See accompanying article on pages 2584–92, this issue.
Portions of this study were presented at the 44th Annual Meeting of the American Society for Hematology, December 6–10, 2002, Philadelphia, Pennsylvania, and at the 45th Annual Meeting of the American Society for Hematology, December 6–9, 2003, San Diego, California.
Despite a lack of long-term data, imatinib has become standard therapy for patients with newly diagnosed chronic-phase chronic myeloid leukemia (CML) who are not candidates for allogeneic stem cell transplantation. In the current study, the authors estimated the incremental cost-effectiveness of imatinib versus interferon-α plus low-dose cytarabine (IFN+LDAC) as first-line therapy for these patients.
Data from the International Randomized Interferon versus STI571 Study and the literature were used to estimate lifetime costs, survival, and quality-adjusted survival. Survival estimates were based on published survival curves for patients who achieved and those who did not achieve a complete cytogenetic response after treatment with interferon-α.
The mean estimated survival with first-line imatinib therapy was 15.30 years, compared with 9.07 years with IFN+LDAC. Undiscounted lifetime costs were approximately $424,600 with imatinib and $182,800 with IFN+LDAC. Using a 3% discount rate, the incremental survival gain with imatinib was 3.93 life-years and 3.89 quality-adjusted life-years (QALYs). Incremental discounted lifetime costs were found to be $168,100 higher with imatinib, resulting in incremental cost-effectiveness ratios of $43,100 per life-year saved (95% confidence interval [95% CI], $37,600–51,100) and $43,300 per QALY (95% CI, $38,300–49,100).
Chronic myeloid leukemia (CML), a malignant myeloproliferative disorder of hematopoietic stem cells, is reported to account for 15–20% of adult leukemia cases.1–3 CML typically presents as a triphasic disease, beginning with a chronic phase that lasts for 3–6 years.1–3 The relatively asymptomatic chronic phase is followed by progressively symptomatic accelerated and blastic phases. The duration of the advanced phases is measured in months.4–11
Imatinib mesylate (Gleevec™; Novartis, Basel, Switzerland) reportedly offers clinical benefits during advanced phases of CML12, 13 and induces high rates of cytogenetic response in the chronic phase among patients who have failed therapy with interferon-α (IFN-α).14, 15 The International Randomized Interferon versus STI571 Study (IRIS) was an open-label trial that compared the efficacy of imatinib versus IFN-α plus low-dose cytarabine (IFN+LDAC) in 1106 patients from 16 countries who were newly diagnosed with chronic-phase CML.16 With a median follow-up of 19 months, 2.0% of the patients randomized to receive imatinib and 57.5% randomized to receive IFN+LDAC crossed over to the alternate treatment after failing first-line therapy. An additional 12.3% of patients in the imatinib group and 31.6% of patients in the IFN+LDAC group discontinued study medication while receiving first-line therapy. Clinical measures were significantly better among the patients receiving imatinib. At 18 months, the rate of complete cytogenetic response (CCyR) to first-line therapy was estimated at 76.2% for imatinib, compared with 14.5% for IFN+LDAC. Estimated rates of progression to accelerated phase or blast crisis were 3.3% in the imatinib group and 8.5% in the IFN+LDAC group.
Despite a lack of long-term data, imatinib has become standard therapy for patients who are newly diagnosed with chronic-phase CML and who are not candidates for allogeneic stem cell transplantation. Nevertheless, in increasingly cost-conscious health care markets around the world, there is demand for evidence of cost-effectiveness—even for breakthrough therapies.17, 18 Moreover, with the passage of the Medicare Prescription Drug, Improvement, and Modernization Act of 2003,19 ready availability of data to demonstrate the value of outpatient pharmaceuticals will be critical to ensuring that patients have access to appropriate treatments. In the current study, we sought to estimate the incremental cost-effectiveness of imatinib compared with IFN+LDAC as the first-line treatment for patients with newly diagnosed chronic-phase CML.
MATERIALS AND METHODS
Using data collected in IRIS and supplemental data from the literature, we conducted an economic evaluation to estimate expected costs, survival, and quality-adjusted survival for patients with newly diagnosed chronic-phase CML who were receiving imatinib or IFN+LDAC as first-line therapy. The institutional review board of Duke University Medical Center approved this study. We assumed that patients receiving imatinib as first-line therapy could switch to IFN+LDAC and then to hydroxyurea until disease progression. Patients receiving IFN+LDAC as first-line therapy were assumed to switch to hydroxyurea at the time of discontinuation of IFN+LDAC.
We designed our analysis to track when patients were in the three phases of CML and were receiving various treatments. We assigned different rates of resource use and utility weights to patients in the chronic and advanced phases of the disease. For patients in chronic phase, we further differentiated assignments of resource use and utility weights according to treatment regimen. The analysis explicitly incorporated the uncertainty associated with each parameter and provided the capability to test assumptions regarding efficacy, survival, duration of treatment, resource use, costs, quality-of-life weights, and discount rates. Table 1 shows the estimates used in the base–case analysis. Cost estimates were valued in 2002 U.S. dollars. We conducted the analysis from the health care system perspective and considered only direct medical costs. In the base–case analysis, we discounted cost and survival estimates at 3% per year.
Table 1. Parameters and Estimates Included in the Base–Case Analysis
SE: standard error; IRIS: International Randomized Interferon versus STI571 Study; IFN + LDAC: interferon-alpha plus low-dose cytarabine; CML: chronic myeloid leukemia; CPT: Current Procedural Terminology; DRG: diagnosis-related group.
Includes all patients who either discontinued the initial treatment (n = 68) or crossed over to treatment with interferon-α plus low-dose cytarabine (n = 11), but excludes patients who died (n = 6).
Includes all patients who discontinued treatment with interferon-α in Years 2 or 3.
Assumed that patients who receive treatment in the advanced phases of CML receive one course of chemotherapy.
Mitoxantrone at a dose of 12 mg/m2/day for 4 days, etoposide at a dose of 100 mg/m2/day for 4 days, and cytarabine at a dose of 1 g/m2 twice daily for 4 days.
Cost differential between facility-based and nonfacility-based fees for CPT code 99214 is $22.99 plus $0.587 per minute for cost of RN-Oncology in 2002 from the U.S. Federal Register for a 20-minute visit.
Medicare reimbursement was divided by the mean length of stay for Medicare patients derived from the Healthcare Cost and Utilization Project available from the Agency for Healthcare Research and Quality.
Complete cytogenetic response at Year 2, proportion (SE)
Survival for the 2 years after the initiation of treatment for chronic-phase CML was based on data from IRIS for patients receiving imatinib as first-line therapy and data from a randomized trial conducted by the Italian Cooperative Study Group on CML for patients receiving IFN+LDAC.20 The IFN+LDAC group in the IRIS was not an appropriate comparison group because of the high rates of crossover to imatinib therapy.
Because to our knowledge there were no long-term survival data for imatinib, survival beyond the first 2 years was based on the relation between cytogenetic response and survival.21–26 We assumed that the survival distribution for patients who achieved a CCyR within 2 years of initiating treatment was equivalent to the survival distribution of a cohort of 322 patients who achieved a CCyR and were studied by the European Study Group on Interferon in CML.24 The survival distribution for patients who did not achieve a CCyR was based on patients without a CCyR in the Italian trial examining IFN+LDAC that was discussed earlier.20 Using these distributions, we estimated log hazard ratios of the increased risk of death among patients with and without CCyRs compared with an age-matched and gender-matched cohort from the general population.27 We used the log hazard ratios and their standard errors to simulate survival curves for patients achieving and those not achieving a CCyR at 2 years of follow-up, conditional on the patients being alive at 2 years. The survival estimation methods are discussed in greater detail in the accompanying article by Anstrom et al. in this issue of Cancer.28
Advanced Phases of CML
Estimates of the efficacy of imatinib from clinical trials involving patients in accelerated phase and blast crisis13, 29 were not thought to be generalizable to patients in advanced phases of the disease who had failed treatment with imatinib in the chronic phase. During advanced phases of CML, we assigned simulated patients in both treatment groups the same distributions for time in accelerated phase and blast crisis and the same distributions of utility weights and estimates of resource use. We derived the distributions assigned to estimate the duration in the accelerated phase prior to blast crisis and the duration in blast crisis prior to death by calculating the area under the curve from published survival curves.5, 6
Changes in Treatment
The estimated proportion of patients who switched from first-line imatinib to second-line IFN+LDAC was based on data collected in the IRIS through July 2002.16 For patients assumed to have switched to second-line IFN+LDAC, we assumed that 50% would discontinue the regimen at 34 months.26 The estimated proportion of patients assumed to have switched from first-line IFN+LDAC to hydroxyurea was based on the Italian Cooperative Study Group trial.22 We did not use data from the IRIS to estimate the proportion of patients switching from IFN+LDAC to imatinib because we believed that patients would be less likely to discontinue IFN+LDAC if imatinib was not a treatment option. In the base–case analysis, we assumed that patients who did not switch to second-line therapy continued to receive first-line therapy until disease progression.
The EuroQol-5D (EQ-5D), a preference-based measure of health-related quality of life,30 was administered every 3 months to patients in the IRIS. We converted responses to categoric items of the EQ-5D into community-weighted utility scores. After stratifying patients by treatment group and clinical phase, we calculated the mean utility weights for patients receiving first-line therapy in the chronic phase. Because data were not available for patients receiving monotherapy with hydroxyurea, we assigned utility values in the base–case analysis to patients receiving hydroxyurea using values from patients receiving imatinib. This assumption results in more conservative estimates (i.e., a higher cost-effectiveness ratio for imatinib) compared with using utilities from patients receiving IFN-α. Because of the small numbers of patients progressing to advanced phases of CML, the utility weights assigned to those phases were based on pooled data from IRIS patients in either of the clinical phases, regardless of their initial treatment assignment.
Estimates of monthly counts of resource use were based on data collected in the IRIS for patients receiving first-line therapy. We assumed conservatively that patients receiving hydroxyurea experienced the same level of resource use as patients receiving imatinib. In the advanced phases of CML, estimates of resource use were based on pooled data from patients in both treatment groups in either the accelerated phase or blast crisis.
We included costs for four medications in the economic analysis: imatinib, IFN-α, cytarabine, and hydroxyurea. Variability in medication costs between patients was based on the dose and duration of treatment with each medication, which we derived from detailed dosing information collected in the IRIS. Based on the total number of daily doses of medication dispensed, 98% of the patients in the imatinib group were assumed to receive 400 mg per day, 1.1% were assumed to receive 600 mg per day, and 0.9% were assumed to receive 800 mg per day. The mean daily dose of IFN-α was 4.81 MU (standard deviation [SD] of 2.27 MU). We assumed that each simulated patient maintained the same dose of medication.
For each population-level simulation run, we simulated 1000 patients and computed mean estimates of costs, survival, and quality-adjusted survival for each treatment strategy. We varied the estimates for the parameters simultaneously according to their assigned distributions, as specified using means and standard errors.
The assigned distributions correspond to second-order uncertainty that is representative of average values across samples of patients from the same population.31 Within each population-level run, estimates were simulated for individual patients using another set of distributions that represent variability at the patient level. We modeled proportions as random variables from beta distributions at the population level, and we used a random number generator to perform patient-level simulations for each parameter. For utility values, we used a beta distribution to vary estimates between population-level runs. We used a normal random variable to simulate values at the patient level. Simulated values were bounded by 0 and 1. We used a normal random variable to generate estimates at the population level for the mean time patients spent in the accelerated phase and blast crisis. We incorporated an exponential distribution into the analysis to replicate the survival distribution for time in the advanced phases of CML for the patient-level simulations. This specification produced a right-skewed distribution of time in the accelerated phase and blast crisis that more closely mimicked what occurs in individual patients. We modeled counts of resource use at the population level as random variables from a beta distribution to generate the mean number of inpatient days or outpatient visits per month. At the patient level, we used a binomial distribution to generate the total number of inpatient days and outpatient visits, dependent on the amount of time each patient spent in each treatment or clinical phase. Then, we distributed counts of resource use evenly across time. This allowed us to better simulate true variability in the use of resources between patients.
The results of the base–case analysis were based on 1000 population-level runs and 95% confidence intervals (95% CIs) for the incremental cost-effectiveness ratios (ICERs) were estimated using the percentile method, whereby the 26th and 975th rankings of the 1000 simulated ICERs were used as the 95% CI limits. We rounded reported estimates of costs and ICERs to the nearest $100.
We conducted several sensitivity analyses to evaluate the impact of varying baseline estimates and assumptions, including the following: 1) applying utility and resource use estimates from patients receiving IFN-α (instead of imatinib) as proxies for patients receiving hydroxyurea; 2) instituting a policy whereby patients could receive first-line therapy with imatinib or IFN+LDAC beyond the first 2 years only if they had achieved a CCyR; 3) increasing or decreasing the cost of study medications by 25%; 4) assuming that all patients receiving IFN-α received the recommended target dose (5 MU/m2/day); 5) using median (instead of mean) survival estimates of 6 months for time spent in the accelerated phase and blast crisis; and 6) increasing the hazard rate of death by 50% for patients receiving first-line imatinib.
We also explored the use of alternative sources for estimating survival for patients with and without CCyRs and the impact of increasing the proportion of patients receiving IFN+LDAC who achieve a CCyR. For patients without a CCyR, we substituted the survival curve in the base–case analysis for the one reported by the French CML Group.26 For patients who achieved a CCyR, we substituted the survival curve in the base–case analysis for the curve reported by Kantarjian et al.21 Because approximately one-third of patients achieving a CCyR while receiving IFN-α therapy do so after 2 years of treatment, we increased the proportion of patients achieving a CCyR while receiving IFN+LDAC by 1.63 to examine the effect of the higher estimate.24 To demonstrate the impact of varying the relative effectiveness of the treatment strategies further, we varied the percentage of patients who achieved a CCyR in the imatinib group from 20–100% while holding the percentage of patients who achieved a CCyR in the IFN+LDAC group constant at 14.2% and plotted the resulting ICERs.
In the base–case analysis, patients receiving first-line therapy with imatinib were estimated to live an average of 15.30 years compared with 9.07 years for patients receiving first-line therapy with IFN+LDAC, an incremental gain of 6.23 years (Table 2). After adjusting for quality of life, the incremental gain was 5.85 quality-adjusted life-years (QALYs). Undiscounted lifetime costs were estimated to be $424,600 for patients receiving imatinib compared with $182,800 for patients receiving IFN+LDAC, a difference of $241,800.
Table 2. Cost-Effectiveness in the Base-Case Analysis
95% confidence intervals were based on the 26th and 975th rankings of the 1000 base-case simulations.
After applying a 3% discount rate, the incremental gain in survival was 3.93 years and 3.89 QALYs. Discounted lifetime costs were $168,100 higher, on average, among patients receiving first-line therapy with imatinib. Combining these estimates resulted in ICERs that were equal to $43,100 per life-year saved and $43,300 per QALY. The upper limits of the 95% CIs for both ratios were < $51,100. Bootstrap scatter plots of the ICERs revealed a relatively low level of variability in the results and a high degree of correlation between incremental differences with regard to costs and survival (Fig. 1). Approximately 10% of the simulated ICERs were < $40,000 per QALY, 75% were < $45,000 per QALY, and 98% were < $50,000 per QALY. All simulated ICERs were < $56,000 per QALY.
Varying the assumptions used to assign utility values and estimates of resource use during treatment with hydroxyurea was found to have little impact on the ICERs (Fig. 2). Likewise, there was little effect noted for assuming that patients would spend a median of 6 months in the advanced phases of CML. Even increasing the hazard rate of death among patients receiving first-line imatinib by 50% was not found to have any substantial impact.
The ICERs were affected most by changes in assumptions regarding discontinuing first-line therapy and changes in the dose or price of imatinib or IFN+LDAC. When we assumed that first-line treatment would be discontinued if patients did not achieve a CCyR within 2 years, the main impact was an increase in the mean QALY estimate and a decrease in the mean cost estimate in the IFN+LDAC group. These changes occurred because patients who discontinued treatment with IFN+LDAC were assumed to then receive hydroxyurea. Because utility values for imatinib were assigned as a proxy during treatment with hydroxyurea, discounted quality-adjusted survival was found to increase by approximately 0.32 QALYs, and because the cost of hydroxyurea is much less than the cost of IFN+LDAC, discounted costs decreased by approximately $50,400, resulting in a larger cost difference between the 2 groups ($213,500). The decrease in costs noted in the imatinib group ($5000) was slight because the majority of patients achieved a CCyR at 2 years, and those who did not switched to treatment with IFN+LDAC. These changes resulted in higher cost-effectiveness ratios ($53,900 per life-year saved and $60,500 per QALY). When we assumed that only patients receiving imatinib who did not achieve a CCyR would be required to discontinue first-line therapy, there was little effect noted on the ICER, because the majority of patients had achieved a CCyR. Furthermore, those patients who did not achieve a CCyR were assumed to have switched to IFN+LDAC, therefore incurring the cost of that therapy.
Previous analyses of the cost-effectiveness of IFN-α have shown that it is highly dependent on the cost and dose of the drug.32–34 Decreasing the cost of imatinib by 25% resulted in ICERs of $26,000 per life-year saved and $26,100 per QALY. Increasing the cost of imatinib by 25% resulted in ICERs of $60,900 per life-year saved and $61,100 per QALY. When we increased the cost of IFN-α by 25%, the ICERs were estimated at $36,700 per life-year saved and $36,600 per QALY. When we decreased the cost of IFN-α by 25%, the resulting ICERs were $48,900 per life-year saved and $49,200 per QALY. In the current study, if we had used the target dose of approximately 9.5 MU per day of IFN-α in the base–case analysis, the ICERs would have been much lower ($20,400 per life-year saved and $20,500 per QALY).
When we used survival curves published by the French CML Group26 to estimate survival for patients who did not achieve a CCyR, the mean discounted survival increased relative to the base–case findings by 0.21 life-years in the first-line imatinib therapy group and by 0.72 life-years in the first-line IFN+LDAC therapy group. Because costs increased along with survival, there was only a slight increase noted in the ICERs over the base–case results ($46,600 per life-year saved and $44,600 per QALY). When we used the survival curve reported by Kantarjian et al.21 to model survival for patients who achieved a CCyR, discounted survival for patients receiving imatinib was reported to decrease by 0.35 life-years relative to the base–case analysis, and there was little change noted for patients receiving IFN+LDAC. When we increased the proportion of patients receiving IFN+LDAC who achieved a CCyR to 22.9%, mean discounted costs and survival in the IFN+LDAC group increased by approximately $11,000 and 0.55 life-years. The resulting ICERs were found to be slightly higher than those in the base–case analysis ($46,300 life-years saved and $44,800 per QALY). Results also were relatively robust when varying the proportion of imatinib-treated patients who achieved a CCyR. When we set the proportion to 100%, the ICER was estimated at $41,400 per QALY; when we set the proportion to 20%, the ICER was estimated at $67,400 per QALY (Fig. 3).
In the current study, the incremental gain in survival among patients receiving first-line imatinib was approximately 6.2 years, or 5.9 QALYs, at an incremental cost of approximately $242,000. After discounting survival and costs at 3% per year, the survival gain in the imatinib group was approximately 3.9 years and 3.9 QALYs, and discounted costs were approximately $168,000 higher. The resulting ICERs of $43,000 per life-year saved and $43,300 per QALY compare favorably with the widely cited cost-effectiveness threshold of $50,000 per QALY.35
Few changes in the parameters were found to have an appreciable effect on the estimated ICERs. Much of the insensitivity was due to the relatively close tracking of survival and costs within treatment groups. For example, a change that increases survival in either treatment group generally increases costs to the same degree, because most patients are assumed to continue to receive therapy. Even when assuming that patients receiving imatinib have a mortality rate that is 50% higher than has been reported historically in patients receiving IFN-α, the ICERs are approximately $51,000 per life-year saved and $47,000 per QALY. Assumptions that affect the cost of treatment have a more substantial effect on the ICERs. When we increased the cost of imatinib by 25%, ICERs were approximately $61,000 per life-year saved or per QALY. When we increased the dose of IFN-α to the target dose of approximately 9.5 MU per day, the ICERs were approximately $20,000 per life-year saved and $20,000 per QALY.
The predictive power of the current analysis relies on the validity of several assumptions. The underlying structure of the analysis is based on the assumption that, conditional on the patient achieving a CCyR at 2 years, patients receiving imatinib have the same survival distribution as patients receiving IFN-α. This survival estimate assumes explicitly that survival after a CCyR will be identical, regardless of the treatment used to induce the CCyR—an assumption that appears to be conservative in that recent evidence suggests improved survival with imatinib among patients with and those without a cytogenetic response.36 Nevertheless, a potential criticism of this assumption is that the relation between CCyR and survival in patients receiving IFN-α exists largely because most patients receiving IFN-α who achieved a CCyR were at low risk according to scores from Sokal et al.37 or Hasford et al.38 and would have had longer survival for this reason alone.24 Although the debate is beyond the scope of the current discussion, we can evaluate the impact of restricting longer survival to patients who achieved a CCyR and who were deemed to be at low risk. Among patients in the IRIS for whom risk scores were available, 43.1% achieved a CCyR and were considered to be at low risk according to the model of Sokal et al.37 The corresponding percentage according to the model of Hasford et al.38 was 34.4%. If we assume that only those patients would benefit from longer survival associated with a CCyR (instead of the 73.8% reported in our base–case analysis) the ICERs would have been $48,800 per QALY and $53,900 per QALY, respectively.
We relied on clinical trials and published data to the extent possible. To our knowledge, the current study is the first to incorporate utility weights derived from patients undergoing treatment as opposed to estimates based on expert opinion. Our average utility estimate of 0.710 for patients receiving IFN+LDAC was lower than the average utility rates of 0.9 and 0.875 assigned in previous analyses.32, 34, 39 Therefore, it appears that the quality-adjusted survival for patients treated with IFN+LDAC based on expert opinion may have been overestimated in the past. Even among patients receiving imatinib, with its superior side effect profile, the average utility weight was 0.854. The current study also relies on counts of resource use for patients receiving first-line therapy in IRIS. Although it is possible that short-term resource use in the trial may not reflect long-term resource use, our results were not sensitive to changes in the estimates of resource use because the cost of imatinib and IFN-α outweighed the effects of changes in other direct medical costs. Even after doubling the estimates of resource use for patients treated with IFN+LDAC and assigning them to the imatinib group, the resulting ICER increased to just $54,100 per QALY.
When data were not available, we relied on conservative assumptions. Although imatinib has been shown to be effective in the advanced phases of CML, we assumed that it provides no additional survival benefit. We also assumed that, unless they were switched to second-line therapy, patients would remain on first-line therapy until entering the accelerated phase of the disease or blast crisis.
The current study addresses the question of whether first-line imatinib in patients with chronic-phase CML is cost-effective compared with IFN+LDAC when imatinib is not available after the discontinuation of IFN+LDAC. We have not addressed whether imatinib is cost-effective as first-line therapy or whether it should be reserved for second-line therapy for patients who do not respond, lose a response, or cannot tolerate therapy with IFN+LDAC. Providing an estimate of the cost-effectiveness of first-line versus second-line therapy is highly dependent on local practice patterns. The IRIS prespecified conditions under which patients randomized to IFN+LDAC could switch to imatinib therapy. If local health care authorities were to adhere to similar criteria, the cost-effectiveness of first-line versus second-line imatinib could be determined, subject to these conditions.
Literature regarding the use and effectiveness of imatinib in clinical trial settings and specialty practice has proliferated. Although studies have documented the efficacy of imatinib in the treatment of patients with CML,40 reports of resistance suggest that there is room for improvement in our understanding of this drug.41–43 Experts have called for trials using higher doses of imatinib, combination therapies, and other novel therapies for CML.44–47 As additional strategies emerge and long-term data become available, it will be important to evaluate the cost-effectiveness of add-on therapies, update our evaluations of existing therapies, and determine how well our analysis reflects real-world experience. Based on a culmination of current evidence, our analysis shows that imatinib is a cost-effective first-line therapy for patients with newly diagnosed chronic-phase CML compared with IFN+LDAC.
The authors thank the coinvestigators; the medical, nursing, and research staffs; and the trial monitors, data managers, and programmers for their contributions to the IRIS study. They also thank Damon Seils for editorial assistance and article preparation.