This study demonstrated that treatment of lymph node–negative HR-positive HER-2–negative breast cancer guided by RS using 21-gene assay was more effective than AOL-guided treatment with regard to life year gained (0.064 years) and QALY gained (0.065 years), at an additional cost of $4,102 per person over a lifetime. Incremental cost-effectiveness ratio for the base case was $63,064 per QALY.
Sensitivity analyses showed that the cost-effectiveness estimates were favorable in younger women. This was due to higher QALY gained with RS strategy in younger patients than in older patients (0.099 QALY in 30-year-olds vs. 0.021 QALY in 70-year-olds) whereas incremental cost with an RS-guided strategy remained similar ($4,036 in 30-year-olds vs. $4,239 in 70-year-olds).
The model was sensitive to changes in the probabilities relating to risk groups and recurrence rates. Varying these values in effect resulted in alteration of test performance of these two strategies. The presence of threshold values within the plausible ranges suggests that the sensitivity and specificity for these two strategies are likely to be comparable. There are little data available in the literature on test properties of a 21-gene assay and AOL. We adapted the results from Bryant et al., the only data available in the literature in which these two methods were compared and a 10-year recurrence rate was assessed in each risk group [12
]. We used an arbitrary range of base value ± 20% for sensitivity analysis, which might have led to an overestimation. The robustness of the model could be more vigorously assessed when this information becomes available.
The ICER was largely driven by the cost of 21-gene assay as it accounted for 22% of total cost in the RS-guided strategy in the base case. Because this was an upfront cost, it was not affected by discounting, yet the effect measured in QALY is heavily discounted as it occurred many years from original diagnosis. The disproportionate change in cost and benefit with time explains the sensitivity of the model to discounting. The utility of various health states and the cost of chemotherapeutic agents had no significant impact on the ICER.
Three studies examining the cost-effectiveness of RS-guided treatment all demonstrated its superiority over conventional methods of risk classification. Hornberger et al. compared RS-guided treatment to treatment guided by National Comprehensive Cancer Network (NCCN) guideline, whereas Kondo et al. compared RS-guided strategy with two other strategies—NCCN guideline or St Gallen recommendation—in the context of Japan's health care system [25
]. Both studies showed increased QALY and acceptable ICER. Lyman et al. showed a gain in life expectancy of 2.2 years with RS-guided treatment compared with treating all
patients with tamoxifen alone; and a net cost savings of $2,256 per patient compared with the cost of treating all
patients with chemotherapy and tamoxifen. QALY was not evaluated in their study [27
It is generally agreed that consensus guidelines are crude tools in risk classification. Risks are determined by one or more clinicopathological features. AOL, however, is a validated tool based on a number of clinicopathological features that gives estimates for outcome at 10 years according to therapy used [16
]. It is seen as a more sophisticated alternative with stronger predictive power and is now widely used to assist in making decisions regarding adjuvant treatment. We therefore chose to use AOL to compare with 21-gene assay.
Our study has limitations. First, there are little data regarding the test performances of 21-gene assay and AOL. We used the data from the only study available and had to estimate the plausible ranges for sensitivity analysis, which might have influenced the robustness of our model. Second, the data we adapted from Bryant et al. grouped the intermediate-risk patients together with the high-risk patients in the RS category. There are currently no data as to whether chemotherapy is beneficial in the intermediate-risk group, and this is the subject of an ongoing phase III trial (TAILORx). In addition, AOL does not provide risk categories. Bryant ranked order output from AOL so that a similar proportion of cases would be categorized as low risk (50%) when compared with 21-gene assay. This method is arbitrary and may not be optimal in identifying patients who are likely to benefit from chemotherapy. Third, we did not consider local recurrence as a health state, and hence might have missed out accounting for its associated cost and utility loss. None of the models in previous studies on 21-gene assay included this state because of the lack of data in validation studies. Fourth, this model did not consider HER-2–positive disease, which represents up to 25% of early breast cancer. It is assumed that patients with HER-2–positive disease are generally advised to receive combined chemotherapy-trastuzumab treatment.
Our study modeled based on the Canadian health system. It would be most interesting to see such analysis performed in the U.S. setting. Because of the likely differences in the practice patterns including patient follow-up, treatment for metastatic disease, and the difference in health care structure, it is anticipated that the estimates of the incremental cost-effectiveness ratio would be even more favorable.