S1 Methods - Proportions of Patients Receiving Endopredict and Chemotherapy, Per Strategy

Online Supplement

Cost-effectiveness analysis of prognostic gene expression signature-based stratification of early breast cancer patients

Patricia R. Blank, Martin Filipits, Peter Dubsky, Florian Gutzwiller, Michael P. Lux, Jan C. Brase, Karsten E. Weber, MargarethaRudas, Richard Greil, SibylleLoibl, Thomas D. Szucs, Ralf Kronenwett, Matthias Schwenkglenks, Michael Gnant.

S1 Methods - Proportions of patients receiving EndoPredict and Chemotherapy, per strategy

Table S1.1. Proportions of patients receiving EndoPredict and Chemotherapy, per strategy

Table S.1.2. Sensitivity and specificity of molecular test or clinical guidelines

Sensitivity and specificity are used to determine the accuracy of a clinical test or clinical guideline. Ideally such tests/ guidelines correctly identify all patients who will develop the condition of interest (i.e. distant recurrence), and also correctly identify all patients who will remain disease free (i.e. have no distant recurrence). Sensitivity and specificity are dependent on the cut-off value above or below which a test is positive or negative, respectively. When the cut-off is changed, increasing sensitivity implies decreasing specificity, and vice versa. Sensitivity and specificity are calculated with the following formula:

• *Sensitivity: True positive / (True positive + False negative) x 100
• **Specificity: True negative / (True negative + False positive) x 100

Whereas:

True positive: number of patients with a positive test (i.e. high risk) and with recurrence

False positive: number of patients with a positive test (i.e. high risk) and without recurrence

False negative: number of patients with a negative test (i.e. low risk) and with recurrence

True negative: number of patients with a negative test (i.e. low risk) and without recurrence

S2 Methods - Clinical model parameters

Table S2.1. Parametric modelling

We used parametric regression modeling to estimate hazards of events by time, over the long-term. The resulting hazard function for overall survival was as follows: h(t) = λ = exp (-Intercept) * 1.15. The hazard function for recurrence was as follows: h(t) = λ = exp (-Intercept) * 0.25. The ten-year survival of the model and the trial data showed accordance.

It was assumed that the probability of transition from the disease-free state to the metastasis state would differ not only between high-risk and low-risk patients, but also between the (partially overlapping but non-identical) high-risk groups defined by the various risk stratification strategies. Hence, annual probabilities of transition to the metastasis state were calculated for each risk stratification strategy and risk group separately. In contrast, transition probabilities from the disease-free state to death, and from the metastasis state to death, were modeled for the pooled patient population, regardless of their risk classification and regardless of strategy, in order to reduce statistical noise. The appropriateness of this simplification was validated with log-rank tests of the ABCSG-based Kaplan-Meier curves for the relevant groups (Figure S.2.2).

Figure S2.2. Kaplan-Meier plots for metastasis-free survival according to the four test strategies by risk group (with number of subjects at risk and 95% CIs)

Metastasis-free survival for low-risk (blue curve), high-risk (red curve) and the whole cohort (both risk groups combined, green curve).p-value of log-rank test (EPclin: EndoPredict test).

S3 Methods - Medical resource use

S3.1. High-risk patients

Disease-free state

In the strategies combining a clinical guideline with EPclin, its use was restricted to those patients classified as being at high risk by the respective guideline.

Patients finally classified as being at high risk were modeled to receive chemotherapy and endocrine therapy in combination, according to current clinical practice. It was assumed that 50% of the patients would receive 5-fluorouracil, epirubicin, and cyclophosphamide (FEC) for six cycles and that the remainder would receive FEC for three cycles followed by three cycles of docetaxel, given that these regimens are used equally in Germany. Half of the patients received endocrine treatment composed of five years of tamoxifen and the remaining received two years of tamoxifen followed by three years of anatrozole, respectively. Drug costs were calculated to include costs of application and surveillance.

Even in absence of symptoms, women with breast-conserving therapy require follow-up care with machine-aided diagnostics. The follow-up care for the disease free group was based on the German-S3 guidelines and included continuous medical visits, mammography, and mamma-sonography, with magnetic resonance imaging (MRI) for unclear findings. Skeletal scintigraphy and chest x-ray or liver-sonography would be used for 20% of patients, respectively[1]. In addition, it was assumed that dual-energy x-ray absorptiometry scans to assess bone density (anastrozole) and vaginal examinations (tamoxifen) would be carried out (Table S3.3).

Costs for irradiation and surgical interventions before the start of adjuvant therapy were assumed to be the same for all patients, hence these costs were not included.

Side effects of chemotherapy

It was assumed that all patients receiving chemotherapy are at risk of side effects. The model included non-hematologic (including nausea, vomiting, arthralgia, and myalgia) and hematologic side effects (covering anemia, neutropenia, and thrombocytopenia) according to the incidence rates seen in patients undergoing chemotherapy [2](Table S3.3).

Costs of side effects of endocrine therapy were not included, since all groups received the same hormonal treatment and hence side effects would occur to the same degree.

Metastasis and end of life

Patients with metastases were assumed to receive non-steroidal aromatase inhibitors, estrogen receptor antagonists, and five cycles of chemotherapy. Besides supportive measures, costs for evaluation, administration, monitoring, and care in the event of progression were also included. Patients were monitored by computer tomography (CT), ultrasound, or scintigraphyevery three months. The health status (including blood-sampling) was assessed every three months and patients were assumed to visit a breast cancer specialist at frequent intervals(Table S3.3). Frequencies of monitoring and diagnostic interventions were estimated according to guidelines and expert consensus[3, 4].

The model assumed that 3% of patients developing a metastasis underwent a loco-regional recurrence before they progressed. These cost were incorporated within the metastatic disease state according to the incidence rate found in our trial population[5-7].

End-of-life care (during the four last weeks of life) included palliative care that included pain and dietetic treatment at home or in the hospice, respectively[3]. The model took into account different medical resource use among the various test strategies which arose from different survival times.

S3.2. Low-risk patients

Low-risk patients received the same therapy scheme as described above, except chemotherapy. This covered endocrine therapy, diagnostics (EPclin where applicable), follow-up care, and care for metastatic disease and at the end of life. No costs for side effects of chemotherapy were included.

Table S3.3. Unit cost input parameters

*incl. vaginal examinations

S3.4.Model implementation

SAS 9.2 was used for statistical analyses. The Markov model was implemented in TreeAge Pro® 2014 (TreeAge Software Inc, Williamstown, MA, USA). In one validation step, 10-year overall and metastasis free survival probabilities seen in the model (76.3% and 90.1%, respectively) were compared with those seen in the original data set (76.1% and 89.9%, respectively).

S4 Results -Deterministic sensitivity analysis

Table S4.1. Deterministic sensitivity analysis (based on 95% CIs, where available)

Parameter values are varied on a one-by-one basis to test the sensitivity of model results to the values of specific parameters or sets of parameters.

*Interpretation of ICER: lower costs and lower effects, the higher the ICER, the better the result

Dominant: a strategy is dominantto the reference the former both costs less and is more effective.

Dominated: a strategy is dominated by another if the former both costs more and is less effective. Dominated strategies are excluded from the calculation of ICERs

NB: Strategies not shown here were constantly dominated by other strategies

EPclin: EndoPredict test; ICER: incremental cost-effectiveness ratio (EUR/QALY); QALY: quality-adjusted life year

S5 Results - Scenario analyses and further sub-analyses

Table S5.1. Scenario analyses (reference strategy: German-S3))

Scenario analyses are chosen to draw attention to variables that materially influence the cost-effectiveness results. They address “what if” questions

*Interpretation of ICER: lower costs and lower effects, the higher the ICER, the better the result

Dominant: a strategy is dominantto the reference the former both costs less and is more effective.

Dominated: a strategy is dominated by another if the former both costs more and is less effective. Dominated strategies are excluded from the calculation of ICERs

Weakly dominated: A strategy is weakly dominated if its cost–effectiveness would be inferior in comparison with at least one more expensive strategy. Weakly dominated strategies are excluded from the calculation of ICERs

EPclin: EndoPredict test; ICER: incremental cost-effectiveness ratio (EUR/QALY); QALY: quality-adjusted life year

NB: Strategies not shown here were constantly dominated by other strategies

Figures S5.2. Sub-analysis: cost-effectiveness plane with reference strategy “no test, no chemotherapy” and additional strategy of chemotherapy to all (regardless of test result; chemo all)

A)Cost per quality-adjusted life years

Chemo all: all patients receive chemotherapy regardless of test result; EPclin: EndoPredict test; Ref: reference strategy; S3: German-S3.

b) Cost per life years gained

Chemo all: all patients receive chemotherapy regardless of test result; EPclin: EndoPredict test; Ref: reference strategy; S3: German-S3.

S6 Results – Probabilistic sensitivity analysis (PSA)

For each recalculation of the model (based on 10,000 sets of parameter values randomly sampled from distributions reflecting the ranges of variation used in deterministic sensitivity analysis, for each variable included in the PSA), strategy-specific cost and effectiveness pairs are plotted on the cost-effectiveness plane.

Figure S6.1. Cost-effectiveness scatter plot (cost per quality-adjusted life years)

Figure S6.2. Cost-effectiveness scatter plot (cost per life year gained)

S7. References to online supplement

1.Wockel A, Kreienberg R: First Revision of the German S3 Guideline 'Diagnosis, Therapy, and Follow-Up of Breast Cancer'. Breast care (Basel, Switzerland) 2008, 3(2):82-86.

2.Martin M, Pienkowski T, Mackey J, Pawlicki M, Guastalla JP, Weaver C, Tomiak E, Al-Tweigeri T, Chap L, Juhos E et al: Adjuvant docetaxel for node-positive breast cancer. The New England journal of medicine 2005, 352(22):2302-2313.

3.Lux MP, Hartmann M, Jackisch C, Raab G, Schneeweiss A, Possinger K, Oyee J, Harbeck N: Cost-utility analysis for advanced breast cancer therapy in Germany: results of the fulvestrant sequencing model. Breast cancerresearchandtreatment2009, 117(2):305-317.

4.al AUe: Stufe-3-Leitlinie Brustkrebs-Frueherkennung in Deutschland, vol. 1. Aktualisierung 2008. München, Germany: W ZuckschwerdtVerlag; 2008.

5.Jakesz R, Jonat W, Gnant M, Mittlboeck M, Greil R, Tausch C, Hilfrich J, Kwasny W, Menzel C, Samonigg H et al: Switching of postmenopausal women with endocrine-responsive early breast cancer to anastrozole after 2 years' adjuvant tamoxifen: combined results of ABCSG trial 8 and ARNO 95 trial. Lancet 2005, 366(9484):455-462.

6.Schmid M, Jakesz R, Samonigg H, Kubista E, Gnant M, Menzel C, Seifert M, Haider K, Taucher S, Mlineritsch B et al: Randomized trial of tamoxifen versus tamoxifen plus aminoglutethimide as adjuvant treatment in postmenopausal breast cancer patients with hormone receptor-positive disease: Austrian breast and colorectal cancer study group trial 6. J ClinOncol2003, 21(6):984-990.

7.Dubsky P, Filipits M, Jakesz R, Rudas M, Singer CF, Greil R, Dietze O, Luisser I, Klug E, Sedivy R et al: EndoPredict improves the prognostic classification derived from common clinical guidelines in ER-positive, HER2-negative early breast cancer. Annals of oncology : official journal of the European Society for Medical Oncology / ESMO 2012.

8.Lux MP, Reichelt C, Karnon J, Tanzer TD, Radosavac D, Fasching PA, Beckmann MW, Thiel FC: Cost-Benefit Analysis of Endocrine Therapy in the Adjuvant Setting for Postmenopausal Patients with Hormone Receptor-Positive Breast Cancer, Based on Survival Data and Future Prices for Generic Drugs in the Context of the German Health Care System. Breast care (Basel, Switzerland) 2011, 6(5):381-389.

1