Cost-Effectiveness Analysis of Prognostic Gene Expression Signature-Based Stratification of Early Breast Cancer Patients

Predictive and prognostic markers are now routinely used to guide patient management in oncology [1]. In women with early-stage estrogen receptor-positive (ER⁺), human epidermal growth factor receptor 2-negative (HER2⁻) node-negative or node-positive primary breast cancer, an individual measure of risk of distant recurrence determines whether the patient is treated with endocrine therapy alone or with additional adjuvant chemotherapy. The decision to use cytotoxic therapy weighs reduced mortality [2] against the morbidity or mortality of adverse effects [3]. Therefore, clinicians need to identify those patients with a low risk of recurrence to avoid the unnecessary use of chemotherapy and its associated toxicities for only marginal risk reduction [4].

Current clinical guidelines consider clinicopathologic factors, including tumor size, nodal status, and histological grade, to assess individual risk of recurrence. In Germany, for example, the interdisciplinary S3 guideline for the “diagnosis, therapy, and follow-up of breast cancer”, which is regularly updated by experts, is used in clinical practice for the management of breast cancer [5]. Over the last few years, several molecular markers have entered the market to aid clinical decision making [6], but their long-term benefits and disadvantages for the patient and the healthcare system need to be balanced [7].

Since 2011, a new, clinically validated gene expression test (EndoPredict^®, Sividon Diagnostics, Köln, Germany) has been used to determine the risk of distant recurrence in patients with ER⁺, HER2⁻ breast cancer [8]. The EndoPredict^® test is based on the assessment of the expression of eight genes of interest by quantitative real-time polymerase chain reaction (PCR) in combination with two clinical risk factors (tumor size, nodal status), which are combined to a hybrid molecular-clinicopathologic score (EPclin) [8]. EPclin adds prognostic information to other routinely measured parameters, including Ki67 [8, 9], and predicts early and late metastasis [9]. When EPclin is used in combination with established guidelines, chemotherapy use is substantially reduced without compromising patient outcome [10].

Here we evaluate the health economic impact of EPclin using a life-long Markov state transition model. We compare seven alternative strategies, three implying the use of different standard guidelines based on clinico-pathological parameters and four using EPclin, either alone or in combination with each of the three standard guidelines.

Our study is the first health economic analysis of the EPclin test. EPclin has acceptable health economic characteristics from the perspective of the German healthcare system. Combining the St. Gallen guideline with EPclin testing of intermediate-/high-risk patients is beneficial and has the potential to aid clinical decision making on the use of adjuvant chemotherapy in ER⁺, HER2⁻ early breast cancer patients. The robustness of this result was confirmed by extensive sensitivity analyses.

Cost-effectiveness thresholds for clinical interventions vary across countries. For example, in the USA they range from $US50,000 to $US100,000 (€36,600–73,200) per QALY gained and from £20,000 to £30,000 (€23,000–35,000) per QALY gained in the UK, although actual resource allocations may differ [36]. In our base-case analysis, St. Gallen/EPclin, German-S3/EPclin, NCCN/EPclin, and EPclin alone were dominant strategies. Compared to current, guideline-based clinical practice, adding EPclin yielded better clinical outcomes on the QALY scale, at lower costs.

Our analysis did not take into account strategies involving other genetic testing platforms. The main reason for this was lack of information on the results the Oncotype DX^® multigene test or other genetic tests would have yielded in the ABCSG6/8 patients, which formed the basis of this analysis.

The Oncotype DX^® multigene test was recommended for use in routine practice in the USA [37] in 2007, and the health economic implications of this have been thoroughly investigated [7, 33, 38, 39, 40]. Although many studies have claimed cost effectiveness, uncertainties around their estimates were not always fully addressed [7]. Input parameters, including amount of chemotherapy use, recurrence rate, quality of life, time horizon, and the test cost seemed to most strongly influence the results [41].

Blohmer et al. [33] developed a Markov model to assess the costs and effects of using Oncotype DX^® prior to adjuvant chemotherapy in Germany, and reported that an Oncotype DX^®-based strategy would save an average of €561 and increase QALYs by 0.06 per patient. Hornberger et al. [40] compared an Oncotype DX^® strategy with the NCCN guideline and reported savings of $US2,028 (€1,669), and a gain of 0.086 QALYs, per patient. A direct comparison between Oncotype and EPclin on the basis of these results and ours would be difficult given different modeling approaches being used, lower test costs of EPclin, and the different proportion of patients classified as low-risk by EPclin compared with Oncotype DX^®. Nevertheless, all these studies showed cost savings and health gains due to the usage of genetic test results. In contrast to our EPclin-specific model, the two Oncotype DX^® models generated by Blohmer et al. [33, 40] and Hornberger et al. [33, 40] considered a predictivity for chemotherapy benefit, i.e., the relative benefit from chemotherapy was assumed to be lower in low-risk patients than in high-risk patients. This assumption was based on a significant test for treatment interaction of the Oncotype DX^® test observed in the randomized (endocrine vs. endocrine–chemotherapy treatment) NSABP (National Surgical Adjuvant Breast and Bowel Project)-B20 and the SWOG (Southwest Oncology Group)-8814 clinical trials in adjuvant breast cancer patients [42, 43]. Results from neoadjuvant studies suggested that chemotherapy sensitivity may also be enhanced in patients at high risk according to EPclin [44], but so far EPclin has not been evaluated in a randomized adjuvant clinical trial to prove that EPclin selects women who will benefit from chemotherapy treatment. In order to approximate the effect of chemotherapy, we used the largest currently available, relevant meta-analysis by Peto et al. [15]. The assumed relative risk of 0.69 for distant recurrence, in patients undergoing adjuvant chemotherapy, may therefore not be entirely correct for low-risk patients, given their low overall probability of developing a recurrence. Nevertheless, the relative benefit of adjuvant chemotherapy has been shown to be independent from standard prognostic parameters such as ER status, grading, or nodal status [15]. It is worth mentioning that in the Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) meta-analysis, non-cancer-related mortality was increased due to chemotherapy (relative risk 1.2, p = 0.05), while overall mortality was reduced (relative risk 0.84, p < 0.0001) [15]. In order to remain conservative, we did not apply the increase in non-cancer-related mortality, since it was unclear whether this was a true effect or a result of competing risk. We assumed that metastases can still occur even in patients with a 10-year RFS, as shown elsewhere [45, 46].

Hall et al. assessed the impact of Oncotype DX^® versus a chemotherapy strategy for early stage lymph node-positive breast cancer patients in the UK [7]. The testing strategy achieved a small increase in life expectancy (0.15 LYG) or quality-adjusted life expectancy (0.16 QALY), but also an increase in cost (£860, €1,096). The costs were the greatest driver of the base-case ICER (£5,529 or €6,428/QALY). The authors clearly demonstrated the potential of molecular stratification, but also showed that there is a risk of a negative balance between costs and health benefits. Importantly, in our model only 28 % of patients were node positive, and therefore had a lower baseline risk of distant metastasis. Hall et al. also included an assessment of the relative risk of death due to chronic heart failure after chemotherapy, which was not considered in our more conservative model (in order to better discriminate the effects of EPclin). Cardiac events, and especially delayed cardiac death, are uncommon (especially with docetaxel-combined chemotherapy), and cannot always be linked directly to chemotherapy [47]. This explains, in part, the reason why a greater effect was observed in terms of both QALYs and LYG in the Oncotype DX^®-guided strategy than in our results.

Our study has both strengths and limitations. The value of a diagnostic test or guideline depends on the ability to differentiate risk. While this is normally presented as sensitivity and specificity [48], here we did not need to explicitly include these parameters because they were indirectly taken into account in the hazard rates for metastasis extracted from the actual ABCSG trial data [48]. In the absence of chemotherapy data, we had to rely on published data for the utilities and effect of chemotherapy, which in part originated from outside Germany. The utility values had to be drawn from European sources, even though clinical treatment schedules or perception of life quality can vary between countries, evaluation methods, and severity of condition [21]. Given the fact that chemotherapy is likely to negatively influence the quality of life of treated patients, we included a disutility to account for this and as similarly used in other studies [7, 22, 23, 33].

Resource use and unit costs were abstracted from a previously published German study, which might introduce uncertainty [49]. However, the model inputs were selected to best match our defined patient population, and extensive sensitivity analyses were performed to take uncertainty into account. A strict follow-up schedule was assumed for all patients, which might be inappropriate, especially for low-risk patients not receiving chemotherapy; follow-up costs might therefore be even smaller for low-risk patients. The results seemed most sensitive to the hazard rates (e.g., for disease free to metastasis) and the risk classification by various strategies. One reason for this might be the low event number in some risk groups (e.g., stratified by NCCN), which may have increased the influence of chance. In order to partially overcome this problem, we pooled the hazard rates for risk of death from general causes and the risk of death in metastatic patients, thereby improving accuracy.

The results presented here were derived using a cohort approach and are valid for an ‘average’ patient. However, specific patient subgroups might respond differently, with better or worse clinical outcomes. ABCSG6/8 were trials for post-menopausal patients with lower than average risk, favoring the adoption of a more conservative model. An additional question is whether patients actually receive their chemotherapy according to guidelines [50, 51]; it has recently been shown that chemotherapy was only given to 69 % of older women with node-positive and/or ER-negative cancers [32], and although the decision to opt for chemotherapy was easy for about 60 % of women, 23 % found it problematic [52]. Our results may suggest that the EPclin test can mostly contribute to avoiding chemotherapy treatment in patients who would be recommended for chemotherapy according to current practice using clinical guidelines. Hence, it may be most obviously relevant for younger, fit patients due to wide use of chemotherapy in this group. However, older patients have a higher risk of experiencing adverse effects of chemotherapy and may be more reluctant to undergo treatment. In this group of older patients, the test may persuade high-risk patients to undergo chemotherapy treatment and may therefore increase survival and costs over current practice. Similarly, use of the genetic test might change decisions on chemotherapy use in lymph node-negative or lymph node-positive (1–3+) patients, with both clinical and economic implications. Given the lack of information on resource use, chemotherapy effect and other parameters across these sub-groups, it was not possible to analyze the cost and effects for specific patient groups. The fact that some patients may not get chemotherapy despite guideline recommendations was addressed in a scenario analysis, where only a proportion of patients were given chemotherapy. In this analysis, the costs and effects of all strategies decreased and the combined St. Gallen/EPclin and the EPclin alone strategies remained dominant. The preferences of cancer patients always need to be considered, especially when the benefits from therapies are disputable.

Biologic and clinical markers can provide prognostic and predictive information and determine the residual risk after standard hormone therapy and the relative risk reduction for additional chemotherapy [53, 54], sparing exposure to hazardous and expensive adverse events [3]. Our results add to the rationale for addressing this approach in terms of the health economics of clinically validated tests and guidelines, to find those patients who benefit most from chemotherapy. The present study suggests that molecular tests such as EPclin can be sensibly combined with clinical guidelines and can help reduce chemotherapy use in ER⁺ breast cancer patients with few other clinical risk factors [10]. Improved risk stratification can be translated into increased quality-adjusted life and is economically viable.