Title: When Evaluating Parameter Uncertainty Is Not Enough: The Case of Dasatinib and Nilotinib for Imatinib-Resistant Chronic Myeloid Leukemia
Abstract: Chronic myeloid leukemia (CML) accounts for one in five cases of leukemia in adults [[1]Sawyers C.L. Chronic myeloid leukemia.N Engl J Med. 1999; 340: 1330-1340Crossref PubMed Scopus (1273) Google Scholar]. Tyrosine kinase inhibitors (TKIs) have revolutionized the treatment paradigm for CML. The first, imatinib (Gleevec/Glivec), was approved in 2002 by the U.S. Food and Drug Administration and the European Agency for the Evaluation of Medicinal Products. Although data on the long-term effectiveness of imatinib were sparse, early cost-effectiveness analyses based on conservative associations between short-term cytogenetic response and long-term quality-adjusted survival indicated that its use in chronic-phase disease was of good value, despite a cost of $29,000 per year in 2002 [[2]Reed S.D. Anstrom K.J. Ludmer J.A. et al.Cost-effectiveness of imatinib versus interferon-alpha plus low-dose cytarabine for patients with newly diagnosed chronic-phase chronic myeloid leukemia.Cancer. 2004; 101: 2574-2583Crossref PubMed Scopus (70) Google Scholar]. Consistent with these findings, guidance from the U.K. National Institute for Health and Clinical Excellence recommends the use of standard-dose imatinib (400 mg/d) as first-line therapy for chronic-phase CML [[3]National Institute for Health and Clinical ExcellenceThe clinical and cost effectiveness of imatinib for first-line treatment of chronic myeloid leukemia.http://guidance.nice.org.uk/TA70Date: October 2003Google Scholar]. Some patients do not respond to standard-dose imatinib (i.e., primary resistance), and some patients lose their response (i.e., secondary resistance). In the landmark International Randomized Study of Interferon and STI571, approximately one in four patients had primary resistance at 18 months [[4]O'Brien S.G. Guilhot F. Larson R.A. et al.Imatinib compared with interferon and low-dose cytarabine for newly diagnosed chronic-phase chronic myeloid leukemia.N Engl J Med. 2003; 348: 994-1004Crossref PubMed Scopus (2991) Google Scholar]; by 5 years, one in four patients had developed secondary resistance [[5]Druker B.J. Guilhot F. O'Brien S.G. et al.Five-year follow-up of 534 patients receiving imatinib for chronic myeloid leukemia.N Engl J Med. 2006; 355: 2408-2417Crossref PubMed Scopus (2889) Google Scholar]. Although stem cell transplant is an option, many patients are not appropriate candidates. Other options include interferon alfa and hydroxyurea/hydroxycarbamide, but increasing the dose of imatinib to 600 or 800 mg/d or switching to treatment with a second-generation TKI is considered to be more effective [[6]Jabbour E. Cortes J. Kantarjian H. Long-term outcomes in the second-line treatment of chronic myeloid leukemia.Cancer. 2011; 117: 897-906Crossref PubMed Scopus (51) Google Scholar]. Two second-generation TKIs are approved for the treatment of patients with resistance to imatinib: dasatinib (Sprycel) and nilotinib (Tasigna). Although a single randomized clinical trial compared dasatinib to high-dose imatinib in patients with imatinib resistance [[7]Kantarjian H. Pasquini R. Lévy V. et al.Dasatinib or high-dose imatinib for chronic-phase chronic myeloid leukemia resistant to imatinib at a dose of 400 to 600 milligrams daily: two-year follow-up of a randomized phase 2 study (START-R).Cancer. 2009; 115: 4136-4147Crossref PubMed Scopus (178) Google Scholar], adequately powered head-to-head trials of nilotinib and dasatinib have not been conducted [[6]Jabbour E. Cortes J. Kantarjian H. Long-term outcomes in the second-line treatment of chronic myeloid leukemia.Cancer. 2011; 117: 897-906Crossref PubMed Scopus (51) Google Scholar]. Furthermore, dissimilarities in inclusion criteria, study designs, and variations in end-point definitions across studies impede the ability to make high-quality direct or indirect comparisons [[6]Jabbour E. Cortes J. Kantarjian H. Long-term outcomes in the second-line treatment of chronic myeloid leukemia.Cancer. 2011; 117: 897-906Crossref PubMed Scopus (51) Google Scholar]. On the basis of best available evidence, clinical experts consider the expected clinical outcomes with dasatinib and nilotinib for second-line treatment to be similar [6Jabbour E. Cortes J. Kantarjian H. Long-term outcomes in the second-line treatment of chronic myeloid leukemia.Cancer. 2011; 117: 897-906Crossref PubMed Scopus (51) Google Scholar, 8National Institute for Health and Clinical ExcellenceLeukaemia (chronic myeloid)—dastatinib, nilotinib, imatinib (intolerant, resistant) (including partial review of TA70): final appraisal determination.http://guidance.nice.org.uk/TA/WaveR/105/FADGoogle Scholar, 9Stein B. Smith B.D. Treatment options for patients with chronic myeloid leukemia who are resistant or unable to tolerate imatinib.Clin Ther. 2010; 32: 804-820Abstract Full Text PDF PubMed Scopus (33) Google Scholar]. In this issue of Value in Health, Hoyle et al. report survival estimates from a decision-analytic model designed to evaluate the cost-effectiveness of nilotinib and dasatinib in comparison with high-dose imatinib for second-line treatment of CML. Among patients with resistance to imatinib, expected survival was an estimated 12.4 years with high-dose imatinib, 13.4 years with dasatinib, and 13.0 years with nilotinib. Although the survival estimates for nilotinib and dasatinib were slightly greater than for high-dose imatinib, all three estimates had 95% confidence intervals ranging from 8 to 17 years. Cumulative discounted costs for high-dose imatinib and nilotinib were an estimated £89,000 and £70,000, respectively, approximately half of the £161,000 estimated for second-line dasatinib. Although the authors report that nilotinib dominated high-dose imatinib (i.e., was less costly and more effective), the incremental cost-effectiveness ratio for dasatinib compared with that for imatinib was an estimated £91,500 per quality-adjusted life-year. One possible explanation is that the price of dasatinib was approximately double the price of nilotinib. However, the monthly cost applied to dasatinib was slightly lower (£1169) than the cost applied to nilotinib (£1217). If two drugs are thought to be equally effective and their costs are similar, why would an economic evaluation show that one was economically dominant and the incremental cost-effectiveness ratio for the other vastly exceeded the £30,000 per quality-adjusted life-year mark compared with high-dose imatinib? The short answer is that patients were assumed to take second-line high-dose imatinib and nilotinib for just 2.7 and 2.4 years, respectively, compared with 6.5 years for patients receiving second-line dasatinib. The authors lacked direct information about treatment duration, and so they derived estimates of treatment duration from estimates of progression-free survival and 3-month discontinuation rates. For patients with imatinib resistance, progression-free survival was an estimated 0.63 at 18 months for nilotinib and 0.77 at 2 years for dasatinib. Although direct comparisons are rife with limitations, there appears to be an advantage with dasatinib. Nevertheless, this potential advantage gave rise to longer treatment duration with dasatinib, resulting in considerably higher costs and qualitatively different findings with regard to cost-effectiveness. Nearly 10 years ago, the ISPOR Good Research Task Force published recommendations on decision-analytic modeling [[10]Weinstein M.C. O'Brien B. Hornberger J. et al.Principles of good practice for decision analytic modeling in health-care evaluation: report of the ISPOR Task Force on Good Research Practices—modeling studies.Value Health. 2003; 6: 9-17Abstract Full Text PDF PubMed Scopus (913) Google Scholar]. Although numerous specific recommendations were included, Garrison's [[11]Garrison L.P. The ISPOR Good Practice Modeling Principles—a sensible approach: be transparent, be reasonable.Value Health. 2003; 6: 6-8Abstract Full Text PDF PubMed Scopus (18) Google Scholar] summary was simple: models should be transparent and reasonable. Without specific information about treatment duration, it may have been reasonable to apply a proxy. Most patients with CML will continue to receive TKIs until disease progression and for some time thereafter [[8]National Institute for Health and Clinical ExcellenceLeukaemia (chronic myeloid)—dastatinib, nilotinib, imatinib (intolerant, resistant) (including partial review of TA70): final appraisal determination.http://guidance.nice.org.uk/TA/WaveR/105/FADGoogle Scholar]. Therefore, the decision by Hoyle et al. to estimate treatment duration as a function of progression-free survival and short-term rates of discontinuation in clinical studies may have been reasonable. To evaluate uncertainty, the authors performed probabilistic sensitivity analysis, which projected that dasatinib was more costly than high-dose imatinib “in virtually all simulations,” whereas nilotinib was more costly than high-dose imatinib in just 9% of simulations. This seemingly compelling finding, however, does not address whether it makes sense that approximately 2.5 years of treatment with nilotinib and 6.5 years of treatment with dasatinib provide the same level of effectiveness. The authors acknowledge that the “duration of treatment is a key input.” Indeed, they report that when progression-free survival for dasatinib was assumed to be the same as for nilotinib, dasatinib became economically dominant relative to high-dose imatinib. Given the structure of the model, the probabilistic sensitivity analysis represented uncertainty associated with estimates of progression-free survival, not uncertainty about the true values for treatment duration. This is unfortunate because some users of cost-effectiveness analyses may rely too readily on findings from probabilistic sensitivity analyses to gauge the reliability of a model's results. In accord with published recommendations for the conduct of cost-effectiveness analyses, reporting on results from probabilistic sensitivity analysis has increased dramatically over the last decade [[12]Jain R. Grabner M. Onukwugha E. Sensitivity analysis in cost-effectiveness studies: from guidelines to practice.Pharmacoeconomics. 2011; 29: 297-314PubMed Google Scholar]. Probabilistic sensitivity analysis is believed to represent an advance over traditional sensitivity analysis in which analysts have discretion over which parameters are tested, the ranges of parameter values applied, and their interpretation [[13]O'Brien B.J. Drummond M.F. Labelle R.J. Willan A. In search of power and significance: issues in the design and analysis of stochastic cost-effectiveness studies in health care.Med Care. 1994; 32: 150-163Crossref PubMed Scopus (358) Google Scholar]. Experts, however, have called for greater awareness of the importance of structural uncertainty [14Briggs A. Sculpher M. Claxton K. Decision modeling for health economic evaluations. Oxford University Press, Oxford, UK2006Google Scholar, 15Bilcke J. Beutels P. Brisson M. Jit M. Accounting for methodological, structural, and parameter uncertainty in decision-analytic models: a practical guide.Med Decis Making. 2011; 31: 675-692Crossref PubMed Scopus (103) Google Scholar]. Structural uncertainty is typically described as uncertainty about the functional form of a model, such as the number of health states in a Markov model or the statistical distribution used in a parametric regression model. In an evaluation of structural uncertainty, the analyst makes the necessary changes to the model and reports the results as a sensitivity analysis. In some cases, modifications to the model structure may be straightforward. In other cases, a complete overhaul may be necessary and, given resource constraints, unlikely to be done. When changes to a model's structure have little impact on the findings, the interpretation is simple. However, when structural changes alter a study's qualitative findings (i.e., switch from “cost-effective” to “not cost-effective” or vice versa), the analyst should further investigate which structure is more appropriate. In the analysis by Hoyle et al., an evaluation of structural uncertainty would have required application of alternative sources or assumptions to model treatment duration for nilotinib and dasatinib. Ideally, the analyst has direct measures of treatment duration. Such information was likely collected in previous clinical studies of dasatinib and nilotinib, albeit not reported in published articles. It is not clear whether the authors requested information on treatment duration from the manufacturers of nilotinib and dasatinib or the clinical investigators leading those studies. However, when access to important information is disallowed, it would be useful for authors to report such occurrences to justify the use of alternative approaches to parameter estimation. Multiple phase 2 and 3 trials are underway to evaluate the use of dasatinib and nilotinib as first-line therapy in patients with newly diagnosed chronic-phase CML. Early results indicate that these second-generation TKIs produce faster and deeper cytogenetic and molecular responses than do imatinib [[16]Kantarjian H.M. Baccarani M. Jabbour E. et al.Second-generation tyrosine kinase inhibitors: the future of frontline CML therapy.Clin Cancer Res. 2011; 17: 1674-1683Crossref PubMed Scopus (49) Google Scholar]. The National Institute for Health and Clinical Excellence has already undertaken an update of its review of first-line therapies for chronic-phase CML. Also, as understanding of resistance to dasatinib, nilotinib, and imatinib improves, discovery efforts are focused on overcoming specific mutations. As the development of third-generation TKIs progresses, another round of technology assessments for second-line and third-line therapy will be in order. Before the next iteration of technology assessments, health economists should clarify the types of information that are necessary for generating credible economic analyses. We should conduct value-of-information analyses to determine when a powerful economic argument can be made for ascertaining the data. Because treatment duration was a key input in the decision model reported by Hoyle et al., it is likely that value-of-information estimates would surpass the cost of measuring treatment duration directly. For years, health economists have argued for a seat at the study design table. We routinely request integrated collection of data on medical resource use, health utilities, and, in some cases, measures of patient time or productivity in clinical protocols. As our field evolves and we become more knowledgeable about the main determinants of cost-effectiveness in specific therapeutic areas, we must use the tools at our disposal to make coherent arguments about the value of collecting specific types of data. Source of financial support: The authors have no financial relationships to disclose.