Cost Effectiveness of Gene Expression Profile Testing in Community Practice

Abstract

Purpose

Gene expression profile (GEP) testing can support chemotherapy decision making for patients with early-stage, estrogen receptor–positive, human epidermal growth factor 2–negative breast cancers. This study evaluated the cost effectiveness of one GEP test, Oncotype DX (Genomic Health, Redwood City, CA), in community practice with test-eligible patients age 40 to 79 years.

Methods

A simulation model compared 25-year societal incremental costs and quality-adjusted life-years (QALYs) of community Oncotype DX use from 2005 to 2012 versus usual care in the pretesting era (2000 to 2004). Inputs included Oncotype DX and chemotherapy data from an integrated health care system and national and published data on Oncotype DX accuracy, chemotherapy effectiveness, utilities, survival and recurrence, and Medicare and patient costs. Sensitivity analyses varied individual parameters; results were also estimated for ideal conditions (ie, 100% testing and adherence to test-suggested treatment, perfect test accuracy, considering test effects on reassurance or worry, and lowest costs).

Results

Twenty-four percent of test-eligible patients had Oncotype DX testing. Testing was higher in younger patients and patients with stage I disease (v stage IIA), and 75.3% and 10.2% of patients with high and low recurrence risk scores received chemotherapy, respectively. The cost-effectiveness ratio for testing (v usual care) was $188,125 per QALY. Considering test effects on worry versus reassurance decreased the cost-effectiveness ratio to $58,431 per QALY. With perfect test accuracy, the cost-effectiveness ratio was $28,947 per QALY, and under ideal conditions, it was $39,496 per QALY.

Conclusion

GEP testing is likely to have a high cost-effectiveness ratio on the basis of community practice patterns. However, realistic variations in assumptions about key variables could result in GEP testing having cost-effectiveness ratios in the range of other accepted interventions. The differences in cost-effectiveness ratios on the basis of community versus ideal conditions underscore the importance of considering real-world implementation when assessing the new technology.

INTRODUCTION

Gene expression profile (GEP) tests, such as Oncotype DX (Genomic Health, Redwood City, CA), have been recommended for use to support treatment decision making for patients with early-stage, node-negative, estrogen receptor (ER)–positive, human epidermal growth factor 2 (HER2)–negative cancers.^1-3 The primary goal of GEP testing is to identify patients at high recurrence risk who will benefit from chemotherapy, while allowing patients with a low recurrence risk to forego chemotherapy, potentially offsetting the test costs with savings from reductions in chemotherapy use.

To date, use of GEP testing in community practice remains low, ranging from 22% to 42% of test-eligible patients.^4-9 Moreover, chemotherapy use is sometimes discordant with test results, with 17% to 26% of patients with high recurrence risk scores not receiving chemotherapy and 8% of patients with low recurrence risk scores receiving chemotherapy.¹⁰

Prior economic analyses of GEP evaluated hypothetical cohorts under ideal conditions and concluded that it had low costs relative to its benefits.^11-17 However, given the divergence of community testing and chemotherapy use from the ideal, it is possible that the expected clinical and economic benefits of GEP are not being fully realized. In this study, we conducted an analysis of the likely cost effectiveness of Oncotype DX testing on the basis of community practice patterns.

METHODS

We constructed a discrete-time state transition simulation model to estimate the likely incremental costs per quality-adjusted life-year (QALY) of community use of Oncotype DX testing versus usual care from a societal perspective. The Georgetown University Oncology and Kaiser Permanente Northern California (KPNC) institutional review boards approved the research.

Intervention and Patients

We selected Oncotype DX because it is the most commonly used GEP test in the United States¹⁰ and the primary focus of prior economic analyses.^11-17 The population included test-eligible patients age 40 to 79 years diagnosed with stage I or IIA, ER-positive, HER2-negative breast cancer between 2000 and 2012. Costs and effects for patients diagnosed from 2000 to 2004 (ie, pre–Oncotype DX period, usual care) were compared with those among patients diagnosed from 2005 to 2012 (ie, period when Oncotype DX testing was used in community practice).

Model Overview

The model was developed using TreeAge Pro 2015 (TreeAge Software, Williamstown, MA). The model captured a 25-year time horizon from diagnosis because the median age of diagnosis is > 60 years¹⁸ and almost all distant recurrences (and deaths from recurrences) occur within 25 years of diagnosis.¹⁹ Events (eg, chemotherapy use or toxicity) were tallied at 1-year transition intervals. The model decision pathways and health states are summarized in Figures 1A and 1B, respectively.

Fig 1.

Decision tree and state transitions for patients with stage I or IIA, node-negative, estrogen receptor (ER)–positive, human epidermal growth factor receptor 2 (HER2)–negative breast cancer. (A) Simulation model schema. The model was developed to compare cost effectiveness of community practice with use of Oncotype DX test versus usual care without the test. The community practice arm included observed testing and chemotherapy use in 2005 to 2012. The usual care arm included chemotherapy use patterns in the pre–Oncotype DX era (2000 to 2005). Testing probabilities were conditional on age and stage. The test results affected the probability of chemotherapy use. (B) State transition. All simulated patients were newly diagnosed with ER-positive, HER2-negative, node-negative, stage I or IIA breast cancer. If death from chemotherapy toxicity did not happen at initial treatment, all simulated patients transitioned to the post-treatment state until breast cancer death (if recurrence occurred) or death from other cause (if no recurrence occurred or death from other cause occurred before recurrence). Patients without distant recurrence only died of non–breast cancer causes or chemotherapy toxicity. Patients remain in the same state until the time of a transition event.

Briefly, the model began with the generation of simulated patients with breast cancer by age and stage on the basis of national incidence rates. In the usual care scenario, patients could receive chemotherapy or not on the basis of their age and stage. In the Oncotype DX testing period, patients were tested or not, and received chemotherapy on the basis of age, stage, and test use and results. If recurrence occurred, it was assumed to progress to breast cancer death within 25 years, unless death occurred earlier as a result of chemotherapy toxicity or other causes. Without recurrence, patients died of other causes, and if chemotherapy was received, it did not provide benefit but could have resulted in toxicity.

Tracking variables built into the model were used to tally starting age, current age, recurrence status, testing status, test result, chemotherapy use, toxicity grade, and cause of death. These tracking variables were used in postprocessing analyses using SAS 9.4 (SAS Institute, Cary, NC), including calculation of life-years for each simulated patient, application of utility weights, cost allocation to each event, and discounting of costs and effects.

Input Parameters

Model inputs used to estimate costs and effects were derived from national data, published research, and KPNC electronic records linking registry data, treatment, and GEP testing (Table 1). Incidence rates were based on SEER data from 2000 to 2012.²¹ Oncotype DX testing and chemotherapy use rates were based on age- and stage-specific use at KPNC.⁸ The marginal distribution of risk score categories and the probability of recurrence conditional on each risk score category were based on published data.³ Using Bayes’ theorem, these data were used to calculate the probability of having each risk score category conditional on whether or not recurrence occurred, as a measure of the test accuracy.

Table 1.

Model Input Parameters for Estimation of Costs and Effects Among Patients With ER-Positive, HER2-Negative, Lymph Node–Negative, Stage I or IIA Breast Cancer

The underlying age- and stage-specific 25-year breast cancer survival in the absence of treatment of ER-positive, HER2-negative, stage I and IIA cancers was based on prior Cancer Intervention and Surveillance Modeling Network analyses.^22,23 This overall survival was partitioned into survival among those who experienced distant recurrence and those who did not and was used to calculate annual risk of breast cancer death given recurrence status in the absence of any adjuvant treatment.

To isolate the effects of Oncotype DX on chemotherapy-related outcomes, we assumed that 100% of patients received hormonal therapy and that adherence was independent of Oncotype DX testing. Treatment with hormonal therapy alone or hormonal therapy and chemotherapy reduced the risk of death among those destined to have distant recurrences but had no effect on breast cancer mortality among those who would never experience recurrence. Treatment effects were based on the most recent meta-analysis from the Early Breast Cancer Collaborative Trialists’ Group.²⁹ The probability of experiencing chemotherapy toxicity and toxicity grade were based on published trials.^24-28 Non–breast cancer mortality was based on US data.³⁰

Survival was weighted by utility values for each health state to estimate QALYs. Utilities were based on female population age-specific values from the EQ-5D reported on the Medical Expenditure Panel Survey.^31,32 Among those who received chemotherapy, utility was further adjusted for the 6 months of administration.^33-35 Patients experiencing recurrence had further decrements in utility (Table 1).

The costs of the Oncotype DX test ($3,416) were based on the Medicare reimbursement rate.³⁸ Age- and stage-specific cancer care costs were based on published national estimates.^38-40,44 Initial care costs were separated into initial care with and without chemotherapy (including toxicity) on the basis of age- and stage-specific proportions of patients receiving chemotherapy nationally. Costs of treatment of chemotherapy toxicity were assumed to include a short hospitalization and emergency room visits for evaluation of adverse events.²⁵

Patients who experienced a distant recurrence were assumed to incur new chemotherapy costs.² On the basis of a median overall survival after distant recurrence of 36 months,^45,46 recurrence costs included 1 year of chemotherapy, 1 year of continuing care, and 1 year of terminal care. Patients without recurrence incurred continuing care costs until the last year of life; they then incurred terminal care costs on the basis of those of the noncancer population.³⁹

Patient time costs for chemotherapy were based on travel and time for standard regimens.^41,42 Time costs for the treatment of toxicity were based on the average length of a hospital stay (eg, for febrile neutropenia) and/or number of emergency room visits. Patient time was valued using the average 2012 US female hourly wage rate.⁴³ All costs were updated to 2015 US dollars (the most current year available) using the medical care component of the Consumer Price Index.⁴⁷ Future costs and QALYs were discounted at 3%.

Analyses

One hundred million simulations were conducted to reduce Monte Carlo error in the estimation of costs and effects. We calculated the incremental cost-effectiveness ratio for community Oncotype DX test and chemotherapy treatment patterns versus usual care in the pre–Oncotype DX era.

Accounting for Uncertainty

To evaluate the impact of uncertainty, we conducted several one-way sensitivity analyses. First, we examined the impact of test misclassification of distant recurrence by varying results across the upper and lower 95% CIs of the derived test operating characteristics for accuracy. Because there were three possible categories of recurrence risk scores (low, intermediate, and high) conditional on two recurrence possibilities (yes or no), to estimate the least misclassification of low-risk scores given no recurrence, the highest probability of having a low-risk score was combined with the lowest probability for having a high-risk score. For the least misclassification of high-risk given recurrence, the highest probability of having a high-risk score was combined with the lowest probability of having a low-risk score. In each calculation, the total is constrained to equal 1, so that the probability of intermediate risk was 1 minus the sum of the probability of the high-risk and low-risk scores.

Next, we varied the cost of the Oncotype DX test from $2,657 to $4,175 on the basis of the difference (± $759) between the retail price of $4,175 and the Medicare reimbursement rate ($3,416). To assess the impact of perfect patient adherence to test-suggested treatment, 100%, 50%, and 0% chemotherapy use was assumed among those with high-, intermediate-, and low-risk scores, respectively.

Scenario Analyses

We assessed the following two alternative scenarios to the base-case analyses: using the insurer (ie, Medicare) perspective by excluding patient time costs, and including the net impact of Oncotype DX testing on possible reassurance or worry about distant recurrence through further utility weighting. For the latter, we assumed that over the first 2 years after diagnosis, patients with low-risk scores gained 0.05 QALY as a result of a decrease in worry, whereas patients with high-risk scores had a 0.05 reduction in QALYs as a result of increased worry about recurrence.⁴⁸

Finally, we conducted a multiway sensitivity analysis of a scenario with the following idealized conditions: 100% test rates and adherence to test-suggested chemotherapy treatment, best test accuracy, inclusion of the impact of testing on utility, and lowest costs. We did not perform a probabilistic sensitivity analysis because the computational burden exceeded available computing resources.

Model Validation

To evaluate the validity of the model outcomes, the code was verified by confirming that results varied in expected directions using extreme values of parameters. Face validity was evaluated by comparing life-years saved among clinically relevant patient subgroups on the basis of age, stage, recurrence, and chemotherapy use (Appendix Table A1, online only).

RESULTS

Community practice Oncotype DX test and chemotherapy rates between 2005 and 2012 were 24.2% and 30.0%, respectively. Tested patients were younger than nontested patients (mean age, 56.2 years [standard deviation, 8.9 years] v 60.7 years [standard deviation, 10.1 years], respectively) and more likely to have stage I disease than stage II disease (data not shown). Tested patients younger than age 50 years had lower chemotherapy rates than untested patients in the same age group (53.0% v 63.6%, respectively). Among older patients, there was more chemotherapy use among tested than untested patients (age 50 to 64 years: 36.5% v 30.8%, respectively; age ≥ 65 years: 17.6% v 8.2%, respectively; Table 2). These patterns resulted in a greater proportion of tested than untested patients who were destined to have distant recurrences receiving chemotherapy (55.3% v 30.4%, respectively).

Table 2.

Chemotherapy Rates by Age and Oncotype DX Test Use and Recurrence Outcomes in Community Practice With Testing and Usual Care in the Pre–Oncotype DX Era Among Patients with Stage I or IIa, Node-Negative, ER-Positive/HER2-Negative Breast Cancer

Incremental Cost Effectiveness

The incremental cost-effectiveness ratio of breast cancer management using Oncotype DX testing as observed in community practice versus usual care without testing was $188,125 per QALY (Table 3). However, varying the values of several factors changed the results substantially, in several cases decreasing the costs to < $75,000 per QALY (Fig 2 and Appendix Table A2, online only). For instance, if Oncotype DX costs were decreased from current Medicare reimbursement rates of $3,416 to $2,657, then the incremental cost-effectiveness ratio of community practice versus usual care decreased to $71,250 per QALY. If Oncotype DX test properties improved, the incremental cost-effectiveness ratio decreased to $28,947 per QALY. If testing had the worst-case accuracy, testing would be dominated (ie, costs more and saves fewer lives than usual care).

Table 3.

Societal Perspective Incremental Cost-Effectiveness Ratio of Oncotype Testing in Community Practice Versus Usual Care Without Testing Among Patients With Stage I or II, Node-Negative, ER-Positive, HER2-Negative Breast Cancer

Fig 2.

Impact of varying single parameters on the societal cost-effectiveness ratios for Oncotype DX testing in community practice versus usual care without testing among patients with stage I or II, node-negative, estrogen receptor–positive, human epidermal growth factor receptor 2–negative breast cancer. This diagram illustrates the changes in the incremental cost-effectiveness ratio (ICER) for the costs per quality-adjusted life-year (QALY) under various parameter values and alternative assumptions. The solid vertical line represents the base-case ICER result comparing community practice with Oncotype DX testing versus usual care without Oncotype DX. The horizontal bars indicate the change from the base ICER when the one individual parameter is varied. If the bar goes to the right of the base case, it indicates that the alternative value or assumption costs more per QALY than the base case, where bars that go to the left indicate that the alternative value or assumption costs less per QALY than the base case. (*) Test costs were varied from the base case of $3,416 to $2,657 and $4,175. The large difference in cost per QALY when test costs were varied is a result of these effects being magnified by the small incremental QALYs between usual care and community care. (†) The accuracy of the test represents the probability of a test score, conditional on actual distant recurrence. The best test accuracy reflects a greater proportion of women who actually experience recurrence having high-risk scores and a smaller proportion having low-risk scores, and among those who do not experience recurrence, fewer have high-risk scores and more have low-risk scores than in the base case on the basis of observed performance in the original validation study. (‡) The worst testing accuracy was dominated. That is, it resulted in community practice being more costly and producing fewer QALYs than usual care without testing. (§) Net reassurance/worry is based on gaining 0.5 QALY or losing 0.5 QALY over the first 2 years after diagnosis with low- and high-risk recurrence scores, respectively. (‖) One hundred percent adherence to test-guided treatment assumes 100% chemotherapy use among patients with high-risk score on gene expression profile testing, 50% chemotherapy use for intermediate risk, and 0% chemotherapy use for low risk.

Under the assumption that having information about recurrence risk affects utilities via worry or reassurance, the incremental cost-effectiveness ratio for Oncotype DX testing as it occurred in community practice versus usual care was $58,431 per QALY gained (Fig 2 and Appendix Table A2). Adherence to test-concordant treatment lowered the cost-effectiveness ratio to $85,490 per QALY, but the insurers’ perspective had less of an effect on the cost-effectiveness ratio ($207,500 per QALY). Finally, in the multiway scenario analyses of ideal circumstances, the likely cost-effectiveness ratio for Oncotype DX testing would be $39,496 per QALY compared with usual care without testing (not shown).

DISCUSSION

This study evaluated the likely cost effectiveness of Oncotype DX testing as integrated into breast cancer care in community practice versus usual care without testing for patients diagnosed with early-stage, ER-positive, HER2-negative breast cancer. The patterns of Oncotype DX use in community practice suggest that there was selection of patients to testing where results may have been most likely to affect treatment decisions. Although Oncotype DX testing has high costs relative to its benefits as deployed in community practice, realistic variations in assumptions about key variables could result in testing having cost-effectiveness ratios in the range of other generally accepted interventions. The variables that resulted in lower cost-effectiveness ratios for community use of Oncotype DX than seen in the base case included lower test costs, higher test accuracy, greater adherence to test-suggested treatment, and consideration of the benefits of testing on quality of life.

GEP testing is primarily recommended to support decisions about adjuvant chemotherapy. Although only 22% to 42% of test-eligible patients undergo Oncotype DX testing in the United States,^4-9 the patterns of care in our study suggest that testing is being used in situations where results are most likely to change management. For instance, although older women were less likely to be tested than younger women, older women who were tested were twice as likely to receive chemotherapy as those who were not tested, especially when they had high recurrence risk scores. In addition, among younger patients in whom chemotherapy is typically recommended, many who were tested and had low-risk results avoided chemotherapy.

The cost-effectiveness ratio in this study is substantially higher than that reported in past analyses of Oncotype DX.^12-14,16 This difference is likely to be the result of several factors. First, past studies assumed ideal conditions and/or large reductions in chemotherapy use with testing.^12-14,16 We found that although rates of chemotherapy decreased in community practice after the introduction of Oncotype DX testing,⁸ testing did not change decisions about chemotherapy as dramatically as earlier analyses assumed it would. Second, in contrast to the assumptions in prior analyses, not all patients who were tested followed the test-suggested decision about chemotherapy.^11,16 Moreover, in community practice, fewer women were receiving chemotherapy under usual care before the introduction of testing than assumed in the earlier studies.

This study was unique in considering the impact of test properties on cost-effectiveness ratios, whereas past analyses generally assumed perfect prediction of recurrence.^12,14,16 In fact, the original validation study found that 70% of patients with high-risk scores did not develop distant recurrence and 7% of patients at low risk had distant recurrences at 10 years.³ When we examined idealized conditions, including perfect test accuracy, the cost-effectiveness ratio decreased to $39,496 per QALY, which is more similar to earlier estimates, given inflation.^13,14

We examined Oncotype DX in this study, but there are several other GEP tests being promoted for clinical use.^49-53 Consequently, it is possible that market forces will decrease future GEP test costs. This analysis demonstrated that if Oncotype DX test costs were lower than present Medicare reimbursement rates, it would have cost-effectiveness ratios similar to many currently covered services.^54,55

A novel contribution of this analysis is the consideration of the impact on the cost-effectiveness ratio of the potential ability of GEP testing to provide reassurance if results indicate a low risk of recurrence (or to increase worry with high-risk results). Given that the majority of patients for whom testing is currently recommended will have low recurrence risk scores, the increase in QALYs from reassurance outweighed any decrease as a result of increasing worry among those with high-risk scores. Consideration of these effects lowered the cost-effectiveness ratio to $58,431. Because our result was based on expert opinion, further research is warranted to determine patient utility and willingness to pay related to this aspect of GEP testing. Furthermore, because selection of test result–concordant therapy affects cost-effectiveness ratios, future studies should explore reasons for discordance between treatment prescribed by GEP results and actual treatments received.

There are several caveats that should be considered in evaluating our results. First, the cost-effectiveness results for community practice used data from a large integrated health plan for GEP testing and chemotherapy rates because there is no national source of community data with registry information, GEP results, and complete chemotherapy data. The data used in this analysis may not generalize to other community settings if financial barriers and other practice factors cause different patterns of patient selection to testing and/or differentially affect events downstream from the decision to use GEP testing. Therefore, costs and effects in other community settings could be better or worse than estimated in this analysis. However, data from the patients included in the integrated health plan have been shown to be representative of the US population in terms of sociodemographic and cancer characteristics,^8,56,57 and the patterns of Oncotype DX use and treatment are similar to those reported in other care settings.⁵⁸

Second, GEP testing does not have a direct effect on survival. GEP testing can only affect QALYs by guiding a greater use of chemotherapy to the small proportion of women at highest risk of recurrence who would not otherwise be treated without testing. Hence, the difference in QALYs between tested and untested patients in this analysis is small. In these situations, factors that lead to even small differences in QALYs between community practice and usual care can magnify differences in the cost-effectiveness ratios. Finally, it will be important to reassess the cost-effectiveness ratios for GEP testing as results of the predictive validity for intermediate-risk scores^59,60 and node-positive disease become available.^59-61

Overall, this economic analysis found that the likely cost-effectiveness ratio for Oncotype DX testing in community practice versus usual care without testing was higher than the ratios for most commonly accepted diagnostic and preventive interventions. However, plausible changes in several factors could change the results and lead to Oncotype DX testing having a cost-effectiveness ratio similar to other commonly accepted practices. The substantial differences in conclusions about cost-effectiveness ratios on the basis of community practice versus more idealized practice underscore the importance of considering real-world implementation when assessing the costs and survival associated with new diagnostic (or treatment) technology.

Appendix

Table A1.

Model Outcomes for Life-Years Among Patients With Stage I or II, Node-Negative, ER-Positive, HER2-Negative Breast Cancer by Chemotherapy, Age, Stage, and Recurrence

Table A2.

One-Way Sensitivity Analysis: Impact of Varying Single Parameters on the Societal Cost Effectiveness of Oncotype DX Testing Use in Community Care Versus Usual Care Without Testing

AUTHOR CONTRIBUTIONS

Conception and design: Young Chandler, Clyde B. Schechter, Charles E. Phelps, Tracy A. Lieu, Scott Ramsey, Jeanne S. Mandelblatt

Administrative support: Aimee Near, Jeanne S. Mandelblatt

Collection and assembly of data: G. Thomas Ray, Tracy A. Lieu, Jeanne S. Mandelblatt

Data analysis and interpretation: Young Chandler, Clyde B. Schechter, Jinani Jayasekera, Aimee Near, Suzanne C. O'Neill, Claudine Isaacs, Charles E. Phelps, Tracy A. Lieu, Scott Ramsey, Jeanne S. Mandelblatt

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Cost Effectiveness of Gene Expression Profile Testing in Community Practice

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.

Young Chandler

No relationship to disclose

Clyde B. Schechter

Consulting or Advisory Role: COHRDATA

Jinani Jayasekera

No relationship to disclose

Aimee Near

No relationship to disclose

Suzanne C. O'Neill

No relationship to disclose

Claudine Isaacs

Consulting or Advisory Role: Pfizer, Genentech, Novartis, AstraZeneca, Medivation, NanoString Technologies

Speakers' Bureau: Genentech, Celgene, Pfizer, AstraZeneca

Research Funding: Novartis (Inst), Pfizer (Inst), Genentech (Inst), Tesaro (Inst)

Patents, Royalties, Other Intellectual Property: UpToDate

Travel, Accommodations, Expenses: Caris Life Sciences

Charles E. Phelps

No relationship to disclose

G. Thomas Ray

Employment: The Permanente Medical Group

Research Funding: Pfizer (Inst)

Tracy A. Lieu

No relationship to disclose

Scott Ramsey

Consulting or Advisory Role: Bayer, Genentech, Bristol-Myers Squibb

Research Funding: Genentech

Travel, Accommodations, Expenses: Bristol-Myers Squibb, Bayer

Jeanne S. Mandelblatt

No relationship to disclose