Presented in part at the 29th Annual Meeting of the Society for Medical Decision Making, Pittsburgh, Pennsylvania, October 21-24, 2007.
The objective of this study was to examine the cost effectiveness of using a pharmacogenetic test for uridine diphosphate glycosyltransferase 1A1*28 (UGT1A1*28) variant homozygosity before administering irinotecan to patients with metastatic colorectal cancer.
A decision-analytic model from the Medicare payer perspective followed hypothetical patients who were treated with combined 5-fluorouracil, leucovorin, and irinotecan. Under usual care, patients received a full dose of irinotecan. With genetic testing, irinotecan dosage was reduced 25% in homozygotes with the UGT1A1*28 variant allele. Test performance, chemotherapy toxicity, and quality-of-life weights were derived from clinical literature and product labels, and costs were derived from 2007 Medicare fee schedules. Chemotherapy efficacy after dose reduction, adverse event risk, and other parameters were varied in 1-way and probabilistic sensitivity analyses. The authors also calculated the value of investing in further studies of chemotherapy efficacy after homozygote dose reductions.
Pretreatment genetic testing costs less ($272 savings per patient tested) and yields slightly improved quality-adjusted life expectancy (0.1 quality-adjusted day per patient tested; approximately 2 quality-adjusted hours). Results depended on treatment efficacy but not adverse event risk assumptions. The results indicated that testing would avoid 84 cases of severe neutropenia, including 4.4 deaths. At a threshold of $100,000 per quality-adjusted life year, the therapeutic efficacy of irinotecan in homozygotes after dose reduction had to be ≥98.4% of full-dose efficacy for genetic testing to remain preferred. Future studies to determine whether this efficacy level can be achieved have an economic value of $22 million.
In 2008, colorectal cancer was diagnosed in >150,000 men and women in the United States.1 Despite advances in screening and early detection, nearly 20% of patients present with metastatic disease.1 Many others will develop recurrent and/or unresectable colorectal cancer after completing adjuvant therapy and will be eligible for palliative chemotherapy to control their disease. Although oxaliplatin-based combination chemotherapy regimens (eg, combined folinic acid, 5-fluorouracil, and oxaliplatin plus bevacizumab) currently are favored for first-line treatment, regimens that contain irinotecan are accepted and sued widely in the first-line, second-line, and third-line (metastatic) settings.2, 3
Neutropenia is a serious side effect of irinotecan treatment, particularly in patients with decreased uridine diphosphate (UDP)-glucunorosyltransferase activity.4 Administered as a prodrug, irinotecan undergoes enzymatic conversion by carboxyesterase-2 to yield the clinically active metabolite, SN-38. This active form interferes with tumor cell division by inhibiting the nuclear enzyme topoisomerase I.5 SN-38 is eliminated from the body through the biliary system after a process of glucuronidation (conjugation to glucuronic acid) through the uridine diphosphate glucuronosyltransferase (UGT)1A1 enzyme. Several retrospective studies have demonstrated that individuals who are homozygous for the UGT1A1*28 variant are at increased risk of irinotecan-related neutropenia and diarrhea, particularly when irinotecan is administered at higher doses.4, 6-9 This likely is because of markedly reduced messenger RNA transcription, leading to decreased overall enzyme production, slower SN-38 glucuronidation, and a greater SN-38 plasma concentration over time, which results in prolonged exposure to the unconjugated form of SN-38.10 Approximately 11% of North Americans may be homozygous for this allele, although it is known that the prevalence is lower among Asian-Americans and African-Americans compared with Caucasians.5
In 2005, the US Food and Drug Administration (FDA) approved a UGT1A1 genotype test to help identify patients with a greater risk for decreased UDP-glucunorosyltransferase activity11 and, thus, with an increased risk of irinotecan-induced neutropenia. The manufacturer's label for irinotecan now recommends that physicians “consider” a dose-reduction of “at least 1 level” (equivalent to approximately a 17% to 34% dose reduction) when administering the drug to patients who are known to be homozygous for UGT1A1*28, but the final label also notes that the precise dose reduction in this population is unknown.12
Data are limited regarding the effects of irinotecan dose reduction on the incidence of side effects or on overall survival in patients with colorectal cancer. A recent meta-analysis indicated that lower doses (<150 mg/m2), versus medium doses (150-250 mg/m2) and high doses (>250 mg/m2), were associated with a lower probability of severe neutropenia in homozygotes.9 In addition, little is known about the costs associated with implementing UGT1A1 testing. Indeed, few cost-effectiveness studies have been conducted of pharmacogenetic tests for inherited mutations of any kind.13 The objective of the current study was to perform an economic evaluation of the use of pharmacogenetic testing to guide dosing of irinotecan versus irinotecan administration according to usual care in patients with metastatic colorectal cancer.
MATERIALS AND METHODS
We created a decision-analytic model from the Medicare payer perspective using TreeAge Pro 2008 (Williamstown, Mass)14 to evaluate the cost effectiveness of testing for the UGT1A1*28 genotype before irinotecan administration in patients with metastatic colorectal cancer compared with usual care (no testing) (Fig. 1). Hypothetical patients in the model were treated with 5-fluorouracil and leucovorin plus irinotecan (FOLFIRI) at a standard intermediate dose (175 mg/m2) that was calculated for a patient who had a body surface area of 1.85 m2 (eg, an individual 5 feet, 10 inches tall who weighed 70 kg).15, 16 We chose to study the FOLFIRI regimen for the model because it is the irinotecan-containing regimen for which the most data on UGT1A1*28 testing and outcomes currently are available. In the usual care strategy, all patients received a standard intermediate dose of irinotecan. In the genetic testing strategy, patients underwent UGT1A1*28 testing before FOLFIRI administration. Those who were identified as homozygous for the UGT1A1*28 variant allele received a 25% dose reduction of irinotecan therapy to 131.25 mg/m2 based on the FDA's label suggesting a dose reduction of at least 1 level, whereas all other patients received the full dose.
We began by assuming that the FOLFIRI regimen had 100% efficacy in all patients and had the same efficacy at a reduced dose among homozygotes as a full dose among heterozygotes or wild types, because there is slower glucuronidation and a greater plasma concentration of SN-38 in homozygotes.6 Then, this assumption was varied along with other model inputs in 1-way and probabilistic sensitivity analyses. A recent study indicated that response rates were not different between homozygotes, heterozygotes, or wild types, although homozygotes experienced more frequent dose reductions because of adverse events.17
We report incremental costs and quality-adjusted life expectancy when comparing the 2 strategies and the economic value of additional investment to reduce uncertainties in the modeling assumptions. We also report the estimated number of severe neutropenia events and deaths avoided by testing.
Adverse Events and Life Expectancy
Severe adverse events associated with irinotecan include grade 3 or 4 (severe) neutropenia.18, 19 The risks of severe neutropenia were derived from randomized trials of irinotecan administration in genotyped populations (Table 1).8, 20 Among homozygotes, the probability of severe neutropenia declined from 14% to 3.7% after genetic testing and dose reduction (vs 3.7% in the rest of the population) based on a study indicating that such dose reductions decreased the risk of a severe adverse event to a risk that was statistically equivalent to that of heterozygotes and wild types (P = .2).8 We reduced this benefit in sensitivity analyses. A fraction (23%) of patients experiencing severe neutropenia were assumed to be admitted to the hospital for management of febrile neutropenia,21 and the remainder were assumed to be treated on an outpatient basis. Because of the very small number of deaths caused by adverse events reported in each trial,15, 22-26 we assumed that the probability of death during a hospitalization for severe or febrile neutropenia was 0.1% (Table 1). The assumed life expectancy after FOLFIRI treatment was 24 months.27
Table 1. Variable Assumptions
UGT1A1 indicates uridine diphosphate glucuronosyltransferase 1A1; FOLFIRI, combined 5-fluorouracil, irinotecan, and leucovorin; FDA, US Food and Drug Administration; CMS-DHHS, Centers for Medicare and Medicaid Services, Department of Health and Human Services; CPT, Common Procedural Terminology; DRG, Diagnosis-related Group.
Probability of severe neutropenia: Homozygotes with full dose
2007 DRG weight×$7500 base rate (CMS-DHHS 200735), DRG 574
Chemotherapy, full dose
Medicare drug acquisition price (CMS-DHHS 200737), multiplied by 75% for reduced dose
Test Characteristics and Allele Prevalence
Genetic testing was assumed to be conducted using the Invader molecular assay developed by Third Wave Technologies (Madison, Wis), an FDA-approved test for UGT1A1*28 genotyping.28 Sensitivity and specificity of the assay were reported on the FDA website as 100%.29 However, we assumed sensitivity and specificity of 99.99% to allow for a small chance of a laboratory error that would affect test results. In the North American population, the prevalence of homozygotes was estimated at 11%.30
Quality of Life and Costs
We used previously published quality-of-life weights (utility values) associated with metastatic colorectal cancer and severe neutropenia (Table 1). A baseline weight of 0.83 was used for the population of metastatic colorectal cancer patients.31 Severe neutropenic events were assumed to last 1 week, leading to an additional decrement in quality of life during that week. Febrile neutropenia alone reportedly has been associated with a quality-of-life value of 0.42 (equivalent to a utility decrement of 0.58 from perfect health).32 The quality of life of patients experiencing both metastatic colorectal cancer and a severe neutropenic event (febrile neutropenia) simultaneously was estimated at 0.42 during 1 week using a lowest value approach.33
We used the Medicare payer perspective to evaluate costs. The model included costs of UGT1A1*28 testing,34 costs of hospitalization for febrile neutropenia,35 cost of a physician office visit,36 and costs of FOLFIRI chemotherapy for a full dose in wild types and heterozygotes and for a reduced dose in homozygotes.37 The model did not include the cost of subsequent chemotherapy, because the wide variety of possible treatments available impaired our ability to calculate this cost with an acceptable degree of accuracy. We wanted to estimate the maximum amount of future medical costs that might be incurred as a result of patients living longer because of the use of the test while still having the test be cost effective. Therefore, we estimated the threshold value at which the benefit of genetic testing would be offset by the additional expected lifetime medical costs from improved survival induced by genetic testing. All costs were reported in 2007 US dollars. Quality-adjusted life expectancy and costs were discounted at an annual rate of 3% per year, in accordance with the recommendation of the Panel on Cost-effectiveness in Health and Medicine.38
We performed 1-way sensitivity analyses on the efficacy of FOLFIRI dose reduction, incidence of and mortality from severe neutropenic events, genetic test performance, cost, and quality-of-life model inputs for a specified range of values (Table 1). It is noteworthy that we varied the efficacy of a dose reduction from 75% to 100% of full-dose efficacy; in the model, a reduction in efficacy resulted in a proportional reduction in life expectancy after FOLFIRI treatment (eg, a 75% reduction in efficacy resulted in a reduction in life expectancy from 24 months to 18 months). We also considered different methods for combining quality-of-life weights for severe neutropenia and metastatic colorectal cancer.33
We evaluated the impact of substituting a societal perspective for the Medicare payer perspective by adding $900 to reflect the value of patient time associated with hospitalization39 and reducing the cost of the UGT1A1*28 test from $103 (the Medicare reimbursement rate) to $75, which represented the estimated cost to conduct the test in a hospital laboratory.
Value of Information
We conducted a probabilistic sensitivity analysis by running 1000 Monte Carlo simulations that varied all noncost variables simultaneously with a willingness-to-pay (WTP) threshold of $100,000 per quality-adjusted life year (QALY).40 We applied triangular distributions41 for all parameters, in which the peak was equal to the initial value and the lower and upper bounds were equal to the lowest or highest values reported in the literature. The exceptions to triangular distributions were those for which data were very limited: chemotherapy efficacy, quality-of-life weights, and probability of hospitalization for neutropenia. We selected uniform distributions for these parameters, because they reflect the highest possible degree of uncertainty and are the least “informative,” as they represent the distribution of possible values as a straight line within a range.
By using results from the probabilistic sensitivity analysis with a WTP threshold of $100,000 per QALY, we estimated the value of obtaining better quality data to reduce uncertainties in our model and, thus, the value that represented the maximum amount that should be spent to conduct a study of dose reductions in homozygotes. First, we calculated the expected value of perfect information (EVPI). EVPI can be interpreted as the value of completely removing the possibility of making an incorrect decision based on the current analysis because of its inherent uncertainty.42 Then, we calculated the expected value of partial perfect information (partial EVPI) for resolving just the uncertainty about the efficacy of a reduced dose of irinotecan in homozygotes. The partial EVPI represents the maximum value society should be willing to pay to eliminate the uncertainty about the efficacy assumption alone. We assumed that the test would continue to be useful for 5 years before being replaced by new technologies and that the size of the population to which the test could be applied was 29,260 patients with metastatic colorectal cancer annually,21 63% of whom we estimate are aged ≥65 years based on Surveillance, Epidemiology, and End Results data from 1973 through 2002.43 Finally we compared the partial EVPI with how much a clinical study might cost to improve decision making concerning dose reductions.
Adverse Events and Cost Effectiveness
For every 10,000 patients tested, 1100 (11%) would receive a 25% dose reduction as a result of being identified as homozygous for UGT1A1*28. On average, this treatment strategy would avoid 84.5 cases of severe neutropenia (including 4.5 deaths because of neutropenia) and would save $2.7 million in treatment costs. Taking into account the cost of testing and assuming no reduction in treatment efficacy, the average cost savings and quality-adjusted life expectancy savings per patient tested were $272.34 and 0.073 quality-adjusted days (Table 2), respectively (not even 2 quality-adjusted hours). If all 29,000 patients with metastatic colorectal cancer diagnosed each year in the US received chemotherapy that included irinotecan, then the estimated net annual savings could be as high as $7.96 million and 6 QALYs as a result of avoiding 245 severe neutropenic events (including 13 fatal events).
Table 2. Cost-Effectiveness Results Depending on the Efficacy of Dose Reduction in Homozygotes
Incremental Cost, $US
QALYs indicates quality-adjusted life years; ICER, incremental cost-effectiveness ratio; QALDs, quality-adjusted life days.
Note: When the efficacy of a reduced dose in homozygotes falls below 98.4%, the no-test strategy is below the willingness-to-pay threshold of $100,000/QALY.
The results were sensitive to the efficacy of a reduced dose and were somewhat sensitive to the cost of the genetic test, but not to the risk of an adverse event. In 1-way sensitivity analyses, the estimated efficacy of a reduced dose of irinotecan administered to homozygous carriers after testing must be at least 98.4% of full-dose efficacy for testing to remain the preferred strategy at a WTP threshold of $100,000 per QALY (Table 2), which means that efficacy in homozygotes must hold nearly constant after a dose reduction. The cost of the genetic test must be <$395 (vs the current Medicare cost of $103) for testing to remain the preferred strategy at a WTP threshold of $100,000 per QALY; at < $375, testing both cost less and increased QALYs. Results were not sensitive to the range of values tested for adverse events (incidence of and mortality from severe neutropenia); for example, if the probability of severe neutropenia after dose reduction in homozygotes was 10.5% (25% reduction) versus 3.7%, then the threshold for treatment efficacy of a reduced dose was 98.3%. Results were also not sensitive to life expectancy, genetic test performance (ie, sensitivity and specificity), or quality-of-life weights (data not shown). Adopting a societal perspective by incorporating patient time costs and a lower test cost did not change the optimal strategy or trends in the sensitivity analyses.
The threshold analysis indicated that, for genetic testing to remain the dominant strategy, the average cost for the 13 patients who would avoid death because of a severe neutropenic event cannot exceed $618,955 per patient during their remaining life expectancy after completion of the FOLFIRI regimen. In probabilistic sensitivity analysis, testing dominated usual care in only 9.1% of simulations. Usual care was preferred in 72% of simulations and in an additional 7.5% when usual care resulted in an increase in QALYs, but it was above the WTP threshold of $100,000 per QALY (Fig. 2).
Value of Studies to Reduce Uncertainties
The partial EVPI for the efficacy variable was $22 million over the next 5 years. In other words, a research investment of up to $22 million to evaluate the efficacy of irinotecan dose reduction in homozygotes would be justified based on a WTP threshold of $100,000 per QALY if irinotecan remained an accepted and widely used treatment alternative for patients with metastatic colorectal cancer in the United States for the next 5 years. Because Medicare primarily provides health insurance coverage for patients aged ≥65 years (including approximately 18,000 patients diagnosed with metastatic colorectal cancer in this age group each year), the partial EVPI just from the Medicare perspective was $13.8 million. The total EVPI was $21.4 million, or $13.4 million from the Medicare perspective, which was slightly lower than the partial EVPI. The higher partial EVPI is probably because of the large and important effect of the efficacy variable in the model, which is not reflected fully in the total EVPI, because uncertainty in other model parameters is correlated with uncertainty in the efficacy parameter and works in the opposite direction.
Pharmacogenetic testing has the potential to fundamentally alter risk-benefit tradeoffs associated with pharmacotherapy, introducing “personalized medicine” concepts into clinical practice. Cost-effectiveness analyses of pharmacogenetic tests can provide important insights into the relative economic value of this new management paradigm, but few of these cost-effectiveness studies have been published.13, 44 To our knowledge, this is the first cost-effectiveness study of UGT1A1*28 testing to guide chemotherapy dosing of irinotecan. Testing itself would cost up to $3 million per year, given the test cost and eligible population. We observed that incorporating UGT1A1*28 testing into the clinical management of patients who have metastatic colorectal cancer treated with irinotecan may result in lower overall medical costs and higher quality-adjusted life expectancy, but our findings were sensitive to several uncertainties in the model's parameter estimates, as indicated by the sensitivity analyses. In particular, maintenance of the clinical efficacy of irinotecan after dose reduction in UGT1A1*28 homozygotes was a key determinant of cost-effectiveness results. That is, if treatment efficacy is not fully maintained after the FDA-recommended dose reduction of irinotecan, then testing will not be a cost-effective alternative. The small reduction in treatment-related risks (severe neutropenia and/or death) gained by testing is outweighed by the risk of shortened survival because of under-treatment. However, limited data in this area do suggest that homozygotes experience slower glucuronidation and a greater plasma concentration of SN-386 and that response rates are not different between homozygotes, heterozygotes, or wild types, even with differential dose reductions across groups.17
Our analysis indicates that the federal government should be willing to invest in further research to reduce the uncertainty associated with UGT1A1 genotype testing and irinotecan dose reduction. Specifically, at a WTP threshold of $100,000 per QALY and with a 5-year time horizon for the genetic testing technology, the federal government should consider investing at least $13.8 million specifically to evaluate the efficacy of a reduced irinotecan dose in UGTA1*28 homozygous patients with colorectal cancer aged ≥65 years. By comparison, the estimated expected cost to conduct an average phase 3 clinical trial is $27.1 million (in 2000 US dollars),45 although some analysts consider this to be a 2-4-fold overestimate.46, 47 The cost to conduct a clinical trial in homozygotes may differ substantially from this amount, however, because irinotecan is not a novel therapeutic entity and already has been proven effective in patients with metastatic colorectal cancer. Alternatively, an observational clinical study could be considered that might provide valuable information more quickly.
This analysis is subject to limitations. First, the probabilities of severe neutropenia and death from neutropenia in patients with colorectal cancer were difficult to estimate, because data were drawn from randomized controlled trials using different regimens, patients were monitored more closely in clinical trials than in usual care (with dose reductions for patients with adverse events), and the number of events or deaths in any 1 trial was small or nil. Because we used estimates drawn from clinical trials, our results may underestimate the value of genetic testing to avoid these events. Conversely, dose reductions reported in some studies were greater than the dose reduction we modeled based on the FDA label. Results were consistent, however, when we varied the amount of dose reduction benefit from avoiding adverse events. Second, there were only 2 published clinical trials that reported results in a relatively small number of genotyped patients. So, although genetic test sensitivity and specificity are fairly well established, estimates of clinical sensitivity and specificity—or the probabilities that a positive test predicts an adverse event or death—have a wide range. Although we accounted for the finding that a patient with severe neutropenia could be hospitalized versus treated on an outpatient basis, this is likely to depend on local care patterns and patient ability to self-manage at home. Varying this parameter, however, did not affect the model results.
We also did not consider alternative dosing approaches. For instance, the results from a study by Toffoli and colleagues8 indicated that successfully administering full-dose irinotecan in UGT1A1*28 homozygotes can produce superior response rates compared with nonhomozygotes (odds ratio, 0.19; 95% confidence interval, 0.04-0.89), albeit with higher rates of hematologic and nonhematologic toxicities (odds ratio, 4.9; 95% confidence interval, 1.36-17.9). This suggests that the magnitude of the recommended dose reduction in homozygotes should be considered carefully, because a potential effectiveness advantage (ie, improved response rate and/or survival) may be compromised significantly. In addition, alternative dosing regimens of irinotecan may have different safety profiles, and the importance of UGT1A1*28 status may be enhanced or reduced depending on pharmacodynamic parameters associated with the dosing schedule. Finally, prophylactic use of agents like pegfilgrastim or filgrastim could help reduce the incidence of neutropenia in homozygotes,48 thereby allowing a full dose of irinotecan to be used in the first cycle but also contributing to the cost of this treatment strategy.
Our analysis focused on the use of irinotecan for the treatment of metastatic colorectal cancer. The broad therapeutic application of irinotecan (eg, tumors of the upper gastrointestinal tract and central nervous system49-51) also should be taken into account when considering both the clinical impact of dosing recommendations and the potential economic value of pharmacogenetic testing. The proportion of patients with metastatic colorectal cancer who eventually will receive irinotecan is not known; however, if <29,214 cancer patients were expected to receive irinotecan in a year, then the value of future research would decrease.
Cost-effectiveness analysis can help policy makers evaluate the clinical utility of new medical technologies like pharmacogenetic tests, make decisions about reimbursement, and identify priorities and investment levels for further research. In the case of UGT1A1*28 testing, we observed that routine testing may improve clinical outcomes and quality of life slightly but is only cost saving if clinical efficacy (survival) is maintained after an irinotecan dose reduction in homozygotes. Further studies to evaluate the impact of irinotecan dose reduction on clinical efficacy would be worth an investment of up to $13.8 million dollars based on the impact of such testing on costs, life expectancy, and quality of life for Medicare beneficiaries alone and even more if all use of irinotecan were considered.
We thank Elisabeth Fenwick, PhD (University of Glasgow, United Kingdom) for methodological expertise and assistance; Debra Leonard, MD, PhD and Hanna Rennert, PhD (Weill Cornell Medical College, New York, NY) for sharing insights about pharmacy testing and billing; and Scott Cantor, PhD (University of Texas M. D. Anderson Cancer Center, Houston, Tex) and participants of the “Using Modeling to Inform Public Health” seminar at Yale University School of Medicine.
Conflict of Interest Disclosures
Supported by each author's institution and in part by the American Cancer Society (Grant Numbers MRSGT-4-002-01-CPHPS to HTG and MRSG-07-232-01-CPHPS to MJH) and the Center for Education and Research on Therapeutics (Agency for Healthcare Research and Quality [Grant Number 1U18-HS016075 (HTG)]).