RE: Treatment Options for Relapsed or Refractory Multiple Myeloma: Effectiveness and Value Draft Report dated April 7, 2016

Transcription

1 Institute for Clinical and Economic Review Two Liberty Square, Ninth Floor Boston, Massachusetts RE: Treatment Options for Relapsed or Refractory Multiple Myeloma: Effectiveness and Value Draft Report dated April 7, 2016 Dear Sir or Madam: Thank you for the opportunity to provide comments on the Treatment Options for Relapsed or Refractory Multiple Myeloma: Effectiveness and Value draft report as part of the public comment period. On behalf of Amgen, Inc. (Amgen), please find enclosed comments on drug therapies for multiple myeloma, as requested. We appreciate the opportunity to comment, and would be pleased to address any questions you may have. Please direct any replies you may have to the enclosed document to either Joshua Ofman at , or me. I would also greatly appreciate a confirmation of successful receipt of this correspondence as soon as it is convenient for you. Amgen is providing you with referenced information. If you would like a reprint of a reference, please contact Amgen Medical Information. Please note that if you are a covered recipient as defined by the Affordable Care Act (ACA) and Amgen provides you with the requested reprint(s), Amgen s cost to obtain such reprint(s) may need to be disclosed and reported in accordance with the requirements under the ACA, state law and related disclosure obligations by Amgen. If you are a non-covered recipient requesting information on behalf of or the benefit of a covered recipient (physician or teaching hospital), the same requirements may apply. This information has been provided to you in response to your unsolicited request. If we may provide further information or assistance, or if you did not request this information, please contact Amgen Medical Information at AMGEN (26436), MedInfo@Amgen.com or visit our website at Sincerely, Suzana Giffin, PharmD Executive Director, Global Scientific Communications MIR

2 April 15, 2016; Submitted Electronically to from Martin Zagari, MD, Vice President, Global Health Economics, Amgen on behalf of Amgen Re: ICER Treatment Options for Relapsed or Refractory Multiple Myeloma: Effectiveness and Value. Dear Dr. Pearson: At Amgen, a science-based company committed to developing and delivering innovative medicines, our mission is to serve patients. We appreciate the opportunity to comment on the CEPAC/ICER draft report Treatment Options for Relapsed or Refractory Multiple Myeloma. Recent Food and Drug Administration (FDA) approvals of several newer therapeutic options for relapsed and/or refractory multiple myeloma (RRMM) represent a substantial advancement for patients and symbolize the spirit of American innovation by making more therapies available to more patients, as called for by The National Cancer Moonshot Initiative. 1 The availability of multiple treatment options in multiple myeloma (MM), a rare and serious cancer of blood cells, enables physicians to provide patients with individualized, precision care. Physicians anticipate that the newest medicines, used in combination and sequentially, may help extend overall survival (OS) of patients with MM from about 6 years ( ) to a decade or more. 2,3 Progress in the treatment of cancer and other serious disease contributes immensely to the improvement in patient wellbeing, which demands that all economic reviews on the value of medicines aim to achieve the highest level of transparency and clinical validity, strive for broad stakeholder engagement, and place scientific rigor and patient interests at the center of the analysis. At the highest level, we do not feel that ICER s evaluation in RRMM embodies these principles, which have been generally agreed on by ISPOR, PhRMA, NPC, and many of the world s leading technology assessment organizations. 4-7 Although the remainder of this response details select technical findings, we feel the overall process, including failing to adhere to such principles, contributed greatly to the issuance of a seriously flawed report that requires substantial revision. After careful review of the ICER RRMM draft report and consistent with our previous technical comments to ICER, we found the methodology inappropriate and disagree with many of ICER s resultant assumptions and conclusions. Specifically, ICER has: 1. Performed a cost-effectiveness evaluation using indirect comparisons when head-to-head data are available. An evaluation using head-to-head data would have achieved more clinically and economically valid results. a. ICER recognized serious limitations (at least 57 mentions in the draft) due to the lack of sufficient evidence to populate key model parameters. b. Lack of sufficient and appropriate data for the analysis should compel ICER to abandon infeasible indirect methods, and avoid the issuance of spurious, misleading and invalid results. 2. Used old and now irrelevant comparative data for the previous standard of care which undermines the modeled effect of new RRMM treatments. 3. Ignored direct patient-derived experience by not appropriately valuing health-related quality of life (HRQL) outcomes, thereby underestimating total quality-adjusted life years (QALYs). 4. Provided final results that not only underestimate the value of MM treatments, but also weighed these low value estimates against inappropriately low value thresholds. 5. Overestimated, in the budget impact analysis, use of newer medicines in late-stage treatment. 6. Failed to correct multiple technical errors, compromising the validity of the results and conclusions. ICER s arbitrary 3-page limit on comments is insufficient to allow for an adequate response to their complex analysis. We have therefore compiled more detailed data in appendices and provided these to ICER as a supplement to this 3-page public response.

3 Detailed Discussion of Issues Inappropriate use of indirect comparisons over head-to-head data: A key limitation and source of bias in any indirect comparison is missing data to account for patient and trial heterogeneity. Consistent with Amgen s comments on the Proposed Scope, ICER notes a lack of sufficient evidence in 12 key areas (57 mentions) to populate key model parameters. ICER further assumed a homogeneous population (Table 7, ICER report), even with evidence of considerable diversity in patient characteristics (Amgen Appendix 1). ICER nevertheless chose to use indirect comparison as the primary analytic method, essentially underweighting the robust head-to-head results for each medicine. An example of a seemingly spurious result from the ICER model is the inexplicably lower hazard ratio (HR) of 0.54 for progression-free survival (PFS) between PAN+BOR+DEX and LEN+DEX. This conflicts with the HR of 0.96 for PFS found in the manufacturer s own response to NICE. 8 Use of an indirect comparison under these circumstances effectively invalidates the results. Recommendation: Use available head-to-head data (between newer agents and LEN+DEN) as primary analysis rather than indirect comparisons. 2. Irrelevant data undermine the treatment effect of newer treatments for RRMM: ICER used data for LEN+DEX that included a highly toxic, high dose DEX from the MM-009/010 trials conducted in 2007, casting doubt on the baseline median time to progression [TTP] of 11.1 and 11.3 months, respectively. 9,10 These data are not reflective of the LEN+DEX efficacy (median PFS, 14.7 to 17.6 months) observed in the more recent direct trials of the newer treatments (2015) Using lower baseline regimen efficacy from older trials from which one extrapolates the relative treatment effect of the newer treatments will underestimate the relative efficacy of every newer intervention (Amgen, Appendix 2). Our replication of ICER s analysis using the (MM-009/010) data above revealed additional survival underestimation by ICER, likely due to an error in translating the published PFS-OS hazard ratio into their final OS curve (Amgen, Appendix 3). Taken together, these two assumptions inappropriately reduce the incremental net benefit and value of all of the newer medicines. Recommendation: (1) Use LEN+DEX data from direct trials of newer medicines. (2) If indirect comparison is performed, correct the methodology error(s) referenced above. 3. Ignoring direct patient-derived data underestimates total QALYs: The systematic underestimation of treatment effect is further worsened by the failure to appropriately weight survival gains with available HRQL data. ICER assumes consistent health state utility values across treatments evaluated in the model during all progression-free (PF) states, and uses Amgen/ASPIRE as the data source for the PF state (Table 12, ICER report). However, these same patient-derived data show higher utilities for CFZ+LEN+DEX compared with LEN+DEX on the basis of improved HRQL for CFZ+LEN+DEX vs LEN+DEX over 18 cycles of treatment (P<0.001 [Appendix 4]). 11 Furthermore, the European Medicines Agency (EMA) elaborates on the relevance of this benefit. 14 Therefore, to assume similar utilities for CFZ+LEN+DEX and LEN+DEX ignores the importance of incorporating patient HRQL into the valuation of the newer treatments. 15,16 Recommendation: Incorporate the impact of patient HRQL into pre-progression utilities by treatment, even if not uniformly available across comparators. 4. Compounding the systematic underestimation of QALYs by exclusively using low QALY thresholds: For reasons noted above, the QALYs gained for CFZ+LEN+DEX in ICER s analysis are considerably lower than those estimated by Amgen in the Kyprolis Global Economic Model (K-GEM). The former are likely systematically lower than justified for some of the other newer treatments as well. The K-GEM was developed in collaboration with input from clinical and health-economic experts worldwide, and uses head-to-head data from the ASPIRE trial, whereas ICER s approach uses indirect comparisons and LEN+DEX data from older trials. These differences, including appropriate weighting of survival with HRQL data from ASPIRE, lead to strikingly different results. The K-GEM estimate for the incremental cost per QALY for CFZ+LEN+DEX

4 vs LEN+DEX is $107,520 for patients with RRMM who have received 1 to 3 prior treatments, which is more favorable (better value) than the estimates generated by ICER for second and third line use of these interventions ($267,464 and $312,840, respectively). Amgen therefore strongly believes that CFZ+LEN+DEX is a cost-effective regimen even under one of the value thresholds applied by ICER. However, Amgen contends that thresholds proposed by ICER still do not reflect patient and clinician perspectives for RRMM. In oncology in the U.S., a broad range of value thresholds have been proposed (around $150,000 to $300,000 per QALY). Specifically, patients with metastatic cancer and other serious illnesses were found to have a threshold closer to $300, Multiple preference studies have suggested higher values for end of life interventions, such as late-stage cancer treatments Recommendation: Use a range of cost-effectiveness thresholds from $150,000-$300,000 for assessing value, in addition to using head-to-head data for each newer treatment. 5. Overestimating late-stage treatment use: For the analysis of budget impact, ICER assumptions regarding the percentage of patients with MM treated for their disease do not align with the cited source (see Amgen Appendix 5). 22 ICER also does not provide a rationale or justification for the high uptake pattern of newer medicines cited for second and third lines (Amgen Appendix 6). Recommendation: Use realistic and justifiable estimates of the patient pool eligible for treatment and for medicine use rates. 6. Failure to correct multiple errors: A detailed description of factual errors identified in the ICER report is provided in Appendix 7. Important flaws potentially affecting value calculations include those related to the overall survival (OS) HRs for ELO-LEN-DEX (0.77 instead of 0.71, Table 3, ICER report); efficacy estimates from the TOURMALINE-MM1 and PANORAMA-1 studies (Table 3, ICER report); inaccurate reference of PFS data from MM-009/MM-010 (actual data reported was TTP); and inaccurate calculation of dose intensities for LEN as part of the ELO-LEN-DEX regimen ( 51% instead of 80% [Appendix 8], Table 8, ICER report), which dramatically influences the cost-effectiveness results. Recommendation: Rectify errors, and prioritize head-to-head comparative models over indirect analysis when rerunning models. Analytical recommendations summary and implications for voting questions: ICER has recognized the serious flaws with its own methodological approach, and despite this has chosen to move forward with an inappropriate analysis resulting in spurious conclusions that are highly discrepant with more valid evidence. Using separate models, ICER should re-conduct analyses of the newer treatments using the head-to-head data available for each newer agent, assign an incremental cost/qaly value for each agent, and compare that value with a reasonable range of value thresholds for cancer in the U.S. (eg, $150,000 through $300,000 per QALY). The resulting valuations should then serve as the primary value benchmark for each newer treatment within its respective labeled indication. Because of the trial heterogeneity noted above, ICER should largely abandon the use of indirect comparisons and newer treatments should not be stack ranked. Amgen believes this revised approach to assessing product cost effectiveness and value will provide more constructive results to the community by highlighting the realistic value of treatments that add years to peoples lives. Voting questions should be modified accordingly to focus on the head-to-head model/results for each new medicine individually. Conclusions: Making the best evidence-based treatment decisions starts by using high-quality evidence in the right ways. Given the methodological flaws and factual errors present in the ICER draft report, Amgen strongly recommends that ICER re-evaluates the effectiveness and value of newer treatment options for MM using a series of simple and transparent models that take into account direct comparisons wherever possible. ICER should also apply appropriate value to improved survival by comparing model results with a range of potential incremental cost/qaly thresholds of up to about $300,000.

5 Response References 1. The White House. FACT SHEET: Investing in the National Cancer Moonshot. Available at: Accessed 12 April Kumar SK, et al. Continued improvement in survival in multiple myeloma: changes in early mortality and outcomes in older patients. Leukemia. 2014;28(5): Winslow, R. New Weapons in the fight against multiple myeloma. The Wall Street Journal. 4 January Available at: Accessed 12 April BIO (Biotechnology Innovation Organization). Bio principles on the value of biopharmaceutical. Available at: Accessed 15 April NCP (National Pharmaceutical Council). Guiding practices for patinet-centered value assessment. Available at: Accessed 15 April PhRMA (Pharmaceutical Research and Manufacturers of America). Principles for value Assessment frameworks. Avaliable at: Accessed 15 April Caro JJ, Briggs AH, Siebert U, Kuntz KM, Force I-SMGRPT. Modeling good research practices--overview: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force--1. Value Health. 2012;15(6): Novartis. Panobinostat for treating multiple myeloma in people who have received at least one prior therapy. ID663 Single technology appraisal (STA) manufacturer s response to the appraisal consultation document Available at: Accessed 13 April Dimopoulos M, et al. Lenalidomide plus dexamethasone for relapsed or refractory multiple myeloma. N Engl J Med. 2007;357(21): Weber D, et al. Lenalidomide plus dexamethasone for relapsed multiple myeloma in North America. N Engl J Med. 2007;357(21): Stewart AK, et al. Carflizomib, lenalidomide, and dexamethasone for relapsed multiple myeloma. N Engl J Med. 2015;372: Lonial S, et al. Elotuzumab therapy for relapsed or refractory multiple myeloma. N Engl J Med. 2015;373(7): Takeda. Data from phase 3 TOURMALINE-MM1 study. 6 December Available at: Accessed 12 April European Public Assessment Report. Kyprolis (carfilzomib): EPAR - Product Information. Available at: Accessed 13 April /12/ Osoba D. Health-related quality of life and cancer clinical trials. Therapeutic Advances in Medical Oncology. 2011;3(2): Kind P, Lafata JE, Matuszewski K, Raisch D. The use of QALYs in clinical and patient decision-making: issues and prospects. Value in Health. 2009;12:S27-S Seabury SA, Goldman DP, Maclean JR, Penrod JR, Lakdawalla DN. Patients value metastatic cancer therapy more highly than is typically shown through traditional estimates. Health Affairs. 2012;31(4): Hillner BE, Smith TJ. Efficacy does not necessarily translate to cost effectiveness: a case study in the challenges associated with 21st-century cancer drug pricing. J Clin Oncol. 2009;27(13): Braithwaite RS, Meltzer DO, King JT, Jr., Leslie D, Roberts MS. What does the value of modern medicine say about the $50,000 per qualityadjusted life-year decision rule? Med Care. 2008;46(4):

6 20. Aldy JE, Viscusi WK. Adjusting the value of a statistical life for age and cohort effects. The Review of Economics and Statistics. 2008;90(3): Hirth RA, Chernew ME, Miller E, Fendrick AM, Weissert WG. Willingness to pay for a quality-adjusted life year: in search of a standard. Med Decis Making. 2000;20(3): Song X, Cong Z, Wilson K. Real-world treatment patterns, comorbidities, and disease-related complications in patients with multiple myeloma in the United States. Curr Med Res Opin. 2016;32(1):

7 Amgen Appendix 1 Amgen Table 1. Baseline patient characteristics MM-009/ MM-010 ASPIRE ELOQUENT-2 TOURMALINE-MM1 LEN+DEX CFZ+LEN+DEX LEN+DEX ELO+LEN+DEX LEN+DEX IX+LEN+DEX LEN+DEX ITT population, N Demographics Median age, years (min, max) 63 (33,86) 64 (38,87) 65 (31,91) 67 (37,88) 66 (38,91) 66 (38,91) 66 (30,89) Female sex, % White/Asian, % 89/NR 95/<1 95/<1 82/10 86/10 86/8 83/9 Disease characteristics del17p, % N/A ECOG PS 0/1-2, % 87/11 90/10 91/9 93/8 90/11 93/5 92/7 Median time since 1 st diagnosis (range) MM stage at baseline Durie-Salmon, % 3.2 years (0.4,15.7) 3.0 years (0.4,19.7) 3.2 years (0.5,27.3) I II III ISS, % 41.6 months (3.6, 208.1) 41.9 months (0.9,194.3) I II III Beta2-microglobulin, % <2.5 mg/l

8 MM-009/ MM-010 ASPIRE ELOQUENT-2 TOURMALINE-MM1 LEN+DEX CFZ+LEN+DEX LEN+DEX ELO+LEN+DEX LEN+DEX IX+LEN+DEX LEN+DEX <3.5 mg/l mg/l mg/l Creatinine clearance (ml/minute) Creatinine clearance <30/30-60/60-90/ 90 ml/minute, % Creatinine clearance, 60 ml/minute, % /19/46/35 0/22/41/37 2/28/38/32 1/22/40/ Heavy chain, IGG/IGA 69/21 71/22 68/22 72/19 57/21 58/15 Prior therapies No. lines of treatment, % Bortezomib, % Carfilzomib, % Bortezomib and carfilzomib, % Thalidomide, % Lenalidomide, % Lenalidomide and thalidomide Doxorubicin, % 56 Melphalan, % NR NR 69 61

9 MM-009/ MM-010 ASPIRE ELOQUENT-2 TOURMALINE-MM1 LEN+DEX CFZ+LEN+DEX LEN+DEX ELO+LEN+DEX LEN+DEX IX+LEN+DEX LEN+DEX PI-naïve, % Stem-cell transplantation, % Refractory/Relapsed/Relapsed and refractory status, % NR NR 35/75/NR 35/65/NR 23/88/11 23/89/12 Refractory to Bortezomib Refractory to Thalidomide Refractory to PI Refractory to IMiD Note: Percentages may not add up to 100 due to rounding. CFZ+LEN+DEX, carfilzomib, lenalidomide, and dexamethasone; ECOG PS, Eastern Cooperative Oncology Group performance status; ELO+LEN+DEX, elotuzumab, lenalidomide, and dexamethasone; IgA, immunoglobulin A; IgG, immunoglobulin G; IMiD, immunomodulatory drug; ISS, International Staging System; IX+LEN+DEX, ixazomib, lenalidomide, and dexamethasone; MM, multiple myeloma; NR, not reported; PI, proteasome inhibitor

10 Amgen Appendix 2 1. Pooled data ASPIRE, ELOQUENT-2, and TOURMALINE-MM1 should be used as the source of the PFS LEN+DEX curve in the ICER model. Clinical parameters were estimated in the ICER model using an evidence base comprised of five trials with a LEN+DEX arm, however, as noted above, the LEN+DEX regimen, in particular the DEX dose in MM-009/010 was higher than that used in the newer trials. (Amgen Figure 1). Indirect comparisons could be made between the carfilzomib plus LEN+DEX (CFZ+LEN+DEX), elotuzumab plus LEN+DEX (ELO+LEN+DEX), and ixazomib plus LEN+DEX (IX+LEN+DEX) treatment arms. The inclusion of MM-009 and MM-010 is an attractive idea since results of long-term follow-up have been published, therefore characterization of time-to-event outcomes would be less dependent upon extrapolation through statistical methods. However, for the reasons discussed below, the inclusion of LEN+DEX data from MM009/010 is inappropriate. Amgen Figure 1. Network diagram Randomization of patients to trial arms ensures balance in patient characteristics between trial arms; however, it cannot be assumed that this balance is held across trials. Therefore, when meta-analyzing data, it is important to ensure that trial and patient characteristics are similarly distributed across trial arms, thereby reducing the likelihood of biasing the results with effect modifiers. TRIAL CHARACTERISTICS In the absence of additional evidence, it is reasonable to assume that the significant heterogeneity in dexamethasone dosing within the different lenalidomide plus dexamethasone regimens used within the network could cause variations in estimates of both efficacy and safety. Therefore the assumption of clinical similarity for lenalidomide plus dexamethasone as the common comparator for the MM- 009/MM-010 and ASPIRE trial as well as trials for other newer agents was not valid as shown below. Trial High- vs Low- DEX DEX Dosing ASPIRE Low-dose 40 mg on days 1, 8, 15, and 22 ELOQUENT-2 a Low-dose 40 mg on days 1, 8, 15, and 22 TOURMALINE-MM1 Low-dose 40 mg on days 1, 8, 15, and 22 MM-009 High-dose then low-dose 40 mg on days 1-4, 9-12, and of each MM-010 High-dose then low-dose 28-day cycle. After the 4 th cycle, DEX was administered only on days 1-4 a On days elotuzumab is administered the dosing of dexamethasone is altered from 40mg PO to 8mg IV and 28 mg PO. Treatment duration was until progression/unacceptable toxicity in each trial

11 Amgen Table 1 in Appendix 1 above compares patient characteristics across LEN+DEX arms of each trial. The patients in the LEN+DEX arms were similar in terms of age, gender, and ECOG performance status. However, there was diversity in the patient composition with regards to stage of disease at baseline. Prior lines of treatment before entering the trial also differed greatly between the MM-009/MM-010 trials and the more recent ASPIRE, ELOQUENT-2, and TOURMALINE-MM1 trials. Of the patients in the LEN+DEX arm of MM-009/010, only 8% of patients were treated with bortezomib prior to baseline. On the other hand, the more recent studies had 66% of patients in the LEN+DEX arms having received bortezomib prior to enrollment, and nearly half of the patients in ELOQUENT-2 and TOURMALINE- MM1 had received thalidomide (vs. 41.9% in MM-009/010). These differences are likely due to the fact that bortezomib had only just been approved for use in treating MM when MM-009/MM-010 were enrolling patients, with thalidomide being subsequently approved. With respect to the number of prior lines of treatment, >80% of patients in MM-009/MM-010 had received 2 lines of treatment prior to enrollment, versus <60% in the newer trials. These differences in prior treatment, and the more experimental nature of treatment before the approval of bortezomib, may have resulted in undefined, residual differences between the patients health status at baseline. The proportion of patients who had been previously treated with lenalidomide was strikingly different across trials, which is an extremely important consideration given each of the newer treatments includes lenalidomide as a component. Finally, since the percentage of patients who were being treated for relapsed or refractory disease was not defined in MM-009 or MM-010, it would be difficult to understand the setting in which patients are being treated. Therefore, clearly knowing the extent of the benefit experienced by LEN+DEX-treated patients, at points in time separated by an entire decade, may further influence treatment response in an unknown direction. The cumulative impact of these differences in trial and patient characteristics likely contributes to the difference in median PFS observed between MM-009/MM-010 (11.1 months) and the newer trials that evaluated innovative treatments ( 14.7 months). Due to the design of the ICER evaluation, these differences likely contributed to a systematic underrepresentation of the true net health benefit of innovative treatments in terms of median PFS and OS (which was directly related to PFS in the model). Therefore, we recommend MM-009/MM-010 be removed from the evidence base. This difference is expected to reduce the risk of bias attributable to differences among patients without substantially reducing the number of patients in the LEN+DEX arm as there are already nearly 1000 patients included across ASPIRE, ELOQUENT-2, and TOURMALINE-MM1. This difference is also likely to result in less uncertainty around the efficacy point estimates. To test this hypothesis we re-estimated the ICER results (Table E2, ICER report) after estimating PFS for LEN+DEX using patients in the ASPIRE trial. These results are shown in Amgen Table 2 and Appendix 3 below. As with the MM-009 and MM-010 trials, TTP/death data was reconstructed based on published Kaplan-Meier (KM) estimates of PFS. Our analysis suggests that life-years are underestimated because PFS for the universal comparator is based on older trials. Amgen Table 2. Estimated life-years from reconstructed data using ASPIRE trial for universal comparator LEN-DEX CFZ-LEN- DEX ELO-LEN- DEX Total LYs PFS LYs Progression LYs IX-LEN-DEX

12 CFZ+LEN+DEX, carfilzomib, lenalidomide, and dexamethasone; ELO+LEN+DEX, elotuzumab, lenalidomide, and dexamethasone; IX+LEN+DEX, ixazomib, lenalidomide, and dexamethasone; LEN+DEX, lenalidomide and dexamethasone; LY, life-year; PFS, progression-free survival

13 Amgen Appendix 3 2. ICER should re-visit their life-year modeling as current estimates of life-years gained are underestimated due to what appears to be an error in applying the methodology. To examine the impact of key modeling assumptions, we attempted to replicate the ICER model for estimating life-years. First, we digitized survival curves from the trials analyzed by ICER and used the methodology developed by Guyot et al to reconstruct individual patient data. Second, similar to ICER, we estimated QALYs using the partition survival approach first introduced by Glasziou et al We began by estimating PFS for the universal comparator (LEN+DEX). To do so, we reconstructed patient data from KM estimates of PFS based on a pooled analysis of the MM-009 and MM-010 trials. 25 Median survival for the reconstructed data matched median survival reported in the published paper very closely (11.2 vs months). Following ICER, we used a Weibull model to extrapolate PFS over a lifetime horizon (30 years in our case). The fitted survival curve tracked closely to the reconstructed KM curve in our analysis, suggesting that the Weibull model we used is appropriate. PFS for each regimen was then estimated by multiplying the hazard ratios reported in Table 9, ICER report by the PFS cumulative hazards from the Weibull model. As was done in the ICER report, we used the estimated relationship between median PFS and median OS from Felix, et al. 26 to derive an OS to PFS hazard ratio. This hazard ratio was applied to the PFS curve to derive OS by assuming proportional hazards and using the following relationship: S OO (t) = S PPP (t) r, where S PPP (t) and S OO (t) are the PFS and OS survival functions respectively, and r is the hazard ratio. 26 To ensure that predicted OS was consistent with observed OS, we reconstructed individual time to death data from KM estimates of OS in the pooled MM-009 and MM-010 trials. Weibull survival curves were fit to the data and compared with the Felix derived OS curves, which matched quite closely. These steps allowed us to estimate discounted life-years and discounted QALYs for the universal comparator and each regimen by integrating the Weibull PFS curve and the predicted OS curve. Following ICER, outcomes were discounted at 3%. Our analysis suggests the following: a. ICER s calculations for life-years following disease progression are underestimated using their own methodology. Using the methodology described above, we estimated discounted life-years when PFS for the universal comparator was based on the MM-009 and MM-010 trials. Results are presented in Amgen Table 3 below and shown side-by-side with ICER s reported results. Amgen Table 3. Estimated life-years from reconstructed data using MM-009 and MM-010 trials for universal comparator LEN+DEX CFZ+LEN+DEX ELO+LEN+DEX IX+LEN+DEX Amgen ICER Amgen ICER Amgen ICER Amgen ICER

14 Total LYs PFS LYs Progression LYs CFZ+LEN+DEX, carfilzomib, lenalidomide, and dexamethasone; ELO+LEN+DEX, elotuzumab, lenalidomide, and dexamethasone; IX+LEN+DEX, ixazomib, lenalidomide, and dexamethasone; LEN+DEX, lenalidomide and dexamethasone; LY, life-year; PFS, progression-free survival Amgen Appendix 3 References 23. Guyot P, Ades A, Ouwens MJ, Welton NJ. Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. BMC medical research methodology. 2012;12(1): Glasziou P, Simes R, Gelber R. Quality adjusted survival analysis. Statistics in medicine. 1990;9(11): Dimopoulos M, et al. Long-term follow-up on overall survival from the MM-009 and MM-010 phase III trials of lenalidomide plus dexamethasone in patients with relapsed or refractory multiple myeloma. Leukemia. 2009;23(11): Félix J, et al. Time-dependent endpoints as predictors of overall survival in multiple myeloma. BMC cancer. 2013;13(1):122.

15 Amgen Appendix 4. Amgen Table 4. Utility data from the K-GEM CFZ+LEN+DEX utility value (mean) At model start At beginning of cycle 3 Pre-Progression At beginning of cycle 6 At beginning of cycle 12 At beginning of cycle 18 Preprogression off treatment On 1st subsequent line Post-Progression On 2nd subsequent line On 3rd subsequent line Best Supportive Care SE LEN+DEX utility value (mean) SE CFZ+LEN+DEX, carfilzomib, lenalidomide and dexamethasone; K-GEM, Kyprolis Global Economic Model; LEN+DEX, lenalidomide and dexamethasone; SE, standard error

16 Amgen believes that ICER s assumption to utilize similar utilities for pre- and post-progression across all treatments without consideration of the patient-reported HRQL from the individual trials is is incorrect. Furthermore, ICER inaccurately cites the data that Amgen has made available to ICER in Amgen s communication to ICER (from Martin Zagari to Sonya Khan, 2/22/2016 in the document titled AMGEN RESPONSE TO ICER DATA REQUEST TO ICER 22 FEB We believe this assumption contradicts the evidence from the trial and the conclusions from the EMA in the European Public Assessment Report (EPAR). Therefore, it cannot be used in the base case. EPAR (pg ): Using a restricted maximum likelihood-based mixed model for repeated measures (MMRM) analysis under the assumption of missing at random (MAR), subjects treated with CFZ+LEN+DEX reported improved global health status with higher QLQ-C30 Global Health Status/QoL scores compared with LEN+DEX over 18 cycles of treatment (p-value = ). The MID is 5 points according to the literature (Cocks 2011; Kvam 2010; Kvam 2011; Delforge 2012), and this MID was met at Cycle 12 (5.56) and approached at Cycle 18 (4.81) when comparing CFZ+LEN+DEX versus LEN+DEX (Results shown in Table 29, EPAR and in Figure 5, EPAR). EPAR (pg. 88): The use of the combination of CFZ+LEN+DEX provided a positive benefit in terms of ORR (CFZ+LEN+DEX 87.1%; LEN+DEX 66.7%; p < ), duration of response (CFZ+LEN+DEX 28.6 months; LEN+DEX 21.2 months), and QoL. The minimal important difference (MID) refers to the global health status of the QLQ-C30, which was reached at Cycle 12 and approached at Cycle 18. Conclusions based on the MID for QLQ-C30 Global Health Status/QoL cannot be directly extrapolated to the EQ-5D (obtained mapping the QLQ-C30 questionnaire) and to the utilities calculated from the EQ-5D. Moreover, estimated difference in mapped utilities between the two arms at Cycle 12 is consistent with the difference estimated at all other time points. Consequently, in order to take into account that there is a clear and statistically significant benefit in HRQL for CFZ+LEN+DEX considering all the time points when HRQL was assessed, Amgen strongly believes that the correct approach consists of using utilities separately estimated in the two arms of the ASPIRE trial.

17 Amgen Appendix 5 To avoid an overestimation of the budget impact with newer treatments, ICER should make the most realistic assumptions about current treatment use. In a follow-up that we conducted with Song et al., who ICER cited for estimating the proportion of untreated patients, we content that ICER should assume no more than 25% of patients based on their analysis be considered untreated. On 15 April 2016, in response to Amgen's request, Xue Song clarified the actual untreated population for the relevant period. Song clarified that 68% of patients in were treated, and given the inclusion of monoclonal gammopathy of undetermined significance (MGUS) patients in SEER (see below), the prevalence estimates account for the total untreated population based on Song et al s results, which should therefore be no more than 25%. The entirety of Song s 15 April 2016 response is as follows: ICER cites from Song, Ze, and Wilson paper (CMRO 2016) that ~50% of patients were not treated for MM. The statement in the paper is Slightly more than half (N ¼ 13,188, 53.8%) of the MM patients initiated treatment during the follow-up period. 53.8% is based on patients diagnosed from The distribution of index years (ie, the year MM was first diagnosed) in the study population (not reported in the manuscript) is: Index Year (N, %) N % , % , % , % , % , % , % , % , % % A total of 3142 patients had their first MM diagnosis in , so a rough estimate would be 68% (=2151/3142) MM patients who received treatment in Consideration of MGUS in SEER Prevalence Estimates Further, ICER assumed that untreated cases were asymptomatic, had MGUS and therefore, would not be represented in the prevalence estimate for MM. The SEER MM prevalence estimates include MGUS patients (8%). 27 Consequently, it would be reasonable for ICER to assume ~25% of patients instead of 50% being untreated which would result in ~27% (24,970) in second line (2L) and 10.5% (9,710) in third line (3L) eligible for treatment. We examined the following SEER dataset SEER 18 Regs Research Data + Hurricane Katrina Impacted Louisiana Cases, Nov 2013 Sub ( varying) dataset was used and the patient selection criteria was solely: Site recode ICD-0-3/WHO 2008 variable = 'Myeloma'. The Myeloma site group parameter includes three ICD-O-3 codes: 9731, 9732, The ICD-O code refers to plasma cell myeloma (alternate name multiple myeloma) and the description of this code states: There are three clinical variants of plasma cell myeloma, all of which are coded to 9732/3:

18 Asymptomatic (smoldering or inactive) PCM: Bone marrow involvement, but no related organ or tissue impairment. Similar to MGUS in its lack of symptoms, but more likely to develop to symptomatic PCM. About 8% of patients are initially asymptomatic. 29 Appendix 5 References 22. Song X, Cong Z, Wilson K. Real-world treatment patterns, comorbidities, and disease-related complications in patients with multiple myeloma in the United States. Curr Med Res Opin. 2016;32(1): SEER. Surveillance, Epidemiology, and End Results Program. Stat Fact Sheets: Myeloma. Accessed 28 March SEER. Surveillance, Epidemiology, and End Results Program. Available at: Accessed 15 April SEER. Surveillance, Epidemiology, and End Results Program. Available at: Accessed 15 April 2016.

19 Amgen Appendix 6 ICER also does not provide rationale or justification for the high uptake pattern of newer interventions in 2L and 3L (Section 6, ICER report). Based on Song et al., 2016, 22 the share of regimens likely to be displaced by newer treatments in 2L and 3L in were a combined 38.1% and 22.2%. Consequently, a 75% uptake rate in each line is unrealistic with 50% uptake rates exceeding the likely upper limits in 2L and 30% uptake rates exceeding the likely upper estimates in 3L. We also note based on Song et al., 2016 current use of carfilzomib in 2L and 3L, and ICER must consider appropriate offsets for the existing share of carfilzomib-based regimens to accurately estimate budget impact. PAN+BOR+DEX budget impact should be considered on the basis of the regimen it is most likely to replace, BOR+DEX, which was the direct comparator in PANORAMA-1. We are not aware of any guidelines/publications that support the use of PAN+BOR+DEX as an alternative to LEN+DEX. Appendix 6 Reference 22. Song X, Cong Z, Wilson K. Real-world treatment patterns, comorbidities, and disease-related complications in patients with multiple myeloma in the United States. Curr Med Res Opin. 2016;32(1):

20 Amgen Appendix 7. Factual Errors, Omissions, Inconsistencies Pag e Error/Omission/Disagreement Correction/Context/Clarification ICER Section 2: Topic in Context p4 However, following disease progression, the relative impact of the level of response is less certain, as complete response (CR) is not consistently predictive of overall or even progression-free survival in these patients p7 In Table 1, ICER report, the dosing of carfilzomib and elotuzumab is not clearly stated, and it is unclear if the correct dosing was used in the ICER model. The number of the number of doses of Kyprolis per cycle is missing; the number of doses of elotuzumab per cycle is also missing. The statement that complete responses are not consistently predictive of OS or PFS depends on one reference (Mohty et al., 2012). Three of the key studies identified by ICER as relevant for this assessment (ASPIRE, PANORAMA-1, and MM-03) have reported prolonged PFS and/or OS associated with deeper responses. For example, in ASPIRE, patients who achieved a CR or better had longer PFS and OS than those who did not. The lack of discussion surrounding CR and/or minimal residual disease in the draft report is a major omission. For CFZ+LEN+DEX, treatment with carfilzomib is for a maximum of 18 cycles with six doses of carfilzomib per cycle in Cycles 1-12, and four doses of carfilzomib per cycle thereafter. It appears to us that in ICERs report 6 doses were considered for carfilzomib in cycle 13 instead of 4 doses (Table E2, ICER Report). p8 Carfilzomib (Kyprolis, Onyx/Amgen) is a newergeneration PI that was first approved in 2012 for use with lenalidomide and dexamethasone in patients with 1-3 prior lines of treatment. FDA approved carfilzomib as monotherapy in 2012 and in combination with lenalidomide and dexamethasone in ICER Section 3: Summary of Coverage Policies and Guidelines

21 p10 The NICE guidance evaluating carfilzomib with lenalidomide and dexamethasone after prior therapy was suspended in January The manufacturer withdrew the submission. ICER IMWG Recommendations p13 The International Myeloma Working Group (IMWG) guidelines recommend the following for the management of relapsed myeloma carfilzomib and pomalidomide should be primarily used for patients refractory and/or intolerant to both bortezomib and lenalidomide [for first relapse]. Section 4: Comparative Clinical Effectiveness p19 p25 The discussion of high-risk disease is inaccurate and misleading. The appraisal of CFZ+LEN+DEX by NICE will be carried out as part of single technology appraisal (STA) ID934. As noted on the NICE website: The appraisals of carfilzomib in combination with lenalidomide and dexamethasone for relapsed multiple myeloma (ID677) and carfilzomib in combination with dexamethasone for relapsed multiple myeloma (ID934) were scheduled into the NICE work program as two separate appraisals because of the difference in timing of their respective marketing authorizations. NICE has agreed to the company s request that these two appraisals now be considered as one STA as the evidence base for the proposed marketing authorization for carfilzomib in combination with dexamethasone will be important in supporting the cost effectiveness of carfilzomib for the broader relapsed/refractory setting. IMWG also states Regimens that can be employed in first or subsequent stages of relapse are summarized in Table 4 [IMWG] with level of evidence and grade of recommendation corresponding to each regimen. Table 4, IMWG includes CFZ+LEN+DEX with level II evidence with a Grade A recommendation. For comparison purposes, the percentage of patients with high-risk disease (Page 19 and Table 3, ICER report) should be calculated from patients who were evaluable for high-risk disease and not from the entire study population. Furthermore, in Table C6, ICER report, the ELOQUENT-2 study did not use a 60% cut-off for del(17p). If any cell was positive, the patient was considered to be del(17p) positive. The ASPIRE trial is the only study under consideration to use the stringent 60% cut-off for del(17p) recommended by IMWG. Finally, since the studies used different cut-offs to define patients with high-risk disease, the results of the studies are not comparable.

22 p20 There are errors in Table 3, ICER report with respect to the TOURMALINE-MM1 study. p20 p21 There is an error in Table 3, ICER report with respect to the ELOQUENT-2 study. p20 There is an error in Table 3, ICER report with respect to the PANORAMA-1 study. p22 The percent of deaths reported in Figure 3, ICER report is misleading. p22 In the trial of ELO+LEN+DEX, survival was statisticallysignificantly improved among patients with 2 prior lines of treatment (HR 0.67, 95% CI ). p25 In the trial of CFZ+LEN+DEX, the hazard ratio relative to LEN+DEX was less favorable in the refractory subgroup (0.89 vs for the overall population). This relationship is consistent with the understanding that double refractory patients tend to have The median number of prior regimens should be one (not two), and the p-value for overall response rate (ORR) should be (not < 0.001). The HR for OS should be 0.77 (95% CI: 0.61, 0.97), not 0.71 (95% CI: 0.54, 0.93). This error is replicated in the network meta-analysis for OS (Table D1, Page 111). Also, it should be noted in the subsequent text (page 21) that the OS data are not statistically significant as described by the sponsor ( at the time of the interim analysis the OS endpoint had not reached the pre-determined threshold for statistical significance. ) 30 Immature OS data are provided in this table (HR: 0.87; 95% CI: 0.69, 1.10; p = 0.26) when final mature OS data are available (HR: 0.94; 95% CI: 0.78, 1.14; p = ). These final mature OS data are provided on the subsequent page, though the p- value is incorrect (should be p = instead of p = ). The immature OS data are also included in the network meta-analysis for OS (p 111; Table D1, ICER report). For PANORAMA-1, the percent of deaths listed are based on immature OS data. The percentage of deaths from the final mature OS data should be reported (: 55%; PAN+BOR+DEX: 53%). For ELOQUENT-2, the percentage of deaths should be from the 3-year OS data (HR: 0.77). If these numbers are not available, then it should be clearly indicated that data from an earlier data cut are being reported. It is inaccurate to state that a statistically significant improvement in patients with 2 prior treatments was observed in the ELOQUENT-2 trial as at the time of the interim analysis, the OS endpoint had not reached the pre-determined threshold for statistical significance in the intention-to-treat population. The discussion of PFS results for patients who are refractory to both bortezomib and an immunomodulatory drug (IMiD) in ASPIRE is misleading. Of 792 patients randomized, only 51 patients were double refractory to bortezomib and an IMiD (24 in the CFZ+LEN+DEX arm and 27 in the LEN+DEX arm). Conclusions based upon such a small subgroup are inconclusive at best.

23 more aggressive disease subtypes. p29 All three of these trials used openlabel designs, raising the concern that gains in quality of life might have been overstated Section 6.3: Incremental Costs per Outcome Achieved p39 Trial populations were sufficiently homogeneous to allow for comparisons via network metaanalysis p41 Base case PFS curves for LEN+DEX were derived from parametric fits to pooled Kaplan- Meier data from the MM-009 and MM-010 trials of LEN+DEX as described above. This is a perplexing statement as two of the trials did not report significant quality of life benefits. For the ASPIRE trial, the only trial to demonstrate significant improvements in quality of life, both treatment arms had similar baseline completion rates and mean baseline QLQ-C30 GHS/QoL scores, indicating limited evidence of bias. We disagree with the assertion that trial populations were sufficiently homogeneous. Please see discussion in appendix. The cited references for these studies do not report PFS data. In fact, according to PFS is not a primary or secondary endpoint of these studies. TTP is the primary endpoint. If TTP KM data were used for the PFS analysis, this should be stated. If PFS KM data were used, accurate references or a clear discussion of how these data were derived should be included in the report. It should also be noted that neither trial was designed to assess PFS. p41 HRs (Table 9, ICER report) with respect to the relationship between PVd vs In addition, as pooled data from the MM-009 and MM-010 trials are being used for comparison, information about these trials should be listed in Table 3, ICER report to understand how similar the patient populations are in the studies under consideration. For example, the percentage of patients receiving prior bortezomib and prior lenalidomide was 10.7% and 0.0%, respectively (MM-009) and 4.5% and 0.0%, respectively (MM- 010). In ASPIRE, 65.9% of patients had received prior bortezomib and 19.7% of patients had received prior lenalidomide, indicating that the patients from these studies had received vastly different prior treatment regimens. The inclusion of these data is critical to assess whether it is valid to use trials that reported results in 2007 as a comparator to modern clinical trials. In the manufacturer s response to the NICE Appraisal Consultation Document for PAN+BOR+DEX, a matching-adjusted indirect

24 p11 6 bortezomib/dexamethasone (Vd) and vs Rd The estimation of drug utilization was derived from several factors, including the relative dose intensity reported in trials or directly provided by manufacturers, and the dosing schedule (Appendix Table E3, ICER report), where the dose may be fixed by weight or by body surface area (BSA), assuming patient characteristics as shown in Table 8, ICER report. p43 Why are the post-progression regimens in the cost-effectiveness model (Table 11, ICER report) not reflective of those assumed for the budget impact model, which assumes that 75% of patients would be on either KRd, ERd, PVd, or ixazomib/lenalidomide/dexametha sone (IRd) at third line (3L)? p44 Utilities (Table 12, ICER report) Section 6.4: Potential Budget Impact p51 However, the authors acknowledge that almost 50% of treatment comparison analysis of PFS was presented for LEN+DEX vs. PAN+LEN+DEX in patients with 2-3 prior treatments (Section 2.2, Pages 16 and 61). The HR (ignoring the lack of proportional hazard) was estimated to be (LEN+DEX vs. PAN+LEN+DEX). This suggests a HR for PAN+LEN+DEX vs.len+dex of (1/1.043), which we recommend ICER use in the analysis. We think that the HRs currently used in the ICER model for PAN+LEN+DEX vs. LEN+DEX (0.64) leads to an overestimation of the QALYs for PAN+LEN+DEX. We note that the treatment intensity of lenalidomide in the ERd regimen is estimated to be 50%, per the citation 66. US Food and Drug Administration. Empliciti Medical/Statistical Review (761035Orig1s000). In: Research CfDEa, ed2015. The FDA report cited actually provides numbers and figures on elotuzumab dose intensity in the ELOQUENT trial. The table and figures suggest that the median dose intensity is 100% for elotuzumab in cycles 1 and 2, and 96% for cycles 3 and beyond, and it s 86% for lenalidomide. Means are not presented numerically, but based on the figures of the cited documents; they are very close to the medians above. This bears no resemblance to the numbers quoted in the ICER report. We disagree since Amgen had provided ICER with arm-specific utilities that varied by cycle See discussion in Amgen Appendix 5 and Appendix 6