Introduction
The use of aromatase inhibitors (AIs) within the adjuvant setting, either upfront or sequentially before or after tamoxifen, has now been established given the results of several international studies [
1‐
7]. Recently Goss et al. reported results of MA.17R that there was a reduction in contralateral breast cancers and increased disease-free survival, even though there was no overall survival benefit, supporting the extended use of an AI as adjuvant endocrine therapy for 10 years [
8]. However, there is still considerable uncertainty as to whether such treatment is necessary for all patients and whether some patients can be treated solely with either tamoxifen or AI alone or switched to AI following tamoxifen treatment such as was done in the IES trial [
3]. The IES (Intergroup Exemestane Study) trial continues to report, even in its final analysis, at a median follow-up time of 12 years, a modest improvement in overall survival for those who ‘switched’ treatments (in preparation).
Whilst women with hormone receptor positive breast cancer can acquire resistance to endocrine treatment, it currently remains uncertain how this resistance occurs, and whether different mechanisms of resistance between the two treatment types exist. Several gene expression assays have been developed with the aim of distinguishing those patients who will relapse early on adjuvant endocrine therapy, including the Prosigna®, Oncotype DX®, EndoPredict® and Breast cancer Index™ but, as yet, none have been evaluated for their capacity to distinguish benefit from different forms of endocrine therapy.
We had established a translational group (PathIES) as part of the IES trial to evaluate the potential role of various candidate biomarkers to distinguish the effectiveness of tamoxifen and AI. This group has already reported the results of ERβ variants and its possible role in helping to predict appropriate endocrine therapy for patients in IES [
9].
The immunohistochemical (IHC) 4 + Clinical (C) score is a prognostic tool based on quantitative assessment of immunohistochemical biomarkers (ER, Progesterone receptor (PgR), HER2 and Ki67) and the clinicopathologic variables (tumour grade, size, nodal status, tumour grade, treatment with AI or tamoxifen) [
10‐
12]. The IHC4+C was developed to predict the residual risk of distant recurrence at 9 years in postmenopausal women with ER positive tumours treated with 5 years of adjuvant endocrine therapy only (i.e. no chemotherapy) [
13,
14]. However, to date, there is no report to evaluate the prognostic value of IHC4+C to patients who have received adjuvant tamoxifen for 2–3 years, followed by subsequent exemestane treatment to complete a total of 5 years endocrine therapy.
We consider it important to examine the role of the IHC4 score in predicting prognosis in patients who are switched from tamoxifen to an AI since this would provide a different cohort from other studies given that we include only those who remain disease-free at 2–3 years. This cohort therefore excludes patients who relapse early but more closely resembles the cohort in whom continuation of therapy beyond 5 years will increasingly be considered. We also report here on the role of IHC4 score in determining the relative sensitivity to either tamoxifen or sequential treatment with tamoxifen and exemestane.
Statistical analyses
The primary endpoint for this study was time to distant recurrence (TTDR) defined as time from random assignment to treatments to distant recurrence, death from breast cancer or unknown cause without prior recurrence. The clinicopathological characteristics of patients selected for this analysis to those not selected (due to unavailable tissue for the analysis, or unavailable markers for IHC4 assessment) were tabulated. No allowance has been made for multiple testing.
Calculation of IHC4 score and evaluation of its prognostic value among PathIES participants
Analysis was limited to ER+ PathIES patients as assessed centrally by either ≥ 1% positive stained cells or H-score ≥ 1 or Allred ≥ 3. The Cuzick et al. [
10] algorithm was adopted as follows, to derive the IHC4 score, which in combination with a clinical score (nodes, grade, age, tumour size) was tested for its prognostic value on our data.
$$ {\text{IHC4}}\;{ = }\; 9 4. 7x\;\;[ - 0.1\;{\text{ER}}_{10} - 0.079\;{\text{PgR}}_{10} \; + \;\;0.586\;{\text{HER2 + 0}} . 2 4 0 { }x\;\ln (1\; + \;4\;x\;{\text{ki67}}) \, ] \, $$
$$ \begin{aligned} {\text{Clinical score = 100 }}x\;[0.417\;N_{1 - 3} + \;1.566\;N > 3\; + \;\;0.93\;x\;(0.497\;T_{1 - 2} + \;0.882\;T_{2 - 3} \hfill \\ + 1.838\;T > 3\; + \;0.559\;Gr_{2} + 0.970\;Gr_{3} \; + \;0.130\;{\text{Age}}\;{ > }\; 6 5 ]\hfill \\ \end{aligned}. $$
In brief, the ER [
10] was equivalent to ER H-score divided by 30 and PgR [
10] equivalent to PR percentage of positive tumour nuclei cells divided by 10. The range of ER10 and PgR would be 0–10. HER2 was considered positive if IHC staining was 3+ and negative for IHC 0, 1+, 2+. Ki67 score was transformed as ln(1 + (4 × ki67)). The IHC4 risk groups were categorised as follows: quartile (Q) 1: < 25%, Q2–Q3: ≥ 25% and < 75%, Q4: ≥ 75%).
For the clinical score, N
j, T
j, G
rj and Agej denote categories of nodal status (N0, 1–3 N+ , > 3 N+), tumour size (< 1 cm (T
0), 1–2 cm (T
1–2), 2–3 cm (T
2–3) and > 3 cm (T >3)), grade (I, II, III) and age (< 65, ≥ 65).
The anastrozole versus tamoxifen effect term was deemed inappropriate for the exemestane effect on the PathIES data to validate the prognostic model of IHC4+C and therefore it was omitted. The IHC4+C risk groups were also categorised based on the quartiles (Q) (Q1: < 25% vs Q2–Q3: ≥ 25% and < 75% vs. Q4: ≥ 75%).
Kaplan–Meier plots, log-rank tests and Cox proportional hazards models, as appropriate, were used to compare how time to distant recurrence varied according to the IHC4 and IHC4+C risk groups. The significance of treatment effect with risk groups was determined by an interaction test in the multivariable Cox model.
A calibration plot comparing the predicted and observed probability of distant relapse by 10 years assessed the performance of the IHC4+C prognostic score. Patients were divided into ten groups according to their 10th percentiles of IHC4+C score; mean predicted values within each group were compared to the observed Kaplan–Meier estimates obtained for each group at 10 years.
Adjusting the prognostic clinical score model using PathIES parameters
To retain the comparability with the original IHC4+C model as reported by Cuzick et al., we used the same criteria to categorise the following variables: age, nodal status, tumour size and grade. The association of clinicopathological parameters, IHC4 score (included as continuous variable) and the PathIES treatment effect (tam→exem vs. tam alone) with survival data was assessed in a univariable Cox regression analysis. For the multivariable Cox regression analysis, we applied a stepwise backward strategy to select the most prognostic factors, whilst forcing the selection to keep treatment in the model, as assessed by a significance level of 10% if not lower.
Discussion
To our knowledge, this is the first time that the IHC4 and IHC4 + Clinical score has been tested for its ability to predict relapse in a cohort of patients who switched to an AI at 2–3 years, thus excluding those who relapsed early and more closely resembling the cohort who would potentially be considered for extended adjuvant therapy beyond 5 years. We found that patients with a IHC4 + Clinical score of ≥ 75th percentile have an approximately 50% risk of recurrence by 10 years after switching at 2–3 years. This may imply that this subgroup should continue adjuvant endocrine therapy beyond the total of 5 years.
Prediction of late relapse is a matter of considerable concern for patients who have switched therapies at 2–3 years and who remain disease-free after 2–3 years, since the current and planned randomised studies are insufficiently mature to assist their decision-making at the current time.
The IHC4 score has been confirmed as being predictive of early relapse by a number of groups, and is known to be especially valuable when combined with clinical prognostic scores [
10]. Recently it has been compared with other scoring systems for its ability to predict both early and late recurrences [
18,
19]; although the PAM50 risk of recurrence (ROR) score was superior in this study, the IHC4 has been found to be an important scoring system.
Our results confirm the prognostic importance of IHC4, alone and in conjunction with clinical scores. Although results from the calibration plot indicated that the prediction based on the published IHC4+C derived from TransATAC study was higher than the actual observed probability in some groups of predicted risk > 10%, One possible reason for this is that PathIES patients were treated with tamoxifen for 2–3 years, and remained recurrence-free before being randomised. Our results, nevertheless, demonstrated the prognostic value of IHC4 to segregate patients associated with differential risks of recurrence. The predictive value of the calculator might be improved by adjusting the weight estimates for each of the factors, given this is a different population and potential prognostic time dependency of some of the clinical pathological variables. Additional study using an independent cohort of patients is needed to investigate the robustness of the estimates.
Several other scoring systems have been advocated for their ability to predict late recurrence in patients with ER positive breast cancer. Sestak et al. [
19] compared IHC4, recurrence score (RS) as well as the PAM50 ROR score in patients enrolled in the ATAC study: here, node status, tumour size and the ROR score, a gene expression profile test, were the factors best able to predict long-term relapse. More recently, the TransATAC group compared the breast cancer index (BCI) (both linear and cubic) the OncotypeDX, as well as the IHC4 score; here the BCI (linear) had the best predictive value [
20]. The components of this score that were most important were HOXB13/IL17BR. The reason for these two factors being so important appears to be that HOXB3 can over-ride the tumour suppressor p21 whilst IL17 is now known to be the prime neutrophil-dependent growth promoter in breast cancer [
21]. The importance of this ratio was also underscored by the reports on retrospective analysis of the ratio in the MA17 study and predicted those who may benefit from extended letrozole therapy [
22]. Recently, TransATAC group reported that EndoPredict (EPclin), an alternative test combining an eight-gene signature (EP score) with tumour size and nodal status, provided more prognostic information than the OncotypeDX score for estimating late recurrence [
23], which may partly due to the reason that the test includes the significant clinicopathological variables.
Previously, an assessment of the predictive effects, in terms of therapy, of Ki67 had been reported by Viale et al. [
24]. This report suggested that high Ki67 levels predicted benefit from aromatase inhibition. However, this result was not amalgamated with the other three components of the IHC4 score, namely ER, PR and HER-2. In the current study, there were too few patients to enable an assessment of the IHC4 score for its capacity to predict which patients benefit from tamoxifen or exemestane after 2–3 years.
Recently, we carried out immunohistochemical staining for ER beta 1 and 2 in a subset of patients. Here, we found that, for those patients whose tumours expressed ER beta 1, the beneficial effect of simply continuing tamoxifen was similar to the patients who switched treatment to exemestane. Although requiring confirmation, this study suggests that it may be possible to ‘tailor’ treatment according to the primary tumour characteristics. This, combined with the IHC4 + clinical score, should enable us to optimise the type and duration of endocrine therapy.
There are a few caveats before translating these results into clinical practice; firstly, these patients did not receive trastuzumab; the study was initiated before the studies of adjuvant trastuzumab were mature and adjuvant trastuzumab became standard practice for patients whose tumours expressed HER-2; however, only 5% of patients had HER2 over-expressed tumours in this study. Secondly, Ki67 measurement, despite being the subject of a recent consortium statement remains a challenging analyte in tissue sections, due principally to heterogeneity of expression [
25,
26]. Thus, all the Ki67 values were analysed and assessed in one central laboratory. Secondly, a large proportion of patients received chemotherapy in this study, and especially this substudy, and caution should be exercised in translating these results to patients who did not receive cytotoxic chemotherapy.
Finally, although tissue markers reflect the biology of breast cancers in large series such as this, they do not enable clinicians to accurately predict the type and duration of treatment for individual patients; this is reflected by our finding here that approximately 50% of those with the highest quartile of the IHC4 + clinical score have not yet relapsed.
Other methods of predicting effectiveness and duration of treatment include the assessment of cell-free DNA. Using sensitive detection methods it is possible to detect circulating DNA from apoptosis residual breast cancer cells. It has now been shown that copy number variation [
27] and detection of mutations [
28] potentially can predict which patients are resistant to therapy.
In summary, the IHC4 score is useful in predicting long-term relapse in patients who remain disease-free after 2–3 years. Future, prospective studies are needed to define the role of IHC4 in selecting patients for long-term therapy.
Acknowledgements
We thank Kelly Mousa for all her work that enabled this study to be done. We thank the women who took part in this study, the pathologists, oncologists, nurses and support staff at local sites, and the data managers, trial coordinators and study managers from the Central and Eastern European Oncology Group (Poland: J. Jassem, A. Brociek, A. Pliszka), the Danish Breast Cancer Group (J. Andersen, B. Bruun Rasmussen), the European Organisation for the Treatment and Research of Cancer (Netherlands: C. van de Velde, E. Meershoek, Belgium: R. Paridaens, A. Delorge), the Gruppo Oncologico Nord Ovest, the Gruppo Oncologico Italiano di Ricerca Clinica, the International Breast Cancer Study Group (Switzerland: A. Coates, R. Camler), the International Collaborative Cancer Group (United Kingdom:, K. Mousa, S. Reed, Belgium: D. Verhoeven, S. Herman), Italian Trials in Medical Oncology (M. Visini), the North West England Group, the Norwegian Breast Cancer Group (P. Lonning), the Yorkshire Breast Group, the Wales Cancer Trials Network, the ICRCTSU (Zsolt Szijgyarto, Lucy Kilburn and Eleftheria Kalaitzaki). We also thank the Breast International Group for their support and the members of the IES steering committee and the PathIES Sub-Committee. This paper is dedicated to James Morden, a very talented scientist who contributed to this paper. James passed away suddenly during the preparation of this manuscript.