The incidence of breast cancer is growing in developed countries due in part to the increase in life expectancy.1 In addition, as a diagnosis anticipating intervention, population screening has led to an increase in cases discovered in early stages.2 An important portion of these cases appear with negative lymph nodes, positive estrogen receptors and negative HER2 in which hormonal treatment has proven effective,3 but there are doubts about the risk-benefit relationship of adjuvant chemotherapy. In some cases, the adverse effects outweigh the benefit due to the low risk of recurrence and, in others, their biological characteristics are less sensitive to chemotherapy.4 In order to assist in the decision about which patients should receive chemotherapy, different clinical-pathological criteria have been developed, such as those of the National Comprehensive Cancer Network (NCCN),5 but the accuracy of their application has been suboptimal. The need for more accurate tools stimulated the development of new multigenic tests that predict the risk of metastasis and allow the selection of cases that are high risk and will therefore benefit from chemotherapy.6–8

The Oncotype test is based on the technique known as reverse transcriptase polymerase chain reaction, that analyzes five reference genes and 16 genes related to cancer and cell-associated proliferation, invasion and hormonal response.9 It was introduced into the market in 2004 and uses a continuous scale to predict the risk of recurrence of cancer at 10 years and the benefit of adjuvant chemotherapy. In 2012, the Department of Health of the Basque Government decided to incorporate it into the services portfolio of the Basque Health Service to be used in cases of early breast cancer according to the indications defined by a group of experts.10 Although Oncotype cost-effectiveness studies have been carried out, most have been developed from theoretical models based on initial indications.11 Further, its use has not been the same in all countries and a number of economic evaluations have been carried out specifically for the use of the test in each country.12,13 To fully understand the specific impact of the use of Oncotype, the cost-utility study presented here should be based on data from clinical practice in the Basque Health Service.14,15

The objective of this work was to carry out a cost-utility study of the use of the Oncotype genomic test in the Basque Health Service during the years 2012-2015 through a probabilistic model, thus addressing the corresponding gaps in the research literature and promoting better modeling of breast cancer treatment alternatives.

The research consisted of a cost-utility study based on a model representing the natural history of breast cancer. The alternative of treatment with chemotherapy according to the risk score criterion based on the Oncotype was compared with the standard alternative based on clinical-pathological criteria.16,17 The technique used was discrete events simulation with the software Arena version 14 (Rockwell Corp.). The population included and therefore simulated in the model were 401 patients with a mean age of 57 years and T1b-c/T2 localized breast cancer, positive hormonal receptors, HER2- and size N0 or N1mi participating in the clinical study.10 Data related to characteristics of patients, Oncotype score and prescribed treatment type were collected over three years of Oncotype clinical practice use in Basque hospitals.10 The treatment that would have been assigned to each patient if Oncotype had not been used, hormonal therapy alone or in combination with chemotherapy, was also recorded by oncologists before testing with Oncotype.10 Though it was not actually applied to patients, that standard alternative based on conventional clinical-pathological factors was used as comparator in the economic evaluation. The treatment recommendation after the genomic test was chemotherapy if test risk-score was high (>30), hormonotherapy alone if the risk was low (<18) and the decision took into account clinical-pathological factors if the risk was intermediate (18-30). The inclusion of Oncotype led to a decrease in the number of chemotherapies prescribed with clinical-pathological criteria, from 56% to 25% of the cases.10

The deterministic cost-utility analysis from the health system perspective estimated an ICER of 17,453 euros/QALY with the discount and 9,613 euros/QALY without the discount (Table 3). The result from the perspective of society as a whole is shown in the same table. Thus, when productivity costs were included, the evaluated intervention was dominant since it determined cost savings and health gains.

The results of the probabilistic analysis are shown in Figures 1, 2 and 3. The 95%CI for the mean ICER was (15,371€/QALY – 16,909€/QALY) and the CI for the total ICERs cloud was (4,722€/QALY – 51,345€/QALY). In the cost-effectiveness plane (Fig. 1) the results of all the simulations appear in the northeast quadrant of the cost-effectiveness plane from the perspective of the Basque Health Service with discount that is characterized by positive incremental costs and utilities; efficiency relies on the adopted threshold. Applying the societal perspective, the results appear in both northeast and southeast quadrants. The cost-effectiveness plane obtained without discount from health service and societal perspectives are shown in Figure A2 in the online Appendix. The acceptability curve (Fig. 2) displays the percentage of simulations in which the ICER is below each threshold of efficiency or willingness to pay from the perspective of the Basque Health Service and with the discount. The percentage of simulations lower than the threshold of 25,000 euros/QALY is 85%. The figure is higher still when the discount is not applied.

Figure 1.

Cost-effectiveness plane for the use of Oncotype from the perspective of the Basque Health Service and Social perspective with discount. QALY: quality adjusted life year.

(0.34MB).

Figure 2.

Acceptability curve for the use of Oncotype from the perspective of the Basque Health Service with discount. INB: incremental net benefit; QALY: quality adjusted life year.

(0.11MB).

Figure 3.

Expected value of perfect information by patient and group of variables from the perspective of the Basque Health Service with discount. EVPI: expected value of perfect information; QALY: quality adjusted life year; RS: risk score.

(0.1MB).

When all the parameters were varied, the total EVPI for the target population was 86,333 euros. The EVPPI synthesized in Figure 3 allowed identifying treatment decision (risk score and chemotherapy distribution) as the group of parameters that provided the greatest variability to the final result not only for the threshold of 25,000 euros/QALY estimated for the Spanish national health system26 but for the entire range of thresholds included in the calculation. The adverse effects hardly generated risk of being wrong with the decision to use Oncotype.

The use of the Oncotype genomic test in the Basque Health Service is a cost-effective intervention since each QALY gained costs less than 25,000 euros. If sick leave costs due to chemotherapy are included in the analysis, the result is a net saving. Thus, from the societal perspective, the Oncotype option is dominant since it generates positive health results and is accompanied by savings in the use of resources.

This decision model has two key characteristics. The first is the precision with which the genomic test establishes the risk of recurrence since the clinical-pathological criteria have important weaknesses. An example is the size of the tumor that is included in these criteria but which presented a non-significant ROC curve in the clinical study.10 In general, clinical-pathological criteria approach decision-making conservatively, prescribing chemotherapy in doubtful cases. The result is unsatisfactory because the percentage of clear cases is small. The number of chemotherapies can be reduced with the test since the improvement of the predictive values allows for a significant reduction in the group of doubtful patients. Thus, patients gain quality of life because a substantial percentage of chemotherapies can be avoided. The reduction of chemotherapies by one third of the total sample is in line with other studies.27 The second key characteristic concerns the cost of chemotherapy. The uncertainty associated with the parameters of the model indicates that in only 85% of the simulations is the ICER lower than the efficiency threshold. The probabilistic analysis allows for identification of groups of parameters that generate uncertainty in treatment decisions. The EVPPI estimation demonstrated that the parametric uncertainty about the result of the test was a relevant issue given the sample size of the clinical study.10 In each simulation, taking a set of probabilities from the two Dirichlet distributions that represented the probability of the score being in a particular group (<18, 18-30,>30) and the cost of chemotherapies, determined a loss of opportunity for the decision to apply Oncotype.

This result is consistent with other studies in which the use of Oncotype is dominant from a health perspective.11,13 Compared with our unit costs, the main difference with respect to other research is in the figure for chemotherapy cost, which is significantly higher in the literature. Thus, Smyth et al.13 applied a total cost of 7,903 euros in 2014 for each patient in Ireland treated with chemotherapy. This cost included pharmacy costs (2,161 euros) and bone marrow stimulators such as granulocyte colony-stimulating factor with pegfilgrastim (4985 euros)13. By contrast, the unit cost of chemotherapy in the Basque Health Service is much lower (2400-2500 euros).10 It is clearly the case that the higher the cost of the chemotherapy, the higher the efficiency of a genomic test that avoids or diminishes it and our probabilistic analysis quantitatively supports this relationship.

Other economic evaluations carried out in France, Mexico and the Netherlands concluded that the use of Oncotype is cost-effective from the Health Service perspective28–30 and dominant from the societal perspective when incorporating lost productivity costs.30 The value of information analysis of Oncotype was addressed in the study by Hall et al.31 to inform the planning of a clinical trial in the United Kingdom. They estimated that Oncotype is dominant but with big confidence intervals, justifying a high total EVPI. We cannot compare their results with ours because they apply Oncotype to patients with higher risk of recurrence. For them, the standard alternative was not clinical-pathological criteria, but prescribing chemotherapy directly. Our conceptual models are similar to those developed in different countries as they are structured on the recurrence risk according to different criteria and treatments. Our results are also consistent with a systematic review of the economic evaluation of genomic tests for early breast cancer care which estimates that genomic testing results in a moderate increase in total costs, but with an ICER below the threshold.32 Another systematic review points out the risk of bias in some of the economic evaluations because of source of the funding.33 This limitation does not affect our study.

From the methodological point of view, we should note a weakness in our study due to selection bias derived from the observational design. But target populations for Oncotype have been tailored to each country and our model had, necessarily, to be tied to the features of Basque clinical practice that were different from others.10,31 Another limitation to be mentioned in our analysis of the EVPI concerns the uncertainty about the use of chemotherapy that is not purely random in statistical terms but reflects differences in clinical practice. Despite these, we have included the distribution of the different treatment schemes because it highlights the added value of probabilistic processing and its key role in the estimation of the efficiency of Oncotype. At the same time, it allows to show that if the cost of oncological medicines increases, the efficiency of the use of the Oncotype will improve. Using a threshold amount the system would be willing to pay in the economic evaluations carried out in the Spanish health sector is, for the time being, purely academic, since the health authorities have not defined a reference figure or a methodology to support it. Traditionally the threshold of 30,000 euros/QALY proposed by Sacristán et al.34 in 2002 from a review of the literature has been used. Recently, Vallejo-Torres et al.26 have proposed a figure between 22,000 and 25,000 euros. The incremental cost of Oncotype Dx is found to be lower than the maximum per QALY suggested as appropriate for Spain, although, at this time, Spanish policy makers do not formally support a fixed definition for a threshold of acceptability.

In conclusion, this case of personalized medicine to inform decision-making in early breast cancer treatment resulted in a cost-effective use of resources in the Basque Health Service according to the results of the cost-utility analysis.

David Epstein.

The corresponding author on behalf of the other authors guarantee the accuracy, transparency and honesty of the data and information contained in the study, that no relevant information has been omitted and that all discrepancies between authors have been adequately resolved and described.

O. Ibarrondo has participated in the design of the study, the construction and processing of the model and the writing of the article. A. Arrospide has participated in the design of the study, the construction and processing of the model and has reviewed the manuscript. I. Álvarez-López, E. Galve-Calvo, M. Gutiérrez-Toribio and A. Plazaola have participated in the design of the study, the field work and the interpretation of the results. J. Mar and F. Freundlich have contributed to the design of the study, collaborated in the interpretation of results and reviewed the article, providing valuable suggestions.

The authors thank Purificación Martínez del Prado, Nerea Ancizar-Lizarraga, Severina Domínguez, Ainhara Lahuerta, Patricia Novas-Vidal, Cristina Churruca, Garbiñe Unanue and Lide Larburu for their contribution to the collection and assembly of data.

This work has been carried out within the doctoral thesis project of O. Ibarrondo and has not received external funding.

None.