5 Committee discussion
5.1 The committee discussed current practice for making adjuvant chemotherapy prescribing decisions. The clinical experts explained that NHS clinical practice has changed since NICE's diagnostics guidance 10 was published in 2013. The PREDICT tool is now used by many NHS trusts rather than the Nottingham Prognostic Index (NPI). Adjuvant! Online is not currently available. The committee also heard that Oncotype DX is currently used in NHS clinical practice and may be used for a broader group than the population defined in the original diagnostics guidance 10, that is, people with oestrogen receptor (ER)-positive, human epidermal growth factor receptor 2 (HER2)-negative and lymph node (LN)‑negative early breast cancer who are assessed as being at intermediate risk using existing risk assessment tools.
5.2 The committee discussed the potential benefits of the tumour profiling tests for people with early breast cancer who are deciding whether to have adjuvant chemotherapy. It acknowledged that chemotherapy is an unpleasant treatment associated with short‑term physical, emotional and financial effects, and also long-term consequences such as infertility and increased risk of cardiomyopathy and leukaemia. The committee heard that there is potential benefit for people with cancer identified as being at low clinical risk, when test results suggest a high risk of distant recurrence. These people could potentially benefit from chemotherapy. It also heard that there is potential benefit for people with cancer categorised as high clinical risk, when test results suggest a low risk of distant recurrence. The committee heard that these people could decide not to have chemotherapy, therefore avoiding toxic side effects and effects on fertility. They could potentially resume normal daily activities earlier, although some may wish to have chemotherapy regardless of the test result. However, the committee noted that the claimed benefits of the tests depend on them having sufficient accuracy and discrimination to correctly classify risk and provide valid clinical information. The clinical experts explained that the additional clinical information provided by the tests may help people discuss further treatment options. This information is particularly helpful for people with cancers identified as intermediate clinical risk when the decision to offer chemotherapy is unclear. However, the final decision to recommend a course of adjuvant chemotherapy would always take into account the person's circumstances and preferences.
5.3 The committee considered the prognostic ability of the tumour profiling tests. It noted that for people with LN-negative disease, all the tests had statistically significant prognostic accuracy over clinical and pathological features or risk assessment tools such as the NPI. It also noted that for people with LN-positive disease, results for prognostic ability were more variable but all tests except IHC4+C showed statistically significant or borderline statistically significant prognostic ability over clinical and pathological features or risk assessment tools. The external assessment group (EAG) explained that there were concerns about bias in all studies reporting prognostic ability. This was because in many of the studies some or all patients had chemotherapy or patients who had not had chemotherapy were selected for analyses. Also, most studies excluded tumour samples with insufficient tissue or missing clinical and pathological data, and some studies included patients who had hormone receptor-negative or HER2-positive disease. The committee concluded that despite the potential spectrum bias, the evidence suggested that all the tumour profiling tests have the ability to predict the risk of distant recurrence in the population included in the assessment. It also concluded that the evidence was weaker in the group with LN-positive disease than in the group with LN-negative disease.
5.4 The committee considered the evidence on micrometastases. The EAG explained that 2 studies with Oncotype DX reported subgroup data on people with micrometastases, but no studies with the other tests reported such data. The EAG noted that in patients with micrometastases and a Recurrence Score result of less than 18, outcomes were more similar to those in patients with LN-negative disease than in those with LN-positive disease. However, in patients with micrometastases and a Recurrence Score result of more than 30, outcomes were more similar to those in patients with LN-positive disease. Results were variable in patients with Recurrence Score results between 18 and 30. The EAG noted that the data were uncertain because of the high risk of confounding. The clinical experts explained that micrometastatic disease is classified as LN-positive disease but treated as LN-negative disease for clinical and shared decision making. In clinical practice some centres send samples from patients with micrometastases for Oncotype DX testing, but others do not. A clinical expert also explained that for patients with micrometastatic disease who have reasons to avoid chemotherapy, such as being older or having comorbidities, tumour profiling tests would be helpful. The committee noted that the ongoing OPTIMA study in LN-positive disease excludes patients with micrometastases, unless the tumour size is 20 mm or more. The EAG reviewed whether all studies in the diagnostics assessment report included or excluded patients with micrometastatic disease. It found that in TransATAC micrometastases were not assessed and therefore the disease was treated as LN-negative, but other studies did not report whether patients with micrometastatic disease were included or not. A company representative explained that in the MINDACT study, micrometastases measuring 0.2 mm to 2 mm were classified as LN-positive and isolated tumour cells were classified as LN‑negative. The clinical experts judged that, on balance, patients with micrometastases were likely to have been included in the studies as having LN-negative disease. The committee concluded that tumour profiling tests should be available as an option for people fulfilling the recommendation requirements and who have micrometastatic disease. Discussion within the multidisciplinary team may be particularly helpful for this group.
5.5 The committee considered the evidence on whether the tumour profiling tests can predict relative treatment effects associated with chemotherapy. The clinical experts stated that it is likely some patients could have a greater relative treatment effect from chemotherapy than others, for example, patients with hormone receptor-positive cancer that is not sensitive to endocrine therapy, but evidence is not available to support this. The EAG explained that the only evidence available to show a relative treatment effect for chemotherapy across different risk groups was for Oncotype DX, and the evidence included in the diagnostics assessment report was weak because it was at high risk of bias from potential confounding. The results of interaction tests (which show whether the tumour profiling test was able to predict a different treatment effect by risk group) in the adjusted analysis in the B20 study by Paik et al. (2006; LN-negative disease) remained statistically significant when adjusting simultaneously for clinical and pathological variables. However, the EAG also explained that the difference in relative treatment effects for chemotherapy in the B20 study may be overestimated because this was the Oncotype DX derivation data set. In the SWOG-8814 study by Albain et al. (2010; LN-positive disease) the results of the interaction tests remained statistically significant when adjusting for some individual clinical and pathological variables, but there was no analysis that adjusted for these simultaneously, and the test was non-significant when adjusted for Allred-quantified ER status. The clinical experts explained that hormone receptor status may also predict relative treatment effects for chemotherapy. The committee considered that if all known clinical and pathological variables were included in the analyses of SWOG-8814 then it was likely that the results of the interaction test would no longer be statistically significant. This suggested highly uncertain relative treatment effects for chemotherapy according to the results of the tumour profiling tests for this group with LN-positive disease. The committee concluded that the evidence on the extent to which tumour profiling tests are able to predict relative treatment effects for chemotherapy is highly uncertain, but there may be some differences between Oncotype DX risk groups. The committee noted that no data were available to assess a difference in relative treatment effects for chemotherapy for EndoPredict, IHC4+C and Prosigna risk groups. However, it considered that it would be unethical to do a randomised controlled trial looking at the benefit of chemotherapy compared with endocrine therapy in patients with a clinically low or high risk of distant recurrence. It also noted that data on MammaPrint suggested no difference in relative treatment effects for chemotherapy.
5.6 The committee considered the evidence on clinical utility, that is, data from studies which assessed the ability of the tumour profiling tests to affect patient outcomes. It discussed the recently published results from TAILORx on Oncotype DX (see section 4.31), which showed that across all patients with Recurrence Score results of 11 to 25, there were no clinically relevant or statistically significant differences between those who had endocrine therapy alone and those who had chemotherapy plus endocrine therapy. The EAG noted that some subgroups, such as those with Recurrence Score results of 21 to 25 and those aged 50 or under, had results with confidence intervals above the non-inferiority margin, which suggested that there could be a clinically relevant difference in these subgroups. The EAG noted that patients included in TAILORx had hormone receptor-positive, HER2-negative and LN‑negative disease, and met National Comprehensive Cancer Network guidelines for recommendation or consideration of chemotherapy. In the group with Recurrence Score results of 11 to 25, 73% to 74% were clinically low risk according to modified Adjuvant! Online. The committee acknowledged that the results from TAILORx may not be generalisable to clinical practice in the UK because the population who had chemotherapy in the study would not be routinely offered chemotherapy in an NHS pathway. It noted that, to fully understand the implications of the study for UK practice, a subgroup analysis of the TAILORx data would be needed investigating the performance of the test in predicting chemotherapy benefit for patients eligible for chemotherapy in the UK. The EAG explained that this had been requested but was not made available. The committee concluded that in principle TAILORx is an important piece of evidence showing the effectiveness of gene profiling to guide adjuvant chemotherapy decisions in breast cancer. But it is uncertain how applicable it is to people with breast cancer in the UK who are considering adjuvant chemotherapy treatment.
5.7 The committee considered the evidence on clinical utility for the other tests. It noted that the only other test with evidence from randomised controlled trials was MammaPrint (the MINDACT study). The committee noted that none of the other tumour profiling tests (EndoPredict, IHC4+C and Prosigna) had similar evidence of clinical utility, but it was aware that this evidence was being collected for Prosigna (see section 5.27). The committee noted that MINDACT (see section 4.32) was a well-designed study. The results suggested that patients with high clinical risk and MammaPrint low-risk scores can forgo chemotherapy without a statistically significant increase in the 5-year risk of distant recurrence. However, a clinical expert explained that the risk of recurrence often continues beyond 5 years and noted that the MINDACT authors (Cardoso et al. 2016) stated that long-term follow-up and outcome data will be essential. These data are being collected and a 10-year follow-up analysis is planned. The committee concluded that none of these tests had strong enough evidence to demonstrate an effect on subsequent patient outcomes.
5.8 The committee was encouraged by the availability of the data set provided in confidence to NICE by Genomic Health. The data set was based on the access scheme operated by NHS England, which provided real world evidence on the use of adjuvant chemotherapy in the NHS following testing with Oncotype DX for the population included in the scope for this assessment. The committee noted that the total number of patients in the data set appeared to be much larger than the number of patients with complete data in the population of interest, and that the advice from clinical experts (see section 5.1) was that the test had been used on a wider group of patients in practice. The committee also noted that the publication on TAILORx (Sparano et al. 2018) may influence chemotherapy decision making in people with a Recurrence Score result of 11 to 25, and therefore the data set may not represent clinical decision making in this group. The committee concluded that the access scheme data set was an important piece of real world evidence for use in the economic model, but that more complete data could have been collected and reported, and that it will be important to continue the data collection to capture the influence of TAILORx. It also concluded that future data collection should be done as part of a national database, rather than by individual companies, to increase transparency and enable it to be linked to outcome data (see section 5.29).
5.9 The committee discussed the analytical validity of IHC4+C. The EAG explained that the evidence has developed since diagnostics guidance 10 was published. The committee noted that the data showed good correlation between different centres when scoring and staining were assessed separately for measurement of the Ki67 marker, which had been achieved with training. But it also noted that when studies looked at staining and scoring combined, the correlation between centres decreased substantially. A clinical expert noted that different antibody clones are available for testing Ki67, ER and progesterone receptor (PR) status. Different studies used different antibody clones which means that the studies are not directly comparable. The committee heard that different methods of assessing ER and PR receptors may be needed for IHC4+C compared with those already used routinely, which may introduce additional complexity. The committee concluded that because of these issues, the reproducibility of IHC4+C was poor. It also concluded that if this test were to be developed further, the antibody clones used in the assays for ER, PR and Ki67 should be specified, and there would need to be substantial investment in staff training and quality assurance.
5.10 The committee discussed the assumptions and inputs used in the model, and considered the extensive stakeholder comments on the model and the EAG responses to these comments. It noted that a specific analysis of the TransATAC data was used for risk classification probabilities and for distant recurrence rates based on test result for Oncotype DX, EndoPredict (EPclin), Prosigna and IHC4+C. The results from this specific analysis of the data set have now been published (Sestak et al. 2018). The EAG explained that this data source was chosen because it included data on 4 of the 5 tests of interest and was specific to the population included in the scope (patients with hormone receptor-positive, HER2-negative disease). The committee heard that although the TransATAC data were slightly older and some patients were not candidates for chemotherapy, the patient characteristics matched well with the more recent MINDACT study. The alternative would be to use different data sources for each test, which would have introduced additional uncertainty and complexity. Also, the group with LN‑negative disease could not have been split according to level of clinical risk. The EAG described the limitations of using data from the TAILORx study (Sparano et al. 2018) for the health economic analysis. It also explained that the distant recurrence-free rates from the TransATAC analysis used in the model were consistent with results from other studies (B14, B20, TAILORx, MD Anderson, Clalit, Memorial Sloan Kettering, SEER and WSG PlanB) both when grouped separately by clinical risk and when all clinical risk groups were pooled together. The committee concluded that the TransATAC analysis had some limitations, but was the best available data for use in the model.
5.11 The committee considered the data on pre- and post-test chemotherapy decisions used in the model. The EAG explained that for 3-level tests (tests with low, intermediate and high-risk categories [IHC4+C, Oncotype DX, Prosigna]), data on pre- and post-test chemotherapy decisions for the group with LN-negative disease and a NPI of more than 3.4 were taken from the Genomic Health access scheme data set (see section 5.8). For other clinical risk subgroups with the 3-level tests, and for all clinical risk subgroups with 2-level tests (tests with low and high-risk categories; EndoPredict, MammaPrint), data on pre-test chemotherapy decisions were taken from different sources to data on post-test chemotherapy decisions. There were also very limited UK data for these groups. The committee considered the modelled impact of these data on chemotherapy use, and noted that although clinical and patient experts thought that the main benefit of the tests was in avoiding unnecessary chemotherapy, most tests were estimated to increase chemotherapy use at least in some subgroups (see section 4.49). The committee concluded that there was much more uncertainty around chemotherapy decision making for the 2-level tests, and for the subgroups who were not included in the original NICE recommendation on tumour profiling tests (LN‑negative disease and a NPI of 3.4 or less, and LN-positive disease).
5.12 The committee considered how adjuvant chemotherapy treatment effects had been applied in the economic model, particularly the relative treatment effects of chemotherapy between the risk groups predicted by the tumour profiling tests. It noted its earlier conclusion that the evidence on whether tumour profiling tests can predict relative treatment effects for chemotherapy is highly uncertain, but that there may be some differences between Oncotype DX risk groups (see section 5.5). It agreed that for EndoPredict, IHC4+C and Prosigna, no evidence was available to show a difference in relative treatment effects of chemotherapy across risk groups, and that data on MammaPrint suggested no difference in relative treatment effects. Therefore for these tests it was appropriate to assume the same relative risk of distant recurrence across all test risk categories. The EAG noted that a relative risk of distant recurrence for chemotherapy compared with no chemotherapy of 0.76 estimated from data reported in a large meta-analysis by the Early Breast Cancer Trialists' Collaborative Group was used in the base case, and that this value had been varied between 0.6 and 0.9 in sensitivity analyses. The committee acknowledged that the ICERs were sensitive to this assumption, increasing as the relative risk moved from 0.6 to 0.9. It concluded that, although the true treatment effect is unknown, the relative risk was unlikely to be 0.9 or more.
5.13 The committee considered stakeholder comments submitted during the first consultation suggesting that Oncotype DX has the ability to predict which patients have disease that will respond to chemotherapy. The EAG noted that in response to the comments it had done additional exploratory analyses for Oncotype DX to show the impact on the incremental cost-effectiveness ratios (ICERs) if a smaller relative treatment effect than that taken from the B20 study (Paik et al. 2006) was applied in the model in the group with LN‑negative disease and a NPI of more than 3.4 (see section 4.51). The EAG noted that the hazard ratios used in these analyses were from comparisons of independent arms of trials and were therefore very uncertain. The EAG also said that using hazard ratios calculated from the B20 and the B14 (Paik et al. 2004) studies resulted in an ICER of around £24,000 per quality-adjusted life year (QALY) gained for Oncotype DX compared with current practice. Using hazard ratios calculated from the B20 and TransATAC studies resulted in an ICER of around £8,000 per QALY gained. Based on the results of the Sparano et al. 2018 publication on TAILORx, the EAG repeated the analysis incorporating an additional assumption of 0 chemotherapy benefit for patients in the Oncotype DX low Recurrence Score result category. It noted that this analysis was based on the strong assumption that Oncotype DX not only identifies patients who will not relapse, but also identifies patients who will relapse but will not respond to chemotherapy. When this assumption was included in the analysis using B20, the analysis using B20 and B14, and the analysis using B20 and TransATAC, the ICERs were below £4,000 per QALY gained. The committee concluded that although these analyses were associated with considerable uncertainty, they gave an indication of Oncotype DX's likely cost effectiveness if the relative treatment effects for chemotherapy did differ between Oncotype DX risk groups, but not to the extent reported in the Paik et al. (2006) study.
5.14 The committee considered stakeholder comments submitted during the first consultation suggesting that chemotherapy adverse events had not been adequately captured in the economic model; in particular, congestive heart failure, permanent hair loss and peripheral neuropathy. The EAG noted that in response to the comments it had done additional exploratory analyses to include these adverse events in the model. Congestive heart failure was added into the model by incorporating estimated lifetime QALY losses and costs taken from an alternative model (Hall et al. 2017). Hair loss and peripheral neuropathy were incorporated using a disutility applied to a proportion of the population for the lifetime of the model. The EAG highlighted the considerable limitations of these analyses, and noted that for tests that increased chemotherapy use in some subgroups, the ICERs became less favourable. The committee noted that including additional adverse events in the model did reduce some of the ICERs, but not enough to change the conclusions. It also noted a further EAG analysis, which suggested that for tests that reduced chemotherapy use but were not cost effective, the QALY gain from avoiding adverse events would have to be in the range of 1.1 to 1.3 to result in cost‑effective ICERs. The committee concluded that it was important to consider potential adverse events that could be caused by chemotherapy. However, the reduction in adverse events from reduced chemotherapy use, although beneficial for patients, was unlikely to affect its conclusions on the cost effectiveness of the tumour profiling tests based on the EAG's analysis.
5.15 The committee considered other assumptions used in the model such as the cost of chemotherapy and how the risk of distant recurrence was applied over time. The EAG explained that there was some uncertainty around these inputs, but all had been tested in sensitivity analyses. The committee concluded that the assumptions and inputs used in the model were reasonable, but they were associated with considerable uncertainty because of the limitations in the data that underpinned them.
5.16 The committee noted its discussion on current practice (see section 5.1) and considered the absence of comparisons of the tumour profiling tests with the PREDICT tool. The EAG explained that in the model it was not possible to compare the tumour profiling tests with PREDICT, or to define the clinical risk groups using PREDICT, because relevant data were not available. The committee noted that the comparisons in the model did not fully reflect current NHS clinical practice, which led to uncaptured uncertainty in the model results. The committee concluded that research on tumour profiling tests should include comparisons with PREDICT (see section 5.26) so that the cost effectiveness of the tests relative to current practice can be fully assessed in future.
5.17 The committee considered the subgroups that were included in the model, that is, people with LN-negative disease and a NPI of 3.4 or less, people with LN-negative disease and a NPI of more then 3.4, and people with LN-positive disease. It noted its earlier conclusion that the evidence suggested that all the tumour profiling tests have the ability to predict risk of distant recurrence (prognosis), but this ability was less certain in the group with LN-positive disease (see section 5.3). The committee also recalled that the test results were particularly helpful for people with cancers identified as intermediate clinical risk when the decision to offer chemotherapy is unclear (see section 5.2). The clinical experts explained that tumour profiling tests were also helpful for people with LN-positive cancer who have comorbidities and therefore an additional reason to want to avoid chemotherapy. The EAG noted that this subgroup of the LN-positive population could not be modelled because of a lack of data. In addition, the committee noted that the EAG's systematic review had highlighted substantial lack of agreement between the tests in risk categorising the group with LN-positive disease. The committee decided to consider the ICERs in the group with LN-negative disease only, but noted that further studies would be helpful to assess the clinical effectiveness of the tests in the group with LN-positive disease (see section 5.27).
5.18 The committee considered the results from the model. It noted that the differences in the QALYs were small, and that the ICERs for all tumour profiling tests were highly uncertain because of the available clinical data and the assumptions used in the modelling (see section 5.10 to section 5.15). It also noted that the base-case ICERs for many of the tumour profiling tests were higher than those normally considered to be cost effective. However, it heard that access proposals had been made by Myriad Genetics (for EndoPredict) and NanoString Technologies (for Prosigna). Genomic Health confirmed that the confidential discount for Oncotype DX would continue in the NHS. The committee concluded that the availability of the access proposals for EndoPredict and Prosigna may reduce the ICERs to a range that could be considered plausibly cost effective despite the clinical uncertainties.
5.19 The committee considered the EndoPredict and Prosigna access proposals. Compared with current practice, the ICERs for EndoPredict and Prosigna in the group with LN-negative disease and a NPI of 3.4 or less were still higher than those normally considered to be a cost-effective use of NHS resources. In the group with LN-negative disease and a NPI of more than 3.4, Prosigna compared with current practice had an ICER of less than £20,000 per QALY gained, and therefore could be considered cost effective. In the same group, EndoPredict compared with current practice had ICERs between £20,000 and £30,000 per QALY gained, which varied depending on whether the testing was done at a local or a central laboratory. The committee noted that local testing was more cost effective than central testing, and that testing became more cost effective as test throughput increased. It also recalled its conclusion that the data on post-chemotherapy decisions were more uncertain for 2-level tests than for 3-level tests (see section 5.11), and noted that the EAG's sensitivity analyses using plausible alternative sources for post-chemotherapy decisions resulted in ICERs that were lower than £20,000 per QALY gained. The committee noted that in sensitivity analyses, when the relative risk of distant recurrence for chemotherapy compared with no chemotherapy was changed to 0.9 from the base‑case value of 0.76, the ICERs increased for both EndoPredict and Prosigna to more than £30,000 per QALY gained. It considered that a relative risk of 0.9 or more across all genomic risk groups was unlikely, but accepted the uncertainty around this parameter (see section 5.12). The committee decided that although there is uncertainty around the ICERs for EndoPredict compared with current practice, sensitivity analyses suggested that the ICER will be around £20,000 per QALY gained, and therefore it could be considered cost effective. The committee concluded that EndoPredict (EPclin) and Prosigna, when provided at the costs stated in the access proposals, were likely to be cost effective in the group with LN-negative disease and a NPI of more than 3.4, but evidence on clinical outcomes will be important to confirm this (see section 5.29).
5.20 The committee considered the ICERs for Oncotype DX compared with current practice. It heard that the proposed confidential test cost for Oncotype DX was the same as in current NHS practice, and that this cost had been used in the EAG's economic model. It noted that compared with current practice, the ICERs for Oncotype DX in the group with LN-negative disease and a NPI of 3.4 or less were higher than those normally considered to be a cost-effective use of NHS resources. In the group with LN-negative disease and a NPI of more than 3.4, the committee noted that in the base-case analyses Oncotype DX was dominated by the comparator. The committee recalled its earlier conclusions; Oncotype DX may be able to predict relative treatment effects for chemotherapy, and the ICERs for Oncotype DX compared with current practice when some relative treatment effect across different risk groups was applied in the model were most likely to be between £2,000 and £25,000 per QALY gained (see section 5.5 and section 5.13). However, it noted that this was very uncertain. The committee concluded that Oncotype DX, when provided at the test cost stated in the access proposal, was likely to be cost effective in the group with LN-negative disease and a NPI of more than 3.4, but evidence on clinical outcomes will be important to confirm this (see section 5.29).
5.21 The committee considered how EndoPredict, Oncotype DX and Prosigna compare with each other. It noted that only pairwise ICERs of each tumour profiling test compared with no testing had been presented, rather than a fully incremental analysis. The EAG explained that a fully incremental analysis could not be done because there was no clinical evidence which directly compared the tests. The committee noted that since the publication of TAILORx (Sparano et al. 2018) evidence on clinical utility was strongest for Oncotype DX. It also noted that it was not possible to determine which test was the most cost-effective use of NHS resources, and that it may not be the test with the lowest acquisition price.
5.22 The committee considered the ICERs for MammaPrint compared with modified Adjuvant! Online. It noted that in the base-case analyses, MammaPrint was dominated by the comparator in the modified Adjuvant! Online high-risk subgroup. In the modified Adjuvant! Online low-risk subgroup, the ICERs were much higher than those normally considered to be cost effective. The committee concluded that MammaPrint would not be a cost-effective use of NHS resources.
5.23 The committee considered the ICERs for IHC4+C compared with current practice. It noted that the ICERs were low or that IHC4+C dominated current practice in all subgroups. The committee felt that the test cost had been underestimated because it did not include any costs for training or for setting up a quality assurance programme. But even if these costs were included, IHC4+C may still be cost effective. However, the committee noted its earlier conclusion on the analytical validity of IHC4+C (see section 5.9) and concluded that it could not be recommended for use in the NHS until issues around reproducibility and implementation had been resolved.
5.24 The committee noted that the model for EndoPredict, IHC4+C, Oncotype DX and Prosigna related only to a postmenopausal population because TransATAC was used as the data source for these tests. It considered whether the model results could also apply to a premenopausal population. A clinical expert explained that the biology of a cancer and its molecular subtype, for example hormone receptor status and HER2 status, is more influential in determining the risk of distant recurrence than menopausal status. Therefore the committee concluded that the model results apply to premenopausal and postmenopausal populations, but noted that clinicians wishing to use a tumour profiling test should first check which populations the test is indicated for (see section 3).
5.25 The committee discussed the generalisability of the data to men. It acknowledged that men make up a small proportion of people with breast cancer. The committee noted that all the clinical and economic evidence was based on trials with women, but that the general subtypes of breast cancer are identical in men and women, and in clinical practice men would have treatment in the same way as women. The committee concluded that the recommendations in this guidance should also apply to men.
5.26 The committee noted that there are several ongoing studies which will provide evidence of long-term patient outcomes: further data collection from the MINDACT study on MammaPrint and the OPTIMA trial on Prosigna. The committee concluded that these studies are relevant to this assessment and data from them may be important when the guidance is considered for updating in the future. It also recalled its earlier conclusion that a subgroup analysis of TAILORx would be welcomed (see section 5.6). But it noted that not all studies would provide UK-specific data and comparisons with the PREDICT tool, which would be important for future updates to fully assess the cost effectiveness of the tests compared with current practice.
5.27 The committee also recalled its previous conclusion on the potential utility of the tests in the group with LN-positive disease (see section 5.17), particularly for people who have comorbidities and who may be particularly affected by the side effects of adjuvant chemotherapy. It noted that further research in this group would be welcome and heard from clinical experts that the ongoing OPTIMA trial may help to reduce some of the uncertainties identified during this assessment.
5.28 The committee considered consultation comments from the Cancer and Society in the 21st century research team about their qualitative research on women's experiences of gene expression profiling for chemotherapy decision making, and noted the importance of this work.
5.29 The committee recalled its previous conclusions on the uncertainties associated with both the clinical and cost effectiveness of EndoPredict (EPclin), Oncotype DX and Prosigna. It had identified clinical uncertainties associated with the effect of the technologies on patient outcomes (see section 5.6 and 5.7) and also on clinical decision making (see section 5.8 and section 5.11). These limitations meant that the estimated cost effectiveness of the technologies in the NHS was highly uncertain (see section 5.18 to section 5.20). On balance the committee concluded that EndoPredict (EPclin), Oncotype DX and Prosigna, when provided at the test cost stated in the access proposal, were likely to be cost effective in the group with LN-negative disease and a NPI of more than 3.4, but evidence on clinical outcomes will be needed to confirm this in the NHS. Further, it considered that this should be addressed through data collection using the National Cancer Registration and Analysis Service which would provide data on NHS use. It also believed that it is necessary that data is collected as part of a national database, rather than by individual companies, to increase transparency, enable the data to be linked to clinical outcomes and ensure evidence is available that can be considered in future updates of this guidance. It therefore decided that its recommendations for EndoPredict (EPclin), Oncotype DX and Prosigna are conditional on data collection arrangements agreed with NICE being put in place. It is anticipated that arrangements will be made to collect timely and complete record-level test data, which can be submitted to the National Cancer Registration and Analysis Service, with the aim of linking test data to chemotherapy use, recurrence and survival outcomes.