Cost–Utility Analysis of Palonosetron-Based Therapy in Preventing Emesis Among Breast Cancer Patients

We modeled emesis-related outcomes and direct medical costs (from a third-party payer perspective within the context of the U.S. health-care system) over a total of four cycles of chemotherapy as patients receiving AC-based regimens usually undergo at least four cycles of AC.¹⁰ We performed all analyses using TreeAge Pro 2009 Suite (Decision Analysis; TreeAge Software, Williamstown, MA). The study was submitted to our institutional review board and was determined to be exempt from review.

Probability Data

Two-drug prophylactic regimens

We estimated the effectiveness of the 5-HT₃ antagonists based on secondary analysis of the raw data from the randomized clinical trial (RCT) directly comparing onda and palo when used alone for prevention of emesis associated with MEC, including 90 breast cancer patients from the palo 0.25-mg arm and 82 from the onda 32-mg arm who received AC-based chemotherapy (Table 1).⁵ Effectiveness estimates for palo 0.25 mg were augmented by data on 117 breast cancer patients on AC-based chemotherapy participating in a multicenter RCT comparing palo with dolasetron (Table 1).⁴ We assumed that dex adds the same relative benefit to either first- or second-generation 5-HT₃ antagonists and obtained the expected additional benefit of dex in preventing acute emesis based on the results of an RCT comparing a single-dose of granisetron in combination with dex vs granisetron given alone to patients undergoing MEC (Table 2).¹¹ Since in the aforementioned study dex was only given on day 1 of chemotherapy, the estimated additional benefit of adding dex to a 5-HT₃ inhibitor on the delayed period was obtained from another RCT; this study, conducted by the Italian Group for Antiemetic Research, compared dex alone, dex plus onda, or placebo on days 2−5 of MEC.¹²

Three-drug prophylactic regimens

We estimated the rate of acute emesis for the three-drug regimens based on data from published studies in which either onda or palo was given in combination with dex and aprepitant on day 1 of MEC (Table 2).^{[5], [7] and [13]} Because aprepitant was either used in combination with dexamethasone or not used on days 2−3 in the trials of palo-based three-drug therapy, we estimated the benefit of adding aprepitant alone to palo on days 2−3 by assuming that the added benefit in the delayed period would be the same as the benefit added to onda. Specifically, we obtained information on the relative risk of delayed emesis control when aprepitant is added on days 2−3 from a large clinical trial of aprepitant combined with onda and dex in breast cancer patients receiving either A or AC chemotherapy (Table 2).⁷

Effectiveness of antiemetics over multiple cycles of chemotherapy

The estimates of changes in the probability of emesis control over multiple cycles of chemotherapy were obtained from a RCT conducted by Herrstedt et al¹⁴ of ondansetron-based two- and three-drug regimens for prevention of chemotherapy-induced nausea and vomiting among breast cancer patients undergoing multiple cycles of AC-based chemotherapy. We assumed that changes in emesis control over four cycles of AC for the palo-based two- and three-drug regimens were similar to the observed changes for the onda-based two- and three-drug strategies, respectively.¹⁴

Resource Utilization and Cost Data

The cost of antiemetic prophylaxis was based on the 2008 Medicare Part B reimbursement rates for pharmaceuticals, which reflects the price of ondansetron following its recent patent expiration (Table 3).¹⁵ The costs of prophylaxis failures were estimated as follows. In the majority of prophylaxis failures, the only cost is the cost of rescue medication. In such cases, we obtained costs by multiplying the individual doses used for rescue treatment of breast cancer patients on AC participating in the clinical trials comparing palo 0.25 mg with single doses of onda or dolasetron by their unit costs based on the 2008 Medicare Part B reimbursement rates.^{[5] and [15]} For the few patients who are seen in the office for uncontrolled emesis, we obtained estimates of the risk of such emesis-related office visits based on the MarketScan Health Productivity Management (HPM) database from Thomson Reuters on 707 breast cancer patients who received their first cycle of AC-based chemotherapy between 1997 and 2002 (Table 2) and its costs from the 2008 Medicare Physician Fee Schedule Reimbursement for a level III office visit (CPT 99213).^{[16] and [17]}

Finally, although hospitalization for emesis is extremely rare in this population, when it occurs, it is quite expensive. For completeness, we obtained estimates of the risk of emesis-related hospitalization from the same population of breast cancer patients from whom we obtained the estimate for the risk of emesis-related office visit, whereas hospital costs were obtained from Healthcare Cost and Utilization Project (HCUP) data on 2,342 breast cancer patients who were hospitalized with a primary or admitting diagnosis of vomiting or dehydration from 1997 to 2003 ([Table 2] and [Table 3]).^{[16] and [18]}

Of note is that since palo was only introduced into the U.S. market in 2003, we anticipated the observed risk of emesis-related office visit and hospital admission obtained from MarketScan data during the period 1997−2002 reflected the risk associated with prophylaxis with onda. As a result, given that, when compared with onda, palo has also shown superiority in reducing the severity of emetic episodes when they occur, we estimated the differential rate of health-care resource utilization for palo and onda based on Haislip et al's reported differential incidence of extreme events associated with chemotherapy-induced nausea and vomiting (CINV) experienced by community-based breast cancer patients who received either palo or onda for emesis prophylaxis following the first cycle of chemotherapy (Table 2).^{[5] and [19]}

Utility Data

We obtained the utility weights for acute and delayed emesis from a published study of preferences elicited from ovarian cancer patients undergoing chemotherapy using a modified visual analog scale (VAS) (Table 2).²⁰ We equally applied these emesis-related utility weights to the initial 5-day period of chemotherapy (the standard duration of follow-up in clinical trials of prophylactic antiemetics) in all six prophylactic strategies of the decision tree. Furthermore, because the risk of CINV after 5 days of chemotherapy is usually so negligible as to be unmeasured in clinical trials of antiemetics, we assumed the utility weights for the remaining 16 days of each of the chemotherapy cycles to be the same as the weight associated with complete emesis control (ie, 0.92). We subsequently converted the resulting estimates of quality-adjusted life days into quality-adjusted life years (QALY).

Analysis

We used a stepwise method to calculate the incremental cost–effectiveness ratios of the different prophylactic therapy strategies, with the generic onda-based two-drug therapy (ie, the lowest cost strategy) as the base comparator (also known as the “anchor”).²¹ We adopted the benchmark range of U.S. $50,000−$100,000 per QALY, which has been commonly cited for oncology-related interventions as the threshold for acceptable cost–effectiveness, and examined the robustness of the results by performing one-way sensitivity analyses of plausible ranges for the model's key parameters based on the data sources used as well as probabilistic sensitivity analysis using Monte Carlo simulation.^{[21] and [22]}

Results

The overall rate of emesis control (on days 1−5) among breast cancer patients following a cycle of AC-based chemotherapy was estimated to be 63% (range 46%−79%) for the onda-based two-drug therapy, 74% (range 66%−85%) for the palo-based two-drug therapy, 76% (range 75%−82%) for the onda-based three-drug therapy, and 92% (range 81%−96%) for the palo-based three-drug therapy. Based on these estimates, relative to the onda-based two-drug therapy, the incremental cost–effectiveness ratios (ICERs) for the palo-based regimens were $115,490/QALY for the two-drug strategy, $199,375/QALY for the two-drug regimen plus aprepitant after emesis, and $200,526/QALY for the three-drug strategy (Table 4). The onda-based two-drug combination plus aprepitant after the onset of emesis was eliminated through extended dominance as it has a greater ICER than the next more effective therapy, the palo-based two-drug treatment strategy (Table 4). The onda-based three-drug strategy was dominated by the palo-based two-drug combination plus aprepitant after the onset of emesis as the former strategy is both less effective and more expensive than the latter (Table 4).

In sensitivity analyses using the commonly accepted cost–effectiveness benchmark range of $50,000−$100,000/QALY, the results were sensitive to changes in the overall emesis control rates for the onda-based two-drug strategy. If the probability of overall emesis control for the onda-based two-drug strategy was as low as its estimated lower bound (46%), the ICER for the palo-based two-drug treatment alternative would drop to $53,892/QALY. The results were also sensitive to changes in the effectiveness for the palo-based two-drug regimen: When its overall control rate was as high as its estimated upper bound (86%), its ICER would be $71,472. In contrast, the results were not sensitive to variations in the probability of overall emesis control for the three-drug strategies, nor were they sensitive to changes in the relative probability of emesis control in subsequent cycles of AC for either the two- or three-drug strategies.

If the probability of emesis-related hospitalization was as high as the upper limit of its 95% confidence interval (CI), the ICER for the palo-based two-drug regimen would be $97,301/QALY. However, changes in the cost of an emesis-related admission (95% CI $3,921−$6,112) did not significantly alter the results, nor did variations in office visit and rescue medicine utilization and their associated costs. The results were also not sensitive to variations in the values for the utility weights throughout their 95% CIs. We performed a threshold analysis to explore the price per dose of palo that would result in an acceptable cost–effectiveness ratio under the $100,000/QALY benchmark and found that the ICER for the palo-based two-drug treatment alternative would only fall to a $100,000/QALY threshold when the cost of palo is decreased by 11%.

Figure 2 shows the cost–effectiveness acceptability curves for each strategy, with the onda-based two-drug therapy as the base comparator. These curves show the proportion of the 100,000 simulations in which the comparing antiemetic regimen was considered more cost-effective than the base comparator at different thresholds. Using the benchmark of U.S. $100,000/QALY, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective in 39% and 26% of the simulations with the onda-based standard therapy as the baseline, respectively, whereas the palo-based and onda-based three-drug strategies and the onda-based two-drug regimen with aprepitant after emesis were cost-effective in fewer than 10% of the simulations. Of note is that the slope of the acceptability curves for the palo-based two-drug strategies are steep when willingness to pay exceeds $50,000/QALY, indicating that the greater the threshold, the greater the increase in the level of confidence that these strategies could be cost-effective. For example, the probability that the palo-based two-drug strategy is more cost-effective than the onda-based two-drug strategy rises to 51% at a threshold value of $125,000/QALY and exceeds 60% at $150,000/QALY.

Figure 3 presents the scatterplot of the results of the probabilistic sensitivity analysis for the palo-based two-drug strategy. Nearly 96% of the simulations fell within the first quadrant of the chart (ie, on the upper right quadrant), which represents the scenario where the palo-based two-drug therapy is more costly but also more effective than the onda-based standard therapy. However, only 39% of the simulations fell below the $100,000/QALY dashed threshold line, which represents the scenario where the palo-based two-drug strategy is more cost-effective than the onda-based standard therapy at the $100,000/QALY benchmark.

Discussion

Our estimates of emesis-related costs and outcomes following four cycles of AC-based chemotherapy in women with breast cancer indicate that at current antiemetic prices and utilities placed on emesis, the additional costs of palo and aprepitant are not warranted at the $100,000/QALY threshold. In probabilistic sensitivity analysis, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective at the $100,000/QALY threshold in only 39% and 26% of the simulations, respectively. The model was sensitive to changes in the values of antiemetic effectiveness for the two-drug regimens and the risk of emesis-related hospitalization.

In threshold analysis, the two-drug palo-based regimen was cost-effective at the $100,000/QALY benchmark when the cost of palo is decreased by 11%. Because the use of the $100,000/QALY threshold is uncommon in clinical practice, the cost-effectiveness of the palo-based two-drug strategy (estimated at $115,490/QALY in our study) compares favorably with other commonly used supportive care measures for women with breast cancer. Such measures include primary prophylaxis with granulocyte colony-stimulating factor in women undergoing chemotherapy with moderate to high myelosuppressive risk (ICER of $116,000/QALY, or $125,948/QALY in 2008 U.S. dollars) and the use of bisphosphonates for the prevention of skeletal complications in breast cancer patients with lytic bone metastases (ICER ranging from $108,200/QALY with chemotherapy as systemic therapy to $305,300 in conjunction with hormonal systemic therapy, or $166,381/QALY to $469,466/QALY in 2008 U.S. dollars, respectively).^{[23] and [24]} Both interventions are considered recommended standards of supportive care for patients with breast cancer and are widely used in breast oncology practices.^{[25] and [26]}

Decision-analytic models, such as the Markov model presented in our study, aim to reflect the reality of clinical practice in a simplified way. Therefore, modelers often need to make decisions regarding the study time frame and model parameters based on the best use of available data. In our study, we obtained estimates for the probability of chemotherapy-induced emesis from studies in which the standard duration of follow-up is 5 days. By so doing, we may have underestimated the cost-effectiveness for the palo-based and aprepitant-based regimens. Although the risk of CINV after 5 days of chemotherapy is usually negligible, anticipation of vomiting may affect a patient's quality of life throughout the cycle of chemotherapy.

In addition, our estimates of costs, which were mostly obtained from Medicare, may differ from those of other third-party payers. However, Medicare is among the largest payers for breast cancer care as 42% of the women diagnosed with cancer in the United States are older than 64 years, and many private organizations set their own reimbursement rates based on the Medicare schedule. Therefore, we believe that Medicare reimbursement data provide a suitable estimate for emesis-related medical costs for all breast cancer patients in the United States.^{[27] and [28]}

The present results should solely be interpreted in light of the cost–effectiveness benchmark of $50,000−$100,000/QALY, which has been frequently used in the context of the U.S. health-care system.^{[22] and [29]} Such a benchmark, however, is a historic, precedent-based threshold set by the cost of caring for patients on dialysis, which was estimated at $50,000/QALY in 1982 ($74,000−$95,000 in 1997 U.S. dollars).^{[30] and [31]} Given the arbitrariness of such a threshold, it has been suggested that the current willingness to pay for medical interventions in the United States probably exceeds $100,000/QALY, with values as high as $300,000/QALY being cited in some oncology publications.^{[22], [29], [31], [32], [33] and [34]} In support of that argument is the public and policy makers' strong negative reaction to the National Institutes of Health Consensus Panel not recommending mammography screening for women aged 40−49 years, a procedure reported to provide an ICER of $105,000 per life-year gained.^{[35] and [36]} As a result, if willingness to pay goes beyond $100,000/QALY, the alternative of adding aprepitant to palo plus dex may also be deemed attractive as the slope of its acceptability curve becomes substantially steep when the willingness to pay for a QALY exceeds $125,000 (Figure 2), suggesting that its marginal gain may exceed its marginal costs at higher thresholds.

In addition, it is worth noting that the present analysis has been conducted from the perspective of a third-party payer within the context of the U.S. health-care system. The large difference in the acquisition cost of palo-based and onda-based therapy observed in the United States is mostly driven by the differential stage of product life cycles for palo and onda. Although at the time of this study palo was still under patent protection, generic onda had entered the U.S. market prior to our study. The large price discrepancy between brand and generic drugs explains the difference in drug costs in this U.S.-based analysis. As such, our results may not reflect the situation in countries with a widely different cost structure, in which the acquisition cost of palo may be substantially lower. When that is the case, the cost–effectiveness profile of the palo-based prophylactic therapy may be deemed substantially more favorable than the profile presented here. Similarly, we anticipate finding a more attractive cost–effectiveness profile for the palo-based therapies as palo reaches the end of its product life cycle in the U.S. market.³⁷ Also of note is that the cost–effectiveness of the palo-based therapy may greatly differ when different perspectives (other than the third-party payer's perspective) are adopted.

Our study, however, has several limitations. First, the utility scores used in our model were derived with a VAS instrument, which does not incorporate patients' preferences under uncertainty. Nevertheless, the VAS approach has been shown to provide utility scores for nausea and vomiting with more variability than scores derived using other methods such as the Standard Gamble (personal communication, Grunberg SM et al, CALGB study 309801). Notwithstanding that, it remains unclear which method gives utility scores for transient health states, such as CINV, with the greatest validity.

Also of note is that due to a lack of information on emesis-related utilities among breast cancer patients in the literature, we used utilities elicited from patients with ovarian cancer. To the best of our knowledge, the utilities in Sun et al²⁰ were the only ones available in the literature that were elicited from a homogeneous population of cancer patients (ie, solely patients with ovarian cancer) and were based on a wide range of health states combining the presence and absence of emesis during either the acute or the delayed period. In addition, the participants in the Sun et al study were treated with carboplatin, which, like the regimen used in our model, is classified as moderately emetogenic in established antiemetic guidelines.^{[8], [9] and [38]} It is also important to emphasize that the population in that study, like our study's population, was composed exclusively of women, who are known to be at increased risk for developing CINV.³⁹

Second, in the absence of clinical trial data, we assumed conservatively that dex and aprepitant add the same relative benefit to both onda and palo. This assumption results in an imperfect estimate of cost–effectiveness. As such, we may have overestimated or underestimated the cost–effectiveness of palo as dex and aprepitant may potentially add less value to the intrinsically more active 5-HT₃ antagonist or uniquely complementary mechanisms of action could contribute to even greater activity with the palo-based therapy. However, our study's estimate of the relative effectiveness of the palo-based two-drug prophylactic therapy versus the onda-based two-drug therapy for preventing delayed emesis is consistent with that reported in a recently published clinical trial comparing palo and granisetron when both drugs are combined with dex following chemotherapy with either AC or cisplatin (1.18 vs 1.17, respectively).⁶

Third, our study did not include the outcomes associated with the adverse effects of antiemetics, and by so doing, we may have underestimated the costs associated with antiemetic prophylaxis. However, the incidence and duration of treatment-related adverse events occurring in the two RCTs comparing palo with either onda or dolasetron were mild and similar across treatment cohorts.^{[4] and [5]}

Fourth, we assumed that changes in emesis control in subsequent cycles of AC for the palo-based regimens were the same as for the onda-based therapy. By so doing, we may have underestimated the cost–effectiveness of palo as the superiority of the more active 5-HT₃ antagonist could be maintained in the subsequent cycles of chemotherapy (or even increased, as seen in the aprepitant-based arm of Herrstedt et al's¹⁴ study). As a result, if future prospective trials of palo-based antiemetic prophylaxis confirm its superiority in maintaining antiemetic efficacy over multiple cycles of AC, the cost–effectiveness profiles for the palo-based strategies may be more favorable than the profiles presented herein.

Last, the incremental gains in QALY observed in cost–utility analysis of interventions associated with transitory and non-life-threatening health states, such as the antiemetic regimens analyzed in our study, tend to render small denominators to be used in the incremental cost–effectiveness ratios. The issue of small denominators has led some researchers to question whether the current methodology of cost–effectiveness analysis is appropriate to determine the cost–effectiveness of treatments for terminal or supportive care.³² However, despite this shortcoming, these types of analysis benefit from having a wider scope as they allow comparisons over different types of health interventions across various diseases. In addition, by incorporating patients' utility levels over different health states (instead of merely looking into cost per additional patient controlled), cost–utility analysis makes explicit the impact of the target population's preferences for the different outcomes. Of importance is that both the Panel on Cost–Effectiveness in Health and Medicine and the Institute of Medicine (IOM) Committee on Regulatory Cost–Effectiveness Analysis recommend the use of QALY as the preferred outcome measure for economic evaluation of health-care interventions.

Conclusion

Although our base-case analysis suggests that, from a third-party payer perspective within the context of the U.S. health-care system, the cost–utility of the palo-based two-drug prophylactic therapy for breast cancer patients receiving four cycles of AC-based chemotherapy exceeds the $50,000–$100,000/QALY threshold, it is comparable to other commonly used supportive care interventions for women with breast cancer. In sensitivity analyses, such a strategy was associated with a 39% chance of being cost-effective at the $100,000/QALY threshold, and the model was sensitive to changes in the values of antiemetic effectiveness and of the probability of emesis-related hospitalization. In threshold analysis, the combination of palo and dex was shown to become cost-effective (at the $100,000/QALY benchmark) when the cost of palo is decreased by 11%. As a result, future research incorporating the price structure of all antiemetics following the recent expiration of onda's patent is needed.

Acknowledgments

The authors thank Dr. Sharon H. Giordano for providing valuable clinical input and Dr. Catherine D. Cooksley for providing suggestions for improving this article.

We modeled emesis-related outcomes and direct medical costs (from a third-party payer perspective within the context of the U.S. health-care system) over a total of four cycles of chemotherapy as patients receiving AC-based regimens usually undergo at least four cycles of AC.¹⁰ We performed all analyses using TreeAge Pro 2009 Suite (Decision Analysis; TreeAge Software, Williamstown, MA). The study was submitted to our institutional review board and was determined to be exempt from review.

Probability Data

Two-drug prophylactic regimens

We estimated the effectiveness of the 5-HT₃ antagonists based on secondary analysis of the raw data from the randomized clinical trial (RCT) directly comparing onda and palo when used alone for prevention of emesis associated with MEC, including 90 breast cancer patients from the palo 0.25-mg arm and 82 from the onda 32-mg arm who received AC-based chemotherapy (Table 1).⁵ Effectiveness estimates for palo 0.25 mg were augmented by data on 117 breast cancer patients on AC-based chemotherapy participating in a multicenter RCT comparing palo with dolasetron (Table 1).⁴ We assumed that dex adds the same relative benefit to either first- or second-generation 5-HT₃ antagonists and obtained the expected additional benefit of dex in preventing acute emesis based on the results of an RCT comparing a single-dose of granisetron in combination with dex vs granisetron given alone to patients undergoing MEC (Table 2).¹¹ Since in the aforementioned study dex was only given on day 1 of chemotherapy, the estimated additional benefit of adding dex to a 5-HT₃ inhibitor on the delayed period was obtained from another RCT; this study, conducted by the Italian Group for Antiemetic Research, compared dex alone, dex plus onda, or placebo on days 2−5 of MEC.¹²

Three-drug prophylactic regimens

We estimated the rate of acute emesis for the three-drug regimens based on data from published studies in which either onda or palo was given in combination with dex and aprepitant on day 1 of MEC (Table 2).^{[5], [7] and [13]} Because aprepitant was either used in combination with dexamethasone or not used on days 2−3 in the trials of palo-based three-drug therapy, we estimated the benefit of adding aprepitant alone to palo on days 2−3 by assuming that the added benefit in the delayed period would be the same as the benefit added to onda. Specifically, we obtained information on the relative risk of delayed emesis control when aprepitant is added on days 2−3 from a large clinical trial of aprepitant combined with onda and dex in breast cancer patients receiving either A or AC chemotherapy (Table 2).⁷

Effectiveness of antiemetics over multiple cycles of chemotherapy

The estimates of changes in the probability of emesis control over multiple cycles of chemotherapy were obtained from a RCT conducted by Herrstedt et al¹⁴ of ondansetron-based two- and three-drug regimens for prevention of chemotherapy-induced nausea and vomiting among breast cancer patients undergoing multiple cycles of AC-based chemotherapy. We assumed that changes in emesis control over four cycles of AC for the palo-based two- and three-drug regimens were similar to the observed changes for the onda-based two- and three-drug strategies, respectively.¹⁴

Resource Utilization and Cost Data

The cost of antiemetic prophylaxis was based on the 2008 Medicare Part B reimbursement rates for pharmaceuticals, which reflects the price of ondansetron following its recent patent expiration (Table 3).¹⁵ The costs of prophylaxis failures were estimated as follows. In the majority of prophylaxis failures, the only cost is the cost of rescue medication. In such cases, we obtained costs by multiplying the individual doses used for rescue treatment of breast cancer patients on AC participating in the clinical trials comparing palo 0.25 mg with single doses of onda or dolasetron by their unit costs based on the 2008 Medicare Part B reimbursement rates.^{[5] and [15]} For the few patients who are seen in the office for uncontrolled emesis, we obtained estimates of the risk of such emesis-related office visits based on the MarketScan Health Productivity Management (HPM) database from Thomson Reuters on 707 breast cancer patients who received their first cycle of AC-based chemotherapy between 1997 and 2002 (Table 2) and its costs from the 2008 Medicare Physician Fee Schedule Reimbursement for a level III office visit (CPT 99213).^{[16] and [17]}

Finally, although hospitalization for emesis is extremely rare in this population, when it occurs, it is quite expensive. For completeness, we obtained estimates of the risk of emesis-related hospitalization from the same population of breast cancer patients from whom we obtained the estimate for the risk of emesis-related office visit, whereas hospital costs were obtained from Healthcare Cost and Utilization Project (HCUP) data on 2,342 breast cancer patients who were hospitalized with a primary or admitting diagnosis of vomiting or dehydration from 1997 to 2003 ([Table 2] and [Table 3]).^{[16] and [18]}

Of note is that since palo was only introduced into the U.S. market in 2003, we anticipated the observed risk of emesis-related office visit and hospital admission obtained from MarketScan data during the period 1997−2002 reflected the risk associated with prophylaxis with onda. As a result, given that, when compared with onda, palo has also shown superiority in reducing the severity of emetic episodes when they occur, we estimated the differential rate of health-care resource utilization for palo and onda based on Haislip et al's reported differential incidence of extreme events associated with chemotherapy-induced nausea and vomiting (CINV) experienced by community-based breast cancer patients who received either palo or onda for emesis prophylaxis following the first cycle of chemotherapy (Table 2).^{[5] and [19]}

Utility Data

We obtained the utility weights for acute and delayed emesis from a published study of preferences elicited from ovarian cancer patients undergoing chemotherapy using a modified visual analog scale (VAS) (Table 2).²⁰ We equally applied these emesis-related utility weights to the initial 5-day period of chemotherapy (the standard duration of follow-up in clinical trials of prophylactic antiemetics) in all six prophylactic strategies of the decision tree. Furthermore, because the risk of CINV after 5 days of chemotherapy is usually so negligible as to be unmeasured in clinical trials of antiemetics, we assumed the utility weights for the remaining 16 days of each of the chemotherapy cycles to be the same as the weight associated with complete emesis control (ie, 0.92). We subsequently converted the resulting estimates of quality-adjusted life days into quality-adjusted life years (QALY).

Analysis

We used a stepwise method to calculate the incremental cost–effectiveness ratios of the different prophylactic therapy strategies, with the generic onda-based two-drug therapy (ie, the lowest cost strategy) as the base comparator (also known as the “anchor”).²¹ We adopted the benchmark range of U.S. $50,000−$100,000 per QALY, which has been commonly cited for oncology-related interventions as the threshold for acceptable cost–effectiveness, and examined the robustness of the results by performing one-way sensitivity analyses of plausible ranges for the model's key parameters based on the data sources used as well as probabilistic sensitivity analysis using Monte Carlo simulation.^{[21] and [22]}

Results

The overall rate of emesis control (on days 1−5) among breast cancer patients following a cycle of AC-based chemotherapy was estimated to be 63% (range 46%−79%) for the onda-based two-drug therapy, 74% (range 66%−85%) for the palo-based two-drug therapy, 76% (range 75%−82%) for the onda-based three-drug therapy, and 92% (range 81%−96%) for the palo-based three-drug therapy. Based on these estimates, relative to the onda-based two-drug therapy, the incremental cost–effectiveness ratios (ICERs) for the palo-based regimens were $115,490/QALY for the two-drug strategy, $199,375/QALY for the two-drug regimen plus aprepitant after emesis, and $200,526/QALY for the three-drug strategy (Table 4). The onda-based two-drug combination plus aprepitant after the onset of emesis was eliminated through extended dominance as it has a greater ICER than the next more effective therapy, the palo-based two-drug treatment strategy (Table 4). The onda-based three-drug strategy was dominated by the palo-based two-drug combination plus aprepitant after the onset of emesis as the former strategy is both less effective and more expensive than the latter (Table 4).

In sensitivity analyses using the commonly accepted cost–effectiveness benchmark range of $50,000−$100,000/QALY, the results were sensitive to changes in the overall emesis control rates for the onda-based two-drug strategy. If the probability of overall emesis control for the onda-based two-drug strategy was as low as its estimated lower bound (46%), the ICER for the palo-based two-drug treatment alternative would drop to $53,892/QALY. The results were also sensitive to changes in the effectiveness for the palo-based two-drug regimen: When its overall control rate was as high as its estimated upper bound (86%), its ICER would be $71,472. In contrast, the results were not sensitive to variations in the probability of overall emesis control for the three-drug strategies, nor were they sensitive to changes in the relative probability of emesis control in subsequent cycles of AC for either the two- or three-drug strategies.

If the probability of emesis-related hospitalization was as high as the upper limit of its 95% confidence interval (CI), the ICER for the palo-based two-drug regimen would be $97,301/QALY. However, changes in the cost of an emesis-related admission (95% CI $3,921−$6,112) did not significantly alter the results, nor did variations in office visit and rescue medicine utilization and their associated costs. The results were also not sensitive to variations in the values for the utility weights throughout their 95% CIs. We performed a threshold analysis to explore the price per dose of palo that would result in an acceptable cost–effectiveness ratio under the $100,000/QALY benchmark and found that the ICER for the palo-based two-drug treatment alternative would only fall to a $100,000/QALY threshold when the cost of palo is decreased by 11%.

Figure 2 shows the cost–effectiveness acceptability curves for each strategy, with the onda-based two-drug therapy as the base comparator. These curves show the proportion of the 100,000 simulations in which the comparing antiemetic regimen was considered more cost-effective than the base comparator at different thresholds. Using the benchmark of U.S. $100,000/QALY, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective in 39% and 26% of the simulations with the onda-based standard therapy as the baseline, respectively, whereas the palo-based and onda-based three-drug strategies and the onda-based two-drug regimen with aprepitant after emesis were cost-effective in fewer than 10% of the simulations. Of note is that the slope of the acceptability curves for the palo-based two-drug strategies are steep when willingness to pay exceeds $50,000/QALY, indicating that the greater the threshold, the greater the increase in the level of confidence that these strategies could be cost-effective. For example, the probability that the palo-based two-drug strategy is more cost-effective than the onda-based two-drug strategy rises to 51% at a threshold value of $125,000/QALY and exceeds 60% at $150,000/QALY.

Figure 3 presents the scatterplot of the results of the probabilistic sensitivity analysis for the palo-based two-drug strategy. Nearly 96% of the simulations fell within the first quadrant of the chart (ie, on the upper right quadrant), which represents the scenario where the palo-based two-drug therapy is more costly but also more effective than the onda-based standard therapy. However, only 39% of the simulations fell below the $100,000/QALY dashed threshold line, which represents the scenario where the palo-based two-drug strategy is more cost-effective than the onda-based standard therapy at the $100,000/QALY benchmark.

Discussion

Our estimates of emesis-related costs and outcomes following four cycles of AC-based chemotherapy in women with breast cancer indicate that at current antiemetic prices and utilities placed on emesis, the additional costs of palo and aprepitant are not warranted at the $100,000/QALY threshold. In probabilistic sensitivity analysis, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective at the $100,000/QALY threshold in only 39% and 26% of the simulations, respectively. The model was sensitive to changes in the values of antiemetic effectiveness for the two-drug regimens and the risk of emesis-related hospitalization.

In threshold analysis, the two-drug palo-based regimen was cost-effective at the $100,000/QALY benchmark when the cost of palo is decreased by 11%. Because the use of the $100,000/QALY threshold is uncommon in clinical practice, the cost-effectiveness of the palo-based two-drug strategy (estimated at $115,490/QALY in our study) compares favorably with other commonly used supportive care measures for women with breast cancer. Such measures include primary prophylaxis with granulocyte colony-stimulating factor in women undergoing chemotherapy with moderate to high myelosuppressive risk (ICER of $116,000/QALY, or $125,948/QALY in 2008 U.S. dollars) and the use of bisphosphonates for the prevention of skeletal complications in breast cancer patients with lytic bone metastases (ICER ranging from $108,200/QALY with chemotherapy as systemic therapy to $305,300 in conjunction with hormonal systemic therapy, or $166,381/QALY to $469,466/QALY in 2008 U.S. dollars, respectively).^{[23] and [24]} Both interventions are considered recommended standards of supportive care for patients with breast cancer and are widely used in breast oncology practices.^{[25] and [26]}

Decision-analytic models, such as the Markov model presented in our study, aim to reflect the reality of clinical practice in a simplified way. Therefore, modelers often need to make decisions regarding the study time frame and model parameters based on the best use of available data. In our study, we obtained estimates for the probability of chemotherapy-induced emesis from studies in which the standard duration of follow-up is 5 days. By so doing, we may have underestimated the cost-effectiveness for the palo-based and aprepitant-based regimens. Although the risk of CINV after 5 days of chemotherapy is usually negligible, anticipation of vomiting may affect a patient's quality of life throughout the cycle of chemotherapy.

In addition, our estimates of costs, which were mostly obtained from Medicare, may differ from those of other third-party payers. However, Medicare is among the largest payers for breast cancer care as 42% of the women diagnosed with cancer in the United States are older than 64 years, and many private organizations set their own reimbursement rates based on the Medicare schedule. Therefore, we believe that Medicare reimbursement data provide a suitable estimate for emesis-related medical costs for all breast cancer patients in the United States.^{[27] and [28]}

The present results should solely be interpreted in light of the cost–effectiveness benchmark of $50,000−$100,000/QALY, which has been frequently used in the context of the U.S. health-care system.^{[22] and [29]} Such a benchmark, however, is a historic, precedent-based threshold set by the cost of caring for patients on dialysis, which was estimated at $50,000/QALY in 1982 ($74,000−$95,000 in 1997 U.S. dollars).^{[30] and [31]} Given the arbitrariness of such a threshold, it has been suggested that the current willingness to pay for medical interventions in the United States probably exceeds $100,000/QALY, with values as high as $300,000/QALY being cited in some oncology publications.^{[22], [29], [31], [32], [33] and [34]} In support of that argument is the public and policy makers' strong negative reaction to the National Institutes of Health Consensus Panel not recommending mammography screening for women aged 40−49 years, a procedure reported to provide an ICER of $105,000 per life-year gained.^{[35] and [36]} As a result, if willingness to pay goes beyond $100,000/QALY, the alternative of adding aprepitant to palo plus dex may also be deemed attractive as the slope of its acceptability curve becomes substantially steep when the willingness to pay for a QALY exceeds $125,000 (Figure 2), suggesting that its marginal gain may exceed its marginal costs at higher thresholds.

In addition, it is worth noting that the present analysis has been conducted from the perspective of a third-party payer within the context of the U.S. health-care system. The large difference in the acquisition cost of palo-based and onda-based therapy observed in the United States is mostly driven by the differential stage of product life cycles for palo and onda. Although at the time of this study palo was still under patent protection, generic onda had entered the U.S. market prior to our study. The large price discrepancy between brand and generic drugs explains the difference in drug costs in this U.S.-based analysis. As such, our results may not reflect the situation in countries with a widely different cost structure, in which the acquisition cost of palo may be substantially lower. When that is the case, the cost–effectiveness profile of the palo-based prophylactic therapy may be deemed substantially more favorable than the profile presented here. Similarly, we anticipate finding a more attractive cost–effectiveness profile for the palo-based therapies as palo reaches the end of its product life cycle in the U.S. market.³⁷ Also of note is that the cost–effectiveness of the palo-based therapy may greatly differ when different perspectives (other than the third-party payer's perspective) are adopted.

Our study, however, has several limitations. First, the utility scores used in our model were derived with a VAS instrument, which does not incorporate patients' preferences under uncertainty. Nevertheless, the VAS approach has been shown to provide utility scores for nausea and vomiting with more variability than scores derived using other methods such as the Standard Gamble (personal communication, Grunberg SM et al, CALGB study 309801). Notwithstanding that, it remains unclear which method gives utility scores for transient health states, such as CINV, with the greatest validity.

Also of note is that due to a lack of information on emesis-related utilities among breast cancer patients in the literature, we used utilities elicited from patients with ovarian cancer. To the best of our knowledge, the utilities in Sun et al²⁰ were the only ones available in the literature that were elicited from a homogeneous population of cancer patients (ie, solely patients with ovarian cancer) and were based on a wide range of health states combining the presence and absence of emesis during either the acute or the delayed period. In addition, the participants in the Sun et al study were treated with carboplatin, which, like the regimen used in our model, is classified as moderately emetogenic in established antiemetic guidelines.^{[8], [9] and [38]} It is also important to emphasize that the population in that study, like our study's population, was composed exclusively of women, who are known to be at increased risk for developing CINV.³⁹

Second, in the absence of clinical trial data, we assumed conservatively that dex and aprepitant add the same relative benefit to both onda and palo. This assumption results in an imperfect estimate of cost–effectiveness. As such, we may have overestimated or underestimated the cost–effectiveness of palo as dex and aprepitant may potentially add less value to the intrinsically more active 5-HT₃ antagonist or uniquely complementary mechanisms of action could contribute to even greater activity with the palo-based therapy. However, our study's estimate of the relative effectiveness of the palo-based two-drug prophylactic therapy versus the onda-based two-drug therapy for preventing delayed emesis is consistent with that reported in a recently published clinical trial comparing palo and granisetron when both drugs are combined with dex following chemotherapy with either AC or cisplatin (1.18 vs 1.17, respectively).⁶

Third, our study did not include the outcomes associated with the adverse effects of antiemetics, and by so doing, we may have underestimated the costs associated with antiemetic prophylaxis. However, the incidence and duration of treatment-related adverse events occurring in the two RCTs comparing palo with either onda or dolasetron were mild and similar across treatment cohorts.^{[4] and [5]}

Fourth, we assumed that changes in emesis control in subsequent cycles of AC for the palo-based regimens were the same as for the onda-based therapy. By so doing, we may have underestimated the cost–effectiveness of palo as the superiority of the more active 5-HT₃ antagonist could be maintained in the subsequent cycles of chemotherapy (or even increased, as seen in the aprepitant-based arm of Herrstedt et al's¹⁴ study). As a result, if future prospective trials of palo-based antiemetic prophylaxis confirm its superiority in maintaining antiemetic efficacy over multiple cycles of AC, the cost–effectiveness profiles for the palo-based strategies may be more favorable than the profiles presented herein.

Last, the incremental gains in QALY observed in cost–utility analysis of interventions associated with transitory and non-life-threatening health states, such as the antiemetic regimens analyzed in our study, tend to render small denominators to be used in the incremental cost–effectiveness ratios. The issue of small denominators has led some researchers to question whether the current methodology of cost–effectiveness analysis is appropriate to determine the cost–effectiveness of treatments for terminal or supportive care.³² However, despite this shortcoming, these types of analysis benefit from having a wider scope as they allow comparisons over different types of health interventions across various diseases. In addition, by incorporating patients' utility levels over different health states (instead of merely looking into cost per additional patient controlled), cost–utility analysis makes explicit the impact of the target population's preferences for the different outcomes. Of importance is that both the Panel on Cost–Effectiveness in Health and Medicine and the Institute of Medicine (IOM) Committee on Regulatory Cost–Effectiveness Analysis recommend the use of QALY as the preferred outcome measure for economic evaluation of health-care interventions.

Conclusion

Although our base-case analysis suggests that, from a third-party payer perspective within the context of the U.S. health-care system, the cost–utility of the palo-based two-drug prophylactic therapy for breast cancer patients receiving four cycles of AC-based chemotherapy exceeds the $50,000–$100,000/QALY threshold, it is comparable to other commonly used supportive care interventions for women with breast cancer. In sensitivity analyses, such a strategy was associated with a 39% chance of being cost-effective at the $100,000/QALY threshold, and the model was sensitive to changes in the values of antiemetic effectiveness and of the probability of emesis-related hospitalization. In threshold analysis, the combination of palo and dex was shown to become cost-effective (at the $100,000/QALY benchmark) when the cost of palo is decreased by 11%. As a result, future research incorporating the price structure of all antiemetics following the recent expiration of onda's patent is needed.

Acknowledgments

The authors thank Dr. Sharon H. Giordano for providing valuable clinical input and Dr. Catherine D. Cooksley for providing suggestions for improving this article.

We modeled emesis-related outcomes and direct medical costs (from a third-party payer perspective within the context of the U.S. health-care system) over a total of four cycles of chemotherapy as patients receiving AC-based regimens usually undergo at least four cycles of AC.¹⁰ We performed all analyses using TreeAge Pro 2009 Suite (Decision Analysis; TreeAge Software, Williamstown, MA). The study was submitted to our institutional review board and was determined to be exempt from review.

Probability Data

Two-drug prophylactic regimens

We estimated the effectiveness of the 5-HT₃ antagonists based on secondary analysis of the raw data from the randomized clinical trial (RCT) directly comparing onda and palo when used alone for prevention of emesis associated with MEC, including 90 breast cancer patients from the palo 0.25-mg arm and 82 from the onda 32-mg arm who received AC-based chemotherapy (Table 1).⁵ Effectiveness estimates for palo 0.25 mg were augmented by data on 117 breast cancer patients on AC-based chemotherapy participating in a multicenter RCT comparing palo with dolasetron (Table 1).⁴ We assumed that dex adds the same relative benefit to either first- or second-generation 5-HT₃ antagonists and obtained the expected additional benefit of dex in preventing acute emesis based on the results of an RCT comparing a single-dose of granisetron in combination with dex vs granisetron given alone to patients undergoing MEC (Table 2).¹¹ Since in the aforementioned study dex was only given on day 1 of chemotherapy, the estimated additional benefit of adding dex to a 5-HT₃ inhibitor on the delayed period was obtained from another RCT; this study, conducted by the Italian Group for Antiemetic Research, compared dex alone, dex plus onda, or placebo on days 2−5 of MEC.¹²

Three-drug prophylactic regimens

We estimated the rate of acute emesis for the three-drug regimens based on data from published studies in which either onda or palo was given in combination with dex and aprepitant on day 1 of MEC (Table 2).^{[5], [7] and [13]} Because aprepitant was either used in combination with dexamethasone or not used on days 2−3 in the trials of palo-based three-drug therapy, we estimated the benefit of adding aprepitant alone to palo on days 2−3 by assuming that the added benefit in the delayed period would be the same as the benefit added to onda. Specifically, we obtained information on the relative risk of delayed emesis control when aprepitant is added on days 2−3 from a large clinical trial of aprepitant combined with onda and dex in breast cancer patients receiving either A or AC chemotherapy (Table 2).⁷

Effectiveness of antiemetics over multiple cycles of chemotherapy

The estimates of changes in the probability of emesis control over multiple cycles of chemotherapy were obtained from a RCT conducted by Herrstedt et al¹⁴ of ondansetron-based two- and three-drug regimens for prevention of chemotherapy-induced nausea and vomiting among breast cancer patients undergoing multiple cycles of AC-based chemotherapy. We assumed that changes in emesis control over four cycles of AC for the palo-based two- and three-drug regimens were similar to the observed changes for the onda-based two- and three-drug strategies, respectively.¹⁴

Resource Utilization and Cost Data

The cost of antiemetic prophylaxis was based on the 2008 Medicare Part B reimbursement rates for pharmaceuticals, which reflects the price of ondansetron following its recent patent expiration (Table 3).¹⁵ The costs of prophylaxis failures were estimated as follows. In the majority of prophylaxis failures, the only cost is the cost of rescue medication. In such cases, we obtained costs by multiplying the individual doses used for rescue treatment of breast cancer patients on AC participating in the clinical trials comparing palo 0.25 mg with single doses of onda or dolasetron by their unit costs based on the 2008 Medicare Part B reimbursement rates.^{[5] and [15]} For the few patients who are seen in the office for uncontrolled emesis, we obtained estimates of the risk of such emesis-related office visits based on the MarketScan Health Productivity Management (HPM) database from Thomson Reuters on 707 breast cancer patients who received their first cycle of AC-based chemotherapy between 1997 and 2002 (Table 2) and its costs from the 2008 Medicare Physician Fee Schedule Reimbursement for a level III office visit (CPT 99213).^{[16] and [17]}

Finally, although hospitalization for emesis is extremely rare in this population, when it occurs, it is quite expensive. For completeness, we obtained estimates of the risk of emesis-related hospitalization from the same population of breast cancer patients from whom we obtained the estimate for the risk of emesis-related office visit, whereas hospital costs were obtained from Healthcare Cost and Utilization Project (HCUP) data on 2,342 breast cancer patients who were hospitalized with a primary or admitting diagnosis of vomiting or dehydration from 1997 to 2003 ([Table 2] and [Table 3]).^{[16] and [18]}

Of note is that since palo was only introduced into the U.S. market in 2003, we anticipated the observed risk of emesis-related office visit and hospital admission obtained from MarketScan data during the period 1997−2002 reflected the risk associated with prophylaxis with onda. As a result, given that, when compared with onda, palo has also shown superiority in reducing the severity of emetic episodes when they occur, we estimated the differential rate of health-care resource utilization for palo and onda based on Haislip et al's reported differential incidence of extreme events associated with chemotherapy-induced nausea and vomiting (CINV) experienced by community-based breast cancer patients who received either palo or onda for emesis prophylaxis following the first cycle of chemotherapy (Table 2).^{[5] and [19]}

Utility Data

We obtained the utility weights for acute and delayed emesis from a published study of preferences elicited from ovarian cancer patients undergoing chemotherapy using a modified visual analog scale (VAS) (Table 2).²⁰ We equally applied these emesis-related utility weights to the initial 5-day period of chemotherapy (the standard duration of follow-up in clinical trials of prophylactic antiemetics) in all six prophylactic strategies of the decision tree. Furthermore, because the risk of CINV after 5 days of chemotherapy is usually so negligible as to be unmeasured in clinical trials of antiemetics, we assumed the utility weights for the remaining 16 days of each of the chemotherapy cycles to be the same as the weight associated with complete emesis control (ie, 0.92). We subsequently converted the resulting estimates of quality-adjusted life days into quality-adjusted life years (QALY).

Analysis

We used a stepwise method to calculate the incremental cost–effectiveness ratios of the different prophylactic therapy strategies, with the generic onda-based two-drug therapy (ie, the lowest cost strategy) as the base comparator (also known as the “anchor”).²¹ We adopted the benchmark range of U.S. $50,000−$100,000 per QALY, which has been commonly cited for oncology-related interventions as the threshold for acceptable cost–effectiveness, and examined the robustness of the results by performing one-way sensitivity analyses of plausible ranges for the model's key parameters based on the data sources used as well as probabilistic sensitivity analysis using Monte Carlo simulation.^{[21] and [22]}

Results

The overall rate of emesis control (on days 1−5) among breast cancer patients following a cycle of AC-based chemotherapy was estimated to be 63% (range 46%−79%) for the onda-based two-drug therapy, 74% (range 66%−85%) for the palo-based two-drug therapy, 76% (range 75%−82%) for the onda-based three-drug therapy, and 92% (range 81%−96%) for the palo-based three-drug therapy. Based on these estimates, relative to the onda-based two-drug therapy, the incremental cost–effectiveness ratios (ICERs) for the palo-based regimens were $115,490/QALY for the two-drug strategy, $199,375/QALY for the two-drug regimen plus aprepitant after emesis, and $200,526/QALY for the three-drug strategy (Table 4). The onda-based two-drug combination plus aprepitant after the onset of emesis was eliminated through extended dominance as it has a greater ICER than the next more effective therapy, the palo-based two-drug treatment strategy (Table 4). The onda-based three-drug strategy was dominated by the palo-based two-drug combination plus aprepitant after the onset of emesis as the former strategy is both less effective and more expensive than the latter (Table 4).

In sensitivity analyses using the commonly accepted cost–effectiveness benchmark range of $50,000−$100,000/QALY, the results were sensitive to changes in the overall emesis control rates for the onda-based two-drug strategy. If the probability of overall emesis control for the onda-based two-drug strategy was as low as its estimated lower bound (46%), the ICER for the palo-based two-drug treatment alternative would drop to $53,892/QALY. The results were also sensitive to changes in the effectiveness for the palo-based two-drug regimen: When its overall control rate was as high as its estimated upper bound (86%), its ICER would be $71,472. In contrast, the results were not sensitive to variations in the probability of overall emesis control for the three-drug strategies, nor were they sensitive to changes in the relative probability of emesis control in subsequent cycles of AC for either the two- or three-drug strategies.

If the probability of emesis-related hospitalization was as high as the upper limit of its 95% confidence interval (CI), the ICER for the palo-based two-drug regimen would be $97,301/QALY. However, changes in the cost of an emesis-related admission (95% CI $3,921−$6,112) did not significantly alter the results, nor did variations in office visit and rescue medicine utilization and their associated costs. The results were also not sensitive to variations in the values for the utility weights throughout their 95% CIs. We performed a threshold analysis to explore the price per dose of palo that would result in an acceptable cost–effectiveness ratio under the $100,000/QALY benchmark and found that the ICER for the palo-based two-drug treatment alternative would only fall to a $100,000/QALY threshold when the cost of palo is decreased by 11%.

Figure 2 shows the cost–effectiveness acceptability curves for each strategy, with the onda-based two-drug therapy as the base comparator. These curves show the proportion of the 100,000 simulations in which the comparing antiemetic regimen was considered more cost-effective than the base comparator at different thresholds. Using the benchmark of U.S. $100,000/QALY, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective in 39% and 26% of the simulations with the onda-based standard therapy as the baseline, respectively, whereas the palo-based and onda-based three-drug strategies and the onda-based two-drug regimen with aprepitant after emesis were cost-effective in fewer than 10% of the simulations. Of note is that the slope of the acceptability curves for the palo-based two-drug strategies are steep when willingness to pay exceeds $50,000/QALY, indicating that the greater the threshold, the greater the increase in the level of confidence that these strategies could be cost-effective. For example, the probability that the palo-based two-drug strategy is more cost-effective than the onda-based two-drug strategy rises to 51% at a threshold value of $125,000/QALY and exceeds 60% at $150,000/QALY.

Figure 3 presents the scatterplot of the results of the probabilistic sensitivity analysis for the palo-based two-drug strategy. Nearly 96% of the simulations fell within the first quadrant of the chart (ie, on the upper right quadrant), which represents the scenario where the palo-based two-drug therapy is more costly but also more effective than the onda-based standard therapy. However, only 39% of the simulations fell below the $100,000/QALY dashed threshold line, which represents the scenario where the palo-based two-drug strategy is more cost-effective than the onda-based standard therapy at the $100,000/QALY benchmark.

Discussion

Our estimates of emesis-related costs and outcomes following four cycles of AC-based chemotherapy in women with breast cancer indicate that at current antiemetic prices and utilities placed on emesis, the additional costs of palo and aprepitant are not warranted at the $100,000/QALY threshold. In probabilistic sensitivity analysis, the palo-based two-drug strategy and the two-drug regimen plus aprepitant following the onset of emesis were shown to be cost-effective at the $100,000/QALY threshold in only 39% and 26% of the simulations, respectively. The model was sensitive to changes in the values of antiemetic effectiveness for the two-drug regimens and the risk of emesis-related hospitalization.

In threshold analysis, the two-drug palo-based regimen was cost-effective at the $100,000/QALY benchmark when the cost of palo is decreased by 11%. Because the use of the $100,000/QALY threshold is uncommon in clinical practice, the cost-effectiveness of the palo-based two-drug strategy (estimated at $115,490/QALY in our study) compares favorably with other commonly used supportive care measures for women with breast cancer. Such measures include primary prophylaxis with granulocyte colony-stimulating factor in women undergoing chemotherapy with moderate to high myelosuppressive risk (ICER of $116,000/QALY, or $125,948/QALY in 2008 U.S. dollars) and the use of bisphosphonates for the prevention of skeletal complications in breast cancer patients with lytic bone metastases (ICER ranging from $108,200/QALY with chemotherapy as systemic therapy to $305,300 in conjunction with hormonal systemic therapy, or $166,381/QALY to $469,466/QALY in 2008 U.S. dollars, respectively).^{[23] and [24]} Both interventions are considered recommended standards of supportive care for patients with breast cancer and are widely used in breast oncology practices.^{[25] and [26]}

Decision-analytic models, such as the Markov model presented in our study, aim to reflect the reality of clinical practice in a simplified way. Therefore, modelers often need to make decisions regarding the study time frame and model parameters based on the best use of available data. In our study, we obtained estimates for the probability of chemotherapy-induced emesis from studies in which the standard duration of follow-up is 5 days. By so doing, we may have underestimated the cost-effectiveness for the palo-based and aprepitant-based regimens. Although the risk of CINV after 5 days of chemotherapy is usually negligible, anticipation of vomiting may affect a patient's quality of life throughout the cycle of chemotherapy.

In addition, our estimates of costs, which were mostly obtained from Medicare, may differ from those of other third-party payers. However, Medicare is among the largest payers for breast cancer care as 42% of the women diagnosed with cancer in the United States are older than 64 years, and many private organizations set their own reimbursement rates based on the Medicare schedule. Therefore, we believe that Medicare reimbursement data provide a suitable estimate for emesis-related medical costs for all breast cancer patients in the United States.^{[27] and [28]}

The present results should solely be interpreted in light of the cost–effectiveness benchmark of $50,000−$100,000/QALY, which has been frequently used in the context of the U.S. health-care system.^{[22] and [29]} Such a benchmark, however, is a historic, precedent-based threshold set by the cost of caring for patients on dialysis, which was estimated at $50,000/QALY in 1982 ($74,000−$95,000 in 1997 U.S. dollars).^{[30] and [31]} Given the arbitrariness of such a threshold, it has been suggested that the current willingness to pay for medical interventions in the United States probably exceeds $100,000/QALY, with values as high as $300,000/QALY being cited in some oncology publications.^{[22], [29], [31], [32], [33] and [34]} In support of that argument is the public and policy makers' strong negative reaction to the National Institutes of Health Consensus Panel not recommending mammography screening for women aged 40−49 years, a procedure reported to provide an ICER of $105,000 per life-year gained.^{[35] and [36]} As a result, if willingness to pay goes beyond $100,000/QALY, the alternative of adding aprepitant to palo plus dex may also be deemed attractive as the slope of its acceptability curve becomes substantially steep when the willingness to pay for a QALY exceeds $125,000 (Figure 2), suggesting that its marginal gain may exceed its marginal costs at higher thresholds.

In addition, it is worth noting that the present analysis has been conducted from the perspective of a third-party payer within the context of the U.S. health-care system. The large difference in the acquisition cost of palo-based and onda-based therapy observed in the United States is mostly driven by the differential stage of product life cycles for palo and onda. Although at the time of this study palo was still under patent protection, generic onda had entered the U.S. market prior to our study. The large price discrepancy between brand and generic drugs explains the difference in drug costs in this U.S.-based analysis. As such, our results may not reflect the situation in countries with a widely different cost structure, in which the acquisition cost of palo may be substantially lower. When that is the case, the cost–effectiveness profile of the palo-based prophylactic therapy may be deemed substantially more favorable than the profile presented here. Similarly, we anticipate finding a more attractive cost–effectiveness profile for the palo-based therapies as palo reaches the end of its product life cycle in the U.S. market.³⁷ Also of note is that the cost–effectiveness of the palo-based therapy may greatly differ when different perspectives (other than the third-party payer's perspective) are adopted.

Our study, however, has several limitations. First, the utility scores used in our model were derived with a VAS instrument, which does not incorporate patients' preferences under uncertainty. Nevertheless, the VAS approach has been shown to provide utility scores for nausea and vomiting with more variability than scores derived using other methods such as the Standard Gamble (personal communication, Grunberg SM et al, CALGB study 309801). Notwithstanding that, it remains unclear which method gives utility scores for transient health states, such as CINV, with the greatest validity.

Also of note is that due to a lack of information on emesis-related utilities among breast cancer patients in the literature, we used utilities elicited from patients with ovarian cancer. To the best of our knowledge, the utilities in Sun et al²⁰ were the only ones available in the literature that were elicited from a homogeneous population of cancer patients (ie, solely patients with ovarian cancer) and were based on a wide range of health states combining the presence and absence of emesis during either the acute or the delayed period. In addition, the participants in the Sun et al study were treated with carboplatin, which, like the regimen used in our model, is classified as moderately emetogenic in established antiemetic guidelines.^{[8], [9] and [38]} It is also important to emphasize that the population in that study, like our study's population, was composed exclusively of women, who are known to be at increased risk for developing CINV.³⁹

Second, in the absence of clinical trial data, we assumed conservatively that dex and aprepitant add the same relative benefit to both onda and palo. This assumption results in an imperfect estimate of cost–effectiveness. As such, we may have overestimated or underestimated the cost–effectiveness of palo as dex and aprepitant may potentially add less value to the intrinsically more active 5-HT₃ antagonist or uniquely complementary mechanisms of action could contribute to even greater activity with the palo-based therapy. However, our study's estimate of the relative effectiveness of the palo-based two-drug prophylactic therapy versus the onda-based two-drug therapy for preventing delayed emesis is consistent with that reported in a recently published clinical trial comparing palo and granisetron when both drugs are combined with dex following chemotherapy with either AC or cisplatin (1.18 vs 1.17, respectively).⁶

Third, our study did not include the outcomes associated with the adverse effects of antiemetics, and by so doing, we may have underestimated the costs associated with antiemetic prophylaxis. However, the incidence and duration of treatment-related adverse events occurring in the two RCTs comparing palo with either onda or dolasetron were mild and similar across treatment cohorts.^{[4] and [5]}

Fourth, we assumed that changes in emesis control in subsequent cycles of AC for the palo-based regimens were the same as for the onda-based therapy. By so doing, we may have underestimated the cost–effectiveness of palo as the superiority of the more active 5-HT₃ antagonist could be maintained in the subsequent cycles of chemotherapy (or even increased, as seen in the aprepitant-based arm of Herrstedt et al's¹⁴ study). As a result, if future prospective trials of palo-based antiemetic prophylaxis confirm its superiority in maintaining antiemetic efficacy over multiple cycles of AC, the cost–effectiveness profiles for the palo-based strategies may be more favorable than the profiles presented herein.

Last, the incremental gains in QALY observed in cost–utility analysis of interventions associated with transitory and non-life-threatening health states, such as the antiemetic regimens analyzed in our study, tend to render small denominators to be used in the incremental cost–effectiveness ratios. The issue of small denominators has led some researchers to question whether the current methodology of cost–effectiveness analysis is appropriate to determine the cost–effectiveness of treatments for terminal or supportive care.³² However, despite this shortcoming, these types of analysis benefit from having a wider scope as they allow comparisons over different types of health interventions across various diseases. In addition, by incorporating patients' utility levels over different health states (instead of merely looking into cost per additional patient controlled), cost–utility analysis makes explicit the impact of the target population's preferences for the different outcomes. Of importance is that both the Panel on Cost–Effectiveness in Health and Medicine and the Institute of Medicine (IOM) Committee on Regulatory Cost–Effectiveness Analysis recommend the use of QALY as the preferred outcome measure for economic evaluation of health-care interventions.

Conclusion

Although our base-case analysis suggests that, from a third-party payer perspective within the context of the U.S. health-care system, the cost–utility of the palo-based two-drug prophylactic therapy for breast cancer patients receiving four cycles of AC-based chemotherapy exceeds the $50,000–$100,000/QALY threshold, it is comparable to other commonly used supportive care interventions for women with breast cancer. In sensitivity analyses, such a strategy was associated with a 39% chance of being cost-effective at the $100,000/QALY threshold, and the model was sensitive to changes in the values of antiemetic effectiveness and of the probability of emesis-related hospitalization. In threshold analysis, the combination of palo and dex was shown to become cost-effective (at the $100,000/QALY benchmark) when the cost of palo is decreased by 11%. As a result, future research incorporating the price structure of all antiemetics following the recent expiration of onda's patent is needed.

Acknowledgments

The authors thank Dr. Sharon H. Giordano for providing valuable clinical input and Dr. Catherine D. Cooksley for providing suggestions for improving this article.