Cost–Utility Analysis of Multigene Assays to Guide Treatment Decisions for Node-Negative Early Breast Cancer

Abstract

Introduction

Clinicopathologic and patient factors, such as tumor grade, size, age, and menopausal status, provide limited prognostic and predictive information in hormone receptor positive (HR +), human epidermal growth receptor 2 negative (HER2−), node-negative early-stage breast cancer, leading to potential over- or under-treatment. Multigene expression profile tests used in clinical practice in the USA, including the 21-gene assay, 70-gene assay, 12-gene assay, and 50-gene assay, offer prognostic information beyond traditional clinicopathologic features to improve treatment decisions. This study aimed to estimate the cost-effectiveness of these four multigene assays compared with clinicopathologic risk assessment alone.

Methods

A decision tree categorized hypothetical patients with HR + /HER2− early-stage invasive breast cancer according to clinical and genomic risk, and integrated clinical expert insights for chemotherapy allocation with literature inputs. A Markov model simulated lifetime costs and outcomes of chemotherapy decisions over a patient’s lifetime. The probability of distant breast cancer recurrence was derived from TAILORx (21-gene assay), MINDACT (70-gene assay), and TransATAC (12-gene assay, 50-gene assay) studies. Costs were calculated from a US societal perspective in 2021 US dollars, considering healthcare costs, lost productivity, and patient out-of-pocket expenses.

Results

The 21-gene assay and 50-gene assay were less costly ( −$12,189 and  −$2410, respectively) and more effective [0.23 and 0.07 quality-adjusted life years (QALYs), respectively] compared with clinicopathologic risk alone. Similarly, the 70-gene assay and 12-gene assay are also cost-effective alternatives [incremental cost-effectiveness ratio (ICER): 27,760 and 7942, respectively].

Conclusions

All four multigene assays were cost-effective from a societal perspective, offering low net lifetime costs or savings with improved outcomes compared with clinicopathologic risk assessment alone. These assays can help refine treatment decisions by providing prognostic risk estimates. In the case of the 21-gene assay, it can also predict chemotherapy benefit leading to the highest lifetime cost savings and greatest QALY gain.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40487-024-00312-4.

Key Summary Points

Why carry out this study?

Clinicopathologic and patient factors, such as tumor grade, size, age, and menopausal, status provide limited prognostic and predictive information in hormone receptor positive (HR +), human epidermal growth receptor 2 negative (HER2−), node-negative early-stage breast cancer, leading to potential over- or undertreatment. Multigene expression profile tests used in clinical practice in the USA, including the 21-gene assay, 70-gene assay, 12-gene assay, and 50-gene assay, offer prognostic information beyond traditional clinicopathologic features to improve treatment decisions.

This study aimed to estimate the cost-effectiveness of multigene expression profile tests to guide adjuvant chemotherapy decisions in HR + and HER2− early breast cancer from a US societal perspective.

What was learned from the study?

In this modelling study, all four multigene assays were found to be cost-effective compared with using clinicopathologic risk assessment alone by adding prognostic, and in the case of the 21-gene assay, predictive information beyond patient and tumor characteristics, which leads to better treatment decisions.

The results were primarily driven by reductions in medical costs and productivity losses when multigene tests were used, along with the avoidance of distant recurrence. An additional benefit was seen with the 21-gene assay, which not only predicts distant recurrence risk but also identifies which patients are likely to benefit from chemotherapy.

Introduction

The current standard treatment of hormone receptor positive (HR +) and human epidermal growth factor receptor 2 negative (HER2−) early breast cancer following breast-conserving surgery involves adjuvant therapy with endocrine therapy (ET) or chemoendocrine (CET) therapy [1, 2]. Traditional assessment of clinicopathologic features, such as tumor grade and size, and nodal involvement alongside patient factors including age and menopausal status, often provide incomplete prognostic information and have limited ability to predict chemotherapy benefit for patients with HR + /HER2−, axillary lymph node-negative (N0) early-stage breast cancer. This can lead to both under- and overtreatment with adjuvant chemotherapy [3].

Gene expression profiling has led to the identification of the human invasion signature and discovery of signaling pathways that are upregulated in the intravasation and eventual metastasis of breast tumor cells. Multigene expression profile tests supplement clinicopathologic information and permit more individualized treatment decisions [4]. Evidence has shown a strong association between the presence of genes coding for members of these pathways and distant metastasis in HR + /HER2− breast cancer [4].

Commercially available assays in the USA include the 21-gene assay (the Oncotype DX Breast Recurrence Score test, Exact Sciences, Madison, US), 70-gene assay (MammaPrint, Agendia, Amsterdam, Netherlands), 12-gene risk score (EndoPredict, Myriad Genetics, Salt Lake City, US), and 50-gene assay (Prosigna risk of recurrence, PAM50, Veracyte, South San Francisco, USA). These assays offer prognostic information beyond traditional clinicopathologic features to inform treatment decisions. However, to date, the 21-gene assay is the only multigene assay that has been shown to predict chemotherapy benefit in node-negative patients [5]; this is supported by National Comprehensive Cancer Network [3] and American Society of Clinical Oncology guidelines [6].

Previous economic evaluations have shown the cost-effectiveness of multigene tests for guiding adjuvant therapy for patients with early-stage breast cancer, especially among those in high-risk groups identified by clinicopathologic factors [7–11]. However, economic evaluations conducted solely from a healthcare payer perspective ignore broader societal costs including out-of-pocket costs [12] and loss of productivity, which can be substantial [13, 14].

To address this knowledge gap, this study aimed to evaluate the cost-effectiveness of four multigene assays compared with clinicopathologic risk alone to guide adjuvant chemotherapy decisions for HR + /HER2− early-stage N0 breast cancer patients from a US societal cost perspective.

Methods

Study Population, Intervention, and Comparators

The model simulated a cohort of patients aged 55 years and older (the median age of the population in the TAILORx study [15]) with N0 HR + /HER2− early-stage invasive breast cancer. The population was categorized into groups according to the results for each test (Table S1 in the electronic supplementary material) [15–17].

The model compared each tumor profiling test (21-gene assay, 70-gene assay, 12-gene assay, or 50-gene assay, further detail in Table S1 in the electronic supplementary material), with clinicopathologic risk assessment alone, which involves evaluating prognostic factors in the absence of genomic tumor profiling (i.e., tumor grade and size, patient age, and menopausal status) to estimate recurrence risk. Given the variability in patient characteristics across studies, and absence of a single study covering all tests, direct comparisons between tests were considered inappropriate. Consequently, each test was compared with clinicopathologic risk alone. Genomic risk score distribution was the same for each pairwise comparison, reflecting consistent underlying genomic risk.

Model Structure

The model consisted of an initial decision tree, followed by a health-state time-dependent discrete-state transition (Markov) cohort model with 6-month cycles, developed in Microsoft Excel. This model structure has been validated in a previously published analyses [18, 19]. In the decision tree (Fig. 1), it was assumed that all patients underwent multigene testing in addition to clinical risk assessment and were assigned to a risk score/group category on the basis of the test results. This additional information influences the probability of receiving either CET or ET alone. Following treatment, the subsequent natural history of breast cancer was then simulated using the Markov model (Fig. 2). Patients could transition to health states representing recurrence-free, distant recurrence, or death. A proportion of patients was assumed to have had a local recurrence prior to experiencing a distant recurrence. Chemotherapy increased the risk of acute myeloid leukemia (AML) and congestive heart failure (CHF). Patients assigned to chemotherapy can move to the AML/CHF health state at any point before or after disease progression. Since the effectiveness of the received treatment was not factored in for these patients, it was assumed that once they enter this state, the only possible transition is to the death state.

Fig. 1.

Decision-tree component of the model. In the gene-assay alternative of the model, chemotherapy assignment was conditional on the assigned genomic risk subgroup according to each test. In the clinicopathologic risk alternative of the model, it was conditional on age and clinical risk. Once patients have been assigned their genomic risk result and assigned adjuvant treatment, they enter the Markov portion of the model

Fig. 2.

Markov component of the model. The model included five health states; the arrows depict patient movement between health states in each model cycle. Patients could move to death from any health state. Patients entered the Markov portion in the recurrence-free health state; the probability of transition to distant recurrence, acute myeloid leukemia (AML), or congestive heart failure (CHF) was conditional on the assigned adjuvant treatment, clinical risk, and genomic risk category (if known)

Modeled costs included Medicare, out-of-pocket expenses, and indirect costs, based on a US societal perspective. The model used a lifetime horizon with 6-month cycles [7, 8], and discounted costs and benefits annually at 3.0% in line with US guidelines [20].

Model Parameters

Transition Probabilities

The patient distribution according to the 21-gene assay recurrence score (RS) result and probability of receiving adjuvant therapy (Table 1) were derived from multiple sources: expert elicitation, data from the UK Breast Cancer Group (UKBCG) [16] and published studies [21–23]. The distribution of genomic risk scores for clinicopathologic risk alone represents the theoretical distribution of underlying genomic risk, which is unknown. Consequently, we assume that the distribution mirrors the observed distribution for patients tested with one of the four assays. The probability of being prescribed adjuvant chemotherapy, either on the basis of clinicopathologic risk alone or after using one of the multigene assays, was derived from a previously reported clinical expert survey [24] and literature (Table 1). Recognizing that chemotherapy use may differ by subpopulation, clinical risk and age-specific data were obtained on the basis of the subgroup definitions used in the TAILORx study for the 21-gene test [15] and clinical risk subgroups according to the MINDACT study for the 70-gene assay [17, 25].

Table 1.

Base case transition probabilities

Parameter	Low risk	Intermediate risk	High risk	Source
Proportion of patients assigned to each risk category, patients with node-negative tumors
21-gene assay	0.1700	0.6900	0.1400	[15]
70-gene assay	0.7500		0.2500	[17]
12-gene assay	0.7387		0.2613	[16]
50-gene signature	0.5503	0.2955	0.1542	[16]
Proportion of patients who receive chemotherapy
21-gene assay	0.0439	0.2168	0.8839	Experts
70-gene assay	0.0605		0.8682	[16, 22]
12-gene assay	0.0670		0.7700	[16, 22]
50-gene signature	0.0590	0.2430	0.9300	[16, 23]
Clinicopathologic risk alone	0.2654			Experts
Probability of distant recurrence for those treated with endocrine therapy
21-gene assay	0.0029	0.0032	0.0251	[15, 30]
70-gene assay	0.0041		0.0093	[25]
12-gene assay	0.0030		0.0111	[31]
50-gene signature	0.0020	0.0065	0.0206	[16]
Probability of distant recurrence for those treated with chemoendocrine therapy
21-gene assay	0.0029	0.0029	0.0068	[15, 30]
70-gene assay	0.0035		0.0073	[25]
12-gene assay	0.0023		0.0084	[31]
50-gene signature	0.0015	0.0049	0.0157	[16]

Probabilities of distant recurrence, determined by the risk categories of each test, were extracted from literature identified via an updated systematic literature review and targeted searches (Table 1). For patients classified using the 21-gene assay, the probability of distant recurrence with CET or ET was based on 12-year distant recurrence-free intervals reported in the TAILORx study [15]. The TAILORx trial randomized patients with intermediate genomic risk (RS 11–25) to CET or ET. To estimate distant recurrence-free intervals for chemotherapy for patients with RS < 11 and for endocrine therapy alone for patients with RS > 25, hazard rates were adjusted using hazard ratios from the NSABP B-20 study [5].

The probability of distant recurrence for patients classified using the 70-gene assay was derived from 8-year follow-up data in the MINDACT study [25]. Probabilities for 12-gene and 50-gene assays were based on a bespoke analysis of TransATAC data reported in the appraisal by the National Institute for Health and Care Excellence in the UK (henceforth referred to as DG34) [16]. Following the approach in DG34, the hazard ratio of 0.76 was applied to all patients assigned to chemotherapy, considering the TransATAC study assigned all patients to endocrine therapy alone.

The incidence of adverse events (AEs) was obtained from the RxPONDER study [26], probability of CHF from Wang et al. [7], and of AML from Petrelli et al. [27]. Given that these AEs are linked to the use of anthracyclines [28, 29], their probability may be related to the distribution of chemotherapy regimens used in clinical practice in the USA, which is uncertain and was tested using scenario analyses.

Mortality

Background mortality was obtained from 2017 US life tables for women [32]. No excess mortality was applied to the recurrence-free state. The probability of death after distant recurrence was derived from the MONARCH 2 trial [33]. The 6-month probability of death after developing AML and CHF was obtained from literature [7, 34].

Costs

This analysis included costs for drug acquisition, administration, follow-up visits, and monitoring. Costs were sourced from literature and reports from the Centers for Medicare and Medicaid Services (CMS) (see Table 2 and further details in Appendix 1 and Tables S2-S12 in the electronic supplementary material). All costs were measured in US dollars and inflated to 2021 prices using Organization for Economic Co-operation and Development indices [35].

Table 2.

Overview of settings, costs, and utilities in the model

Parameter	Base case	Sensitivity analysis	Source
Discount rate costs	3.00%	1.00%–5.00%	[20]
Discount rate outcomes	3.00%	1.00–5.00%	[20]
MGA costs
21-gene assay	$3873	Not reported	[36]
70-gene assay	$3873	Not reported	[36]
12-gene assay	$3873	Not reported	[36]
50-gene signature	$2510	Not reported	[36]
Chemotherapy costs
AC followed by P	$20,423	Not reported	[37–44]
AC followed by weekly P	$14,849	Not reported	[37–44]
TC	$7365	Not reported	[37–44]
EC	$20,930	Not reported	[37–44]
AC followed by T	$13,225	Not reported	[37–44]
TAC	$14,351	Not reported	[37–44]
AC	$11,153	Not reported	[37–44]
CMF	$5870	Not reported	[37–44]
Weekly P + carboplatin	$13,290	Not reported	[37–44]
T + carboplatin	$12,751	Not reported	[37–44]
Endocrine therapy costs	$127	Not reported	[37–44]
Adverse events costs (weighted average)	$1230	Not reported	[7]
Health states
Recurrence-free (annual)
Year 1	$754	603–905	[8]
Year 2	$681	545–817	[8]
Year 3	$608	486–729	[8]
Year 4	$535	428–642	[8]
Year 5 +	$462	369–554	[8]
Local recurrence	$26,235	20,988–31,483	[7]
Distant recurrence disease management	$45,028	Not reported	[45–47]
AML	$21,349	17,080–25,619	[7]
CHF one-off	$35,769	28,616–42,923	[7]
CHF (ongoing per six-month cycle)	$3424	2739–4109	[7]
Terminal care	$31,394	25,115–37,673	[8]
Indirect costs
OOP	$3460	Not reported	[12]
Productivity lost—Adjuvant chemotherapy	$7373	Not reported	[13, 48]
Productivity lost—distant recurrence	$17,049	Not reported	[13, 48]
Utilities
Recurrence-free	0.8240	0.7850–0.8570	[49]
Distant recurrence	0.6850	0.6200–0.7350	[49]
AML	0.2600	0.2080–0.6600	[34, 50]
CHF	0.7100	0.6206–0.7994	[51]
Local recurrence (decrement)	−0.1080	0.0280–0.1880	[52]
Chemotherapy administration (decrement)	−0.0380	0.0304–0.0456	[52]

AC doxorubicin and cyclophosphamide, AML acute myeloid leukemia, CHF congestive heart failure, CMF cyclophosphamide, methotrexate, and 5-fluorouracil, EC, epirubicin and cyclophosphamide, MGA multigene assay, OOP out-of-pocket, P paclitaxel, T docetaxel, TAC docetaxel, doxorubicin, and cyclophosphamide, TC docetaxel and cyclophosphamide

Health-Related Quality of Life

Health-related quality of life (HRQoL) utility values for each health state and decrements representing one-off decreases in HRQoL associated with chemotherapy AEs and local recurrence were derived from existing literature [16, 34, 49, 51, 52]. Further details on these utility inputs can be found in Table S13 in the electronic Supplementary Material.

Base-Case Analysis

The analysis was based on pairwise comparisons between each of the four genomic tests and clinicopathologic risk alone. An incremental cost-effectiveness ratio (ICER), which measures the additional cost per QALY gained was estimated. Additionally, net monetary benefit, the percentage change in chemotherapy use, and change in years spent in distant recurrence were all reported in the electronic supplementary material.

Sensitivity Analysis

Parameter uncertainty was tested using deterministic sensitivity analysis (Table 2 and Table S14 in the electronic supplementary material). The results are presented in a tornado diagram using the metric of net monetary benefit (Figure S1 in the electronic supplementary material). For the probabilistic analysis, gamma distributions were used for cost calculations, beta distributions for probabilities and utilities, and log-normal distributions for hazard ratios associated with distant recurrence post-chemotherapy. Parameters were randomly sampled from these distributions across 5000 simulations. The results were summarized using the cost-effectiveness plane and cost-effectiveness acceptability curves (CEAC). Univariate and probabilistic analyses were run with a willingness-to-pay per QALY threshold ranging from $50,000 to $100,000 per additional QALY [53].

Scenario analyses explored the impact of structural assumptions and parameter value choices, including alternative risk score distribution sources, chemotherapy probabilities, reduction in distant recurrence rates, chemotherapy benefit, costs, and a US Medicare perspective (Table S17 in the electronic Supplementary Material).

Ethical approval

This research is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

Results

Base-Case Results

The results indicated that the 21-gene assay, 70-gene assay, 12-gene assay, and 50-gene assay were cost-effective when compared with standard clinicopathologic risk assessment for informing adjuvant treatment decisions in N0 early breast cancer (Table 3). The 21-gene assay and 50-gene assay were dominant (less costly and more effective) compared with clinicopathologic risk alone, yielding savings of $12,189 and $2410, respectively. The 70-gene assay and 12-gene assay incurred incremental costs of $1079 and $392, respectively. Treatment decisions based on the 21-gene, 70-gene, 12-gene, and 50-gene assays resulted in QALY gains of 0.23, 0.04, 0.05, and 0.07, respectively. A breakdown of the model results is shown in Table S15 and S16 in the electronic supplementary material.

Table 3.

Expected costs, QALYs, life-years, and ICERs

Variable	Comparison 1		Comparison 2		Comparison 3		Comparison 4
	21-gene assay	Clin. path. risk	70-gene assay	Clin. path. risk	12-gene assay	Clin. path. risk	50-gene signature	Clin. path. risk
Total expected cost ($)	81,656	93,485	97,321	96,242	90,509	90,117	96,341	98,751
Incremental cost ($)	−12,189	Reference	1079	Reference	392	Reference	−2410	Reference
Total expected QALYs	13.95	13.72	13.75	13.71	13.83	13.78	13.73	13.65
Incremental QALYs	0.23	Reference	0.04	Reference	0.05	Reference	0.07	Reference
ICER ($)	Dominant	Reference	27,760	Reference	7,942	Reference	Dominant	Reference
NMB ($)	35,226	Reference	2808	Reference	4544	Reference	9768	Reference

Clin. path clinicopathologic, ICER incremental cost-effectiveness ratio, NMB net monetary benefit, QALY quality-adjusted life-year

One-Way Sensitivity Analyses

The model results were most sensitive to the parameters related to the probability of distant recurrence, percentage of patients assigned to chemotherapy, discount rates, cost of genomic tests, and utilities related to patients in the recurrence-free health state (Figure S1 in the electronic supplementary material).

Scenario Analyses

Full results of scenario analyses considering the comparison of each assay compared with using clinicopathologic risk alone are reported in Table S17 in the electronic supplementary material. The 21-gene assay remained dominant in all the scenarios considered. The 70-gene assay was cost-effective in all scenarios, except when the probability of chemotherapy in the absence of multigene testing was low. The 12-gene assay remained cost-effective across all scenarios, except when the hazard ratio for chemotherapy benefit was set to its upper limit of 0.99. The 50-gene assay remained dominant across all scenarios, except when the chemotherapy benefit hazard ratio was set to its upper limit of 0.99. In this case, it was no longer dominant but remained cost-effective.

Probabilistic Analyses

The results of the probabilistic analysis were broadly consistent with the results of the deterministic analysis (Table S18, scatterplots in Fig. S2, cost-effectiveness acceptability curves in Fig. S3 in the electronic Supplementary Material). On the basis of a willingness-to-pay threshold of $50,000 per QALY, the probability of the 21-gene, 70-gene, 12-gene, and 50-gene assays of being cost-effective compared with clinicopathologic risk alone were 98.66%, 61.32%, 75.70%, and 91.42%, respectively. For a willingness-to-pay threshold of $100,000 per QALY, the probability of the 21-gene, 70-gene, 12-gene, and 50-gene assays of being cost-effective compared with clinicopathologic risk alone were 98.66%, 61.32%, 75.70%, and 91.42%, respectively. Alternative PSA results are presented in Fig. S4 and associated text in electronic Supplementary Material to test the inclusion of the hazard ratio for chemotherapy benefit for patients with RS results of 26–100, which was the largest source of uncertainty in the model.

Discussion

Multigene assays appeared cost-effective options for guiding chemotherapy decisions in N0 HR + /HER2− early-stage invasive breast cancer from a US societal perspective when compared with using clinicopathologic risk alone. Reduced risk of distant recurrence led to cost reductions and increased HRQoL. This was consistent across sensitivity analyses.

At WTP thresholds of $100,000 per QALY and the lower bound of $50,000 per QALY, both the 21-gene assay and 50-gene assay were dominant (less costly and better outcomes) compared with clinicopathologic risk assessment alone; however, the 21-gene assay demonstrated the highest lifetime cost savings (−$12,189) and greatest QALY gain (0.23).

The 21-gene assay is the only multigene assay validated by clinical trials for predicting chemotherapy benefit [5]. This means that the relative treatment effect of chemotherapy differed according to RS subgroup in the model. For the 70-gene test, the effect of chemotherapy was derived directly from the subgroups that were randomized to both CET and ET alone in the MINDACT trial, and a constant treatment effect applied for other subgroups. The 50-gene and 12-gene assays were assumed to be prognostic-only, and a constant treatment effect was assumed for all patients assigned to chemotherapy according to the approach used in DG34. Larger cost savings and QALY gains for the 21-gene assay were generated from avoidance of distant recurrence, as the 21-gene assay can predict which patients benefit from chemotherapy (represented by relative treatment effect), in addition to giving information on the absolute risk of distant recurrence.

Previous studies have reported multigene tests as cost-effective options to guide adjuvant therapy decisions for patients with early-stage breast cancer [9–11]. However, the cost-effectiveness of these tests varies between patient subgroups, with patients at high risk of recurrence based on clinicopathologic factors alone benefiting the most [9]. Additionally, a more consistent alignment of results has been noted in N0 breast cancer patients across studies [9].

The cost-effectiveness of the 21-gene assay in the US has been supported by economic evaluations using PREDICT for clinicopathologic risk categorization [7, 8]. Kunst et al. classified patients into three subgroups on the basis of the absolute 10-year survival benefit from chemotherapy using the PREDICT tool: < 3% for low clinical risk, 3–5% for intermediate clinical risk, and over > 5% for high clinical risk. The findings indicated that the 21-gene assay is cost-effective for patients with intermediate and high clinical risk but not for those with low clinical risk [8]. The present analysis incorporates the latest evidence from phase 3 randomized controlled trials to inform the long-term effectiveness of chemotherapy decisions, offering a more up-to-date perspective.

Studies in Japan and the UK showed the 70-gene assay is a cost-effective option from societal and health service perspective compared with current practice, though not consistently across all analysis scenarios, highlighting gaps in the evidence [9, 54]. Hall et al. concluded that the 50-gene assay has > 80% probability of being cost-effective from a UK National Health Service perspective [55]. A Canadian study concluded that genomic profiling using the 21-gene and 70-gene assays in early-stage invasive breast cancer is a cost-effective alternative to the standard of care from a healthcare payer perspective [56].

Comparing the findings of this analysis with those from existing studies presents challenges, primarily because of different population stratification criteria. The TransATAC study is the only study that directly compared the 21-gene, 12-gene, and 50-gene assays in the same cohort. There is a notable gap in the literature concerning direct comparisons between the 21-gene assay and 70-gene assay. For this reason, a direct comparison between tests was not considered appropriate. Merging data from different studies of multigene assays that classify genomic risk differently could lead to bias, affecting the reliability of the analysis.

Previous analyses evaluating multigene assays have faced criticism for overlooking the influence of clinicopathologic factors [57]. The present analysis specifically targeted N0 HR + /HER2− breast cancer. Stratification of chemotherapy assignment before and after using the 21-gene assay according to age (older or younger than 50 years) and clinical risk (low or high) was conducted to overcome this limitation of previous studies.

There is a significant challenge in adopting a broader perspective owing to the limited data available to inform a cost-effectiveness threshold. Although the analysis focused on a societal perspective, the same thresholds were used as there is no established or widely accepted cost-effectiveness threshold for the US societal perspective. Common thresholds for the healthcare perspective are $50,000, $100,000, and $150,000 per QALY gained [58]. Under the assumption that the healthcare budget is fixed, a broader perspective would typically be associated with a lower threshold. Therefore, the acceptability curve was reported across a range of willingness-to-pay (WTP) thresholds, up to $100,000, and results were specifically provided for WTP thresholds of $50,000 and $100,000. In scenarios using a narrower perspective, the NMB decreased slightly in all cases, but multigene tests remained cost-effective options.

This study has several limitations. The cost-effectiveness analysis required merging evidence from multiple studies for specific patient subgroups. For example, to estimate chemotherapy benefit for patients with RS > 25, data from the NSABP B-20 clinical trial were used. Although the use of these data did not change study conclusions when tested using scenario analyses, the wide CI for this parameter introduced substantial uncertainty into the probabilistic sensitivity analysis (Fig. S3 in the electronic supplementary material). To estimate the probability of distant recurrence for patients assigned to CET after being tested with the 12-gene and 50-gene assays, the probability of distant recurrence risk for patients assigned to ET informed by the TransATAC study was adjusted using the treatment effect estimate from the EBCTCG meta-analysis [risk ratio (RR) = 0.76 reported in DG34] [16, 59]. Uncertainty around this value was addressed in the sensitivity analysis, which did not alter the findings [59].

The probability of chemotherapy according to RS category was based on clinical expert opinion, in the absence of recently published decision impact studies in the USA. The estimated proportion of patients allocated to chemotherapy varied across subgroups, highlighting variability in US clinical practice. It is important to note that the Surveillance, Epidemiology, and End Results (SEER) studies included patients diagnosed before 2016 and therefore did not capture changes in clinical practice following the publication of the TAILORx results. Alternative inputs, based on studies informed by the SEER registry [21, 60], were tested in scenario analyses, with no changes observed in the study’s primary findings. In contrast, for the other tests, the probability estimates for chemotherapy were primarily sourced from DG34 [16]. However, because no reported variability measures were available, an arbitrary range of ± 20% was applied. Consequently, this approach may lead to underestimating of the uncertainty surrounding the cost-effectiveness results for the additional multigene tests evaluated in this study.

In the model, all patients assigned to chemotherapy were presumed to carry an increased risk of AML and CHF. There is evidence that these latent effects are primarily linked to the use of anthracycline chemotherapy, which may have led to an overestimate of long-term AE risk in the model. The impact of this assumption was examined by setting the probability of AML and CHF to 0, with no change to the study conclusions. The utility value for AML was based on a previous HTA [9], and was originally reported by Barr et al.[61], which provided a utility for patients with AML in second complete remission. In alignment with this, it was assumed that patients remain in this health state until death. Although this value was lower than those reported in more recent economic evaluations owing to the availability of new treatments, it was considered appropriate as it likely reflected a mix of patients in both pre- and postprogression states, resulting in a lower average. To explore the impact of an alternative source, a higher utility value from a more recent study by Tremblay et al. [50] was applied as the upper bound in the sensitivity analysis. Minimal impact on the model results was observed from this change.

Conclusions

The present study found that the 21-gene and 50-gene assays are dominant (less costly and better outcomes) compared with using clinicopathologic risk alone to guide adjuvant chemotherapy decisions for women with N0, HR + , and HER2− early breast cancer. The 70-gene and 12-gene assays were also found to be cost-effective options. This study adds to the existing evidence base on the cost-effectiveness of tumor profiling assays from a US societal perspective, with potential implications for the use of testing in clinical practice, and value-based decision making for genomic testing in the USA. The results of this study should be considered in light of the uncertainty in the parameter inputs influenced by data gaps requiring synthesis from multiple studies, this has been described earlier in this article. The uncertainty of the results could be reduced using future real-world evidence and decision impact studies examining the impact of multigene assays on chemotherapy decisions in the USA, given emerging clinical trial evidence on the treatment effect of chemotherapy guided by a combination of clinicopathologic factors and information from tumor profiling assays.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Vladislav Berdunov and Gebra Cuyún-Carter were responsible for the study conception and study design; Vladislav Berdunov and Yaneth Gil Rojas performed model programming and analysis; Yara Abdou advised the study team as an expert in breast cancer; Christy Russell, Sara Campbell, and Jennifer Racz provided guidance toward the collection and validation of specific inputs, as well as in the review and editing of the manuscript. All authors contributed to the first and subsequent drafts of the manuscript. All authors read and approved the final manuscript.

Funding

Sponsorship for this study and Rapid Service Fee were funded by Exact Sciences.

Data Availability

The parameter inputs used in the model were identified from sources in the open domain or peer-reviewed journal articles. All parameter values and their sources are reported in the article and the electronic Supplementary Material.

Declarations

Conflict of Interest

Vladislav Berdunov was an employee of Putnam during the development of the research and received funding from Exact Sciences. His current affiliation is Evidera. Gebra Cuyún-Carter is an employee and stockholder of Exact Sciences. Yaneth Gil Rojas is an employee of Putnam and have received funding from Exact Sciences. Christy Russell is an employee and stockholder of Exact Sciences. Sara Campbell is an employee and stockholder of Exact Sciences. Jennifer Racz is an employee and stockholder of Exact Sciences. Yara Abdou served as an uncompensated consultant for Exact Sciences.

Ethical Approval

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.