An illustration · invented records · this is a real, openly licensed paper, reproduced in full

Tumour profiling tests to guide adjuvant chemotherapy decisions in lymph node-positive early breast cancer: a systematic review and economic evaluation

Tappenden P, Cooper K, Hamilton J, Nalbant G, Essat M, Rayner A, Wong R, Battisti NML, Wyld L, Asghar U, et al. (2025). Tumour profiling tests to guide adjuvant chemotherapy decisions in lymph node-positive early breast cancer: a systematic review and economic evaluation. Health Technology Assessment 29(49). NIHR award NIHR135822. DOI 10.3310/KGFD4040

License: CC BY (Creative Commons Attribution). This license permits copying and redistribution with attribution, which is why the whole paper can sit in this folder. Source: https://www.journalslibrary.nihr.ac.uk/hta/KGFD4040 · PubMed 41074908 · PMC12683682 How this file was made: the publisher’s JATS XML was obtained from the NIHR Journals Library on 2026-09-08 and split into locator-tagged paragraphs and tables. This Markdown was rendered from those passages on 2026-09-08. Figures are represented by their captions only; chart data is not extracted. References are omitted. The report’s nine appendices are included at the end of this file, extracted by a one-off step described there; the full PDF is 182 pages. Wording is the publisher’s; section headings and paragraph numbers match the locators used in the source note.

Abstract

Breast cancer is the most commonly diagnosed cancer in women in England. Breast cancer and chemotherapy treatment can impact upon patients’ quality of life and survival. Tumour profiling tests can help to identify whether patients will benefit from chemotherapy.

_¶1

To evaluate the effectiveness and cost-effectiveness of four tumour profiling tests (Oncotype DX, Prosigna, EPclin and MammaPrint), compared with current decision-making (no testing), to guide use of adjuvant chemotherapy in people with hormone-receptor positive, human epidermal growth factor receptor 2 negative, early-stage breast cancer with one to three positive lymph nodes.

_¶2

A systematic review identified studies via a literature search in April 2023 and from our previous review. The economic analysis included a review of existing models and development of an independent model.

_¶3

Fifty-five articles were included, 42 for prognostic and predictive ability and 13 for impact on chemotherapy decisions. All four tests showed prognostic ability for determining risk of relapse. The RxPONDER randomised controlled trial of Oncotype DX indicated no chemotherapy benefit in post-menopausal lymph node-positive patients with a recurrence score of 0–25, but a statistically significant benefit in pre-menopausal patients with a recurrence score of 0–25. An older randomised controlled trial reanalysis (Southwest Oncology Group-8814) indicated lower relative chemotherapy benefit with lower recurrence score, with statistically significant interactions between recurrence score and chemotherapy benefit in some but not all analyses. There was no clear evidence of prediction of relative chemotherapy benefit for Prosigna, EPclin or MammaPrint. Decision impact studies in lymph node-positive populations in the United Kingdom and Europe were only available for Oncotype DX, and they reported a reduction of 12–75% in chemotherapy recommendations following testing.

_¶4

Based on the list prices of the tests and downstream treatments, the independent model suggests the following:

_¶5

This test dominates current decision-making in post-menopausal lymph node-positive women, provided an assumption of predictive benefit holds, but the test is dominated if this assumption does not hold. The test is dominated by current decision-making in pre-menopausal lymph node-positive women.

_¶6

The probabilistic incremental cost-effectiveness ratio for Prosigna versus current decision-making in post-menopausal lymph node-positive women is £39,357 per quality-adjusted life-year gained.

_¶7

The probabilistic incremental cost-effectiveness ratio for EPclin versus current decision-making in post-menopausal lymph node-positive women is £4113 per quality-adjusted life-year gained.

_¶8

Within clinical high-risk pre-/post-menopausal lymph node-positive women, MammaPrint is dominated by current decision-making.

_¶9

There are limited data on the prediction of chemotherapy benefit; evidence for Oncotype DX may support a predictive benefit, but this is uncertain. Decision impact studies in a lymph node-positive population were available only for Oncotype DX. The economic model relies on an assumption of predictive benefit for Oncotype DX, and broader assumptions around the way that Prosigna, MammaPrint and EPclin test results would affect chemotherapy decisions.

_¶10

All four tests provide prognostic information on the risk of relapse. The evidence on prediction of relative chemotherapy benefit is weaker and mostly limited to Oncotype DX. The economic analyses indicate that Oncotype DX and EPclin may have favourable cost-effectiveness profiles in post-menopausal lymph node-positive subgroups, although this is uncertain.

_¶11

This study is registered as PROSPERO CRD42023425638.

_¶12

This award was funded by the National Institute for Health and Care Research (NIHR) Evidence Synthesis Programme (NIHR award ref: NIHR135822) and is published in full in Health Technology Assessment; Vol. 29, No. 49. See the NIHR Funding and Awards website for further award information.

_¶13

Breast cancer is the most common cancer in women in England. Breast cancer, and its treatment, can affect a person’s quality of life and how long they live for. Most women with early-stage breast cancer which has spread to one to three lymph nodes receive chemotherapy to stop the cancer from coming back and spreading elsewhere in the body, but this treatment can cause side effects, including damage to the heart and secondary cancers. Currently, some doctors use computer tools which use information about the patient and the tumour to decide if chemotherapy is needed. Tumour profiling tests are used to help women with early breast cancer to decide whether they should have chemotherapy. They test small samples of a patient’s tumour to find out whether the genes in it mean that a person has a high or low risk of the disease returning. If this risk is low, the patient might not need chemotherapy and, therefore, they can avoid its side effects. Some tests might also be able to identify which patients are more likely to respond to chemotherapy.

_¶1

This study looked at the evidence for four tumour profiling tests. Fifty-four clinical studies were identified. The results suggest that all of the tests can give information on the risk of the cancer returning. There was some information about whether one of the tests (Oncotype DX) can predict which patients will respond to chemotherapy. There was information about how using one test (Oncotype DX) affects the decision to have chemotherapy. Our study also looked at whether or not these tests represent good value for money for the National Health Service through cost-effectiveness analyses. The analyses showed that two of the tests (Oncotype DX and EPclin) may represent a good use of National Health Service resources for some patient groups.

_¶2

Breast cancer is the most commonly diagnosed cancer in women and the fourth most common cause of cancer-related death in the UK. Most people with lymph node-positive (LN+) breast cancer receive adjuvant chemotherapy due to their increased risk of recurrence. However, chemotherapy is associated with considerable adverse effects. Currently, adjuvant chemotherapy decisions may be informed by clinical and pathological information, sometimes via a risk prediction tool. Improved information on a patient’s risk of recurrence (i.e. their prognostic risk) and/or their likely response to chemotherapy (i.e. predictive benefit) may help clinicians to target chemotherapy to patients who will benefit most. Tumour profiling tests aim to improve decisions on chemotherapy use by improving the categorisation of patients according to risk and the identification of patients who will benefit most from chemotherapy.

_¶1

In 2018, the National Institute for Health and Care Excellence (NICE) published Diagnostics Guidance (DG) No. 34. DG34 recommends the use of Oncotype DX, Prosigna and EndoPredict (EPclin) for guiding chemotherapy decisions in people with oestrogen receptor (ER)-positive, human epidermal growth factor receptor 2 (HER2) negative, lymph node-negative (LN0) early breast cancer, including those with micrometastases. Two other tests assessed in DG34 (MammaPrint and immunohistochemical 4 (IHC4)) were not recommended in the LN0 population. While DG34 also assessed these tests in women with LN+ early breast cancer, the Appraisal Committee did not make any specific recommendations on the use of any test within LN+ patients. This assessment provides an updated systematic literature review and economic analysis of four tumour profiling tests (Oncotype DX, Prosigna, EPclin and MammaPrint) compared to current decision-making in women with ER-positive [and/or progesterone receptor (PR)-positive], HER2-negative, early breast cancer with one to three positive lymph nodes.

_¶2

The main research question is: ‘Do tumour profiling tests used for guiding adjuvant chemotherapy decisions in patients with ER-positive (and/or PR-positive), HER2-negative, early-stage breast cancer with 1 to 3 positive lymph nodes represent a clinically effective and cost-effective use of NHS resources?’

_¶3

To conduct a systematic review of effectiveness and cost-effectiveness of four tumour profiling tests (Oncotype DX, Prosigna, EPclin and MammaPrint).

_¶4

To develop a health economic model to assess the cost-effectiveness of tumour profiling tests compared with current decision-making (no testing) on the use of chemotherapy from the perspective of the NHS and Personal Social Services (PSS).

_¶5

The External Assessment Group (EAG) undertook a systematic review of Oncotype DX, Prosigna, EPclin and MammaPrint for guiding adjuvant chemotherapy decisions in women with ER+/PR+, HER2− early breast cancer where the study population was at least 80% LN+. Studies were identified from the previous review which informed NICE DG34 (searches conducted in 2017) plus an updated search (April 2023) covering MEDLINE, EMBASE, Cochrane, and other sources. Eligible data types included prospective randomised controlled trials (RCTs) and studies of prognostic ability, prediction of relative chemotherapy benefit, impact of tests on chemotherapy decisions (restricted to UK and European studies), and health-related quality of life (HRQoL) and anxiety associated with testing.

_¶6

The EAG undertook a systematic review of existing economic analyses of Oncotype DX, Prosigna, EPclin and MammaPrint for guiding adjuvant chemotherapy decisions in women with ER+, HER2−, LN+ early breast cancer. Studies included published analyses which were identified within the previous systematic review undertaken to inform NICE DG34 and economic analyses in LN+ populations published since 2017. The EAG also critically appraised economic analyses of Oncotype DX and MammaPrint submitted to NICE by the test manufacturers.

_¶7

The EAG also developed a de novo health economic model to assess the cost-effectiveness of Oncotype DX, MammaPrint, Prosigna, and EndoPredict (EPclin), each compared against current decision-making. The economic analysis was undertaken from the perspective of the NHS and PSS and was largely based on the model developed to inform NICE DG34, with updates to reflect changes in the breast cancer treatment pathway and updated evidence on the tests identified from the clinical effectiveness review. The EAG model adopts a hybrid decision tree/Markov structure. Model parameter values were informed by the RxPONDER, TransATAC, SWOG-8814 and MINDACT trials, a recent UK decision impact study undertaken in women with LN+ early breast cancer, previous economic models, routine costing sources and other literature. All results presented in this report reflect the list prices of the tumour profiling tests; additional analyses including price discounts for the tests and downstream treatments were provided in a separate confidential appendix to NICE.

_¶8

In total, 55 articles were included, 42 relating to prognostic and predictive ability, and 13 relating to impact on chemotherapy decisions. Data were reported for two prospective RCTs (RxPONDER and MINDACT). In RxPONDER, LN+ patients with an Oncotype DX Breast Recurrence Score (RS) of 0–25 were randomised to chemotherapy versus no chemotherapy. In MINDACT, patients with discordant MammaPrint risk and clinical risk were randomised to chemotherapy versus no chemotherapy. In addition, the ongoing OPTIMA RCT compares Prosigna test-directed chemotherapy use versus standard chemotherapy use; however, results are not yet available.

_¶9

All four tests demonstrated prognostic ability for determining risk of relapse in LN+ populations, both with and without adjustment for clinical factors.

_¶10

No predictive data in a LN+ population were identified for Prosigna or EPclin. For Oncotype DX, a reanalysis of the SWOG-8814 RCT using cut-offs of RS < 18 and > 30 indicated no effect of chemotherapy on 10-year disease-free survival in the low-risk group, a non-significant effect in the intermediate-risk group, and a borderline statistically significant effect in the high-risk group, with statistically significant interaction tests in some but not all analyses. The RxPONDER prospective RCT reported no benefit of chemotherapy in post-menopausal patients with an RS of 0–25, but a statistically significant benefit in pre-menopausal patients with an RS of 0–25, while the test for interaction between RS (within the range 0–25) and effect of chemotherapy was not statistically significant in either group. The National Cancer Database reported 5-year overall survival within post-menopausal or older-age subgroups with RS ≤ 25; some analyses showed a statistically significant chemotherapy benefit while others did not. For MammaPrint, prediction of chemotherapy benefit could not be determined from the LN+ subgroup of the MINDACT prospective RCT, because all patients in the clinical high-risk, MammaPrint high-risk group, were offered chemotherapy (there was a non-significant effect of chemotherapy in the LN+ MammaPrint low-risk group). A cohort reanalysis from 2009 reported a non-significant interaction test between MammaPrint score and effect of chemotherapy on breast cancer-specific survival (p = 0.95).

_¶11

Studies on chemotherapy decisions in LN+ populations in the UK and Europe indicated a net reduction in the percentage of patients recommended chemotherapy pre-test to post-test of between 12% and 75%, with greater reductions in groups with lower RS. All studies used Oncotype DX; no decision impact studies were identified for EPclin, Prosigna or MammaPrint.

_¶12

No studies reported HRQoL or anxiety associated with using tumour profiling tests in a LN+ population. Therefore, studies in a LN0 or mixed nodal status population were briefly summarised, with mixed results regarding the impact of testing and anxiety.

_¶13

The results of the EAG’s probabilistic base-case analyses are summarised below.

_¶14

Within the pre-menopausal LN+ population, Oncotype DX is dominated by current decision-making. These results are driven by the estimated reduction in the use of adjuvant chemotherapy due to the test in women who would have benefitted from treatment.

_¶15

Within the post-menopausal LN+ subgroup, Oncotype DX dominates current decision-making, provided the assumption of predictive benefit holds. These results are driven by an estimated large reduction in the use of adjuvant chemotherapy in women who would not have benefitted from treatment. As was the case with the economic analyses in the LN+ subgroup undertaken to inform DG34, removing this assumption of predictive benefit results in a situation whereby Oncotype DX is dominated by current decision-making, driven by a large reduction in the use of adjuvant chemotherapy in women who would have benefitted from treatment and an increase in the lifetime probability of developing distant metastases (DM). This assumption of predictive benefit remains subject to some uncertainty, and it strongly influences the conclusions of the economic analysis in the post-menopausal LN+ subgroup.

_¶16

The incremental cost-effectiveness ratio (ICER) for Prosigna versus current decision-making is expected to be £39,357 per quality-adjusted life-year (QALY) gained. The model suggests that the use of Prosigna will result in a small decrease in the use of chemotherapy, a small reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify any evidence to support a predictive benefit for Prosigna in the LN+ population.

_¶17

The ICER for EPclin versus current decision-making is expected to be £4113 per QALY gained. The model suggests that the use of EPclin will result in a small decrease in adjuvant chemotherapy use, a small reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify any evidence to support a predictive benefit for EPclin in the LN+ population.

_¶18

MammaPrint is dominated by current decision-making. These results are driven by a large reduction in the use of adjuvant chemotherapy in women who would have benefitted from treatment, an increase in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify sufficient evidence to support a predictive benefit for MammaPrint in the LN+ population.

_¶19

Strengths of the clinical evidence base include the fairly substantial evidence for prognostic ability of all four tests. A major limitation is the difficulty in collecting new data on predictive ability, as it is not considered ethical to randomise patients who are high risk on any test to chemotherapy versus no chemotherapy. Therefore, although there are prospective RCTs for the effect of chemotherapy in low-risk to intermediate-risk patients, data for high-risk patients are limited to retrospective reanalyses of trials, plus observational data in which test results may have influenced treatment. Decision impact data in a LN+ population were only available for Oncotype DX. Anxiety and HRQoL data associated with testing were not identified in a LN+ population.

_¶20

The EAG’s model has several strengths: the economic analysis is consistent with the NICE Reference Case and relates specifically to the LN+ population under consideration within this appraisal; the model structure is generally consistent with most published economic models of tumour profiling tests as well as the two economic models submitted by the test manufacturers; for each individual test, risk classification probabilities and distant recurrence-free interval estimates have been taken from same source where data permit, which avoids potential spectrum bias; the analysis uses a recent relevant UK decision impact study undertaken in LN+ women; and a broad assessment of uncertainty around all key model inputs has been presented, including testing assumptions around whether Oncotype DX is predictive of chemotherapy benefit.

_¶21

The EAG’s economic analyses are subject to several weaknesses: the economic analyses of Oncotype DX based on RxPONDER indirectly assume a predictive benefit which reflects a plausible clinical assumption about the effect of chemotherapy in women who were excluded from the trial (external data from SWOG-8814 are used to inform the benefit of chemotherapy in women with an RS of > 25), rather than a statistical test of interaction across the full RS spectrum; there are inconsistencies in Oncotype DX RS cut-offs between sources used in the model; the analyses rely on a decision impact study of Oncotype DX to estimate post-test probabilities for all 2- and 3-level tests, which is highly uncertain; and there is insufficient evidence to allow for the economic analyses of EPclin and MammaPrint in an exclusively pre-menopausal subgroup. There is uncertainty around the potential negative effects of chemotherapy on infertility which may not be fully captured in the analysis of Oncotype DX in the pre-menopausal LN+ subgroup. The EAG’s analyses of net health benefit provide a means for considering whether any missing health effects are likely to impact on the conclusions drawn from the economic analysis.

_¶22

Oncotype DX, Prosigna and EPclin are already recommended for use in the NHS for women with ER+ (and/or PR+), HER2−, LN0 early breast cancer. Depending on the specific test and population under consideration, tumour profiling may result in fewer women receiving adjuvant chemotherapy (reducing costs and increasing capacity), but this may lead to more women requiring further treatment for DM (increasing costs and reducing capacity).

_¶23

MammaPrint is not currently recommended for use in the NHS. MammaPrint testing can be undertaken either as an off-site service with samples sent to a laboratory in the USA or through a decentralised testing service for laboratories with next-generation sequencing (NGS) capability. The per-sample pricing of MammaPrint remains the same regardless of testing location. Not all laboratories will have NGS capabilities which has implications for how MammaPrint testing is organised and delivered. For the other tests, only one sample processing approach is available – for Oncotype DX, samples are processed centrally at the Exact Sciences laboratory in the USA, whereas for Prosigna and EPclin, samples are processed in local laboratories.

_¶24

Research priorities include the following:

_¶25

Further studies assessing the ability of all four tests to predict long-term relative chemotherapy benefit in LN+ populations would help to address uncertainty. This may require observational or registry data to assess outcomes across the full range of test scores. In addition, the OPTIMA trial is ongoing, comparing Prosigna test-directed chemotherapy use versus standard chemotherapy use.

_¶26

Longer-term studies to further quantify the negative impact of adjuvant chemotherapy using a preference-based instrument would be valuable, to estimate both short-term toxicity and longer-term negative effects, including impacts on fertility in pre-menopausal women.

_¶27

Further UK and European studies assessing the impact of tumour profiling tests on recommendations for adjuvant chemotherapy in LN+ populations may reduce uncertainty around clinical impact and cost-effectiveness.

_¶28

The integration of tumour profiling tests with decision aid tools to support shared decision-making may constitute a useful research direction.

_¶29

The role of tumour profiling tests in older adults, who may be more prone to chemotherapy complications in the context of limited life expectancy, is also a research priority, as is research on test performance in males and in ethnically diverse populations.

_¶30

This study is registered as PROSPERO CRD42023425638.

_¶31

_¶32

Background and definition of the decision problem

This report includes some material which has been reproduced from the National Institute for Health and Care Excellence (NICE) scope and the protocol for this appraisal. © NICE 2024 Tumour profiling tests to guide adjuvant chemotherapy decisions in early breast cancer. Available from www.nice.org.uk/guidance/dg58. All rights reserved. Subject to Notice of rights (www.nice.org.uk/terms-and-conditions#notice-of-rights). NICE guidance is prepared for the National Health Service in England. All NICE guidance is subject to regular review and may be updated or withdrawn. NICE accepts no responsibility for the use of its content in this product/publication.

_¶1

Condition and aetiology

Breast cancer is the most commonly diagnosed cancer and the fourth most common cause of cancer-related death in the UK. During the period 2016–8, an average of 46,479 women and 319 men were diagnosed with breast cancer in England each year.1 Initial treatment for breast cancer usually involves surgery to remove the primary tumour and some or all of the axillary lymph nodes. Depending on the breast cancer characteristics, this may be followed by one or more of the following treatments: radiotherapy, endocrine (hormone) therapy, targeted therapy, bisphosphonates and/or chemotherapy. A proportion of patients also receive neoadjuvant therapy prior to surgery, although this is primarily aimed at women with triple negative or human epidermal growth factor receptor 2 (HER2)-positive breast cancers.

Figure 3 presents estimates of breast cancer incidence by age and sex for the UK, based on data from 2016 to 2018 reported by Cancer Research UK.6 Breast cancer incidence varies most according to gender. Women are considerably more likely to develop breast cancer than men. For both males and females, incidence generally increases with age. Over 82% of cases of breast cancer occur in people aged 50 years and over and approximately 24% of cases are in people aged 75 years and older.

_¶1

Significance in terms of ill-health (burden of disease)

Breast cancer is the second most common cause of cancer death in women after lung cancer, with an age-standardised mortality rate of 32.8 per 100,000 women. The age-standardised mortality rate in men is substantially lower at 0.3 per 100,000 men. During the period 2017–9, an average of 9509 women and 69 men died from breast cancer in England each year.7

_¶1

Current methods for staging of breast cancer

Breast cancer staging takes into account three main factors: (1) tumour size; (2) metastases to the regional lymph nodes; and (3) the presence/absence of distant metastases.8 The tumour/node/metastases (TNM) staging system was developed and is maintained by the American Joint Committee on Cancer (AJCC) and the Union International Contre le Cancer. Version 8 of the AJCC TNM staging system was published in 2018.9 According to this staging system, T stage is classified according to size of the tumour and degree of local infiltration; N stage is classified according to the number and location of metastases to the lymph nodes in the axilla, between the ribs (internal mammary nodes) and above or below the collarbone (supraclavicular and infraclavicular nodes); and M stage is classified by the presence of metastases beyond the breast and regional lymph nodes. The overall TNM stage of the cancer is defined as shown in Table 1. Early breast cancer is generally defined as cancer which has not spread beyond the breast or the ipsilateral axillary lymph nodes, and is confined to stages I, II or IIIA.

_¶1

Management of early breast cancer

National Institute for Health and Care Excellence Guideline (NG) 1013 provides recommendations on the diagnosis and management of early and locally advanced breast cancer. The guideline was first published in 2009 and was updated in 2018 and again in July 2023. The general treatment pathway for women with early breast cancer is summarised in Figure 4. Key recommendations for the diagnosis and management of early breast cancer are summarised in the subsequent sections, based on NG101,3 Harnan et al.,10 the summary provided in the NICE scope11 and additional information provided by the clinical advisors to the External Assessment Group (EAG).

_¶1

Surgical resection and neoadjuvant treatments

The initial treatment for early and some locally advanced breast cancers usually involves the surgical resection of the primary tumour. Surgical options to remove the disease in the breast include breast conserving surgery or mastectomy (where the whole breast is removed). If appropriate, women are offered the option to have reconstruction at the time of the initial surgery, or at a later date. Neoadjuvant systemic treatment may be given prior to surgery, with the aim of reducing the size of the tumour to enable breast conserving surgery. Depending on whether clinical or ultrasound visible axillary disease is present, axillary surgery is also performed, involving a sentinel lymph node biopsy if the nodes are not thought to be involved and an axillary clearance if there is upfront nodal disease. Increasingly, for women with clinically involved nodes, where a good response to neoadjuvant chemotherapy is anticipated (triple negative or HER2+ breast cancer), chemotherapy will be given first to attempt to downstage the axilla.

_¶1

In women who have a sentinel lymph node biopsy, if there is heavy nodal disease then a subsequent clearance is performed. If only micrometastases, isolated tumour cells or just one or two nodes are involved, then axillary radiotherapy or no formal axillary treatment is indicated. These strategies reduce the risk of adverse events (AEs) such as lymphoedema without a negative impact on survival.

_¶2

Adjuvant therapy planning

After surgery, adjuvant treatment may be needed to treat residual micrometastatic disease following surgery and to reduce the risk of local and distant relapse. Adjuvant treatment may involve chemotherapy, endocrine therapy (ET), targeted therapy, radiotherapy or a combination of these treatments. The decision to offer, and the selection of, adjuvant therapy is made taking into account the patient’s clinical history, the patient’s fitness and health status, the stage of disease, the patient’s likely prognosis, the molecular characteristics of the tumour and the patient’s preferences. NG1013 makes the following recommendations on adjuvant treatment planning:

_¶1

The EAG’s clinical advisors also commented that PREDICT version 2.0 has not been validated in pregnant women and that it may be less accurate for patients treated with neoadjuvant systemic therapy, patients aged 65 years and over, patients with a high comorbidity burden and patients with multifocal breast cancer, bilateral breast cancer, rare breast cancer subtypes or two different breast cancers. They also commented that PREDICT may be less accurate in the context of contemporary systemic treatment standards of care.

_¶2

While NG1013 recommends the use of PREDICT to provide prognostic information on breast cancer recurrence and absolute chemotherapy benefit to guide decisions about the use of adjuvant chemotherapy, several other prognostic tools are also available which can help to predict the likelihood of breast cancer recurrence. These tools are described in Prognostic risk prediction tools.

_¶3

The potential value of tumour profiling tests to guide chemotherapy decisions for women with lymph node-positive early breast cancer

Meta-analyses of randomised clinical trials (RCTs) reported by the EBCTCG have indicated that the use of adjuvant chemotherapy is associated with a reduction in the risk of distant recurrence and death in women with early-stage breast cancer.16 Lymph node involvement is associated with an increased risk of recurrence; hence, the majority of women with LN+ early breast cancer in England currently receive adjuvant chemotherapy.17,18 However, chemotherapy is also associated with considerable AEs, including both short- and long-term effects. These AEs negatively impact on patients’ health-related quality of life (HRQoL) and result in additional healthcare costs. Short-term toxicity that occurs during chemotherapy is usually temporary and reversible and commonly includes: nausea; vomiting; mouth soreness; diarrhoea; tiredness; liver damage; diarrhoea and constipation; skin rash and nail changes; hair loss and temporary lowering of the blood counts which can lead to hospitalisation due to neutropenic sepsis and death. Chemotherapy is also associated with a risk of late effects, including damage to the heart, temporary or permanent amenorrhea, peripheral neuropathy, and a small increase in the risk of secondary malignancies including leukaemia.19 Adjuvant chemotherapy may prevent distant recurrence for some women with early breast cancer, while others will not obtain benefit from treatment, with many women remaining recurrence-free at 10 years without chemotherapy.20 This presents a challenge for clinicians in estimating prognosis and making the most appropriate therapeutic decisions regarding whether or not to offer adjuvant chemotherapy to women with early-stage breast cancer. Improved information on a patient’s risk of recurrence (i.e. prognostic risk) and/or likely response to chemotherapy (i.e. predictive benefit) may help target chemotherapy at those patients who will benefit the most from treatment. Avoiding chemotherapy in patients who have a lower risk of recurrence, who would therefore obtain limited benefit, avoids the unpleasant side effects of chemotherapy and reduces expenditure on both the chemotherapy itself and the treatment of AEs resulting from its use.

_¶1

Oncotype DX (Exact Sciences)

Oncotype DX Breast Recurrence Score (RS) (Oncotype DX) is a Conformité Européene (CE) marked assay designed to quantify the 9-year risk of distant recurrence. The company claims that the test can also predict the likelihood of chemotherapy benefit. The test also reports the underlying tumour biology: ER, PR and HER2 status. The test is intended for use in people with early breast cancer that has the following clinical features:

_¶1

Oncotype DX quantifies the expression of 21 genes. Of these, 16 are cancer-related genes correlated with distant recurrence-free survival (DRFS), and 5 are reference genes for normalising the expression of the cancer-related genes. This information is used to calculate the Breast RS.

_¶2

Oncotype DX is offered as a test service to the NHS. Samples are processed centrally at the Exact Sciences centralised laboratory in the USA, which is accredited by the American Association for Laboratory Accreditation and the College of American Pathologists. The test requires a formalin-fixed paraffin-embedded (FFPE) breast cancer tissue sample from a biopsy or surgical resection, which can be sent as a paraffin embedded block or as 15 unstained charged slides. The test process uses reverse transcription-quantitative polymerase chain reaction (RT-qPCR).

_¶3

The test gives an RS of between 0 and 100, which is used to estimate the 9-year risk of distant recurrence, assuming 5 years of standard ET. The company claims that the RS also predicts the benefit of chemotherapy in terms of reducing the risk of distant recurrence. For LN+ disease (one to three positive nodes), the Instructions For Use document states that a score below 18 predicts little to no chemotherapy benefit, a score between 18 and 30 predicts a potential chemotherapy benefit, and a score of 31 or more predicts a large benefit from chemotherapy. However, the company’s website (accessed by NICE on the 27 February 2023) states that an RS of 25 or less predicts no chemotherapy benefit for post-menopausal women and 2.9% benefit at 5 years for pre-menopausal women. The company’s website states that in both groups, a score of 26–100 is inferred to predict substantial chemotherapy benefit.

_¶4

The Oncotype DX Breast RS results are typically reported within 7–10 calendar days after the sample is received at the laboratory.

The relevant comparator is current decision-making, which includes clinical and pathological features used to assess risk, and clinicopathological tools outlined in Current service provision (current tools include PREDICT and NPI while older tools include AOL). Due to the lack of availability of end-to-end studies comparing decision-making based on the test versus current tools, different evidence types were sought and are linked via the EAG’s health economic model (see Independent External Assessment Group economic analysis).

_¶1

Outcomes

The clinical review aimed to identify the following types of data:

Evidence on pre-test and post-test decisions/recommendations for receiving chemotherapy was identified for Oncotype DX; this included five UK studies (within six references)35–40 and seven other European studies.41–47 Of these, two UK studies and four other European studies reported data by Oncotype DX risk groups. No decision impact studies were identified which assessed EPclin, Prosigna or MammaPrint.

_¶1

No studies (or subgroups) reporting HRQoL or anxiety associated with use of tumour profiling tests were identified in a mainly LN+ population. Therefore, a brief summary of the evidence in a LN0 or mixed population is provided.

_¶1

Risk of bias in included studies

A summary of risk of bias in the included studies is provided here, with further details in Appendix 3.

_¶1

The two prospective RCTs (RxPONDER29 and MINDACT30), assessed using the Cochrane RoB2 tool,27 scored low risk of bias on all domains, and low risk of bias overall. As noted in Summary of evidence identified for each outcome type, there may have been selection bias in RxPONDER since patients had knowledge of their RS result before agreeing to be randomised.

_¶2

Risk of bias in prognostic and predictive studies was assessed using the PROBAST tool.28 For prognostic studies, the following factors may have affected results to some extent. Studies varied in terms of whether people received chemotherapy or not; studies are therefore reported separately according to chemotherapy use in the section on prognostic ability (see Results: prognostic ability). In some studies, some participants did not match the review question (either not ER+, not HER2− or not LN1–3); these factors were taken into account when selecting studies for inclusion in the economic model. Most studies excluded a proportion of patients for various reasons including insufficient tissue, missing data, failed tests and others, which may have influenced results to some extent, though the impact is difficult to assess. In terms of outcomes, chemotherapy decisions were not influenced by the test result in studies of retrospective use of the test (i.e. reanalyses of RCTs and cohorts), whereas in observational studies in which the test was used prospectively, chemotherapy decisions may have been influenced by the test result; therefore, observational studies are reported separately in the section on prognostic ability (see Results: prognostic ability).

_¶3

For predictive studies, the following factors may have affected results to some extent. Only the SWOG-8814 study32 was a reanalysis of a RCT in which chemotherapy use was randomised; in the remaining studies, chemotherapy use was not randomised. This limitation is reflected in the section on prediction of chemotherapy benefit (see Results: prediction of chemotherapy benefit). In some studies, some participants did not match the review question (either not ER+, not HER2− or not LN1–3). Most studies excluded a proportion of patients for various reasons including insufficient tissue, missing data, failed tests and others, which may have influenced results to some extent, though the impact is difficult to assess. In terms of outcomes, chemotherapy decisions were not influenced by the test result in two studies of retrospective use of the test, whereas in the three observational registries in which the test was used prospectively, chemotherapy decisions may have been influenced by the test result; therefore, observational studies are reported separately in the section on prediction of chemotherapy benefit (see Results: prediction of chemotherapy benefit).

_¶4

Overview of prognostic data in this report

The prognostic ability of a genomic test describes its ability to differentiate between patients with good versus poor outcomes. Studies of prognostic ability provide risk classification probabilities, that is, the proportion of patients allocated to each risk group. They also provide the risk of distant metastases (DM) or OS per risk group, and HRs for the difference in outcomes between risk groups (both unadjusted and after adjustment for clinical and pathological factors). The evidence on prognostic ability in this review includes the following types of evidence and key studies:

_¶1

A summary of prognostic data for distant recurrence across the four tests, based on reanalyses of trials or cohorts, is provided in Table 4. Full details of the prognostic data are provided in Appendix 4, which includes additional outcomes (such as DFS, OS and BCSS). Data for the two prospective RCTs (RxPONDER29,51 and MINDACT30,52) are presented in Tables 5 and 6. Data on observational studies of prospective use of Oncotype DX are provided in Table 7.

_¶2

Summary of distribution of genomic risk groups and distant recurrence risk

Table 4 summarises prognostic data from studies reporting 10-year distant recurrence outcomes. For patients receiving endocrine monotherapy, the review identified one study of Oncotype DX,20 no studies of MammaPrint, two studies of Prosigna20,56 and three studies of EPclin.20,63,65 A third study of Prosigna58 which used different cut-offs is shown in Table 4 for completeness, but is not included in this textual summary. In terms of distribution, the study of Oncotype DX20 (which used cut-offs of RS < 18 and RS 30) assigned more patients to the low-risk group (57%) than the studies of Prosigna (4–8% low risk)20,56 or EPclin (19–35% low risk).20,63,65 Freedom from distant recurrence at 10 years in the low-risk group was 81% (one study of Oncotype DX),20 100% (two studies of Prosigna)20,56 and 94–100% (three studies of EPclin).20,63,65 Freedom from distant recurrence at 10 years in the high-risk group was 62% (one study of Oncotype DX),20 69–76% (two studies of Prosigna)20,56 and 70–81% (three studies of EPclin).20,63,65 Further details of the prognostic data can be found in Appendix 4 (Table 49 for Oncotype DX, Table 50 for MammaPrint, Table 51 for Prosigna and Table 52 for EPclin).

_¶1

Table 4 also presents 10-year distance recurrence data from further studies in which some or all patients received chemotherapy, including one study of Oncotype DX,49 three studies of MammaPrint,33,53,54 one study of Prosigna59,60 and one study of EPclin.59,60 The distributions and 10-year distant recurrence data in these studies follow a similar pattern to the studies of ET monotherapy. MammaPrint, for which there were no studies of ET monotherapy, assigned 38–48% of patients to the low-risk group, while freedom from distant recurrence at 10 years ranged from 79% to 95% in the low-risk group and 54–81% in the high-risk group.33,53,54 Further details of the prognostic data can be found in Appendix 4.

As described in Results of clinical review: overview, the ongoing OPTIMA study34 is a RCT of Prosigna test-directed chemotherapy use versus standard chemotherapy use. The review did not identify any published results of OPTIMA so far.

_¶1

Observational data: prospective use of Oncotype DX

Several publications report observational studies or registry data for the prospective use of Oncotype DX in clinical practice. In these studies, the Oncotype DX result likely influenced the use of chemotherapy and therefore outcomes. As such, these data have limited use in comparing outcomes between test groups (prognostic ability), though they do provide large-sample data on real-world outcomes. These studies included the Clalit registry66 in Israel (n = 709), the SEER registry67–69 in the USA (n = 6483), the NCDB70–74 in the USA (n = 25,029) and a few smaller prospective studies.75–77 An overview of results is described here, with full results in Table 53 (see Appendix 5), while data on distant recurrence are shown in Table 7.

_¶1

In terms of distribution using cut-offs of RS < 18 and > 30 (see Appendix 5), across the Clalit66 and SEER67,68 registries, 53–58% were low risk (RS 0–17), 35–36% intermediate risk (RS 18–30) and 7–10% high risk (RS ≥ 30), which is similar to the distribution in the TransATAC study20 (57% low, 32% intermediate, 11% high). A study in younger patients75 (age ≤ 40 years) reported a greater proportion of high-risk patients (33% low, 42% intermediate, 25% high). Using an RS cut-off of > 25, across the Clalit66 and NCDB70 registries plus a German study,77 the distribution was in the range of 81–88% (RS 0–25) and 13–19% (RS ≥ 26). The NCDB also reports the distribution using RS cut-offs of < 11 and > 25 as follows: 24% (RS 0–10), 64% (RS 11–25) and 13% (RS ≥ 26).

_¶2

Distant recurrence data from two sources (Clalit registry66 and the Young Women’s Breast Cancer Study75) are shown in Table 7. Within Clalit,66 using the RS cut-offs of < 18 and > 30, the 5-year DRFI was 97% in low-risk patients (7% chemotherapy use), 94% in intermediate risk (40% chemotherapy use) and 83% in high risk (86% chemotherapy use), with Oncotype DX being significantly prognostic despite the greater chemotherapy use in higher-risk patients (see Table 7). Using the cut-offs of RS < 11 and > 25, 5-year DRFI was 96% (RS 0–10), 96% (RS 11–25) and 87% (RS ≥ 26), with Oncotype DX again being statistically significantly prognostic. In younger patients, both Clalit66 and the Young Women’s Breast Cancer Study75 show a statistically significant prognostic effect (see Table 7). However, in older patients (≥ 70 years), there was no statistically significant prognostic effect on 5-year DRFI in Clalit66 (see Table 7).

The RxPONDER29 prospective RCT of Oncotype DX randomised patients with an RS of 0–25 to chemotherapy plus ET versus ET monotherapy (Table 9). In terms of distant recurrence, the results indicated that chemotherapy had little benefit in post-menopausal patients with an RS of 0–25; the 5-year DRFI was 95.8% with chemotherapy versus 96.6% with no chemotherapy, an absolute difference of 0.8% favouring no chemotherapy (adjusted HR 1.12; 95% CI 0.82 to 1.52; p = 0.49). Conversely, there was a benefit of chemotherapy in pre-menopausal patients with an RS of 0–25; the 5-year DRFI was 96.3% with chemotherapy versus 93.9% with no chemotherapy, an absolute difference of 2.4% favouring chemotherapy (adjusted HR 0.64; 95% CI 0.43 to 0.95; p = 0.026).29 Similar data are presented for 5-year DRFS and IDFS, again showing a statistically significant benefit of chemotherapy in pre-menopausal patients, but not in post-menopausal patients (see Table 9).

_¶1

A test for interaction was reported between RS (within the range 0–25) and the effect of chemotherapy on IDFS; no interaction test was reported for distant recurrence. The test did not show a statistically significant interaction across all patients (HR for interaction 1.02; 95% 0.98 to 1.05; p = 0.35), with similar non-significant results in the pre-menopausal and post-menopausal subgroups; (see Table 9).29 Separate data on the effect of chemotherapy on IDFS were also presented within smaller RS ranges (RS 0–10, 11–15, 16–20, 21–25; and 0–13 and 14–25). However, there was no clear pattern or trend in the HR for the effect of chemotherapy, within either the pre-menopausal or post-menopausal groups (see Table 9). This indicates no statistically significant predictive effect within the RS 0–25 group, though RxPONDER cannot provide data on whether there is a predictive effect between the RS 0–25 and RS 26–100 groups.

_¶2

Prediction of chemotherapy benefit: reanalysis of cohort (MammaPrint)

In terms of prediction of CT benefit, the only data identified for MammaPrint were a reanalysis of two cohorts (Mook et al., 2009)33 presenting an interaction test between MammaPrint score and effect of chemotherapy on BCSS. The adjusted interaction test had a non-significant p-value of 0.95 (Table 10).

_¶1

Chemotherapy effect within groups: prospective randomised controlled trial of MammaPrint (MINDACT)

As noted in Results: prognostic ability, the prospective RCT of MammaPrint (MINDACT)30 randomised patients to chemotherapy versus no chemotherapy if they had a discordant genomic risk (via MammaPrint) and clinical risk (via mAOL). Data were presented for the effect of chemotherapy versus no chemotherapy on outcomes for the clinical high-risk, MammaPrint low-risk, LN+, HR+ HER2− subgroup. However, data were not analysed for the clinical low-risk, MammaPrint high-risk group due to small numbers of LN+ patients.30 This is consistent with the company’s focus on the clinical high-risk group in the Agendia submission to NICE.85

_¶1

Within the clinical high-risk, MammaPrint low-risk, LN+, HR+, HER2− subgroup, 8-year DMFS was 91.2% with chemotherapy versus 89.9% with no chemotherapy, an absolute difference of 1.3% favouring chemotherapy,30 with a non-significant HR (HR 0.84; 95% CI 0.51 to 1.37; p = NR; Table 11). Similar data for this subgroup are presented for 8-year DMFI, DFS and OS, though no HRs were presented for the effect of chemotherapy for these outcomes (see Table 11).30

_¶2

The DMFS HR (above) indicates that the effect of chemotherapy in clinical high-risk, MammaPrint low-risk patients was not statistically significant, but the point estimate was in favour of chemotherapy. Since all patients in the clinical high-risk, MammaPrint high-risk group were offered chemotherapy, it was not possible to determine from MINDACT whether MammaPrint was predictive for chemotherapy benefit.

_¶3

Effect of chemotherapy within recurrence score groups: registry data (Oncotype DX)

Several publications report registry data for the prospective use of Oncotype DX in clinical practice, with outcomes per risk group for patients with and without chemotherapy. These studies included analyses of the Clalit registry66,79 in Israel (n = 709), the SEER registry80 in the USA (n = 2588) and the NCDB71–74,81–84 in the USA (n = 28,591). However, use or non-use of chemotherapy was not randomised, and it may correlate with clinical factors which affect outcomes; therefore, the interpretation of the data on effect of chemotherapy from these studies should be approached with caution. An overview of results is described here, with full results in Table 54 (see Appendix 6). Data on distant recurrence are shown in Table 12, while Table 13 presents data on post-menopausal or older age groups, for comparison with the RxPONDER findings in post-menopausal patients.

_¶1

Data on distant recurrence for chemotherapy versus no chemotherapy are only reported for the Clalit registry66,79 (SEER only reports BCSS while NCBD only reports OS). Within Clalit (see Table 12), using the cut-offs of RS < 18 and > 30, the relationship between Oncotype DX risk group and effect of chemotherapy was unclear, with the absolute difference in 5-year DRFI favouring chemotherapy for the intermediate-risk group (difference 8.7%, p = 0.019) but favouring no chemotherapy for the low-risk (4.8%, p = 0.245) and high-risk (8.0%, p = NR) groups. However, using the cut-offs of RS 11 and 25, there appeared to be a trend towards a greater effect of chemotherapy in high-risk groups, with the absolute difference in 5-year DRFI favouring no chemotherapy in the low-risk group (13%, p = NR) but favouring chemotherapy in the intermediate-risk (3.4%, p = NR) and high-risk (17.8%, p = 0.017) groups. Across all patients with an RS of ≤ 25 (irrespective of age or menopausal status), the difference in 5-year DRFI was 2.1% favouring chemotherapy, though this was not statistically significant (p = 0.521). Data for all outcomes and subgroups are presented in Appendix 6.

Of the 12 Oncotype DX studies, 2 UK studies36,37,40 and 4 European studies43,45–47 presented data by Oncotype DX risk groups (Table 15). Five studies36,37,40,45–47 used RS 18 and 30 cut-offs, while two studies used the newer cut-offs of RS 11 and 25 (one study43) or RS 13 and 25 (one study36,40).

_¶1

Among the studies that used the cut-offs of RS 18 and 30,36,37,40,45–47 the net change in chemotherapy recommendations or decisions (pre-test to post-test) was as follows: a decrease of 20%,46 68%,40 91%45 and 93%37 in the RS 0–17 risk group; a decrease of 19%,46 35%,40 37%37 and 54%45 in the RS 18–30 risk group; and either a 17%37 decrease, no change,45,46 or a 1.7%40 increase in the RS > 30 risk group.

_¶2

In the study that used cut-offs of RS 11 and 25,43 the net change in chemotherapy recommendations (pre-test to post-test) was as follows: 52% decrease in the RS < 11 risk group; 18% decrease in the RS 11–25 risk group; and 0% change in the RS > 26 risk group. In the study that used cut-offs of RS 13 and 25 (UK),36,40 the net change was as follows: 67% decrease in the RS 0–13 risk group; 56% decrease in the RS 14–25 risk group; and 5% increase in the RS 26–100 risk group. This study also reported results for pre- and post-menopausal subgroups and, within the post-menopausal subgroup, pre- and post-publication of the RxPONDER results.

_¶3

Conclusions for decision impact data

The net changes in the percentage of patients with a chemotherapy recommendation or decision (pre-test to post-test) among the UK studies were reductions of 28–75% across five Oncotype DX studies.35–40 The net changes across European (non-UK) studies41–47 were reductions of 12%42 to 73% for Oncotype DX. Within studies reporting data by Oncotype DX risk group, there were greater reductions in chemotherapy recommendations in the low-risk and intermediate-risk groups than in the high-risk groups.

_¶1

No studies (or subgroups) were identified which assessed HRQoL or anxiety associated with the use of tumour profiling tests in a LN+ population. Therefore, a brief summary of such studies in a LN0 or mixed nodal status population is provided below (these were all included in the previous review by Harnan et al.10 for NICE DG34 and no subsequent studies were identified). The studies described below are not counted as included studies for the purposes of the PRISMA flow chart; see Appendix 2.

_¶1

Oncotype

Of two studies of Oncotype DX in LN0/LN+ patients in the USA, one (Evans et al., 2016)86 reported no difference between pre- and post-test values on the Impact of Events Scale (p = 0.09) and no statistically significant interaction with RS risk group. The other (Lo et al., 2010)87 reported a statistically significant improvement in overall State-Trait Anxiety Inventory (STAI) score between pre- and post-test values (p = 0.007), but no statistically significant change in HRQoL measured via Functional Assessment of Cancer Therapy – Breast cancer (FACT-B) or Functional Assessment of Cancer Therapy – General (FACT-G) (p = 0.55 and p = 0.49, respectively).

_¶1

MammaPrint

A study of MammaPrint (Retel et al., 2013)88 which included LN0/LN+ patients screened for MINDACT in the Netherlands used Lynch’s distress scale and Lerman’s Cancer Worry Scale, and reported statistically significantly higher distress when the genomic test failed; when the patient was categorised as high risk by both clinical scoring and MammaPrint; and in patients with discordant results when the treatment matched the MammaPrint risk but not the clinical risk. Only patients with high clinical risk and no genomic test result, or high clinical risk and high genomic risk, had a statistically significant decrease in HRQoL via FACT-B.

_¶1

EndoPredict

One study of EndoPredict89 in LN0/LN+ patients in England reported a statistically significant decrease in anxiety (measured via the STAI) for those whose treatment decision changed from chemotherapy to no chemotherapy on the basis of EndoPredict (p = 0.045), and an increase in anxiety (via the STAI) for those whose treatment decision changed from no chemotherapy to chemotherapy (p = 0.001).

_¶1

Prosigna

Two studies assessed Prosigna in LN0 patients in Spain (Martin et al., 2015)90 and Germany (Wuerstlein et al., 2016).91 In both studies, state anxiety reduced significantly in low-risk patients (p < 0.001 and p = 0.008) but not in the intermediate- or high-risk groups. Both studies reported FACT-G; one90 reported no change in overall scores, whereas the other91 reported a statistically significant change in emotional and physical well-being (p = 0.030, p = 0.005, respectively).

_¶1

No studies of HRQoL or anxiety were identified in a LN+ population. Across studies undertaken in a LN0 or mixed population, some reported a significant improvement in anxiety before and after testing, while others reported no significant change in anxiety or HRQoL. Patients reported a decrease in anxiety after a low-risk test result or when their treatment was downgraded to no chemotherapy post-test, but an increase in anxiety when treatment was upgraded to chemotherapy, or after scoring high risk both on the test and clinical measures. It is unclear how far the results of these studies can be generalised to a LN+ population.

_¶1

Cost-effectiveness

This chapter presents a systematic review of published economic evaluations of tumour profiling tests to guide treatment decisions in people with ER+, HER2−, LN+ early breast cancer (see Review of existing economic analyses), a summary and critique of the economic models submitted to NICE by the test manufacturers (see Review and critique of economic analyses of tumour profiling tests submitted by the test manufacturers) and the methods and results of an independent economic analysis undertaken by the EAG (see Independent External Assessment Group economic analysis). A discussion of the key issues around the cost-effectiveness of the tumour profiling tests is presented in Discussion. Details of the review and planned economic analyses can be found in the final EAG protocol (www.nice.org.uk/).

_¶1

Cost-effectiveness review: methods

Systematic searches were undertaken to identify existing economic evaluations of tumour profiling tests to guide treatment decisions in people with ER+, HER2−, LN+ early breast cancer. The review includes studies identified within the previous review undertaken to inform NICE DG34 (Harnan et al.10) as well as more recent studies published since February 2017 (the cut-off date for the search applied in Harnan et al.10). The review was undertaken with the purpose of exploring methodological choices and their potential relevance to the current decision problem, rather than to assess the results of the published economic evaluations or the potential sources of bias which might affect these.

_¶1

A systematic search was undertaken to identify all economic evaluations of the four tumour profiling tests listed in the NICE scope11 (Oncotype DX, EndoPredict, MammaPrint and Prosigna) for breast cancer.

_¶2

Literature searching for economic evaluation studies was undertaken in May 2023 in the following electronic databases:

_¶3

The search strategies comprised medical subject heading (MeSH) or Emtree Thesauri terms and free-text synonyms for: (1) ‘tumour profiling tests’ and ‘breast cancer’ and (2) ‘breast cancer’ only. Searches for all four tests were limited by publication date from 2017. Searches were translated across databases and were not limited by language. The search strategies are presented in Appendix 1. Search filters designed to identify economic evaluations and reviews were used in MEDLINE and other databases, where appropriate. Reference and citation searching of included papers was also undertaken. In addition, economic studies listed in the Cytel CEA report85 [provided as part of the Agendia company submission (CS)], the Exact Sciences CS23 and RFI documents provided to NICE by Veracyte26 and Myriad25 were checked to ensure that no relevant studies had been missed by the electronic searches.

_¶4

In order to be considered potentially relevant for inclusion in the review, studies were required to meet all of the following criteria:

_¶5

Cost-effectiveness review results: summary of included studies

A PRISMA flow diagram summarising study selection is presented in Appendix 7. Following de-duplication, the electronic searches identified a total of 404 studies. Of these, 65 studies were deemed to be potentially includable and full texts were obtained for further scrutiny. Five of these studies met the inclusion criteria for the review.10,92–95 Seven further studies96–102 which were included in the previous systematic review10 reflected a LN+ population and were also included in this review. No additional studies were identified from hand-searching the reference lists for the systematic literature reviews (SLRs) reported in the submission from Exact Sciences,23 the Cytel CEA report85 or from the information provided to NICE by Veracyte and Myriad.25,26 The scope addressed within the 12 included studies and the key aspects of the modelling approaches are summarised Tables 16 and 17, respectively.

_¶1

The 12 included economic studies were undertaken to reflect a range of settings, including the UK, the USA, Canada and Germany. Most of the included studies adopted a direct healthcare perspective. Where reported, the time horizon ranged from 25 years to the patient’s remaining lifetime. Ten of the 12 included studies reflected an exclusively LN+ population or reported separate subgroup analyses for women with LN+ disease; the remaining two studies93,95 included mixed cohorts in which the majority of patients were reported to have LN+ disease. Where reported, the modelled populations range between the ages of 56 and 62 years for most studies. All but one of the included studies93 evaluated Oncotype DX. EndoPredict, Prosigna and MammaPrint were each included in less than half of the included studies. Across all studies, the comparator was consistently either current decision-making (i.e. no tumour profile testing) or chemotherapy for all patients. None of the studies reported incremental cost-effectiveness ratios (ICERs) comparing tumour profiling tests against each other.

_¶2

With the exception of one study,101 all of the included economic analyses adopted a cohort-level hybrid modelling approach comprising a decision tree to determine genomic risk classification and a state transition (Markov) component to estimate long-term outcomes. The cycle lengths applied in the Markov models ranged from 1 month to 1 year. The Markov models typically included three key health states: (1) relapse-free; (2) DM; and (3) dead. Several models also included further health states describing the impact of short- and/or long-term complications associated with chemotherapy, including nausea/vomiting or other toxicity; febrile neutropenia (FN); acute myeloid leukaemia (AML); heart failure (HF); and chronic myeloid leukaemia (CML). One model included a separate health state for LR.97 The majority of models which evaluated Oncotype DX assumed that this test is predictive of chemotherapy benefit, whereby the relative treatment effect of CET versus ET alone is assumed to differ according to Oncotype DX RS. Only one study95 reported an analysis which included an assumption of predictive benefit for MammaPrint and Prosigna. None of the studies included an assumption of predictive benefit for EndoPredict. There was variation among the included models regarding assumptions about the extent of chemotherapy use with and without tumour profiling tests – some studies compared tumour profiling tests against a strategy of chemotherapy for all, while others applied estimates of the proportion of patients receiving chemotherapy with/without testing from published literature and/or from routine data.

_¶3

Only one of the included studies included an analysis of all four tumour profiling tests listed in the final NICE scope11 for this appraisal (Harnan et al.10). As newer relevant clinical evidence has been published since this economic model was developed – in particular, RxPONDER29 and longer-term follow-up data from MINDACT30 – and because treatment pathways for breast cancer have changed since the publication of NICE DG34,13 none of the existing published studies identified by the review provide a sufficient basis for informing the current appraisal.

_¶4

Review and critique of economic analyses of tumour profiling tests submitted by the test manufacturers

This section provides a summary and critique of the economic analyses submitted by the test manufacturers. Executable economic models were submitted to NICE by Exact Sciences (Oncotype DX) and Agendia (MammaPrint). No submissions were received from Myriad (EPclin) or Veracyte (Prosigna).

_¶1

Summary of economic analysis submitted by Exact Sciences

In May 2023, Exact Sciences submitted an executable economic model and an accompanying written submission which details the methods and results of the model (hereafter referred to as the Exact Sciences CS.23). The company also provided responses to clarification questions from the EAG in June 2023,103 which included an updated version of the economic model. The executable model is an adaptation of the earlier economic analysis reported by Berdunov et al.92 (see Cost-effectiveness review results) which in turn was based largely on the EAG’s model developed to inform NICE DG34 (Harnan et al.10). The Exact Sciences model differs from the model developed to inform DG34, in that it includes evidence on test risk classifications and DRFI from the RxPONDER trial,78 as well as other updated parameter estimates which are intended to reflect changes in the downstream breast cancer pathway since DG34 was published in 2018.

_¶1

The Exact Sciences CS23 presents cost-effectiveness estimates for Oncotype DX versus clinical–pathological tools alone in terms of the incremental cost per QALY gained from the perspective of the NHS and PSS in England over a 45-year (lifetime) time horizon. The model applies a 6-month cycle length and includes half-cycle correction to account for the timing of events. Health outcomes and costs are discounted at a rate of 3.5% per annum. Costs are valued at 2020 prices.

_¶2

The Exact Sciences base-case analysis is presented across three populations: (1) the overall ER+, HER2−, LN+ (one to three nodes) early breast cancer population; (2) pre-menopausal women with ER+, HER2−, LN+ early breast cancer; and (3) post-menopausal women with ER+, HER2−, LN+ early breast cancer. Men with early breast cancer are not considered in the model. Comparisons of Oncotype DX versus other tumour profiling tests (MammaPrint, EPclin and Prosigna) are not included in the company’s base-case analyses but are included in additional exploratory analyses presented in the CS.

_¶3

The general structure of the Exact Sciences model is similar to the model used to inform DG34.13 The model structure adopts a hybrid approach comprising an initial decision tree component which stratifies patients according to their genomic risk based on the tumour profiling test result, followed by a Markov component which estimates long-term health outcomes and costs conditional on genomic risk and whether the patient receives adjuvant CET or ET alone. The decision tree component includes three levels (low, intermediate and high risk), although DRFI is assumed to be the same for patients with low risk (RS < 13) and intermediate risk (RS 13–25). The long-term Markov model includes four health states: (1) recurrence-free; (2) DM; (3) AML; and (4) dead. LR is captured as a transient event in a proportion of patients who develop DM. Within the base-case analysis and all scenario analyses, the model assumes that Oncotype DX is predictive of chemotherapy benefit, with an HR for distant recurrence for CET versus ET alone of 0.89 assumed in the Oncotype DX RS 0–25 category and an HR of 0.59 assumed in the RS > 25 category for the overall LN+ population.32,78 The equivalent HRs for CET versus ET in the RS 0–25 category for the pre-menopausal and post-menopausal subgroups are 0.64 and 1.12, respectively; the HR of 0.59 is also applied to the RS > 25 category in these subgroup analyses.32,78

_¶4

QALYs are modelled as a function of whether patients receive CET or ET alone, which subsequently determines the risk of DM, AML and death. The model includes a short-term disutility associated with chemotherapy-related toxicity in the first model cycle which corresponds to once-only QALY loss of 0.038 for all patients receiving chemotherapy. A once-only QALY loss is also applied for patients experiencing LR in any model cycle. QALYs are adjusted for increasing age using utility multipliers based on Ara and Brazier104 which are aggregated into age bands (one band for patients aged < 30 years, one band for those aged > 85 years, and 5-year bands for those aged 30–85 years).

_¶5

The model includes resource costs associated with:

_¶6

The Exact Sciences model assumes that 11% of women in the target population will have an Oncotype RS of > 25. This estimated proportion is based on the number of women who were screened for eligibility for entry into RxPONDER who had an RS > 25 as the numerator (N = 1035) and the overall number of women who were registered for screening in RxPONDER as the denominator (N = 9383).29 Kalinsky et al.29 report that a total of 4300 women were excluded from RxPONDER and that these women were excluded for various reasons: ineligible (N = 164); no RS (N = 84); had RS > 26 (N = 1035); declined to participate (N = 2372); had recurrence (N = 23) or had other or unknown reason (N = 622). It is likely that some of the 3265 women who were excluded for other reasons would actually have had an RS of > 25. The EAG believes that the denominator for this calculation should reflect those women who were eligible for the trial and for whom a known Oncotype DX test result was available. As such, the company’s model likely underestimates the probability that a woman will have an RS of > 25. The EAG believes that within the RxPONDER trial, the proportion of women who had an RS of > 25 lies somewhere between a minimum value of 0.11 [assuming that all other excluded patients have an RS of < 25 (1035/9383)] and a maximum value of 0.17 [including only patients with a known RS in the calculation (1035/6118)]. This range may vary across study populations.

_¶1

In response to a request for clarification from the EAG,103 the company highlighted that RxPONDER is an independently conducted study and that the company only had access to the information provided in the study protocol and the trial publication.29 The company’s response also highlights that the proportion of women with an RS of > 25 in RxPONDER is broadly similar to that in Stemmer et al.122 and SEER.80

_¶2

Uncertainty around relevant cut-offs for National Institute for Health and Care Excellence decision-making

The Exact Sciences CS23 does not clearly state how the company intends the Oncotype DX test results to be used in clinical practice. Page 19 of the CS states that ‘The RS result is typically defined as low (0 to 25) or high (26–100) however clinicians may apply different thresholds based on their interpretation of the evidence’. This suggests that Oncotype DX would be used as a two-level test (giving results as either low risk or high risk) based on an RS cut-off of 25. However, Table 2B of the Instructions For Use document for Oncotype DX123 refers to chemotherapy benefit by RS in LN+ patients based on three levels: low – RS 0–17; intermediate – RS 18–30; and high – RS 31–100. The company’s economic model is based specifically on the cut-offs applied in RxPONDER29 (RS 0–25 and RS > 25). This is the only scenario in which the cost-effectiveness of Oncotype DX has been evaluated within the CS.

_¶1

In response to a request for clarification from the EAG,103 the company stated that ‘the validated RS result cut-offs used in the RxPONDER study should be used to categorise patients according to their risk of distant recurrence’ and that the Instructions For Use document will be updated to reflect this. At the time of writing, this document had not been updated and so there remains some uncertainty around the most relevant cut-offs for NICE decision-making.

_¶2

Model errors and other minor implementation issues

The EAG double-programmed the deterministic version of the Exact Sciences model to verify its implementation. The EAG was able to generate almost identical estimates of life years gained (LYGs), QALYs and costs for both the Oncotype DX and the comparator group in the LN+ population. Overall, the EAG considers the Exact Sciences model to be well programmed and free from major errors. During this double-programming exercise, the EAG identified the following minor issues:

_¶1

Additional analyses undertaken using the Exact Sciences model

As part of their response to clarification questions from the EAG,103 the company provided an updated version of the model which addresses errors (a) and (b) listed in External Assessment Group critique of the Exact Sciences model. The company’s written response103 noted that the presence of these errors had a negligible impact on the model results. The EAG further amended the company’s revised model to also address issues (c), (d), (e), (g) and (i). Issue (f) could not be easily resolved in the company’s existing model structure and issue (h) was not resolved as it applies only to the company’s exploratory analyses of other tumour profiling tests. The inclusion of these model amendments by the EAG had only a small impact on the results and did not affect the company’s original base-case economic conclusions, with Oncotype DX dominating clinical–pathological tools alone in the overall LN+ population and in the post-menopausal LN+ subgroup, and clinical–pathological tools alone dominating Oncotype DX in the pre-menopausal LN+ subgroup.

_¶1

Summary of economic analysis submitted by Agendia

In May 2023, Agendia submitted an executable model and an accompanying written document prepared by Cytel which details the methods and results of the model (hereafter referred to as the Cytel CEA report85). The company also provided responses to clarification questions from the EAG in May and June 2023,126 which included an updated version of the economic model. The Cytel CEA report (p. 5) states that the objective of the report is

_¶1

The Cytel CEA report (page 30) also states ‘Agendia is seeking reimbursement consideration to include LN+ [1–3, nodes] patients’.

_¶2

The Cytel CEA report85 presents cost-effectiveness estimates for MammaPrint in terms of the incremental cost per QALY gained compared with three other tumour profiling tests and usual care (current decision-making using clinical–pathological tools alone) from the perspective of the UK NHS and PSS over a 45.5-year (lifetime) horizon. The model includes a cycle length of 6 months and includes half-cycle correction. Health outcomes and costs are discounted at 3.5% per annum. Costs are valued at 2021–2 prices, including uplifting of unit cost estimates using Hospital and Community Health Services (HCHS) indices and NHS Cost Inflation Indices (NHSCII) where necessary.

_¶3

The Cytel CEA report85 presents a base-case analysis and five additional scenario analyses. The modelled population and the comparators under consideration differ between these analyses (Table 21). The base-case analysis compares the use of MammaPrint versus Oncotype DX, EPclin, Prosigna and usual care in a population of women who are ER+ and HER2−, who have either LN0 or LN+ (one to three nodes), and who are considered to be clinical high risk according to NPI or mAOL. In the base-case analysis, at model entry, the population is assumed to be aged 58.9 years based on the NHS England (NHSE) Access Scheme Dataset, which reflects LN0 patients who received Oncotype DX testing following DG10.128 The modelled population is intended to reflect both pre- and post-menopausal women, although post-menopausal women are considered as a specific subgroup in Scenario 4 (see Table 21). Men with early breast cancer are not reflected in the modelled population.

_¶4

The general structure of the Agendia model is similar to the model used to inform DG34.13 The model uses a hybrid approach and is comprised of an initial decision tree component which stratifies patients according to their genomic risk based on the tumour profiling test, followed by a Markov component which estimates long-term outcomes and costs conditional on genomic risk and whether the patient receives adjuvant CET or ET alone. The decision tree component of the model stratifies patients into either high risk or low risk for 2-level tests, or high, intermediate or low risk for 3-level tests, and determines whether the patient receives adjuvant chemotherapy. The long-term Markov model includes four health states: (1) recurrence free; (2) DM; (3) AML; and (4) dead. The model also includes separate tunnel states to reflect the impact of LR (prior to DM), which is assumed to impact on QALYs and costs, but does not affect the patient’s underlying health state or mortality risk. The benefit of chemotherapy is modelled using RRs applied to the risk of DM with ET alone. Within the base-case analysis and Scenarios 1–4, the company’s model assumes that MammaPrint is predictive of chemotherapy benefit, with RRs for DM for CET versus ET alone of 1.0 and 0.38 applied to the genomic low-risk and genomic high-risk groups, respectively. In Scenario 5, which reflects a pure LN+ population, RRs of 0.97 and 0.28 are applied in the genomic low-risk and high-risk groups. All other tests are assumed to be prognostic only, except for Oncotype DX in Scenario 4 (see Table 21). This is a key assumption which favours MammaPrint over all of the other comparator tests and usual care.

_¶5

QALYs are modelled as a function of whether patients receive adjuvant chemotherapy and the long-term trajectory of patients through the Markov model health states. Lower utility values are applied to patients receiving CET versus ET alone for 3 years, which are intended to represent the disutility resulting from toxicity associated with adjuvant chemotherapy (net loss per patient treated with CET vs. no ET alone = 0.29 QALYs). The model applies comparatively lower utility values to the DM and AML states than the recurrence-free state. The model also includes a disutility value associated with LR of −0.11.112 The model includes age adjustment of utility values based on Ara and Brazier.104

_¶6

The Agendia model assumes that MammaPrint is predictive of chemotherapy benefit, and that the other tumour profiling tests are prognostic only (with the exception of Oncotype DX in Scenario 4, see Table 21). In the base-case scenario, the model applies an RR for distant recurrence for CET versus ET of 1.00 (i.e. no chemotherapy benefit) for patients who are MammaPrint low-risk and an RR of 0.38 (i.e. substantial chemotherapy benefit) for patients who are MammaPrint high-risk. In the pure LN+ subgroup (Scenario 5), RRs of 0.97 and 0.28 are applied. The company’s justification for this assumption of predictive benefit is based on the non-significant HR obtained from an adjusted Cox model fitted to MINDACT IPD by the company to estimate the effect of CET versus ET alone in HR+, HER2− women (including both LN0 and LN+) who were clinical high risk and genomic low risk (HR 0.74, 95% CI 0.43 to 1.28).126 The company’s model assumes that because this HR is not statistically significant, it should be interpreted as being equivalent to chemotherapy having no effect (i.e. HR = 1.00), and because chemotherapy is known to be clinically effective overall,16 a considerably greater treatment effect must therefore apply to the clinical high-risk genomic high-risk group who were not randomised in MINDACT. The logic underpinning the company’s calculations of chemotherapy treatment effects by genomic risk group in the base-case analysis is as follows:

_¶1

The EAG notes that the updated MINDACT publication by Piccart et al.30 reports an HR for DMFS for the overall clinical-high genomic-low risk group of 0.66 (95% CI 0.48 to 0.92) and an HR for DRFI in this same population of 0.66 (95% CI 0.46 to 0.95). After excluding the 21% of participants in this group that are not relevant to the decision problem, the HR for the HR+, HER2− population (regardless of nodal status) provided by the company was 0.74 (95% CI 0.43 to 1.28). The wider CI is expected due to the reduction in sample size and, although no longer statistically significant, the point estimate of 0.74 is still the best estimate of the treatment effect for this group. Based on the Z-score (calculated on a log scale), this implies that 86% of individuals in this group are likely to have a beneficial response to chemotherapy (HR < 1.0). If the estimated HR of 0.74 is applied directly (rather than assuming that it is 1.0), then the HR for the 39% of patients who are MammaPrint high-risk would be 0.79 (i.e. 0.61 × 0.74 + 0.39 × 0.79 = 0.76). This is similar to the HR for the MammaPrint low-risk group.

_¶2

Piccart et al.30 also report an estimate of the relative treatment effect on DMFS specifically for the clinical high genomic low LN1–3 population, including those with HER2+ and ER− or PR− disease (HR 0.84, 95% CI 0.51 to 1.37; see Table 11). This analysis indicates a comparatively lesser effect of chemotherapy in women with LN1–3 which is again not statistically significant, and the sample size is small (N = 658). It is unclear why the point estimate of the effect of chemotherapy is less pronounced in this subgroup.

_¶3

Overall, the EAG considers that there is insufficient evidence from the MINDACT trial to support the argument that MammaPrint is predictive of chemotherapy benefit, and there is no external evidence to inform the relative treatment effect of chemotherapy in the clinical high genomic high group. As such, the EAG believes that the company’s interpretation of the results of their IPD analysis are flawed and that it is more reasonable to apply the same treatment effect estimate for chemotherapy to both the MammaPrint low-risk and MammaPrint high-risk groups.

_¶4

The Cytel CEA report85 (page 30) states that ‘A disutility associated with short-term AEs related to adjuvant chemotherapy is applied once during the first model cycle only (whilst the patient is receiving treatment)’. This is not an accurate description of the assumptions employed in the Agendia model. Instead, the model applies treatment-specific utility values for patients receiving adjuvant CET and for patients receiving ET alone, based mostly on values reported by Lidgren et al.,111 with higher utility scores applied to the ET group. These differences are assumed to persist for 3 years, although the source of the assumption of a 3-year disutility is unclear. The utility values applied for patients receiving CET or ET alone in the Agendia model are summarised in Table 25. The model suggests that a patient who is treated with CET who survives for 3 years without experiencing relapse will lose 0.29 QALYs compared with an equivalent patient who receives ET alone. This modelled QALY loss is substantially larger than the QALY losses applied in the majority of other economic models of tumour profiling tests in LN+ women included in the EAG’s review and the Exact Sciences Model (see Review of existing economic analyses, Table 26 and Summary of economic analysis submitted by Exact Sciences). The only studies in which a similar or higher chemotherapy-related QALY loss is applied are those reported by Vanderlaan et al.102 and Wong et al.101 In both these studies, QALY losses associated with adjuvant chemotherapy appear to be based on assumptions rather than empirical evidence. Overall, the EAG has concerns that the QALY loss associated with chemotherapy in the Agendia model has likely been overestimated. The EAG also notes that the Agendia model includes age adjustment of utility values based on Ara and Brazier;104 the EAG believes that it would be more appropriate to use more recent estimates by Hernández-Alava et al.113

_¶1

Concerns regarding costs

The Agendia model includes the same estimates of mortality risk and HRQoL for AML as those used in the EAG’s model developed to inform DG34.10 However, the Agendia model applies a substantially higher lifetime cost estimate for AML compared with the DG34 model (Harnan et al.10 cost = £10,600 vs. Russell-Smith et al.137 cost = £132,039). This cost estimate has been taken from an economic modelling study of gemtuzumab ozogamicin plus SC chemotherapy in people with de novo CD33-positive AML. The population included in the analysis reported by Russell-Smith et al. is not consistent with the patient population reflected in the Agendia model, as the latter has secondary (therapy-related) AML. In addition, as the AML lifetime cost estimate has been updated in the Agendia model, but the mortality risk and health utility value have not, this implies that the treatment of AML has become substantially more expensive without any improvement in health outcomes. The company’s approach favours all tumour profiling tests which reduce the incidence of AML as a consequence of fewer patients receiving chemotherapy. The EAG considers that it would be preferable for the model to reflect mortality risks, QALYs and costs associated with current therapies estimated in patients with secondary AML.

_¶1

In addition, the company has applied inflation indices to uplift the prices of all four tumour profiling tests considered in the model. However, since DG3413 was published in 2018, the list prices of EPclin and Oncotype DX have not changed and the marginal cost per Prosigna test has decreased (see Independent External Assessment Group economic analysis, Table 37).

_¶2

Model errors

The EAG identified three sets of programming errors in the Agendia model:

_¶1

Additional exploratory analysis undertaken by the External Assessment Group

The EAG undertook an additional analysis which attempts to address six of the issues identified in the EAG’s critique:

_¶1

The results of this analysis are shown in Table 27 together with the results of the company’s deterministic base-case analysis. The EAG’s additional analysis suggests that MammaPrint leads to a small reduction in survival, a small increase in QALYs and a small decrease in costs; hence, MammaPrint remains dominant. However, the EAG remains concerned that there are still some minor errors in this model – the EAG ran a scenario in which no patients receive chemotherapy in either the MammaPrint or usual care groups and the model still suggests that MammaPrint generates additional QALYs over usual care – this clearly reflects an error. The EAG also notes that this model does not reflect all of the EAG’s preferred assumptions and evidence sources (see Independent External Assessment Group economic analysis). As such, the EAG believes that the results of this reanalysis should be interpreted with caution.

_¶2

Scope of the External Assessment Group economic analysis

The EAG developed a health economic model to assess the cost-effectiveness of Oncotype DX, Prosigna, EPclin and MammaPrint versus current decision-making. The model was programmed using Microsoft Excel. The scope of the EAG’s model is summarised in Table 28. The model assesses the health outcomes and costs associated with each tumour profiling test and current decision-making over a lifetime horizon (up to age 100 years) from the perspective of the NHS and PSS. All costs and health outcomes are discounted at a rate of 3.5% per annum. The analysis adopts a formal price year of 2022–3, including uplifting of older cost estimates using inflation indices151 where necessary.

_¶1

Population

Overall, the population reflected in the economic model relates to women with ER+/PR+, HER2−, LN+ (one to three nodes) early breast cancer. This is consistent with the final NICE scope.11 The following issues should be noted with respect to the modelled population:

The EAG’s model is based on the economic analysis used to inform DG34,13 together with updated evidence on the tumour profiling tests and updated evidence and assumptions regarding downstream events, health outcomes and costs. The general model structure is consistent with the majority of studies identified in economic review (see Review of existing economic analyses) as well as the models submitted by Agendia85 and Exact Sciences23 (see Review and critique of economic analyses of tumour profiling tests submitted by the test manufacturers). The EAG’s model is intended to capture the key trade-offs in the use of tumour profiling tests in guiding the decision to receive or forgo chemotherapy. Specifically, the model reflects the benefits associated with adjuvant chemotherapy in terms of the reduction in the risk of developing DM and the avoidance of adverse impacts of relapse on HRQoL, survival and costs, as well as its negative effects, which include short-term toxicities and late effects (AML) and the costs of the adjuvant chemotherapy itself. Within the model, the benefits of the tumour profiling tests are modelled by changing the probability that patients receive chemotherapy. In scenarios in which the test is assumed to be predictive of chemotherapy benefit, the relative treatment effect for chemotherapy versus ET alone differs between genomic risk classification groups.

_¶1

The model takes the form of a hybrid decision tree and long-term Markov model. The decision tree component stratifies patients according to their genomic risk [low, intermediate or high risk for 3-level tests (Prosigna and Oncotype DX using the older RS cut-offs) or low risk or high risk for 2-level tests (Oncotype DX using the newer RS cut-offs, EPclin and MammaPrint)] and according to whether the patient receives chemotherapy conditional on their genomic risk classification (Figure 5). As such, the decision tree determines the distribution of patients across up to six categories:

_¶2

Each of these branches is linked to a long-term Markov model which predicts lifetime QALY gains and costs conditional on the patient’s risk of DM, whether or not they receive chemotherapy, and the magnitude of the treatment effect of chemotherapy on DM.

_¶3

The structure of the long-term Markov submodels is illustrated in Figure 6. Each Markov submodel is evaluated using 6-monthly cycles until the patient cohort has reached age 100. Patients enter the model aged 62 years if post-menopausal, or aged 44 years if pre-menopausal, and the evaluation is continued until the cohort has reached age 100 years. Each Markov submodel includes four mutually exclusive and jointly exhaustive health states: (1) recurrence-free; (2) DM; (3) AML; and (4) dead. Each submodel differs with respect to the patient’s risk of developing DM, as determined by their genomic risk classification and whether or not they receive chemotherapy. For all Markov submodels, patients enter the model in the distant recurrence-free state. During each 6-month cycle, patients who are recurrence-free can remain recurrence-free, develop DM, develop AML (if they have previously received adjuvant chemotherapy) or die. Congestive heart failure (CHF) was not explicitly included as a late effect in the model because the EAG’s clinical experts stated that oncologists are generally able to select out those patients who are at risk of this event based on clinical risk factors, baseline cardiac function and biochemical tests.152 Patients who are alive with DM can remain in their current health state, develop AML or die. For patients who have developed AML, the only remaining transition is to the dead state. Patients may die from breast cancer, AML or other causes. Adverse effects of chemotherapy are captured through the inclusion of a short-term (1-year) toxicity-related QALY loss and additional AE management costs applied in the first model cycle, and through the inclusion of the AML state which impacts on survival, HRQoL and costs. The benefit of chemotherapy is modelled through the application of an HR which is applied to the probability of developing DM with ET alone within each genomic risk category. In all scenarios, the use of the tumour profiling test impacts on health outcomes and costs by influencing the probability that a patient receives chemotherapy. In BC1–3, a predictive benefit is assumed for Oncotype DX; hence, the HR for distant recurrence for CET versus ET alone is assumed to differ between the risk classification groups.

_¶4

QALYs gained are estimated by assigning health utility values to each of the Markov submodel health states. The model also includes a short-term QALY loss associated with AEs resulting from the use of adjuvant chemotherapy and a QALY loss associated with the incidence of LR. Health utility values are adjusted for increasing age.

_¶5

The model includes costs associated with: the tumour profiling tests; acquisition and administration of adjuvant chemotherapy and supportive medications, ET, other drug treatments (bisphosphonates and/or ovarian suppression), routine follow-up visits and tests, treatments for LR, treatments for DM, treatments for AML and end-of-life care.

_¶6

The post-test chemotherapy probabilities applied in the model are shown in Table 34. Post-test chemotherapy probabilities were selected based on consideration of the studies included in the systematic review of decision impact studies (see Results: decision impact):

_¶1

Age-specific probabilities of all-cause death were estimated using Office for National Statistics life tables for England (years 2018–20).155 These mortality risks are applied to all women who remain in the recurrence-free state. These probabilities are also used as constraints in the DM and AML states to ensure that the risk of death with DM and AML remain at least as high as the risk of death in the general population in every model cycle.

_¶1

The utility values and QALY losses applied in the EAG’s model are summarised in Table 35. The derivation of each individual utility value/QALY loss is described in further detail in the subsequent sections. Within the economic model, all utility values were adjusted for increasing age using EuroQol-5 Dimensions, three-level version (EQ-5D-3L) estimates for the general population of the UK reported by Hernández-Alava et al.113

_¶1

The model developed to inform DG3485 applied utility values to the recurrence-free and DM health states based a cross-sectional observational study of 361 patients with a previous diagnosis of breast cancer who attended the outpatient clinic at the Karolinska University Hospital in Sweden between April and May 2005 (Lidgren et al.111). Within this study, patients were asked to complete both the EQ-5D-3L and a direct time trade-off question. Patients were then divided into mutually exclusive groups based on their breast cancer disease state: State ‘P’ – first year after primary breast cancer; State ‘R’ first year after recurrence; State ‘S’ – second and following years after primary breast cancer or recurrence and State ‘M’ – metastatic disease. Lidgren et al. report a utility value of 0.824 for patients in State S who were receiving ET and a utility value of 0.685 for patients in State M. These values were applied to the recurrence-free and DM states in the model. The disutility associated with chemotherapy was derived from a previous economic model reported by Campbell et al.112

_¶1

The EAG undertook a further review to identify other potentially relevant studies which have been published since 2017 (the cut-off date in Harnan et al.10). Systematic searches were undertaken to identify studies reporting on HRQoL associated with different health states for women with breast cancer. Searches were undertaken in May 2023 in the following electronic databases:

_¶2

The searches focused specifically on studies which report HRQoL estimates for health states measured and valued using the EQ-5D. The search strategy comprised sensitive MeSH or Emtree Thesauri terms and free-text synonyms for ‘breast cancer’ combined with free-text synonyms for ‘EQ-5D’. The search strategies are presented in Appendix 1. Studies were considered potentially relevant if they reported EQ-5D valuations for both non-metastatic/early breast cancer and DM states, thereby reflecting key health states in the model. Studies which reported disutilities associated with AEs resulting from the use of chemotherapy were also retained for separate consideration. Studies were sifted by title and abstract according to the inclusion criteria. Full texts were retrieved for all potentially relevant studies identified at the title/abstract stage. In order to be considered for inclusion in the review, studies had to meet the following criteria:

_¶3

The searches identified a total of 404 studies. The full texts of 23 studies were retrieved for more detailed review. None of these studies reported EQ-5D-3L estimates for patients with non-metastatic breast cancer receiving ET and for patients with DM. One ‘near miss’ was identified in which the authors reported EQ-5D-3L utility values for patients with early and metastatic breast cancer (Verrill et al.162). This study was a UK cross-sectional study of 299 adult patients with HER2+ early or metastatic breast cancer. The authors report mean EQ-5D-3L values of 0.73 for early breast cancer on treatment post surgery, 0.73 for early breast cancer after completion of adjuvant treatment, and 0.60 for metastatic breast cancer. Given that the population in Verrill et al.162 reflects a HER2+ population, whereas the target population for this appraisal relates to a HER2− population, this study was considered only in sensitivity analyses. As such, the EAG’s model retains the use of Lidgren et al.111 as the primary source of utility values.

_¶4

_¶1

Where sufficient information was available, distribution parameters were characterised using reported standard errors (SEs) or 95% CIs. Where insufficient information was provided, SEs were assumed to be equal to 10% of the mean.

_¶2

Alongside the PSA, the EAG also undertook a number of DSAs to explore alternative evidence sources and assumptions. The following analyses were undertaken across each of BCs 1–7 (where relevant):

_¶3

In addition to these sensitivity analyses, the EAG also estimated the impact of changes in chemotherapy use following the use of tumour profiling testing on the number of infusion chair hours, based on infusion times obtained from the EAG’s clinical advisors (see Appendix 8, Table 55).

_¶4

Model verification methods

A number of approaches were used to ensure the credibility of the EAG’s model. These included:

_¶1

Results of the External Assessment Group economic analysis

This section presents the results of the EAG’s economic analysis. The results of the EAG’s base-case analysis generated using the probabilistic and deterministic versions of the model are presented in Tables 40 and 41, respectively. The results of the DSAs are presented in Table 42. A summary of the model-predicted impact of tumour profiling testing on chemotherapy use, clinical outcomes, costs and net health benefits (NHBs) per 1000 women tested is presented in Table 43. CEACs for each comparison are shown in Appendix 9, Figures 7–13. The results of these analyses are summarised together in the subsequent sections.

_¶1

Oncotype DX versus current decision-making (BC1-4)

The probabilistic version of the model for the pre-menopausal LN+ subgroup suggests that compared with current decision-making, Oncotype DX is expected to result in 0.66 fewer LYGs, 0.18 fewer QALYs and additional costs of £1810 per patient tested. Consequently, Oncotype DX is dominated by current decision-making in this population. These results are driven by the estimated reduction in the use of adjuvant chemotherapy in women who would have benefitted from treatment. Assuming willingness-to-pay (WTP) thresholds of £20,000 and £30,000 per QALY gained, the probability that Oncotype DX generates more net benefit than current decision-making is approximately 0.06. The DSAs indicate that Oncotype DX remains dominated across all analyses, except for DSA23 (cost of chemotherapy doubled).

_¶1

Within the post-menopausal LN+ subgroup, the probabilistic version of the model suggests that compared with current decision-making, Oncotype DX is expected to generate 0.21 additional LYGs, 0.11 additional QALYs and cost savings of £4273 per patient tested. Consequently, Oncotype DX dominates current decision-making in this population, provided the assumption of predictive benefit holds. These results are driven by an estimated reduction in the use of adjuvant chemotherapy in women who would not have benefitted from chemotherapy. Assuming WTP thresholds of £20,000 and £30,000 per QALY gained, the probability that Oncotype DX generates more net benefit than current decision-making is approximately 1.00. The DSAs indicate that Oncotype DX remains dominant across all analyses except for those in which the assumption of a predictive benefit of chemotherapy is removed (DSAs 11–13); within these scenarios, the ICER for Oncotype DX is in the South-West quadrant and ranges from £9772 to £279,599 saved per QALY lost. The analyses of Oncotype DX within the post-menopausal LN+ subgroup based on TransATAC20 using the older RS cut-offs suggest a similar finding – Oncotype DX dominates current-decision-making when a predictive benefit is assumed, but it is dominated by current decision-making when this assumption is removed. These results remain generally consistent across the range of DSAs tested.

_¶2

Prosigna versus current decision-making (BC5)

The probabilistic version of the model suggests that compared with current decision-making, Prosigna is expected to result in 0.06 additional LYGs, 0.03 additional QALYs and additional costs of £1084 per patient tested; the ICER for Prosigna versus current decision-making is expected to be £39,357 per QALY gained. The model suggests that the use of Prosigna will result in a small decrease in the use of chemotherapy, a small reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. Assuming WTP thresholds of £20,000 and £30,000 per QALY gained, the probability that Prosigna generates more net benefit than current decision-making is approximately 0.16 and 0.34, respectively. The DSAs resulted in ICERs ranging from £23,853 per QALY gained to dominated. The DSAs indicate that the ICER is sensitive to the source of test risk classification probabilities and associated DRFI estimates, the HR for chemotherapy, and the costs of adjuvant chemotherapy and downstream treatments for DM.

_¶1

EPclin versus current decision-making (BC6)

The probabilistic version of the model suggests that compared with current decision-making, EPclin is expected to result in 0.13 additional LYGs, 0.06 additional QALYs and additional costs of £231 per patient tested; the ICER for EPclin versus current decision-making is expected to be £4113 per QALY gained. The model suggests that the use of EPclin will result in a small decrease in the use of chemotherapy, a reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. Assuming WTP thresholds of £20,000 and £30,000 per QALY gained, the probability that EPclin generates more net benefit than current decision-making is approximately 0.82 and 0.86, respectively. The DSAs resulted in ICERs ranging from dominating to dominated. The DSAs indicate that the ICER is sensitive to the test risk classification probabilities and associated DRFI estimates, the baseline probability of receiving chemotherapy, the HR for chemotherapy, and the costs of adjuvant chemotherapy and downstream treatments for DM.

_¶1

MammaPrint versus current decision-making (BC7)

The probabilistic version of the model suggests that compared with current decision-making, MammaPrint is expected to result in 0.22 fewer LYGs, 0.07 fewer QALYs and additional costs of £786 per patient tested; hence, MammaPrint is dominated by current decision-making. The model suggests that the use of MammaPrint will result in a large decrease in the use of chemotherapy, an increase in the lifetime probability of developing DM and additional net costs due to the cost of the test. Assuming WTP thresholds of £20,000 and £30,000 per QALY gained, the probability that MammaPrint generates more net benefit than current decision-making is approximately 0.01. The DSAs suggest that MammaPrint is either dominated or results in a South-West quadrant ICER which is less than £30,000 per QALY gained across all scenarios tested.

_¶1

Discussion

The EAG undertook a systematic review of published economic evaluations of tumour profiling tests to guide adjuvant chemotherapy decisions in women with ER+, HER2−, LN+ early breast cancer. A total of 12 studies were included in the review, including five studies identified from the new searches and seven studies which were included in the previous systematic review by Harnan et al.10 The economic models included in the review adopted similar structures based on a hybrid decision tree and state transition approach, built around three core health states which were defined according to the presence or absence of DM and survival status. Only one of the studies (Harnan et al.) included all four tumour profiling tests listed in the final NICE scope for this appraisal.

_¶1

Two of the test manufacturers, Exact Sciences and Agendia, submitted model-based economic analyses to inform the appraisal. The structures of these models are broadly similar to the approaches used in the published economic analyses identified by the EAG’s systematic review. The model of Oncotype DX provided by Exact Sciences presents separate base-case analyses for: (1) pre-menopausal women with LN+ early breast cancer; (2) post-menopausal women with LN+ early breast cancer; and (3) a blended analysis which reflects a mixed pre- and post-menopausal LN+ population. The model is informed by RxPONDER29,78 in women with an Oncotype DX RS of 0–25 and by external data (TransATAC20 and SWOG-881432) for women with an RS of > 25. Pre- and post-test chemotherapy probabilities are based on an unpublished UK decision impact study on the use of Oncotype DX undertaken in women with LN+ early breast cancer.18 All three base-case analyses include an assumption that Oncotype DX is predictive of chemotherapy benefit, with different relative treatment effects for adjuvant chemotherapy versus ET applied to women who are low risk (RS 0–25) and those who are high risk (RS > 25). The company’s model suggests that Oncotype DX dominates current decision-making in post-menopausal women with LN+ disease and that Oncotype DX is dominated by current decision-making in pre-menopausal women with LN+ disease. Within the overall LN+ population, the model suggests that Oncotype DX dominates current decision-making; however, this analysis is misleading as it masks the cost-ineffectiveness of the test in the pre-menopausal subgroup.

_¶2

The model provided by Agendia compares MammaPrint to other tumour profiling tests and usual decision-making across a range of populations, including women with LN0 disease. The company’s analysis includes a separate scenario analysis which focuses on a pure LN+ subgroup. The company’s analyses include an assumption that MammaPrint is predictive of chemotherapy benefit, based on the finding of a non-significant HR for DRFI for chemotherapy versus no chemotherapy for women who are clinical high-risk and MammaPrint low-risk, which was calculated through a reanalysis of IPD for women with HR+, HER2− disease (LN0 or LN+) from the MINDACT trial.30 The company’s submitted model suggests that MammaPrint dominates current decision-making in the LN+ subgroup. The EAG does not consider the company’s assumption of predictive benefit to be a reasonable interpretation of the results of their reanalysis of the MINDACT IPD. In addition, the EAG believes that the company’s model likely overestimates the negative HRQoL impact of chemotherapy toxicity. The EAG also identified some programming errors which affect the model results. The EAG undertook a reanalysis of this model which removes the assumption of predictive benefit, down-weights the chemotherapy-related QALY loss and corrects the programming errors. This reanalysis suggests that MammaPrint leads to a small loss in survival, a small QALY gain and a small cost saving; hence MammaPrint remains dominant. However, the EAG has concerns that this model is still subject to some programming errors and notes that it does not include all of the EAG’s preferred assumptions and evidence sources.

_¶3

The EAG developed a de novo health economic model to assess the cost-effectiveness of Oncotype DX, Prosigna, EPclin and MammaPrint, each versus current decision-making. The economic analysis was undertaken from the perspective of the NHS and PSS and was largely based on the model structure used to inform NICE DG34.10 The EAG’s model adopts a hybrid decision tree and state transition structure. Key updates to the previous version of the EAG model include:

_¶4

The EAG’s base-case analyses suggest the following results:

_¶5

The EAG’s model is subject to the following strengths:

_¶6

The EAG’s economic analyses are also subject to several weaknesses, many of which stem from uncertainties and gaps in the available evidence:

The ICER for Prosigna versus current decision-making is expected to be £39,357 per QALY gained. The model suggests that the use of Prosigna will result in a small decrease in the use of chemotherapy, a small reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify any evidence to support a predictive benefit for Prosigna in the LN+ population.

_¶1

EndoPredict (EPclin)

The ICER for EPclin versus current decision-making is expected to be £4113 per QALY gained. The model suggests that the use of EPclin will result in a small decrease in the use of chemotherapy, a reduction in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify any evidence to support a predictive benefit for EPclin in the LN+ population.

_¶1

MammaPrint

MammaPrint is dominated by current decision-making. These results are driven by the large, estimated reduction in the use of adjuvant chemotherapy in women who would have benefitted from treatment, an increase in the lifetime probability of developing DM and additional net costs due to the cost of the test. The EAG’s systematic review did not identify sufficient evidence to support a predictive benefit for MammaPrint in the LN+ population.

_¶1

Strengths and limitations in the clinical evidence base

Strengths of the clinical evidence base include the fact that there is fairly substantial evidence for prognostic ability of all four tests. A major limitation is that it is difficult to collect new data on predictive ability because it is not considered ethical to randomise patients who are high risk on any of the tests to chemotherapy versus no chemotherapy. Therefore, although there are prospective RCTs for the effect of chemotherapy within low- to intermediate-risk patients, data for high-risk patients are limited to retrospective reanalyses of trials, plus observational data in which test results may have influenced treatment. Decision impact data in a LN+ population were available for Oncotype DX, but not for the other three tests. Anxiety and HRQoL data were not identified in a LN+ population.

_¶1

Strengths and limitations relating to the health economic analysis

The EAG’s model is subject to several strengths. In particular, the economic analysis is consistent with the NICE Reference Case and relates specifically to the LN+ population under consideration within this appraisal; the model structure is consistent with most published economic models of tumour profiling tests as well as the two economic models submitted by the test manufacturers; where data permit, risk classification probabilities and DRFI estimates for each individual test have been taken from the same source, which improves consistency and avoids the potential for spectrum bias; the analysis uses a recent UK decision impact study undertaken in LN+ women, and a broad assessment of uncertainty around all key model inputs has been presented, including testing assumptions around whether Oncotype DX is predictive of chemotherapy benefit. The EAG notes that under similar assumptions around the benefits of each tumour profiling test, the EAG’s model and the Exact Sciences model indicate similar economic conclusions.

_¶1

The EAG’s economic analyses are subject to several weaknesses: the EAG’s analyses of Oncotype DX based on RxPONDER indirectly assume a predictive benefit which reflects a plausible clinical assumption about the effect of chemotherapy in women who were excluded from the trial (those with an RS of > 25), rather than a statistical test of interaction across the full RS spectrum; there are inconsistencies in RS cut-offs between sources used in the model; the analyses rely on a decision impact study of Oncotype DX to estimate post-test probabilities for all 2- and 3-level tests, which is highly uncertain; and there is insufficient evidence to allow for the economic analysis of EPclin and MammaPrint in an exclusively pre-menopausal subgroup. There is uncertainty around the potential negative effects of chemotherapy on infertility which may not be fully captured in the analyses of Oncotype DX in the pre-menopausal LN+ subgroup. The EAG’s analyses of NHB provide a means for the Appraisal Committee to decide whether any missing health effects are likely to impact on the conclusions drawn from the economic analysis.

_¶2

Uncertainties

As was the case when NICE DG34 was undertaken, evidence relating to the impact on patient outcomes where the test is used in clinical practice remains largely absent, and is impeded by the long-term follow-up required, the large sample sizes required, and ethical problems associated with withholding chemotherapy from clinically high-risk patients.

_¶1

Evidence relating to key subgroups defined in the scope is generally lacking. Where possible, separate data and analyses have been presented for pre-menopausal and post-menopausal women. Limited data were available by clinical risk subgroups as defined by risk assessment tools such as NPI or PREDICT. There were limited data in male-only subgroups or cohorts, and data relating to people of different ethnicities were difficult to interpret due to differences in treatment practices in different countries. No data were identified which could allow for a separate analysis of the value of tumour profiling tests in people with comorbidities who would be particularly affected by the adverse effects of chemotherapy.

_¶2

There were no relevant decision impact studies on the use of MammaPrint, Prosigna or EPclin in a UK or European LN+ population. This remains a key area of uncertainty.

_¶3

Generalisability

The economic analyses of EPclin and Prosigna are informed by the TransATAC trial which relates only to a post-menopausal population. It is expected that EPclin may also be used in pre-menopausal women. It was not possible to undertake separate economic analyses for MammaPrint or EPclin in a pre-menopausal LN+ population.

_¶1

Implications for service provision

Oncotype DX, Prosigna and EPclin are already recommended by NICE for use in the NHS for women with ER+ (and/or PR+), HER2−, LN0 early breast cancer. The EAG’s model suggests that all of the tumour profiling tests are expected to result in fewer women receiving adjuvant chemotherapy, thereby reducing costs and increasing capacity. However, for some of the tests, these initial benefits may lead to more women later requiring further treatment for DM, thereby offsetting cost savings and capacity improvements.

_¶1

MammaPrint is not currently recommended for use in the NHS. MammaPrint testing can be undertaken either as an off-site service with samples sent to a laboratory in the USA or through a decentralised testing service for laboratories with NGS capability. The per-sample pricing of MammaPrint remains the same regardless of where the testing is performed. Not all laboratories will have NGS capability which will impact how testing services are delivered. For the other tests, only a single testing option is available – for Oncotype DX, samples are processed centrally, whereas for Prosigna and EPclin, samples are processed in local laboratories.

_¶2

Suggested research priorities

Research priorities include the following:

_¶1

The use of patient and public involvement

There was no patient and public involvement in producing the draft version of this report. However, the report was circulated for consultation to stakeholders, which included patients and the public, and the final EAG report was discussed by the NICE Technology Appraisal Committee at a meeting which included representation by patients and the public. The report was amended and further analyses were conducted to address points raised during these stages of the appraisal.

_¶1

Equality, diversity and inclusion

As this report is secondary research, no patient participation was involved and the EAG did not need to consider the equality, diversity and inclusion of participants. The primary research team was part of the SCHARR Technology Assessment Group contracted by the Department of Health, and this team is a diverse group representing a wide range of protected characteristics, consisting of seniority, ages, ethnicity and religious beliefs, and including both males and females. The clinical team represents experts within their field who have successfully worked with the SCHARR Technology Assessment Group.

_¶1

Tables

TABLE 1 Summary of TNM stages

Stage	T	N	M
Stage 0	Tis	N0	M0
Stage IA	T1	N0	M0
Stage IB	T0	N1mi	M0
	T1	N1mi	M0
Stage IIA	T0	N1	M0
	T1	N1	M0
	T2	N0	M0
Stage IIB	T2	N1	M0
	T3	N0	M0
Stage IIIA	T0	N2	M0
	T1	N2	M0
	T2	N2	M0
	T3	N1	M0
	T3	N2	M0
Stage IIIB	T4	N0	M0
	T4	N1	M0
	T4	N2	M0
Stage IIIC	Any T	N3	M0
Stage IV	Any T	Any N	M1

TABLE 2 Breast cancer risk prediction tools

Tool	NPI	AOL	PREDICT (Version 2.2)
Factors included in the prediction algorithm	Tumour size	Age at diagnosis	Age at diagnosis
	Nodal status	Comorbidity factors	Menopausal status
	Tumour grade	ER status	Mode of detection
		Tumour size	Invasive tumour size
		Tumour grade	Tumour grade
		Nodal status	Number of positive nodes
			ER status
			HER2/ERBB2 status
			Ki67 status
			Generation of chemotherapy regimen
Outcome(s) predicted	Mortality	Mortality or relapse	Mortality

TABLE 3 Summary of tumour profiling tests

Test	Oncotype DX Breast RS	Prosigna	EndoPredict EPclin score	MammaPrint
Manufacturer	Exact Sciences	Veracyte	Myriad	Agendia
Purpose	Recurrence risk and chemotherapy benefit	Intrinsic subtype and recurrence risk	Distant recurrence risk and chemotherapy benefit	Distant recurrence risk and chemotherapy benefit
Description	21-gene assay (16 cancer genes; RT-qPCR)	50-gene assay (50 cancer genes; direct mRNA counting) + clinical factors	12-gene assay (8 cancer genes; RT-qPCR) + clinical factors	70-gene assay (microarray)
Testing location	Test service (USA)	Local laboratory	Local laboratory	Local laboratory (NGS) or test service (USA)
Stage	Early-stage (stage I–IIIa)	Early-stage (stage I–IIIA)	Early-stage	Early-stage (stage I, II or operable stage III)
Lymph node status	LN0 or LN + (up to 3 positive nodes)	LN0 and LN + (up to 3 positive nodes, and 4 + nodes)	LN0 and LN + (up to 3 positive nodes)	LN0 or LN + (up to 3 positive nodes)
Hormone receptor status	HR+	HR+	ER+	HR+
HER2 status	HER2−	HER2− or HER2+	HER2−	HER2−
Menopausal status	Pre- and post-menopausal	Post-menopausal only	Pre- and post-menopausal	Pre- and post-menopausal
Test result	RS	Risk category (low, intermediate, high)	Risk category (low, high)	Risk category (low, ultra-low, high)
	Chemotherapy benefit	Intrinsic subtype	Chemotherapy benefit (%)	Chemotherapy benefit
	Probability of distant recurrence (%)	Probability of distant recurrence (%)	Probability of distant recurrence (%)
Assumptions	Score assumes 5 years of endocrine treatment	Score assumes 5 years of endocrine treatment	Scores assume 5 years of endocrine treatment	Assumes no adjuvant therapy

TABLE 4 Summary of prognostic data for 10-year distant recurrence (all four tests)

Test	ET/CT	ReferenceStudy	Design	N pts	Outcome	Nodal status	HR, HER2	Meno status	Test cut-offs	Distribution %	DR free 0–10 years %	aSig prog 10 years?	aSig prog 10 years adj?
Low	Int	High	Low	Int	High
Oncotype DX	ET alone	Sestak 201820(TransATAC)	RCT-R	N = 183	DRFI	LN1−3	HR + HER2−	Post	18, 30	57	32	11	81	71	62	N	N
	All CT + ET	Mamounas 201849(NSABP-28)	RCT-R	N = 722	DRFI	LN1−3	ER+NR HER2	Pre/post	18, 30	37	34	28	85	72	63	Y	Y
MammaPrint	Variable ET/CT	Drukker 201453	Cohort-R	N = 144	DMFS	74% LN1−326% LN4 +	77% ER+NR HER2	Pre/post (age < 53)	0.4	38	–	62	79	–	54	Y	–
		Mook 200933	Cohort-R	N = 241	DMFS	LN1−3 + Lnmicro	79% ER+84% HER2−	Pre/post	NR	41	–	59	91	–	76	Y	N
		Vliek 201754(RASTER)	Cohort-R	N = 134	DRFI	LN1−3	83% ER+85% HER	Pre/post	NR	48	–	52	95	–	81	Y	–
Prosigna	ET alone	Sestak 201820(TransATAC)	RCT-R	N = 183	DRFI	LN1−3	HR + HER2−	Post	16, 40	8	32	60	100	79	69	N	Y
		Gnant 201456/Filipits 201457(ABCSG-8)	RCT-R	N = 413	DMFS	89% LN1−311% LN4 +	ER + HER2−	Post	16, 40	4	34	62	100	94	76	–	Y
		Laenkholm 201858(DBCG)	Cohort-R	N = 1395	DRFS	LN1−3	HR + HER2−	Post	Varies by N nodes	26	28	46	97	89	78	Y	Y
	All CT + ET	Martin 2016/1459,60(GEICAM 9906)	RCT-R	N = 536	DMFS	64% LN1−336% LN4 +	ER + HER2−	54% pre46% post	18, 65	19	56	26	92	74	66	Y	N
EPclin	ET alone	Sestak 201820(TransATAC)	RCT-R	N = 183	DRFI	LN1−3	HR + HER2−	Post	3.3	23	–	77	94	–	70	Y	Y
		Filipits 201963(ABCSG-6/8)	RCT-R	N = 453	DRFR	LN1−3	ER + HER2−	Post	3.3	35	–	65	96	–	81	Y	Y
		Constantinidou 202265	Cohort-R	N = 62	DRFS	LN1−3	ER + HER2−	Pre	3.3	19	–	81	100	–	75	N	Y
	All CT + ET	Martin 2016/1459,60(GEICAM 9906)	RCT-R	N = 555	DMFS	64% LN1−336% LN4 +	ER + HER2−	54% pre46% post	3.3	13	–	87	100	–	72	Y	Y

TABLE 5 Prognostic data from prospective RCT of Oncotype DX (RxPONDER)

ReferenceStudy	Outcome	N, ET/CTDesign	Nodal status HR, HER2	Meno status	Test cut-offs	Risk 0–5 years %	HR between test groups (95% CI)	Sig?a*Adj
Low RS ≤ 25	High
CT	No
Oncotype DX: prospective RCT: distant recurrence
Kalinsky 202129RxPONDER	DRFI (0–5 years)	CT vs. noneProsp RCTn = 3353	LN1–3 (65% 1 node, 25% 2 nodes, 9% 3 nodes)	Post-meno	All ≤ 25	94.9	93.9	–	–	–
		n = 1665	100% HR+100% HER2−	Pre-meno	All ≤ 25	96.3	93.9	–	–	–
Oncotype DX: prospective RCT: IDFS
Kalinsky 202129RxPONDER	IDFS (0–5 years)	n = 5018	LN1–3	All meno (67% post)	All ≤ 25	92.2	91.0	–	*0–5 years: HR per unit-RS (within RS 0–25, adj for meno and CT): 1.05 (1.04 to 1.07), p < 0.001	Y*
			100% HR+	Post-meno	All ≤ 25	91.2	91.9	–	*0–5 years: HR per unit-RS (within RS 0–25, adj for CT, nodes, grade, tumour size, age): 1.05 (1.03 to 1.07), p < 0.001	Y*
			100% HER2−	Pre-meno	All ≤ 25	93.9	89.0	–	*0–5 years: HR per unit-RS (within RS 0–25, adj for CT, nodes, grade, tumour size, age): 1.06 (1.02 to 1.09), p = 0.001	Y*
Oncotype DX: prospective RCT: IDFS by ethnicity
Abdou 202351RxPONDER	IDFS	n = 4015CT+ET vs. ET	LN1–3	White (n = 2833)	All ≤ 25	91.5	–	–	–
		Prosp RCT	100% HR+	Black (n = 248)	All ≤ 25	87.0	–	–	–
			100% HER2−
				Asian (n = 324)	All ≤ 25	93.9	–	–	–
				Hispanic (n = 610)	All ≤ 25	91.4	–	–	–

TABLE 6 Prognostic data from prospective RCT of MammaPrint (MINDACT)

ReferenceStudy	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Menostatus	Test cut-offs	Distribution %	Risk 0–8 years %	HR between groups	Sig?a*Adj
Low	High	Low	High
MammaPrint: prospective RCT: distant recurrence
Piccart 202130,bMINDACT	DMFI	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	31	92.3(50% CT)	80.9(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	8	95.2(no CT)	–c	–	–
	DMFS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	31	91.0(50% CT)	79.1(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	8	94.0(no CT)	–c	–	–
Lopes Cardozo 202252MINDACT	DMFI	N = 201 (ultra-low)Var ET/CTProsp-RCT	LN1–399% ER+97% HER2−	–	> 0.355 ultra-low	Ultra-low:15	–	Ultra-low:95.2	–	–	–
MammaPrint: prospective RCT: DFS
Piccart 202130,bMINDACT	DFS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	31	84.5(50% CT)	74.5(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	8	85.6(no CT)	–	–	–
MammaPrint: prospective RCT: OS
Piccart 202130,bMINDACT	OS	n = 1176CT+ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	31	95.1(50% CT)	89.1(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	8	98.1(no CT)	–	–	–

TABLE 7 Observational data for Oncotype DX (distant recurrence)

Cohort	Reference	Nodal statusHR, HER2	Outcome	N ET/CT	MenoAgeClin	Test cut-offs	Distribution %	% risk of outcome	HR between test risk groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High
Oncotype: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic: 42%	DRFI (0–5 years)	n = 709	All meno	18, 30	53	36	10	96.8 (7% CT)	93.7 (40% CT)	83.1 (86% CT)	0–5 years: low vs. high: HR 0.19 (0.09 to 0.40)	Y
		LN1–3 : 58%		Var ET/CT									0–5 years: int vs. high: HR 0.39 (0.20 to 0.79), p < 0.001	Y
		100% ER+											*0–5 years: adj HR: low vs. high: HR 0.23 (0.11 to 0.50)	Y*
		100% HER2−											*0–5 years: adj HR: int vs. high: HR 0.42 (0.20 to 0.86), p = 0.001	Y*
						11, 25	≤ 25:81	19	95.7 (5% CT)	96.0 (18% CT)	86.9 (77% CT)	0–5 years: p < 0.001	Y
						≤ 25, 26–30				96.0 (15% CT)	91.5 (67% CT)	–	–
						18–25					94.4 (31% CT)		–	–
				n = 109Var ET/CT	Age < 50	18, 30	48	37	16	96.2 (12% CT)	100.0 (48% CT)	64.2 (100% CT)	0–5 years: p < 0.001	Y
				n = 464Var ET/CT	Age 50–69	18, 30	54	37	9	97.6 (6% CT)	93.5 (42% CT)	87.8 (90% CT)	0–5 years: p = 0.017	Y
				n = 136Var ET/CT	Age ≥ 70	18, 30	57	33	10	94.7 (7% CT)	88.7 (22% CT)	92.9 (57% CT)	0–5 years: p = 0.458	N
Young Women’s Breast Cancer Study	Poorvu 202075	LNmic, LN1–3	DRFS (0–6 years)	n = 163	Age ≤ 40	18, 30	33	42	25	0–6 years: 85.9 (83% CT)	0–6 years: 87.3 (97% CT)	0–6 years: 62.8 (98% CT)	0–6 years: p = 0.004	Y
		100% ER+		Var ET/CT
		100% HER2−
						11, 25	9	54	37	0–6 years: 92.3 (79% CT)	0–6 years: 85.2 (92% CT)	0–6 years: 71.3 (97% CT)	0–6 years: p = 0.10	N

TABLE 8 Prediction of chemotherapy benefit: RCT reanalysis (Oncotype DX)

StudyReferenceDesign	Nodal statusHR, HER2	Outcome	N	Menopausal status	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Interaction	Preda*Adj
Low	Int	High	Low	Int	High	Low	Int	High	Adj
CT	No	CT	No	CT	No
Oncotype DX: RCT-reanalysis: distant recurrence
No data
Oncotype DX: RCT-reanalysis: DFS
SWOG-8814	LN1–3 : 62%	DFS	n = 367	Post-meno	18, 30	–	–	–	–	–	–	–	–	–	1.34 (0.47 to 3.82)	0.95 (0.43 to 2.14)	0.59 (0.32 to 1.11)	Y	p = 0.029 (adj nodes)	Y*
Albain 201032	LN4 +: 38%	0–5 years
RCT-R	100% HR+
	88% HER2−
		DFS0–10 years	n = 367	Post-meno	18, 30	64	60	–	–	55	43	4	–	12	1.02 (0.54 to 1.93)p = 0.97	0.72 (0.39 to 1.31)p = 0.48	0.59 (0.35 to 1.01)p = 0.033	Y	p = 0.053(adj nodes)p = sig (NR) (adj various)p = 0.15 (adj Allred-ER)	NYN*
		DFS5–10 years	n = 367	Post-meno	18, 30	–	–	–	–	–	–	–	–	–	0.88 (0.38 to 1.92)	0.52 (0.21 to 1.27)	0.60 (0.22 to 1.62)	Y	p = 0.58(cont RS, adj nodes)	N*
Oncotype DX: RCT-reanalysis: BCSS and OS
SWOG-8814	LN1–3 : 62%	BCSS
Albain 201032	LN4+: 38%
RCT-R	100% HR+
	88% HER2−	0–10vyears	n = 367	Post-meno	18, 30	–	–	–	–	73	54	–	–	19	p = 0.56	p = 0.89	p = 0.033	Y	–	–
		OS0–10 years	n = 367	Post-meno	18, 30	–	–	–	–	68	51	–	–	17	1.18 (0.55 to 2.54); p = 0.68Log-rank p = 0.63	0.84 (0.40 to 1.78); p = 0.65Log-rank p = 0.85	0.56 (0.31 to 1.02); p = 0.057Log-rank p = 0.027	Y	Int (adj nod):0–10 years: p = 0.0260–5 years: p = 0.0165–10 years: p = 0.87	YYN*

TABLE 9 Prediction of chemotherapy benefit: prospective RCT of Oncotype DX (RxPONDER)

StudyReferenceDesign	Nodal statusHR, HER2	Outcome	N	Menopausal status	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Interaction	Preda*Adj
Low RS ≤ 25	High	Low RS ≤ 25	High
CT	No	CT	No
Distant recurrence: full population
RxPONDERKalinsky 202129(Kalinsky SABCS 2021 slides78)bProsp RCT	LN1–3(65% 1 node, 25% 2 nodes, 9% 3 nodes)100% HR+100% HER2−	DRFS(0–5 years)	n = 5018	All (67% post)	All ≤ 25	94.9	93.9	–	–	1.0	–	RS ≤ 25 : 0.88 (0.71 to 1.09), p = 0.25	–	–
Distant recurrence: post-menopausal
RxPONDER	LN1–3100% HR+100% HER2−	DRFS(0–5 years)	n = 3353	Post-meno	All ≤ 25	94.494.3b	94.494.8b	–	–	0.1-0.5b	–	RS ≤ 25: HR 1.05 (0.81 to 1.37), p = 0.70RS ≤ 25b: adj HR 1.12 (0.88 to 1.44), p = 0.35	–	–
		DRFI(0–5 years)	n = 3353	Post-meno	All ≤ 25	95.8b	96.6b	–	–	−0.8b	–	RS ≤ 25b: adj HR 1.12 (0.82 to 1.52), p = 0.49	–	–
Distant recurrence: pre-menopausal
RxPONDER	LN1–3100% HR+100% HER2−	DRFS(0–5 years)	n = 1665	Pre-meno	All ≤ 25	96.195.9b	92.893.4b	–	–	3.32.5b	–	RS ≤ 25: HR 0.58 (0.39 to 0.87), p = 0.009RS ≤ 25b: adj HR 0.66 (0.45 to 0.97), p = 0.033	–	–
		DRFI(0–5 years)	n = 1665	Pre-meno	All ≤ 25	96.3b	93.9b	–	–	2.4b	–	RS ≤ 25b: adj HR 0.64 (0.43 to 0.95), p = 0.026	–	–
			NR	Pre-meno	0–13	–	–	–	–	2.3	–	–	–	–
					14–25	–	–	–	–	2.8	–	–	–	–
IDFS: full population
RxPONDER	LN1–3100% HR+100% HER2−	IDFS(0–5 years)	n = 5018	All (67% post)	All ≤ 25	92.2	91.0	–	–	1.2	–	RS ≤ 25 : 0.86 (0.72 to 1.03), p = 0.10	HR 1.02 (0.98 to 1.05), p = 0.35 (adj meno)	N
IDFS: post-menopausal
RxPONDER	LN1–3100% HR+100% HER2−	IDFS(0–5 years)	n = 3353	Post-meno	All ≤ 25	91.391.2b	91.991.9b	–	–	−0.6−0.7b	–	RS ≤ 25 : HR 1.02 (0.82 to 1.26), p = 0.89RS ≤ 25b : adj HR 1.06 (0.87 to 1.30), p = 0.55	HR 1.01 (0.97 to 1.06), p = 0.48	N
			NR	Post-meno	0–10	92.7	92.7	–	–	0.0	–	RS 0–10 : 0.72 (0.44 to 1.18)	–	–
					11–15	93.5	95.8	–	–	−2.3	–	RS 11–15 : 1.30 (0.88 to 1.92)	–	–
					16–20	93.2	90.8	–	–	2.4	–	RS 16–20 : 0.91 (0.57 to 1.43)	–	–
					21–25	84.8	93.2	–	–	−8.4	–	RS 21–25 : 1.13 (0.75 to 1.70)	–	–
					0–13	–	–	–	–	–	–	RS 0–13 : 1.01 (0.71 to 1.44)	–	–
					14–25	–	–	–	–	–	–	RS 14–25 : 1.01 (0.77 to 1.33)	–	–
IDFS: pre-menopausal
RxPONDER	LN1–3100% HR+100% HER2−	IDFS(0–5 years)	n = 1665	Pre-meno	All ≤ 25	93.993.9b	89.089.0b	–	–	4.94.9b	–	RS ≤ 25 : HR 0.60 (0.43 to 0.83), p = 0.002RS ≤ 25b : adj HR 0.64 (0.47 to 0.87), p = 0.004	HR 1.04 (0.97 to 1.12), p = 0.26	N
			NR	Pre-meno	0–10	96.6	92.4	–	–	4.2	–	RS 0–10 : 0.47 (0.18 to 1.20)	–	–
					11–15	95.5	93.3	–	–	2.2	–	RS 11–15 : 0.68 (0.33 to 1.37)	–	–
					16–20	91.5	83.8	–	–	7.7	–	RS 16–20 : 0.57 (0.35 to 0.94)	–	–
					21–25	92.4	85.2	–	–	7.2	–	RS 21–25 : 0.63 (0.30 to 1.31)	–	–
					0–13	–	–	–	–	–	–	RS 0–13 : 0.49 (0.24 to 0.99)	–	–
					14–25	–	–	–	–	–	–	RS 14–25 : 0.63 (0.43 to 0.91)	–	–

TABLE 10 Prediction of chemotherapy benefit: reanalysis of cohort (MammaPrint)

Cohort	Reference	Nodal statusHR, HER2	Outcome	N	Menopausal status	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT	Interaction	Preda*Adj
Low	High	Low	High	Low	High
CT	No	CT	No
MammaPrint: cohort reanalysis: BCSS
NKI, Italy, VdVCohort-R	Mook 200933	LN1micro to LN379% ER+84% HER2−	BCSS0–10 years	n = 347	All meno	NR	–	–	–	–	–	–	–	–	Int p = 0.95 (adj)	N*

TABLE 11 Chemotherapy effect within risk groups: prospective RCT of MammaPrint (MINDACT)

StudyReferenceDesign	Nodal statusHR, HER2	Outcome	N	Clinical risk	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Interaction	Preda*Adj
Low MMP	High MMP	Low MMP	High MMP	Low MMP	High MMP
CT	No	CT	No
MammaPrint: prospective RCT: distant recurrence
MINDACTPiccart 202130,bProsp RCT	LN1–3100% HR+100% HER2−	DMFS(0–8 years)	N = 658	High mAOLc	> 0 low, ≤ 0 high	91.2	89.9	–	–	1.3	–	0.84 (0.51 to 1.37), p = NR	–	–	–
		DMFI(0–8 years)	N = 658	High mAOLc	> 0 low, ≤ 0 high	92.3	90.9	–	–	1.4	–	–	–	–	–
MammaPrint: prospective RCT: other outcomes
MINDACTPiccart 202130,bProsp RCT	LN1–3100% HR+100% HER2−	DFS (0–8 years)	N = 658	High mAOLc	> 0 low, ≤ 0 high	85.3	82.8	–	–	2.5	–	–	–	–	–
		OS (0–8 years)	N = 658	High mAOLc	> 0 low, ≤ 0 high	95.5	94.9	–	–	0.6	–	–	–	–	–

TABLE 12 Effect of chemotherapy within risk groups: registry data for Oncotype DX (distant recurrence)

Cohort	Reference	Nodal status HR, HER2	Outcome	N	Menopausal status	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Inter-action	Preda*Adj
Low	Int	High	Low	Int	High	Low	Int	High
CT	No	CT	No	CT	No
Oncotype DX: observational registry: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic : 42%LN1–3 : 58%100% ER+100% HER2−	DRFI0–5 years	n = 709	All meno	18, 30	92.3	97.1	99	90.3	82	90	−4.8	8.7	−8.0	p = 0.245	p = 0.019	–	–	–
						11, 25	83.3	96.3	98.8	95.4	97.5	79.7	−13.0	3.4	17.8	–	–	p = 0.017	–	–
						All ≤ 25	–	–	97.7	95.6	–	–	2.1	–	p = 0.521	–	–	–
						All 18–25	–	–	100	91.8	–	–	–	8.2	–	–	p = 0.058	–	–	–
	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	DRFS0–7 years	n = 140	All meno	All 26–30	–	–	–	–	89.4	78.0	–	–	11.4	–	–	Not sig	–	–

TABLE 13 Chemotherapy effect within risk groups: registry data for Oncotype DX (post-menopausal or older age groups)

Cohort	Reference	Nodal statusHR, HER2	Outcome	N	Age,Clinical factors	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Inter-action	Preda*Adj
Low	Int	High	Low	Int	High	Low	Int	High
CT	No	CT	No	CT	No
Oncotype: observational registry: OS
NCDB	Cao 2022 (abst)82	LN1–3100% ER+100% HER2−	OSNR	n = NR	Age > 50	All 20–25	–	–	–	–	–	–	–	–	–	–	Unadj: 0.521 (NR), p = 0.019	–	–	–
NCDB (cont)	Ibraheem 201971	LN1–3 : 97%LN4–9 : 3%100% HR+100% HER2−	OS0–5 years	n = 8886	Age > 50	All 11–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.64 (0.47 to 0.86), p = 0.004	–	–	–
NCDB (cont)	Weiser 202273	LN1–3100% HR+100% HER2−	OS0–5 years	NR	Age 50–75Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–	Adj: 1.12 (0.86 to 1.46)	–	–	–	–
NCDB (cont)	Weiser 202174	LN1–3100% HR+100% HER2−	OS0–5 years	NR	Age 51–70	All ≤ 25	–	–	–	–	–	–	1.6	–	Adj: 1.49 (1.12 to 1.97), p = 0.006	–	–	–
						All 12–17	–	–	–	–	–	–	–	3.6	–	–	Adj: 2.80 (1.45 to 5.24)	–	–	–
						All 18–25	–	–	–	–	–	–	–	3.2	–	–	Adj: 1.37 (0.92–2.05)	–	–	–
				NR	Age > 70	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 1.1 (0.68 to 1.78), p = 0.69	–	–	–

TABLE 14 Decision impact: Oncotype DX (not split by test risk group)

ReferenceCountry	Study, settingYears	HR, HER2	Nodal statusClinical risk	Recom/decision	Menopausal status	Test group	RxP	N pts	No CT	No CT to CT	CT	CT to no CT	Pre-test CT	Post-test CT	Net change CT
Battisti 2019 (abst)35UK	PONDx; 30 centres2017–8	ER+ HER2−	LN1–3	R-R	All (65% post)	All RS	–	567	–	–	–	–	371 (65%)	162 (29%)	−209 (−37%)
				R-D	All (65% post)	All RS	–	567	–	–	–	–	371 (65%)	140 (25%)	−231 (−41%)
Holt 2023 (abst)36Holt 202440UK	14 centres2017–22	ER+ HER2−	LN1–3	R-D	All (77% post)	All RS	–	664	117 (18%)	17 (3%)	171 (26%)	359 (54%)	530 (80%)	188 (28%)	−342 (−52%)
					Pre-meno	All RS	–	152	23 (15%)	6 (4%)	65 (43%)	58 (38%)	123 (81%)	71 (47%)	−52 (−34%)
					Post-meno	All RS	–	512	94 (18%)	11 (2%)	106 (21%)	301 (59%)	407 (79%)	117 (23%)	−290 (−57%)
Loncaster 201737UK	Greater Manchester (NR centres)2012–5	ER+ HER2−	LN+CT indicated.Post-test decision based on RS	R-D	Post-meno	All RS	–	65	0 (0%)	0 (0%)	20 (31%)	45 (69%)	65 (100%)	20 (31%)	−45 (−69%)
Malam 202238UK	Norfolk and Norwich (1 centre)2014–20	ER+ HER2−	LN1–3Post-test decision based on RS	R-R	All meno	All RS	–	69	36 (52%)	1 (1.4%)	8 (12%)	24 (35%)	32 (46%)	9 (13%)	−19 (−28%)
Nanda 2021 (abst)39UK	Oxford + Swansea (2 centres)2013–9	ER+ HER2−	LN1–3 (inc micromets)CT indicated	R-R	All meno	All RS	–	173	0 (0%)	0 (0%)	44 (25%)	129 (75%)	173 (100%)	44 (25%)	−129 (−75%)
Eiermann 201347Germany	15 centres2010–1	ER+ HER2−	LN1–3	R-R	All meno	All RS	–	122	18 (15%)	12 (10%)	58 (46%)	34 (28%)	92 (75%)	70 (57%)	−22 (−18%)
				R-D	All meno	All RS	–	122	–	–	–	–	92 (75%)	57 (47%)	−35 (−29%)
Cognetti 202141Italy	PONDx; 27 centres2016–7	ER+ HER2−	LN1–3	R-R	All (55% post)	All RS	–	414	–	–	–	–	258 (62%)	110 (28%)	−148 (−55%)
Dieci 201943Italy	ROXANE; 9 centres2017–8	HR+ HER2−	LN1–394% high clin risk (mAOL)	R-R	All (55% post)	All RS	–	99	42 (42%)	3 (3%)	24 (24%)	30 (30%)	54 (55%)	27 (27%)	−27 (−27%)
Dieci 201842Italy	Breast DX, 9 centres2014–6	ER+ HER2−	LN1–3Int clin risk	R-R	All (55% post)	All RS	–	126	49 (39%)	5 (4%)	52 (41%)	20 (16%)	72 (57%)	57 (45%)	−15 (−12%)
				R-D	All (55% post)	All RS	–	126	–	–	–	–	72 (57%)	54 (43%)	−18 (−14%)
Zambelli 202046Italy	BONDX (4 centres, Lombardy)2017–8	ER+ HER2−	LN1–3Int clin risk	R-R	All meno	All RS	–	127	79 (62%)	0 (0%)	25 (20%)	23 (18%)	48 (38%)	25 (20%)	−23 (−18%)
Fernandez-Perez 2021 (abst)44Spain	9 centres (Galicia)2013–8	HR+ HER2−	LN1–3 (inc micromets)	R-R	All (50% post)	All RS	–	229	–	–	–	–	159 (69%)	59 (26%)	−100 (−44%)
Llombart-Cussac 202345Spain	KARMA Dx (8 centres)2016–7	ER+ HER2−	LN1–3High clin riskCT indicated	R-R	All meno	All RS	–	150	0 (0%)	0 (0%)	41 (27%)	109 (73%)	150 (100%)	41 (27%)	−109 (−73%)

TABLE 15 Decision impact: Oncotype DX (results by test risk group)

ReferenceCountry	Study, settingYears	HR, HER2	Nodal statusClinical risk	Recom/decision	Meno status	Test group	RxP	N pts	No CT	No CT to CT	CT	CT to no CT	Pre-test CT	Post-test CT	Net change CT
Oncotype DX: cut-offs of RS 18 and 30
Holt 2023 (abst)36Holt 202440UK	14 centres2017–22	ER+ HER2−	LN1–3	R-D	All meno	RS 0–17	–	400	95 (24%)	3 (1%)	28 (7%)	274 (69%)	302 (76%)	31 (8%)	−271 (−68%)
						RS 18–30	–	204	20 (10%)	12 (6%)	88 (43%)	84 (41%)	172 (84%)	100 (49%)	−72 (−35%)
						RS > 30	–	58	0 (0%)	2 (3%)	55 (95%)	1 (1.7%)	56 (97%)	57 (98%)	+ 1 (+ 1.7%)
Loncaster 201737UK	Greater Manchester (NR centres)2012–5	ER+ HER2−	LN+CT indicated.Post-test decision based on RS	R-D	Post-meno	RS 0–17	–	40	0 (0%)	0 (0%)	40 (100%)	37 (93%)	40 (100%)	3 (8%)	−37 (−93%)
						RS 18–30	–	19	0 (0%)	0 (0%)	19 (100%)	7 (37%)	19 (100%)	12 (63%)	−7 (−37%)
						RS > 30	–	6	0 (0%)	0 (0%)	6 (100%)	1 (17%)	6 (100%)	5 (83%)	−1 (−17%)
Eiermann 201347Germany	15 centres2010–1	ER+ HER2−	LN1–3	R-R	All meno	RS 0–17	–	67	–	2 (3%)	–	30 (45%)	–	–	–
						RS 18–30	–	44	–	8 (18%)	–	4 (9%)	–	–	–
						RS 31 +	–	11	–	2 (18%)	–	0 (0%)	–	–	–
Llombart-Cussac 202345Spain	KARMA Dx (8 centres)2016–7	ER+ HER2−	LN1–3High clinical riskCT indicated	R-R	All meno	RS 0–17	–	86	0 (0%)	0 (0%)	8 (9%)	78 (91%)	86 (100%)	8 (9%)	−78 (−91%)
						RS 18–30	–	57	0 (0%)	0 (0%)	26 (46%)	31 (54%)	57 (100%)	26 (46%)	−31 (−54%)
						RS 31–100	–	7	0 (0%)	0 (0%)	7 (100%)	0 (0%)	7 (100%)	7 (100%)	No change
Zambelli 202046Italy	BONDX (4 centres, Lombardy)2017–8	ER+ HER2−	LN1–3Int clinical risk	R-R	All meno	RS 0–17	–	71	56 (79%)	0 (0%)	1 (1%)	14 (20%)	15 (21%)	1 (1%)	−14 (−20%)
						RS 18–30	–	48	23 (48%)	0 (0%)	16 (33%)	9 (19%)	25 (52%)	16 (33%)	−9 (−19%)
						RS 31–100	–	8	0 (0%)	0 (0%)	8 (100%)	0 (0%)	8 (100%)	8 (100%)	No change
Oncotype DX: cut-offs of RS 11 and 25
Holt 2023 (abst)36Holt 2023 (unpub)UK	14 centres2017–22	ER+ HER2−	LN1–3	R-D	All meno	RS 0–13	–	261	68 (26%)	2 (1%)	13 (5%)	178 (68%)	191 (73%)	15 (6%)	−176 (−67%)
						RS 14–25	–	305	48 (16%)	7 (2%)	72 (24%)	178 (58%)	250 (82%)	79 (26%)	-171 (-56%)
						RS 26–100	–	98	1 (1%)	8 (8%)	86 (88%)	3 (3%)	89 (91%)	94 (96%)	+ 5 (+ 5%)
					Pre-meno	RS 0–25	–	127	23 (18%)	4 (3%)	43 (34%)	57 (45%)	100 (79%)	47 (37%)	−53 (−42%)
						RS 26–100	–	25	0 (0%)	2 (8%)	22 (88%)	1 (4%)	23 (92%)	24 (96%)	+ 1 (+ 4%)
					Post-meno	RS 0–25	–	439	93 (21%)	5 (1%)	42 (10%)	299 (68%)	341 (78%)	47 (11%)	−294 (−67%)
							Pre-RxP	292	57 (20%)	1 (0.3%)	40 (14%)	194 (66%)	234 (80%)	41 (14%)	−193 (−66%)
							Post-RxP	147	36 (24%)	4 (3%)	2 (1%)	105 (71%)	107 (73%)	6 (4%)	−101 (−69%)
						RS 26–100	–	73	1 (1%)	6 (8%)	64 (88%)	2 (3%)	66 (90%)	70 (96%)	+ 4 (+ 5%)
							Pre-RxP	44	1 (2%)	5 (11%)	36 (82%)	2 (5%)	38 (86%)	41 (93%)	+ 3 (+ 7%)
							Post-RxP	29	0 (0%)	1 (3%)	28 (97%)	0 (0%)	28 (97%)	29 (100%)	+ 1 (+ 3%)
Dieci 201943Italy	ROXANE; 9 centres2017–8	HR+ HER2−	LN1–3Most high clinical risk (mAOL)	R-R	All (55% post)	RS < 11	–	31	–	–	–	–	19 (61%)	3 (10%)	−16 (−52%)
						RS 11–25	–	61	–	–	–	–	28 (46%)	17 (28%)	−11 (−18%)
						RS 11–17	–	NR	–	–	–	–	−(49%)	−(19.5%)	NR (−29.5%)
						RS 18–25	–	NR	–	–	–	–	−(40%)	−(45%)	NR (+ 5%)
						RS ≥ 26	–	7	–	–	–	–	7 (100%)	7 (100%)	No change

TABLE 16 Existing economic evaluations – analytic scope

Author	Population	Per cent LN+	Age	Intervention	Comparator	Country	Perspective	Time horizon	Discount rate (%)
Berdunov et al. (2021)92	Patients with ER+/HER2− EBC and one to three positive axillary lymph nodes, unrestricted by clinical or genomic risk	100%	Starting age unclear	Oncotype DX	Clinical risk tools alone	UK	NHS and PSS	Lifetime	3.5
Hinde et al. (2019)93	Women with ER+, HER2− EBC	95%a	Mean age 56.5 years	EndoPredict (EPclin)	Standard risk tools only	UK	NHS	Lifetime	3.5
Masucci et al. (2019)94	Patients with ER+, HER2−, LN+ EBC	100%	Mean age 60 years	Oncotype DX, MammaPrint, Prosigna, MammaTyper, IHC4-AQUA, IHC4	Current practice	Canada	Health care payer	Lifetime	1.5
Harnan et al. (2019)10	Patients with ER+, HER2−, LN+ EBC	100% in LN + subgroup	Mean age 58 years	Oncotype DX, EPclin, Prosigna, IHC4 + C, MammaPrint	Current practice	UK	NHS ans PSS	Lifetime	3.5
Hall et al. (2017)95	Women aged 40 years or older with ER+, HER2−, clinically high risk (1–9 axillary lymph nodes, or LN0 with a tumour size ≥ 30 mm) surgically treated early invasive breast cancer	81%	Starting age unclear	Oncotype DX, MammaPrint, Prosigna, MammaTyper, IHC4-AQUA, IHC4	Chemotherapy for all	UK	NHS	Lifetime	3.5
Stein et al. (2016)96	ER+, HER2− ESBC patients	100%	Median age 58 years	Oncotype DX; MammaPrint/Bluetest; Prosigna	Chemotherapy for all	UK	NHS	Lifetime (up to age 100 years)	3.5
Hannouf et al. (2014)97	Post-menopausal women with ER+/PR+ axillary LN+ ESBC	100%	Mean age 61 years	Oncotype DX	Current practice	Canada	Canadian public health care system	Lifetime	5.0
Blohmer et al. (2013)98	Patients with ER+, HER2−, LN0 or LN+ (up to 3 nodes) ESBC.	100% in LN + subgroup	Mean age 56.3 years	Oncotype DX	Conventional diagnostic procedures	Germany	Health care payer	30 years	3.0
Lamond et al. (2012)99	ER-sensitive, LN0 and LN+ BC	100% in LN+ subgroup	Median age 50 years	Oncotype DX	Current practice (population-based study)	Canada	Canadian health care system perspective	25 years	3.0
Hall et al. (2012)100	LN+, ER+ ESBC	100%	Baseline age 60 years	Oncotype DX	SC (chemotherapy for all)	UK	NHS	Lifetime (up to maximum age 100 years)	3.5
Wong et al. (2012)101	Women with LN+ HR+ breast cancer (1–3 nodes)	100%	Reflective of RxPONDER	Oncotype DX	Current care (US NCCN guidelines)	USA	Payer	Lifetime (40 years)	3.0
Vanderlaan et al. (2011)102	Minimally LN+, ESBC	100%	Mean age 62 years	Oncotype DX	Current care (US NCCN guidelines)	USA	US payer (managed care) perspective	30-years	3.0

TABLE 17 Existing economic evaluations – modelling approach and assumptions regarding predictive benefit and chemotherapy use

Author	Model approach	Cycle length	Model type	Does model claim predictive benefit for test?	Assumptions on chemotherapy use	Long-term health states
Berdunov et al. (2021)92	Decision tree and Markov model	6 months	Classification to low-, intermediate- and high risk	Yes – predictive benefit included in base-case analysis. Scenarios assuming no predictive benefit also presented	Chemotherapy use following RS based on the Clalit registry. Chemotherapy use in current practice based on NCRAS data used in DG34	4 states: (1) recurrence-free; (2) distant recurrence; (3) AML; (4) dead
Hinde et al. (2019)93	Decision tree and Markov model	1 year	Classification to low-, intermediate- and high risk	No – single HR applied for chemotherapy benefit	Chemotherapy with and without EPclin drawn directly from trial (Bloomfield et al.)	3 states: (1) disease-free; (2) metastases; (3) death
Masucci et al. (2019)94	Markov model	1 year	Classification to low-, intermediate- and high risk	Yes – HRs based on SWOG-8814 and clinical expert opinion	Based on literature and clinical opinion	9 states: (1) chemotherapy; (2) chemotherapy nausea/vomiting; (3) chemotherapy FN; (4) chemotherapy; (5) disease-free; (6) distant recurrence; (7) CHF; (8) leukaemia; (9) death
Harnan et al. (2019)10	Decision tree and Markov model	6 months	Classification to low-, intermediate- and high risk	Base-case analyses assume no predictive benefit for any test. Scenario analysis presented for predictive benefit for Oncotype DX only	Chemotherapy use following test based on literature. Chemotherapy use in current practice based on NCRAS data	4 states: (1) recurrence-free; (2) distant recurrence; (3) AML; (4) dead
Hall et al. (2017)95	Decision tree and modified Markov model	1 year	Classification to low and high risk	Yes – predictive benefit incorporated by modelling log HR for 10-year RFS as linear function of Oncotype DX RS.	All high-risk patients receive chemotherapy	6 health states: (1) disease-free; (2) distant recurrence; (3) LR; (4) disease-free after LR; (5) HF; (6) dead.
Stein et al. (2016)96	Decision tree and modified Markov model	1 year	Classification to low and high risk	Separate analyses undertaken including predictive benefit and assuming constant benefit across risk groups	All high-risk patients receive chemotherapy	7 states: (1) disease-free; (2) distant recurrence; (3) LR; (4) disease-free after LR; (5) CHF; (6) CML; (7) dead.
Hannouf et al. (2014)97	Markov	1 month	Classification to low-, intermediate- and high risk with separate Markov nodes for CT + ET vs. ET alone (accounting for chemotherapy-related AEs)	Unclear – appears to assume predictive benefit	Model assumes 50% IR patients receive chemotherapy	ET only model – 5 states: (1) remission; (2) LR; (3) distant recurrence; (4) dead. CT + ET model – 5 states: (1) remission with chemotherapy SAEs; (2) remission without chemotherapy SAEs; (3) LR; (4) distant recurrence; (5) dead.
Blohmer et al. (2013)98	Decision tree and Markov model	1 year	Classification to low-, intermediate- and high risk	Yes – relative risk reductions of 0% applied to LR and IR, relative risk reduction of 41% applied to HR	Based on data reported by Eiermann et al.	3 states: (1) recurrence-free; (2) distant recurrence; (3) dead
Lamond et al. (2012)99	Markov	1 month	Classification to low-, intermediate- and high risk	Yes – only in low risk and high risk	For no test, based on Canadian population-based study; for test, based on RS score. Usage in intermediate group assumed to be the same in both groups	10 states: (1) chemotherapy; (2) CINV; (3) FN; (4) disease-free; (5) local relapse; (6) distant relapse; (7) treated local relapse; (8) AML/MDS; (9) CHF; (10) dead.
Hall et al. (2012)100	Decision tree and modified Markov model	NR	Classification to low and high risk	Unclear – data contained within the appendices appear to suggest predictive benefit is modelled	All high-risk patients receive chemotherapy	6 states: (1) disease-free; (2) distant recurrence; (3) LR; (4) disease-free after LR; (5) CHF; (6) dead.
Wong et al. (2012)101	Decision tree with partitioned survival approach to determine sojourn time	NR	For patients whose treatment decision was based on US NCCN criteria classification to low risk or high risk. For patients whose treatment was based on the Oncotype DX test results classification to low-, intermediate- or high risk	Yes – different treatment effects applied for each risk category	~55% women assumed to receive chemotherapy	Not clearly reported – appears to be 3 states: (1) disease-free; (2) relapsed; (3) dead.
Vanderlaanet al. (2011)102	Appears to be Markov	NR	Classification to low and high risk.	No – same recurrence rates for all high-risk patients	71% of women in usual care assumed to receive chemotherapy treatment	3 states: (1) non-progressed disease; (2) progressed disease; (3) death.

TABLE 18 Summary of economic comparisons presented in the Exact Sciences CS

Analysis	Population	Intervention and comparators	Key sources of DRFI risk and chemotherapy benefit	Chemotherapy benefit assumptions	Additional EAG comments
Base-case analysis, overall LN+ population	ER+, HER2−, LN+ (1–3 nodes), pre- and post-menopausal	Oncotype DX Clinical-pathological tools alone	RxPONDER,78TransATAC,20 SWOG-881432	Predictive benefit assumed for Oncotype DX	Analysis uses weighted DRFI risk and HRs for the pre-menopausal and post-menopausal subgroups.
Base-case analysis, pre-menopausal LN+ subgroup	ER+, HER2−, LN+ (1–3 nodes), pre-menopausal	Same as overall LN + population	RxPONDER (pre-menopausal subgroup),78 TransATAC,20 SWOG-881432	Predictive benefit assumed for Oncotype DX	Relevant to decision problem set out in final NICE scope.11
Base-case analysis, post-menopausal LN+ subgroup	ER+, HER2−, LN+ (1–3 nodes), post-menopausal	Same as overall LN+ population	RxPONDER (post-menopausal subgroup),78 TransATAC,20 SWOG-881432	Predictive benefit assumed for Oncotype DX	Relevant to decision problem set out in final NICE scope.11
Exploratory analyses	ER+, HER2−, LN+ (1–3 nodes)Data for comparators reflect exclusively or mostly post-menopausal patients	Oncotype DX MammaPrint EndoPredict EPclin Prosigna Clinical-pathological tools aloneAll analyses presented as pairwise comparisons of test vs. clinical–pathological tools alone	RxPONDER,78TransATAC,20 SWOG-8814,32 MINDACT,30 EBCTCG16	Predictive benefit assumed for Oncotype DX	Relevant to decision problem set out in final NICE scope.11

TABLE 19 Key evidence sources used to inform the Exact Sciences base-case analyses (overall LN+ population)

Parameter group	Source	EAG comments
Clinical parameters
Start age	Unclear	The start age of 55 years is not cited in Exact Sciences CS.23
Test risk classification probabilities	RS 0–13 and RS 14–25: RxPONDER29RS > 25: Number of women who were excluded from RxPONDER29 due to RS > 25 divided by number of women registered for screening in the trial	Assumes that all other women who were screened and excluded from RxPONDER29 had an RS of 0–25.
DRFI probabilities for ET alone	RS 0–25: RxPONDER, ET arm29RS > 25: TransATAC20	Use of external data from TransATAC20 is necessary because patients with an RS of > 25 were excluded from RxPONDER.29
Risk tapering	Ward et al.105 and expert opinion	Same as DG34 model up to 15 years.
Chemotherapy probability under current decision-making	Holt et al.18	Unpublished decision impact study of Oncotype DX in women with LN+ early breast cancer. These data were submitted to NICE as part of the Exact Sciences CS and the submission from the Peony Breast Cancer Unit.
Chemotherapy probability conditional on Oncotype DX result	Holt et al.18	Unpublished decision impact study of Oncotype DX in women with LN+ early breast cancer. These data were submitted to NICE as part of the Exact Sciences CS and the submission from the Peony Breast Cancer Unit.
Chemotherapy benefit	RS 0–25: RxPONDER, CET arm29	The inclusion of the HR from SWOG-881432 for the RS > 25 group indirectly introduces an assumption of predictive benefit for Oncotype DX.
	RS > 25: HR from SWOG-881432 applied to ET risk from TransATAC20
Death risk with DM	Abemaciclib plus fulvestrant arm of MONARCH2 trial106	Not fully consistent with model assumptions about treatments for DM, whereby only 65% of patients receive a CDK4/6 inhibitor as first-line therapy.
Probability of AML	Petrelli et al.107	The Exact Sciences CS23 justifies the use of Petrelli et al.107 on the basis that it is more recent than Wolff et al.108 However, Wolff et al. is more recent than Petrelli et al.
Death risk with AML	NICE TA552109 (liposomal cytarabine–daunorubicin for untreated AML)	Reflects a more recent source than that used in the DG34 model. The use of the median OS underestimates mean OS.
Probability of LR	De Bock et al.110	Same as DG34 model.
HRQoL parameters
Utility, recurrence-free	Lidgren et al.111	Same as DG34 model.
Utility, DM	Lidgren et al.111	Same as DG34 model.
Utility, AML	NICE TA552109	Consistent with source of modelled AML mortality risk.
QALY loss, chemotherapy	Campbell et al.112	Same as DG34 model.
QALY loss, LR	Campbell et al.112	Same as DG34 model.
Utility age adjustment	Ara and Brazier104 (banded estimates)	Values reported by Hernández-Alava et al.113 are more up to date. The use of age bands is unnecessary.
Resource use and cost parameters
Tumour profiling test costs	NICE DG3413 (list prices)	Price discounts for other tumour profiling tests are not known to the company.
Adjuvant chemotherapy regimens used and associated resource use	Clinical opinion. Costing approach includes acquisition, administration and supportive medications.	The EAG’s clinical advisors agreed that the assumed distribution of chemotherapy regimens generally reflects current practice but noted that there is an increasing shift away from anthracycline-based regimens in certain patient groups.
ET usage	Ward et al.105	Based on DG34 model.
Treatments for DM	Clinical opinion, Kurosky et al.114 and MONARCH2 trial106	The model assumes that 65% of women with DM will receive a CDK4/6i as first-line treatment. The EAG’s clinical advisors commented that CDK4/6i treatment would now be offered as first-line therapy for the vast majority of women with distant recurrence.
Cost AML (initial one-off cost and ongoing cyclical cost)	Zeidan et al.115	The model applies a once-only cost of intensive therapy in the first 6 months after diagnosis and an ongoing 6-monthly cost to reflect the cost of BSC for patients surviving beyond the initial 6 months.
AE frequency	TACT trial116	The model applies the frequency of AEs associated with FEC-D to anthracycline–taxane combinations and the costs of FEC60 to all other regimens.
Unit costs	NHS Reference Costs,117,118 eMIT,119 BNF,120 PSSRU121	Appropriate sources applied
Cost of death	Hinde et al.93	–

TABLE 20 Summary of cost-effectiveness results presented in the Exact Sciences CS (based on the company’s revised model provided as part of their clarification response)

Analysis type	Base case, overall LN+ population	Base case, LN+ pre-menopausal subgroup	Base case, LN+ post-menopausal subgroup
Deterministic ICERs	Oncotype DX dominates clinical–pathological tools alone	Oncotype DX is dominated by clinical–pathological tools alone	Oncotype DX dominates clinical–pathological tools alone
Probabilistic ICERs	Oncotype DX dominates clinical–pathological tools alone	Oncotype DX is dominated by clinical–pathological tools alone	Oncotype DX dominates clinical–pathological tools alone
Probability test is cost-effective at WTP = £20,000/QALY gained	Probability = 0.93	Probability = 0.07	Probability = 1.00
DSAs	Assuming WTP = £20,000/QALY, NMB is positive for all DSAs except lower value of HR from RxPONDER	Assuming WTP = £20,000/QALY, NMB is negative for all DSAs except upper value of HR from RxPONDER	Assuming WTP = £20,000/QALY, NMB is positive for all DSAs
Additional scenario analyses	Oncotype DX dominates clinical–pathological tools alone in all scenarios tested	Oncotype DX is dominated by clinical–pathological tools alone in all scenarios tested	Oncotype DX dominates clinical–pathological tools alone in all scenarios tested
Exploratory analyses of other tests vs. clinical–pathological tools alone	Oncotype DX dominates clinical–pathological tools alone.	Not presented	Not presented
	MammaPrint: ICER vs. clinical–pathological tools alone = £50,626 per QALY gained
	EPclin: ICER vs. clinical–pathological tools alone = £9355 per QALY gained
	Prosigna: ICER vs. clinical–pathological tools alone = £41,773 per QALY gained

TABLE 21 Summary of economic analyses presented in the Cytel CEA report on MammaPrint

Analysis	Population	Intervention and comparators	Key sources of DRFI risk and chemotherapy benefit	Chemotherapy benefit assumptions	Additional EAG comments
Base case.	Clinical high-risk. LN0 NPI > 3.4 and LN+ patients weighted in blended analysis. Clinical high-risk patients (ER+, HER2−).	MammaPrint Oncotype DX Prosigna EPclin Usual care	TransATAC,20 MINDACT,30 EBCTCG16	Predictive benefit included for MammaPrint. All other tests assumed to be prognostic only.	Assumes mAOL high risk is equivalent to NPI > 3.4.Analysis includes a minority of LN+ patients.
Scenario 1. Full ER+, HER2−population stratified by 2-level clinical test.	Clinical low-risk and clinical high-risk patients. LN0 NPI > 3.4 and LN+ patients weighted in blended analysis.	Same as Agendia base case			Clinical low-risk patients included in model, but only clinical high-risk patients get the genomic test.
					Analysis includes a minority of LN+ patients.
Scenario 2. Full ER+, HER2–population – stratified by 3-level clinical test.	Clinical low-risk and clinical high-risk patients. LN0 NPI > 3.4 and LN+ patients weighted in blended analysis.	Same as Agendia base case			Analysis partitions population into clinical low-risk, clinical high-risk and LN+, but only clinical high-risk patients receive the genomic test.
					Analysis includes a minority of LN+ patients.
Scenario 3. ER+, HER2− post-menopausal women stratified by 2-level clinical test.	Post-menopausal clinical low-risk and clinical high-risk patients. LN0 NPI > 3.4 and LN+ patients weighted in blended analysis.	Same as Agendia base case			Clinical low-risk patients included, but only clinical high-risk patients receive the genomic test.
					Analysis includes a minority of LN+ patients.
Scenario 4. TAILORx clinical study stratified by 2-level clinical test.	Clinical low-risk and clinical high-risk patients. LN0 patients only.	MammaPrint Oncotype DX Usual care.	TAILORx127 MINDACT30	Predictive benefit included for MammaPrint and Oncotype DX.	Clinical low-risk patients included, but only clinical high-risk patients receive the genomic test. Analysis excludes LN+ patients.
Scenario 5. ER+, HER2−, LN+ subgroup.	Clinical high risk. LN+ patients only.	Same as base-case scenario, but restricted to LN+ subgroups from MINDACT30 and TransATAC20			This is the only analysis which directly addresses the decision problem set out in the NICE scope.11

TABLE 22 Key evidence sources used to inform Agendia model – base-case analysis and Scenario 5 (pure LN+ subgroup)

Parameter group	Source	EAG comments
Clinical parameters
Start age	NHSE Access Dataset129	Reflects patients with LN0 disease.
Test risk classification probabilities	MINDACT,30 TransATAC20	Analyses between MammaPrint and other tests assume equivalence of populations enrolled in TransATAC20 and MINDACT.30MammaPrint data are based on the HR+/HER2− population of MINDACT (LN0 and/or LN+)30
DRFI probabilities	MINDACT,30 TransATAC20	DRFI for MammaPrint low-risk ET and MammaPrint high-risk chemotherapy plus ET groups are based on reanalyses of MINDACT IPD.85
Risk tapering	Ward et al.105	Same as DG34 model.
Chemotherapy probability under usual care	NCRAS17 and expert opinion85	The Cytel CEA report85 states that ‘there are no empirical evidence sources which provide estimates of baseline chemotherapy use for patients who are mAOL high-risk or mAOL low risk’, hence the need to rely on expert opinion.
Chemotherapy probability conditional on genomic tests	MammaPrint – Kuijer et al.130Oncotype DX – Crolley et al.131Prosigna – UKBCG survey (3-level)10EPclin – UKBCG survey (2-level)10	Use of different sources for each test implicitly assumes that 2-level tests (MammaPrint and EPclin) are interpreted differently and that 3-level tests (Oncotype DX and Prosigna) are interpreted differently.Crolley et al.,131 was undertaken in a purely LN0 population and Kuijer et al.130 was undertaken in women without axillary lymph node involvement (pN0 or pN1mi).
Chemotherapy benefit for MammaPrint	Assumptions of predictive benefit based on interpretation of MINDACT data30	Assumes predictive benefit – RRs of 1.00 and 0.38 are applied to patients receiving CET in the MammaPrint low- and high-risk patients, respectively. RRs of 0.97 and 0.28 are applied in Scenario 5 (LN+ subgroup).
Chemotherapy benefit for other tests	EBCTCG meta-analysis16	An RR of 0.76 is applied to all patients receiving CET in the Oncotype DX, Prosigna, EPclin and usual care comparator groups.
Death risk with DM	Wang et al.132	This data set reflects patients diagnosed between 2010 and 2015 and therefore is unlikely to reflect survival for patients receiving current first-line treatments for DM (e.g. CDK4/6i therapies).
Probability of AML	Wolff et al.108	Same as DG34 model.
Death risk with AML	Edlin et al.133	Same as DG34 model.
LR transition probabilities	Geurts et al.134	LR is assumed to impact only on QALYs and costs without affecting the patient’s underlying health state or survival.
HRQoL parameters
Utility values, recurrence-free with adjuvant CET or ET alone	Lidgren et al.111	Disutility value applied for patients receiving CET for 3 years. The source of the assumed the duration of the disutility is unclear.
Utility value, DM	Lidgren et al.111	Same as DG34 model.
Utility value, AML	Younis et al.135	Same as DG34 model.
Disutility, LR	Campbell et al.112	Same as DG34 model.
Utility age adjustment	Ara and Brazier104	Values reported by Hernández-Alava et al.113 are more up to date.
Resource use and cost parameters
Tumour profiling test costs	NICE DG3411 (list prices)	Price discounts for other tests are not known to the company. Test prices are uplifted using inflation indices.
Adjuvant chemotherapy regimens used and associated resource use	Clinical opinion. Structure of costing approach based on Hall et al.95	Updated from DG34 model.
ET usage	Ward et al.105	Same as DG34 model.
Bisphosphonates	Ward et al.105	Similar to DG34 model.
Cost DM	Thomas et al.136	Same as DG34 model.
Cost AML	Russell-Smith et al.137	The population of the model reported by Russell-Smith et al. relates to patients with de novo AML which is not therapy-related.
AE frequency	Based on various trials138–142	Values used in the company’s original model are unclear and appear highly inflated. These values were amended in the company’s revised model.
Unit costs	NHS Reference Costs,143 eMIT,144 MIMS,145 PSSRU121	Appropriate sources applied.
Cost of death	Georgiou and Bardsley146	–

TABLE 23 Summary of cost-effectiveness results presented in the Cytel CEA report (includes company’s correction of errors at the clarification stage)

Analysis type	Base case – clinical high-risk patients	Scenario 1 – full population stratified by 2-level clinical test	Scenario 2 – full population – stratified by 3-level clinical test	Scenario 3 – post-menopausal women stratified by 2-level clinical test	Scenario 4 – TAILORx clinical study stratified by 2-level clinical test	Scenario 5 – LN+ subgroup
Deterministic ICERs	MammaPrint dominates all comparators	MammaPrint dominates all comparators	MammaPrint dominates all comparators	MammaPrint dominates all comparators	MammaPrint dominates Oncotype DX and usual care	MammaPrint dominates all comparators
Probabilistic ICERs	MammaPrint dominates all comparators	Not presented	Not presented	Not presented	Not presented	Not presented
Probability test is cost-effective at λ = £20,000	Probability = 0.91	Not presented	Not presented	Not presented	Not presented	Not presented
DSAs	Highest ICER for MammaPrint vs. comparators across all analyses = £392 per QALY gained	Not presented	Not presented	Not presented	Not presented	Not presented
Additional scenario analyses	Highest ICER for MammaPrint vs. comparators across all analyses = £3647 per QALY gained	Not presented	Not presented	Not presented	Not presented	Not presented

TABLE 24 Probability of receiving chemotherapy conditional on genomic risk classification applied in the Agendia model

Test	Low-risk	High-risk	Source
2-level tests
MammaPrint	0.05	0.97	Kuijer et al.130
EPclin	0.15	0.92	UKBCG survey10
3-level tests
Oncotype DX	0.03	0.91	Crolley et al.131
Prosigna	0.04	0.92	UKBCG survey10

TABLE 25 Utility values associated with CET applied in the Agendia model

Model health state	Utility value	Source
Recurrence-free, CET, year 1	0.620	Lidgren et al.111 State p ‘First year after primary breast cancer’, receiving adjuvant chemotherapy
Recurrence-free, CET, year 2–3	0.743	Value used in model NR in Lidgren et al.111
Recurrence-free, CET, year 4+	0.824	Lidgren et al.111 State S ‘Second and following years after primary breast cancer/recurrence’, receiving ET
Recurrence-free, ET alone, year 1	0.744	Lidgren et al.111 State p ‘First year after primary breast cancer’, receiving ET
Recurrence-free, ET alone, year 2+	0.824	Lidgren et al.111 State S ‘Second and following years after primary breast cancer/recurrence’, receiving ET

TABLE 26 Adjuvant chemotherapy disutility values and QALY losses applied in the Agendia model and other models included in the EAG’s systematic review

Model	Disutility/QALY loss associated with adjuvant chemotherapy toxicity per patient treated	Source of disutility value/QALY loss
Agendia model85	QALY loss of 0.29	Lidgren et al.111
Harnan et al. (2019)10	QALY loss of 0.04	Campbell et al.112
Berdunov et al. (2021)92	Disutility of −0.04, 1-year duration	Campbell et al.112
Blohmer et al. (2013)98	QALY loss of 0.07	Peasgood et al.147
Hall et al. (2012)100	QALY loss of 0.08	Lidgren et al.111
Hall et al. (2017)95	Disutility of −0.096, 1-year duration	Campbell et al.112
Hannouf et al. (2014)97	QALY loss in first year of approximately 0.025. Treatment-specific utility values favouring ET over chemotherapy are applied thereafter.	Assumptions
Hinde et al. (2019)93	QALY loss of 0.12	Lidgren et al.111
Lamond et al. (2012)99	QALY loss appears to be around 0.06 (excluding the impact of CINV and FN)	Tufts,148 Tengs,149 Ward105
Masucci et al. (2019)94	QALY loss of 0.06	Lidgren et al.111
Stein et al. (2016)96	Disutility of −0.096, 1-year duration	Lidgren et al.111
Vanderlaan et al. (2011)102	QALY loss of 0.50 over lifetime	Assumption
Wong et al. (2012)101	Disutility of −0.30, duration unclear	Assumption

TABLE 27 Results of additional analysis undertaken by the EAG

Option	LYGsa	QALYs	Costs	Inc. LYGs	Inc. QALYs	Inc. costs	ICER
Company’s Scenario Analysis 5 (LN+), other tests excluded
MammaPrint	23.45	11.57	£23,327	1.43	0.70	−£4676	MammaPrint dominating
Usual care	22.02	10.87	£28,003	–	–	–
Company’s Scenario Analysis 5 (LN+), other tests excluded, including EAG amendments
MammaPrint	23.86	11.60	£21,570	−0.08	0.02	−£26	MammaPrint dominating
Usual care	23.93	11.59	£21,596	–	–	–

TABLE 28 Scope of the EAG economic analysis

Population	Women with ER+/PR+, HER2−, LN+ early breast cancer (1–3 nodes).
	For the evaluation of Oncotype DX using the newer cut-offs (RS 0–25 and RS > 25), pre-menopausal and post-menopausal subgroups are considered separately.
	For the evaluation of MammaPrint, the modelled population reflects the mAOL clinical high-risk ER+, HER2−, LN+ subgroup within the MINDACT trial.30
Interventions	1. Oncotype DX [two sets of cut-offs assessed: (a) new cut-offs – low RS 0–25, RS high > 25; (b) old cut-offs – low RS < 18; intermediate RS 18–30; high RS > 30].0
	2. Prosigna (cut-offs LN+: low 0–15, intermediate 16–40, high 41–100).
	3. EPclin (cut-off: 3.3: low < 3.3; high ≥ 3.3).
	4. MammaPrint (cut-off: low > 0, high ≤ 0).
Comparator	Current decision-making, which may include any tool, or clinical and pathological features, used to assess risk.
	For MammaPrint, the clinical high-risk subgroup is based on mAOL, as per the design of the MINDACT trial.30,150
	Due to evidence limitations, the tumour profiling tests are not compared incrementally against each other.a
Main economic outcome	Incremental cost per QALY gained
Additional model outcomes	Incremental LYGs
	Incremental QALYs gained
	Incremental costs
	Impact on chemotherapy use
Perspective	NHS and PSS
Time horizon	Lifetime
Discount rate	3.5% per annum
Price year	2022–3

TABLE 29 Evidence sources used in the EAG’s base-case model

Parameter group	Source
Patient characteristics
Patient age	Holt et al.18
Mean BSA	Assumption
Clinical parameters
Risk classification probabilities	TransATAC,20 RxPONDER,29 MINDACT30
6-month DRFI on ET alone	TransATAC,20 RxPONDER,78 MINDACT IPD85
Pre-test probability of receiving adjuvant chemotherapy	Holt et al.18,36
Post-test probability of receiving chemotherapy (3-level tests)	Holt et al.18,36
Post-test probability of receiving chemotherapy (2-level tests)	Holt et al.18,36
HRs for DM, CET vs. ET	EBCTCG,16 RxPONDER,78 SWOG-881432
6-month probability of death due to DM	Rebuilt model based on Suri et al.153
Probability of developing LR	De Bock et al.110
6-month probability of developing AML	Wolff et al.108
6-month probability of death due to AML	Rebuilt model based on Bewersdorf et al.154
Other-cause mortality	ONS life tables for England155
HRQoL parameters
Utility, recurrence-free	Lidgren et al.111
Utility, distant recurrence	Lidgren et al.111
Utility AML	Rebuilt model based on Bewersdorf et al.154
QALY loss due to chemotherapy-related AEs	Campbell et al.112
QALY loss due to LR	Campbell et al.112
Cost parameters
Tumour profiling test costs	Test manufacturers23–26
Adjuvant chemotherapy and supportive medications	Proportions based on expert opinion.92 Unit costs taken from eMIT156 and BNF.157
ET	Distribution and duration of treatments based on expert opinion. Drug costs taken from eMIT156 and BNF
Bisphosphonates	Proportion based on expert opinion. Unit costs taken from eMIT156 and BNF.157
AEs	Frequency based on Ellis et al.116 Costs taken from NHS Reference Costs 2021–2.143
Routine follow-up	Frequency based on expert opinion. Unit costs taken from Ward et al.105 and NHS Reference Costs 2021–2.143
LR (once-only cost)	Karnon et al.124
DM (lifetime cost)	Mean cost reported by Suri et al.153
AML (lifetime cost)	Costs of intensive therapy, HSCT and subsequent BSC from Zeidan et al.115 applied to rebuilt model based on Bewersdorf et al.154

TABLE 30 Risk classification probabilities used in the EAG’s model

Scenario	Scenario description	Test	Test risk classification probability	Source
Low	Int	High
BC1	RxPONDER pre-menopausal	Oncotype DX	0.89	N/A	0.11a	Kalinsky et al.29
BC2	RxPONDER post-menopausal	Oncotype DX	0.89	N/A	0.11a	Kalinsky et al.29
BC3	TransATAC, predictive	Oncotype DX	0.57	0.32	0.11	Sestak et al.20
BC4	TransATAC, non-predictive	Oncotype DX	0.57	0.32	0.11	Sestak et al.20
BC5	TransATAC, non-predictive	Prosigna	0.08	0.32	0.60	Sestak et al.20
BC6	TransATAC, non-predictive	EPclin	0.23	N/A	0.77	Sestak et al.20
BC7	MINDACT clinical high-risk	MammaPrint	0.69	N/A	0.31	Piccart et al.30

TABLE 31 Cumulative DRFI probabilities for ET alone used in the EAG’s model

Scenario	Scenario description	Test	DRFI time point reported	Cumulative DRFI	Source
Low	Int	High
BC1	RxPONDER pre-menopausal	Oncotype DX	RS0-25 : 5 yearsRS > 25 : 10 years	0.06	N/A	0.38	RS0-25: Kalinsky et al.78RS > 25: Sestak et al.20
BC2	RxPONDER post-menopausal	Oncotype DX	RS0-25 : 5 years	0.03	N/A	0.38	RS0-25: Kalinsky et al.78
			RS > 25 : 10 years				RS > 25: Sestak et al.20
BC3	TransATAC, predictive	Oncotype DX	10 years	0.19	0.29	0.38	Sestak et al.20
BC4	TransATAC, non-predictive	Oncotype DX	10 years	0.19	0.29	0.38	Sestak et al.20
BC5	TransATAC, non-predictive	Prosigna	10 years	0.00	0.21	0.31	Sestak et al.20
BC6	TransATAC, non-predictive	EPclin	10 years	0.06	N/A	0.30	Sestak et al.20
BC7	MINDACT clinical high-risk	MammaPrint	8 years	0.09	N/A	0.26a	MINDACT IPD85

TABLE 32 Hormone receptors for DM for chemotherapy vs. no chemotherapy applied in the EAG’s model

Scenario	Scenario description	Test	Test risk	Source of HRs
Low	Int	High
BC1	RxPONDER pre-menopausal	Oncotype DX	0.64	N/A	0.59	RS 0–25: Kalinsky et al.78
						RS > 25: Albain et al.3
BC2	RxPONDER post-menopausal	Oncotype DX	1.12	N/A	0.59	RS 0–25: Kalinsky et al.78
						RS > 25: Albain et al.32
BC3	TransATAC, predictive	Oncotype DX	1.02	0.72	0.59	Albain et al.32
BC4	TransATAC, non-predictive	Oncotype DX	0.71	0.71	0.71	EBCTCG16
BC5	TransATAC, non-predictive	Prosigna	0.71	0.71	0.71	EBCTCG16
BC6	TransATAC, non-predictive	EPclin	0.71	N/A	0.71	EBCTCG16
BC7	MINDACT clinical high-risk	MammaPrint	0.71	N/A	0.71	EBCTCG16

TABLE 33 Pre-test probability of receiving chemotherapy

Scenario	Scenario description	Test	Pre-test probability	Source
BC1-7	All scenarios	All tests	0.80	Holt et al.18

TABLE 34 Post-test probability of receiving chemotherapy

Scenario	Scenario description	Test	Test risk	Source
Low	Int	High
BC1	RxPONDER pre-menopausal	Oncotype DX	0.37	N/A	0.96	Holt et al.,18 pre-menopausal subgroup
BC2	RxPONDER post-menopausal	Oncotype DX	0.11	N/A	0.96	Holt et al.,18 post-menopausal subgroup
BC3	TransATAC, predictive	Oncotype DX	0.08	0.49	0.98	Holt et al.36
BC4	TransATAC, non-predictive	Oncotype DX	0.08	0.49	0.98	Holt et al.36
BC5	TransATAC, non-predictive	Prosigna	0.08	0.49	0.98	Holt et al.36
BC6	TransATAC, non-predictive	EPclin	0.11	N/A	0.96	Holt et al.,18 post-menopausal subgroup
BC7	MINDACT clinical high-risk	MammaPrint	0.19	N/A	0.96	Holt et al.,18 post-menopausal subgroup

TABLE 35 Utility values and QALY losses applied in the EAG’s model

Parameter	Mean value	Source
Utility, recurrence-free	0.824	Lidgren et al.111
Utility, DM	0.685	Lidgren et al.111
Utility AML	0.59	Estimated by dividing the mean QALYs by mean the LYGs in the rebuilt model based on Bewersdorf et al.154
QALY loss chemotherapy	−0.038	Campbell et al.112
QALY loss LR	−0.108	Campbell et al.112

TABLE 36 Summary of costs applied in the EAG’s model

Resource use component	Mean cost
Tumour profiling tests (list prices)
Oncotype DX	£2580
Prosigna	£1896
EPclin	£1500
MammaPrint	£2616
Adjuvant chemotherapy (once-only)	£7410.48
ET years 1–2 (per cycle)	£66.95
ET years 3–5 (per cycle)	£66.44
ET years 6–10 (per cycle)	£53.16
Bisphosphonates (per cycle)a	£320.84
Ovarian suppression (per cycle)b	£496.73
AEs (once-only)	£1249.58
Follow-up, year 1 (per year)	£360.48
Follow-up, years 2–5 (per year)	£139.00
LR (once-only)	£16,494.23
DM (once-only)	£117,482.09
AML (once-only)	£132,185.91
End of life care (once-only)	£4898.17

TABLE 37 Costs of tumour profiling tests

Test	List price excluding VAT	Source
Oncotype DX	£2580.00	Exact Sciences CS,23 May 2023. Price includes costs of all activities required to conduct the testing service, including shipping, materials, customer support, online customer portal for accessing orders and results information.
Prosigna	£1896.00	Veracyte RFI document,26 February 2023. Price reflects in-house NHS testing, including costs of gene signature assay, nCounter DX analysis, nCounter servicing, RNA isolation kit and laboratory staff costs.
EPclin	£1500.00	Myriad RFI document,25 March 2023. Price includes all reaction agents and consumables. Price reflects locally run testing service.
MammaPrint	£2616.00	Agendia value dossier,24 February 2023. Price includes transport, specimen processing and all other costs associated with reporting the result.

TABLE 38 Per-cycle adjuvant chemotherapy costs applied in the EAG’s model

Regimen	Proportion (%)	Drug acquisition cost per model cycle	Administration, pharmacy, visits and monitoring per model cycle	Total cost per model cycle
FEC75 (6 cycles)	0.00	£821.13	£4110.70	£4931.83
FEC100-T (3 + 3 cycles)	23.75	£1581.22	£4110.70	£5691.92
TC (4 cycles)	10.00	£1640.50	£2735.53	£4376.03
EC90/T75 (4 + 4 cycles)	28.75	£2101.75	£5485.87	£7587.62
EC90 (4 cycles)	0.00	£547.90	£2735.53	£3283.43
C-D (6 cycles)	2.50	£2458.68	£4110.70	£6569.38
TAC (6 cycles)	1.25	£2509.74	£4110.70	£6620.44
Accelerated EC90/P (4 + 4 cycles)	23.75	£3381.97	£5485.87	£8867.84
Weekly P (12 weeks)	2.50	£155.13	£8236.21	£8391.34
EC/weekly P (4 cycles, 12 weeks)	7.50	£703.03	£10,986.55	£11,689.58
Weighted cost	–	£2096.50	£5313.98	£7410.48

TABLE 39 Frequency of AEs and unit costs applied in the EAG’s model

AE	FEC-D – frequency	Control – frequency	Unit cost	Cost source
Anaemia	0.006	0.007	£1439.66	NHS Reference Costs 2021–2:143 SA04G-L, non-elective short and long stay
Febrile neutropoenia	0.071	0.029	£3676.55	NHS Reference Costs 2021–2:143 SA35A-E, non-elective long stay
Leucopoenia	0.246	0.175	£501.80	NHS Reference Costs 2021–2:143 weighted average of JA12D-L, non-elective short and long stay and NHS Reference Costs 2020–1: consultant-led outpatient visit, WF01A medical oncology
Neutropoenia	0.455	0.384	£501.80	NHS Reference Costs 2021–2:143 weighted average of JA12D-L, non-elective short and long stay and NHS Reference Costs 2020–1: consultant-led outpatient visit, WF01A medical oncology
Thrombocytopenia	0.006	0.013	£2163.16	NHS Reference Costs 2021–2:143 SA12G-K, non-elective short and long stay
Alopecia	0.102	0.103	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Diarrhoea	0.037	0.028	£1446.84	NHS Reference Costs 2021–2:143 FD10J-M, non-elective short and long stay
Infection	0.142	0.088	£1628.07	NHS Reference Costs 2021–2:143 DZ22K-Q, non-elective short and long stay
Lethargy	0.221	0.131	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Musculoskeletal (other)	0.070	0.015	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Myalgia/arthralgia	0.050	0.001	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Nausea/vomiting	0.097	0.099	£1579.14	NHS Reference Costs 2021–2:143 FD11K, non-elective short and long stay
Neuropathy	0.048	0.005	£1886.35	NHS Reference Costs 2021–2:143 AA26C-H, non-elective short and long stay
Oedema	0.008	0.003	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Pain	0.028	0.001	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Skin disorder (including nail changes)	0.033	0.012	£221.48	NHS Reference Costs 2021–2:143 consultant-led outpatient visit, WF01A medical oncology
Stomatitis	0.076	0.036	£1978.14	NHS Reference Costs 2021–2:143 CB01F, non-elective short and long stay

TABLE 40 Central estimates of cost-effectiveness, all EAG base-case comparisons, probabilistic

Option	LYGsa	QALYs	Costs	Inc. LYGsa	Inc. QALYs	Inc. costs	Incremental cost per QALY gained
BC1 – Oncotype DX, RxPONDER pre-menopausal (predictive benefit)
Oncotype DX	32.73	14.25	£41,631	−0.66	−0.18	£1810	Dominated
Current DM	33.39	14.43	£39,821	–	–	–	–
BC2 – Oncotype DX, RxPONDER post-menopausal (predictive benefit)
Oncotype DX	21.82	11.18	£26,546	0.21	0.11	-£4273	Dominating
Current DM	21.61	11.07	£30,818	–	–	–	–
BC3 – Oncotype DX, TransATAC, post-menopausal (predictive benefit)
Oncotype DX	19.29	10.11	£47,762	0.05	0.04	−£1942	Dominating
Current DM	19.24	10.07	£49,704	–	–	–	–
BC4 – Oncotype DX, TransATAC, post-menopausal (non-predictive benefit)
Oncotype DX	19.28	10.11	£47,806	−0.44	−0.17	£1811	Dominated
Current DM	19.72	10.28	£45,994	–	–	–	–
BC5 – Prosigna, TransATAC, post-menopausal (non-predictive benefit)
Prosigna	19.73	10.28	£47,427	0.06	0.03	£1084	£39,357
Current DM	19.67	10.25	£46,342	–	–	–	–
BC6 – EPclin, TransATAC, post-menopausal (non-predictive benefit)
EPclin	19.88	10.34	£45,786	0.13	0.06	£231	£4113
Current DM	19.75	10.29	£45,555	–	–	–	–
BC7 – MammaPrint, MINDACT, LN+ subgroup (non-predictive benefit)
MammaPrint	24.50	12.04	£40,614	−0.22	−0.07	£786	Dominated
Current DM	24.72	12.10	£39,828	–	–	–	–

TABLE 41 Central estimates of cost-effectiveness, all base-case comparisons, deterministic

Option	LYGsa	QALYs	Costs	Inc. LYGsa	Inc. QALYs	Inc. costs	Incremental cost per QALY gained
BC1 – Oncotype DX, RxPONDER pre-menopausal (predictive benefit)
Oncotype DX	32.69	14.24	£41,814	−0.65	−0.18	£1787	Dominated
Current DM	33.34	14.42	£40,027	–	–	–	–
BC2 – Oncotype DX, RxPONDER post-menopausal (predictive benefit)
Oncotype DX	21.81	11.23	£26,630	0.21	0.11	−£4283	Dominating
Current DM	21.60	11.12	£30,913	–	–	–	–
BC3 – Oncotype DX, TransATAC, post-menopausal (predictive benefit)
Oncotype DX	19.26	10.15	£48,145	0.08	0.05	−£2300	Dominating
Current DM	19.18	10.10	£50,444	–	–	–	–
BC4 – Oncotype DX, TransATAC, post-menopausal (non-predictive benefit)
Oncotype DX	19.27	10.16	£47,986	−0.45	−0.17	£1862	Dominated
Current DM	19.72	10.33	£46,124	–	–	–	–
BC5 – Prosigna, TransATAC, post-menopausal (non-predictive benefit)
Prosigna	19.71	10.32	£47,650	0.06	0.03	£1108	£40,220
Current DM	19.65	10.30	£46,543	–	–	–	–
BC6 – EPclin, TransATAC, post-menopausal (non-predictive benefit)
EPclin	19.86	10.39	£46,080	0.12	0.05	£305	£5580
Current DM	19.74	10.33	£45,775	–	–	–	–
BC7 – MammaPrint, MINDACT, LN+ subgroup (non-predictive benefit)
MammaPrint	24.50	12.06	£40,621	−0.22	−0.07	£792	Dominated
Current DM	24.71	12.13	£39,830	–	–	–	–

TABLE 42 Deterministic sensitivity analysis results for all base-case comparisons – test vs. current decision-making

DSA	BC1 – Oncotype DX, RxPONDER pre-menopausal, predictive	BC2 – Oncotype DX, RxPONDER post-menopausal, predictive	BC3 – Oncotype DX, TransATAC post-menopausal, predictive	BC4 – Oncotype DX, TransATAC post-menopausal, non-predictive	BC5 – Prosigna, TransATAC post-menopausal, non-predictive	BC6 – EPclin, TransATAC post-menopausal, non-predictive	BC7 – MammaPrint, MINDACT, non-predictive
Deterministic base case ICER	Dominated	Dominating	Dominating	Dominated	£40,220	£5580	Dominated
DSA1: 17% of women assumed to be in RS > 25 group (Oncotype DX only)	Dominated	Dominating	N/A	N/A	N/A	N/A	N/A
DSA2: Prosigna test classification probabilities and DRFI from Gnant et al.,56	N/A	N/A	N/A	N/A	£23,853	N/A	N/A
DSA3: EPclin test classification probabilities and DRFI from Filipits et al.163	N/A	N/A	N/A	N/A	N/A	Dominated	N/A
DSA4: 3-level post-test chemotherapy probabilities – Llombart Cussac et al.45	N/A	N/A	Dominating	Dominated	£37,959	N/A	N/A
DSA5: 3-level post-test chemotherapy probabilities – Loncaster et al.37	N/A	N/A	Dominating	Dominated	Dominated	N/A	N/A
DSA6: 3-level post-test chemotherapy probabilities – Zambelli et al.46	N/A	N/A	Dominating	Dominated	£62,801	N/A	N/A
DSA7: 2-level post-test chemotherapy probabilities – Dieci et al.43	Dominated	Dominating	N/A	N/A	N/A	£6448	Dominated
DSA8: 3-level post-test chemotherapy probabilities – UKBCG survey (3-level tests)10	N/A	N/A	Dominating	Dominated	£37,092	N/A	N/A
DSA9: 2-level post-test chemotherapy probabilities – UKBCG survey (2-level tests)10	Dominated	Dominating	N/A	N/A	N/A	£12,606	Dominated
DSA10: Risk tapering to 50% at 10 years then 0% at 15 years	Dominated	Dominating	Dominating	Dominated	£40,876	£7097	Dominated
DSA11: CET vs. ET HR = 0.60 in all genomic risk groups (non-predictive)	Dominated	£9772 (SWQ)	Dominated	Dominated	£24,584	Dominating	Dominated
DSA12: CET vs. ET HR = 0.71 in all genomic risk groups (non-predictive)	Dominated	£42,518 (SQW)	Dominated	Dominated	£40,220	£5580	Dominated
DSA13: CET vs. ET HR = 0.80 in all genomic risk groups (non-predictive)	Dominated	£279,599 (SWQ)	Dominated	Dominated	£60,336	£14,493	Dominated
DSA14: Chemotherapy QALY loss halved	Dominated	Dominating	Dominating	Dominated	£44,427	£5820	Dominated
DSA15: Chemotherapy QALY loss doubled	Dominated	Dominating	£757,556 (SWQ)	Dominated	£49,618	£6080	Dominated
DSA16: Chemotherapy QALY loss tripled	Dominated	Dominating	£106,021 (SWQ)	Dominated	£53,808	£6267	Dominated
DSA17: Baseline probability of chemotherapy = 0.90	Dominated	Dominating	Dominating	Dominated	Dominated	£13,402	Dominated
DSA18: Start age + 5 years	Dominated	Dominating	Dominating	Dominated	£52,697	£8137	Dominated
DSA19: Start age – 5 years	Dominated	Dominating	Dominating	Dominated	£33,567	£4379	Dominated
DSA20: Utility values from Verrill et al.162	Dominated	Dominating	Dominating	Dominated	£44,393	£6172	Dominated
DSA21: AML removed from model	Dominated	Dominating	Dominating	Dominated	£47,629	£7274	Dominated
DSA22: Chemotherapy cost halved	Dominated	Dominating	Dominating	Dominated	£46,376	£8253	Dominated
DSA23: Chemotherapy cost doubled	£5007 (SWQ)	Dominating	Dominating	£10,361 (SWQ)	£27,908	£235	£29,702 (SWQ)
DSA24: DM lifetime cost halved	Dominated	Dominating	Dominating	£524 (SWQ)	£46,275	£12,758	£1239 (SWQ)
DSA25: DM lifetime cost doubled	Dominated	Dominating	Dominating	Dominated	£28,111	Dominating	Dominated
DSA26: AML costs halved	Dominated	Dominating	Dominating	Dominated	£41,175	£6066	Dominated
DSA27: AML costs doubled	Dominated	Dominating	Dominating	Dominated	£38,311	£4608	Dominated

TABLE 43 Model-predicted incremental clinical and economic outcomes per 1000 women tested – test vs. current decision-making

Incremental model outcome (test vs. current decision-making)	BC1 – Oncotype DX, RxPONDER pre-menopausal, predictive	BC2 – Oncotype DX, RxPONDER post-menopausal, predictive	BC3 – Oncotype DX, TransATAC post-menopausal, predictive	BC4 – Oncotype DX, TransATAC post-menopausal, non-predictive	BC5 – Prosigna, TransATAC post-menopausal, non-predictive	BC6 – EPclin, TransATAC post-menopausal, non-predictive	BC7 – MammaPrint, MINDACT, non-predictive
Number of women receiving chemotherapy	−361	−594	−491	−491	−46	−39	−370
Number of infusion chair hours	−1854	−3051	−2520	−2520	−235	−203	−1900
Number of women experiencing DM during their lifetime	41	−13	−2	46	−3	−8	24
LYGs (undiscounted)	−650	214	81	−447	59	122	−217
QALYs gained (discounted)	−178	113	53	−171	28	55	−66
Additional costs to NHS/PSS (discounted)	£1,786,628	−£4,282,569	−£2,299,836	£1,862,075	£1,107,509	£305,191	£791,671
Net health benefit (£20,000 per QALY gained)	−267	327	168	−265	−28	39	−105
Net health benefit (£30,000 per QALY gained)	−237	255	130	−233	−9	45	−92

Table

Host	Database	Dates covered by databasea	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	1191
Ovid	Embase	1974–present	3184
Wiley	Cochrane Database of Systematic Reviews (Cochrane Library)	1996–present	132
Wiley	Cochrane Central Register of Controlled Trials (Cochrane Library)	1898–present	507
INAHTA	INAHTA	1989–present	77
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	1846
NIH	ClinicalTrials.gov		58
WHO	WHO International Clinical Trials Registry Platform		43
	Total		7038
	Unique records		4195

Table

#	Searches	Results
1	exp Breast Neoplasms/	339,439
2	exp mammary neoplasms/	23,367
3	exp breast/	52,644
4	exp neoplasms/	3,822,842
5	3 and 4	32,428
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	471,637
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	44,846
8	1 or 2 or 5 or 6 or 7	495,731
9	(endopredict or epclin or “ep score”).mp.	150
10	(mammaprint or 70-gene or “70 gene”).mp.	882
11	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	1967
12	(prosigna or pam50 or 50-gene or “50 gene”).mp.	805
13	or/9-12	3400
14	8 and 13	1978
15	limit 14 to yr=“2017 -Current”	1191

Table

#	Searches	Results
1	breast tumor/	94,110
2	exp breast/	127,654
3	exp neoplasm/	5,482,710
4	2 and 3	82,927
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	748,134
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	43,926
7	1 or 4 or 5 or 6	766,300
8	(endopredict or epclin or “ep score”).mp.	390
9	(mammaprint or 70-gene or “70 gene”).mp.	2013
10	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	4933
11	(prosigna or pam50 or 50-gene or “50 gene”).mp.	2103
12	or/8-11	8314
13	7 and 12	5481
14	limit 13 to yr=“2017 -Current”	3184

Table

#	Searches	Results
1	MeSH descriptor: [Breast Neoplasms] explode all trees	17,635
2	MeSH descriptor: [Neoplasms, Ductal, Lobular, and Medullary] explode all trees	865
3	MeSH descriptor: [Breast] explode all trees	1428
4	MeSH descriptor: [Neoplasms] explode all trees	110,452
5	#3 and #4	563
6	(breast* near/5 (neoplasm* or cancer* or tumor* or tumour* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary))	44,803
7	(mammar* near/5 (neoplasm* or cancer* or tumor* or tumour* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar))	354
8	#1 or #2 or #5 or #6 or #7	45,124
9	(endopredict or epclin or “ep score”)	31
10	(mammaprint or “70 gene”)	138
11	(oncotype or “recurrence score” or “21 gene”)	289
12	(prosigna or pam50 or “50 gene”).mp.	19,573
13	#9 or #10 or #11 or #12	19,949
14	#8 and #13 with Cochrane Library publication date Between Jan 2017 and Jan 2023, in Cochrane Reviews, Trials	639

Table

#	Searches	Results
1	(breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)) (Topic)	613,247
2	(mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)) (Topic)	24,383
3	#1 OR #2	626,070
4	(endopredict OR epclin OR “ep score”) (Topic)	188
5	(mammaprint OR 70-gene OR “70 gene”) (Topic)	1676
6	(oncotype OR “recurrence score” OR 21-gene OR “21 gene”) (Topic)	4301
7	(prosigna OR pam50 OR 50-gene OR “50 gene”) (Topic)	1777
8	#7 OR #6 OR #5 OR #4	7417
9	#8 AND #3	3322
10	#8 AND #3 and 2023 or 2022 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	1846

Table

#	Searches	Results
1	endopredict	7
2	epclin	1
3	“ep score”	231
4	mammaprint	21
5	oncotype	31
6	“recurrence score”	2
7	prosigna	13
8	pam50	2
9	breast*	903
10	mammar*	9
11	#10 OR #9	908
12	“70-gene”	371
13	“70 gene”	371
14	“21-gene”	371
15	“21 gene”	371
16	“50-gene”	371
17	“50 gene”	371
18	#17 OR #16 OR #15 OR #14 OR #13 OR #12	371
19	#18 AND #11	58
20	#8 OR #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1	272
21	#20 OR #19	309
22	((pam50) OR (prosigna) OR (“recurrence score”) OR (oncotype) OR (mammaprint) OR (“ep score”) OR (epclin) OR (endopredict)) OR (((“50 gene”) OR (“50-gene”) OR (“21 gene”) OR (“21-gene”) OR (“70 gene”) OR (“70-gene”)) AND ((mammar) OR (breast))) 2017 to 2023	77

Table

#	Searches	Results
1	endopredict OR epclin OR “ep score”	7
2	mammaprint OR 70-gene OR “70 gene”	19
3	oncotype OR “recurrence score” OR 21-gene OR “21 gene”	35
4	prosigna or pam50 or “50 gene”	20
5	or/1-4 (limit to 2017-present)	43

Table

#	Searches	Results
1	endopredict OR epclin OR “ep score”	7
2	mammaprint OR 70-gene OR “70 gene”	26
3	oncotype OR “recurrence score” OR 21-gene OR “21 gene”	77
4	prosigna or pam50 or “50 gene”	2
5	or/1-4 (limit to 2017-present)	58

Table

#	Searches	Results
1	endopredict	2
2	epclin	3
3	“ep score”	9
4	mammaprint	160
5	oncotype	212
6	“recurrence score”	465
7	prosigna	14
8	pam50	57
9	“70-gene”	23
11	“21-gene”	57
13	“50-gene”	12

Table

#	Searches	Results
1	endopredict	24
2	epclin	16
3	“ep score”	9
4	mammaprint	32
5	oncotype	57
6	“recurrence score”	63
7	prosigna	29
8	pam50	90
9	“70-gene”	27
11	“21-gene”	32
13	“50-gene”	10

Table

#	Searches	Results
1	endopredict	0
2	epclin	0
3	“ep score”	0
4	mammaprint	3
5	oncotype	7
6	“recurrence score”	7
7	prosigna	0
8	pam50	3
9	“70-gene”	2
11	“21-gene”	4
13	“50-gene”	2

Table

#	Searches	Results
1	endopredict	0
2	epclin	0
3	ep score	9
4	mammaprint	0
5	oncotype	0
6	“recurrence score”	1
7	prosigna	0
8	pam50	0
9	70-gene	0
11	“21-gene”	0
13	“50-gene”	0

Table

Host	Database	Dates covered	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	77
Ovid	EMBASE	1974–present	317
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	155
	Total retrieved	–	549
	Unique records		404

Table

#	Searches	Results
1	exp Breast Neoplasms/	339,611
2	exp mammary neoplasms/	23,370
3	exp breast/	52,667
4	exp neoplasms/	3,824,859
5	3 and 4	32,448
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	471,982
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	44,866
8	1 or 2 or 5 or 6 or 7	496,080
9	(endopredict or epclin or “ep score”).mp.	149
10	(mammaprint or 70-gene or “70 gene”).mp.	880
11	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	1963
12	(prosigna or pam50 or 50-gene or “50 gene”).mp.	805
13	or/9-12	3396
14	8 and 13	1977
15	limit 14 to yr=“2017 -Current”	1190
16	exp “Costs and Cost Analysis”/	264,079
17	Economics/	27,499
18	exp Economics, Hospital/	25,708
19	exp Economics, Medical/	14,388
20	Economics, Nursing/	4013
21	exp models, economic/	16,199
22	Economics, Pharmaceutical/	3101
23	exp “Fees and Charges”/	31,352
24	exp Budgets/	14,104
25	budget*.tw.	35,158
26	ec.fs.	442,581
27	cost*.ti.	142,156
28	(cost* adj2 (effective* or utilit* or benefit* or minimi*)).ab.	188,331
29	(economic* or pharmacoeconomic* or pharmaco-economic*).ti.	59,859
30	(price* or pricing*).tw.	51,979
31	(financial or finance or finances or financed).tw.	120,944
32	(fee or fees).tw.	21,211
33	(value adj2 (money or monetary)).tw.	2985
34	quality-adjusted life years/	15,581
35	(qaly or qalys).af.	14091
36	(quality adjusted life year or quality adjusted life years).af.	23,657
37	or/16-36	895,282
38	15 and 37	77

Table

#	Searches	Results
1	breast tumor/	94,109
2	exp breast/	127,715
3	exp neoplasm/	5,486,388
4	2 and 3	82,947
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	748,784
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	43,946
7	1 or 4 or 5 or 6	766,964
8	(endopredict or epclin or “ep score”).mp.	390
9	(mammaprint or 70-gene or “70 gene”).mp.	2013
10	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	4934
11	(prosigna or pam50 or 50-gene or “50 gene”).mp.	2102
12	or/8-11	8316
13	7 and 12	5481
14	limit 13 to yr=“2017 -Current”	3184
15	exp breast tumor/	642,405
16	exp breast/	127,715
17	exp neoplasm/	5,486,388
18	16 and 17	82,947
19	Socioeconomics/	159,524
20	Cost benefit analysis/	93,753
21	Cost effectiveness analysis/	179,610
22	Cost of illness/	21,158
23	Cost control/	75,866
24	Economic aspect/	123,726
25	Financial management/	120,747
26	Health care cost/	222,179
27	Health care financing/	13,847
28	Health economics/	35,524
29	Hospital cost/	25,189
30	(fiscal or financial or finance or funding).tw.	286,082
31	Cost minimization analysis/	3974
32	(cost adj estimate*).mp.	4184
33	(cost adj variable*).mp.	320
34	(unit adj cost*).mp.	5524
35	or/19-34	1,107,927
36	14 and 35	317

Table

#	Searches	Results
1	(breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)) (Topic)	614,142
2	(mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)) (Topic)	24,399
3	#1 OR #2	626,970
4	(endopredict OR epclin OR “ep score”) (Topic)	1778
5	(mammaprint OR 70-gene OR “70 gene”) (Topic)	4308
6	(oncotype OR “recurrence score” OR 21-gene OR “21 gene”) (Topic)	1682
7	(prosigna OR pam50 OR 50-gene OR “50 gene”) (Topic)	188
8	#7 OR #6 OR #5 OR #4	7430
9	#8 AND #3 and 2023 or 2022 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	1854
10	TS=(cost* and (effective* or utilit* or benefit* or minimi)) OR TS=(cost) OR TI=(economic* or pharmacoeconomic* or pharmaco-economic) OR TS=(price or pricing) OR TS=(financial or finance or finances or financed) OR TS=(fee or fees) OR TS=(value and (money or monetary)) OR TS=(economic) OR TS=(economic* and (hospital or medical or nursing or pharmaceutical)) OR TS=(“quality adjusted life year” or “quality adjusted life years”) OR TS=(qaly or qalys) OR TS=(budget*)	2,940,280
11	#9 AND #10	155

Table

Host	Database	Dates covered	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	139
Ovid	EMBASE	1974–present	391
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	139
	Total retrieved	–	669
	Unique records		404

Table

#	Searches	Results
1	exp Breast Neoplasms/	340,146
2	exp mammary neoplasms/	23,375
3	exp breast/	52,745
4	exp neoplasms/	3,830,835
5	3 and 4	32,512
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).ti.	256,329
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).ti.	15,265
8	1 or 2 or 5 or 6 or 7	402,600
9	(euroqol or euro qol or eq5d or “eq 5d” or eq-5d).tw.	16,230
10	8 and 9	203
11	limit 10 to yr=“2017 -Current”	139

Table

#	Searches	Results
1	exp breast tumor/	643,804
2	exp breast/	127,820
3	exp neoplasm/	5,497,634
4	2 and 3	83,006
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).ti.	360,376
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).ti.	16,188
7	1 or 4 or 5 or 6	687,386
8	(euroqol or euro qol or eq5d or “eq 5d” or eq-5d).tw.	29,905
9	7 and 8	597
10	limit 9 to yr=“2017 -Current”	391

Table

#	Searches	Results
1	((breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary))) (Title)	348,943
2	((mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar))) (Title)	11,684
3	#1 OR #2	359,016
4	#1 OR #2	359,016
5	(euroqol or euro qol or eq5d or eq 5d or “eq-5d”) (Topic)	19,973
6	#4 AND #5	213
7	#4 AND #5 and 2022 or 2023 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	139

TABLE 44 Risk of bias and applicability (adapted from PROBAST)

Number	Criterion	Scoring for this review
Risk of bias
Domain 1 participants	Were appropriate data sources used?	Yes (prognosis): reanalysis of RCT or cohort or nested case control AND patients did not receive chemotherapy
		Yes (predicting chemotherapy benefit): RCT or reanalysis of RCT
		No (prognostic): non-nested case control or case series AND/OR some/all patients had chemotherapy
		No (predicting chemotherapy benefit): patients not randomised to chemotherapy vs. no chemotherapy
Domain 1 participants	Were all inclusions and exclusions of participants appropriate?	Yes: all eligible patients from trial or consecutive eligible patients from prospective registry
		No: some eligible patients excluded (e.g. not sent for testing, insufficient tissue, test failures, missing data, AND/OR non-prospective registry)
		Unclear: if unclear
Domain 2 predictors (tests)	Were the tests (predictors) defined and assessed in a similar way for all participants?	Yes: if test assessed in a similar way for all participants (most/all studies in this review likely to score Yes as uses standardised test)
		No: test not assessed in a similar way for all participants
Domain 2 predictors (tests)	Were the tests (predictor assessments) made without knowledge of outcome data?	Yes: if test assessors blinded to clinical outcomes
		No: if not blinded
		Unclear: if unclear
Domain 3 outcomes	Were the outcome definitions standardised or defined a priori?	Yes: at least one outcome was standardised (e.g. DRFS, OS) or defined a priori
		No: all outcomes non-standardised and not defined a priori
		Unclear: if unclear
Domain 3 outcomes	Were the outcomes determined without knowledge of test (predictor) information?
		Yes: if outcome assessors blinded to test results
		No: if not blinded
		Unclear: if unclear
Domain 3 outcomes	Was chemotherapy decision made before test result known?	Yes: test did not influence use of chemotherapy (Yes if retrospective use of test on stored tumour samples, i.e. reanalyses of RCTs or cohorts)
		No: test result may have influenced use of chemotherapy (No for observational studies of prospective use of test)
		(This item is not in PROBAST but is important for this review)
Domain 4 analysis	Were there a reasonable number of participants with outcome data?	Yes: at least 100 patients with outcome data
		No: < 100 patients with outcome data
Domain 4 analysis	Were all enrolled participants included in \|the analysis?	Yes: if all enrolled participants included in the analysis
		No: if some enrolled patients not analysed
Applicability
Domain 1 participants	Did the included participants match the review question?	Yes: all patients in scope (HR+, HER2–, LN1–3)
		Mostly: < 20% out of scope
		No: > 20% out of scope
		Unclear: if unclear
Domain 2 predictors (tests)	Did the definition and assessment of tests (predictors) match the review question?	Yes: same as commercially available tests
		No: different from commercially available tests (e.g. FFPE vs. fresh samples, test methods)
Domain 3 outcomes	Did the outcomes match the review question?	Yes: at least one outcome matched the review question
		No: no outcomes matched the review question

TABLE 45 Risk of bias in prospective RCTs (using Cochrane RoB2)

RCT	Risk of bias due to …
Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported result	Overall risk of bias
RxPONDERKalinsky 202129	Low	Low	Low	Low	Low	Low
MINDACTPiccart 202130	Low	Low	Low	Low	Low	Low

TABLE 46 Risk of bias in prognostic studies (retrospective reanalyses of RCTs and cohorts)

Reference	Cohort	DesignDerivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match review question?	Tests match review question?	Outcomes match review question?
Albain 201032	SWOG-8814	RCT-RV	Y (RCT-R, ET only)	N (InT, TF)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Constantinidou 202265	Cyprus + Notts	Cohort-RV	Y (cohort-R, ET only)	N (InT, MD)	Y	Y	Y	Y	Y	N	N	Y	Y	Y
Drukker 201453	VdV cohort, the Netherlands	Cohort-RV (21% also in derivation set)	N (cohort-R, some CT)	Y	Y	UC	Y	Y	Y	Y	Y	N (> 20% ER− and > 20% LN4 +)	Y	Y
Filipits 201963	ABCSG-6/8	RCT-RV	Y (RCT-R, ET only)	UC	Y	UC	Y	Y	Y	Y	Y	Y	Y	Y
Gnant 2014;56 Filipits 201457	ABCSG-8, Austria	RCT-RV	Y (RCT-R, ET only)	N (InT, MS, TF, no consent)	Y	Y	Y	Y	Y	Y	N	Mostly (11% LN4 +)	Y	Y
Jackisch 2022 (abst)55	Germany, PATH	Cohort-RV	N (cohort-R, some CT)	N (reason NR)	Y	UC	Y	Y	Y	N	N	UC	Y	Y
Laenkholm 201858	DBCG, Denmark	Cohort-RV	Y (cohort-R, ET only)	N (FT, MD)	Y	Y	Y	Y	Y	Y	N	Y	Y	Y
Lundgren 202262	SBII:pre trial	RCT-RV	Y (RCT-R, ET only)	N (InT, FT, MD)	Y	N	Y	Y	Y	Y	Y	Y	Y	Y
Mamounas 201849	NSABP-28	RCT-RV	N (RCT-R, all CT)	N (InT, MS)	Y	UC	Y	Y	Y	Y	Y	UC (HER2 NR)	Y	Y
Martin 2016;59 Martin 201460	GEICAM 9906, Spain	RCT-RV	N (RCT-R, adjuvant CT)	N (MD)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Mook 200933	NKI and Italy	Cohort-RV	N (cohort-R, some CT)	N (InT, RNA quality)	Y	Y	Y	Y	Y	Y	Y	N (> 20% ER−, 16% HER2 +)	Y	Y
Penault-Llorca 201850	PACS01	RCT-RV	N (RCT-R, some CT)	N (FT, InT, MS)	Y	Y	Y	Y	Y	Y	Y	N (> 20% LN4 +)	Y	Y
Pu 202061	WHEL Study	RCT-R	N (RCT-R, some CT)	N (InT, MS, TF)	Y	UC	Y	Y	Y	Y	N	UC (NR N nodes)	Y	Y
Sestak 2018;20 201748	TransATAC	RCT-RV	Y (RCT-R, ET only)	N (InT; FT)	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Sestak 202064	Lobular subgroup (TransATAC + ABCSG-6/8)	RCT-RV	Y (RCT-R, ET only)	UC	Y	Y	Y	Y	Y	Y	N	Mostly (20% LN4 +)	Y	Y
Vliek 201754	RASTER	Cohort-RV	N (cohort-R, some CT)	N (InT, MS, no consent)	Y	UC	Y	Y	Y	Y	Y	Mostly (17% ER−, 15% HER2 +)	Y	Y

TABLE 47 Risk of bias in prognostic studies (observational studies of prospective use of test)

Reference	Cohort	DesignDerivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match review question?	Tests match review question?	Outcomes match review question?
Braun 202277	Red Cross Hospital, Munich, Germany	ObservationalV	N (prospective use of test, some CT)	Y	Y	Y	Y	UC	N	Y	N	Mostly (20% LNmic)	Y	Y
Ibraheem 202070	NCDB	ObservationalV	N (prospective use of test, CT)	N (MD, SFT)	Y	Y	Y	UC	N	Y	Y	Y	Y	Y
Massarweh 201869Petkov 201667Roberts 201768	SEER	ObservationalV	N (prospective use of test, some CT)	N (InT, MS, SFT, no consent)	Y	Y	Y	Y	N	Y	Y	UC (% LNmic NR)	Y	Y
Nitz 201776	WSG PlanB	ObservationalV	N (prospective use of test, some CT)	N (dropout, screening failure)	Y	Y	Y	UC	N	Y	Y	Y	Y	Y
Poorvu 202075	Young Women’s Breast Cancer Study	ObservationalV	N (part prospective use of test, part stored samples, some CT)	N (InT, MS, no consent)	Y	Y	Y	UC	N	Y	N	UC (% LNmic NR)	Y	Y

TABLE 48 Risk of bias in prediction studies

Reference	Cohort	Derivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match protocol?	Tests match review question?	Outcomes match review question?
Albain 201032	SWOG-8814	RCT-RV	Y (RCT-R)	N (InT, TF)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Mook 200933	NKI and Italy	Cohort-RV	N (not RCT)	N (InT, RNA qual)	Y	Y	Y	Y	Y	Y	Y	N (> 20% ER−, 16% HER2 +)	Y	Y
Abel 202281	NCDB	ObservationalV	N (not RCT)	N (MD, SFT)	Y	Y	Y	Y	N	Y	N	Y	Y	Y
Cao 2022 (abst)82
Ibraheem 201971
Iorgulescu 201983
Kumar 2023 (abst)84
Nash 202372
Weiser 202174
Weiser 202273
Petkov 2020 (abst)80	SEER	ObservationalV	N (not RCT)	N (InT, MS, SFT, no consent)	Y	Y	Y	Y	N	Y	Y	UC (% LNmic NR)	Y	Y
Rotem 2022 (abst)79Stemmer 201766	Clalit, Israel	ObservationalV	N (not RCT)	N (SFT)	Y	Y	Y	Y	N	Y	N	N (> 20% LNmic)	Y	Y

TABLE 49 Prognostic data (Oncotype DX)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Menostatus	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
Oncotype DX: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	18, 30	57	32	11	95.9	84.8	83.6	0–10 years80.65–10 years82.1	0–10 years70.95–10 years80.5	0–10 years62.05–10 years72.5	0–5 years: int vs. low: HR 3.84 (1.31 to 11.23)0–5 years: high vs. low: HR 4.45 (1.19 to 16.58)0–10 years: int vs. low: HR 1.66 (0.86 to 3.23)0–10 years: high vs. low: HR 2.35 (0.99 to 5.60)0–10 years: per 1 SD change: 1.39 (1.05 to 1.85)*Adj: LR vs. CTS (p = 0.06) and NPI (p = 0.1)	YYNNYN*
Oncotype DX: distant recurrence, variable ET/CT
Mamounas 201849NSABP-28	DRFI	n = 722All CT + ETRCT-R	LN1–3100% ER+NR HER2	All meno	18, 30	37	34	28	–	–	–	84.7	71.5	63.1	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 2.42 (NR); p < 0.001	YY*
Penault-Llorca 201850PACS01	DRFI	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	93.7	87.3	69.3	–	–	–	0–5 years: HR per 50-RS: 4.14 (2.67 to 6.43); p < 0.001*0–5 years: adj HR 3.36 (1.88 to 6.00), p < 0.001	YY*
Oncotype DX: DFS
Albain 201032SWOG-8814	DFS	n = 148ET monoRCT-R	LN + 100%LN4 +: 37%100% HR+91% HER2–	Post-meno	18, 30	37	31	32	–	–	–	60	49	43	0–5 years: HR 5.55 (2.32 to 3.28); p = 0.00020–10 years: between risk groups: p = 0.0170–10 years: HR per 50-RS: 2.64 (1.33 to 5.27); p = 0.0065–10 years: HR 0.86 (0.27 to 2.74); p = 0.80	YYYN
Mamounas 201849NSABP-28	DFS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	79.8	64.8	57	0–10 years: p < 0.0010–5 years: adj HR per 50-RS 3.81 (2.67 to 5.43); p < 0.0010–10 years: adj HR per 50-RS 2.53 (1.90 to 3.38); p < 0.001*5–10 years: adj HR per 50-RS 1.39 (0.88 to 2.19); p = 0.16	YYYN*
Penault-Llorca 201850PACS01	DFS	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	90.8	84.9	64.6	–	–	–	0–5 years: HR per 50-RS: 3.28 (2.18 to 4.94); p < 0.001*0–5 years: adj HR 2.66 (1.62 to 4.37), p < 0.001	YY*
Kalinsky 202129RxPONDER	IDFS	n = 5018CT + ET vs. ETProsp RCT	LN1–3100% HR+100% HER2−	All meno (67% post)	All ≤ 25	–	–	–	See prediction tables for outcomes per risk group	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.05 (1.04 to 1.07), p < 0.001 (adj meno and CT)	Y*
				Post-meno	All ≤ 25	–	–	–	–	–	–	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.05 (1.03 to 1.07), p < 0.001 (adj CT, nodes, grade, tumour size, age)	Y*
				Pre-meno	All ≤ 25	–	–	–	–	–	–	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.06 (1.02 to 1.09), p = 0.001 (adj CT, nodes, grade, tumour size, age)	Y*
Abdou 202351RxPONDER	IDFS	n = 4015CT + ET vs. ETProsp RCT	LN1–3100% HR+100% HER2−	Whiten = 2833	All ≤ 25	–	–	–	91.5			–	–	–	–	–
				Blackn = 248	All ≤ 25	–	–	–	87.0			–	–	–	–	–
				Asiann = 324	All ≤ 25	–	–	–	93.9			–	–	–	–	–
				Hispanicn = 610	All ≤ 25	–	–	–	91.4			–	–	–	–	–
Oncotype DX: OS and BCSS
Albain 201032SWOG-8814	OS	n = 148ET monoRCT-R	LN + 100%LN4 +: 37%100% HR+, 91% HER2–	Post-meno	18, 30	37	31	32	–	–	–	77	68	51	0–10 years: between risk groups: p = 0.0030–10 years: HR per RS-50 : 4.42 (1.96 to 9.97), p = 0.0006	YY
Penault-Llorca 201850PACS01	OS	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	99	95.6	85.6	–	–	–	0–5 years: HR per 50-RS: 5.0 (3.01 to 8.28); p < 0.001	Y
Mamounas 201849NSABP-28	OS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	93.3	79.2	70.7	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 3.09 (CI NR); p < 0.001	YY*
Mamounas 201849NSABP-28	BCSS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	98	82.9	75.6	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 3.38 (CI NR); p < 0.001	YY*`

TABLE 50 Prognostic data (MammaPrint)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Meno status	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years /other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
MammaPrint: distant recurrence, ET monotherapy
No studies
MammaPrint: distant recurrence, variable ET/CT
Piccart 202130	DMFS	n = 1176	LN1–3	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	95.7(50% CT)	–	89.0(all CT)	8 years91.0(50% CT)	–	8 years79.1(all CT)	–	–
MINDACT (not on prognostics summary table since CT use per risk group was influenced by test result)		CT + ET vs. ET	100% HR+	Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	96.3(no CT)	–	–	8 years94.0(no CT)	–	–	–	–
		Prosp-RCT	100% HER2−
	DMFI	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	96.3(50% CT)	–	89.3(all CT)	8 years92.3(50% CT)	–	8 years80.9(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	97.5(no CT)	–	–	8 years95.2(no CT)	–	–	–	–
Lopes Cardozo 202252	DMFI	N = 201 (ultra-low)	LN1–3	–	> 0.355 ultra-low	Ultra-low: 15	–	–	Ultra-low: 97.4	–	–	8 yearsUltra-low:95.2	–	–	–	–
MINDACT		Var ET/CT	99% ER+
		Prosp-RCT	97% HER2−
Drukker 201453	DMFS	n = 144	LN1–3 : 74%	Age < 53 years	0.4	38	–	62	94.5	–	64.7	10 years 78.6	–	10 years 54.325 years 44.5	0–25 years: HR 2.24 (1.25 to 4.00); p = 0.01	Y
VdV cohort, the Netherlands		Var ET/CT	LN4 +: 26%									25 years
		Cohort-R	77% ER+									NE
			NR HER2
Mook 200933NKI and Italy	DMFS	n = 241Var ET/CTCohort-R	LN1–3 : 100% inc micromets79% ER+84% HER2−	All meno	NR	41	–	59	98	–	80	91	–	76	0–10 years: HR 4.13 (1.72 to 9.96); p = 0.002*0–10 years: adj HR: 2.99 (0.996 to 8.99); p = 0.051	YN*
Vliek 201754RASTER	DRFI	N = 134Var ET/CTCohort-R	LN1–383% ER+85% HER2−	All ages	NR	48	–	52	98.4	–	86.9	94.9	–	80.7	0–10 years: low vs. high: HR 4.7 (1.3 to 16.2); p = 0.008	Y
				All agesHigh mAOL (n = 109)	NR	40	–	60	97.7	–	86.1	95.2	–	79.5	0–10 years: low vs. high: HR 4.8 (1.1 to 21.4), p = 0.022	Y
MammaPrint: DFS
Piccart 202130MINDACT	DFS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	91.6(50% CT)	–	85.9(all CT)	8 years84.5(50% CT)	–	8 years74.5(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	92.6(no CT)	–	–	8 years85.6(no CT)	–	–	–	–
MammaPrint: OS and BCSS
Piccart 202130MINDACT	OS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	98.3(50% CT)	–	95.8(all CT)	8 years95.1(50% CT)	–	8 years89.1(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	98.1(no CT)	–	–	8 years98.1(no CT)	–	–	–	–
Drukker 201453VdV cohort, the Netherlands	OS	n = 144Var ET/CTCohort-R	LN1–3 : 74%LN4 +: 26%77% ER+NR HER2	Age < 53 years	0.4	38	–	62	98.2	–	76.9	10 years 92.525 years 42.2	–	10 years 58.725 years 47.1	0–25 years: HR 1.83 (1.07 to 3.11), p = 0.03	Y
Jackisch 2022 (abst)55Germany, PATH	OS	n = 38Var ET/CTCohort-R	LN+NR	All meno (assumed)	NR	53	–	47	–	–	–	93.3	–	40.4	–	–
Mook 200933NKI and Italy	BCSS	n = 241Var ET/CTCohort-R	LN1–3 : 100% inc micromets79% ER+84% HER2−	All meno	NR	41	–	59	99	–	88	96	–	76	0–10 years: HR 5.70 (2.01 to 16.23); p = 0.001*0–10 years: adj HR 7.17 (1.81 to 28.43); p = 0.005	YY*
				All menoHigh AOL (n = 209)	NR	–	–	–	–	–	–	94	–	76	0–10 years: HR 4.12 (1.45 to 11.76); p = 0.008	Y

TABLE 51 Prognostic data (Prosigna)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Menostatus	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
Prosigna: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	NR; assume 16, 40	8	32	60	100	91.7	87.4	0–10 years1005–10 years100	0–10 years79.35–10 years87.0	0–10 years69.35–10 years75.0	0–5 years: int vs. high: HR 1.30 (0.47 to 3.60)0–10 years: int vs. high: HR 1.37 (0.69 to 2.72)HR per 1 SD change: 1.58 (1.16 to 2.15)*LR vs. CTS (p = 0.04) and NPI (p = 0.09)	NNYY, N*
Gnant 2014,56 Filipits 201457ABCSG-8, Austria	DMFS	n = 413ET monoRCT-R	LN1–3 : 89%LN4 +: 11%100% ER+ 100% HER2−	Post-meno	16, 40	4	34	62	–	–	–	0–10 years 1005–15 years100	0–10 years 93.65–15 years87.0	0–10 years 76.15–15 years75.0	5–15 years: low risk: no events5–15 years: int vs. high: HR 3.15 (1.20 to 8.24); p = 0.0200–10 years: prognostic over clinical factors (p < 0.0001)5–15 years: prognostic over clinical factors (p = 0.003)	–YYY
Laenkholm 201858DBCG, Denmark	DRFS	n = 1395ET monoCohort-R	LN1–3100% HR+ 100% HER2−	Post-meno	BespokeVaries by N nodes	26	28	46	–	–	–	96.5	88.5	77.9	0–10 years: unadj: p < 0.0010–10 years: low vs. int: adj HR 0.39 (0.20 to 0.77)0–10 years: high vs. int: adj HR 1.54 (1.04 to 2.26); p < 0.001	YYY
						40 only	–	–	–	–	–	–	95.2(low to int)	78.1	–	–
Prosigna: distant recurrence, variable ET/CT
Martin 2016,59 Martin 201460GEICAM 9906, Spain	DMFS	n = 536All CT + ET RCT-R	LN1–3 : 64%LN4 +: 36%100% ER+ 100% HER2−	All meno (46% post)	18, 65	19	56	26	–	–	–	92	74	66	0–10 years: low vs. int: HR 4.4 (NR)0–10 years: low vs. high: HR 5.8 (NR), p < 0.0001*Prosigna vs. EPclin + clinical factors (p = 0.567)	–YN*
Prosigna: DFS
Pu 202061WHEL Study	DFS	n = 344Var ET/CTRCT-R	LN+100% ER+ 100% HER2−	All meno	NR	26	53	21	–	–	–	81	64	56	0–10 years: p = 0.02	Y
Prosigna: OS and BCSS
Lundgren 202262SBII:pre trial	OS	n = 123ET/noneRCT-R	LN1–3100% ER+ 100% HER2−	Pre-meno	16, 40	2	42	57	–	–	–	–	–	–	0–10 years: int vs. high: HR 1.84 (0.91 to 3.74); p = 0.090–10 years: int vs. high: adj HR 1.32 (0.61 to 2.88); p = 0.48> 10 years: int vs. high: HR 1.02 (0.54 to 1.93); p = 0.96> 10 years: int vs. high: adj HR 1.29 (0.66 to 2.53); p = 0.46	NNNN
Lundgren 202262SBII:pre trial	BCFI	n = 123ET/noneRCT-R	LN1–3100% ER+ 100% HER2−	Pre-meno	16, 40	2	42	57	–	–	–	–	–	–	0–10 years: int vs. high: HR 1.99 (1.08 to 3.66); p = 0.030–10 years: int vs. high: adj HR 1.85 (0.95 to 3.58); p = 0.07> 10 years: int vs. high: HR 1.19 (0.50 to 2.80); p = 0.70> 10 years: int vs. high: adj HR 1.13 (0.43 to 2.95); p = 0.81	YNNN
Pu 202061WHEL Study	BCSS	n = 344Var ET/CTRCT-R	LN+100% ER+ 100% HER2−	All meno	NR	26	53	21	–	–	–	90	84	77	0–10 years: p = 0.003	Y

TABLE 52 Prognostic data (EPclin)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	MenoClin risk	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
EPclin: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	3.3	23	–	77	97.9	–	87.6	0–10 years94.45–10 years96.7	0–10 years-5–10 years-	0–10 years69.75–10 years76.4	0–5 years: high vs. low: HR 6.00 (0.80 to 44.93)0–10 years: high vs. low: HR 6.77 (1.63 to 28.07)0–10 years: per 1 SD change: 1.69 (1.29 to 2.22)*LR vs. CTS (p = 0.20) or NPI (p = 0.02)	NYYN, Y*
Filipits 201963ABCSG-6/8	DRFR	n = 453ET monoRCT-R	LN1–3100% ER+ 100% HER2−	Post-meno	3.3	35	–	65	–	–	–	0–10 years 95.60–15 years 84.75–10 years 98.25–15 years 87.0	–	0–10 years 80.90–15 years 75.15–10 years 90.55–15 years 84.0	0–10 years: HR 3.65 (1.73 to 7.68); p = 0.00030–10 years: adj HR 2.68 (1.77 to 4.08); p < 0.00015–15 years: HR 3.00 (1.03 to 8.71); p = 0.0345–15 years: adj HR 3.43 (1.74 to 6.76); p = 0.0005	YYYY
Sestak 202064Lobular (from TransATAC + ABCSG-6/8)	DRFS	n = 144ET monoRCT-R	LN1–3 : 80%LN4 +: 20%100% HR+ 100% HER2−	Post-menoLobular	3.3	26	–	74	–	–	–	93.6	–	68.8	HR 3.70 (2.49 to 5.50); p < 0.0001*EPclin vs. clinical factors (p = 0.0026)	YY*
Constantinidou 202265Cyprus + Notts	DRFS	n = 62ET monoCohort-R	LN1–3100% ER+ 100% HER2−	Pre-meno	3.3	19	–	81	–	–	–	100	–	75	High vs. low: p = 0.066*Adj HR (cont score): 2.91 (1.70 to 4.97), p < 0.001	NY*
EPclin: distant recurrence, variable ET/CT
Martin 2016,59 Martin 201460GEICAM 9906, Spain	DMFS	n = 555All CT + ETRCT-R	LN1–3 : 64%LN4 +: 36%100% ER+ 100% HER2−	All meno (46% post)	3.3	13	–	87	–	–	–	100	–	72	Low vs. high: HR not estimable, p < 0.0001*EPclin vs. clinical factors (p = 0.0018)	YY*
				Pre-meno	3.3	12	–	88	–	–	–	100	–	70	Low vs. high: HR NR, p = 0.0006	Y
				Post-meno	3.3	13	–	87	–	–	–	100	–	76	Low vs. high: HR NR, p = 0.0023	Y

TABLE 53 Observational data for Oncotype DX (all outcomes and analyses)

Cohort	Reference	Nodal statusHR, HER2	Outcome	NET/CT	MenoAgeClin	Test cut-offs	Distribution %	% risk of outcome	HR between test risk groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High
Oncotype DX: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	DRFI(0–5 years)	n = 709Var ET/CT	All meno	18, 30	53	36	10	96.8 (7% CT)	93.7 (40% CT)	83.1 (86% CT)	0–5 years: low vs. high: HR 0.19 (0.09 to 0.40)0–5 years: int vs. high: HR 0.39 (0.20 to 0.79); p < 0.0010–5 years: adj HR: low vs. high: HR 0.23 (0.11 to 0.50)0–5 years: adj HR: int vs. high: HR 0.42 (0.20 to 0.86); p = 0.001	YYYY
						11, 25	≤ 25:81	19	95.7(5% CT)	96.0(18% CT)	86.9(77% CT)	0–5 years: p < 0.001	Y
						≤ 25, 26–30				96.0 (15% CT)	91.5 (67% CT)	–	–
						18–25					94.4 (31% CT)		–	–
				n = 109Var ET/CT	Age < 50	18, 30	48	37	16	96.2 (12% CT)	100.0 (48% CT)	64.2 (100% CT)	0–5 years: p < 0.001	Y
				n = 464Var ET/CT	Age 50–69	18, 30	54	37	9	97.6 (6% CT)	93.5 (42% CT)	87.8 (90% CT)	0–5 years: p = 0.017	Y
				n = 136Var ET/CT	Age ≥ 70	18, 30	57	33	10	94.7 (7% CT)	88.7 (22% CT)	92.9 (57% CT)	0–5 years: p = 0.458	N
Young Women’s Breast Cancer Study	Poorvu 202075	LNmic, LN1–3100% ER+100% HER2−	DRFS(0–6 years)	n = 163Var ET/CT	Age ≤ 40	18, 30	33	42	25	0–6 years:85.9 (83% CT)	0–6 years:87.3 (97% CT)	0–6 years:62.8 (98% CT)	0–6 years: p = 0.004	Y
						11, 25	9	54	37	0–6 years:92.3 (79% CT)	0–6 years:85.2 (92% CT)	0–6 years:71.3 (97% CT)	0–6 years: p = 0.10	N
Oncotype DX: DFS
WSG PlanB	Nitz 201776	LN1–3100% HR+100% HER2−	DFS(0–5 years)	n = 110Var ET/CT	All meno	0–10				94.4(No CT)	–	–	–	–
Red Cross Hospital, Munich, Germany	Braun 202277	LNmic: 20%LN1–3 : 80%100% HR+100% HER2−	DFS(0–5 years)	n = 217Var ET/CT	All meno (63% post)	≤ 25, 26 +	86	14	RS 0–25 : 90.3 (19% CT)	71.0 (93% CT)	–	–
Oncotype DX: OS and BCSS
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	BCSS(0–5 years)	n = 709Var ET/CT	All meno	18, 30	53	36	10	99.5 (7% CT)	96.6 (40% CT)	94.3 (86% CT)	0–5 years: p < 0.001	Y
						11, 25	RS≤ 25:81		19	99.1 (5% CT)	98.8 (18% CT)	93.5 (77% CT)	0–5 years: p < 0.001	Y
						≤ 25, 26–30					98.9 (15% CT)	92.6 (67% CT)	–	–
						18–25					97.8 (31% CT)		–	–
SEER registry	Petkov 201667	LN1mic, LN1–3100% HR+100% HER2−	BCSS(< 5 years)	n = 4691Var ET/CT	All	18, 30	57	36	7	99.0 (23% CT)	97.7 (47% CT)	85.7 (75% CT)	< 5 years: high vs. low: HR 11.0 (7.8 to 15.5)< 5 years: int vs. low: HR 3.1 (2.3 to 4.3), p < 0.001< 5 years: adj HR: high vs. low: HR 7.8 (5.3 to 11.6)< 5 years: adj HR: int vs. low: HR 3.0 (2.1 to 4.2); p < 0.001	YYYY
				n = 328Var ET/CT	Black ethnicity	18, 30	54	36	9	99.4 (CT NR)	98.9	91.3	< 5 years: p = 0.4117	N
				n = 4021Var ET/CT	White ethnicity	18, 30	58	36	7	99 (CT NR)	97.6	84.1	< 5 years: p < 0.0001	Y
				n = 320Var ET/CT	Other ethnicity	18, 30	57	34	8	98.5 (CT NR)	99.1	100	< 5 years: p = 0.8427	N
SEER registry	Roberts 201768	LN1mic, LN1–3100% HR+100% HER2−	BCSS(0–5 years)	n = 6483Var ET/CT	All	18, 30	58	35	7	98.8 (CT NR)	97.3	88.5	0–5 years: p < 0.001*0–5 years: adj: p < 0.001	YY*
			OS(0–5 years)	n = 6483Var ET/CT	All	18, 30	58	35	7	92.1 (CT NR)	90.9	81.7	0–5 years: p < 0.001*0–5 years: adj: p < 0.001	YY*
SEER registry	Massarweh 201869	LN1mic, LN1–3100% HR+100% HER2−	BCSS(0–5 years)	n = 6437Var ET/CT	Women	18, 30	59	35	7	98.8 (23% CT)	97.3 (48% CT)	89.2 (77% CT)	0–5 years: p < 0.001	Y
				n = 46Var ET/CT	Men	18, 30	52	26	22	100 (33% CT)	100 (50% CT)	N/A (60% CT)	0–5 years: p = 0.02	Y
			OS(0–5 years)	n = 6437Var ET/CT	Women	18, 30	59	35	7	92.2 (23% CT)	90.8 (48% CT)	83.2 (77% CT)	0–5 years: p < 0.001	Y
				n = 46Var ET/CT	Men	18, 30	52	26	22	78.9 (33% CT)	100 (50% CT)	N/A (60% CT)	0–5 years: p = 0.002	Y
NCDB	Ibraheem 202070	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 25,029Var ET/CT	All meno	11, 25	24	64	13	–	–	–	0–5 years: int vs. low: HR 1.15 (0.97 to 1.36)High vs. low: HR 2.94 (2.43 to 3.56)Per 10-RS: HR 1.38 (1.31 to 1.44)	NYY
NCDB	Ibraheem 201971	LN1–3 : 97%LN4-9 : 3%	OS(0–5 years)	n = 13,163Var ET/CT	All meno	11, 25	–	–	–	–	–	–	0–5 years:RS 18–25 vs. 11–17: HR 1.20 (1.07 to 1.35); p < 0.001RS 18–25 vs. 11–17: adj HR 1.15 (1.03 to 1.29); p < 0.001RS 26–30 vs. 11–17: HR 1.91 (1.65 to 2.22); p < 0.001RS 26–30 vs. 11–17: adj HR 1.62 (1.38 to 1.89); p < 0.001	YYYY
NCDB	Nash 202372	LN1–3100% HR+100% HER2−	OS(NR, med FU 5.5 years)	n = 4124Var ET/CT	Age 40–50	11, 25	–	–	–	–	–	–	RS 26–30 vs. 0–25: adj HR 2.29 (1.49 to 4.86)RS 31–50 vs. 0–25: adj HR 3.70 (2.03 to 6.75)*RS 51–100 vs. 0–25: adj HR 2.31 (0.78–6.86); p < 0.001	YYN*
NCDB	Weiser 202273	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 2691Var ET/CT	Lobular	11, 25	–	–	–	95.5	95.5	83.8	0–5 years: p = 0.0004*Adj: sig	YY*
NCDB	Weiser 202174	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 28,591Var ET/CT	All	≤ 25	–	–	–	–	–	–	*0–5 years: RS 18–25 vs. RS 12–17: adj HR 1.30 (1.00 to 1.68)	Y*

TABLE 54 Chemotherapy effect within risk groups: Registry data for Oncotype DX (all outcomes)

Cohort	Reference	Nodal statusHR, HER2	Outcome	N	MenoAgeClin	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Inter-action	Preda*Adj
Low	Int	High	Low	Int	High	Low	Int	High
CT	No	CT	No	CT	No
Oncotype DX: observational: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	DRFI0–5 years	n = 709	All meno	18, 30	92.3	97.1	99	90.3	82	90	−4.8	8.7	−8.0	p = 0.245	p = 0.019	–	–	–
						11, 25	83.3	96.3	98.8	95.4	97.5	79.7	−13.0	3.4	17.8	–	–	p = 0.017	–	–
						All ≤ 25	–	–	97.7	95.6	–	–	2.1	–	p = 0.521	–	–	–
						18–25	–	–	100	91.8	–	–	–	8.2	–	–	p = 0.058	–	–	–
	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	DRFS0–7 years	n = 140	All meno	All26–30	–	–	–	–	89.4	78.0	–	–	11.4	–	–	Not sig	–	–
Oncotype DX: observational: BCSS and OS
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	BCSS0–5 years	n = 709	All meno	18, 30	100.0	99.4	98.9	95.1	93.4	100	0.6	3.8	−6.6	–	–	–	–	–
						11, 25	100.0	99.1	100.0	98.6	97.1	84.0	0.9	1.4	13.1	–	–	–	–	–
						All ≤ 25	–	–	100.0	98.7	–	–	1.3	–	–	–	–	–	–
						18–25	–	–	100.0	96.8	–	–	–	3.2	–	–	–	–	–	–
	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	BCSS0–7 years	n = 140	All meno	26–30	–	–	–	–	98.7	93.8	–	–	4.9	–	–	p = 0.024	–	–
SEER	Petkov 2020 (abst)80	LN1mic-LN3100% HR+100% HER2−	BCSS0–5 years	n = 2588	Age≤ 50	0–10	100	100	–	–	–	–	0	–	–	–	–	–	–	–
						11–15	–	–	97.7	99.5	–	–	–	−1.8	–	–	–	–	–	–
						16–20	–	–	98.4	98.7	–	–	–	−0.3	–	–	–	–	–	–
						21–25	–	–	98.8	98.4	–	–	–	0.4	–	–	–	–	–	–
						26–100	–	–	–	–	93.9	95.6	–	–	−1.7	–	–	–
Clalit, Israel	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	OS0–7 years	n = 140	All meno	26–30	–	–	–	–	96.3	93.8	–	–	2.5	–	–	Not sig	–	–
NCDB	Abel 202281	LN1–3100% HR+100% HER2−	OS0–5 years	n = 21,370	Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–	p = 0.278	–	–	–
				n = 6356	Lobular	All ≤ 25	–	–	–	–	–	–	–	–	–	p = 0.532	–	–	–
				n = 4251	Age < 50 Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: 0.44 (0.22 to 0.86), p = 0.016	–	–	–
				n = 1062	Age < 50 Lobular	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: 0.54 (0.14 to 2.18), p = 0.39	–	–	–
NCDB (cont)	Cao 2022 (abst)82	LN1–3100% ER+100% HER2−	OSNR	n = 28,427	Age ≤ 50	All20–25	–	–	–	–	–	–	–	–	–	–	Unadj: 0.334 (NR), p = 0.002	–	–	–
					Age > 50	All20–25	–	–	–	–	–	–	–	–	–	–	Unadj: 0.521 (NR), p = 0.019	–	–	–
NCDB (cont)	Ibraheem 201971	LN1–3 : 97%LN4–9 : 3%100% HR+100% HER2−	OS0–5 years	n = 13,163	All meno	11–17	–	–	97.7	96.5	–	–	–	1.2	-	–	Adj: 0.63 (0.40 to 0.99), p = 0.044Threshold: RS > 13 sig CT benefit	–	–	–
						18–25	–	–	96.0	92.7	–	–	–	3.3	–	–	Adj: 0.53 (0.37 to 0.76), p = 0.001	–	–	–
						26–30					92.2	85.5			6.7			Adj: 0.50 (0.28 to 0.89), p = 0.018
				n = 3101	Age ≤ 50	All11–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.68 (0.35 to 1.32), p = 0.25	–	–	–
				n = 8886	Age > 50	All11–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.64 (0.47 to 0.86), p = 0.004	–	–	–
NCDB (cont)	Iorgulescu 201983	LN1–3100% ER+100% HER2−	OS0–5 years	n = 2735	All meno	18, 30	93	92	93.2	85.7	92.4	66.9	1.0	7.5	25.5	Unadj: p = 0.27Adj: 0.81 (0.33 to 1.98), p = 0.64	Unadj: p = 0.02Adj: 0.67 (0.35 to 1.27), p = 0.22	Unadj: p < 0.001Adj: 0.24 (0.13 to 0.47), p < 0.001	–	–
NCDB (cont)	Kumar 2023 (abst)84	LN1–3: > 90%LN4 +: < 10%100% HR+100% HER2−	OS0–10 years	n = 8628	Age ≤ 50	0–11	–	–	–	–	–	–	–	–	–	Adj: 0.56 (0.22 to 1.42)	–	–	–	–
						12–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.55 (0.38 to 0.80)	–
						All ≤ 25	–	–	93.0	91.0	–	–	2.0	–	Unadj: 0.60 (0.48 to 0.75), p < 0.0001Adj: 0.54 (0.39 to 0.76), p = 0.0004	–	–	–
				n = 8628	Age 18–40	All ≤ 25	–	–	86.0	82.8	–	–	3.2	–		Adj: 0.43 (0.22 to 0.85)		–	–	–
				n = 8628	Age 40–50	All ≤ 25	–	–	94.7	92.2	–	–	2.5	–		Adj: 0.59 (0.39 to 0.87)		–	–	–
NCDB (cont)	Nash 202372	LN1–3100% HR+100% HER2−	OSNR, med FU 5.5 years	N = 4124	Age 40–50	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: p = 0.41Adj: 0.72 (0.47 to 1.12), p = 0.15	–	–	–
						25–30	–	–	–	–	–	–	–	–	–	–		Unadj: p = 0.28
						31–50	–	–	–	–	–	–	–	–	––	–		Unadj: p = 0.002
																		Adj: 0.29 (0.10 to 0.85), p = 0.02
						> 50	–	–	–	–	–	–	–	–	–	–		Not sig (few events)
NCDB (cont)	Weiser 202273	LN1–3100% HR+100% HER2−	OS0–5 years	n = 16,646	All menoDuctal	11–25	–	–	96.7	95.1	–	–	–	1.6	–	–	Unadj: p = 0.004Adj: non-sig	–	–	–
				NR	Age < 50Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 2.32 (1.19 to 4.49)	–	–	–
				NR	Age 50–75Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 1.12 (0.86 to 1.46)	–	–	–
				n = 2691	All menoLobular	0–10	94.7	95.7	–	–	–	–	−1.0	–	–	Unadj: p = 0.888Adj: non–sig	–	–	–	–
						–11–25	–	–	96.6	94.9	–	–	–	1.7	–	–	Unadj: p = 0.381Adj: non-sig	–
NCDB (cont)	Weiser 202174	LN1–3100% HR+100% HER2−	OS0–5 years	n = 28,591	All meno	All ≤ 25	–	–	96.6	93.2	–	–	3.4	–	Unadj: p < 0.001Adj: 1.63 (1.28 to 2.07)	–	–	–
				NR	Age ≤ 50	All ≤ 25	–	–	–	–	–	–	1.4	–			Adj: 1.88 (1.05 to 3.37), p = 0.032	–	–	–
						12–17	–	–	–	–	–	–	–	1.3	–	–	Adj: 2.49 (0.80 to 7.76)	–	–	–
						18–25	–	–	–	–	–	–	–	4.4	–	–	Adj: 3.30 (1.38 to 7.84)	–	–	–
				NR	Age 51–70	All ≤ 25	–	–	–	–	–	–	1.6	–			Adj: 1.49 (1.12 to 1.97), p = 0.006	–	–	–
						12–17	–	–	–	–	–	–	–	3.6	–	–	Adj: 2.80 (1.45 to 5.24)	–	–	–
						18–25	–	–	–	–	–	–	–	3.2	–	–	Adj: 1.37 (0.92–2.05)	–	–	–
				NR	Age > 70	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 1.1 (0.68 to 1.78), p = 0.69	–	–	–
				NR	Age ≤ 70	0–10	–	–	–	–	–	–	–	–	–	p = 0.44	–	–
						12–25	–	–	–	–	–	–	–	3.0	–	–	Adj: 1.91 (1.42 to 2.57)	–	–	–
						12–17	–	–	–	–	–	–	–	3.4	–	–	Adj: 3.04 (1.78 to 5.21), p < 0.001	–	–	–
						18–25	–	–	–	–	–	–	–	3.8	–	–	Adj: 2.02 (1.42 to 2.87), p < 0.001	–	–	–

TABLE 55 Infusion time for each chemotherapy regimen included in the EAG model

Regimen	Doses per course	Infusion time per dose (hours)	Infusion time per course
FEC75 (6 cycles)
Fluorouracil 600 mg/m²	6	0.08	0.50
Epirubicin 75 mg/m²	6	0.08	0.50
Cyclophosphamide 600 mg/m²	6	0.50	3.00
FEC100 +T (3 + 3 cycles)
Fluorouracil 500 mg/m²	3	0.08	0.25
Epirubicin 100 mg/m²	3	0.08	0.25
Cyclophosphamide 500 mg/m²	3	0.50	1.50
Docetaxel 100 mg/m²	3	1.00	3.00
TC (4 cycles)
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Docetaxel 75 mg/m²	4	1.00	4.00
EC90/T75 (4 + 4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Docetaxel 75 mg/m²	4	0.50	2.00
EC90 (4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Accelerated EC90/P (4 + 4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Paclitaxel 175 mg/m²	4	1.00	4.00
C-D (6 cycles)
Carboplatin AUC 6 (assumed 600 mg)	6	1.00	6.00
Docetaxel 75 mg/m²	6	1.00	6.00
TAC (6 cycles)
Docetaxel 75 mg/m²	6	1.00	6.00
Doxorubicin 50 mg/m²	6	0.08	0.50
Cyclophosphamide 500 mg/m²	6	0.08	0.50
Weekly P (12 weeks)
Paclitaxel 80 mg/m²	12	1.00	12.00
EC90/weekly P (4 cycles, 12 weeks)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Paclitaxel 80 mg/m²	12	1.00	12.00

Figure captions

Figure Five-year net survival by time since diagnosis, adults with breast cancer diagnosed in 2016–20, followed up to 2021. CDSR, Cochrane Database of Systematic Reviews; CENTRAL, Cochrane Central Register of Controlled Trials; INAHTA, International Network of Agencies for Health Technology Assessment. Source: NHS Digital, National Disease Registration Service.
Figure Five-year net survival by stage, adults with breast cancer diagnosed in 2016–20, followed up to 2021. Source: NHS Digital, National Disease Registration Service.
Figure Breast cancer incidence by age and sex, 2016–8, UK. Source: Cancer Research UK.
Figure Diagnosis and management pathway for early breast cancer. FISH, fluorescence in situ hybridisation; IHC, immunohistochemistry. Notes: Ki67 is tested for following biopsy in some centres; however, the methodology for this is not standardised. Extended endocrine therapy may be given for 10 years. If abemaciclib is given, this would normally be started after completion of a course of adjuvant chemotherapy. Ovarian suppression should be considered only in pre-menopausal women. Bisphosphonates are recommended for use only in post-menopausal women. Adjuvant and neoadjuvant treatment for HER2+ early breast cancer may include pertuzumab and trastuzumab or trastuzumab alone. In the neoadjuvant setting in poor responders, TDM1 may be given after surgery instead of continuing with the neoadjuvant regimen. This population is out of scope. Women with triple negative breast cancer who have neoadjuvant chemotherapy and respond poorly may now be offered postoperative capecitabine chemotherapy. Immunotherapy may be used in the neoadjuvant setting for women with triple receptor-negative breast cancer. This population is out of scope. Women with known BRCA1 or 2 gene mutations are now eligible for adjuvant PARP inhibitors. Radiotherapy may also be offered depending on the type of surgery done and the patient’s risk of recurrence. ER, oestrogen receptor; FISH, fluorescence in situ hybridisation; HER2, human epidermal growth factor receptor 2; IHC, immunohistochemistry; PR, progesterone receptor.
Figure External Assessment Group’s model – decision tree component. Notes: For Oncotype DX under the new cut-offs, EPclin, MammaPrint, four branches are used due to the absence of an intermediate-risk category for these tests. All patients are also assumed to receive ET.
Figure External Assessment Group’s model – long-term Markov model component. Note: In line with the model developed to inform DG34, once-only costs and QALY losses associated with LR are modelled for women who develop DM (see Evidence sources used to inform the model parameters)
Figure Cost-effectiveness acceptability curves, BC1 – Oncotype DX, RxPONDER pre-menopausal (predictive benefit).
Figure Cost-effectiveness acceptability curves, BC2 – Oncotype DX, RxPONDER post-menopausal (predictive benefit).
Figure Cost-effectiveness acceptability curves, BC3 – Oncotype DX, TransATAC, post-menopausal (predictive benefit).
Figure Cost-effectiveness acceptability curves, BC4 – Oncotype DX, TransATAC, post-menopausal (non-predictive benefit).
Figure Cost-effectiveness acceptability curves, BC5 – Prosigna, TransATAC, post-menopausal (non-predictive benefit).
Figure Cost-effectiveness acceptability curves, BC6 – EPclin, TransATAC, post-menopausal (non-predictive benefit).
Figure Cost-effectiveness acceptability curves, BC7 – MammaPrint, MINDACT, LN+ subgroup (non-predictive benefit).

Appendices

Extracted 2026-09-08 from the same publisher XML. Locators follow the same style. Appendix tables are listed under their appendix; where a table’s label already appeared in the Tables section above, it is repeated here in context.

Clinical effectiveness searches

Ovid MEDLINE(R) and Epub Ahead of Print, In-Process, In-Data-Review & Other Non-Indexed Citations and Daily 1946–25 April 2023

_¶1

Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)

_¶2

WHO International Clinical Trials Registry Platform

_¶3

American Society of Clinical Oncology (ASCO) www.asco.org/

_¶4

European Society for Medical Oncology (ESMO) www.esmo.org/

_¶5

American Association for Cancer Research (AACR) www.aacr.org/

_¶6

European Cancer Organisation (ECO) www.europeancancer.org/

_¶7

Myriad genetics https://myriad.com/publications/

_¶8

Exact Sciences (aka Genomic Health) www.exactsciences.com/

_¶9

Cost-effectiveness searches

Ovid MEDLINE(R) and Epub Ahead of Print, In-Process, In-Data-Review & Other Non-Indexed Citations and Daily 1946–3 May 2023

_¶1

Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)

_¶2

EuroQol-5 Dimensions searches

Ovid MEDLINE(R) and Epub Ahead of Print, In-Process, In-Data-Review & Other Non-Indexed Citations and Daily 1946–3 May 2023

_¶1

_¶2

The following factors either were judged low risk or were unlikely to have affected results. For Domain 2 (predictors, i.e. the tests themselves), all studies used the same version of the test for all participants (as the tests are standardised). Some studies blinded test assessors to patient outcomes, while for other studies this was unclear; however, since the tests are based on objective measures of gene expression, this is unlikely to have affected interpretation of test results. For Domain 3 (outcomes), all studies used standardised outcomes relating to recurrence or survival. It was assumed that blinding of outcome assessors to test results applied within studies of retrospective use of the test, while in studies of prospective use, blinding to test results was generally unclear; however, as most outcomes were standardised cancer outcomes, this is unlikely to have affected outcome reporting. For Domain 4 (analysis), most studies included a reasonable number of participants (over 100). In terms of applicability to the review question, the test and outcomes matched the review question in all studies.

_¶3

Results: risk of bias in prediction studies

The risk of bias in prediction studies, assessed using the PROBAST tool,28 is presented in Table 48.

_¶1

The following factors may have affected results to some extent. For Domain 1 (participants), only the SWOG-8814 study32 was a reanalysis of a RCT in which chemotherapy use was randomised; in the remaining studies, chemotherapy use was not randomised. This limitation is reflected in the section on prediction of chemotherapy benefit (see Results: prediction of chemotherapy benefit). In some studies, some participants did not match the review question (either not ER+, not HER2− or not LN1–3). Most studies excluded a proportion of patients for various reasons including insufficient tissue, missing data, failed tests and others, which may have influenced results to some extent, though the impact is difficult to assess. For Domain 3 (outcomes), chemotherapy decisions were not influenced by the test result in the two studies of retrospective use of the test, whereas in the three observational registries in which the test was used prospectively, chemotherapy decisions may have been influenced by the test result; therefore, observational studies are reported separately in the section on prediction of chemotherapy benefit (see Results: prediction of chemotherapy benefit).

_¶2

_¶3

PRISMA 2020 flow diagram for economic evaluations of tumour profiling tests

_¶1

For more information, visit: www.prisma-statement.org/

_¶2

PRISMA 2020 flow diagram for HRQoL associated with different health states for women with breast cancer

_¶3

For more information, visit: www.prisma-statement.org/

_¶4

Tables in Appendix 1: Literature search strategies

Table

Host	Database	Dates covered by databasea	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	1191
Ovid	Embase	1974–present	3184
Wiley	Cochrane Database of Systematic Reviews (Cochrane Library)	1996–present	132
Wiley	Cochrane Central Register of Controlled Trials (Cochrane Library)	1898–present	507
INAHTA	INAHTA	1989–present	77
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	1846
NIH	ClinicalTrials.gov		58
WHO	WHO International Clinical Trials Registry Platform		43
	Total		7038
	Unique records		4195

_{aIndicates dates covered by database as a whole. Searches were then restricted to years 2017 onwards, as indicated in the search strategies below.}

Table

#	Searches	Results
1	exp Breast Neoplasms/	339,439
2	exp mammary neoplasms/	23,367
3	exp breast/	52,644
4	exp neoplasms/	3,822,842
5	3 and 4	32,428
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	471,637
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	44,846
8	1 or 2 or 5 or 6 or 7	495,731
9	(endopredict or epclin or “ep score”).mp.	150
10	(mammaprint or 70-gene or “70 gene”).mp.	882
11	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	1967
12	(prosigna or pam50 or 50-gene or “50 gene”).mp.	805
13	or/9-12	3400
14	8 and 13	1978
15	limit 14 to yr=“2017 -Current”	1191

Table

#	Searches	Results
1	breast tumor/	94,110
2	exp breast/	127,654
3	exp neoplasm/	5,482,710
4	2 and 3	82,927
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	748,134
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	43,926
7	1 or 4 or 5 or 6	766,300
8	(endopredict or epclin or “ep score”).mp.	390
9	(mammaprint or 70-gene or “70 gene”).mp.	2013
10	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	4933
11	(prosigna or pam50 or 50-gene or “50 gene”).mp.	2103
12	or/8-11	8314
13	7 and 12	5481
14	limit 13 to yr=“2017 -Current”	3184

Table

#	Searches	Results
1	MeSH descriptor: [Breast Neoplasms] explode all trees	17,635
2	MeSH descriptor: [Neoplasms, Ductal, Lobular, and Medullary] explode all trees	865
3	MeSH descriptor: [Breast] explode all trees	1428
4	MeSH descriptor: [Neoplasms] explode all trees	110,452
5	#3 and #4	563
6	(breast* near/5 (neoplasm* or cancer* or tumor* or tumour* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary))	44,803
7	(mammar* near/5 (neoplasm* or cancer* or tumor* or tumour* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar))	354
8	#1 or #2 or #5 or #6 or #7	45,124
9	(endopredict or epclin or “ep score”)	31
10	(mammaprint or “70 gene”)	138
11	(oncotype or “recurrence score” or “21 gene”)	289
12	(prosigna or pam50 or “50 gene”).mp.	19,573
13	#9 or #10 or #11 or #12	19,949
14	#8 and #13 with Cochrane Library publication date Between Jan 2017 and Jan 2023, in Cochrane Reviews, Trials	639

Table

#	Searches	Results
1	(breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)) (Topic)	613,247
2	(mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)) (Topic)	24,383
3	#1 OR #2	626,070
4	(endopredict OR epclin OR “ep score”) (Topic)	188
5	(mammaprint OR 70-gene OR “70 gene”) (Topic)	1676
6	(oncotype OR “recurrence score” OR 21-gene OR “21 gene”) (Topic)	4301
7	(prosigna OR pam50 OR 50-gene OR “50 gene”) (Topic)	1777
8	#7 OR #6 OR #5 OR #4	7417
9	#8 AND #3	3322
10	#8 AND #3 and 2023 or 2022 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	1846

Table

#	Searches	Results
1	endopredict	7
2	epclin	1
3	“ep score”	231
4	mammaprint	21
5	oncotype	31
6	“recurrence score”	2
7	prosigna	13
8	pam50	2
9	breast*	903
10	mammar*	9
11	#10 OR #9	908
12	“70-gene”	371
13	“70 gene”	371
14	“21-gene”	371
15	“21 gene”	371
16	“50-gene”	371
17	“50 gene”	371
18	#17 OR #16 OR #15 OR #14 OR #13 OR #12	371
19	#18 AND #11	58
20	#8 OR #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1	272
21	#20 OR #19	309
22	((pam50) OR (prosigna) OR (“recurrence score”) OR (oncotype) OR (mammaprint) OR (“ep score”) OR (epclin) OR (endopredict)) OR (((“50 gene”) OR (“50-gene”) OR (“21 gene”) OR (“21-gene”) OR (“70 gene”) OR (“70-gene”)) AND ((mammar) OR (breast))) 2017 to 2023	77

Table

#	Searches	Results
1	endopredict OR epclin OR “ep score”	7
2	mammaprint OR 70-gene OR “70 gene”	19
3	oncotype OR “recurrence score” OR 21-gene OR “21 gene”	35
4	prosigna or pam50 or “50 gene”	20
5	or/1-4 (limit to 2017-present)	43

Table

#	Searches	Results
1	endopredict OR epclin OR “ep score”	7
2	mammaprint OR 70-gene OR “70 gene”	26
3	oncotype OR “recurrence score” OR 21-gene OR “21 gene”	77
4	prosigna or pam50 or “50 gene”	2
5	or/1-4 (limit to 2017-present)	58

Table

#	Searches	Results
1	endopredict	2
2	epclin	3
3	“ep score”	9
4	mammaprint	160
5	oncotype	212
6	“recurrence score”	465
7	prosigna	14
8	pam50	57
9	“70-gene”	23
11	“21-gene”	57
13	“50-gene”	12

Table

#	Searches	Results
1	endopredict	24
2	epclin	16
3	“ep score”	9
4	mammaprint	32
5	oncotype	57
6	“recurrence score”	63
7	prosigna	29
8	pam50	90
9	“70-gene”	27
11	“21-gene”	32
13	“50-gene”	10

Table

#	Searches	Results
1	endopredict	0
2	epclin	0
3	“ep score”	0
4	mammaprint	3
5	oncotype	7
6	“recurrence score”	7
7	prosigna	0
8	pam50	3
9	“70-gene”	2
11	“21-gene”	4
13	“50-gene”	2

Table

#	Searches	Results
1	endopredict	0
2	epclin	0
3	ep score	9
4	mammaprint	0
5	oncotype	0
6	“recurrence score”	1
7	prosigna	0
8	pam50	0
9	70-gene	0
11	“21-gene”	0
13	“50-gene”	0

Table

Host	Database	Dates covered	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	77
Ovid	EMBASE	1974–present	317
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	155
	Total retrieved	–	549
	Unique records		404

Table

#	Searches	Results
1	exp Breast Neoplasms/	339,611
2	exp mammary neoplasms/	23,370
3	exp breast/	52,667
4	exp neoplasms/	3,824,859
5	3 and 4	32,448
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	471,982
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	44,866
8	1 or 2 or 5 or 6 or 7	496,080
9	(endopredict or epclin or “ep score”).mp.	149
10	(mammaprint or 70-gene or “70 gene”).mp.	880
11	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	1963
12	(prosigna or pam50 or 50-gene or “50 gene”).mp.	805
13	or/9-12	3396
14	8 and 13	1977
15	limit 14 to yr=“2017 -Current”	1190
16	exp “Costs and Cost Analysis”/	264,079
17	Economics/	27,499
18	exp Economics, Hospital/	25,708
19	exp Economics, Medical/	14,388
20	Economics, Nursing/	4013
21	exp models, economic/	16,199
22	Economics, Pharmaceutical/	3101
23	exp “Fees and Charges”/	31,352
24	exp Budgets/	14,104
25	budget*.tw.	35,158
26	ec.fs.	442,581
27	cost*.ti.	142,156
28	(cost* adj2 (effective* or utilit* or benefit* or minimi*)).ab.	188,331
29	(economic* or pharmacoeconomic* or pharmaco-economic*).ti.	59,859
30	(price* or pricing*).tw.	51,979
31	(financial or finance or finances or financed).tw.	120,944
32	(fee or fees).tw.	21,211
33	(value adj2 (money or monetary)).tw.	2985
34	quality-adjusted life years/	15,581
35	(qaly or qalys).af.	14091
36	(quality adjusted life year or quality adjusted life years).af.	23,657
37	or/16-36	895,282
38	15 and 37	77

Table

#	Searches	Results
1	breast tumor/	94,109
2	exp breast/	127,715
3	exp neoplasm/	5,486,388
4	2 and 3	82,947
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).mp.	748,784
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).mp.	43,946
7	1 or 4 or 5 or 6	766,964
8	(endopredict or epclin or “ep score”).mp.	390
9	(mammaprint or 70-gene or “70 gene”).mp.	2013
10	(oncotype or “recurrence score” or 21-gene or “21 gene”).mp.	4934
11	(prosigna or pam50 or 50-gene or “50 gene”).mp.	2102
12	or/8-11	8316
13	7 and 12	5481
14	limit 13 to yr=“2017 -Current”	3184
15	exp breast tumor/	642,405
16	exp breast/	127,715
17	exp neoplasm/	5,486,388
18	16 and 17	82,947
19	Socioeconomics/	159,524
20	Cost benefit analysis/	93,753
21	Cost effectiveness analysis/	179,610
22	Cost of illness/	21,158
23	Cost control/	75,866
24	Economic aspect/	123,726
25	Financial management/	120,747
26	Health care cost/	222,179
27	Health care financing/	13,847
28	Health economics/	35,524
29	Hospital cost/	25,189
30	(fiscal or financial or finance or funding).tw.	286,082
31	Cost minimization analysis/	3974
32	(cost adj estimate*).mp.	4184
33	(cost adj variable*).mp.	320
34	(unit adj cost*).mp.	5524
35	or/19-34	1,107,927
36	14 and 35	317

Table

#	Searches	Results
1	(breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)) (Topic)	614,142
2	(mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)) (Topic)	24,399
3	#1 OR #2	626,970
4	(endopredict OR epclin OR “ep score”) (Topic)	1778
5	(mammaprint OR 70-gene OR “70 gene”) (Topic)	4308
6	(oncotype OR “recurrence score” OR 21-gene OR “21 gene”) (Topic)	1682
7	(prosigna OR pam50 OR 50-gene OR “50 gene”) (Topic)	188
8	#7 OR #6 OR #5 OR #4	7430
9	#8 AND #3 and 2023 or 2022 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	1854
10	TS=(cost* and (effective* or utilit* or benefit* or minimi)) OR TS=(cost) OR TI=(economic* or pharmacoeconomic* or pharmaco-economic) OR TS=(price or pricing) OR TS=(financial or finance or finances or financed) OR TS=(fee or fees) OR TS=(value and (money or monetary)) OR TS=(economic) OR TS=(economic* and (hospital or medical or nursing or pharmaceutical)) OR TS=(“quality adjusted life year” or “quality adjusted life years”) OR TS=(qaly or qalys) OR TS=(budget*)	2,940,280
11	#9 AND #10	155

Table

Host	Database	Dates covered	Results
Ovid	Epub Ahead of Print, In-Process & Other Non-Indexed Citations, MEDLINE(R) Daily and MEDLINE(R)	1946–present	139
Ovid	EMBASE	1974–present	391
Clarivate	Web of Science Science Citation Index Expanded (1900–), Conference Proceedings Citation Index – Science (1990–)	1900–present	139
	Total retrieved	–	669
	Unique records		404

Table

#	Searches	Results
1	exp Breast Neoplasms/	340,146
2	exp mammary neoplasms/	23,375
3	exp breast/	52,745
4	exp neoplasms/	3,830,835
5	3 and 4	32,512
6	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).ti.	256,329
7	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).ti.	15,265
8	1 or 2 or 5 or 6 or 7	402,600
9	(euroqol or euro qol or eq5d or “eq 5d” or eq-5d).tw.	16,230
10	8 and 9	203
11	limit 10 to yr=“2017 -Current”	139

Table

#	Searches	Results
1	exp breast tumor/	643,804
2	exp breast/	127,820
3	exp neoplasm/	5,497,634
4	2 and 3	83,006
5	(breast* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary)).ti.	360,376
6	(mammar* adj5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar)).ti.	16,188
7	1 or 4 or 5 or 6	687,386
8	(euroqol or euro qol or eq5d or “eq 5d” or eq-5d).tw.	29,905
9	7 and 8	597
10	limit 9 to yr=“2017 -Current”	391

Table

#	Searches	Results
1	((breast* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullary))) (Title)	348,943
2	((mammar* NEAR/5 (neoplasm* or cancer* or tumo?r* or carcinoma* or adenocarcinoma* or sarcoma* or dcis or ductal or infiltrat* or intraductal* or lobular or medullar))) (Title)	11,684
3	#1 OR #2	359,016
4	#1 OR #2	359,016
5	(euroqol or euro qol or eq5d or eq 5d or “eq-5d”) (Topic)	19,973
6	#4 AND #5	213
7	#4 AND #5 and 2022 or 2023 or 2021 or 2020 or 2019 or 2018 or 2017 (Publication Years)	139

Tables in Appendix 3: Risk-of-bias assessment

TABLE 44 Risk of bias and applicability (adapted from PROBAST)

Number	Criterion	Scoring for this review
Risk of bias
Domain 1 participants	Were appropriate data sources used?	Yes (prognosis): reanalysis of RCT or cohort or nested case control AND patients did not receive chemotherapy
		Yes (predicting chemotherapy benefit): RCT or reanalysis of RCT
		No (prognostic): non-nested case control or case series AND/OR some/all patients had chemotherapy
		No (predicting chemotherapy benefit): patients not randomised to chemotherapy vs. no chemotherapy
Domain 1 participants	Were all inclusions and exclusions of participants appropriate?	Yes: all eligible patients from trial or consecutive eligible patients from prospective registry
		No: some eligible patients excluded (e.g. not sent for testing, insufficient tissue, test failures, missing data, AND/OR non-prospective registry)
		Unclear: if unclear
Domain 2 predictors (tests)	Were the tests (predictors) defined and assessed in a similar way for all participants?	Yes: if test assessed in a similar way for all participants (most/all studies in this review likely to score Yes as uses standardised test)
		No: test not assessed in a similar way for all participants
Domain 2 predictors (tests)	Were the tests (predictor assessments) made without knowledge of outcome data?	Yes: if test assessors blinded to clinical outcomes
		No: if not blinded
		Unclear: if unclear
Domain 3 outcomes	Were the outcome definitions standardised or defined a priori?	Yes: at least one outcome was standardised (e.g. DRFS, OS) or defined a priori
		No: all outcomes non-standardised and not defined a priori
		Unclear: if unclear
Domain 3 outcomes	Were the outcomes determined without knowledge of test (predictor) information?
		Yes: if outcome assessors blinded to test results
		No: if not blinded
		Unclear: if unclear
Domain 3 outcomes	Was chemotherapy decision made before test result known?	Yes: test did not influence use of chemotherapy (Yes if retrospective use of test on stored tumour samples, i.e. reanalyses of RCTs or cohorts)
		No: test result may have influenced use of chemotherapy (No for observational studies of prospective use of test)
		(This item is not in PROBAST but is important for this review)
Domain 4 analysis	Were there a reasonable number of participants with outcome data?	Yes: at least 100 patients with outcome data
		No: < 100 patients with outcome data
Domain 4 analysis	Were all enrolled participants included in \|the analysis?	Yes: if all enrolled participants included in the analysis
		No: if some enrolled patients not analysed
Applicability
Domain 1 participants	Did the included participants match the review question?	Yes: all patients in scope (HR+, HER2–, LN1–3)
		Mostly: < 20% out of scope
		No: > 20% out of scope
		Unclear: if unclear
Domain 2 predictors (tests)	Did the definition and assessment of tests (predictors) match the review question?	Yes: same as commercially available tests
		No: different from commercially available tests (e.g. FFPE vs. fresh samples, test methods)
Domain 3 outcomes	Did the outcomes match the review question?	Yes: at least one outcome matched the review question
		No: no outcomes matched the review question

TABLE 45 Risk of bias in prospective RCTs (using Cochrane RoB2)

RCT	Risk of bias due to …
Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported result	Overall risk of bias
RxPONDERKalinsky 202129	Low	Low	Low	Low	Low	Low
MINDACTPiccart 202130	Low	Low	Low	Low	Low	Low

TABLE 46 Risk of bias in prognostic studies (retrospective reanalyses of RCTs and cohorts)

Reference	Cohort	DesignDerivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match review question?	Tests match review question?	Outcomes match review question?
Albain 201032	SWOG-8814	RCT-RV	Y (RCT-R, ET only)	N (InT, TF)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Constantinidou 202265	Cyprus + Notts	Cohort-RV	Y (cohort-R, ET only)	N (InT, MD)	Y	Y	Y	Y	Y	N	N	Y	Y	Y
Drukker 201453	VdV cohort, the Netherlands	Cohort-RV (21% also in derivation set)	N (cohort-R, some CT)	Y	Y	UC	Y	Y	Y	Y	Y	N (> 20% ER− and > 20% LN4 +)	Y	Y
Filipits 201963	ABCSG-6/8	RCT-RV	Y (RCT-R, ET only)	UC	Y	UC	Y	Y	Y	Y	Y	Y	Y	Y
Gnant 2014;56 Filipits 201457	ABCSG-8, Austria	RCT-RV	Y (RCT-R, ET only)	N (InT, MS, TF, no consent)	Y	Y	Y	Y	Y	Y	N	Mostly (11% LN4 +)	Y	Y
Jackisch 2022 (abst)55	Germany, PATH	Cohort-RV	N (cohort-R, some CT)	N (reason NR)	Y	UC	Y	Y	Y	N	N	UC	Y	Y
Laenkholm 201858	DBCG, Denmark	Cohort-RV	Y (cohort-R, ET only)	N (FT, MD)	Y	Y	Y	Y	Y	Y	N	Y	Y	Y
Lundgren 202262	SBII:pre trial	RCT-RV	Y (RCT-R, ET only)	N (InT, FT, MD)	Y	N	Y	Y	Y	Y	Y	Y	Y	Y
Mamounas 201849	NSABP-28	RCT-RV	N (RCT-R, all CT)	N (InT, MS)	Y	UC	Y	Y	Y	Y	Y	UC (HER2 NR)	Y	Y
Martin 2016;59 Martin 201460	GEICAM 9906, Spain	RCT-RV	N (RCT-R, adjuvant CT)	N (MD)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Mook 200933	NKI and Italy	Cohort-RV	N (cohort-R, some CT)	N (InT, RNA quality)	Y	Y	Y	Y	Y	Y	Y	N (> 20% ER−, 16% HER2 +)	Y	Y
Penault-Llorca 201850	PACS01	RCT-RV	N (RCT-R, some CT)	N (FT, InT, MS)	Y	Y	Y	Y	Y	Y	Y	N (> 20% LN4 +)	Y	Y
Pu 202061	WHEL Study	RCT-R	N (RCT-R, some CT)	N (InT, MS, TF)	Y	UC	Y	Y	Y	Y	N	UC (NR N nodes)	Y	Y
Sestak 2018;20 201748	TransATAC	RCT-RV	Y (RCT-R, ET only)	N (InT; FT)	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Sestak 202064	Lobular subgroup (TransATAC + ABCSG-6/8)	RCT-RV	Y (RCT-R, ET only)	UC	Y	Y	Y	Y	Y	Y	N	Mostly (20% LN4 +)	Y	Y
Vliek 201754	RASTER	Cohort-RV	N (cohort-R, some CT)	N (InT, MS, no consent)	Y	UC	Y	Y	Y	Y	Y	Mostly (17% ER−, 15% HER2 +)	Y	Y

_{Cohort-R, reanalysis of cohort study; CT, chemotherapy; FT, failed test; InT, insufficient tissue; LN, number of positive lymph nodes; MD, missing data; MS, missing samples; N, no; NR, not reported; RCT-R, reanalysis of RCT; TF, test failure; UC, unclear; V, validation study; WHEL, Women’s Healthy Eating and Living; Y, yes.}

TABLE 47 Risk of bias in prognostic studies (observational studies of prospective use of test)

Reference	Cohort	DesignDerivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match review question?	Tests match review question?	Outcomes match review question?
Braun 202277	Red Cross Hospital, Munich, Germany	ObservationalV	N (prospective use of test, some CT)	Y	Y	Y	Y	UC	N	Y	N	Mostly (20% LNmic)	Y	Y
Ibraheem 202070	NCDB	ObservationalV	N (prospective use of test, CT)	N (MD, SFT)	Y	Y	Y	UC	N	Y	Y	Y	Y	Y
Massarweh 201869Petkov 201667Roberts 201768	SEER	ObservationalV	N (prospective use of test, some CT)	N (InT, MS, SFT, no consent)	Y	Y	Y	Y	N	Y	Y	UC (% LNmic NR)	Y	Y
Nitz 201776	WSG PlanB	ObservationalV	N (prospective use of test, some CT)	N (dropout, screening failure)	Y	Y	Y	UC	N	Y	Y	Y	Y	Y
Poorvu 202075	Young Women’s Breast Cancer Study	ObservationalV	N (part prospective use of test, part stored samples, some CT)	N (InT, MS, no consent)	Y	Y	Y	UC	N	Y	N	UC (% LNmic NR)	Y	Y

_{CT, chemotherapy; InT, insufficient tissue; LNmic, lymph node micrometastases; MD, missing data; MS, missing samples; N, no; NR, not reported; SFT, only those sent for test included; UC, unclear; V, validation study; WSG, West German Study Group; Y, yes.}

TABLE 48 Risk of bias in prediction studies

Reference	Cohort	Derivation or validation?	Risk of bias	Applicability
Domain 1 participants	Domain 2 predictors	Domain 3 outcomes	Domain 4 analysis	Participants	Predictors	Outcomes
Appropriate data sources?	Appropriate exclusions?	Tests same for all participants?	Blinded test assessors to outcomes?	Outcomes standardised or a priori?	Blinded outcome assessors to test?	CT decision made before test result known?	Participants N > 100?	All analysed?	Participants match protocol?	Tests match review question?	Outcomes match review question?
Albain 201032	SWOG-8814	RCT-RV	Y (RCT-R)	N (InT, TF)	Y	Y	Y	Y	Y	Y	N	N (> 20% LN4 +)	Y	Y
Mook 200933	NKI and Italy	Cohort-RV	N (not RCT)	N (InT, RNA qual)	Y	Y	Y	Y	Y	Y	Y	N (> 20% ER−, 16% HER2 +)	Y	Y
Abel 202281	NCDB	ObservationalV	N (not RCT)	N (MD, SFT)	Y	Y	Y	Y	N	Y	N	Y	Y	Y
Cao 2022 (abst)82
Ibraheem 201971
Iorgulescu 201983
Kumar 2023 (abst)84
Nash 202372
Weiser 202174
Weiser 202273
Petkov 2020 (abst)80	SEER	ObservationalV	N (not RCT)	N (InT, MS, SFT, no consent)	Y	Y	Y	Y	N	Y	Y	UC (% LNmic NR)	Y	Y
Rotem 2022 (abst)79Stemmer 201766	Clalit, Israel	ObservationalV	N (not RCT)	N (SFT)	Y	Y	Y	Y	N	Y	N	N (> 20% LNmic)	Y	Y

_{Cohort-R, reanalysis of cohort study; InT, insufficient tissue; MD, missing data; MS, missing samples; N, no; NR, not reported; LN, number of positive lymph nodes; LNmic, lymph node micrometastases; RCT-R, reanalysis of RCT; SFT, only those sent for test included; TF, test failure; UC, unclear; V, validation study; Y, yes.}

Tables in Appendix 4: Additional tables for prognostic ability

TABLE 49 Prognostic data (Oncotype DX)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Menostatus	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
Oncotype DX: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	18, 30	57	32	11	95.9	84.8	83.6	0–10 years80.65–10 years82.1	0–10 years70.95–10 years80.5	0–10 years62.05–10 years72.5	0–5 years: int vs. low: HR 3.84 (1.31 to 11.23)0–5 years: high vs. low: HR 4.45 (1.19 to 16.58)0–10 years: int vs. low: HR 1.66 (0.86 to 3.23)0–10 years: high vs. low: HR 2.35 (0.99 to 5.60)0–10 years: per 1 SD change: 1.39 (1.05 to 1.85)*Adj: LR vs. CTS (p = 0.06) and NPI (p = 0.1)	YYNNYN*
Oncotype DX: distant recurrence, variable ET/CT
Mamounas 201849NSABP-28	DRFI	n = 722All CT + ETRCT-R	LN1–3100% ER+NR HER2	All meno	18, 30	37	34	28	–	–	–	84.7	71.5	63.1	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 2.42 (NR); p < 0.001	YY*
Penault-Llorca 201850PACS01	DRFI	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	93.7	87.3	69.3	–	–	–	0–5 years: HR per 50-RS: 4.14 (2.67 to 6.43); p < 0.001*0–5 years: adj HR 3.36 (1.88 to 6.00), p < 0.001	YY*
Oncotype DX: DFS
Albain 201032SWOG-8814	DFS	n = 148ET monoRCT-R	LN + 100%LN4 +: 37%100% HR+91% HER2–	Post-meno	18, 30	37	31	32	–	–	–	60	49	43	0–5 years: HR 5.55 (2.32 to 3.28); p = 0.00020–10 years: between risk groups: p = 0.0170–10 years: HR per 50-RS: 2.64 (1.33 to 5.27); p = 0.0065–10 years: HR 0.86 (0.27 to 2.74); p = 0.80	YYYN
Mamounas 201849NSABP-28	DFS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	79.8	64.8	57	0–10 years: p < 0.0010–5 years: adj HR per 50-RS 3.81 (2.67 to 5.43); p < 0.0010–10 years: adj HR per 50-RS 2.53 (1.90 to 3.38); p < 0.001*5–10 years: adj HR per 50-RS 1.39 (0.88 to 2.19); p = 0.16	YYYN*
Penault-Llorca 201850PACS01	DFS	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	90.8	84.9	64.6	–	–	–	0–5 years: HR per 50-RS: 3.28 (2.18 to 4.94); p < 0.001*0–5 years: adj HR 2.66 (1.62 to 4.37), p < 0.001	YY*
Kalinsky 202129RxPONDER	IDFS	n = 5018CT + ET vs. ETProsp RCT	LN1–3100% HR+100% HER2−	All meno (67% post)	All ≤ 25	–	–	–	See prediction tables for outcomes per risk group	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.05 (1.04 to 1.07), p < 0.001 (adj meno and CT)	Y*
				Post-meno	All ≤ 25	–	–	–	–	–	–	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.05 (1.03 to 1.07), p < 0.001 (adj CT, nodes, grade, tumour size, age)	Y*
				Pre-meno	All ≤ 25	–	–	–	–	–	–	–	–	–	*0–5 years: HR per unit-RS (within RS 0–25): 1.06 (1.02 to 1.09), p = 0.001 (adj CT, nodes, grade, tumour size, age)	Y*
Abdou 202351RxPONDER	IDFS	n = 4015CT + ET vs. ETProsp RCT	LN1–3100% HR+100% HER2−	Whiten = 2833	All ≤ 25	–	–	–	91.5			–	–	–	–	–
				Blackn = 248	All ≤ 25	–	–	–	87.0			–	–	–	–	–
				Asiann = 324	All ≤ 25	–	–	–	93.9			–	–	–	–	–
				Hispanicn = 610	All ≤ 25	–	–	–	91.4			–	–	–	–	–
Oncotype DX: OS and BCSS
Albain 201032SWOG-8814	OS	n = 148ET monoRCT-R	LN + 100%LN4 +: 37%100% HR+, 91% HER2–	Post-meno	18, 30	37	31	32	–	–	–	77	68	51	0–10 years: between risk groups: p = 0.0030–10 years: HR per RS-50 : 4.42 (1.96 to 9.97), p = 0.0006	YY
Penault-Llorca 201850PACS01	OS	n = 530All CT74% ETRCT-R	LN1–3 : 60%LN4 +: 40%100% HR+90% HER2−	All meno (39% post)	18, 30	39	30	31	99	95.6	85.6	–	–	–	0–5 years: HR per 50-RS: 5.0 (3.01 to 8.28); p < 0.001	Y
Mamounas 201849NSABP-28	OS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	93.3	79.2	70.7	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 3.09 (CI NR); p < 0.001	YY*
Mamounas 201849NSABP-28	BCSS	n = 722All CT + ETRCT-R	LN1–3100% ER +NR HER2	All meno	18, 30	37	34	28	–	–	–	98	82.9	75.6	0–10 years: p < 0.001*0–10 years: adj HR per 50-RS: 3.38 (CI NR); p < 0.001	YY*`

_{Adj, adjusted; CT, chemotherapy; CTS, Clinical Treatment Score (set of clinical factors); int, intermediate; LN, lymph nodes (number positive); LR, likelihood ratio; meno, menopausal; N, no; NR, not reported; prosp, prospective; RCT-R, RCT reanalysis; SD, standard deviation; sig, significant; Y, yes. aThe last column indicates whether each hazard ratio between test risk groups is statistically significant at the 5% level. Asterisk (*) denotes analyses adjusted for clinical factors.}

TABLE 50 Prognostic data (MammaPrint)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Meno status	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years /other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
MammaPrint: distant recurrence, ET monotherapy
No studies
MammaPrint: distant recurrence, variable ET/CT
Piccart 202130	DMFS	n = 1176	LN1–3	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	95.7(50% CT)	–	89.0(all CT)	8 years91.0(50% CT)	–	8 years79.1(all CT)	–	–
MINDACT (not on prognostics summary table since CT use per risk group was influenced by test result)		CT + ET vs. ET	100% HR+	Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	96.3(no CT)	–	–	8 years94.0(no CT)	–	–	–	–
		Prosp-RCT	100% HER2−
	DMFI	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	96.3(50% CT)	–	89.3(all CT)	8 years92.3(50% CT)	–	8 years80.9(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	97.5(no CT)	–	–	8 years95.2(no CT)	–	–	–	–
Lopes Cardozo 202252	DMFI	N = 201 (ultra-low)	LN1–3	–	> 0.355 ultra-low	Ultra-low: 15	–	–	Ultra-low: 97.4	–	–	8 yearsUltra-low:95.2	–	–	–	–
MINDACT		Var ET/CT	99% ER+
		Prosp-RCT	97% HER2−
Drukker 201453	DMFS	n = 144	LN1–3 : 74%	Age < 53 years	0.4	38	–	62	94.5	–	64.7	10 years 78.6	–	10 years 54.325 years 44.5	0–25 years: HR 2.24 (1.25 to 4.00); p = 0.01	Y
VdV cohort, the Netherlands		Var ET/CT	LN4 +: 26%									25 years
		Cohort-R	77% ER+									NE
			NR HER2
Mook 200933NKI and Italy	DMFS	n = 241Var ET/CTCohort-R	LN1–3 : 100% inc micromets79% ER+84% HER2−	All meno	NR	41	–	59	98	–	80	91	–	76	0–10 years: HR 4.13 (1.72 to 9.96); p = 0.002*0–10 years: adj HR: 2.99 (0.996 to 8.99); p = 0.051	YN*
Vliek 201754RASTER	DRFI	N = 134Var ET/CTCohort-R	LN1–383% ER+85% HER2−	All ages	NR	48	–	52	98.4	–	86.9	94.9	–	80.7	0–10 years: low vs. high: HR 4.7 (1.3 to 16.2); p = 0.008	Y
				All agesHigh mAOL (n = 109)	NR	40	–	60	97.7	–	86.1	95.2	–	79.5	0–10 years: low vs. high: HR 4.8 (1.1 to 21.4), p = 0.022	Y
MammaPrint: DFS
Piccart 202130MINDACT	DFS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	91.6(50% CT)	–	85.9(all CT)	8 years84.5(50% CT)	–	8 years74.5(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	92.6(no CT)	–	–	8 years85.6(no CT)	–	–	–	–
MammaPrint: OS and BCSS
Piccart 202130MINDACT	OS	n = 1176CT + ET vs. ETProsp-RCT	LN1–3100% HR+100% HER2−	High mAOL (n = 989)	> 0 low, ≤ 0 high	69	–	31	98.3(50% CT)	–	95.8(all CT)	8 years95.1(50% CT)	–	8 years89.1(all CT)	–	–
				Low mAOL (n = 187)	> 0 low, ≤ 0 high	92	–	8	98.1(no CT)	–	–	8 years98.1(no CT)	–	–	–	–
Drukker 201453VdV cohort, the Netherlands	OS	n = 144Var ET/CTCohort-R	LN1–3 : 74%LN4 +: 26%77% ER+NR HER2	Age < 53 years	0.4	38	–	62	98.2	–	76.9	10 years 92.525 years 42.2	–	10 years 58.725 years 47.1	0–25 years: HR 1.83 (1.07 to 3.11), p = 0.03	Y
Jackisch 2022 (abst)55Germany, PATH	OS	n = 38Var ET/CTCohort-R	LN+NR	All meno (assumed)	NR	53	–	47	–	–	–	93.3	–	40.4	–	–
Mook 200933NKI and Italy	BCSS	n = 241Var ET/CTCohort-R	LN1–3 : 100% inc micromets79% ER+84% HER2−	All meno	NR	41	–	59	99	–	88	96	–	76	0–10 years: HR 5.70 (2.01 to 16.23); p = 0.001*0–10 years: adj HR 7.17 (1.81 to 28.43); p = 0.005	YY*
				All menoHigh AOL (n = 209)	NR	–	–	–	–	–	–	94	–	76	0–10 years: HR 4.12 (1.45 to 11.76); p = 0.008	Y

_{Adj, adjusted; cohort-R, cohort reanalysis; CT, chemotherapy; int, intermediate; LN, lymph nodes (number positive); meno, menopausal; N, no; NR, not reported; prosp, prospective; sig, significant; var, variable; Y, yes. aThe last column indicates whether each hazard ratio between test risk groups is statistically significant at the 5% level. Asterisk (*) denotes analyses adjusted for clinical factors.}

TABLE 51 Prognostic data (Prosigna)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	Menostatus	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
Prosigna: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	NR; assume 16, 40	8	32	60	100	91.7	87.4	0–10 years1005–10 years100	0–10 years79.35–10 years87.0	0–10 years69.35–10 years75.0	0–5 years: int vs. high: HR 1.30 (0.47 to 3.60)0–10 years: int vs. high: HR 1.37 (0.69 to 2.72)HR per 1 SD change: 1.58 (1.16 to 2.15)*LR vs. CTS (p = 0.04) and NPI (p = 0.09)	NNYY, N*
Gnant 2014,56 Filipits 201457ABCSG-8, Austria	DMFS	n = 413ET monoRCT-R	LN1–3 : 89%LN4 +: 11%100% ER+ 100% HER2−	Post-meno	16, 40	4	34	62	–	–	–	0–10 years 1005–15 years100	0–10 years 93.65–15 years87.0	0–10 years 76.15–15 years75.0	5–15 years: low risk: no events5–15 years: int vs. high: HR 3.15 (1.20 to 8.24); p = 0.0200–10 years: prognostic over clinical factors (p < 0.0001)5–15 years: prognostic over clinical factors (p = 0.003)	–YYY
Laenkholm 201858DBCG, Denmark	DRFS	n = 1395ET monoCohort-R	LN1–3100% HR+ 100% HER2−	Post-meno	BespokeVaries by N nodes	26	28	46	–	–	–	96.5	88.5	77.9	0–10 years: unadj: p < 0.0010–10 years: low vs. int: adj HR 0.39 (0.20 to 0.77)0–10 years: high vs. int: adj HR 1.54 (1.04 to 2.26); p < 0.001	YYY
						40 only	–	–	–	–	–	–	95.2(low to int)	78.1	–	–
Prosigna: distant recurrence, variable ET/CT
Martin 2016,59 Martin 201460GEICAM 9906, Spain	DMFS	n = 536All CT + ET RCT-R	LN1–3 : 64%LN4 +: 36%100% ER+ 100% HER2−	All meno (46% post)	18, 65	19	56	26	–	–	–	92	74	66	0–10 years: low vs. int: HR 4.4 (NR)0–10 years: low vs. high: HR 5.8 (NR), p < 0.0001*Prosigna vs. EPclin + clinical factors (p = 0.567)	–YN*
Prosigna: DFS
Pu 202061WHEL Study	DFS	n = 344Var ET/CTRCT-R	LN+100% ER+ 100% HER2−	All meno	NR	26	53	21	–	–	–	81	64	56	0–10 years: p = 0.02	Y
Prosigna: OS and BCSS
Lundgren 202262SBII:pre trial	OS	n = 123ET/noneRCT-R	LN1–3100% ER+ 100% HER2−	Pre-meno	16, 40	2	42	57	–	–	–	–	–	–	0–10 years: int vs. high: HR 1.84 (0.91 to 3.74); p = 0.090–10 years: int vs. high: adj HR 1.32 (0.61 to 2.88); p = 0.48> 10 years: int vs. high: HR 1.02 (0.54 to 1.93); p = 0.96> 10 years: int vs. high: adj HR 1.29 (0.66 to 2.53); p = 0.46	NNNN
Lundgren 202262SBII:pre trial	BCFI	n = 123ET/noneRCT-R	LN1–3100% ER+ 100% HER2−	Pre-meno	16, 40	2	42	57	–	–	–	–	–	–	0–10 years: int vs. high: HR 1.99 (1.08 to 3.66); p = 0.030–10 years: int vs. high: adj HR 1.85 (0.95 to 3.58); p = 0.07> 10 years: int vs. high: HR 1.19 (0.50 to 2.80); p = 0.70> 10 years: int vs. high: adj HR 1.13 (0.43 to 2.95); p = 0.81	YNNN
Pu 202061WHEL Study	BCSS	n = 344Var ET/CTRCT-R	LN+100% ER+ 100% HER2−	All meno	NR	26	53	21	–	–	–	90	84	77	0–10 years: p = 0.003	Y

_{Adj, adjusted; BCFI, breast cancer-free interval; cohort-R, cohort reanalysis; CT, chemotherapy; CTS, Clinical Treatment Score (set of clinical factors); int, intermediate; LN, lymph nodes (number positive); LR, likelihood ratio; meno, menopausal; N, no; NPI, Nottingham Prognostic Index; NR, not reported; RCT-R, RCT reanalysis; SD, standard deviation; sig, significant; Y, yes. aThe last column indicates whether each hazard ratio between test risk groups is statistically significant at the 5% level. Asterisk (*) denotes analyses adjusted for clinical factors.}

TABLE 52 Prognostic data (EPclin)

ReferenceStudy/cohort	Outcome	N, ET/CTDesign	Nodal statusHR, HER2	MenoClin risk	Test cut-offs	Distribution %	Risk 0–5 years %	Risk 0–10 years/other %	HR between test groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High	Low	Int	High
EPclin: distant recurrence, ET monotherapy
Sestak 2018,20 201748TransATAC	DRFI	n = 183ET monoRCT-R	LN1–3100% HR+ 100% HER2−	Post-meno	3.3	23	–	77	97.9	–	87.6	0–10 years94.45–10 years96.7	0–10 years-5–10 years-	0–10 years69.75–10 years76.4	0–5 years: high vs. low: HR 6.00 (0.80 to 44.93)0–10 years: high vs. low: HR 6.77 (1.63 to 28.07)0–10 years: per 1 SD change: 1.69 (1.29 to 2.22)*LR vs. CTS (p = 0.20) or NPI (p = 0.02)	NYYN, Y*
Filipits 201963ABCSG-6/8	DRFR	n = 453ET monoRCT-R	LN1–3100% ER+ 100% HER2−	Post-meno	3.3	35	–	65	–	–	–	0–10 years 95.60–15 years 84.75–10 years 98.25–15 years 87.0	–	0–10 years 80.90–15 years 75.15–10 years 90.55–15 years 84.0	0–10 years: HR 3.65 (1.73 to 7.68); p = 0.00030–10 years: adj HR 2.68 (1.77 to 4.08); p < 0.00015–15 years: HR 3.00 (1.03 to 8.71); p = 0.0345–15 years: adj HR 3.43 (1.74 to 6.76); p = 0.0005	YYYY
Sestak 202064Lobular (from TransATAC + ABCSG-6/8)	DRFS	n = 144ET monoRCT-R	LN1–3 : 80%LN4 +: 20%100% HR+ 100% HER2−	Post-menoLobular	3.3	26	–	74	–	–	–	93.6	–	68.8	HR 3.70 (2.49 to 5.50); p < 0.0001*EPclin vs. clinical factors (p = 0.0026)	YY*
Constantinidou 202265Cyprus + Notts	DRFS	n = 62ET monoCohort-R	LN1–3100% ER+ 100% HER2−	Pre-meno	3.3	19	–	81	–	–	–	100	–	75	High vs. low: p = 0.066*Adj HR (cont score): 2.91 (1.70 to 4.97), p < 0.001	NY*
EPclin: distant recurrence, variable ET/CT
Martin 2016,59 Martin 201460GEICAM 9906, Spain	DMFS	n = 555All CT + ETRCT-R	LN1–3 : 64%LN4 +: 36%100% ER+ 100% HER2−	All meno (46% post)	3.3	13	–	87	–	–	–	100	–	72	Low vs. high: HR not estimable, p < 0.0001*EPclin vs. clinical factors (p = 0.0018)	YY*
				Pre-meno	3.3	12	–	88	–	–	–	100	–	70	Low vs. high: HR NR, p = 0.0006	Y
				Post-meno	3.3	13	–	87	–	–	–	100	–	76	Low vs. high: HR NR, p = 0.0023	Y

_{Adj, adjusted; cohort-R, cohort reanalysis; cont, continuous; CT, chemotherapy; CTS, Clinical Treatment Score (set of clinical factors); int, intermediate; LN, lymph nodes (number positive); LR, likelihood ratio; meno, menopausal; N, no; NR, not reported; RCT-R, RCT reanalysis; SD, standard deviation; sig, significant; Y, yes. aThe last column indicates whether each hazard ratio between test risk groups is statistically significant at the 5% level. Asterisk (*) denotes analyses adjusted for clinical factors.}

Tables in Appendix 5: Additional tables for observational data

TABLE 53 Observational data for Oncotype DX (all outcomes and analyses)

Cohort	Reference	Nodal statusHR, HER2	Outcome	NET/CT	MenoAgeClin	Test cut-offs	Distribution %	% risk of outcome	HR between test risk groups (95% CI)	Sig?a*Adj
Low	Int	High	Low	Int	High
Oncotype DX: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	DRFI(0–5 years)	n = 709Var ET/CT	All meno	18, 30	53	36	10	96.8 (7% CT)	93.7 (40% CT)	83.1 (86% CT)	0–5 years: low vs. high: HR 0.19 (0.09 to 0.40)0–5 years: int vs. high: HR 0.39 (0.20 to 0.79); p < 0.0010–5 years: adj HR: low vs. high: HR 0.23 (0.11 to 0.50)0–5 years: adj HR: int vs. high: HR 0.42 (0.20 to 0.86); p = 0.001	YYYY
						11, 25	≤ 25:81	19	95.7(5% CT)	96.0(18% CT)	86.9(77% CT)	0–5 years: p < 0.001	Y
						≤ 25, 26–30				96.0 (15% CT)	91.5 (67% CT)	–	–
						18–25					94.4 (31% CT)		–	–
				n = 109Var ET/CT	Age < 50	18, 30	48	37	16	96.2 (12% CT)	100.0 (48% CT)	64.2 (100% CT)	0–5 years: p < 0.001	Y
				n = 464Var ET/CT	Age 50–69	18, 30	54	37	9	97.6 (6% CT)	93.5 (42% CT)	87.8 (90% CT)	0–5 years: p = 0.017	Y
				n = 136Var ET/CT	Age ≥ 70	18, 30	57	33	10	94.7 (7% CT)	88.7 (22% CT)	92.9 (57% CT)	0–5 years: p = 0.458	N
Young Women’s Breast Cancer Study	Poorvu 202075	LNmic, LN1–3100% ER+100% HER2−	DRFS(0–6 years)	n = 163Var ET/CT	Age ≤ 40	18, 30	33	42	25	0–6 years:85.9 (83% CT)	0–6 years:87.3 (97% CT)	0–6 years:62.8 (98% CT)	0–6 years: p = 0.004	Y
						11, 25	9	54	37	0–6 years:92.3 (79% CT)	0–6 years:85.2 (92% CT)	0–6 years:71.3 (97% CT)	0–6 years: p = 0.10	N
Oncotype DX: DFS
WSG PlanB	Nitz 201776	LN1–3100% HR+100% HER2−	DFS(0–5 years)	n = 110Var ET/CT	All meno	0–10				94.4(No CT)	–	–	–	–
Red Cross Hospital, Munich, Germany	Braun 202277	LNmic: 20%LN1–3 : 80%100% HR+100% HER2−	DFS(0–5 years)	n = 217Var ET/CT	All meno (63% post)	≤ 25, 26 +	86	14	RS 0–25 : 90.3 (19% CT)	71.0 (93% CT)	–	–
Oncotype DX: OS and BCSS
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	BCSS(0–5 years)	n = 709Var ET/CT	All meno	18, 30	53	36	10	99.5 (7% CT)	96.6 (40% CT)	94.3 (86% CT)	0–5 years: p < 0.001	Y
						11, 25	RS≤ 25:81		19	99.1 (5% CT)	98.8 (18% CT)	93.5 (77% CT)	0–5 years: p < 0.001	Y
						≤ 25, 26–30					98.9 (15% CT)	92.6 (67% CT)	–	–
						18–25					97.8 (31% CT)		–	–
SEER registry	Petkov 201667	LN1mic, LN1–3100% HR+100% HER2−	BCSS(< 5 years)	n = 4691Var ET/CT	All	18, 30	57	36	7	99.0 (23% CT)	97.7 (47% CT)	85.7 (75% CT)	< 5 years: high vs. low: HR 11.0 (7.8 to 15.5)< 5 years: int vs. low: HR 3.1 (2.3 to 4.3), p < 0.001< 5 years: adj HR: high vs. low: HR 7.8 (5.3 to 11.6)< 5 years: adj HR: int vs. low: HR 3.0 (2.1 to 4.2); p < 0.001	YYYY
				n = 328Var ET/CT	Black ethnicity	18, 30	54	36	9	99.4 (CT NR)	98.9	91.3	< 5 years: p = 0.4117	N
				n = 4021Var ET/CT	White ethnicity	18, 30	58	36	7	99 (CT NR)	97.6	84.1	< 5 years: p < 0.0001	Y
				n = 320Var ET/CT	Other ethnicity	18, 30	57	34	8	98.5 (CT NR)	99.1	100	< 5 years: p = 0.8427	N
SEER registry	Roberts 201768	LN1mic, LN1–3100% HR+100% HER2−	BCSS(0–5 years)	n = 6483Var ET/CT	All	18, 30	58	35	7	98.8 (CT NR)	97.3	88.5	0–5 years: p < 0.001*0–5 years: adj: p < 0.001	YY*
			OS(0–5 years)	n = 6483Var ET/CT	All	18, 30	58	35	7	92.1 (CT NR)	90.9	81.7	0–5 years: p < 0.001*0–5 years: adj: p < 0.001	YY*
SEER registry	Massarweh 201869	LN1mic, LN1–3100% HR+100% HER2−	BCSS(0–5 years)	n = 6437Var ET/CT	Women	18, 30	59	35	7	98.8 (23% CT)	97.3 (48% CT)	89.2 (77% CT)	0–5 years: p < 0.001	Y
				n = 46Var ET/CT	Men	18, 30	52	26	22	100 (33% CT)	100 (50% CT)	N/A (60% CT)	0–5 years: p = 0.02	Y
			OS(0–5 years)	n = 6437Var ET/CT	Women	18, 30	59	35	7	92.2 (23% CT)	90.8 (48% CT)	83.2 (77% CT)	0–5 years: p < 0.001	Y
				n = 46Var ET/CT	Men	18, 30	52	26	22	78.9 (33% CT)	100 (50% CT)	N/A (60% CT)	0–5 years: p = 0.002	Y
NCDB	Ibraheem 202070	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 25,029Var ET/CT	All meno	11, 25	24	64	13	–	–	–	0–5 years: int vs. low: HR 1.15 (0.97 to 1.36)High vs. low: HR 2.94 (2.43 to 3.56)Per 10-RS: HR 1.38 (1.31 to 1.44)	NYY
NCDB	Ibraheem 201971	LN1–3 : 97%LN4-9 : 3%	OS(0–5 years)	n = 13,163Var ET/CT	All meno	11, 25	–	–	–	–	–	–	0–5 years:RS 18–25 vs. 11–17: HR 1.20 (1.07 to 1.35); p < 0.001RS 18–25 vs. 11–17: adj HR 1.15 (1.03 to 1.29); p < 0.001RS 26–30 vs. 11–17: HR 1.91 (1.65 to 2.22); p < 0.001RS 26–30 vs. 11–17: adj HR 1.62 (1.38 to 1.89); p < 0.001	YYYY
NCDB	Nash 202372	LN1–3100% HR+100% HER2−	OS(NR, med FU 5.5 years)	n = 4124Var ET/CT	Age 40–50	11, 25	–	–	–	–	–	–	RS 26–30 vs. 0–25: adj HR 2.29 (1.49 to 4.86)RS 31–50 vs. 0–25: adj HR 3.70 (2.03 to 6.75)*RS 51–100 vs. 0–25: adj HR 2.31 (0.78–6.86); p < 0.001	YYN*
NCDB	Weiser 202273	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 2691Var ET/CT	Lobular	11, 25	–	–	–	95.5	95.5	83.8	0–5 years: p = 0.0004*Adj: sig	YY*
NCDB	Weiser 202174	LN1–3100% HR+100% HER2−	OS(0–5 years)	n = 28,591Var ET/CT	All	≤ 25	–	–	–	–	–	–	*0–5 years: RS 18–25 vs. RS 12–17: adj HR 1.30 (1.00 to 1.68)	Y*

_{Adj, adjusted; CT, chemotherapy; int, intermediate; LN, lymph nodes (number positive); meno, menopausal; N, no; NR, not reported; sig, significant; var, variable; Y, yes. aThe last column indicates whether each hazard ratio between test risk groups is statistically significant at the 5% level. Asterisk (*) denotes analyses adjusted for clinical factors.}

Tables in Appendix 6: Additional tables for chemotherapy effect within risk groups

TABLE 54 Chemotherapy effect within risk groups: Registry data for Oncotype DX (all outcomes)

Cohort	Reference	Nodal statusHR, HER2	Outcome	N	MenoAgeClin	Test cut-offs	% risk of outcome	Abs diff CT vs. no CT	HR for CT vs. no CT (95% CI)	Inter-action	Preda*Adj
Low	Int	High	Low	Int	High	Low	Int	High
CT	No	CT	No	CT	No
Oncotype DX: observational: distant recurrence
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	DRFI0–5 years	n = 709	All meno	18, 30	92.3	97.1	99	90.3	82	90	−4.8	8.7	−8.0	p = 0.245	p = 0.019	–	–	–
						11, 25	83.3	96.3	98.8	95.4	97.5	79.7	−13.0	3.4	17.8	–	–	p = 0.017	–	–
						All ≤ 25	–	–	97.7	95.6	–	–	2.1	–	p = 0.521	–	–	–
						18–25	–	–	100	91.8	–	–	–	8.2	–	–	p = 0.058	–	–	–
	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	DRFS0–7 years	n = 140	All meno	All26–30	–	–	–	–	89.4	78.0	–	–	11.4	–	–	Not sig	–	–
Oncotype DX: observational: BCSS and OS
Clalit, Israel	Stemmer 201766	LN1mic: 42%LN1–3 : 58%100% ER+100% HER2−	BCSS0–5 years	n = 709	All meno	18, 30	100.0	99.4	98.9	95.1	93.4	100	0.6	3.8	−6.6	–	–	–	–	–
						11, 25	100.0	99.1	100.0	98.6	97.1	84.0	0.9	1.4	13.1	–	–	–	–	–
						All ≤ 25	–	–	100.0	98.7	–	–	1.3	–	–	–	–	–	–
						18–25	–	–	100.0	96.8	–	–	–	3.2	–	–	–	–	–	–
	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	BCSS0–7 years	n = 140	All meno	26–30	–	–	–	–	98.7	93.8	–	–	4.9	–	–	p = 0.024	–	–
SEER	Petkov 2020 (abst)80	LN1mic-LN3100% HR+100% HER2−	BCSS0–5 years	n = 2588	Age≤ 50	0–10	100	100	–	–	–	–	0	–	–	–	–	–	–	–
						11–15	–	–	97.7	99.5	–	–	–	−1.8	–	–	–	–	–	–
						16–20	–	–	98.4	98.7	–	–	–	−0.3	–	–	–	–	–	–
						21–25	–	–	98.8	98.4	–	–	–	0.4	–	–	–	–	–	–
						26–100	–	–	–	–	93.9	95.6	–	–	−1.7	–	–	–
Clalit, Israel	Rotem 2022 (abst)79	LN+100% ER+100% HER2−	OS0–7 years	n = 140	All meno	26–30	–	–	–	–	96.3	93.8	–	–	2.5	–	–	Not sig	–	–
NCDB	Abel 202281	LN1–3100% HR+100% HER2−	OS0–5 years	n = 21,370	Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–	p = 0.278	–	–	–
				n = 6356	Lobular	All ≤ 25	–	–	–	–	–	–	–	–	–	p = 0.532	–	–	–
				n = 4251	Age < 50 Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: 0.44 (0.22 to 0.86), p = 0.016	–	–	–
				n = 1062	Age < 50 Lobular	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: 0.54 (0.14 to 2.18), p = 0.39	–	–	–
NCDB (cont)	Cao 2022 (abst)82	LN1–3100% ER+100% HER2−	OSNR	n = 28,427	Age ≤ 50	All20–25	–	–	–	–	–	–	–	–	–	–	Unadj: 0.334 (NR), p = 0.002	–	–	–
					Age > 50	All20–25	–	–	–	–	–	–	–	–	–	–	Unadj: 0.521 (NR), p = 0.019	–	–	–
NCDB (cont)	Ibraheem 201971	LN1–3 : 97%LN4–9 : 3%100% HR+100% HER2−	OS0–5 years	n = 13,163	All meno	11–17	–	–	97.7	96.5	–	–	–	1.2	-	–	Adj: 0.63 (0.40 to 0.99), p = 0.044Threshold: RS > 13 sig CT benefit	–	–	–
						18–25	–	–	96.0	92.7	–	–	–	3.3	–	–	Adj: 0.53 (0.37 to 0.76), p = 0.001	–	–	–
						26–30					92.2	85.5			6.7			Adj: 0.50 (0.28 to 0.89), p = 0.018
				n = 3101	Age ≤ 50	All11–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.68 (0.35 to 1.32), p = 0.25	–	–	–
				n = 8886	Age > 50	All11–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.64 (0.47 to 0.86), p = 0.004	–	–	–
NCDB (cont)	Iorgulescu 201983	LN1–3100% ER+100% HER2−	OS0–5 years	n = 2735	All meno	18, 30	93	92	93.2	85.7	92.4	66.9	1.0	7.5	25.5	Unadj: p = 0.27Adj: 0.81 (0.33 to 1.98), p = 0.64	Unadj: p = 0.02Adj: 0.67 (0.35 to 1.27), p = 0.22	Unadj: p < 0.001Adj: 0.24 (0.13 to 0.47), p < 0.001	–	–
NCDB (cont)	Kumar 2023 (abst)84	LN1–3: > 90%LN4 +: < 10%100% HR+100% HER2−	OS0–10 years	n = 8628	Age ≤ 50	0–11	–	–	–	–	–	–	–	–	–	Adj: 0.56 (0.22 to 1.42)	–	–	–	–
						12–25	–	–	–	–	–	–	–	–	–	–	Adj: 0.55 (0.38 to 0.80)	–
						All ≤ 25	–	–	93.0	91.0	–	–	2.0	–	Unadj: 0.60 (0.48 to 0.75), p < 0.0001Adj: 0.54 (0.39 to 0.76), p = 0.0004	–	–	–
				n = 8628	Age 18–40	All ≤ 25	–	–	86.0	82.8	–	–	3.2	–		Adj: 0.43 (0.22 to 0.85)		–	–	–
				n = 8628	Age 40–50	All ≤ 25	–	–	94.7	92.2	–	–	2.5	–		Adj: 0.59 (0.39 to 0.87)		–	–	–
NCDB (cont)	Nash 202372	LN1–3100% HR+100% HER2−	OSNR, med FU 5.5 years	N = 4124	Age 40–50	All ≤ 25	–	–	–	–	–	–	–	–	–	Unadj: p = 0.41Adj: 0.72 (0.47 to 1.12), p = 0.15	–	–	–
						25–30	–	–	–	–	–	–	–	–	–	–		Unadj: p = 0.28
						31–50	–	–	–	–	–	–	–	–	––	–		Unadj: p = 0.002
																		Adj: 0.29 (0.10 to 0.85), p = 0.02
						> 50	–	–	–	–	–	–	–	–	–	–		Not sig (few events)
NCDB (cont)	Weiser 202273	LN1–3100% HR+100% HER2−	OS0–5 years	n = 16,646	All menoDuctal	11–25	–	–	96.7	95.1	–	–	–	1.6	–	–	Unadj: p = 0.004Adj: non-sig	–	–	–
				NR	Age < 50Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 2.32 (1.19 to 4.49)	–	–	–
				NR	Age 50–75Ductal	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 1.12 (0.86 to 1.46)	–	–	–
				n = 2691	All menoLobular	0–10	94.7	95.7	–	–	–	–	−1.0	–	–	Unadj: p = 0.888Adj: non–sig	–	–	–	–
						–11–25	–	–	96.6	94.9	–	–	–	1.7	–	–	Unadj: p = 0.381Adj: non-sig	–
NCDB (cont)	Weiser 202174	LN1–3100% HR+100% HER2−	OS0–5 years	n = 28,591	All meno	All ≤ 25	–	–	96.6	93.2	–	–	3.4	–	Unadj: p < 0.001Adj: 1.63 (1.28 to 2.07)	–	–	–
				NR	Age ≤ 50	All ≤ 25	–	–	–	–	–	–	1.4	–			Adj: 1.88 (1.05 to 3.37), p = 0.032	–	–	–
						12–17	–	–	–	–	–	–	–	1.3	–	–	Adj: 2.49 (0.80 to 7.76)	–	–	–
						18–25	–	–	–	–	–	–	–	4.4	–	–	Adj: 3.30 (1.38 to 7.84)	–	–	–
				NR	Age 51–70	All ≤ 25	–	–	–	–	–	–	1.6	–			Adj: 1.49 (1.12 to 1.97), p = 0.006	–	–	–
						12–17	–	–	–	–	–	–	–	3.6	–	–	Adj: 2.80 (1.45 to 5.24)	–	–	–
						18–25	–	–	–	–	–	–	–	3.2	–	–	Adj: 1.37 (0.92–2.05)	–	–	–
				NR	Age > 70	All ≤ 25	–	–	–	–	–	–	–	–	–		Adj: 1.1 (0.68 to 1.78), p = 0.69	–	–	–
				NR	Age ≤ 70	0–10	–	–	–	–	–	–	–	–	–	p = 0.44	–	–
						12–25	–	–	–	–	–	–	–	3.0	–	–	Adj: 1.91 (1.42 to 2.57)	–	–	–
						12–17	–	–	–	–	–	–	–	3.4	–	–	Adj: 3.04 (1.78 to 5.21), p < 0.001	–	–	–
						18–25	–	–	–	–	–	–	–	3.8	–	–	Adj: 2.02 (1.42 to 2.87), p < 0.001	–	–	–

_{Abs diff, absolute difference; adj, adjusted; CT, chemotherapy; int, intermediate; LN, lymph nodes (number positive); meno, menopausal; NR, not reported; prosp, prospective; pred, predictive of CT benefit; RS, Recurrence Score (Oncotype DX); sig, significant; unadj, unadjusted. aThe last column indicates whether interaction test (between risk group and CT use) indicates a significant predictive effect for CT benefit at the 5% level. Asterisk (*) denotes interaction adjusted for clinical factors.}

Tables in Appendix 8: Adjuvant chemotherapy infusion time by regimen

TABLE 55 Infusion time for each chemotherapy regimen included in the EAG model

Regimen	Doses per course	Infusion time per dose (hours)	Infusion time per course
FEC75 (6 cycles)
Fluorouracil 600 mg/m²	6	0.08	0.50
Epirubicin 75 mg/m²	6	0.08	0.50
Cyclophosphamide 600 mg/m²	6	0.50	3.00
FEC100 +T (3 + 3 cycles)
Fluorouracil 500 mg/m²	3	0.08	0.25
Epirubicin 100 mg/m²	3	0.08	0.25
Cyclophosphamide 500 mg/m²	3	0.50	1.50
Docetaxel 100 mg/m²	3	1.00	3.00
TC (4 cycles)
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Docetaxel 75 mg/m²	4	1.00	4.00
EC90/T75 (4 + 4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Docetaxel 75 mg/m²	4	0.50	2.00
EC90 (4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Accelerated EC90/P (4 + 4 cycles)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Paclitaxel 175 mg/m²	4	1.00	4.00
C-D (6 cycles)
Carboplatin AUC 6 (assumed 600 mg)	6	1.00	6.00
Docetaxel 75 mg/m²	6	1.00	6.00
TAC (6 cycles)
Docetaxel 75 mg/m²	6	1.00	6.00
Doxorubicin 50 mg/m²	6	0.08	0.50
Cyclophosphamide 500 mg/m²	6	0.08	0.50
Weekly P (12 weeks)
Paclitaxel 80 mg/m²	12	1.00	12.00
EC90/weekly P (4 cycles, 12 weeks)
Epirubicin 90 mg/m²	4	0.08	0.33
Cyclophosphamide 600 mg/m²	4	0.08	0.33
Paclitaxel 80 mg/m²	12	1.00	12.00

Source note · Summary · Home

Diagnosis binder

Files

Tumour profiling tests to guide adjuvant chemotherapy decisions in lymph node-positive early breast cancer: a systematic review and economic evaluation

Abstract

Background and definition of the decision problem

Condition and aetiology

Aetiology

Pathology

Prognosis

Epidemiology and incidence

Significance in terms of ill-health (burden of disease)

Current methods for staging of breast cancer

Management of early breast cancer

Surgical resection and neoadjuvant treatments

Adjuvant therapy planning

Endocrine therapy

Adjuvant chemotherapy

Bisphosphonates

Ovarian suppression

Adjuvant targeted therapy

Radiotherapy

Prognostic risk prediction tools

Nottingham Prognostic Index

Adjuvant! Online

PREDICT (Version 2.2)

The potential value of tumour profiling tests to guide chemotherapy decisions for women with lymph node-positive early breast cancer

Oncotype DX (Exact Sciences)

Prosigna (Veracyte)

EndoPredict (Myriad)

MammaPrint (Agendia)

Current usage of tumour profiling tests in the National Health Service

Description of decision problem

Interventions

Comparators

Population and important subgroups

Subgroups

Outcomes

Patient-reported outcomes:

Aims and objectives of the assessment

Clinical effectiveness

Overview of systematic review methodology

Population and subgroups

Interventions

Comparators

Outcomes

Date and language limits

Search strategy

Study selection and data extraction strategy

Quality assessment strategy

Methods of analysis/synthesis

Quantity and type of included studies

Prognostic ability: summary of evidence

Prediction of chemotherapy benefit: summary of evidence

Prospective randomised controlled trial of Oncotype DX: RxPONDER

Prospective randomised controlled trial of MammaPrint: MINDACT

Ongoing prospective randomised controlled trial of Prosigna: OPTIMA

Impact of test results on chemotherapy decisions: summary of evidence

Health-related quality of life and anxiety associated with use of the tests: summary of evidence

Risk of bias in included studies

Overview of prognostic data in this report

Summary of distribution of genomic risk groups and distant recurrence risk

Summary of prognostic ability across tests

Prognostic data from prospective randomised controlled trial of Oncotype DX (RxPONDER)

Prognostic data from prospective randomised controlled trial of MammaPrint (MINDACT)

Ongoing prospective randomised controlled trial of Prosigna: OPTIMA

Observational data: prospective use of Oncotype DX

Conclusions for prognostic data

Overview of predictive data in this report

Prediction of chemotherapy benefit: randomised controlled trial reanalysis (Oncotype DX)

Prediction of chemotherapy benefit: prospective randomised controlled trial of Oncotype DX (RxPONDER)

Prediction of chemotherapy benefit: reanalysis of cohort (MammaPrint)

Chemotherapy effect within groups: prospective randomised controlled trial of MammaPrint (MINDACT)

Effect of chemotherapy within recurrence score groups: registry data (Oncotype DX)

Effect of chemotherapy for older patients with recurrence score ≤ 25: registry data (Oncotype DX)

Conclusions for prediction of chemotherapy benefit data

Decision impact: overview and study characteristics

Decision impact results for Oncotype DX: all patients

Decision impact results for Oncotype DX: by risk group

Conclusions for decision impact data

Overview of data on health-related quality of life and anxiety