Reply: Cost-efficiency analyses for the US of biosimilar filgrastim-sndz, reference filgrastim, pegfilgrastim, and pegfilgrastim with on-body injector in the prophylaxis of chemotherapy-induced (febrile) neutropenia

Original Article: https://www.tandfonline.com/doi/full/10.1080/13696998.2017.1358173

The aim of our study¹ was to replicate and extend to the US market a 2012 European economic evaluation² that is one of the most cited publications on the savings that can be achieved by conversion from filgrastim (Neupogen, Amgen) and pegfilgrastim (Neulasta, Amgen, Thousand Oaks, CA) to the biosimilar filgrastim-sndz/EP2006 (Zarzio/Zarxio, Sandoz, Princeton, NJ) in the prophylaxis of chemotherapy-induced (febrile) neutropenia (CIN/FN). This was followed by an independent simulation study³ evaluating how these savings can be applied to expand patient access to medications by purchasing additional anti-neoplastic therapy on a budget-neutral basis. These studies are part of an established portfolio of cost-efficiency and expanded access investigations related to biosimilars^1–6. In this context, we use the term cost-efficiency to refer not only to reducing costs, but also to the efficiencies, such as expanded access, that can be gained from the savings.

Beyond the “canon of the C’s” of pharmacoeconomics

The economic evaluation of healthcare interventions has evolved significantly over the past three or so decades—to witness, for instance, how the classic book by Drummond et al. has changed in volume, focus, and applicability. The 1987 first edition⁷, under 200 pages, focused mainly on cost-effectiveness, cost-utility, and cost-benefit analysis. At 464 pages, the 2015 fourth edition⁸ adopted a thematic approach targeting technical principles and methods which cuts across many types and levels of economic evaluation. This approach reflects that economic research questions drive the selection of methods, not that methods pre-determine the type and scope of research questions that may be asked. Further, the fourth edition implies moving beyond the narrow scope of the “C classifications” that have long defined (and constrained) health economic analysis. The five C’s mentioned by Bensink et al.⁹, citing secondary sources^10–12, mirror a classical perspective on pharmacoeconomic analysis where methods drive research questions. More generally, questioning terms, positing a canon, and defending orthodoxy detract unnecessarily from the primary scientific responsibility of contributing transparent economic evaluations to inform policy and decision-making, and to assure (may we say so?) cost-responsible healthcare to patients.

Cost-efficiency is not merely a matter of minimizing cost, but of evaluating the cost of various scenarios of delivering healthcare interventions—as we did in our study. Contrary to the belief of some, the goal is not to determine the lowest cost, but to find the most efficient cost structure contributing to the value equation. Efficiency and value are intrinsically intertwined¹³, as Porter¹⁴ emphasizes with his “health outcomes achieved per dollar spent” definition. This is critical in cancer care for two reasons¹⁵. First, new insights into disease, highly specific treatment options, and greatly improved (survival) outcomes all come with a significant price tag. We need to know where and how we can be more (cost-)efficient in our choice of treatment options, so that we can achieve better outcomes for less money. Second, cost-responsible cancer care is critical as payment models move beyond merely administering cancer treatments. For instance, the Oncology Care Model¹⁶ integrates, within a context of performance and accountability, enhanced services such as care coordination and navigation, evidence-based treatment, and (expanded) access to care—thus addressing the ethical mandate to be both cost-efficient and patient-centric¹⁷.

A meta-analysis of non-inferiority trials is not a superiority trial

Non-inferiority trials are a common—and logical—approach to gaining regulatory approval of improved formulations of a drug. Pegfilgrastim was approved on the basis of two phase III non-inferiority trials^18,19 comparing it to standard daily filgrastim. As to the primary endpoint, both studies showed that the mean number of days of grade 4 neutropenia in pegfilgrastim-treated patients did not exceed that of filgrastim-treated patients by more than 1 day. Both studies also documented that pegfilgrastim and filgrastim were comparable on the secondary endpoints of days of severe neutropenia in cycles 2–4, the incidence of febrile neutropenia, absolute neutrophil count (ANC) nadir depth, and time to ANC recovery. With these two non-inferiority trials as the primary evidence base for pegfilgrastim, there is no phase III evidence of superiority of pegfilgrastim over filgrastim.

Meta-analysis does not solve this problem. Meta-analyses of non-inferiority trials do not establish superiority, including when phase II trials are mixed in^20–22. Further, meta-analyses are only as good as the studies included are comparable in objective, design, measurement, and analysis. This questions the three meta-analyses cited above.

Focusing first on Pinto et al.²⁰ and Cooper et al.²¹: these included the Green et al.¹⁹ and Holmes et al.¹⁸ phase III double-blind non-inferiority trials, as well as three phase II studies^23–25. Two of these were open-label^24,25, and two were dose-finding^23,25. All but one study¹⁹ used weight-based pegfilgrastim dosing, and all studies used weight-based instead of the prevailing fixed dosing options for filgrastim. Trials also differed in indication, chemotherapy regimen, number and length of chemotherapy cycles, and patient age.

If indeed the (implicit) intent of these meta-analyses was to show the superiority of pegfilgrastim over filgrastim, it seems reasonable that they would have focused on the duration of grade 4 neutropenia, which was the primary end-point in all five studies. This is not the case, and no explanation is given. After extracting the duration data from the five studies (Table 1), we performed a random-effects meta-analysis of the differences in means of this primary end-point. At p = .709, this difference is not statistically significant (Figure 1). This is consistent with the non-inferiority observed in the phase III trials and, certainly, does not warrant any claim of superiority of pegfilgrastim over filgrastim. Sub-analyses by type of trial (not shown) were in line with these results.

Figure 1. Pooled difference in means meta-analysis.

Table 1. Duration (days) of grade 4 neutropenia in cycle 1.

Study	Pegfilgrastim			Filgrastim
	n	Mean	SD	n	Mean	SD
Green et al.^19 (phase III)	77	1.8	1.4	75	1.6	1.1
Holmes et al.^18 (phase III)	149	1.7	1.5	147	1.8	1.4
Holmes et al.^23 (phase II)	46	1.3	1.1	25	1.6	1.3
Vose et al.^24 (phase II)	29	2.8	2.6	31	2.4	2.7
Grigg et al.^25 (phase II)	14	1.5	1.0	13	0.8	1.2

Cooper et al.²¹ reported only FN incidence in all cycles, and it is unclear why other secondary end-points were not included. To their credit, Pinto et al.²⁰ corrected a calculation error in the all-cycle FN incidence analysis—attenuating the overall p-value from .033 to .04. In both, the pooled adjusted risk ratio for Holmes et al.¹⁸ favored pegfilgrastim. While the other four studies favored neither agent, the weight (45.1%) of Holmes et al.¹⁸ in Cooper et al.²¹ tilted the pooled risk estimate into marginal significance (p = .04). The possibility of publication bias cannot be excluded. Pinto et al.²⁰ also reported pooled risk ratios for FN incidence in cycle 1 (p = .100), and the incidence of neutropenia grade 4 in cycles 1–4 (p = .365), and pooled standardized difference in means for time to ANC recovery (p = .669); none of which were statistically significant.

Further, Cooper et al.’s²¹ meta-analysis is an extraction from a larger study that also used Bayesian mixed-treatment comparison using Markov chain Monte Carlo simulations²⁶. This analysis failed to show a difference in FN rate between pegfilgrastim and filgrastim across the randomized trials.

The meta-analysis by Wang et al.²² on FN incidence includes the five studies above and adds a phase II trial²⁷ comparing the biosimilar pegfilgrastim DA-3031 to daily filgrastim to the all- cycles analysis. This additional trial reported FN incidence rates, but only for cycle 1 (rates were statistically similar), not for any other cycles. This makes the inclusion of this additional study questionable at best. Bensink et al.⁹ also refers to a combined [sic] analysis²⁸ of the two phase III non-inferiority studies^18,19. Again, the trials’ primary end-point of duration of grade 4 neutropenia is not included, and this analysis mixes weight-based and fixed pegfilgrastim dosing vs weight-based filgrastim dosing.

In summary, Bensink et al.⁹ state we “do not provide sufficient rationale, nor a sufficiently robust critical appraisal, to justifying excluding this evidence” from two non-inferiority studies, three meta-analyses, and a combined analysis. The evidence presented herein should remedy this. Most importantly, through scientific analysis, argumentation, and discourse rather than insufficiently detailed categorical statements, we demonstrate (1) that the strongest evidence (phase III trials) documents unequivocally, and solely, the non-inferiority of pegfilgrastim relative to filgrastim on the primary end-point of duration of grade 4 neutropenia episodes; (2) that focusing on secondary end-points does not alter this conclusion; (3) that neither the inclusion of phase II trials does; and (4) that, apart from a challengeable finding on all-cycle FN incidence, pegfilgrastim and filgrastim are comparable on the secondary end-points as well. Despite what Bensink et al.⁹ claim, there is insufficient scientific evidence to sustain their statement that “filgrastim and pegfilgrastim are not equivalent” (they are) and that results of the analyses cited “collectively provide evidence that pegfilgrastim provides significant benefit over filgrastim” (they do not). Note that our conclusions are aligned with the evidence-based EORTC²⁹ and NCCN³⁰ guidelines.

A note on including 1–14 doses of filgrastim in our analyses is in order. Our study was a simulation of costs and savings across a continuum. Therefore, the full range of possible doses was included. In our follow-on papers on savings-enabled expanded access^3,6 we considered real-world evidence on filgrastim practice patterns from Weycker et al.³¹ in NHL (mean ± SD = 6.6 ± 3.1 days), breast cancer (6.1 ± 2.9 days), and lung cancer (4.3 ± 3.1 days). We also applied real-world findings from the MONITOR-GCSF study of EP2006 for all patients (5.1 ± 2.3 days), patients with solid tumors (5.1 ± 2.2 days), and patients with hematological malignancies (5.0 ± 2.6 days)³². The evidence that the “use of inappropriately low doses of filgrastim in real-world clinical practice has been shown to increase the risk of febrile neutropenia”⁹ is associative, not causative; and has not been evaluated in RCTs. More generally, and more importantly so, the MONITOR-GCSF study has shown that clinicians are foregoing normative filgrastim administration patterns and opting for shorter regimens³²; that clinical outcomes and CIN/FN-related hospitalizations and chemotherapy disturbances are not necessarily higher and, moreover, can be predicted at both the patient- and the cycle-levels³³; that clinicians often “over-prophylact” patients relative to guidelines³⁴, and that this is associated with better outcomes³³. Add to this the reality that US payers are increasingly denying prophylaxis with pegfilgrastim and are authorizing up to seven injections of filgrastim instead, bringing a risk-management perspective to the debate on the appropriateness of fewer daily injections.

Assumptions prevail over opinions

The objective of our study¹ was to evaluate the cost-efficiency of CIN/FN prophylaxis in terms of drug acquisition and drug administration costs of prophylaxis under different scenarios. The objective was not to conduct cost-effectiveness analyses, as Bensink et al.⁹ recommend that we should have done, but fail to specify in terms of the outcome denominator(s) to be used. Further, adding in the “cost of cancer-related neutropenia or fever hospitalizations”⁹ would be relevant if there was RCT-based evidence of differences in hospitalization rates between filgrastim and pegfilgrastim. In the absence of such evidence, and considering the comparable prophylaxis efficacy, adding in this cost would not be a differentiator, as it would apply equally across all agents, and any differences would likely disappear in sensitivity analyses. Our study was transparent, perhaps exceedingly so, about its assumptions, scenarios, and methodologies, and how these translated into our results.

Bensink et al.⁹ conclude by alluding to additional concerns. Lacking clarity, specificity, and substantiation, we are not able to reply to these concerns.

Conclusions

This Reply has given us the opportunity to provide the evidence that was presumably missing in our paper. We have been able to counter, with scientific evidence and logic of inference, the many statements by Bensink et al.⁹ that were premised on postulations, perspectives, and opinions; rather than being grounded in careful scientific appraisal of objectives, designs and methodologies, statistical analyses, results, and conclusions. The scientific evidence provided herein reaffirms the conclusions from our recent¹ and prior study².

Transparency

Declaration of funding

This work was sponsored by Sandoz, Inc.

Declaration of financial/other relationships

AM serves on a Speakers Bureau and Steering Committee for Sandoz, Inc. He was subcontracted by Matrix45 for work on this project. KC, MB, and SB are employees of Sandoz, Inc., the study sponsor. IA and KM are employees of Matrix45, which was contracted by Sandoz, Inc. to conduct the simulations. By company policy, employees are prohibited from owning equity in client organizations (except through mutual funds or other independently administered collective investment instruments) or contracting independently with client organizations. Matrix45 provides similar services to those described in this article to other biopharmaceutical companies on a non-exclusivity basis. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

None.