Cost-effectiveness of Novel Treatment Sequences for Transplant-Ineligible Patients With Multiple Myeloma

Key Points Question What is the optimal sequence of treatments for patients with multiple myeloma from the perspective of the patient, physician, and society? Findings This economic evaluation found that sequences starting with daratumumab-bortezomib-melphalan-prednisone (second line: carfilzomib-lenalidomide-dexamethasone or elotuzumab-lenalidomide-dexamethasone) or bortezomib-melphalan-prednisone-thalidomide-maintenance bortezomib-thalidomide (VMPT-VT) (second line: daratumumab-lenalidomide-dexamethasone) had the largest expected overall survival (7.5 years); total costs per patient for these sequences ranged between $786 024 and $1 085 794. The sequence VMPT-VT-carfilzomib-lenalidomide-dexamethasone-panobinostat-bortezomib-dexamethasone had the most favorable cost-effectiveness ratio ($98 585 per life-year gained, and $132 707 per quality-adjusted life-year). Meaning These findings can support clinical decision-making and guideline development, reimbursement decisions, and price negotiations.


Introduction
Multiple myeloma (MM) is the second most common type of blood cancer, with 50 918 patients with new diagnoses annually in Europe 1 and 32 270 such patients annually in North America. 2 Like many cancers, it is incurable, and treatment aims at prolonging the time to disease progression to control disease symptoms and increase overall survival (OS). Eventually, the disease will progress (again), and patients, therefore, often receive several lines of treatment. [3][4][5] In the past 12 years, the number of treatments for patients with MM has increased substantially.
Many randomized clinical trials have been conducted adding next-generation proteasome inhibitors (carfilzomib 6 and ixazomib 7 ), monoclonal antibodies (daratumumab 8,9 and elotuzumab 10 ), new immunomodulatory agents (pomaliomide 11 ), and a histone deacetylase inhibitor (panobinostat 12 ) to melphalan-prednisone or a 2-drug backbone of either lenalidomide-dexamethasone (LenDex) or bortezomib-dexamethasone (BorDex), which were the standard therapies for relapsed MM. In addition to these 3-drug regimens, carfilzomib-dexamethasone (CarDex) 13 has been compared with BorDex in the relapsed refractory setting. The 3-drug regimens and CarDex resulted in improved progression-free survival (PFS) compared with the 2-drug regimens LenDex and BorDex. Moreover, for CarDex and carfilzomib-lenalidomide-dexamethasone (CarLenDex) an improvement in OS was observed. 14,15 Although the availability of effective regimens is embraced by both patients and hematologists, it also imposes challenges. First, as the number of treatment options increases, it is vital to know which sequence is most effective and to compare the OS and quality-adjusted life-years (QALYs) of different sequences. Unfortunately, trials investigating treatment sequences are lacking. 16 Second, to preserve global access to affordable and effective therapies, it becomes increasingly important to investigate their costs and cost-effectiveness. The prices of novel agents are often high, and, in most regimens, either an expensive drug is added to standard therapy or a more expensive drug replaces standard therapy. Furthermore, many of the novel agents are not limited to a prespecified number of cycles but are administered until disease progression, further increasing the costs. 17 Although costeffectiveness studies 3,18-21 for elderly patients with non-transplant-eligible (NTE) MM have been conducted, they did not investigate treatment sequences or include recently introduced agents. Therefore, we estimated the clinical effects, costs, and cost-effectiveness of treatment sequences, including all currently available novel agents, for NTE MM.

Methods
This study uses data from the PHAROS registry. 22,23 Data collection for that registry and the use of those data for the current study were approved by the ethical committee of the Erasmus University Medical Center Rotterdam in the Netherlands, which waived the need for informed consent because the data were deidentified. This study follows the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) reporting guideline.

Patient-Level Simulation Model
A model is necessary to combine different data sources and extrapolate data to calculate lifetime costs and effects. We adapted our previously developed patient-level simulation model for elderly patients with NTE MM, 3 which was based on data from a Dutch observational registry (ie, PHAROS registry). 22,23 The model is a discrete event simulation consisting of objects and events. 24 Objects are individual patients and are obtained from a real-world Dutch population of patients with MM aged 65 years or older and, therefore, NTE (median age at first-line treatment, 75 years); events were initiation of a new treatment line or death. Regression models per line of treatment (including coefficients for treatment, and, for the first-line treatment, patient and disease characteristics also) were used to estimate the time to event (TTE), which was defined as the time since the start of treatment to an event) (eAppendix in the Supplement). For the first-and second-line treatments, events were either the start of the next line of treatment or death. The type of event (ie, next line of treatment or death) was based on a logistic regression per line of treatment. From the start of thirdline treatment, only time to death was modeled. TTE (as a proxy for time to progression) was selected as the outcome measure because the initiation of a new line of treatment or death is associated with changing costs and effects. The model simulates individual patients, each with his or her own patient and disease characteristics. For each patient, costs and effects were estimated for a maximum of 3 treatment lines according to regression models (eAppendix in the Supplement). We modeled TTE, OS (defined as time from the start of first-line treatment to death), QALYs, costs, and cost-effectiveness. Utility values (on a scale of 0 to 1, where 1 denotes perfect health, and 0 denotes death) for patients treated in clinical practice per treatment and per line of treatment are not available in the literature for patients with NTE MM. To obtain QALYs, we used, as in our previous model, 3 a mean (SD) utility value of 0.76 (0.21). 25

Sequential Treatment Strategies
Novel treatment options for newly diagnosed NTE MM and relapsed or refractory MM were identified from 2 systematic literature reviews and network meta-analyses (NMAs) of randomized phase 3 trials. 26,27 Both NMAs were updated to include the most recent results from the MAIA, 28 ALCYONE, 29 and OPTIMISMM trials. 30 To estimate the effectiveness of novel treatment sequences, the outcomes from the NMAs were combined with the Weibull regression models for lines 1, 2, and 3.
Each model included a reference category for treatment (ie, line 1, melphalan-thalidomide; line 2, BorDex; and line 3, LenDex), and the relative effectiveness of the novel treatments (ie, hazard ratios [HRs] for PFS obtained from the NMAs) was used in the regression models for lines 1, 2, and 3 in the patient-level simulation model (see eFigure 1, eFigure 2, and eFigure 3 in the Supplement). The 95% CIs were used to incorporate uncertainty of treatments' effectiveness. We identified 30 treatment sequences on the basis of the outcomes of the NMAs and clinical relevance (eTable 1 in the Supplement). We assumed that patients would not receive 2 lenalidomide-based (or 2 bortezomibbased) regimens in a treatment sequence. Hence, patients receiving lenalidomide as the first-line

Other Model Parameters Treatment Costs
We retrieved dosing schemes and timing of administrations from Dutch clinical guidelines or, if not available, from randomized clinical trials (eTable 2 in the Supplement). We distinguished time periods and different costs per day for treatments with changing dosing schedules, frequencies, or regimen composition. Regimens that were given continuously (ie, until progression) were assumed to be given until the start of the next treatment or until death, whichever occurred first. For treatments with a maximum treatment duration, we used TTE or the maximum duration, whichever was shortest. Unit costs for drugs as of March 2019 were retrieved from a Dutch pharmaceutical price database (ie, including 6% value-added tax) (see eTable 3 and eTable 4 in the Supplement for details). eTable 5 in the Supplement shows detailed drug costs per day and month.

Other Costs
Other resource use included hospitalizations, outpatient visits, and laboratory tests. For agents requiring intravenous administration, we assumed a visit to the day ward per administration. Apart from intravenous drug administration, we retrieved average resource use in clinical practice per month by line of treatment from all elderly patients (aged Ն65 years) of the PHAROS registry (eTable 4 in the Supplement). Unit prices for hospitalization days, intravenous administrations, and outpatient visits were obtained from the Dutch costing manual 31 and a Dutch MM costing study 32 (eTable 5 in the Supplement).

Statistical Analysis
For the cost-effectiveness analysis, costs (presented in US dollars; conversion rate as of January 26, 2021, €1 = $1.2143) and effects were calculated for all treatment sequences from a hospital perspective with a lifetime horizon. The sequence bortezomib-melphalan-prednisone (VMP)-LenDex-pomalidomide-dexamethasone (PomDex) was selected as the base case for the costeffectiveness analysis to obtain the incremental (ie, additional) effects and costs and the incremental cost-effectiveness ratio (ICER) per life-year (LY) and per QALY gained. To account for the uncertainty of the estimates, all results were obtained using probability distributions for input parameters. Input parameters for resource use and unit costs followed gamma distributions, and the utility values followed a beta distribution. The uncertainty regarding the effectiveness of the treatments was incorporated by drawing HRs from the 95% CIs and 95% credible intervals from the NMAs. We obtained mean ICERs from 5000 simulations. Costs and effects in this study were calculated by applying (1) no discount rates and (2) a discount rate of 4% for future costs and 1.5% for future effects as recommended by the Dutch Costing Manual. 31 We performed regression analysis in Stata MP statistical software version 16

Results
Outcomes by Line  [24] months vs 53 [19] months for VMPT-VT). DaraLenDex had the highest total, annual, and drug costs. Although TTEs for VMP and LenDex were similar (32 vs 28 months), costs for LenDex were more than doubled compared with those for VMP (total costs, $281 241 vs $119 052; annual costs, $121 618 vs $45 230). The difference in costs is mainly associated with differences in treatment duration; LenDex is given until progression (ie, 28 months), whereas VMP is given with a maximum duration of 9 cycles of 35 days.
In the second and third lines, DaraLenDex had the longest TTE and the highest mean total costs per patient but not the highest annual costs ( Figure 1C and 1D and Figure 1E   However, total costs in line 2 for CarLenDex (3-drug regimen) were lower than total costs for CarDex (2-drug regimen) ($456 492 vs $481 745). Although CarDex is given until progression (ie, 23 months), carfilzomib is discontinued after 18 cycles of 28 days each in the 3-drug regimen.  DaraVMP was more effective as a first-line treatment than other first-line treatments such as

ICERs for Treatment Sequence
The   dominated by the base case, which means that effectiveness of these sequences was lower, whereas the costs were higher ($568 884 vs $285 619). Figure 3 shows the incremental OS in months and incremental costs for each of the 5000 simulations per treatment sequence compared with the base case. The individual circles form a cloud, and the spreading of this cloud shows the uncertainty in the outcomes. Figure 3 shows that the 2 dominated sequences, LenDex-CarDex-DaraBorDex and LenDex-CarDex-PomDex, were in

Discussion
To our knowledge, this study is the first to provide evidence of sequences in terms of clinical effects, costs, and cost-effectiveness in clinical practice for patients with NTE MM. We show the optimal sequences for clinical effects and cost-effectiveness, as well as relative differences. These insights provide valuable evidence additional to data from registration studies and can be used for clinical decision-making, guideline development, reimbursement decisions, and price negotiations. In addition to identifying the optimal sequence in terms of effectiveness, our modeling study also provides evidence on costs. For example, total costs of the most effective treatment sequences ranged from $786 024 to $1 085 794 per patient. Compared with our base case VMP-LenDex-PomDex, the additional costs for treating 1 patient with MM may require the health care budget to increase with up to $800 000 per patient. Furthermore, our results show that sequences with similar effects may greatly differ in health care costs. Although the expected outcomes of the sequences VMP-LenDex-PomDex and LenDex-CarDex-PomDex were similar, the costs of the latter sequence were twice as high ($568 884 vs $285 619). This difference was mainly due to higher costs of LenDex compared with VMP as first-line treatment (total mean costs per patient were $119 052 for VMP and $281 241 for LenDex). In addition, we showed that the most effective sequences did not accrue the highest costs. The sequence DaraLenDex-CarDex-PomBorDex had the highest total costs ($1 139 944) but lower mean OS (5.4 years) compared with the most effective sequences.
Using public list prices, we were able to provide evidence for reimbursement decisions by modeling the ICERs for treatment sequences including novel agents. These ranged from $98 585 to

Limitations
This study has limitations that should be addressed. Although modeling treatment sequences provides additional information, assumptions had to be made to make a comparison of treatments that are not head-to-head comparisons. To compare the regimens, we used relative effectiveness between treatment regimens because the relative difference preserves randomization within the trials. 36 36 Because the relative effectiveness of DaraVMP compared with VMP was more favorable (HR for PFS in the ALCYONE trial, 0.42) 29  We assumed that the HR for PFS would be representative for the HR for TTE to use the NMA outcomes. Although we cannot verify whether this assumption is valid for all treatments, it is supported by the results from the VISTA trial (HR for VMP vs MP for PFS, 0.56 [P < .001]; HR for time to next treatment, 0.52 [P < .001]) 37 and GIMEMA0305 trial (VMPT-VT vs VMP, HR, 0.58 [P < .001] and time to next treatment, HR, 0.52 [P < .001]) 38 . The number of patients decreases per line of treatment, which is based on the association between TTE and the type of event, as observed in Dutch clinical practice data (ie, PHAROS data). This is, however, comparable to proportions reported for other European countries. 5 We assumed that the association between TTE and type of event also exists for the novel therapies. Future clinical practice data should confirm this modeled proportion for novel treatments (60%-70% for second-line treatments and 35%-50% for third-line treatments).
In addition, the actual acquisition price may be lower in other countries because of negotiated discounts. Our results (Figure 1) show that drug costs are the major cost factor and illustrate the sensitivity of our results to the drug prices. If the negotiated price reduction has a similar ratio for the drugs, the order of the results will not substantially differ. An advantage of our model is that negotiated discounted prices and prices from other countries can be easily incorporated.