In a randomized controlled trial (RCT) of patients with recurrent, platinum-sensitive ovarian cancer, the combination weekly docetaxel and carboplatin was associated a with progression-free survival (PFS) of 13.7 months compared with 8.4 months for sequential, single-agent docetaxel followed by carboplatin. The objective of the current study was to construct a cost-utility model to compare these 2 regimens with the incorporation of prospectively collected quality-of-life (QoL) data.
An RCT of concurrent docetaxel and carboplatin (cDC) versus docetaxel followed by carboplatin (sequential docetaxel and carboplatin [sDC]) was the basis for a Markov decision model, and the primary effectiveness outcome was PFS. Costs were estimated using US dollars based on Medicare reimbursements for chemotherapy regimens, bone marrow support, and management of adverse events. QoL data obtained using the Functional Assessment of Cancer Therapy-General questionnaire were converted to utilities. Costs and incremental cost-effectiveness ratios (ICERs) were reported in US dollars per quality-adjusted life year (QALY). Extensive 1-way sensitivity analyses and a Monte Carlo probabilistic sensitivity analysis were performed.
The least expensive strategy was sDC, which cost an average of $20,381, compared with cDC, which cost an average of $25,122. cDC had an ICER of $25,239 per QALY compared with sDC. cDC remained cost-effective, with an ICER <$50,000 per QALY, over a range of costs and estimates. In Monte Carlo sensitivity analysis using a $50,000 per QALY willingness-to-pay threshold, cDC was either dominant or cost-effective with an ICER <$50,000 per QALY in 83% of simulations.
The standard treatment for recurrent, platinum-sensitive ovarian cancer is combination chemotherapy using platinum doublets, which have demonstrated superior progression-free survival (PFS) and, in some patients, superior overall survival (OS) compared with single-agent carboplatin.1-3 Although platinum-based combinations have yielded improved clinical outcomes compared with single-agent carboplatin, they also have been associated with greater toxicity. This led to a recently completed investigation of planned sequential monotherapy for platinum-sensitive recurrence with the intent of achieving similar outcomes while introducing less toxicity than combination therapy.4
In a recently reported, randomized phase 2 trial, we studied combination carboplatin/docetaxel (cDC) and sequential docetaxel followed by carboplatin (sDC). We observed that cDC resulted in higher PFS at the expense of lower quality of life (QoL) during treatment.4 To better quantify the tradeoff between PFS and QoL in the platinum-sensitive population, we performed a cost-utility analysis, which was informed by our randomized trial of cDC versus sDC.
MATERIALS AND METHODS
A cost-utility analysis was performed using a Markov state transition model, which was constructed using TreeAge Pro software (TreeAge Software Inc., Williamstown, Mass). Two chemotherapy regimens were compared in a decision tree format for the treatment of recurrent, platinum-sensitive ovarian cancer (Fig. 1): 1) cDC with docetaxel at a dose of 30 mg/m2 given intravenously on Days 1 and 8 combined with carboplatin at an area under the curve (AUC) of 6 mg/mL per minute intravenously on Day 1 every 3 weeks; or 2) sDC, with docetaxel at a dose of 30 mg/m2 given intravenously on Days 1 and 8 every 3 weeks followed by carboplatin at an AUC of 6 intravenously every 3 weeks at first progression or after 6 cycles of docetaxel for stable disease or a partial response. These 2 chemotherapy regimens were evaluated as described separately in a prospective, phase 2 randomized controlled trial with regard to PFS and QoL4; costs were not collected during the clinical trial but were obtained as described below using national reimbursement data.5 Investigators were required to obtain approval from their respective institutional review boards; the trial was registered in the clinical trials database of the United States National Institutes of Health (National Clinical Trials no. 00090610). The time horizon for the cost-utility analysis was 24 months, and costs and outcomes were not discounted because of the relatively short time horizon. A third-party payer perspective was used for this analysis. Sensitivity analyses were performed to account for uncertainty in assumptions.
Baseline Model and Estimates
PFS was modeled for the 2 treatment strategies as reported by Alvarez Secord et al.4 Briefly, the median PFS was 13.7 months in the cDC arm and 8.4 months in the sDC arm (hazard ratio for sDC, 1.62; 95% confidence interval, 1.08-2.45). Survival was modeled by using raw PFS data from the trial and was based on measurable disease only (according to Response Evaluation Criteria in Solid Tuomors6). One Markov cycle of the model was made equivalent to 1 chemotherapy cycle (21 days). The time horizon of the model was set at 24 months, at which time >95% of patients had experienced recurrence or had died in each arm. Discounting was not performed because of the short time horizon.
Rates of adverse events for which a significant difference was documented between treatment arms were modeled, and costs were applied. Specifically, the rate of significant (Common Toxicity Criteria grade 2-3) neurotoxicity, the use of erythropoietin, and the use of granulocyte–colony-stimulating factors differed between arms and were included in the model (Table 1). Adverse events that were not different between the arms were not included in the model.
Table 1. Clinical Estimates and Ranges
Abbreviations: cDC, combination docetaxel and carboplatin; GCSF, granulocyte colony-stimulating factor; EPO, erythropoietin; sDC, planned sequential docetaxel followed by carboplatin.
aAll estimates were derived from Alvarez Secord A, Berchuck A, Higgins R, et al. A multicenter, randomized, phase II study evaluating the efficacy and safety of combination docetaxel and carboplatin and sequential therapy with docetaxel then carboplatin in patients with recurrent platinum-sensitive ovarian cancer [abstract]. Gynecol Oncol. 2010;116(suppl):A3).4
Probability of GCSF support
Probability of EPO support
Probability of severe neurologic toxicity
End of study
End of study
Costs associated with treatment and adverse events are listed in Table 2. The cost of one 21-day cycle of treatment included the cost of 1 physician visit, costs of all chemotherapy drugs administered during the cycle, all infusion/treatment charges, and costs of standard pretreatment medications. Routine laboratory work was excluded, because we assumed that this was similar between regimens. The costs of each treatment modality were estimated using national 2010 Medicare reimbursement data (www.cms.gov; accessed July 1, 2010). Costs were applied to the treatment of adverse events whose rates differed significantly between treatment groups in the randomized trial (grade 3-4 neutropenia, grade 2-3 neurotoxicity). For hematologic toxicities, only the costs of administering granulocyte–colony-stimulating factor (pegfilgrastim; 6 mg every 3 weeks) or erythropoietin (500 μg every 3 weeks) to the percentage of women who received them were incorporated. Because the number of cycles over which bone marrow support factors were administered was not collected in the trial, we assumed that anyone who received growth factors incurred the costs of this treatment over 3 cycles of chemotherapy. The cost of grade 3 and 4 neurologic toxicity was estimated as the cost of an outpatient neurology consultation visit plus the cost of pregabalin at 100 mg 3 times daily for 6 months ($680.28; http://www.drugstore.com; accessed July 1, 2010); the costs of medication were estimated as 60% of charges. With regard to the additional costs of cancer recurrence, a simplifying assumption was made. We assumed that patients who progressed, experienced disease recurrence, or died (events on the PFS curve) incurred the cost of 3 additional cycles of outpatient chemotherapy. We obtained the costs of 4 commonly used regimens for recurrent ovarian cancer (liposomal doxorubicin, weekly paclitaxel, gemcitabine, and weekly topotecan) and used the mean cost of treatment with 3 cycles of these regimens as the “cost of next chemotherapy,” with variation over the entire range examined for sensitivity analysis. All costs were inflated to 2010 dollars using the medical care component of the Consumer Price Index (http://www.bls.gov/schedule/archives/cpi_nr.htm; accessed July 1, 2010).
Abbreviations: CPT, Current Procedural Terminology; N/A, not available.
Charges were used to calculate cost using a cost-to-charge ratio of 0.6 when direct Medicare reimbursement data were not available.
Initial recurrence chemotherapy regimens
“Next” chemotherapy regimens
Neurology consultation for peripheral neuropathy (CPT code 99245)
Pregabalin 100 mg 3 times daily for 6 mo
Pegfilgrastim 6 mg subcutaneous injection
Darbepoeitin 500 mcg subcutaneous injection
In the randomized trial, QoL measures were obtained using the Functional Assessment of Cancer Therapy (FACT) scale.7 To convert patient-reported QoL measures to utilities for quality adjustment of trial results, we used a previously described and validated method that incorporates 4 items from the FACT-general version (FACT-G) into a formula to calculate a utility.8 Therefore, a utility was calculated for each patient at 4 time points: at study entry, before Cycle 4, before Cycle 6, and at the end of the study (Table 1). Utilities were applied at study entry and were maintained at that value until the next cycle, when a new utility was available. The end-of-study time differed for each patient, and we assumed that the end-of-study utility occurred at Cycle 12 of treatment. We also assumed that any differences in utilities eventually became insignificant, so that the entered utility was calculated as the average of the end-of-study utilities calculated for each treatment arm after 16 cycles.
One-Way Sensitivity Analysis
We performed 1-way sensitivity analyses on adverse event rates by varying these rates over clinically reasonable ranges (Table 1). We varied the percentage of patients whose neurologic toxicity was severe enough to warrant a neurology physician consultation plus use of a commonly used medication for neuropathic pain (pregabalin). We did not examine inpatient treatment, because no patients in the trial developed grade 4 neuropathy.
We varied cost estimates over a wide range (from 50% to 150% of the value used in the base case) to reflect possible variation in reimbursement rates (Table 2).
The 95% confidence intervals were calculated around each utility using raw QoL data and the number of QoL responses available at each time point (Table 1). Utility values were varied over their 95% confidence intervals at each time point sampled for Monte Carlo probabilistic sensitivity analysis.
Sensitivity Analysis: Monte Carlo Simulation
To simultaneously account for uncertainty in all estimates, including survival, cost, adverse event rates, bone marrow support rates, and QoL, we performed a Monte Carlo probabilistic sensitivity analysis using 10,000 repetitions. During each repetition, sampling was performed from distributions that represented each estimate used to construct the model. Triangular distributions were used for cost, whereas beta distributions were used for probabilities and survival data.
In the base case, the least expensive strategy was sDC, which had an average cost of $20,381, compared with cDC, which had an average cost of $25,122. cDC had an incremental cost-effectiveness ratio (ICER) of $25,239 per quality-adjusted life year (QALY) compared with sDC.
One-Way Sensitivity Analyses
The costs of chemotherapy, bone marrow supportive treatment, and neurotoxicity were varied over the ranges listed in Table 2. When the cost of cDC varies to 150% of the estimate, the ICER for this regimen becomes $61,759 per QALY compared with sDC. When the cost of sDC is varied to 50% of the estimate, cDC remains cost-effective with an ICER of $49,384 per QALY. At the high-end estimate of the cost of sDC (150%), cDC is extremely cost-effective with an ICER of $1093 per QALY compared with sDC. Variation in the cost of single-agent carboplatin over its range does not change the ICER of cDC significantly ($22,062 to $28,415 per QALY). Likewise, variation in the cost of additional chemotherapy regimens over its range results in an ICER between $25,000 and $26,000 per QALY (Table 3).
Table 3. Key One-Way Sensitivity Analyses
Range for Sensitivity Analysis
ICER Range for Combination Versus Sequential Arms, $US per QALY
Cost of 3 cycles of next chemotherapy regimen, $US
Probability of CSF support in combination arm
Probability of EPO support in combination arm
Probability of severe neurologic toxicity in combination arm
We varied the probability of treatment for bone marrow and neurotoxicity over the ranges listed in Table 1. In each case, cDC remained cost-effective compared with sDC, with an ICER <$50,000 per QALY (Table 3).
Monte Carlo Simulation
Monte Carlo probabilistic sensitivity analysis was performed with 10,000 repetitions, and the results are depicted as a cost-effectiveness scatter plot in Figure 2. In the Monte Carlo simulation, cDC was the dominant strategy (more effective and less costly) in 11% of simulations, it was costlier but more effective with an ICER <$50,000 per QALY in 72% of simulations, costlier but more effective with an ICER >$50,000 per QALY in 16% of simulations, and dominated (more costly but less effective) in 1% of simulations. The median ICER in the Monte Carlo simulation was $21,233 per QALY (95% confidence interval, from cDC dominant to $135,105 per QALY).
Decisions regarding the treatment of recurrent cancer are guided optimally by consideration of the expected additional PFS afforded by a treatment as well as by consideration of the adverse effects of each treatment, its impact on QoL, and its costs. In the current study, we performed a cost-effectiveness analysis that accounted for each of these factors and observed that cDC was cost-effective compared with sDC. Our results indicate that the slightly lower QoL, higher costs, and additional need for bone marrow support observed in the cDC group are offset by the superior PFS achieved in this group.
Our results are in agreement with a previous cost-effectiveness analysis of the available regimens for the treatment of platinum-sensitive recurrence, in which we observed that combined carboplatin and paclitaxel potentially was cost-effective compared with single-agent carboplatin.9 Women with platinum-sensitive recurrences often are able to achieve durable responses and lengthy second remissions during which they have good QoL, factors that favor the use of aggressive regimens in this setting.10 In the current study, combination therapy using cDC remained cost-effective even when prospectively collected QoL data indicating lower QoL during treatment were incorporated into the analysis. In contrast, platinum-resistant recurrence is associated with a dismal prognosis. Rocconi et al previously reported that the only potentially cost-effective management strategies for platinum-resistant recurrence were single-agent treatments and supportive care.11
Our study is strengthened by the results from a probabilistic sensitivity analysis, which simultaneously accounts for uncertainty in each of the model's parameters. In Monte Carlo analysis using $50,000 per QALY as the “willingness to pay” threshold, cDC is the clear treatment of choice in 83% of simulations. If the willingness to pay threshold is set at $100,000 per QALY, which is becoming common practice in modern cost-effectiveness studies in oncology,12, 13 then cDC becomes the treatment of choice in 94% of simulations. The 95% confidence intervals for the ICER of cDC are fairly wide, ranging from negative values (which means that cDC is both more effective and less expensive than sDC) to $135,105 per QALY. The wide confidence intervals of the ICER may be related to the small size of the randomized controlled trial (n = 75 in each arm) and the uncertainty surrounding utility estimates, which also were incorporated using their 95% confidence intervals.
There are several limitations to the current study. First is the lack of prospectively collected health care use data from which to calculate costs. We used trial-reported adverse event rates and rates of bone marrow support to approximate cost differences between groups, and this method may result in over estimating or under estimating costs. Second, we were not able to incorporate differences in productivity or caregiver expenses between treatment arms, because these also were not collected prospectively. A third limitation is that, despite the availability of prospectively collected QoL data using the FACT scale, we lacked prospectively collected utilities data. Although we were able to convert the FACT scale to a utility using a previously described method,8 the referenced study used the preference of a set of nonovarian cancer patients to derive utilities from the FACT-G questionnaire. The standard in a cost-effectiveness analysis performed from the standpoint of a third party is to use societal-based rather than patient-based preferences for utility derivation.14 With this limitation acknowledged, we believe that the incorporation of a patient preference-based utility lends credibility to our assessment, because QoL is a critical factor in the treatment of recurrent cancer.
In conclusion, we used prospectively collected results from a randomized controlled trial to construct a cost-utility model to demonstrate that combined docetaxel and carboplatin is cost-effective compared with sequential docetaxel followed by carboplatin for the treatment of recurrent, platinum-sensitive ovarian cancer. Decisions about the treatment of platinum-sensitive recurrence optimally should be made with attention to the potential PFS afforded by each treatment, its treatment costs, adverse event rates, and differences in treatment-related QoL.
This study was funded by Sanofi-Aventis US.
CONFLICT OF INTEREST DISCLOSURES
Laura Havrilesky received research support from BD TriPath Oncology. Robin Pokrzywinska received research support from Sanofi-Aventis. Dennis A. Revicki received research support from Sanofi-Aventis. Robert V. Higgins is a member of the Merck Speakers Bureau. Angeles Alvarez Secord received research support from Sanofi-Aventis, GlaxoSmithKline, Lilly, and Bristol-Myers Squibb.