Cost-utility analysis of 21-gene assay for node-positive early breast cancer.

Background
For women with lymph node (ln)-positive, estrogen receptor-positive, and her2 (human epidermal growth factor receptor 2)-negative breast cancer (bca), current guidelines recommend treatment with both hormonal therapy and chemotherapy. The 21-gene Recurrence Score (rs) assay might be helpful in selecting patients with bca who can be spared chemotherapy when they have 1-3 positive lns and a lower risk of recurrence. In the present study, we performed a cost-utility analysis comparing use of the 21-gene rs assay with current practice from the perspective of a Canadian health care payer.


Methods
A Markov model was developed to determine costs and quality-adjusted life-years (qalys) over a patient's lifetime. Patient outcomes in both study groups were examined based on published clinical trials. Costs were derived primarily from published Canadian sources. Costs and outcomes were discounted at 1.5% annually, and costs are reported in 2016 Canadian dollars. A probabilistic analysis was used, and the model parameters were varied in a sensitivity analysis.


Results
The results indicate that use of the 21-gene rs assay was less costly ($432 less) and more effective (0.22 qalys) than current practice. The probabilistic analysis revealed that 70% of the 10,000 simulated incremental cost-effectiveness ratios were in the southeast quadrant. The results were sensitive to the probability of a low rs and to the probability of receiving chemotherapy in the low-risk rs category and in current practice.


Conclusions
Use of the 21-gene rs assay could be a cost-effective strategy for Ontario patients with estrogen receptor-positive, her2-negative early bca and 1-3 positive lns.


INTRODUCTION
Breast cancer (bca) is the most common cancer and the 2nd leading cause of death from cancer among Canadian women 1 . For women with lymph node (ln)-positive, estrogen receptor (er)-positive bca, current guidelines recommend treatment with both hormonal therapy and chemotherapy 2,3 . In er-positive disease, the addition of adjuvant chemotherapy to tamoxifen reduces the risk of bca recurrence by approximately 30% and the risk of bca mortality by approximately 20%.
The 21-gene Recurrence Score (rs) assay is a genomic test used in determining the likelihood of chemotherapy benefit after surgical treatment for early-stage bca 4 . Patients are defined as having a high (rs ≥ 31), intermediate (rs = 18-30), or low (rs < 18) risk of distant recurrence. The 21-gene rs assay might be helpful in selecting patients with a lower risk of recurrence, for whom chemotherapy could be omitted. It is routinely used to guide adjuvant treatment decisions in patients with node-negative bca.
The clinical utility of the rs assay has also been demonstrated in studies involving ln-positive patients [4][5][6][7] . A growing body of evidence is showing that the rs assay can lead to tailored treatment by sparing patients with 1-3 positive lns and a low rs unnecessary chemotherapy that could be associated with an increased risk of toxicity and a reduction in health-related quality of life 8 . In addition, tailored treatment might save the health care system unnecessary drug and administration costs, and the costs associated with complications 9 .
Studies have shown that the 21-gene rs assay might be a cost-effective option for patients with ln-negative, er-positive bca [10][11][12][13] . A few studies 9,[14][15][16] have also examined the cost-effectiveness of the 21-gene rs assay in patients with ln-positive, er-positive bca and have shown that it might be a cost-effective option.
Currently in Ontario, the most populous province in Canada, the assay is reimbursed by the publicly funded health care system only for ln-negative and micrometastatic lnpositive early-stage bca. In the present study, we performed a cost-utility analysis of the 21-gene rs assay compared with current practice (no rs assay) for patients with lnpositive (1-3 positive lns), er-positive, and her2 (human epidural growth factor receptor 2)-negative early bca from the perspective of the Ontario Ministry of Health and Long-Term Care.

Model Structure
Using the TreeAge Pro 2014 software application (TreeAge Software, Williamstown, MA, U.S.A.), a Markov model with yearly cycle lengths ( Figure 1) was developed to compare the 21-gene rs assay with current practice in terms of lifetime clinical consequences [quality-adjusted life-years (qalys)] and economic consequences (cost). The model simulated a cohort of patients 60 years of age diagnosed with ln-positive (1-3 positive lns), er-positive, her2-negative early bca. The starting age of 60 years was chosen because that is the median age of patients with bca in Canada (clinical expert opinion, 2018).
Patients in the 21-gene rs assay strategy were stratified into 3 categories: low risk (rs < 18), intermediate risk (rs = 18-30), and high risk (rs ≥ 31) of recurrence. Patients in the high-risk category started in one of three chemotherapy health states: chemotherapy with no toxicities; chemotherapy with febrile neutropenia; or chemotherapy with nausea and vomiting. Patients in the chemotherapy health states could move to these health states: disease free without recurrence, distant recurrence, or death. Patients who received chemotherapy could also go on to develop congestive heart failure (chf) or leukemia 1-7 years after treatment. Of patients in the intermediate-risk category, 52% received adjuvant chemotherapy and started in one of the three chemotherapy health states (Table i) 7 . Those who did not receive chemotherapy (48%) started in the disease-free health state. All patients in the low-risk category did not receive adjuvant chemotherapy and started in the disease-free health state, where they received tamoxifen for 5 years, followed by aromatase inhibitors for another 5 years. From the disease-free without recurrence health state, patients could remain in that health state, develop distant recurrence, or die. In the distant recurrence health state, patients stayed in that health state until death. All patients in the model eventually died from bca or background mortality. A lifetime horizon was taken, and future costs and benefits were discounted at 1.5% annually based on guidelines from the Canadian Agency for Drugs and Technologies in Health 28 .

Clinical Input Data
The proportion of patients assigned to each risk category based on the 21-gene rs assay was obtained from a prospective study that evaluated the effect of the assay on treatment decisions for women with ln-positive early bca in Ontario (Table i) 7 . A phase iii randomized controlled trial (swog-8814) found that the rs was a prognostic factor for disease-free survival in patients with ln-positive, er-positive bca 25 . Because recurrence-free survival was not available in swog-8814, we used its reported disease-free survival, which consisted of time to death, recurrence, or new primary cancer. We made an assumption that a large proportion of events were related to recurrence rather than to death or a new primary cancer. To calculate the annual probability of distant recurrence, we fitted a Weibull distribution to Kaplan-Meier curves for each rs category in the tamoxifen-only group (that is, no chemotherapy) and calculated an annual probability of recurrence 29 . To obtain the risk of distant recurrence with chemotherapy for each risk group, we applied the hazard ratios from swog-8814 (Table i) 25 . For patients in the current-practice strategy, we estimated the weighted average risk of recurrence with and without adjuvant chemotherapy in the 3 rs categories.
Once patients developed distant recurrence, it was assumed that the transitional probability for bca death was the same for all patients regardless of rs classification. Based on clinician input, we also assumed that all patients would die after year 20 in that health state (clinical expert opinion, 2018). We accounted for background mortality using Canadian life tables from Statistics Canada 27 .

Chemotherapy Assignment and Toxicities
For patients in the 21-gene rs assay strategy, the proportion receiving chemotherapy was obtained from the post-rs-test treatment recommendations of oncologists in an Ontario study 7 . Patients in all rs categories were given hormonal treatment with tamoxifen for 5 years, followed by an aromatase inhibitor (letrozole) for another 5 years. Patients who received adjuvant chemotherapy started tamoxifen after chemotherapy completion.  Based on clinical opinion, 70% of those in the highrisk rs category received 6 cycles of fec-d: 3 cycles of 5-fluorouracil 500 mg/m 2 , epirubicin 100 mg/m 2 , and cyclophosphamide 500 mg/m 2 , followed by 3 cycles of docetaxel 100 mg/m 2 (clinical expert opinion, 2018). The remaining 30% received 8 cycles of ddac-t: 4 cycles of doxorubicin 60 mg/m 2 and cyclophosphamide 600 mg/m 2 , followed by 4 cycles of paclitaxel 175 mg/m 2 . For patients assigned to the intermediate-risk rs category, 52% received chemotherapy. All patients in that category who received chemotherapy were treated with 4 cycles of tc chemotherapy (docetaxel 75 mg/m 2 and cyclophosphamide 600 mg/m 2 ). All patients treated with adjuvant chemotherapy received 300 μg of filgrastim daily (clinical expert opinion, 2018). As well, patients in all chemotherapy health states received zoledronic acid every 6 months for 3 years. For patients in the current-practice strategy, the proportion receiving chemotherapy and the type of chemotherapy regimen assigned were obtained from the pre-rs-test treatment recommendations of oncologists in an Ontario study (Table i) 7 .
Four different types of chemotherapy toxicities were considered: febrile neutropenia, nausea and vomiting, chf, and leukemia. Febrile neutropenia and nausea and Ranges used in the sensitivity analyses were, for the most part, ±10% of the base case. Where available, additional literature was used for the sensitivity analysis ranges. As a result, some values have wide ranges. RS = Recurrence Score (Oncotype DX: Genomic Health, Redwood City, CA, U.S.A.); CTx = chemotherapy; FEC-D = 5-fluorouracil-epirubicincyclophosphamide, followed by docetaxel; dd AC-T = dose-dense doxorubicin-cyclophosphamide, followed by paclitaxel; TC = docetaxelcyclophosphamide.
vomiting could occur during treatment. The probability of febrile neutropenia while receiving fec-d or tc and granulocyte colony-stimulating factor was obtained from a meta-analysis 17 . The probability of febrile neutropenia while on ddac-t was obtained from an Ontario study 20 . The probability of a patient experiencing nausea and vomiting was obtained from the respective clinical trials for each of the treatments 18,21,22 .
Congestive heart failure and leukemia are chemotherapy toxicities that can occur after treatment. Patients treated with fec-d and ddac-t are at risk for those toxicities because of their anthracycline content. The proportions of patients experiencing leukemia, by treatment type, and experiencing chf were obtained from two studies that used Ontario population data 20 . The proportion of patients experiencing chf was obtained from a study examining the risk of heart failure associated with adjuvant trastuzumab in patients with bca 19 . The probability of experiencing chf that we used was the same for both fec-d and ddac-t. The risk of death from chf for years 1-3 was obtained from a study that provided the Kaplan-Meier probabilities of survival for women with non-ischemic stroke over 4 years 23 . Afterward, a constant probability of death was assumed. The probability of death associated with leukemia from years 1-3 was taken from a study that examined patients with treatment-related secondary acute myeloid leukemia in Sweden 24 . After year 3, a constant probability of death was assumed.

Utilities
The utility values (

Costs
Costs were obtained from various sources (Table ii). To obtain a total cost for filgrastim for each chemotherapy regimen, we assumed that 50% of patients would receive filgrastim type 1 and the other 50% would receive filgrastim type 2 (Sunnybrook Health Sciences Centre database, 2017).
Each chemotherapy regimen was associated with a different quality-based procedure band for systemic therapy, which included the cost of nursing and pharmacy time (Sunnybrook Health Sciences Centre database, 2017). We also included the costs of a physician consultation (fee code A445), medical-specific assessment (fee code A444), and partial assessment (fee code A448) for each chemotherapy regimen 35 . Lastly, we included the cost of a peripherally inserted central catheter and technologist for each chemo therapy regimen ($376) (Sunnybrook Health Sciences Centre database, 2017).
The costs of ongoing care for the patients who were disease-free (that is, recurrence-free) and the costs of treating distant recurrence were obtained from a Canadian study that estimated the health care costs associated with the lifetime management of women diagnosed with bca 33 . The cost averaged for all stages of bca was used. Costs for treating distant recurrence might be underestimated given that costs have increased with the introduction of cyclin-dependent kinase 4/6 inhibitors in the treatment of those patients.
The costs associated with chemotherapy toxicities were obtained from a variety of published sources (Table ii) 36 . We assumed that a patient could experience, at most, 1 episode of febrile neutropenia during chemotherapy (clinical expert opinion, 2018) and that nausea and vomiting happen after each chemotherapy cycle and last for 1 week. The costs of nausea and vomiting therefore differed with the chemotherapy regimen (clinical expert opinion, 2018). The costs associated with leukemia came from a Canadian study that examined the cost of cancer by phase of care. The analysis included the direct health care costs for the initial phase of cancer (that is, the first year after the cancer diagnosis) for patients who survived beyond year 1 39 . Lastly, the costs associated with chf were obtained from a cost-effectiveness study 38 . All costs were inflated to 2016 Canadian dollars using the Consumer Price Index for health care services in Ontario 40 .

Analysis
We estimated an incremental cost-effectiveness ratio (icer) representing the incremental cost for 1 qaly gained with use of the 21-gene rs assay instead of current practice. We also estimated an incremental cost per life-year gained. For the base case, a probabilistic analysis was conducted to assess the uncertainty of all parameters. We used a Dirichlet distribution for multinomial data, gamma distributions for costs, and beta distributions for probabilities and utilities. All parameters were randomly sampled from their assigned distributions and 10,000 simulations were performed. We estimated the likelihood of each strategy being more favourable across a range of willingness-to-pay thresholds using cost-effectiveness acceptability curves. The results were also summarized on a cost-effectiveness plane. A deterministic 1-way sensitivity analysis was conducted on all key model parameters (such as the probability of recurrence with chemotherapy and the rs stratification).

Validation
The model structure and outcomes were evaluated. For each rs category, we obtained the proportion of deaths attributable to chf, leukemia, distant recurrence, short-term toxicity, and background mortality. Those estimates were shown to clinician experts to determine whether they were in line with clinical practice (supplementary Tables i and ii).

RESULTS
The results indicated that the 21-gene rs strategy dominated current practice. The 21-gene rs strategy was less costly ($432 less) and more effective (0.22 qalys) than current practice over a lifetime horizon (Table iii). In addition, 0.17 life-years were gained with the 21-gene rs assay strategy.
The probabilistic analysis revealed that 70% of the 10,000 simulated icers were located in the southeast quadrant, meaning that the 21-gene rs assay strategy was less costly and more effective than current practice ( Figure 2). If 1 qaly gained was valued at $50,000, then 95% of the simulated icers were considered cost-effective ( Figure 3). The tornado diagram in supplementary Figure 1 shows the main variables that influenced the results. All other sensitivity analyses (within the specified ranges) did not affect the results. If the probability of receiving chemotherapy in the low-risk rs category increased to 50% from 0%, then the icer was approximately $21,299 per qaly gained (the 21-gene rs assay strategy was more costly and more effective). If the probability of being in the low-risk rs category decreased to 0.39 from 0.57, the 21-gene rs assay strategy was dominated by current practice. Lastly, if the probability of receiving chemotherapy in the current practice arm decreased to 0.64 from 0.79, the icer was $2,159 per qaly gained.

DISCUSSION
We found that the 21-gene rs assay strategy could be a costeffective option for Ontario patients with ln-positive (1-3 lns), er-positive, her2-negative early bca. Many costeffectiveness studies have been conducted for ln-negative patients 10,11,13,14,16,41 . A number of studies found that the 21-gene rs assay is primarily a cost-effective option (that is, the icer falls below the willingness-to-pay threshold) 11,13,14,41 ; a few found that the 21-gene rs assay is not cost-effective for the ln-negative population 10,16 .
Studies involving patients with ln-positive bca have been conducted in various countries, and most have determined that the 21-gene rs assay strategy might be a costeffective option. To our knowledge, our Canadian costutility analysis is the only one focused on 1-3 ln-positive, er-positive, her2-negative bca. Of two cost-utility studies conducted in Canada 14,16 , the first 14 examined the costs and effects of the 21-gene rs assay from the perspective of the Canadian health care payer over a 25-year time horizon. The population consisted of patients 50 years of age with ln-positive, endocrine-sensitive bca. The number of lns was not specified. The results indicated that the 21-gene rs assay cost $864 more and led to 0.06 qalys gained (icer: $14,844 per qaly gained). All 1-way sensitivity analyses were Filgrastim type 1, 0.50; filgrastim type 2, 0.50. Dx = diagnosis; RS = Recurrence Score (Oncotype DX: Genomic Health, Redwood City, CA, U.S.A.); CTx = chemotherapy; FEC-D = 5-fluorouracilepirubicin-cyclophosphamide, followed by docetaxel; TC = docetaxel-cyclophosphamide; dd AC-T = doxorubicin-cyclophosphamide, followed by paclitaxel; PICC = peripherally inserted central catheter.
within the $50,000-per-qaly willingness-to-pay threshold (icer: $6,199-$14,297). That study took into consideration drug costs associated with 2 chemotherapy regimens (fec-100 and fec-d). The cost per regimen was higher in the present study because the costs of the treatment regimen, the care associated with chemotherapy, and granulocyte colony-stimulating factors were included. The present study was therefore more comprehensive in its costing approach. Also, the other cost-utility study assumed that, after rs stratification, chemotherapy use would be 7%, 70%, and 92% for the low-risk, intermediate-risk, and high-risk rs categories respectively. In contrast, our study assumed that no patients in the low-risk rs category would receive adjuvant chemotherapy and that 100% of patients in the high-risk rs category would.
The second Canadian cost-utility study 16 examined the costs and effects of using the 21-gene rs assay strategy over the lifetime of a patient diagnosed with early-stage er-positive or progesterone receptor-positive, axillary ln-positive (1-3 lns) bca. It found that in postmenopausal women, the 21-gene rs assay strategy cost $36.20 more and led to 0.08 qalys gained (icer: $464 per qaly gained). The sensitivity analysis revealed that, if fewer than 36% of the women in the intermediate-risk rs group received adjuvant chemotherapy, then the 21-gene rs assay led to a negative cost and effect. Their model differs from ours in that they also modelled local recurrence, a second primary bca, and remission with or without chemotherapy-associated serious adverse events. For the chemotherapy-associated adverse events, they considered 8 different diagnoses that resulted in hospitalizations. The types of serious adverse events considered were not reported in the publication. The treatment regimens that the authors considered also differed slightly from ours. The costs of their chemotherapy regimens were lower than the costs of ours. (They might not have included the costs associated with nursing and pharmacy time, a peripherally inserted central catheter, physician visits, and granulocyte colony-stimulating factor). Our study adds to the literature given that we used-and made explicit-detailed costs of the chemotherapy regimens.
Three studies conducted in the United States and the United Kingdom also considered the patient population considered in the present work 9,15,42 . Chandler et al. 42 found that, compared with current clinical practice and taking a health care payer perspective, the 21-gene rs assay strategy resulted in an icer of $188,125 per qaly gained. The results were most sensitive to test accuracy and utilities. With perfect test accuracy, the cost-effectiveness ratio was $28,947 per Cost-effectiveness plane, with 10,000 simulated incremental cost-effectiveness ratios from probabilistic analysis. WTP = willingness to pay. qaly. If the utilities took into consideration the effect on the patient of decreased worry by receiving information about recurrence risk, the icer was $58,431 per qaly. Vanderlaan et al. 15 examined the cost-effectiveness of the 21-gene rs assay strategy from a health care payer perspective over a 30-year time horizon. That study separated patients in the 21-gene rs assay arm into low-risk and high-risk rs categories. The results indicated that the intervention was less costly than current practice ($384 less) and more effective (0.127 qalys gained). The intervention dominated current practice. The sensitivity analysis revealed that if the proportion of tested patients who received chemotherapy after assay testing decreased to 42% from 54%, then the intervention resulted in even more cost savings. One study from the United Kingdom examined the same patient population from the perspective of the health care payer over a lifetime horizon 9 . That study stratified patients into low-risk and high-risk rs categories. The results revealed a mean incremental cost of £860 and incremental effectiveness of 0.16 qalys gained (icer: £5,529). The results were sensitive to the price of the 21-gene rs assay, the model time horizon, the relative risk of chf, quality of life while on chemotherapy, and rs cut-offs for low-risk and high-risk.
It is difficult to compare our overall results with the results from that study, given that its rs stratification omitted the intermediate-risk rs category.

Strengths and Limitations
The major strength of our study is that most of the data were obtained from published Canadian sources and reflect current practice. That approach makes our model applicable to the Ontario setting and potentially helpful for guiding practice in the province. As well, our model takes into consideration current treatment regimens within the Canadian setting. We addressed most of the modelling concerns reported in a systematic review of cost-effectiveness analyses of the 21-gene rs assay 43 . Specifically, we considered both the short-term and long-term toxicities associated with chemotherapy; we selected a diagnosis age of 60 years, because that is the average age of patients presenting with bca; and we assumed that that patients in the low-risk rs group had the same risk of distant recurrence whether they were treated with adjuvant chemotherapy or not. Our study has a number of limitations to highlight. First, because of a lack of published data, it did not consider local recurrence. Having that information might make the 21-gene rs assay strategy look even more economically attractive. Second, the cost of the disease-free health state might have been underestimated given that the costs were obtained from a study that reported costs for women diagnosed with bca in 1995 33 . Third, our model did not include the cost of antiemetic drugs to treat nausea and vomiting. Adding the cost of those drugs would make the 21-gene assay strategy more economically attractive. Fourth, a number of assumptions were made that might have influenced the findings. For example, the risks of recurrence for patients in the low-risk rs category receiving and not receiving chemotherapy were assumed to be the same. In addition, the risk of recurrence became zero after 20 years, and the risks of chf and leukemia occurred only during years 1-7 after chemotherapy. Nevertheless, those assumptions were made based on existing literature and were confirmed by clinicians to be clinically appropriate. Lastly, because our model was Ontario-specific, the generalizability of the findings might be limited to settings with a similar context.
It is important to note that, although bca recurrence with chemotherapy was not shown in the 1-way sensitivity analysis to have a significant effect on the results (within the clinically plausible ranges), the results could change in favour of current practice depending on the method used for estimation of recurrence. The recurrence rates used in our study were obtained for each rs risk category and for ln-positive women 25 . A study by Pan et al. 44 of bca recurrence in women with er-positive bca and 1-3 positive lns reported that the annual rate of distant recurrence was 2.2% in year 1 and 1.8% in year 20. We could not directly compare their estimates with the estimates from Albain et al. 25 , because Pan et al. did not separate their estimates by rs risk category.

CONCLUSIONS
The 21-gene rs assay was found to be an economically attractive option compared with no 21-gene rs assay in women with ln-positive (1-3 lns), er-positive, her2negative bca in Ontario , given certain model assumptions. Those findings might be of considerable interest to policymakers and health care providers in their decision-making process for personalized medicine for bca patients in similar settings.