Background
‘Personalized medicine’ (PM) is an increasingly relevant concept in clinical oncology. The term PM refers to an evolving approach to clinical decision making which seeks “to improve the stratification and timing of health care by utilizing biological information and biomarkers on the level of molecular disease pathways, genetics, proteomics as well as metabolomics” [
1]. Although genomic information is considered to be the cornerstone of this discipline [
2], clinical and sociodemographic characteristics of the patient and individual preferences can also be utilized to personalize medicine [
3]. Because treatment strategies can be tailored in such a way that only patients who stand to benefit receive treatment [
4], PM is particularly relevant in diseases, such as breast cancer, where in some cases the potential adverse effects may outweigh the benefits of treatment [
5].
Breast cancer is among the most common types of cancer and a leading cause of cancer deaths in women. In Austria, breast cancer accounts for 30% of all tumors and for 16% of all cancer deaths in women [
6]. The incidence of breast cancer in Austria in 2012 was about 76 cases per 100,000 women [
6]. One of 13 females born in 2011 will develop breast cancer by the age of 75 years [
7]. Aside from a small percentage of familial breast cancer, the risk factors for this malignancy are rather broad and vague: e.g., age, early menarche, late menopause, and obesity [
8]. Although many treatment options are available [
9,
10], the standard of care for early breast cancer is surgical resection, often followed by adjuvant radiation. Additional adjuvant systemic therapy depends on the hormone receptor status, such as estrogen receptor (ER) status, postmenopausal status, human epidermal growth factor receptor 2/neu (HER-2/neu) status, stage of the disease and co-morbidities.
For women with lymph node negative, estrogen receptor (ER) positive early-stage breast cancer who have relatively low recurrence risk adjuvant chemotherapy decision is complex and uncertain. While adjuvant systemic therapy can be beneficial for women at higher risk of a distant recurrence, it can cause more harm than benefit for low risk patients. Several prognostic tests are available to help identify women most likely to benefit from adjuvant systemic therapy in order to help guiding adjuvant therapy decision-making. For example, Adjuvant!Online is a free online tool that estimate risks and benefits of adjuvant therapy after breast cancer surgery based on factors, such as the patient’s stage, pathologic features, age and comorbidity level [
11]. Mammaprint and OncotypeDX (ODX) are gene expression assays that “combine the measurements of gene expression levels within the tumor to produce a number associated with the risk of distant disease recurrence. These genetic tests aim to improve on risk stratification schemes based on clinical and pathologic factors currently used in clinical practice” [
12]. As clinical decisions are increasingly based on the predictions of these tests, the additional costs should be considered in decision analyses of cancer management and treatment, similar to other companion diagnostics in PM [
13].
Several studies have been conducted to evaluate the cost effectiveness or cost utility of different chemotherapy regimens. In our systematic literature search in CRD (Center for Reviews and Dissemination) [
14], we found 24 cost-effectiveness studies that evaluate various chemotherapeutic regimens including capecitabine, cyclophosphamide, docetaxel, doxorubicin, epirubicin, eribulin, flourouracil, gemcitabine, ixabepilone, methotrexate, mytomycin, paclitaxel, vinblastine, and vinorebline. These studies were performed in the healthcare contexts of China, Canada, Germany, France, Italy, South Korea, Spain, UK, USA, The Netherlands, and Thailand. In particular, none were conducted for the Austrian healthcare context and none of these take into account personalized treatment decisions based on risk classification by AO and the new 21 gene assay ODX.
Our study focused on patients with ER and/or progesterone receptor (PR) positive, HER-2/neu negative and lymph node negative early breast cancer for whom AO and ODX risk classification may provide additional information that impacts decision-making. An advanced literature search conducted in PubMed [
15] yielded no further studies on the combined prognostic approach of AO and ODX for these risk groups. Only Paulden et al. evaluate in a secondary analysis cost effectiveness of chemotherapy within risk groups according to AO and ODX. However, this was in a Canadian setting which differs from Austria (e.g., due to provided chemotherapy regimens and costs) [
16].
The goal of the current study was to evaluate risk-group specific cost effectiveness of adjuvant chemotherapy for Austrian women with resected ER and/or PR positive, HER-2/neu negative, and lymph node negative early breast cancer. All potential risk groups according to the joint application of AO and ODX are considered. Additionally, we then compare these results to those from the Canadian study by Paulden et al. [
16].
Results
Base case
The results of the base-case analysis for the Austrian and the Canadian settings are displayed in Table
1. For each risk group, two lines depict the estimated, discounted LYs, QALYs and costs when chemotherapy is provided and when it is not. The ICER summarizes the results of chemotherapy or none.
Table 1
Discounted life-years, QALYs and incremental cost-effectiveness ratios of chemotherapy in the Austrian setting versus Canadian setting
Low | Low | No | 15.51 | 12.04 | 11,021 | D | 15.36 | 11.93 | 10,088 | D |
Yes | 15.16 | 11.65 | 23,383 | 15.14 | 11.69 | 11,817 |
Low | Int. | No | 14.96 | 11.61 | 12,063 | D | 14.85 | 11.52 | 11,279 | 61,861 |
Yes | 15.05 | 11.57 | 23,608 | 15.04 | 11.61 | 16,562 |
Low | High | No | 12.06 | 9.31 | 17,520 | 3361 | 12.07 | 9.32 | 17,368 | 3118 |
Yes | 14.73 | 11.31 | 24,240 | 14.73 | 11.37 | 23,743 |
Low | N/A | No | 14.80 | 11.48 | 9230 | 566,277 | 14.69 | 11.40 | 8347 | 3768 |
Yes | 14.97 | 11.50 | 20,554 | 14.96 | 11.55 | 8925 |
Int. | Low | No | 15.21 | 11.80 | 11,557 | D | 15.10 | 11.72 | 10,708 | D |
Yes | 15.02 | 11.54 | 23,641 | 15.01 | 11.59 | 12,105 |
Int. | Int. | No | 13.71 | 10.62 | 14,421 | 13,504 | 13.69 | 10.60 | 13,872 | 4094 |
Yes | 14.77 | 11.34 | 24,163 | 14.78 | 11.40 | 17,150 |
Int. | High | No | 9.45 | 7.25 | 21,733 | 698 | 9.51 | 7.30 | 22,401 | 416 |
Yes | 14.61 | 11.21 | 24,500 | 14.60 | 11.26 | 24,049 |
Int. | N/A | No | 12.63 | 9.77 | 13,372 | 4790 | 12.63 | 9.76 | 12,927 | 548 |
Yes | 14.78 | 11.35 | 20,960 | 14.79 | 11.41 | 13,833 |
High | Low | No | 15.21 | 11.81 | 11,561 | D | 15.09 | 11.71 | 10,721 | D |
Yes | 15.03 | 11.55 | 23,666 | 15.01 | 11.59 | 12,116 |
High | Int. | No | 13.76 | 10.66 | 14,471 | 14,496 | 13.66 | 10.58 | 13,969 | 3940 |
Yes | 14.75 | 11.33 | 24,143 | 14.77 | 11.40 | 17,186 |
High | High | No | 9.43 | 7.23 | 21,815 | 676 | 9.48 | 7.27 | 22,474 | 400 |
Yes | 14.60 | 11.21 | 24,502 | 14.59 | 11.26 | 24,068 |
High | N/A | No | 12.09 | 9.34 | 14,257 | 3097 | 12.08 | 9.33 | 14,078 | 2816 |
Yes | 14.88 | 11.43 | 20,729 | 14.87 | 11.48 | 20,121 |
The results for the Austrian setting indicate that chemotherapy is dominated in the risk groups L-L (low AO, low ODX), L-I (low AO, intermediate ODX), I-L (intermediate AO, low ODX) and H-L (high AO, low ODX). Patients in these risk groups do not on average benefit from chemotherapy with respect to the clinical outcomes (LYs, QALYs). These results are consistent with the results for the Canadian setting with the exception of the L-I risk group (low AO and intermediate ODX).
In high risk ODX patients, chemotherapy seems to clearly be cost effective because an additional QALY can be gained at a low additional cost (ICER less than 3500 EUR/QALY). Chemotherapy is also cost effective in patients with an intermediate ODX risk and an intermediate or high AO risk chemotherapy with a WTP threshold of 15,000 EUR/QALY. These results are also consistent with the results from the Canadian setting. For patients in our model that are tested only with AO, chemotherapy is mainly cost effective with the exception of those who are AO low risk (L-N). These results differ slightly to the Canadian setting where chemotherapy for L-N patients is cost effective.
Sensitivity analyses
Results of the sensitivity analyses are displayed in Table
2 (assuming WTP 50,000 EUR/QALY) and in the additional files (Additional file
2: Table S2A, Additional file
3: Table S2B, Additional file
4: Table S2C and Additional file
5: Table S2D assuming WTP 100,000 EUR/QALY). We ran the analysis for the four main risk groups (ODX low, ODX intermediate, ODX high, ODX not provided). In each block, we considered the respective AO risk in three columns. The first row in the table provides the results of the cost effectiveness of chemotherapy for each risk group in the base case. For example, for patients that have a low risk according to ODX and a low risk according to AO (L-L), chemotherapy was dominated (D) in the base case. In the following section, we display the results of the lower and upper bound when the parameters are varied. For example, we first consider a patient cohort age 40 (lower bound) and a patient cohort age 70 (upper bound). For the above risk group L-L, we observe that chemotherapy is still dominated, even if we vary the parameter age within the range of 40–70 years. We marked parameters depending on their impact on cost-effectiveness results and the following decisions: a) if the parameters that were changed led to the same decision based on the cost-effectiveness result, we use a white background, b) if those parameters that were varied led to a different decision based on the cost-effectiveness results, cells were colored with a dark grey. These were done assuming a WTP threshold of 50,000 EUR/QALY (Table
2) or 100,000 EUR/QALY (Additional file
2: Table S2A, Additional file
3: Table S2B, Additional file
4: Table S2C and Additional file
5: Table S2D).
Table 2
Sensitivity analyses of cost effectiveness of chemotherapy
In summary, in one-way sensitivity analyses results were robust to changes in utilities, costs of chemotherapy and the genetic test ODX, a discount rate of 2.5% and patients at 40 years of age. For older age groups, the decision would be similar assuming a WTP of 65,000 EUR/QALY. The results, however, were sensitive to the probabilities of distant recurrence (with and without chemotherapy) especially within the risk groups L-I, I-L, I-I, H-L, H-I, L-N.
In the risk groups L-I, I-L and H-L, chemotherapy was dominated in the base case but was cost effective when the probabilities of distant recurrence were varied. In the risk groups I-I and H-I, chemotherapy was dominated when the probability of distant recurrence was varied. L-N became cost effective when the lower range of the probability of distant recurrence following chemotherapy was used.
Discussion
In our cost-effectiveness analysis of adjuvant chemotherapy for early stage breast cancer patients, we evaluated upfront testing within the cancer management process similar to the evaluation of companion diagnostics. Our analysis shows that in the Austrian setting chemotherapy is effective and potentially cost effective for patients with an intermediate or high risk of disease according to ODX, independent from the AO risk classification (with the only exception of risk group L-I). In other words, low ODX risk suggests chemotherapy should not be considered but low AO risk may benefit from chemotherapy only if ODX risk is high.
Our results demonstrate that if the ODX test was not applied, chemotherapy would be considered cost effective for AO intermediate and high risk patients. However, based on the additional results of the ODX test, chemotherapy is dominated (less effective and more costly) for ODX low risk patients within these AO risk groups. Therefore, in the decision process we would not favor chemotherapy and consequently, reduce harms and costs. For low risk patients per the AO test, chemotherapy would very likely not be cost effective. However, after taking into account the additional information provided by the ODX test as well as the additional costs of this test, chemotherapy became cost effective for AO low and ODX high risk patients. In particular, these patients would greatly benefit from chemotherapy, both clinically and economically. In summary, our results demonstrate the importance of considering personalized information and additional costs in the evaluation of chemotherapy.
Sensitivity analyses demonstrate that the results are relatively robust with respect to the decisions about almost all model parameters except for the probability of distant recurrence within the risk groups L-I, I-L, I-I, H-L, H-I, L-N. For high risk patients per ODX and those classified only based on AO, the results are robust.
The advantage of our modeling approach is that, in addition to providing risk group-specific cost effectiveness of chemotherapy, we are also able to evaluate the effectiveness and cost effectiveness of the risk classification tools as previously shown [
19]. Within this analysis, we considered that decisions regarding chemotherapy are based on the risk classification and additional factors. Therefore, only a percentage of patients would finally agree or not agree on chemotherapy in the respective risk groups. Our modular modeling structure approach allows one to adapt the model to evaluate additional test information or other innovative personalized test-treatment decisions.
Our results are consistent with the analysis of Paulden et al. [
16] that showed a similar cost effectiveness of chemotherapy in the Canadian setting when using comparable risk classifications. In our systematic literature review, we identified no cost-effectiveness study for Austria nor any study that applied Adjuvant!Online or OncotypeDX. We identified four studies that sought to evaluate the same adjuvant chemotherapy regimen (FEC and TC). However, only one of these studies compared chemotherapy versus no chemotherapy. Campbell et al. [
32] compared four strategies including one strategy without chemotherapy and three with different chemotherapy regimens. They found that “with an average to high risk of recurrence […], FEC-D appeared most cost effective assuming a threshold of £20,000 per QALY for the National Health Service (NHS). For younger low risk women, E-CMF (epirubicin, cyclophosphamide, methotrexate, fluorouracil) /FEC tended to be the optimal strategy and, for some older low risk women, the model suggested a policy of no chemotherapy was cost effective” [
32]. These results were consistent with our results that also suggest that adjuvant chemotherapy is not cost effective in low-risk groups but is in high risk groups.
In Austria, there is currently no explicit threshold for health technologies to be considered cost effective. In other countries, thresholds vary and they are rarely disease or cancer specific. For example, in Canada, an oncology-specific ceiling threshold value of C$75,000 (equivalent to EUR 51,528) has been suggested and NICE (National Institute for Health and Care Excellence) provides a general threshold in 2012 of £18,317/QALY (EUR 23,180) that can be revised based on other factors [
33].
Our study has several limitations. Although modeling studies allow information to be combined from different sources, we included as much Austrian data and local information on cancer management as possible. However, due to a lack of information about utility parameters and estimates for the risk of distant recurrence, we applied results from international studies. The underlying causes of hospitalizations were adapted for the Austrian context based on information of local clinical experts.
For some of the risk groups of interest, the decision regarding the provision of adjuvant chemotherapy may be predefined. However, we analyzed all potential groups for completeness. In addition to AO and ODX, there are other risk classification scores and genetic tests that may be used for this purpose. Since AO is continuously updated and free of charge, it was considered as first choice. Although not currently covered, ODX is a genetic test that has shown convincing analytical and clinical validity and therefore is likely to be implemented in clinical practice in Austria in the near future.
The ability to compare these results with the Canadian study results is limited due to the different health care settings (e.g., type of chemotherapy recommended, follow up treatment, cost structure), however, the results fall in a reassuringly similar direction.
In the future, our analysis could be applied in cost-effectiveness analyses based on risk classifications that are obtained from combinations of various multi-parameter molecular marker assays. At the fourteenth St. Gallen International Breast Cancer Conference, one expert panel discussed the role of multi-parameter molecular marker assays for prognosis and their value in selecting patients who require chemotherapy. “Oncotype DX®, MammaPrint®, PAM-50 ROR® score, EndoPredict® and the Breast Cancer Index® were all considered usefully prognostic for years 1-5” [
34]. Beyond 5 years, reports suggest that these tests are prognostic [
34]. The Panel agreed the PAM50 ROR® score to be clearly prognostic beyond 5 years. However, a clear majority rejected the prognostic value of MammaPrint®. For Oncotype DX®, the majority of the panel agreed with the potential value in predicting the usefulness of chemotherapy. Improved evidence supporting our modeling study will be provided by the TAILORx trail. After the full TAILORx trial results on ODX become available, we will rerun the analysis using updated input parameters including the probabilities of distant recurrence. Although there are promising alternative tests that allow personalized treatment decisions, multi-parameter molecular assays are expensive and may not be widely available [
34].
Nevertheless, for reimbursement decisions, there have been strong efforts to enhance patient access to PM in Europe [
35]. Decision-analytic modeling demonstrating cost effectiveness of combined test-treatment decisions may, therefore, provide particularly important information to decision makers and, potentially, improve accessibility to PM. For example, the ISPOR Personalized Medicine Special Interest Group notes that outcomes research and economic modeling can inform the assessment of PM at an early stage and supports prioritization of further research by early-stage decision modeling of potential cost-effectiveness and value of information (VOI) analyses [
36]. Payne et al. derived recommendations to improve market access for companion diagnostics. Economic modeling is prescribed as a possible approach “to describe and quantify gaps in the evidence base and the added value of future research to reduce current uncertainties to support the introduction of companion diagnostics” [
37].
The role of patient involvement has changed in recent years such that patients are increasingly included in the clinical decision making process. Therefore, personalized treatment has to account more actively for patient preferences and their individual state of health [
4]. For decision-analytic modeling, as demonstrated in this study, future models should allow for patient-specific utility values. Further research in the field of companion diagnostics has identified an additional contribution of complementary diagnostics that goes beyond the usual health gains and cost savings. It highlights for example the value to the patient of having greater certainty of treatment benefit [
38]. The upcoming publication of the EPEMED OHE study 2015 (European Personalized Medicine Association, Office of Health Economics) will provide insights on “how to articulate a value based evaluation per its economic, medical and full social appreciation and how a broader conception of value can be the path toward improving the HTA process” [
39].