–1.13)	Lognormal (95% CI)	Goldenberg et al., 1996; Bennett et al., 1988; Carette et al., 1986; 1994; 1995; Heymann et al., 2001; Quimby, 1989; Scudds et al., 198935,43–49
Milnacipran 100 mg	SNRI	20.7	1.76 (1.32–2.37)	Lognormal (95% CI)	Clauw, 2008; Mease et al., 200936,37
Milnacipran 200 mg	SNRI	25.6	2.18 (1.69–2.84)	Lognormal (95% CI)	Clauw, 2008; Mease et al., 2009; Gendreau et al., 200536–38
Pregabalin 300 mg	ANTI	20.5	1.75 (1.29–2.37)	Lognormal (95% CI)	Crofford et al., 2005; Arnold et al., 2008; Mease et al., 20083,39,40
Pregabalin 450 mg	ANTI	26.9	2.29 (1.73–3.04)	Lognormal (95% CI)	Crofford et al., 2005; Arnold et al., 2008; Mease et al., 20083,39,40
Tramadol 300 mg	OPIOID	25.0	2.13 (1.32–3.45)	Lognormal (95% CI)	Bennett et al., 200341,†
Pramipexole	NEDA	11.7	1.00 (n/a)	Lognormal (95% CI)	Assume placebo‡

*RR data and CI taken from the results of a mixed-treatment comparison.

†Tramadol drop-out rates are problematic: the single trial has very low drop-out rates for placebo and treatment, so there is great uncertainty.

‡Set to placebo level because Holman42 does not report drop-out rates.

ANTI, anticonvulsant; CI, confidence interval; n/a, not applicable; NEDA, nonergot dopamine agonist; PSA, probabilistic sensitivity analysis; RR, relative risk; SNRI, serotonin norepinephrine reuptake inhibitor; TCA, tricyclic antidepressant.

The calculated response rates (Table 2) and drop-out rates (Table 3) were then used as direct inputs to the Markov model.

The lack of large clinical datasets for some of the comparator treatments could clearly be seen in the wider confidence intervals identified for both the relative risks of response and drop-out rates, particularly for TCAs (amitriptyline) and SNRIs (milnacipran) (Tables 2 and 3).

Long-term treatment adherence

The model allowed for a proportion of patients to drop out from current treatment in each 3-month model cycle following the initial 3-month treatment period. These patients were switched to the next treatment in the sequence. Pooled outcome data from two, longer-term, 6-month, randomized clinical trial extensions for duloxetine showed that patients who entered the extension period (i.e., those already remaining on treatment for at least 6 months) experienced a 12.5% adverse event-related treatment drop-out rate in the subsequent 6 months of therapy50. This was equivalent to an annualized rate of approximately 25%. A conservative assumption was taken so that all the treatments had the same expected level of long-term treatment adherence (Table 4). This annualized rate parameter value was explored across a 20–30% range in the sensitivity analyses.

Table 4.  Additional model parameter data.

Clinical parameter	Value	PSA distribution	Source
Long-term drop out rate	25% per annum	Uniform (±20%)	Choy et al., 200950
Baseline BPI	6.5	Not applied	Arnold et al., 2005; Russell et al., 200829,31
BPI change: full responder	Baseline + 70%	Beta (SE = 0.008)	Arnold et al., 2005; Russell et al., 200829,31
BPI change: responder	Baseline + 38%	Beta (SE = 0.003)	Arnold et al., 2005; Russell et al., 200829,31
BPI change: inadequate	Baseline + 0%	Not applied	Arnold et al., 2005; Russell et al., 200829,31
Utility by pain severity score
BPI = 0	0.90	Beta (SE = 0.02)	Russell et al., 2008; Chappell et al., 200929,30
BPI = 1	0.79	Beta (SE = 0.02)
BPI = 2	0.74	Beta (SE = 0.01)
BPI = 3	0.68	Beta (SE = 0.01)
BPI = 4	0.63	Beta (SE = 0.01)
BPI = 5	0.61	Beta (SE = 0.01)
BPI = 6	0.52	Beta (SE = 0.01)
BPI = 7	0.40	Beta (SE = 0.02)
BPI = 8	0.25	Beta (SE = 0.02)
BPI = 9	0.12	Beta (SE = 0.03)
BPI = 10	0.05	Beta (SE = 0.03)
Treatment switching
Primary care physician visit	46%	Uniform (±20%)	Penrod, 200454
Specialist outpatient visit	54%	Uniform (±20%)	Penrod, 200454
Primary care physician visit	$66.87 per visit	Not applied	CPT code 99211-553
Specialist outpatient visit	$135.03 per visit	Not applied	CPT code 99241-553
Nondrug/societal cost
Nondrug: full response	$1615 cost per year	Uniform (±20%)	White et al., 2009; Sicras-Mainar et al., 2009; White et al., 200822,24,55
Nondrug: response	$2017 cost per year	Uniform (±20%)
Nondrug: inadequate	$2496 cost per year	Uniform (±20%)
Societal: full response*	$173 cost per year	Uniform (±20%)
Societal: response*	$849 cost per year	Uniform (±20%)
Societal: inadequate*	$1654 cost per year	Uniform (±20%)
Drug (target daily dose)
Duloxetine (60 mg)	$4.00 cost per day	Not included	WAC52
Amitriptyline (75 mg)	$0.10 cost per day	Not included
Milnacipran (100 mg)	$3.38 cost per day	Not included
Milnacipran (200 mg)	$3.38 cost per day	Not included
Pregabalin (300 mg)	$4.28 cost per day	Not included
Pregabalin (450 mg)	$4.28 cost per day	Not included
Tramadol (300 mg)	$1.04 cost per day	Not included
Pramipexole (4.5 mg)	$5.46 cost per day	Not included

*Societal costs were applied in the PSA and were excluded from the base-case evaluations.

BPI, Brief Pain Inventory; CPT, current procedural terminology; PSA, probabilistic sensitivity analysis; SE, standard error; WAC, wholesale acquisition cost.

Health utility data

QALYs were calculated in the model and were based on the estimated time spent by patients in alternative levels of pain response and pain severity. The model linked utility weight data to an 11-point pain severity scale, using the generic EQ-5D assessment instrument and standard UK tariffs, in line with recommendations from the National Institute for Health and Clinical Excellence25,51. The EQ-5D was applied to patient data taken at the initial and final assessments in two pivotal duloxetine clinical trials, which provided a dataset of 2144 individual utility-pain severity assessments29,30.

The model was based on univariate mean average utility values by the 11-point pain score (Table 4). However, as part of the sensitivity analysis, a linear-based regression with fixed- and random-effects analyses also was used to consider alternative modeled utility-pain score mapping derived from the patient-level data. The EQ-5D results showed a strong negative correlation between pain severity and HRQoL, which was a robust finding when limited to final assessment data only and when stratified by treatment arm.

Resource and costing data

For the base-case analysis, the model used a final target dose of 60 mg for duloxetine, 75 mg for amitriptyline, 200 mg for milnacipran, and 450 mg for pregabalin (Table 4). The dose levels were taken from the recommended and approved dosages, clinical trial designs, and clinical opinion. The drug treatment costs are the Wholesale Acquisition Cost (WAC) prices taken from the Red Book52. The model allowed for titration to the target dose, where appropriate, when calculating overall drug costs.

Treatment switching costs were based on one primary care physician visit and/or one outpatient specialist visit and used Current Procedural Terminology codes 99211-5 and 99241-5, respectively53 (Table 4).

For a consideration of the wider nondrug clinical management costs of FM, the model was populated with US direct costs from a published study22. In the absence of further appropriate data for the US, these costs then were broken down by pain severity by applying a relative cost index by severity, as seen in a recent Spanish study24 (Table 4).

The cost year was set at 2009. A US-based annual discounting rate of 3.0% per year was applied to costs and benefit outcomes; sensitivity analyses used alternative 0% and 6% rates, derived from the AMCP Format for Formulary Submissions, Version 3.0 (AMCP, 2009)56.

Sensitivity analyses

To assess the potential impact of parameter and data uncertainty and to identify those parameters that had the most effect on the economic analysis, we conducted an extensive set of one-way sensitivity analyses, using the upper and lower limits of the 95% confidence intervals of the parameter values or credible ranges of values where parameters were populated from clinical opinion. A probabilistic sensitivity analysis (PSA) was conducted for the 2-year base-case analysis, to confirm the robustness of the cost-effectiveness conclusions given parameter uncertainty. In the PSA, the model was run repeatedly for 1000 iterations, and parameter values were varied simultaneously within assigned statistical distributions (Tables 2–4); Lognormal (for relative risks, bounded 0–infinity), beta (for utility and probabilities, bounded 0–1), and uniform (where no information on distribution shape).

The PSA focused on a cost-per-QALY outcome assessed for first-line and second-line duloxetine versus a base-case treatment strategy. The results of the PSA are presented in scatter plots and in associated cost-effectiveness acceptability curves.

Results

The introduction of duloxetine in any treatment sequence (i.e., first-, second-, or third-line placement) resulted in additional QALYs and response-based clinical benefits over the standard US treatment approach for FM in the model (Table 5).

Table 5.  Cost-effectiveness results (per 1000 patients).

	Base case	Duloxetine sequence placement
		A First line	B Second line	C Third line	D Fourth line	E Fifth line	F Sixth line
Total cost	$5,170,290	$5,753,200	$5,314,042	$5,274,407	$5,296,697	$5,189,409	$5,220,160
Direct: drug	$1,448,445	$2,054,327	$1,605,370	$1,557,125	$1,573,186	$1,466,527	$1,500,891
Direct: nondrug	$3,721,844	$3,698,873	$3,708,672	$3,717,282	$3,723,511	$3,722,882	$3,719,269
SCMs	18,963	19,627	19,423	19,296	19,251	19,240	19,291
QALYs	1,361	1,373	1,369	1,367	1,365	1,365	1,366
Incremental cost and benefits
Total cost		$582,911	$143,752	$104,118	$126,408	$19,119	$49,870
SCMs		665	460	334	288	277	329
QALYs		12.3	8.7	5.9	4.0	4.2	5.3
ICERs
Cost per SCM		$877	$312	$312	$439	$69	$152
Cost per QALY		$47,560	$16,565	$17,548	$31,848	$4,506	$9,471

ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; SCM, symptom control month.

First-line treatment with duloxetine resulted in an additional 665 SCMs and 12.3 QALYs per 1000 patients, at an additional cost of $582,911 over base case. This equated to a cost per additional SCM of $877 and a cost per additional QALY of $47,560 when compared with the base-case treatment sequence. Second-line treatment resulted in an additional 460 SCMs and 8.7 QALYs per 1000 patients, at an additional cost of $143,752. This equated to a cost per additional SCM of $312 and a cost per additional QALY of $16,565 when compared with the base-case treatment sequence.

While benefits were achieved in the base-case treatment sequence, a significant percentage of patients (approximately 9%) failed to achieve adequate levels of long-term pain control or experienced a drop-out from therapy and subsequently went untreated over the remaining model time horizon. Placement of duloxetine early in the treatment sequence resulted in a greater proportion (60.8 vs. 58.5%) of time spent in higher level of pain control (≥50% improvement), which was closely correlated with improved HRQoL and utility scores.

In a relative step-wise comparison of the possible duloxetine strategies, first-line use resulted in ICERs of $2145 per SCM and $122,727 per QALY over second-line use (Figure 2). Second-line use resulted in ICERs of $715 per SCM and $28,101 per QALY over a later placement, with third-line treatment being ruled out through extended dominance (see sequence C in Figure 3, as second-line use of duloxetine provides a greater benefit at a lower cost per SCM or cost per QALY).

Figure 2.  Cost-effectiveness frontier plots (per 1000 patients). A = first line; B = second line; C = third line. D = fourth line; E = fifth line; F = sixth line. CE, cost effectiveness; QALY, quality-adjusted life-year.

Figure 3.  PSA cost-effectiveness acceptability curves. CEAC, cost-effectiveness acceptability curve; PSA, probabilistic sensitivity analysis; QALY, quality-adjusted life-year.

Sensitivity analysis

The sensitivity scenario analyses confirmed that two key parameters drove the cost-effectiveness levels for duloxetine within a standard treatment sequence for FM: the level of pain response assumed for TCAs (in the absence of adequate quality of published 50% response data), and the proportion of long-term drop-out patients who were assumed to be lost to any subsequent treatment (Table 6).

Table 6.  Sensitivity scenario results.

Outcome	Cost per SCM ($)			Cost per QALY ($)
	First line vs. base case	Second line vs. base case	First line vs. second line	First line vs. base case	Second line vs. base case	First line vs. second line
Base-case analysis	877	312	2,145	47,560	16,565	122,727
Discount rate for cost and benefits = 0%	884	317	2,158	47,543	16,644	124,517
Discount rate for cost and benefits = 6%	870	308	2,133	47,576	16,586	121,066
Benefits undiscounted/ cost discounted (3%)	877	312	2,145	47,170	16,317	123,780
TCA 50% response: set to placebo levels	848	312	2,051	29,288	16,565	39,738
Pain response rates: set to 120% of base-case values	1,187	370	2,250	61,765	17,866	130,271
Pain response rates: set to 80% of base-case values	638	284	2,028	36,027	16,039	114,407
3-month switching: set to 120% of base-case values	788	273	2,437	44,791	15,168	149,975
3-month switching: set to 80% of base-case values	927	336	1,997	48,706	17,170	110,262
Long-term drop-out lost to subsequent treatment: 5%	852	304	1,713	46,110	15,870	98,361
Long-term drop-out lost to subsequent treatment: 20%	772	265	1,184	42,044	13,497	68,283
Long-term drop-out lost to subsequent treatment: 50%	685	194	872	38,411	10,057	50,435
Long-term drop limited to 1-year horizon	890	300	1,984	47,808	15,648	112,727
Pain response achieved at end of 3-month period	877	312	2,145	63,841	21,948	170,144
Pain response achieved at beginning of 3-month period	877	312	2,145	42,863	14,952	110,201
Utility assigned to pain: upper 95% CI level	877	312	2,145	47,640	16,563	123,476
Utility assigned to pain: lower 95% CI level	877	312	2,145	47,753	16,665	122,646
Assumed all visits as GP	879	312	2,153	47,682	16,549	123,185
Assumed all visits as outpatient	875	313	2,139	47,455	16,579	122,336
Inclusion of societal cost	830	263	2,103	44,990	13,939	120,295

CI, confidence interval; GP, general practitioner; QALY, quality-adjusted life-year; SCM, symptom control month; TCA, tricyclic antidepressant.

When the 50% response levels of the TCA were set to levels approaching those of placebo (rather than assuming a relative effect size), the first-line use of duloxetine was seen to be much more cost-effective relative to base case (a cost per QALY of $29,288) and to second-line use of duloxetine (cost per QALY of $39,738) (Table 6). Because of the lack of published data on the 50% defined response rate for TCA, it was difficult to have certainty on relative differences to duloxetine and other treatments. In the base case, we assumed a similar relative level between the 30% and the 50% defined responses seen with duloxetine (42.5 ÷ 56.8 = 0.75), i.e. the model assumed 75% of TCA responders achieved the 50% improvement definition of response.

When we assumed that 50% of all long-term drop-out patients were lost to subsequent treatment and we applied this percentage equally for all active treatments, again the first-line duloxetine treatment compared with base case remained cost effective (at a reduced cost per QALY of $38,411). The additional benefits of first-line duloxetine treatment relative to second-line treatment approached thresholds of cost effectiveness of $50,000 (with a cost per QALY of $50,435). The cost effectiveness of second-line duloxetine compared with base case was further reduced to $10,057 from $16,565 under this scenario of long-term drop-out.

The assumption of the timing of achieving the response also had an effect on the level of additional QALYs for duloxetine, with an additional 6.5–9.6 QALYs per 1000 patients achieved with second-line duloxetine (and 9.1 to 13.6 QALYs with first-line duloxetine). This introduced a significant variation in the cost-per-QALY measure (Table 6). However, it was much more in line with the clinical evidence that the treatment effect, in terms of pain control, was achieved within the first 2–4 weeks of treatment, which was in line with our base-case settings.

PSA

In the PSA analysis, the cost-effectiveness acceptability curves were presented for both first-line and second-line duloxetine treatment strategies (Figure 3). The results showed that, given the underlying uncertainty in all the parameter values, there was a 60% or greater likelihood that second-line use of duloxetine (immediately after failure on TCAs) would provide additional QALY benefits at a cost per QALY that was lower than $50,000. Using a lower threshold of cost effectiveness, $30,000 per QALY, reduced this percentage to a 53% likelihood that second-line use of duloxetine would be cost effective. The same PSA analysis showed a corresponding 41% likelihood of first-line duloxetine reaching similar levels of cost effectiveness.

Discussion

Summary of research findings

The economic model confirmed that, as part of a standard sequential treatment of typically prescribed and FDA-approved drugs for FM in the US, the introduction of duloxetine resulted in significant additional clinical benefits that could be expected to translate into HRQoL benefits to patients. The level of this additional benefit (12.3 QALYs per 1000 patients for first-line duloxetine and 8.7 QALYs per 1000 patients for second-line duloxetine) is achieved at an additional cost that is likely to be seen as an acceptable investment in healthcare (at a cost per QALY of under $50,000). These cost-effectiveness results were robust to a full range of parameter and sensitivity analysis assumptions. The cost per QALY for second-line duloxetine treatment versus a base-case sequence remained within the range $13,000–22,000 in the sensitivity scenarios. For first-line duloxetine treatment, the corresponding sensitivity range was $29,000–62,000.

An important part of the model’s design is the assumption that treatment-failure patients would be challenged with an alternative drug therapy as part of an overall treatment sequence. A recently published economic model of pregabalin18 looked only at a first-line positioning of pregabalin and alternative treatments for FM, including TCAs. However, the Choy model was not developed to consider a second-line positioning of these same drugs (after failure of TCAs, for example), nor did the Choy model consider the ongoing treatment experience of patients who did experience a treatment failure. This is the definite weakness in nonsequence-based economic models for pain control because it seems unrealistic that patients would not try alternative treatments. Our model design, as used in this study, allowed us to explore the importance of assuming subsequent treatments for drop-out patients (i.e., treatment switching), and this was a significant model parameter.

Another noteworthy aspect of the current model is that the underlying clinical evidence base is much weaker for the older off-label treatments (as we described in the Methods section). Also, many of the studies of more established treatments had lower placebo response rates, making cross-comparisons to other treatments difficult, particularly when using a standard indirect adjustment approach based on relative risks of response to a common placebo comparator. This issue was dealt with through the use of a MTC model in the data synthesis, which was employed to allow the variation in individual trial placebo levels to be reflected in the overall relative risk data used in the model (Tables 2 and 3).

Significantly, the clinical trial EQ-5D data available from the duloxetine studies demonstrated a clear and strong correlation between the BPI pain-scale severity scores (0–10) and standard HRQoL utility weights. At the most severe levels of pain, the observed utility dropped to around 0.05–0.12 (BPI 9–10), compared with lowest levels of pain (BPI 0–1), which had utilities in the range 0.79–0.90. This in itself is useful research data to evidence the steep decline in HRQoL experienced by patients with FM, where pain is a significant burden of the disease.

Study limitations

The model has a number of specific limitations in its evidence base that should be acknowledged. First, our review of the clinical evidence base supporting the assumptions for treatment comparators in the model identified a great deal of underlying uncertainty in parameter values, in particular response rates at both the 30% and 50% improvement thresholds. As previously indicated, the published trials for many comparator treatments were limited in both number and cohort size, and this resulted in a widening of confidence intervals for the estimates of relative response rates and drop-out rates (Tables 2 and 3). Also, many of the older clinical trials had absolute placebo response rates much lower than those seen in more recent pivotal trials, resulting in much higher relative response rates. To a certain extent, these limitations were dealt with through the PSAs, which explored the full range of potential values given this level of uncertainty. However, it remained difficult to make any indirect comparisons with long-term off-label treatments, such as TCAs.

The advantage of duloxetine (and other therapies for FM) is in the clear clinical evidence of patients achieving a higher level of response that corresponds to at least a 50% improvement in baseline pain-severity scores. There remain limited published data using similar definitions of pain response for TCAs, typically based on the lower 30% improvement threshold. However, even when we assumed that TCAs would provide a 50% response rate level similar to the relative 30% and 50% response rates for other treatments, the clinical and economic advantage of introducing a second-line duloxetine treatment remained evident.

The model assumed that the treatment response rates were independent of the exact placement in the treatment sequence (i.e., that treatment effect is independent of any prior therapeutic failures). This is a necessary assumption for the sequence-based model because none of the trial data were reported for patient groups explicitly defined by prior treatment failure.

The model also assumed that patients maintained their level of pain control, provided that they remained on active treatment. However, the model did allow for a proportion of patients to drop out from active treatment over time. Clinical data for duloxetine from the 6-month clinical trials certainly suggested that such a prolonged response could be achieved if treatment adherence is met. However, long-term data for duloxetine and comparator treatments remain limited at the present time.

Finally, the model assumed the same level of average pain improvement from baseline (70%) for all full responders, who by definition achieved a ≥ 50% improvement irrespective of the treatment taken. This level of improvement was used to calculate the overall pain improvement and hence the utility score for responders. We based this average value on the observed change in pain scores of full responders from the four duloxetine trials. However, there were no equivalent data for the treatment comparators; therefore, the assumption could not be fully validated.

There remains a clear need for more head-to-head clinical trials of treatments for FM and the continued use of standardized outcome measures for assessing pain in patients with FM.

Conclusion

The introduction of duloxetine into a standard FM treatment sequence is a cost-effective way to improve pain control in patients with FM. This result was observed in particular for second-line duloxetine immediately after failure on TCAs, such as amitriptyline, and was robust to sensitivity analyses. Although first-line use provides additional benefits beyond second-line use the cost effectiveness in this position is uncertain given the assumptions around the current use of TCAs.

While there are clear limitations in the quality of and consistency in the underlying clinical data for FM treatments, making comparative economic modeling a challenge, it is also clear that there is a large and significant proportion of patients who remain inadequately treated with current, standard pharmacotherapy for FM-related pain symptoms and who therefore can be expected to have a significantly reduced quality of life.

Transparency

Declaration of funding

The funding for this paper was provided by Eli Lilly and Company.

Declaration of financial/other relationships

S.M.B. and N.R. are employees of RTI Health Solutions, an independent contract research organization that has received research funding for this and other studies from Eli Lilly and Company and pharmaceutical companies that market drugs for use in the treatment of patients with fibromyalgia. T.K.L., Y.Z., and A.C. are employees of Eli Lilly and Company, a pharmaceutical company which manufacturers duloxetine. D.A. and K.L. have received funding support from Eli Lilly and Company.

Acknowledgments

No assistance in the preparation of this article is to be declared.