Tumour profiling tests to guide adjuvant chemotherapy decisions in early breast cancer

The prognostic ability of a test describes its ability to differentiate between people who will have good or poor outcomes. There was some evidence that all the tests had statistically significant prognostic ability for distant recurrence within 10 years. Clinical experts noted that the absolute benefit of chemotherapy is dependent on the absolute level of risk, so people with low risk of recurrence will have a lower absolute benefit from chemotherapy than people with a high risk of recurrence. So, tests with prognostic ability are useful to help guide chemotherapy decisions even if they are unable to predict chemotherapy benefit.

The predictive ability of a test is determined by whether the relative effect of chemotherapy versus no chemotherapy on clinical outcomes differs between risk groups or ranges. The EAG did not identify any evidence on the predictive ability of EndoPredict or Prosigna in a population that was mostly people with LN‑positive breast cancer. So, these tests could only be considered to have prognostic ability.

The committee concluded that there was not enough evidence to definitively say whether or not MammaPrint is predictive of chemotherapy benefit. Two studies were available that evaluated the predictive ability of MammaPrint. A reanalysis of 2 cohorts reported by Mook et al. (2009) did not find a significant interaction between MammaPrint result and the effect of chemotherapy on breast cancer-specific survival. In the MINDACT trial, people were randomised to chemotherapy or no chemotherapy only if their clinical risk and MammaPrint risk were discordant. Everyone in the clinical high-risk and MammaPrint high-risk group was offered chemotherapy, so the EAG stated that it was not possible to determine whether MammaPrint was predictive of chemotherapy benefit from MINDACT data. The company stated that MINDACT was designed to demonstrate lack of chemotherapy benefit in the clinical high-risk, MammaPrint low-risk group and that chemotherapy benefit on distant recurrence in this group was non-significant (hazard ratio [HR] 0.84, 95% confidence interval [CI] 0.51 to 1.37). So, people in this group would have low or no benefit from chemotherapy. The EAG responded that a non-significant benefit is not necessarily the same as no benefit, and that there was no evidence to estimate the size of chemotherapy benefit in the MammaPrint high-risk group. A company representative noted that MINDACT was not powered to detect differences in chemotherapy benefit by LN status. So, the predictive ability of MammaPrint remained uncertain.

The committee thought that it was likely that Oncotype DX has some predictive ability for chemotherapy benefit in postmenopausal women with oestrogen receptor (ER)- or progesterone receptor (PR)-positive, HER2-negative early breast cancer with 1 to 3 positive lymph nodes. There were 2 studies that provided evidence on the predictive ability of Oncotype DX for chemotherapy benefit. A reanalysis of the SWOG-8814 study found a difference in chemotherapy benefit on disease-free survival between people with a recurrence score (RS) of 0 to 17, 18 to 30 and 31 to 100 over 10 years. The statistical significance of this difference was dependent on which factors were adjusted for in the analysis. The RxPONDER trial used Oncotype DX for screening and excluded people with RS 26 to 100, because they were likely to benefit from chemotherapy based on the results of SWOG-8814. So, it was not possible to determine whether Oncotype DX was predictive of chemotherapy benefit with a cutoff of RS 25 using results from RxPONDER alone. People with RS 0 to 25 were randomised to chemotherapy or no chemotherapy. Within this group, the hazard ratio for the effect of chemotherapy was non-significant but favoured no chemotherapy for postmenopausal women (HR 1.12, 95% CI 0.82 to 1.52). A clinical expert noted that RxPONDER was not powered as a non-inferiority trial, so this finding could be considered uncertain.

The EAG noted that in SWOG-8814, 38% of people had 4 or more positive lymph nodes and 12% had HER2-positive cancer, so the population did not exactly match the scope, and the effect of chemotherapy may be overestimated. In RxPONDER, the overall clinical risk was relatively low for a population with 1 to 3 positive lymph nodes. The EAG noted that of the people randomised, 65% had 1 positive lymph node, 25% 2 positive nodes and 9% 3 positive nodes. Some people had micrometastases and 24% had low-grade cancer. Additionally, the EAG identified possible selection bias because people had their test result before agreeing to randomisation, and there was some crossover between trial arms. Therefore, the results may underestimate the effect of chemotherapy in a wider population with LN-positive breast cancer with higher overall clinical risk. The company highlighted a subgroup analysis of RxPONDER which indicated that the effect of chemotherapy on invasive disease-free survival was non-significant in people with 1 positive node and in people with 2 or 3 positive nodes. The committee felt that the evidence was uncertain, but commented that it is unlikely that more evidence will be generated in this population to reduce this uncertainty. It concluded that the results of SWOG-8814 and RxPONDER together suggest it is likely that Oncotype DX is predictive of chemotherapy benefit in postmenopausal women with ER- or PR-positive, HER2-negative early breast cancer with 1 to 3 positive lymph nodes.

The committee heard that it is likely that chemotherapy is more effective in premenopausal women than in postmenopausal women because of the effect of chemotherapy on ovarian function. The RxPONDER trial reported a significant benefit of chemotherapy on 5-year distant recurrence in premenopausal women with RS 0 to 25, but not in postmenopausal women (see section 3.9). Clinical experts stated that chemotherapy can suppress ovarian function in premenopausal women and this may be responsible for the treatment effects seen. So, the committee concluded that Oncotype DX should not be used to help guide chemotherapy decisions for premenopausal women with LN-positive early breast cancer.

The committee noted that the tests measure the expression of different genes (see sections 2.10 to 2.25). So, the effects of menopausal status seen in studies of Oncotype DX may not be generalisable to the other tumour profiling tests. There was limited data for other tests stratified by menopausal status. Most of the evidence for EndoPredict was in postmenopausal women. Data from MINDACT for MammaPrint was in a mixed population, in which 33% were premenopausal women. Prosigna is only intended for postmenopausal women. Clinical experts agreed that there is a plausible biological explanation for the difference in chemotherapy benefit between pre- and postmenopausal women (see section 3.11). They also highlighted that the risk of incorrectly foregoing chemotherapy is higher in premenopausal women than in postmenopausal women. So, the committee concluded that EndoPredict and MammaPrint should not be used to help guide chemotherapy decisions in premenopausal women, and that Prosigna is not intended for use in this population.

The committee concluded that evidence on how test results affected chemotherapy decisions could reasonably be generalised between tumour profiling tests. The only available evidence on how test results influenced chemotherapy recommendations or decisions in a LN-positive population was for Oncotype DX. Clinical experts stated that the decision-making process is similar regardless of test identity, and that there are many other factors that can affect decisions on chemotherapy in addition to tumour profiling test results (see section 2.5). The committee recalled that different tests measure the expression of different genes, but that there is some overlap between tests. However, the way different tests define risk groups resulted in large differences in the number of people who would be assigned as having low, intermediate or high risk, even between tests with the same number of risk categories (see section 3.22), so there was some uncertainty. The committee concluded that it was reasonable to use the available data from studies of Oncotype DX to evaluate the other tests, because alternative methods would be more uncertain. More evidence on how the results of tests other than Oncotype DX affect chemotherapy decision making for people with LN-positive early breast cancer would be helpful to reduce uncertainty (see section 3.22).

The committee concluded that more evidence was needed on how well tumour profiling tests can predict risk in different ethnic groups (see section 4.1). There was not enough evidence to say whether the ability of tumour profiling tests to predict risk may differ across ethnic groups. In RxPONDER, differences in 5-year invasive disease-free survival within the RS 0 to 25 group were reported according to ethnicity (White, 92%; Black, 87%; Asian, 94%), but no prognostic or predictive data were reported. In a subgroup analysis of the SEER database, the prognostic ability of Oncotype DX was only statistically significant in White participants, but this was based on small numbers. There was no evidence stratified by ethnicity for other tumour profiling tests.

There was very little data on using tumour profiling tests for men with breast cancer. Clinical experts stated that ER- or PR-positive, HER2-negative, LN-positive breast cancer generally responds to treatment with chemotherapy in the same way for men and women. But it is unclear whether tumour profiling tests would have the same prognostic and predictive ability for men and women. No data was identified on using tumour profiling tests for trans, non-binary or intersex people. It is likely that decisions on adjuvant chemotherapy in these populations would be individualised to the person considering their hormonal profile, their circumstances and any gender-affirming treatment, in addition to the factors described in section 2.5. Tumour profiling tests may be used if clinical judgement determines them to be suitable for the person and as part of the shared decision-making process. Further research is needed on the prognostic and predictive ability of tumour profiling tests in these populations (see section 4.1).

The committee concluded that EndoPredict was likely to be a cost-effective use of NHS resources when used to guide adjuvant chemotherapy decision making with postmenopausal women. In the EAG's base-case analysis, the probabilistic incremental cost-effectiveness ratio (ICER) for EndoPredict compared with decision making without tumour profiling was £4,113 per quality-adjusted life year (QALY) gained. This result was driven by a small decrease in the number of people having chemotherapy and in the number of people with distant recurrence, and additional costs of testing. With confidential price discounts applied, EndoPredict dominated standard care (it cost less and produced more QALYs). The committee considered a scenario analysis that used an alternative source for test risk classification probabilities and distant recurrence-free interval (Filipits et al. 2019). In this scenario, EndoPredict was instead dominated by standard care (it cost more and produced fewer QALYs). The EAG noted that the data from Filipits et al. was taken at 15 years, when there were few people remaining in the study. The data used in the base-case analysis (from the TransATAC study) was taken at 10 years. Ten-year data from Filipits et al. produced similar results to the base case. Clinical experts explained that the benefit of chemotherapy is primarily in the first few years, and that the risk of recurrence after 10 years is less likely to be affected by chemotherapy. So, the committee preferred the base-case analysis. The committee recalled that there was no data on how EndoPredict results would affect chemotherapy decisions for people with LN-positive breast cancer, and that interpretation of results may not be generalisable between tests (see section 3.13). It noted that scenario analyses using data from different studies of Oncotype DX to inform impact on chemotherapy decisions did not have a large effect on the cost-effectiveness estimates for EndoPredict. So, although this parameter was uncertain, the committee felt that the conclusions of the economic model were likely to be robust.

The committee concluded that MammaPrint was unlikely to be a cost-effective use of NHS resources. In the EAG's base-case analysis, MammaPrint was considered to have prognostic ability only, and was dominated by standard care when used for a population with clinical high risk and mixed menopausal status (33% premenopausal). This result was driven by an increase in the number of people who developed distant recurrence as well as the additional costs of testing, even though there was a large decrease in the number of people having chemotherapy. The company stated that results from the MINDACT trial had been misrepresented, and that if MammaPrint was considered to have predictive ability then it would dominate standard care. The committee recalled that it was uncertain whether MammaPrint was predictive of chemotherapy benefit (see section 3.8). Clinical experts commented that the data from the 33% of premenopausal women in the MINDACT cohort may be obscuring a smaller effect of chemotherapy on distant recurrence in postmenopausal women (see sections 3.11 and 3.12). The company submitted data from an exploratory subgroup analysis of MINDACT participants with 1 to 3 lymph nodes aged over 50 (n=430), and suggested that the effect of chemotherapy on distant recurrence in this group could be used in the economic model (HR 0.88, 95% CI 0.46 to 1.68). The committee considered this analysis, and noted that the confidence interval included the value used in the EAG's base case (HR 0.71) which was derived from the Early Breast Cancer Trialists' Collaborative Group (EBCTCG) 2012 meta-analysis. The EAG explained that using the MINDACT subgroup analysis HR across both risk groups would imply a lower benefit of chemotherapy overall than was seen in the EBCTCG meta-analysis, which included a much larger population than MINDACT. So, the committee concluded that the EAG's base case was more appropriate.

The committee concluded that Oncotype DX was likely to be a cost-effective use of NHS resources when used to help guide adjuvant chemotherapy decision making with postmenopausal women. In the EAG's base-case analysis, Oncotype DX dominated decision making without tumour profiling in postmenopausal women. However, this was dependent on the assumption that the test has predictive ability for chemotherapy benefit. In scenario analyses in which Oncotype DX had prognostic ability only, testing resulted in reduced costs but also fewer QALYs than standard care (savings of more than £30,000 per QALY lost with confidential price discounts applied). RxPONDER only provided data for people with RS 0 to 25, so model inputs for the RS 26 to 100 group had to be sourced from the TransATAC and SWOG-8814 studies, which used a cutoff of RS 30 for the high-risk group. The committee recalled its discussions on the uncertainty around the predictive ability of Oncotype DX for postmenopausal women (see sections 3.9 and 3.10), and that the effect of chemotherapy was likely overestimated in SWOG-8814 and underestimated in RxPONDER. It noted a scenario analysis, in which the hazard ratio for chemotherapy effect in the RS 0 to 25 group was reduced to 1.0 from 1.12, did not have a large effect on the cost-effectiveness estimates. The committee also noted that the economic model used a constant hazard ratio for the effect of chemotherapy over time. This may overestimate the effect because the greatest benefit of chemotherapy is seen in the first few years (see section 3.16). The EAG provided scenario analyses in which the effect of chemotherapy was lost after 5 or 10 years, and Oncotype DX remained dominating over current practice. The committee concluded that it was likely that Oncotype DX has some predictive ability for chemotherapy benefit in this population, but that the difference in chemotherapy benefit between risk groups was possibly overestimated in the economic model.

The committee concluded that Prosigna was likely to be a cost-effective use of NHS resources when used to help guide adjuvant chemotherapy decision making with postmenopausal women. In the EAG's base-case analysis, the probabilistic ICER for Prosigna was £24,547 per QALY gained using list prices. This result was driven by a small decrease in the number of people having chemotherapy and in the number of people with distant recurrence, and additional costs of testing. With confidential price reductions applied, the ICER was below £20,000 per QALY gained. The committee considered several scenario analyses which demonstrated that the ICER was sensitive to factors including sources of key inputs and the effectiveness and cost of chemotherapy. However, in most scenarios the ICER remained below £20,000 per QALY gained.

The EAG's economic model predicted that using tumour profiling tests would reduce the number of people having chemotherapy. The committee recognised that infusion services are often under a lot of pressure. Clinical experts explained that it is unlikely that reduced numbers of people having chemotherapy for early breast cancer would allow people with other types of cancer to access treatment faster, because generally people do not have to wait long for chemotherapy. However, it could improve patient and staff experience at infusion centres and reduce medical errors. Patient experts emphasised that people with LN-positive breast cancer should not be pressured to forego chemotherapy for capacity reasons, and that they should be offered an initial discussion with an oncologist even if tumour profiling tests suggest that they have a low risk of recurrence. The committee concluded that benefits resulting from reduced use of chemotherapy are already captured in the economic model through reduced costs and chemotherapy-related adverse events, and so did not need to be considered qualitatively.

The committee acknowledged that different tests were likely to reduce the number of people having chemotherapy by different amounts, but that EndoPredict, Oncotype DX and Prosigna were all likely to be cost effective compared with not using tumour profiling tests. The EAG's economic model predicted that using Oncotype DX would result in 594 fewer women having chemotherapy per 1,000 tested, whereas with EndoPredict or Prosigna the reduction was around 40. However, the committee noted that there was not enough comparative evidence to compare the tests directly, and that all 3 tests had ICERs below £20,000 per QALY gained when compared to not using tumour profiling tests.

The only available evidence on how test results influenced chemotherapy recommendations or decisions was for Oncotype DX. The committee considered that this data may not be fully generalisable to other tests because they measure the expression of different sets of genes, and because the risk classification probabilities of each test result in different numbers of people being assigned to low-, intermediate- or high-risk groups (see section 3.13). However, uncertainty in this parameter did not have a large effect on the cost-effectiveness results for EndoPredict (see section 3.16). Further evidence on how EndoPredict and Prosigna results affect chemotherapy decisions for people with LN-positive early breast cancer would be helpful to reduce this uncertainty.

The OPTIMA trial is an ongoing randomised controlled trial of Prosigna, comparing test-directed chemotherapy use with standard chemotherapy prescribing. The population includes people with a high clinical risk of recurrence and mostly people with LN-positive cancer (1 to 9 positive nodes). No results have yet been published. The committee recognised that OPTIMA may be able to address some of the uncertainty around the results for Prosigna, and encouraged clinicians to continue to promote enrolment in OPTIMA.