Cost-effectiveness analysis of baricitinib versus dupilumab for moderate to severe atopic dermatitis: an Italian healthcare system perspective

Abstract

Aims

To assess, within the Italian healthcare system, the cost-effectiveness of baricitinib versus dupilumab, both in combination with topical corticosteroids (TCS), in adults with moderate to severe atopic dermatitis (AD) who are eligible for but have failed, have contraindications to, or cannot tolerate ciclosporin.

Materials and methods

Using the perspective of the Italian healthcare payer, direct medical costs associated with each intervention were estimated over a lifetime horizon. A Markov cohort model utilized the proportions of patients with ≥75% improvement Eczema Area and Severity Index obtained from clinical trials. Health outcomes were evaluated in quality-adjusted life years (QALYs) to assess the cost effectiveness of baricitinib against a willingness-to-pay threshold of €35,000 per QALY gained.

Results

In the base case, with secondary censoring applied, patients treated with dupilumab or baricitinib, in combination with TCS, accumulated total costs of €135,780 or €129,586, and total QALYs of 18.172 or 18.133, respectively. The incremental cost-effectiveness ratio of dupilumab versus baricitinib was estimated at €160,905/QALY.

Limitations

Core assumptions were needed to extrapolate available short-term clinical trial data to lifelong data, adding uncertainty. Benefits of baricitinib seen in clinical trials and not assessed in dupilumab clinical trials were not included. Discontinuation rates for each treatment were derived from different sources potentially introducing bias. Results may not be generalizable to other populations.

Conclusions

This cost-effectiveness analysis shows that, from the Italian healthcare payer perspective, in the treatment of patients with moderate to severe AD who have experienced failure on, are intolerant to, or have contraindication to ciclosporin, dupilumab cannot be considered cost-effective when compared with baricitinib. Given its oral administration, favorable risk/benefit profile and lower acquisition cost compared with dupilumab, baricitinib may offer a valuable, cost-effective treatment option—after failure on conventional systemic agents—for patients with moderate to severe AD in Italy.

PLAIN LANGUAGE SUMMARY

Baricitinib is the first oral systemic treatment for patients with moderate to severe atopic dermatitis (AD). The drug was effective for treating patients with AD in clinical trials, producing improvements in skin inflammation, itch, sleep disturbances due to itch and skin pain, as well as the quality of life of patients. However, it is important to ensure that healthcare funds are well spent. We therefore compared the cost-effectiveness of baricitinib, with another new systemic treatment for AD, dupilumab, (both in combination with topical corticosteroids) in patients with moderate to severe AD who are eligible for but have failed or are unable to take ciclosporin, in Italy. We found that using dupilumab to treat these patients with AD cost more than using baricitinib, although dupilumab was more effective. Combining these considerations showed that the cost of obtaining the additional benefit from dupilumab over baricitinib was not cost-effective for the Italian healthcare system. Baricitinib may be a better treatment option because it is given orally, has a favorable balance between the risks and benefits of treatment, and costs less than dupilumab.

Keywords:

• Atopic dermatitis
• baricitinib
• cost effectiveness
• dupilumab
• Italy

JEL Classification Codes:

• H51
• H5
• H
• I19
• I1
• I
• P46
• P4
• P

This article is related to:

Re: Johansson E, Giovannitti M, Mezzetti M, et al. Cost-effectiveness analysis of baricitinib versus dupilumab for moderate to severe atopic dermatitis: an Italian healthcare system perspective. J Med Econ. 2023;26(1):1155–1166

Introduction

Moderate to severe atopic dermatitis (AD) can cause widespread inflamed skin all over the body and constant itching, with areas of affected skin occasionally becoming infected.^1–3 AD causes the highest number of disability-adjusted life years among skin diseases,⁴ and patients with moderate to severe disease often experience unbearable symptoms that interfere with practically every daily activity.^3,5–7 An Italian study estimated the total annual financial burden of the disease to be €4,284 per patient,⁸ and a recent United States study estimated the average lifetime cost of usual care with emollients for moderate to severe AD at approximately $271,450 per patient.⁹ However, limited treatment options are available for these patients.

Topical therapies are the backbone of AD treatment.^10,11 However, patients with moderate to severe AD who have been unable to achieve adequate disease control with first line topical corticosteroids (TCS), with or without adjunct topical calcineurin inhibitors (TCI) are candidates for additional therapies, such as phototherapy, high-potency TCS, and systemic treatment.^10,11 Ciclosporin is the only conventional systemic therapy approved in the European Union (EU), and although efficacious, it has a narrow therapeutic index and monitoring of renal function and blood pressure is recommended, as use can lead to renal impairment and increased blood pressure.¹¹ Current AD guidelines list off-label use of systemic corticosteroids (only as rescue therapy), methotrexate, and azathioprine as alternative therapeutic options; however, these therapies are limited by low quality evidence of efficacy, various contraindications, and, in some instances, severe toxicity.¹¹

In 2017, the European Medicines Agency (EMA) approved dupilumab,¹² an immunoglobin (Ig) G4 monoclonal antibody that inhibits interleukin (IL)-4 and IL-13, for subcutaneous injection, with or without TCS, for patients with moderate to severe AD. Dupilumab was the first new treatment approved for moderate to severe AD since 2002.^12,13 However, not all patients achieve disease control with this drug,¹⁴ injections cause anxiety for some patients, with injection site reactions occurring in 8 to 19% of patients,^14,15 and conjunctivitis occurring more frequently than with placebo in clinical trials¹⁴ and within 16 weeks of starting treatment in 34% of patients in long-term registries.¹⁶

Baricitinib, the first oral systemic treatment for moderate to severe AD, was approved by the EMA for adult patients who are candidates for systemic therapy in 2020.¹⁷ Baricitinib inhibits Janus Kinase (JAK) 1 and JAK2 with high selectivity, and targeted and reversible inhibition, suggesting it affects several cytokines involved in AD pathogenesis.¹⁸ Three Phase 3, randomized, double-blind, placebo-controlled trials have confirmed the efficacy of baricitinib (e.g. improvements in skin inflammation, itch, sleep disturbances due to itch and skin pain), when used as monotherapy or in combination with TCS, in the approved patient population.^19,20 A further Phase 3, randomized, double-blind, placebo-controlled, trial confirmed the efficacy of baricitinib in combination with TCS in patients with prior failure of ciclosporin, or intolerance or contraindication to ciclosporin.²¹ Furthermore, rapid, and sustained efficacy has been observed for many patient-reported outcomes, demonstrating the important impact baricitinib may have on patients’ quality of life.^19–21

In this analysis, we assessed the cost-effectiveness of baricitinib, as compared with dupilumab, both in combination with TCS therapy, in patients with moderate to severe AD who are eligible for but have failed, have contraindications to, or cannot tolerate ciclosporin, within the context of the Italian healthcare system.

Methods

General

Target population

The target population were adults (≥18 years) who had been diagnosed with moderate to severe AD for at least 12 months, had an inadequate response to existing topical medication within the last 6 months, were eligible for systemic therapy, and had experienced failure to ciclosporin, or were intolerant to, or had a contraindication to ciclosporin. The target population was selected as it reflected the population identified by the Italian Medicines Agency (AIFA) as eligible for treatment with the comparator, dupilumab.²²

Study perspective

Estimates of direct medical costs associated with each intervention were derived from the perspective of the Italian healthcare payer. These included the costs of the comparators (i.e. baricitinib plus TCS and dupilumab plus TCS), concomitant topical treatments, and rescue medications to treat flares often used in the management of moderate to severe AD, ongoing administration and monitoring costs, costs associated with adverse events (AEs), and the costs of best supportive care (BSC) for non-responders.

Comparators

Dupilumab was selected as the comparator for baricitinib as it is the only drug currently reimbursed in Italy for the treatment of moderate to severe AD after cyclosporin failure. In the absence of head-to-head clinical study data, new AIFA guidelines on the pricing and reimbursement of medicines (which came into effect March 2021) require a cost-effectiveness analysis (CEA) comparing any new drug with the appropriate comparator(s) in the relevant indication to be included in the submission dossier.²³ Both baricitinib and dupilumab were assumed to be administered with adjunct emollients and TCS,^10,11,21 in line with standard clinical practice. Dose regimens for both comparators reflect the main regimens recommended for AD in the EU: baricitinib 4 mg/day administered orally²⁴ and dupilumab 600 mg starting dose followed by 300 mg once every 2 weeks administered by subcutaneous injection.²⁵

Time horizon

A lifetime horizon was adopted for the base case given the Italian healthcare payer perspective, as this was aligned with the earlier dupilumab submission.^7,26 An alternative 3-year time horizon was explored in a scenario analysis, as this was considered long enough to capture all effects of both comparator treatments on benefits and costs.

Discount rate

Both costs and health outcomes were discounted at an annual rate of 3%, as recommended by Italian pharmacoeconomic guidelines.²³ Sensitivity analyses explored annual discount rates between 0% and 5%, as per Italian guidelines.

Censoring

Secondary censoring (i.e. censoring of data after permanent study drug discontinuation) was used for all utility and efficacy variables included in the models. This censoring method is considered to be reflective of clinical practice.

Outcomes

Choice of outcomes

The health outcomes of each intervention were evaluated in quality-adjusted life years (QALYs) to assess the cost effectiveness of baricitinib against the willingness-to-pay threshold of €35,000 per QALY gained. This threshold is the approximate mid-value of published ranges for Italy of €12,000–€60,000.²⁷

Measurement of effectiveness

Estimates of clinical effectiveness were derived from an indirect treatment comparison (ITC) of baricitinib 4 mg/day versus dupilumab 300 mg once every 2 weeks (following a loading dose of 600 mg), in combination with TCS, in adult patients with moderate to severe AD who had failed, had contraindications to, or were intolerant of both topical agents and ciclosporin (ITC population B).²⁸ The ITC used Bucher methodology²⁹ for meta-analysis of data when more than one study contributed, and placebo was used as the common comparator.

Relative efficacy was calculated as relative risk versus placebo for all outcomes, based on primary and secondary endpoints common across trials, including the proportion of patients with ≥50%, ≥75%, or ≥90% improvement Eczema Area and Severity Index (EASI50, EASI75, EASI90) at Week 16, with missing data accounted for by non-responder imputation.²⁸ Clinical trials included in the ITC population B were BREEZE-AD4 (NCT03428100) for baricitinib²¹ and LIBERTY AD CAFÉ (NCT02755649) for dupilumab.³⁰ Estimates of clinical effectiveness for BSC were derived from the placebo response rate in BREEZE-AD4.

Treatment response for the CEA base case was defined as an EASI75 response at the end of the induction period, measured using secondary censoring (Table 1). An alternative definition of treatment response, EASI50, was explored in sensitivity and scenario analyses.

Table 1. Outcomes—base case model inputs.

	Baricitinib		Dupilumab		BSC
Model input	Value	Source^a	Value	Source^a	Value	Source^a
EASI75 treatment response rate at 16 weeks, secondary censoring,^b % (SE%)	34.8 (5)	ITC (BREEZE-AD4)^28	38.5 (5)	ITC (CAFÉ)^28	19.4 (4)	BREEZE-AD4^21
Probability of treatment discontinuation between Week 16 and Week 52	0.213	BREEZE-AD4^21; [Data on file]	0.163	Real-world study^31	0.383	BREEZE-AD4^21; [Data on file]
Annual probability of discontinuation after 52 weeks	0.277	BREEZE-AD4^21; [Data on file]	0.227	Real-world study^31	0.430	BREEZE-AD4^21; [Data on file]
Annual probability of adverse event		BREEZE-AD4^21; [Data on file]		NICE TA534^26		NICE TA534^26
Injection site reaction	0.00		0.09		0.00
Allergic conjunctivitis	0.04		0.40		0.19
Infectious conjunctivitis	0.00		0.26		0.03
Oral herpes	0.17		0.06		0.11
Upper respiratory tract infection	0.07		0.00		0.00
Blended age-adjusted morality rate		Italian life tables^33 & BREEZE-AD4^21		Italian life tables^33 & BREEZE-AD4^21		Italian life tables^33 & BREEZE-AD4^21
Baseline utility, secondary censoring,^b mean (SD)	0.7996 (0.1441)					BREEZE-AD4^21; [Data on file]
Utility change from baseline at 16 weeks, secondary censoring,^b mean (SD)						BREEZE-AD4^21; [Data on file]
<EASI50	0.0776 (0.0597–0.0954)
EASI50 (to < EASI75)	0.0978 (0.0805–0.1150)
≥EASI75	0.1078 (0.0907–0.1250)

^aData were either derived or extracted from the listed sources.

^bSecondary censoring, that is censoring of data after permanent study drug discontinuation, was applied.

BSC, best supportive care; CAFÉ, LIBERTY AD CAFÉ trial; EASI50/EASI75, ≥50%/≥75% improvement Eczema Area and Severity Index; ITC, indirect treatment comparison; SD, standard deviation; SE, standard error.

Responders were assumed to enter a maintenance phase, during which patients could discontinue treatment. Maintenance phase discontinuation rates for baricitinib and BSC were derived from the BREEZE-AD4 study. Source study permanent discontinuation rates between Week 0 and Week 52 (representing all-cause discontinuation; i.e. discontinuation of treatment due to lack of long-term efficacy, adverse event, patient or physician preference) were converted into discontinuation rates for the 36-week timeframe; that is, the probability of treatment discontinuation between the end of the induction period (time of response assessment; Week 16 in base case) and Week 52 (Table 1). The rate for dupilumab was derived from a real-world persistence study,³¹ and was based on persistence at 12 months, with conversion to the appropriate timeframe (36 weeks). Loss of response was assumed to occur at a continuous and constant rate, and loss of response led to treatment discontinuation.

Annual probabilities of discontinuation after 52 weeks, representing the annual rates at which patients discontinue baricitinib or dupilumab each year due to lack of long-term efficacy, adverse events, patient or physician preference, were applied to patients in the maintenance health state starting in the second year of the model. These probabilities were derived as follows (Table 1): for dupilumab, a probability of 22.7% was set based on the abovementioned real-world persistence study,³¹ and for baricitinib, post hoc analysis was performed on the long-term extension phase of the BREEZE-AD4 study ranging from Week 52 to Week 104. Patients were included if they started the trial receiving baricitinib 4 mg at baseline and remained on the same treatment after re-randomization after 52 weeks. As no similar analysis was possible for the placebo arm of the BREEZE-AD4 study (most patients had been re-randomized to baricitinib 2 mg or 4 mg treatment), annual all-cause discontinuation data (Week 0–52) were used for BSC. Patients who discontinued baricitinib or dupilumab were switched to BSC.

The average number of flares per patient per year was assumed to be eight with BSC and in non-responders,³² and three while patients were receiving baricitinib or dupilumab; the average duration of a flare was assumed to be 15.2 days. AEs included were based on the most frequent and serious reported for each treatment, and derived from 16-week data from the BREEZE-AD4 study, rescaled to annual probabilities, for baricitinib, and from annual probabilities from National Health and Care Excellence (NICE) Technology Assessment (TA) 534 for dupilumab and BSC (Table 1). The risk of AEs was assumed to remain constant over the treatment duration, and AEs were not modelled as separate health states, rather AE rates and cost consequences accumulated in each cycle of treatment.

Age-adjusted mortality was stratified by sex and based on Italian Life Tables,³³ with no adjustment for AD-specific mortality. Gender-specific rates were combined to a blended rate based on the BREEZE-AD4 trial population gender proportions.

Measurement and valuation of preference-based outcomes

Utility values, baseline, and change from baseline at Week 16 based on observed and secondary censored responder criteria (e.g. EASI75), were derived directly from the BREEZE-AD4 study population, using linear repeated measurement models with utility values as dependent variables and response levels (using secondary censoring) as independent variables. Health state utility values were calculated by mapping EQ-5D-5L onto EQ-5D-3L response levels,³⁴ which were then valued against the Italian value set.³⁵

Given the base case lifetime horizon and the expectation that health utility declines with age, an annual adjustment factor for age, derived from United Kingdom (UK) data, was included in the model.³⁶ The annual adjustment factor Y enters the model multiplicatively and is derived using the following formula, where [38.2] is the mean age of the patients in the Phase 3 baricitinib clinical trials and [0.903] the overall utility at baseline age: $$\text{Y }=\left(1.0708-0.0044*\left[38.2\right]\right)/\left[0.903\right]$$ Included AEs were generally mild and transient, and generally balanced between treatment arms. No significant decrement to health-related quality of life (HRQoL) was expected due to AEs. Therefore, AE disutilities were not incorporated to avoid double counting, in line with previous AD economic assessments.^7,37

Costs

Estimating resource use and costs

Table 2 summarizes all cost inputs included in the model. Drug acquisition costs for baricitinib and dupilumab were derived from both the Compendio Farmadati and the Information Hospital Service (IHS) database,^38,39 and calculated based on the cost per unit of each treatment and the required number of units per cycle for the approved AD regimens (Supplementary material Table S1). The cost of BSC, assumed to consist of TCS (clobetasol), was also derived from Compendio Farmadati, and calculated based on the pack cost and recommended daily dose (3 g/day)^38,40 (Supplementary material Table S2). Patients receiving BSC were assumed not to accrue any drug acquisition costs during BSC response to avoid double counting costs with concomitant background TCS medication.

Table 2. Costs—summary of base case model inputs.

Cost category		16-week induction cost (€)	Annual maintenance cost (€)	Sources
Drug acquisition	Baricitinib	2,207.52	7,194.15	[38,39]
	Dupilumab	3,077.46	8,890.44
	BSC*	18.26	59.36
Administration and monitoring	Baricitinib	417.14	799.68	[8,16,41–44,47]
	Dupilumab	671.93	1,633.79
	BSC	1,271.75	2,273.21
Concomitant medication	Baricitinib	141.42	425.08	[38]
	Dupilumab	141.42	425.08
	BSC	257.16	824.48
Rescue medication for flares	Baricitinib	214.79	644.37	[38]
	Dupilumab	214.79	644.37
	BSC	644.37	1,718.32
Adverse events	Baricitinib	36.22	113.58	[7,26,43,48,49]
	Dupilumab	52.64	171.08
	BSC	16.59	53.90
Non-responders	BSC	–	5,002.92	[8,16,38,41–44,47]

*BSC drug acquisition cost is for TCS (clobetasol) 3 g/day.

BSC, best supportive care.

Costs of concomitant medication (bathing and emollient products, and background TCS) were derived from Compendio Farmadati³⁸ (Supplementary material Table S3). Costs of rescue medications for flares, often required due to the relapsing remitting nature of AD, were based on five potential treatment options (with weight-based doses calculated based on the mean weight of patients in the BREEZE-AD4 study²¹) and derived from Compendio Farmadati³⁸ (Supplementary material Table S4).

Resource use related to the administration and monitoring of treatment during induction and maintenance, which varied due to the differences in modes of action and administration routes of the comparators, was aligned with published Italian sources, and associated unit costs were derived from published Italian sources^8,16,41–46 (Supplementary material Table S5). Resource use and associated costs in the BSC non-response state, which comprised physician visit costs and the costs of other resource use for ongoing management, concomitant medication costs and the cost of rescue medication for flares, were derived from published Italian sources^{8,16,38,41–44,47} (Supplementary material Table S6).

Costs associated with AEs were derived from the Italian Ministry of Health,⁴³ the Italian Diagnosis Related Groups (DRG)^48,49 and NICE TA534⁷ (Supplementary material Table S7).

Currency, price date, and conversion

Costs were calculated in 2020 Euro, with no adjustment for time.

Model

Choice of model

A Markov cohort model was built with four health states: induction, maintenance, non-response, and death (Figure 1). It was programmed to facilitate pairwise comparisons of treatments as well as the implementation of different lengths of induction periods depending on treatment, with the same structural framework as the NICE TA534⁷ and United States Institute for Clinical and Economic Review⁵⁰ models identified in a systematic literature review of cost-effectiveness model structures adopted in previous analyses for treatments of AD. A cycle length of 4 weeks (28 days) was sufficiently short to capture variations in costs and benefits resulting from changes in treatment.

Figure 1. Markov cohort model structure.

*Arrows to the Dead health state removed for simplification; Dead can be reached from any other health state at any time.

Treatment was dichotomized into two periods (induction and maintenance) as per other cost-effectiveness models undertaken in AD.^7,37 A tunnel state framework was implemented in the induction period to reflect patients’ stay on treatment (baricitinib, dupilumab, or BSC) until first response assessment at 16 weeks (i.e. trial induction treatment period). Patients who responded to treatment at the end of the induction period remained on treatment by moving to the maintenance period. For patients on maintenance treatment, it was assumed that the level of response was constant at the level observed at the end of the induction period. Patients remained in this maintenance period until they transitioned to the next treatment in the sequence (i.e. BSC) or to non-response (i.e. after loss of response to BSC). Patients who did not respond to treatment at the end of the induction period transitioned to the BSC induction phase. Patients transitioned to the non-response health state when they lost response to BSC. Death was possible from every state.

Model assumptions

The model included a number of assumptions (Table 3) including that only EASI75 responders continued treatment beyond 16 weeks; on average, patients maintained the level of response achieved at the end of the induction period until they discontinued treatment; patients with an initial response at Week 16 but without a response at Week 52 discontinued treatment at a continuous and constant rate between 16 and 52 weeks; patients in the non-response health state were assumed to remain in that state until the end of the model or death; and the risks of AEs remained constant over the treatment duration. In addition, health state utility values were derived from a baricitinib clinical trial capturing the impact of baricitinib and BSC on utility; although, given the different modes of action and routes of administration for baricitinib and dupilumab, the utility values may not reflect all AEs associated with dupilumab. It was also assumed that health utility values decline with age and that patients had 100% treatment adherence.

Table 3. Key model assumptions.

Assumption	Rationale^a
Only EASI75 responders continue treatment beyond 16 weeks	EASI75 is an effective (if conservative) proxy for clinical improvement justifying treatment continuation after the induction period
	Based on the less stringent AIFA assessment that patients receiving dupilumab and achieving EASI50 should continue with dupilumab treatment^22
On average, patients maintain the level of response achieved at the end of the induction period until they discontinue treatment	Patients on average appear to maintain their response over time in open label extension studies^51^,^52
	Loss of response was assumed to result in treatment discontinuation and is thus reflected in the all-cause discontinuation rate, as previously modelled in psoriasis^53
Patients with an initial response at Week 16 but without a response at Week 52 discontinued treatment at a continuous and constant rate between 16 and 52 weeks	It appears plausible that treatment discontinuation is a random process—reasonably approximated by a constant and continuous rate
Patients in the non-response health state remained in that state until the end of the model or death	Patients who received BSC had failed, had a contraindication to, or could not tolerate all other treatment options
The risks of AEs remain constant over the treatment duration	Time-correlated data on the incidence of AEs were not available. Thus, the conservative assumption that AEs are constant over time was used
Health state utility values were derived from a baricitinib clinical trial capturing the impact of baricitinib and BSC on utility	Derived from data obtained from patients receiving baricitinib/best supportive care; however, the utility values may not reflect AEs associated with different modes of action, such as injection site reactions for dupilumab, and some of the AEs may be too short-lived to be reflected in the utilities estimated from clinical trials
Health state utility values decline with age	The model applied an age adjustment factor to account for this decrease over time
Adherence to treatment was 100%	There were high compliance rates in trials of both dupilumab and baricitinib.
	Should compliance be lower, it may be assumed that the effectiveness and costs would decrease proportionally, limiting the impact on the ICER from changes to compliance
Efficacy occurred at the end of the induction (i.e. trial) period and, therefore, only patients who entered the maintenance phase benefited from treatment	This is conservative given that novel systemic therapies incur all the costs, but not the benefits of treatment during the trial period
	The impact is expected to be limited given the short duration of the trial period (16 weeks)

^aCaveats and additional details are also included as appropriate.

AE, adverse event; AIFA, Italian Medicines Agency; BSC, best supportive care; EASI50/75, 50%/75% improvement Eczema Area and Severity Index; ICER, incremental cost-effectiveness ratio.

Analytical methods

A half-cycle correction was not included in the model due to the short cycle length (4 weeks). However, given the different time references for model inputs (e.g. annual [maintenance] and per 16-week period [induction]), variables were rescaled to 4-week durations. Probabilities (e.g. annual) were converted to an instantaneous rate and then converted back to the desired length probability (e.g. 4-week) as follows: $$r=–\text{ln }\left(1–P_{\mathit{annual}}\right)\text{ and}{P}_{4w}=\text{ 1 }–e^{–r*(28/365.25)}$$

For absolute inputs (e.g. annual number of flares), linear conversions were applied by dividing the value to be adjusted by the number of days per year that the input was referring to, for example, 4-week frequency = annual frequency * (28/365.25).

Response data were calculated using non-responder imputation to account for missing data. Utility values were determined using linear repeated measurement models with utility as the dependent variable and response levels (using secondary censoring) as independent variables.

Base case results are presented as total and incremental deterministic costs and QALYs for dupilumab (followed by BSC) and baricitinib (followed by BSC), and the pairwise incremental cost-effectiveness ratio (ICER) for dupilumab (followed by BSC) versus baricitinib (followed by BSC). A cost-effectiveness plane illustrates the base case results in the context of the Italian payer willingness-to-pay threshold.

One-way sensitivity analysis (OWSA) was performed to consider the influence of individual parameters on model outputs, using pre-specified upper and lower limits for pre-selected parameters (Supplementary material Table S8); results are presented in the form of a Tornado diagram that includes the 10 most influential variables. Probabilistic sensitivity analysis (PSA) was undertaken to address uncertainty around the means of input parameters by assigning appropriate distributions around the means (Supplementary material Table S9); results are presented as probabilistic mean costs and QALYs, with a scatterplot of individual simulations on the cost-effectiveness plane and a cost-effectiveness acceptability curve, both generated using 3,000 simulations of the PSA.

Scenario analyses were run to explore the impact of the assumptions and structural uncertainty of the model (Table 4). Results are presented as total and incremental deterministic costs and QALYs for baricitinib (followed by BSC) and dupilumab (followed by BSC), total life years for each treatment, and the pairwise ICER for dupilumab versus baricitinib.

Table 4. Scenario analyses.

Scenario	Base case	Scenario values
Model time horizon	Lifetime	3 years
Caregiver utilities	No health state utility value impact in caregivers	Positive health state utility value impact on caregivers for responders, i.e. EASI75: +0.01
Treatment response definition	EASI75 at 16 weeks	EASI50 at 16 weeks
BSC loss of response	BREEZE-AD4 study data	36.6% annual discontinuation rate from BSC maintenance phase—Peserico et al.^54
Number of flares	8 treated flares per year in BSC maintenance period	15.5 treated flares per year in BSC maintenance phase—Simpson et al.^14

BSC, best supportive care; EASI50/75/90, 50%/75%/90% improvement Eczema Area and Severity Index.

Results

Incremental costs and outcomes

In the base case, in the BREEZE-AD4 study population (patients with moderate to severe AD who were eligible for systemic therapy and had experienced failure to ciclosporin, or were intolerant to, or had a contraindication to ciclosporin), dupilumab and baricitinib, in combination with TCS, accumulated total costs of €135,780 and €129,586, and total QALYs of 18.172 and 18.133, respectively (Table 5). The ICER of dupilumab versus baricitinib was estimated at €160,905/QALY. As seen in the cost-effectiveness plane shown in Figure 2, this ICER is above the willingness-to-pay threshold of €35,000/QALY in Italy.²⁷

Figure 2. Base case results—cost effectiveness plane (dupilumab vs. baricitinib). ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year; WTP, willingness-to-pay.

Table 5. Deterministic base case results.

	Dupilumab	Baricitinib	Incremental
Total cost (€)	135,780	129,586	6,194
Drug acquisition costs (€)	15,186	9,349	5,837
Administration and monitoring (€)	58,683	57,896	787
Concomitant medication (€)	19,678	19,826	−148
Rescue medication for flares (€)	40,708	41,107	−399
Adverse events (€)	1,524	1,408	117
Total QALYs	18.172	18.133	0.038
ICER (€/QALY)	–	–	160,905

ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

Characterizing uncertainty

The 10 most influential variables in the OWSA are displayed in Figure 3. In this diagram, each bar represents the impact of uncertainty in an individual variable on the ICER. The five variables with the largest impact on the ICER were the dupilumab and baricitinib EASI75 responses, the year 2+ treatment discontinuation rate for baricitinib, the dupilumab pack cost, and the year 2+ treatment discontinuation rate for dupilumab. All other variables had minimal effect.

Figure 3. One-way sensitivity analysis results—Tornado diagram (dupilumab vs. baricitinib). Secondary censoring, that is censoring of data after permanent study drug discontinuation, was applied to treatment outcomes. BSC, best supportive care; EASI75, 75% improvement Eczema Area and Severity Index.

A summary of the results of the PSA is presented in Table 6 along with a scatterplot of total costs and QALYs in Figure 4, and a cost-effectiveness acceptability curve in Figure 5. The results of the PSA align well with the deterministic results with only minor deviations in incremental costs and QALYs.

Figure 4. Probabilistic sensitivity analysis results—cost-effectiveness plane (dupilumab vs. baricitinib). Scatterplot of total costs and QALYS. Generated using 3,000 simulations of probabilistic sensitivity analysis. BSC, best supportive care; QALY, quality-adjusted life year.

Figure 5. Probabilistic sensitivity analysis results—cost-effectiveness acceptability curve (dupilumab vs. baricitinib). Generated using 3,000 simulations of probabilistic sensitivity analysis. BSC, best supportive care.

Table 6. Probabilistic base case results.

Treatment	Total costs (€)	Total QALYs	Total life years	Incremental cost (€)	Incremental QALYs	Incremental ICER (€/QALY)
Baricitinib	129,766	18.11	24.58	Reference	Reference	Reference
Dupilumab	136,233	18.14	24.58	6,467	0.04	180,574

ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

The cost-effectiveness results for each scenario analysis described in Table 4 are reported in Table 7. The scenarios with the largest impact on the ICER of dupilumab versus baricitinib were the scenarios on model time horizon and the definition of treatment response.

Table 7. Scenario analyses results.

Treatment	Total costs (€)	Total QALYs	Total life years	Incremental cost (€)	Incremental QALYs	Incremental ICER (€/QALY)
Base case
Baricitinib	129,586	18.13	24.58	Reference	Reference	Reference
Dupilumab	135,780	18.17	24.58	6,194	0.04	160,905
Model time horizon—3 years
Baricitinib	19,126	2.40	2.90	Reference	Reference	Reference
Dupilumab	22,655	2.41	2.90	3,529	0.01	290,890
Caregiver utilities: +0.01
Baricitinib	129,586	18.15	24.58	Reference	Reference	Reference
Dupilumab	135,780	18.19	24.58	6,194	0.04	147,246
Treatment response definition—EASI50
Baricitinib	135,650	18.42	24.58	Reference	Reference	Reference
Dupilumab	176,533	18.86	24.58	40,883	0.44	93,525
BSC loss of response
Baricitinib	129,582	18.14	24.58	Reference	Reference	Reference
Dupilumab	135,776	18.18	24.58	6,194	0.04	160,829
Number of flares
Baricitinib	130,082	18.13	24.58	Reference	Reference	Reference
Dupilumab	136,281	18.17	24.58	6,189	0.04	160,784

EASI50, 50% improvement Eczema Area and Severity Index; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

Discussion

This study estimated that, from the perspective of the Italian healthcare payer, treatment with baricitinib accrued fewer total costs but also slightly fewer total QALYs compared with dupilumab in the treatment of patients with moderate to severe AD who have experienced failure on, are intolerant to, or have contraindications to ciclosporin. The base case results suggested that, for the EASI75 response criteria, dupilumab was associated with 0.038 more QALYs and additional costs of €6,194 over a lifetime horizon than baricitinib, resulting in an ICER of €160,905/QALY. This ICER is significantly above the Italian payer willingness-to-pay threshold of €35,000/QALY,²⁷ suggesting that dupilumab cannot be considered cost-effective in comparison to baricitinib from the Italian healthcare payer perspective. Further explorations suggested that this conclusion was not altered substantially under the range of plausible uncertainty in input parameters.

A recent Italian consensus statement confirmed the inclusion of the itch Numeric Rating Scale in the definition of moderate to severe AD,⁵ and in the ITC, baricitinib produced faster and numerically greater itch improvement at Week 16 than dupilumab in a population similar to our analysis population (patients with moderate to severe AD who had experienced failure on, were intolerant to, or had contraindications to ciclosporin).²⁸ In contrast, the EASI75 results at Week 16 numerically favored dupilumab. These findings suggest that, if itch is considered, the ICER for dupilumab versus baricitinib may be even greater than obtained in the current CEA.

This CEA had several strengths. These include that the Markov cohort model structure used for this analysis has been used previously in AD by users from academia and the pharmaceutical industry, as well as healthcare payers and decision makers,^7,37,50,53 and results from this type of model have been widely published in the literature. All input cost and resource use data were aligned with the current Italian 2020 setting. The model was designed with functionality to include treatment sequencing should a new market entry occur. Thus, given the restrictions of that, tunnel states were implemented. The model also allows flexible induction input so that we could also change the response for itch at 4 weeks and use the same model structure. Additionally, where appropriate, sensitivity and/or scenario analyses were conducted to test assumptions around key inputs, in line with previously published economic models.^7,26,37 Notably, these analyses included large variations in discontinuation rate that did not change the overall cost-effectiveness conclusions.

Several limitations must also be mentioned. First, in the absence of lifetime data, several core assumptions were applied within the model to extrapolate available short-term clinical trial data. This extrapolation introduced a prominent source of uncertainty to this CEA, although long-term drug survival is difficult to estimate given the innovative nature of these treatments. Second, although important benefits of baricitinib treatment on sleep disturbance due to improvements in itch and skin pain have been observed in Phase 3 clinical trials,^19,20 skin pain and sleep disturbance have not been assessed in dupilumab clinical trials and were, therefore, not included in the ITC used to inform this CEA. These benefits may not be fully captured by the EQ-5D, given that the EQ-5D does not include domains for itch and sleep. Furthermore, benefits in relation to itch and sleep may also extend further to other aspects of life such as social and sexual functioning.⁷ In addition, baricitinib may provide added value to patients over and above that captured in this analysis; for example, oral administration (particularly in patients who cannot receive or fear injections). Third, discontinuation rates for baricitinib/BSC and dupilumab were derived from different sources and may therefore bias one treatment over another. Baricitinib/BSC discontinuation rates were derived from a Phase 3 clinical trial in adults with moderate to severe AD, the majority of whom had severe disease.²¹ In contrast, dupilumab discontinuation rates were derived from a real-world study in which disease severity was not reported (72% had received prior systemic corticosteroids but only 10% had received prior cyclosporin).³¹

Results of this CEA may not be generalizable to other populations. The analysis was based on data from a clinical trial and therefore has limitations compared with real-world populations; additionally, the analysis was conducted from an Italian healthcare payer perspective (including, for example, a target population reflective of the AIFA dupilumab population, age adjustments based on Italian life tables, treatment, AE, and resource costs derived from published Italian sources, and health state utilities based on an Italian value set) and therefore may not be generalizable to different countries. However, we expect that similar results would be seen in other countries where dupilumab is more costly than baricitinib.

Data supporting most treatments used for AD (including ciclosporin) are very limited. At the time of these analyses, dupilumab was the only drug currently reimbursed in Italy for the treatment of moderate to severe AD after cyclosporin failure, and baricitinib was the only reimbursed drug in this indication in other EU countries. Literature searches show that the two best evaluated drugs, with the strongest evidence for use in this indication, are baricitinib and dupilumab, although other drugs are being developed. For example, tralokinumab, upadacitinib, and abrocitinib are approved for the treatment of moderate to severe AD in Europe.^55–57 Thus, the options for this population have recently expanded, and the most suitable treatment options appear to be the JAK inhibitors and monoclonal antibodies against IL-4/IL-13 or IL-13. Relative cost-effectiveness may therefore be an important decision-making tool.

Conclusions

This CEA shows that dupilumab is more effective but also more costly from the Italian healthcare payer perspective compared with baricitinib in the treatment of patients with moderate to severe AD who have experienced failure on, are intolerant to, or have contraindication to ciclosporin. Based on ICERs determined in this patient population from the BREEZE-AD4 study, it is plausible to conclude that dupilumab cannot be considered cost-effective when compared with baricitinib in these patients. Given its oral administration, favorable risk/benefit profile and lower acquisition cost compared with dupilumab, baricitinib may offer a valuable, cost-effective treatment option, after failure on conventional systemic agents, for patients with moderate to severe AD in Italy.

Transparency

Declaration of financial/other relationships

Erin Johansson, Maurizio Mezzetti, Na Lu and Silvia Sabatino are full-time employees and may hold shares in Eli Lilly and Company. Massimo Giovannitti was a full-time employee of Eli Lilly and Company during the conception and development of the work.

Acknowledgements

None stated.

Reviewer disclosure statements

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Additional information

Funding

The authors would like to acknowledge Caroline Spencer and Sue Williamson (Rx Communications, Mold, UK) for medical writing assistance with the preparation of this manuscript, funded by Eli Lilly and Company. This work was funded by Eli Lilly and Company.