Cost-effectiveness of adding carfilzomib to lenalidomide and dexamethasone in relapsed multiple myeloma from a US perspective

Abstract

Objective: To assess the economic value of carfilzomib (Kyprolis), this study developed the Kyprolis Global Economic Model (K-GEM), which examined from a United States (US) payer perspective the cost-effectiveness of carfilzomib-lenalidomide-dexamethasone (KRd) versus lenalidomide-dexamethasone (Rd) in relapsed multiple myeloma (RMM; 1–3 prior therapies) based on results from the phase III ASPIRE trial that directly compared these regimens.

Methods: A partitioned survival model that included three health states of progression-free (on or off treatment), post-progression, and death was developed. Using ASPIRE data, the effect of treatment regimens as administered in the trial was assessed for progression-free survival and overall survival (OS). Treatment effects were estimated with parametric regression models adjusting for baseline patient characteristics and applied over a lifetime horizon. US Surveillance, Epidemiology and End Results (1984–2014) registry data were matched to ASPIRE patients to extrapolate OS beyond the trial. Estimated survival was adjusted to account for utilities across health states. The K-GEM considered the total direct costs (pharmacy/medical) of care for patients treated with KRd and Rd.

Results: KRd was estimated to be more effective compared to Rd, providing 1.99 life year and 1.67 quality-adjusted life year (QALY) gains over the modeled horizon. KRd-treated patients incurred $179,393 in total additional costs. The incremental cost-effectiveness ratio (ICER) was $107,520 per QALY.

Limitations: Extrapolated survival functions present the greatest uncertainty in the modeled results. Utilities were derived from a combination of sources and assumed to reflect how US patients value their health state.

Conclusions: The K-GEM showed KRd is cost-effective, with an ICER of $107,520 per QALY gained against Rd for the treatment of patients with RMM (1–3 prior therapies) at a willingness-to-pay threshold of $150,000. Reimbursement of KRd for patients with RMM may represent an efficient allocation of the healthcare budget.

Keywords:

• Carfilzomib
• Lenalidomide
• Dexamethasone
• Multiple myeloma
• Relapsed
• Cost-effectiveness

Introduction

Multiple myeloma (MM) is a rare disease that accounts for ∼1% of all cancers and is the second most common hematological malignancy^1,2. At diagnosis, nearly 40% of patients are under the age of 65 years³. MM is characterized by a pattern of remission and relapse and returns more aggressively and in a shorter period of time with each successive relapse, ultimately leading to death. Median overall survival (OS) after first relapse is ∼3–3.5 years in patients who have received a second-line therapy^4,5. Many patients with relapsed MM (RMM) have worsened health status as a result of disease progression, comorbidities and treatment-related toxicities⁶; patients’ health-related quality-of-life (HRQoL) also degrades as the disease progresses^7,8. Furthermore, the cost of managing MM increases as the disease progresses and worsens^9,10.

Given the relapsing nature of the disease, the goals of first and subsequent lines of treatment are to achieve the longest possible progression-free survival (PFS) with minimal treatment-related toxicities, thereby extending OS while maintaining or improving HRQoL. For the last decade, lenalidomide and/or bortezomib in combination with dexamethasone and/or other alkylating agents have been the most common treatment options for managing RMM. Real-world treatment pattern data from the US show that, in 2013–2014, lenalidomide- and bortezomib-based combinations represented the mainstay of treatment in patients with RMM after one prior therapy (second line), used in 37.4% and 34.7% of patients, respectively¹¹.

The clinical value for the addition of carfilzomib (Kyprolis) to lenalidomide plus dexamethasone (KRd) for patients with RMM has been clearly established and endorsed by National Comprehensive Cancer Network (NCCN) Guidelines¹². In ASPIRE, a phase III, multi-center, randomized, controlled trial, KRd, achieved an unprecedented median PFS of 26.3 months, an increase in median PFS of 8.7 months over lenalidomide plus dexamethasone (Rd) (hazard ratio [HR] for progression or death = 0.69; 95% confidence interval [CI] = 0.57–0.83; p = 0.0001). While median OS was not reached, the Kaplan–Meier 24-month OS rates showed a strong trend in favor of the KRd group (HR for death = 0.79; 95% CI = 0.63–0.99; p = 0.04). Almost nine of 10 patients (87.1%) responded to KRd and the rate of patients achieving a complete response was more than tripled (31.8% of patients in the KRd group vs 9.3% of patients in the Rd group), and was associated with longer PFS and OS in the trial. In addition to efficacy, HRQoL was significantly improved (p < 0.001) in the KRd arm compared with the Rd arm over the 18 cycles of carfilzomib treatment¹³. Importantly, a post-hoc secondary analysis of ASPIRE data showed that the clinical benefit of adding carfilzomib to Rd was consistent across prior lines of treatment (HR for improvement in PFS vs Rd was 0.69 for patients with one prior line of therapy and ≥2 prior lines of therapy)¹⁴, regardless of the cytogenetic risk status. The safety profile of KRd remained manageable relative to Rd with comparable rates of ≥ grade 3 adverse events (AEs) (83.7% of patients in the KRd group; 80.7% of patients in the Rd group) and comparable rates of discontinuation¹³. Based on ASPIRE data, the US Food and Drug Administration and the European Medicines Agency approved KRd for the treatment of relapsed MM and the NCCN Guidelines for MM recommended KRd as a preferred category 1 option for previously treated MM¹². In fact, KRd is the only regimen evaluated by NCCN Evidence Blocks™ to achieve the maximum score for efficacy (5/5) in this patient population¹⁵.

In recent times, the US has witnessed a heightened discussion of drug costs and their associated value, not only in policy and payer forums, but more broadly in the public media^16,17. Now, more than ever before, data on clinical efficacy and safety are necessary but not sufficient to define therapeutic value; economic analyses that quantify value over the long-term from a patient and societal perspective are essential. Cost-effectiveness analyses examine the overall value of a treatment in terms of incremental costs over incremental health benefits. Health benefits are generally expressed as quality-adjusted life-years (QALYs) and are calculated based on the duration of life adjusted for the quality-of-life. Results are expressed as incremental cost-effectiveness ratios (ICERs) and are evaluated against threshold values that typically vary by country. The use of direct data from comparative clinical trials in cost-effectiveness analyses, when available, is preferred over methods that make indirect comparisons derived by combining data from separate trials, given the large number of uncontrolled variables and heterogeneous patient populations across disparate studies. In order to demonstrate the economic value of carfilzomib from a US payer perspective, the Kyprolis Global Economic Model (K-GEM) compared KRd with Rd using the direct, head-to-head comparative data from the ASPIRE trial.

Patients and methods

Model description and structure

The K-GEM is a cohort-based, partitioned survival model with a 30-year horizon that was developed in Microsoft Excel. K-GEM represents patients in three health states: (1) progression-free (PF), (2) post-progression (PP), and (3) death (Figure 1). Initially, all patients are assumed to be in the PF state, in which they could be on or off treatment. Subsequently, patients may progress or die. In the PP health state, patients may receive subsequent line(s) of active treatment or best supportive care (BSC). Upon discontinuation of subsequent line(s) of active treatment, all patients switch to BSC. The distribution of patients among these health states was estimated using a partitioned survival approach, i.e. using parametric survival functions fitted on ASPIRE PFS and OS data to estimate the proportion of patients in the PF and death states in a given cycle. The length of a cycle in K-GEM was 4 weeks, to correspond with treatment cycles within the ASPIRE trial. The proportion of patients in PP was the difference between the proportion of patients in the death state and the proportion of patients in the PF state.

Figure 1. Schematic of how patients flow through the Kyprolis Global Economic Model (K-GEM).

Model parameters and data sources

Model population

The target population is patients with RMM who had received 1–3 prior therapies (i.e. mirroring the baseline characteristics of patients enrolled in the ASPIRE trial)¹³. Prior exposure to lenalidomide plus dexamethasone was allowed, as was prior exposure to bortezomib.

Data sources

The K-GEM incorporated data from the phase III ASPIRE trial, registry data from Surveillance, Epidemiology, and End Results (SEER), publicly available information from product labels, publicly available cost data, published literature, and expert opinions.

Regimens

In ASPIRE, patients received KRd or Rd in 28-day cycles. Carfilzomib was administered intravenously (IV) over 10 min on days 1, 2, 8, 9, 15, and 16 of cycles 1–12 and on days 1, 2, 15, and 16 during cycles 13–18, after which all remaining patients discontinued carfilzomib. The starting dose was 20 mg/m² on days 1 and 2 of cycle 1 and then 27 mg/m² thereafter. Patients with a body surface area (BSA) greater than 2.2 m² received a dose of carfilzomib based upon a 2.2-m² BSA (i.e. no more than a 60-mg vial was used, per administration) and this dose cap was also applied in K-GEM. For all cycles, lenalidomide (25 mg, oral) was administered on days 1–21. Dexamethasone (40 mg) was assumed in K-GEM to be orally administered on days 1, 8, 15, and 22¹³. Both lenalidomide and dexamethasone were administered until progression in each arm.

The ASPIRE study protocol was approved by the institutional review boards of all participating institutions and was conducted in accordance with the Declaration of Helsinki.

Treatment effect

Survival analyses for PFS and OS were conducted according to National Institute for Health and Care Excellence (NICE) guidelines¹⁸. ASPIRE OS data were relatively immature, with 58% and 54% of patients alive after 36 months in the KRd and Rd arm, respectively.

In order to test whether the treatment effect is proportional over time and the survival curves fitted to each treatment group have a similar shape, the proportional hazards assumptions were tested with log-cumulative hazard plots and log odds survival plots, and with tests of interaction between treatment effect and time with a Cox model¹⁹. Non-significant results would suggest that the null hypothesis could not be rejected and that the hazards could be assumed proportional. Once determined, joint parametric models would then be fitted to the entire dataset. Distinct parametric multiple regression models were generated with treatment effect as a covariate to estimate the difference between KRd and Rd on PFS and OS.

Probability of survival for the baseline Rd curve beyond the trial follow-up, specifically after the time of the last death event in the Rd arm, was derived by fitting a Weibull parametric distribution to real-world observational data from the US SEER registry (1984–2014). Data from SEER were matched to the ASPIRE population on age, gender, and time since MM diagnosis. Decade of diagnosis was controlled and results were adjusted to the most recent decade to account for the progressive availability of new treatments and better management of the disease.

Adverse events

AEs of grade 3 or greater severity were determined from the treatment-related AEs reported in ASPIRE. Those with incidence ≥2% in either treatment arm of the ASPIRE trial¹³ or identified based on clinical expert opinion as relevant in RMM (i.e. less frequent AEs with high impact to HRQoL and/or costs, such as cardiac failure and dyspnea) were included in K-GEM unless they were redundant (e.g. platelet count decreased was not included as AE, but thrombocytopenia was included) (Table 1). Monthly probabilities of experiencing a given AE were calculated from the percentages of patients experiencing an AE over the course of the ASPIRE trial and from the mean time on treatment in ASPIRE (KRd = 88.1 weeks; Rd =70.7 weeks)¹³. Patients were assumed to be at risk for AEs only while on treatment. The impact of AEs was evaluated in the ‘PF, on-treatment’ state as well as in the ‘PP, on subsequent treatment’ state.

Table 1. Percentages of patients receiving KRd or Rd reporting ≥ grade 3 adverse events over the course of the ASPIRE trial and associated costs.

Adverse events	KRd	Rd	Cost (USD)*
Blood and Lymphatic System Disorders
Neutropenia	27.04%	23.39%	8593
Anemia	8.42%	9.25%	5734
Thrombocytopenia	14.29%	10.03%	5020
Leukopenia	2.55%	3.60%	8593
Febrile Neutropenia	2.04%	0.77%	23,392
Lymphopenia	2.55%	1.54%	8593
Gastrointestinal Disorders
Diarrhea	1.53%	2.06%	1774
Nausea	0.26%	0.77%	3242
Constipation	0.26%	0.26%	5885
Cardiac Disorders
Cardiac failure congestive	0.51%	N/A	5949
Vascular Disorders
Hypertension	0.77%	0.77%	2456
Thrombo-Embolic Event†	5.10%	4.11%	14,182‡
Respiratory, Thoracic, and Mediastinal Disorders
Dyspnea	1.53%	0.26%	1185
General Disorders and Administration Site Conditions
Fatigue	5.87%	4.63%	2038
Asthenia	2.55%	2.06%	2038
Metabolism and Nutrition Disorders
Hypophosphatemia	5.10%	1.29%	2555
Nervous System Disorders
Neuropathy Peripheral	1.53%	1.54%	1708
Renal and Urinary Disorders
Renal failure†	1.28%	0.51%	9304
Infections and Infestations
Infections and Infestations   (including pneumonia)†	10.46%	8.74%	1192

* Costs for constipation and renal failure were from the H⋅CUPnet National (Nationwide) Inpatient Sample (NIS) 2013²⁴; costs for all other adverse events were from Amgen Inc. (data on file)²⁵.

† Represents a pooling of preferred terms.

‡ Calculated as a weighted average of the costs for ‘Deep vein thrombosis’ and for ‘Pulmonary embolism’ in the source, using the respective numbers of events in ASPIRE (with KRd and Rd arms pooled) as weights.

KRd, Carfilzomib plus lenalidomide plus dexamethasone; Rd, lenalidomide plus dexamethasone; USD, United States dollars.

Treatment discontinuations

Discontinuation of treatments due to reasons other than progression was modeled for each of the individual components of both regimens. A patient could discontinue one or two of the components of the regimen while remaining on the other component(s) (e.g. in the KRd arm, a patient could discontinue carfilzomib while staying on lenalidomide and dexamethasone). Exponential distributions were fitted to observed time to treatment discontinuation data in ASPIRE.

Subsequent treatments

The impact of subsequent therapy was captured in K-GEM from a cost perspective and also from an efficacy perspective, implicitly by relying on the OS data from the trial that allowed subsequent therapies. Upon disease progression, after a treatment-free interval of three cycles as observed in ASPIRE, patients received subsequent active treatment(s) as observed in each arm of the ASPIRE trial (Table 2). The dosing schedules applied for each subsequent treatment are presented in Table 3. The relative dose intensity (i.e. the ratio of actual dose vs planned dose) was assumed to be 100% for all subsequent treatments, and overall duration for subsequent therapies was assumed to be a maximum of 17 cycles for both comparators, based on data from a chart-audit conducted across seven European countries²⁰. We considered a scenario where carfilzomib, pomalidomide plus dexamethasone, bortezomib plus dexamethasone, and daratumamab were distributed equally as subsequent treatment options for patients who progress.

Table 2. Subsequent therapies.

n	KRd	Rd
	396	396
Proportion of progressing patients receiving active  subsequent treatment(s)	74.38%	73.60%
Proportion of patients receiving Best  Supportive Care	25.62%	26.40%
Proportion of patients receiving each  subsequent treatment*
Bortezomib	37.09%	58.70%
Cyclophosphamide	33.77%	35.87%
Doxorubicin	13.25%	13.59%
Melphalan	10.60%	9.78%
Bendamustine	7.95%	5.98%
Etoposide	6.62%	5.98%
Cisplatin	6.62%	4.89%
Pomalidomide	7.95%	6.52%
Carfilzomib	5.96%	4.35%
Lenalidomide	18.54%	14.67%
Thalidomide	10.60%	6.52%
Dexamethasone	70.20%	70.65%
Prednisone	9.27%	13.59%

* The sum of proportions for a given treatment arm (i.e. KRd or Rd) may exceed 100% due to patients receiving more than one treatment at the same time (e.g. combination therapies) and due to the aggregated approach (patients may receive several lines of subsequent treatments).

KRd, Carfilzomib plus lenalidomide plus dexamethasone; Rd, lenalidomide plus dexamethasone.

Table 3. Subsequent-line(s) drug costs.

Drug	Average dose per administration	No. of administrations in treatment cycle	Length of treatment cycle (weeks)	Cost per tablet/capsule/vial (USD)	Drug cost per model cycle (USD)	Sources*
Bortezomib	1.3 mg/m^2	5.8	6	$1610.00 (3.5 mg vial)	$4468	Velcade FDA label (FDA, 2014)^47
Cyclophosphamide	3.0 mg/kg	28	4	$2.22 (25 mg tablet) $4.09 (50 mg tablet)	$543	Cyclophosphamide FDA label (FDA, 2013a)^48
Doxorubicin	57.5 mg/m^2	1	3.5	$538.95 (10 mg vial) $1077.90 (20 mg vial) $2694.75 (50 mg vial)	$6617	Doxorubicin hydrochloride FDA label (FDA, 2013b)^49
Melphalan	6.0 mg	14	6	$10.58 (2 mg tablet)	$296	Alkeran FDA label (FDA, 2007)^50
Bendamustine	120.0 mg/m^2	2	3	$639.50 (25 mg vial) $2557.98 (100 mg vial)	$15,301	pCODR, 2012^51
Etoposide	150.0 mg	4.5	3.5	$85.37 (50 mg capsule)	$1317	VePesid FDA label (FDA, 2004)^52
Cisplatin	60.0 mg/m^2	1	3.5	$15.00 (50 mg vial) $30.00 (100 mg vial)	$39.20	CISplatin FDA label (FDA, 2015a)^53
Pomalidomide	4.0 mg	21	4	$582.22 (1, 2, 3, or 4 mg capsules)	$12,227	Pomalyst FDA label (FDA, 2013c)^54
Carfilzomib	27.0 mg/m^2	6	4	$1861.95 (60 mg vial)	$9525	Assumed same as in KRd arm in ASPIRE, for cycles 2–12
Lenalidomide	25.0 mg	21	4	$506.29 (5, 10, 15, or 25 mg capsules)	$10,632	Assumed same as in KRd or Rd arms in ASPIRE
Thalidomide	200.0 mg	21	3	$11.28 (200 mg capsule)	$316	Assumed same as VTd arm in MMVAR trial^55
Dexamethasone	40.0 mg	4	4	$0.07 (0.5 mg tablet)	$22.40	Assumed same as in KRd or Rd arms in ASPIRE
Prednisone	5.0 mg	28	4	$0.24 (1 mg tablet) $0.47 (5 mg tablet)	$13.16	Rayos FDA label (FDA, 2012b)^56

* The source for drug costs and available strengths is the Truven Health Analytics’ RED BOOK²¹.

FDA, Food and Drug Administration; KRd, carfilzomib plus lenalidomide plus dexamethasone; pCODR, pan-Canadian Oncology Drug Review; Rd, lenalidomide plus dexamethasone; VTd, bortezomib plus thalidomide plus dexamethasone.

Direct costs

The K-GEM considered the total direct costs (pharmacy and medical) of care for patients treated with KRd and Rd, while indirect costs were not included. The following cost components were included: drug costs, costs for treatment of AEs, drug administration and routine monitoring costs, subsequent treatment (PP), and associated AE costs. All costs were estimated from a US payer perspective in 2015 US dollars (USD).

Drug costs

Medication prices were estimated with the average wholesale acquisition cost (WAC) obtained from Truven Health Analytics’ RED BOOK²¹; the average wholesale price (AWP) was used when WAC prices were unavailable. To calculate drug costs, mean weight or BSA of patients, available strengths for a vial/capsule/tablet, price of a pack, number of vials/capsules/tablets in a pack and the average dose per administration and their respective subsequent dose escalations were considered. In the case of drugs for which the dose administered depends on the weight or BSA of patients, drug costs were estimated per actual administered dose for ∼90% of patients and, for the remaining 10% of patients, drug costs were rounded up to include the actual drug administered and any unused drug remaining in the vial based on the available vial size(s). The dosing for each comparator was based on the ASPIRE trial. Information on the dosing for each treatment, along with the unit cost and cost per cycle for each drug, are presented in Table 4.

Table 4. KRd and Rd drug costs.

Treatment regimen	Regimen components	Planned regimen	Cost per tablet/capsule/vial (USD)	Dose intensity	Drug cost per model cycle (USD)
KRd	Carfilzomib (Cycle 1)	20 mg/m^2 via IV on days 1 and 2, escalating to 27 mg/m^2 on days 8, 9, 15, and 16 of Cycle 1	$1861.95	92%	$8047.20
	Carfilzomib (Cycles 2–12)	27 mg/m^2 IV on days 1, 2, 8, 9, 15, and 16	$1861.95	92%	$8715.69
	Carfilzomib (Cycles 13–18)	27 mg/m^2 IV on days 1, 2, 15, and 16	$1861.95	90%	$5734.26
	Carfilzomib (Cycle 19+)	Not administered	$1861.95	—	0
	Lenalidomide	25 mg orally on days 1–21	$506.29	81%	$8558.83
	Dexamethasone	40 mg orally on days 1, 8, 15, and 22	$0.07	85%	$19.11
Rd	Lenalidomide	25 mg orally on days 1–21	$506.29	83%	$8803.37
	Dexamethasone	40 mg orally on days 1, 8, 15, and 22	$0.07	86%	$19.24

IV, Intravenous; KRd, Carfilzomib plus lenalidomide plus dexamethasone; Rd, lenalidomide plus dexamethasone; USD, United States dollars.

Administration and routine monitoring

Administration costs were based on the unit costs for IV and subcutaneous administration costs derived from the 2015 Centers for Medicare & Medicaid Services (CMS) National Physician Fee Schedule ($100.7 IV administration [healthcare common procedure coding system (HCPCS) code 96409] for carfilzomib, $125.37 IV administration [HCPCS code 96413] for doxorubicin and bendamustine, $153.61 IV administration [HCPCS codes 96413 + 96415] for cisplatin; $64.58 subcutaneous administration [HCPCS code 96401] for bortezomib). For carfilzomib, the administration costs were adjusted to account for skipped doses when no IV drug was given based on ASPIRE data.

It was assumed that all patients would require periodic routine monitoring in both PF and PP states, which included office visits, laboratory tests, and procedures. The pre-progression monthly monitoring costs were estimated to be $188.91 (whether on or off treatment) and the post-progression monthly monitoring costs to be $445.03 (whether on active subsequent treatment or on BSC), based on resource use data reported in a recent publication²². The resource use costs for on-treatment patients were assumed to be the same across both comparators, and all unit costs were based on the national limit price from the CMS 2014 fee schedule. Additional resource use data were considered on the basis of clinical expert opinion: outpatient office visit at a frequency of 16 times/year in PF, 0 in PP, with a unit cost of $42.12⁵⁸; and serum free light chain assay at a frequency of 12 times/year in PF, 0 in PP, with a unit cost of $18.54^22,23.

Adverse event costs

The costs for treatment-related AEs associated with KRd and Rd, as well as with the subsequent treatments, were included in K-GEM and were obtained from two sources^24,25. Costs for constipation and renal failure were from the H⋅CUPnet National (Nationwide) Inpatient Sample (NIS) 2013²⁴; and costs for all other AEs were from a retrospective observational study using the MarketScan database that identified adult patients with MM (ICD-9-CM code 203.0x) who initiated MM treatment between January 1, 2006 and December 31, 2014²⁵. AEs and associated costs while on MM treatment were identified using diagnosis/procedure and medication claim codes. It was assumed that the cost of managing a grade 3 event was the same as the cost of managing a grade 4 event of the same type, and that cost would be independent of the course of the disease (e.g. pre-progression neutropenia costs are the same as post-progression neutropenia costs). A summary of mean costs by AE is presented in Table 1. For each regimen, the average cycle costs of an AE were calculated from the proportion of patients having AEs in each model cycle and from the costs for the management of the AE.

Subsequent treatment costs

Prices for subsequent treatments were obtained from Truven Health Analytics’ RED BOOK, as described above. Information on the unit cost and the cost per model cycle for each subsequent treatment included in K-GEM are presented in Table 3.

Health-state utilities

Since in the ASPIRE trial preference-based utility data were not collected, utility input parameters for the model were estimated by using external information and trial-based patient-reported outcomes. In particular, in the PF health state, cycle-specific utility estimates were derived by combining two sources of information (i.e. ‘reference’ utility value and adjustment factor). The ‘reference’ utility value (0.81, irrespective of time since treatment initiation and the treatment itself), which has been used in several previous economic evaluations for treatments in MM^26,27, was derived directly from the EuroQol five dimensions questionnaire (EQ-5D) collected in patients with MM²⁶. The reference utility value was then adjusted by an adjustment factor to estimate time- and treatment arm-dependent utilities. Adjustment factors were estimated in two steps. First, patient-reported EORTC QLQ-C30 data gathered in the ASPIRE trial were mapped onto EQ-5D for both treatment arms by applying an existing relevant mapping algorithm²⁸. The mapping function had been derived using a relapsed/refractory MM patient population comparable to the ASPIRE trial population, and has been employed in a similar context before²⁹. Second, the relative difference in mapped utilities between cycle 1 (baseline) and subsequent assessments (cycles 3, 6, 12, and 18) was used to calculate time- and treatment-specific adjustment factors, i.e. the ratio of predicted utility beyond baseline to the utility at baseline. Utilities in the pre-progression, off-treatment state were assumed to be equal to the utilities estimated in cycle 18. The same difference between pre- and post-progression utilities (0.17) was adopted for the K-GEM as it was published for the reference utility source²⁶. That is, post-progression utilities were estimated by subtracting this difference (0.17) from the weighted average utility in the PF state (cycles 1–18) where the treatment-specific weighted average utility was calculated by adjusting for the proportion of patients in the PF state at each assessment in ASPIRE. Resulting utility values and adjustment factors applied by cycle in K-GEM are reported in Table 5.

Table 5. Utility values used in K-GEM.

State	Reference utility^26	Adjustment factor*		Utility used in K-GEM†
		KRd	Rd	KRd	Rd
Pre-progression (Cycle 1)	0.810	1.00	1.00	0.810	0.810
Pre-progression (Cycle 3)	0.810	1.01	0.98	0.818	0.798
Pre-progression (Cycle 6)	0.810	1.02	1.00	0.829	0.808
Pre-progression (Cycle 12)	0.810	1.04	1.01	0.840	0.818
Pre-progression (Cycles 18+)‡	0.810	1.05	1.02	0.851	0.829
Post-progression	0.64			0.664†	0.643§

* The pre-progression adjustment factor was calculated as a ratio of time-dependent utility vs baseline utility from the Proskorovsky et al.²⁸ mapping algorithm for each of the KRd and Rd treatment arms in ASPIRE.

† Since HRQoL data in ASPIRE were collected at certain fixed intervals (Cycles 1, 3, 6, 12, 18, and end-of-treatment visit), mapped utilities between those cycles in K-GEM were calculated using an interpolation (e.g. utility in cycle 2 was the result of the interpolation between utility in cycle 1 and utility in cycle 3).

‡ Utilities in the pre-progression but off treatment state were assumed to be equal to the utilities estimated in cycle 18.

§ The post-progression utility was estimated by subtracting the disutility related to progression²⁶ (0.17) from the weighted average utility in PFS state by arm (cycles 1–18, as adjusted for the present analyses). The latter was calculated considering the relative proportion of patients in PF at each time interval in ASPIRE.

K-GEM, Kyprolis Global Economic Model; KRd, Carfilzomib plus lenalidomide plus dexamethasone; Rd, lenalidomide plus dexamethasone.

The impact of AEs on HRQoL was considered as part of this evaluation. A utility decrement associated with an AE and the average duration of this AE were identified from the data reported in the Evidence Review Group (ERG) report for the NICE technology appraisal of pomalidomide (Table 6)³⁰. The disutility per model cycle for each grade ≥3 AE for each treatment was obtained by multiplying the AE disutility per event by its per-cycle probability.

Table 6. Adverse event-related utility decrement values in K-GEM.

Adverse event	Disutility^30^,^57	Duration (days)	Duration-adjusted utility decrement (decrement per event)*
Blood and Lymphatic System Disorders
Neutropenia	0.145	13.2	0.005
Anemia	0.250	10.7	0.007
Thrombocytopenia	0.008	14.1	0.000
Leukopenia	0.000	15.5	0.000
Febrile Neutropenia	0.150	9.4	0.004
Lymphopenia	0.000	15.5	0.000
Gastrointestinal Disorders
Diarrhea	0.103	12.0	0.003
Nausea	0.103	24.3	0.007
Constipation	0.103	12.4	0.003
Cardiac Disorders
Cardiac failure congestive	0.063	21.0	0.004
Vascular Disorders
Hypertension	0.000	28.0	0.000
Thrombo-Embolic Event	0.063	21.0	0.004
Respiratory, Thoracic, and Mediastinal Disorders
Dyspnea	0.000	11.0	0.000
General Disorders and Administration Site Conditions
Fatigue	0.000	14.6	0.000
Asthenia	0.000	22.0	0.000
Metabolism and Nutrition Disorders
Hypophosphatemia	0.000	11.4	0.000
Nervous System Disorders
Neuropathy Peripheral	0.065	8.0	0.001
Renal and Urinary Disorders
Renal failure	0.110^57	29.8	0.009
Infections and Infestations
Infections and Infestations   (including pneumonia)	0.140	12.0	0.005

* Calculated by multiplying the expected disutility for each adverse event by its duration in days divided by the number of days in a year (365.25).

K-GEM, Kyprolis Global Economic Model.

Analyses

In each 4-week cycle, the K-GEM generated outcomes that were aggregated to estimate life-years (LYs), progression-free life-years (PFLYs), quality-adjusted progression-free life-years (QAPFLYs), QALYs and lifetime costs (PF + PP) for KRd and Rd over a 30-year horizon. The ICER was calculated as incremental costs per QALY gained. Future costs and benefits were discounted at a 3% rate per year.

Univariate deterministic sensitivity analyses were conducted to test the effects of parameter uncertainty within K-GEM. The key model parameters were varied using 95% CIs or a standard error (SE) based on empirical data, where available. In the absence of data, ±20% of the base case values were used.

Probabilistic sensitivity analyses were conducted using 1000 second-order Monte Carlo simulations, such that probability distributions were assigned to the relevant model parameters.

All analyses were conducted from a US payer perspective. A threshold of $150,000 per QALY gained was used for the probabilistic analyses^31,32.

In addition to the primary analysis per the parameters described for the intent-to-treat (ITT) population included in ASPIRE, the following scenarios were examined: (1) Add-on therapy costs, i.e. only carfilzomib costs were considered, since the cost-effectiveness of Rd has already been evaluated previously^33,34; (2) HR for OS from primary ASPIRE publication¹³; (3) utilities modeled to be equal for KRd and Rd; (4) an analysis by number of prior therapies (1 prior and ≥2 prior therapies) using the results from the parametric regression models; (5) discount rate of 5% instead of 3% for costs and for benefits; (6) model horizon of 20 years instead of 30 years; (7) drug costs rounded up to include cost for calculated dose and any unused drug in the vial for 100% of patients; and (8) 100% relative dose intensity for carfilzomib as opposed to the dose intensity in the ASPIRE trial.

Results

The results from the test of the interaction between treatment effect and time (p = 0.08 for PFS; p = 0.41 for OS), and visual examination of the log-cumulative hazard plots and log odds survival plots (parallel curves, data on file³⁵), suggested that proportional hazards and proportional odds assumptions were valid. Consequently, the KRd and Rd arms could be jointly fitted, assuming the same parametric distribution.

In addition to treatment effect, the parametric models retained pre-specified baseline covariates that were significant, as well as those considered to be influential prognostic factors (i.e. age, time since last prior regimen, prior stem cell transplant status and number of prior regimens) for PFS and OS. No evidence of treatment-covariate interaction was detected for PFS or OS, which confirmed a consistent treatment effect across sub-groups. PFS was modeled with a log-logistic parametric distribution and OS was modeled with a Weibull parametric distribution, since these distributions had the best fit visually (Figure 2) as well as statistically, based on Akaike information criterion (AIC) and Bayesian information criterion (BIC) statistics (Table 7). The treatment effect estimated from parametric regression models with KRd relative to Rd was statistically significant for both OS (HR of the Weibull multiple parametric model = 0.70, 95% CI = 0.53–0.87) and PFS (odds ratio [OR] of the log-logistic multiple parametric model = 0.51, 95% CI = 0.38–0.67).

Figure 2. Progression-free survival and overall survival parametric distributions. KM, Kaplan–Meier; KRd, carfilzomib plus lenalidomide plus dexamethasone; OS, overall survival; PFS, progression free survival; Rd, lenalidomide plus dexamethasone; SEER, Surveillance, Epidemiology, and End Results.

Table 7. AIC and BIC values for all parametric multiple regression models.

Parametric multiple model	OS		PFS
	AIC	BIC	AIC	BIC
Weibull	2879.526	2967.383	3616.997	3700.230
Log-normal	2896.742	2984.599	3627.049	3710.282
Log-logistic	2881.649	2969.506	3613.262	3696.495
Exponential	2891.937	2975.170	3621.829	3700.438
Generalized gamma	2881.182	2973.663	3613.726	3701.583
Gompertz	2879.669	2967.527	3620.746	3703.979

AIC, Akaike information criterion; BIC, Bayesian information criterion; OS, overall survival; PFS, progression-free survival.

Primary analysis

The ICER with KRd over Rd was $107,520, and the incremental cost per LY gained was $89,957 (Table 8). Patients on KRd incurred $483,845 in lifetime costs compared with $304,452 for patients on Rd. The QALYs gained were 5.88 and 4.21 for patients on KRd and Rd, respectively. It was estimated that patients treated with KRd benefited from an additional 1.99 LYs (7.83 vs 5.84), 1.20 PFLYs (3.79 vs 2.59), and 1.67 QALYs (5.88 vs 4.21) compared with patients on Rd. The pre-progression drug costs for the multi-year KRd regimen were $356,041; 30.8% ($109,779) was attributable to carfilzomib (which was administered for a maximum of 18 cycles or until progression, whichever occurred first) and the remainder, 69.2% ($246,262), was attributable to Rd (administered until progression). Subsequent therapy drug costs in the PP state were not very different in the KRd ($73,355) and Rd arms ($76,018).

Table 8. Primary analysis results.

	KRd	Rd
Effectiveness outcomes
PFLYs	3.79	2.59
QAPFLYs	3.20	2.13
LYs	7.83	5.84
QALYs	5.88	4.21
Cost outcomes (USD)
Treatment costs	363,432	189,923
Drug costs* (Total)	356,041	189,923
Drug costs - carfilzomib	109,779	—
Drug costs - Rd	246,262	189,923
Administration costs* (Total)	7391	0
Administration costs - carfilzomib†	7391	—
Administration costs - Rd	0	0
AE costs (Total)	15,097	13,087
Progression-free AE costs	8445	6131
Post-progression AE costs	6652	6956
Subsequent treatment costs – PP (Total)	75,141	78,236
Drug costs	73,355	76,018
Administration costs	1786	2218
Monitoring costs (Total)	30,175	23,205
On treatment	5103	3741
Off treatment - PF	3492	2139
Post-progression	21,580	17,324
Total Pre-progression costs	380,472	201,934
Total Post-progression costs	103,373	102,517
Total Lifetime costs	483,845	304,452
Incremental Results
Incremental PFLYs		1.20
Incremental QAPFLYs		1.07
Incremental LYs		1.99
Incremental QALYs		1.67
Incremental costs (USD)		179,393
Incremental Cost-Effectiveness Ratios
Incremental cost per PFLY (USD)		149,834
Incremental cost per QAPFLY (USD)		167,379
Incremental cost per LY (USD)		89,957
Incremental cost per QALY (USD)		107,520

* Drug costs and administration costs were accrued over the following average times on treatment (for a 30-year time horizon):

— Carfilzomib: 15.1 cycles (13.9 months).

— Lenalidomide, in KRd arm: 31.2 cycles (28.7 months); in Rd arm: 23.1 cycles (21.3 months).

— Dexamethasone, in KRd arm: 30.0 cycles (27.6 months); in Rd arm: 21.8 cycles (20.1 months).

† For carfilzomib, the administration costs were adjusted to account for skipped doses when no IV drug was given based on ASPIRE data.

AE, adverse event; KRd, carfilzomib plus lenalidomide plus dexamethasone; LYs, life-years; PF, progression-free; PFLYs, progression-free life-years; PP, post-progression; QALYs, quality-adjusted life-years; QAPFLYs, quality-adjusted progression-free life-years; Rd, lenalidomide plus dexamethasone; USD, United States Dollars.

Sensitivity analyses

Results of univariate deterministic sensitivity analyses are presented in a form of a Tornado diagram (Figure 3). The ICER was most influenced by the HR for OS beyond the trial duration, followed by the drug costs per cycle of lenalidomide in the KRd regimen, the drug costs per cycle of lenalidomide in the Rd regimen, the HR for OS over the trial period, the drug costs per cycle of carfilzomib and the OR for PFS. The model results were less sensitive to the utility values.

Figure 3. Tornado diagram illustrating results of deterministic sensitivity analyses. KRd, carfilzomib plus lenalidomide plus dexamethasone; OS, overall survival; PFS, progression-free survival; QALYs, quality-adjusted life-years; Rd, lenalidomide plus dexamethasone.

The results of the probabilistic sensitivity analysis (PSA) are presented next. As shown, 90.6% of the simulations resulted in a positive ICER, indicating that KRd was more effective and more costly than Rd (North-East quadrant in Figure 4). The cost-effectiveness acceptability curve from K-GEM (Figure 5) showed that, at a willingness-to-pay (WTP) level of $150,000 per QALY gained, the estimated probability of KRd being cost-effective was 61.9%. At $250,000 per QALY gained, the estimated probability of KRd being cost-effective was 76.8%.

Figure 4. Incremental cost-effectiveness plane illustrating the probabilistic sensitivity analysis results. ICER, Incremental cost-effectiveness ratio; QALYs, Quality adjusted life-years.

Figure 5. Cost-effectiveness acceptability curve. QALY, Quality adjusted life-year.

Scenario analyses

Results from the scenario analyses are summarized in Table 9. Consideration of carfilzomib costs only resulted in the ICER being below $75,000/QALY gained. Reduction of the time horizon or increase of the discount rates both increased the ICER to values just below $130,000. When assuming that all unused drug was costed, 100% dose intensity for carfilzomib, or equivalent utilities in both treatment arms considered, the ICER was increased by ∼$10,000 relative to the primary analysis ($107,520). An alternative scenario regarding survival modeling had the greatest impact on the estimated outcomes. When the HR from the primary ASPIRE publication was used for OS, the LY and QALY estimates for both arms were reduced and the ICER increased to $139,682/QALY gained. Results for the patient group with one prior line of therapy, who represented 43% of the ASPIRE population, were in line with the overall population, resulting in an ICER of $101,914/QALY gained. The ICER for the group of patients with ≥2 prior therapies was relatively higher ($130,138). In a scenario where carfilzomib, pomalidomide plus dexamethasone, bortezomib plus dexamethasone and daratumamab each had 25% share for patients who progress, the ICER was estimated to be $106,512.

Table 9. Scenario analyses results.

Scenarios	KRd					Rd					KRd vs Rd
	PFLYs	LYs	QALYs	Pre-progression costs (USD)	Post-progression costs (USD)	PFLYs	LYs	QALYs	Pre-progression costs (USD)	Post-progression costs (USD)	Incremental LYs	Incremental QALYs	Incremental total costs (USD)	ICER (USD/QALY gained)
Base case	3.79	7.83	5.88	380,472	103,373	2.59	5.84	4.21	201,934	102,517	1.99	1.67	179,393	107,520
Carfilzomib costs only*	3.79	7.83	5.88	134,210	103,373	2.59	5.84	4.21	12,011	102,517	1.99	1.67	123,054	73,753
OS HR from primary ASPIRE publication	3.79	7.08	5.38	380,472	92,120	2.59	5.74	4.15	201,934	98,561	1.34	1.23	172,097	139,682
1 prior line of therapy	4.06	7.38	5.63	371,761	95,418	2.79	5.47	4.01	206,092	95,948	1.91	1.62	165,139	101,914
2 or more prior lines of therapy	3.60	8.18	6.08	389,262	108,855	2.46	6.13	4.38	169,841	107,198	2.05	1.70	221,078	130,138
Utilities are equal for KRd and Rd arms	3.79	7.83	5.72	380,472	103,373	2.59	5.84	4.21	201,934	102,517	1.99	1.50	179,393	119,389
Discount rate of 5% for both costs and utilities	3.45	6.77	5.11	368,065	92,943	2.40	5.18	3.75	195,465	94,845	1.59	1.36	170,699	125,872
Time horizon – 20 years	3.62	7.15	5.40	380,065	99,519	2.51	5.54	4.00	201,738	100,790	1.62	1.40	177,056	126,881
Costs for administered and unused drug for all patients	3.79	7.83	5.88	400,218	111,502	2.59	5.84	4.21	201,934	114,391	1.99	1.67	195,396	117,111
100% dose intensity for carfilzomib	3.79	7.83	5.88	390,992	103,373	2.59	5.84	4.21	201,934	102,517	1.99	1.67	189,914	113,826

* Rd drug costs were excluded from both arms.

AE, adverse event; HR hazard ratio; KRd, carfilzomib plus lenalidomide plus dexamethasone; LYs, life-years; OS, overall survival; PFLYs, progression-free life-years; PP, post-progression; QALYs, quality-adjusted life-years; Rd, lenalidomide plus dexamethasone; USD, United States Dollars.

Discussion

In the pivotal phase III ASPIRE trial, KRd achieved greater than 2 years of PFS (median PFS = 26.3 months) and was superior to Rd for the treatment of patients with RMM who had received 1–3 prior therapies. This magnitude of absolute PFS has been previously observed only among newly-diagnosed patients with previously untreated myeloma. The efficacy of carfilzomib was also clearly demonstrated in ENDEAVOR, the first ever head-to-head, phase III trial of two proteasome inhibitors in RMM. Median PFS was doubled with carfilzomib plus dexamethasone (Kd) compared with bortezomib plus dexamethasone (Vd) (18.7 months vs 9.4 months; HR = 0.53, 95% CI = 0.44–0.65; p < 0.0001)³⁶. In addition to the magnitude of PFS benefit achieved with carfilzomib, the efficacy of KRd was consistent across sub-groups (e.g. prior exposure to bortezomib, number of prior therapies, age, risk group)¹⁴, and its safety profile was manageable, thus offering clinicians a highly potent option to tailor treatment according to the needs of the individual patient.

To assess the economic value of carfilzomib, the primary analysis focused on the ITT population data from the ASPIRE clinical trial, which directly compared KRd and Rd in patients with RMM. From the primary analysis, we see that the ICER is $107,520, and the incremental cost per LY gained is $89,957. The model-estimated total LYs achieved with KRd and Rd are 7.83 and 5.84, respectively, with an estimated gain of nearly 2 LYs with KRd.

Interestingly, the estimated total survival for Rd based on the K-GEM is similar to that estimated in a life-time cost-effectiveness (CE) model from the UK National Health Service (NHS) perspective, which showed that Rd achieved a total of 5.37 life-years³⁴. These data serve as a source of external validation for the approach utilized in the K-GEM to extrapolate survival data beyond the trial with the use of SEER registry data. The SEER data are generalizable to the US population, enhancing the relevance of the model findings from the perspective of both patients and payers in the US.

While Rd is a common treatment for patients with RMM, it is a relatively novel agent. In this case, where an innovative product such as carfilzomib is added onto an existing regimen (Rd) that already contains an innovative product, the cost of the two products simply added together could generate the perception that KRd is an expensive regimen. However, patients, when treated with KRd, receive carfilzomib until cycle 18 only, and, furthermore, the number of doses per cycle beyond cycle 12 are 33% fewer than the number of doses up to cycle 12. Therefore, carfilzomib does not represent a large share of the drug costs during the PF state. In the K-GEM, carfilzomib represented 30.8% of the total pre-progression drug costs, while a majority of the drug costs (69.2%) were represented by Rd. Consequently, the majority of the KRd drug costs are attributable to a historical base treatment generally considered cost-effective. Therefore, while KRd was highly cost-effective in the primary analysis, the ICER improved dramatically in the scenario that considered only carfilzomib costs during the pre-progression period.

In other scenario analyses, we demonstrated the robustness of the ICERs for KRd over Rd from the K-GEM by conducting an analysis with the HR for OS from the primary publication, and by examining extreme scenarios for key parameters. In particular, extreme scenarios assumed equal utilities for both regimens (despite a clear significant difference in HRQoL between KRd and Rd), accounted for the cost of any unused drug, considered the maximum dose intensity for carfilzomib (i.e. patients received 100% of the planned doses, which is rarely realized in real-world settings), used a 5% discount rate that is greater than usual, and applied a significantly shorter time horizon (20 years). In each case, the incremental cost per QALY increased as expected, but remained within accepted thresholds.

The results for the analysis by number of prior therapies should be interpreted with the consideration that the population of interest in the trial, and the resulting indication approved by the FDA for KRd, included patients who have received 1–3 prior therapies. In general, the results did not deviate considerably from the primary analysis highlighting the consistency of the treatment effect achieved with KRd vs Rd for patients who had received 1 or ≥2 prior therapies. The model predicted longer life expectancy for patients with ≥2 prior therapies since this specific ASPIRE population was, on average, more than 2 years younger than the population with one prior therapy (63.0 vs 65.2 years, respectively). In contrast to the K-GEM, an analysis by the Institute for Clinical and Economic Review examining the cost-effectiveness of KRd v Rd via an indirect comparison³⁷ concluded that the ICER was higher than reported here, but the lowest among recently approved triplet combinations (KRd, elotuzumab-Rd, ixazomib-Rd). The indirect comparison did not address patient heterogeneity across trials, nor did it consider differences in definitions for influential trial parameters (for e.g. cytogenetic risk). The analysis utilized baseline Rd PFS estimates from the MM-009/010^38,39 trials, which were considerably lower than those observed in the more recent trials^13,40,41 to project PFS and OS for the standard of care (Rd), which led to an under-estimation of the gain in both PFS and OS for the novel regimens including KRd. Furthermore, the analysis did not appropriately value HRQoL outcomes by treatment; thereby under-estimating total QALYs for treatments such as KRd, which showed an improvement in HRQoL over Rd in ASPIRE¹³. Collectively, these limitations resulted in an over-estimation of the ICER for KRd in ICER’s analysis, which have come under severe scrutiny and criticism from a number of organizations representing MM clinicians such as the American Society of Hematology (ASH) and patient support groups including Cancer Support Community and the Cutaneous Lymphoma Foundation.

In the primary analysis, we concluded that KRd was cost-effective vs Rd, on the basis of a $150,000 threshold, which was recently recommended as a threshold for decision-makers in the US to anchor upon³². However, it should be noted that, in the US, a broad range of ICER thresholds have been proposed ($150,000–$300,000 per QALY gained)^31,42, although no strict limits have been set. Based on the WHO criterion, an ICER threshold that is 3-times the gross domestic product (GDP) per capita, the US threshold (with a GDP per capita of $50,000) would be $150,000/QALY gained³¹. US physicians have valued a treatment on top of standard of care at $125,000 if it yielded 6 additional months of life, which results in a threshold of $250,000 per life-year gained⁴³. Consistent with this, another survey of oncologists showed that the clinicians valued 1 QALY to be ∼$300,000⁴⁴. Additionally, studies based on labor market data have concluded that the willingness of consumers to pay for reducing mortality risk is estimated to be at $300,000 per LY⁴⁵.

At the current WTP thresholds in the US, this analysis demonstrates that reimbursement of KRd for the treatment of patients with RMM who have received 1–3 prior therapies represents an efficient allocation of a US payer’s healthcare budget. While a CE analysis is a very meaningful approach for decision-makers to judge the value of a new technology relative to the current standard of care, it does not provide insight into incremental gains achieved with the introduction of several technologies over time. Ten years ago, patients with MM lived 2–3 years from diagnosis. Today, thanks to the development and considerable use of novel medicines, one out of four patients live 10 years or longer^3,46, and this estimate would be expected to be higher with the recent spate of approved treatments for RMM. Evaluation of the value of a combination/sequence of novel treatments in a single disease area would be an interesting analysis rather than individually comparing each treatment with another, but such an analysis is beyond the scope of this body of work.

The analysis had various limitations associated with the underlying data and methods. Despite the large size of the ASPIRE trial and a mature PFS dataset, the survival functions had to be extrapolated to more than 25 years after the follow-up period in the trial, which increases uncertainty in the model results. Although the SEER population was matched to ASPIRE in terms of age, gender, and disease duration, additional prognostic factors such as the International Staging System (ISS) stage, cytogenetic risk, and prior treatments were not available to further improve the comparability of the patients in the two datasets. The application of the SEER data to extrapolate survival beyond trial duration is influential in the model and, as shown, results are sensitive to the survival modeling approach chosen. The SEER data do offer a real-world representation of survival outcomes, which are often meaningful for payer audiences making decisions at a population level. As noted, the survival estimates for the Rd arm produced from this extrapolation with data up to 2012 corresponded with previously published estimates of LYs gained with Rd.

Utility data were not collected in the ASPIRE trial; therefore, a mixed methodology was used to derive utilities from multiple sources. Previously published reference utility estimates from a Dutch study²⁶, which source data from patients who underwent intensive chemotherapy were combined with ASPIRE trial results via a utility-mapping exercise, however were not adjusted for preference weights reflecting the US population. Utility decrements associated with AEs were adjusted on the basis of a UK assessment³⁰. Utility for all subsequent therapies were also assumed to be the same for the rest of the patients’ life. Therefore, the utility estimates are subject to uncertainty, and future studies are needed to determine utility estimates that correspond closely with how patients value these health states.

A further data gap, contributing to some—although limited—uncertainty, was the lack of subsequent therapy patterns, including the length, the therapy choices and the treatment-free periods between further lines, specifically for the US, in the published literature. There is, therefore, uncertainty about the composition of subsequent therapies, since trial data are not necessarily representative of real-world treatment patterns in the US. Finally, assumptions and data from a published study sourcing these from key opinion leader interviews were used to provide resource use data for the health states. To account for these data limitations, several scenarios have been tested and detailed univariate analyses were run.

While the partition survival modeling approach undertaken in the K-GEM is the most common approach to model CE of oncology therapies, there is always some structural uncertainty inherent to the modeling process that would be revealed only if an alternative model was also built or became apparent with observational data based on patient experience with KRd. While long-term observational data will yield true survival estimates, the K-GEM utilizes data from a well-designed, robust trial that conducted a direct comparison between KRd and Rd, and extrapolates survival gains with the use of real-world data from SEER.

Conclusions

The K-GEM was a model that simulated treatment of the RMM population who had received 1–3 prior lines of therapy in the ASPIRE trial. The model showed that KRd is cost effective, with an ICER of $107,520 per QALY gained against Rd; the ICER is below the WTP threshold of $150,000, a generally accepted threshold in the US. This model is based on actual direct costs of treatment and largely driven by the superior efficacy of KRd over Rd demonstrated in the ASPIRE trial, which was observed to be consistent across sub-groups. For the population studied in the ASPIRE trial, i.e. relapsed MM patients who have received one-to-three prior treatments, KRd represents a highly effective and cost-effective treatment option. For payers, reimbursement of KRd for this patient population may represent an efficient allocation of healthcare dollars for the management of relapsed MM.

Transparency

Declaration of funding

This work was supported by Amgen, Inc., Thousand Oaks, CA.

Declaration of financial/other relationships

AJ has served as an advisory board member and consultant for Amgen and Celgene, has received sponsorships/grants from Amgen and Celgene, and has received honoraria from Celgene. SA, BB, MC, and SP are employed by and own stock in Amgen Inc. AB, IH, and ET are employees of Evidera, Inc., and this analysis was developed in collaboration with Evidera and sponsored by Amgen. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

The authors would like to thank Julie Gegner, PhD, and Shawn Lee, PhD, employees of Amgen, for writing and editorial support.