Proposed triaging approach for remaining on NeoET, considering alternative treatments or additional diagnostic procedures
The data presented in the sections below support the approach of identification of three groups of naturally postmenopausal patients that depends on the baseline expression of estrogen receptor (ER) and progesterone receptor (PgR) and an early on-treatment biopsy for Ki67 where indicated. Figure1 summarises the approach separately for ER/PgR Allred scores (a) and percentage of cells positive (b). In some centres where Ki67 is routinely measured at diagnosis, baseline Ki67 can provide an additional means of identifying those patients who may be excluded from on-treatment biopsy.
Allred scores are created by summing two separate scores: (the overall intensity of stained cells, none = 0, weak = 1, intermediate = 2, strong = 3) + (percentage of cells staining positive none = 0, <1% = 1, 1–10% = 2, 11–33% = 3, 34–66% = 4, ≥67% = 5).
Group 1: (~5%) should NOT be considered for NeoET
This is the group most likely to show no response and possible progression during protracted NeoET. Their endocrine responsiveness is poor overall, and they have a high incidence of on-treatment Ki67 > 10% which portends a poor prognosis.
Group 2: (~35%) may be considered for NeoET, provided that endocrine responsiveness is demonstrated as follows
In patients with Ki67 ≥ 15% at diagnosis, a core-biopsy should be taken after 2–4 weeks or later if more convenient. If on-treatment Ki67 > 10%, patients should be considered for other options such as surgery or neoadjuvant chemotherapy rather than continuing on NeoET; if on-treatment core biopsy and/or Ki67 analysis are unavailable, enhanced monitoring should be performed. Regarding the feasibility of on-treatment biopsies, if a surgical marker has not been placed, a patient can be started on NeoET and the surgical marker placed at the time of the on-treatment biopsy.
Group 3: (~60%) may remain on NeoET for at least 6 months
This group have very good endocrine responsiveness and a low incidence of on-treatment Ki67 > 10% overall. However, if baseline Ki67 is known to be >30%, an on-treatment biopsy should be considered as in Group 2.
Marti and Sanchez-Mendez2 have recently advocated the use of NeoET with biopsy for Ki67 and dichotomisation at 10% but provided no strategy for minimizing the numbers requiring on-treatment biopsy.
ER, PgR at diagnosis and clinical response
Studies in which ER+ PgR+ or simply high-ER score tumors (eg., H-score > 100) were selected for NeoET reported <5% of tumors progressing de novo on ET3,4. In the first, a phase 2 study conducted in Edinburgh, all tumors were ER+ and 92% were also PgR+. Patients were treated with neoadjuvant letrozole and 23/24 (96%) had a >25% reduction in tumor volume over the 3-month study period. In a second study, only 1/47 (~2%) of tumors with an H-score of >100 had de novo progression during the preoperative phase of ET treatment. The Edinburgh group recently audited their data on 456 postmenopausal patients treated with letrozole for a mean of 206 days between 2001 and 2016. The 4% of cases with Allred ER scores of 5 or 6 showed substantially poorer regression than those with Allred scores >6 (mean absolute reduction in largest volume = 7.2% and 34.4%, respectively).
These data suggest that high initial ER levels or a combination of both ER+ and PgR+ could be used to select a group of tumors highly likely to be controlled on NeoET. This approach alone still leaves a significant proportion of ER+ tumors that may progress during NeoET, particularly if NeoET must be prolonged, and thus additional approaches to response assessment are required if NeoET is to be used optimally and without adversely affecting long-term outcomes.
Early on-treatment Ki67 and clinical response
Presurgical use of ET can be either short-term, generally for about 2–4 weeks during the “window-of-opportunity” prior to scheduled surgery, or longer-term, usually for at least 3 months and in many cases up to 6 months and sometimes beyond5. The focus in the former, so-called window trials, is on obtaining biological response data including the proliferation marker Ki67. In the longer term, the primary goal of NeoET is downstaging of the disease and sometimes deriving biological response data for adjuvant treatment planning.
In several clinical trials of therapeutic NeoET, a biopsy at 2–4 weeks after starting NeoET has been taken for Ki67 measurement. There is substantial evidence for the on-treatment values providing strong prognostic information6,7,8. There are, however, few published data where the Ki67 data are considered in relation to clinical response or local tumor control.
We have now examined the data from two previously published NeoET clinical trials in postmenopausal patients to examine this relationship: anastrozole±gefitinib (IL1839/223)9 and letrozole±palbociclib (PALLET)10. Recent studies have consistently assessed a Ki67 cut-off at 10% as either a primary end-point or key factor in treatment allocation (POETIC5, WSG-ADAPT11, Z10317, ALTERNATE12), and this pre-specified cut-off has been employed in all of the data analyses performed below.
Patients with HER2+ disease were excluded at recruitment from PALLET, and those entered into IL1839/223 were excluded from the current analysis. Gefitinib has no significant impact on either Ki67 or clinical response9, so all patients in IL1839/223 starting on anastrozole for 2 weeks were included in this analysis; for PALLET only those patients in the letrozole alone arm were included, because palbociclib has profound antiproliferative effects in addition to those seen with letrozole10. For each study, 4 categories of residual, tumor size at the end of NeoET were prospectively developed based on 1-dimensional ultrasound measurements: >90%; <90%, ≥70%; <70%, ≥50%; <50% of baseline. The proportion of cases with 2-week Ki67 values dichotomized at 10% in each of these categories is are shown in Table2. It is clear that the better the Ki67 suppression at 2 weeks the greater the tumor size reduction from baseline, with the two sets of data producing similar results: in both trials, about 5 times as many patients had 2-week Ki67 > 10% among those whose tumors regressed poorly (<10% decrease), compared with those whose tumors regressed by >50%. For IL1839/223, p-for trend = 0.0504, and for the letrozole arm of PALLET, p = 0.02; summating the two sets of data, p = 0.002.
In the Z1031B study of neoadjuvant aromatase inhibitors13, of the 214 patients with Ki67 measured after 2–4 weeks, 165 (77%) patients had values ≤ 10% and of this group three patients showed progressive disease at 16 weeks. Building on these data, the ALTERNATE trial triaged postmenopausal patients with Ki67 > 10% after 4 weeks’ NeoET (anastrozole alone, fulvestrant alone, or the combination) to chemotherapy, whereas those with Ki67 ≤ 10% are maintained on anastrozole ± fulvestrant for 24 weeks before surgery. Reassuringly, only about 2% of patients maintained on NeoET experienced progression over that time, of whom half were confirmed radiologically.
Overall, these combined data provide a strong rationale for assessment of Ki67 after 2–4 weeks as a criterion for selecting patients who can safely be maintained on NeoET. However, during the COVID-19 pandemic, imaging-guided biopsies are not always readily available. In view of the direct relationship with clinical response as described above, we have evaluated whether ER and PgR expression at diagnosis can be used to reduce the number of patients for whom on-treatment biopsy is needed, and whether baseline Ki67 (if available) can further inform the decision.
Hormone receptor status and early on-treatment Ki67: a means of triaging patients
Higher levels of ER expression are known to relate to greater proportional benefit from adjuvant endocrine therapy14 and to relate to greater Ki67 suppression between baseline and 2 weeks, with aromatase inhibitors or tamoxifen15. Although PgR+ and PgR− cases show similar proportional benefit from adjuvant tamoxifen, PgR+ tumors have better prognosis, demonstrate greater clinical response to endocrine therapy and show greater Ki67 suppression with aromatase inhibitors or tamoxifen. We have therefore examined the degree to which ER and PgR status are associated with 2-week Ki67 in POETIC, ADAPT, PALLET and IL1839/223.
Our training analysis was conducted on data from a case-control (relapses vs non-relapses) study underway in POETIC. Receptor expression categories were derived from this exercise, then subsequently tested in the other trials to examine their validity. In each case, the data were categorised according to the Allred scores and subsequently according to percent ER+ cells to provide wide applicability. For POETIC, PALLET, and IL1839/223, the Allred scores were derived as close approximations from the ER and PgR H-scores.
Table3 shows the number and proportion of Ki67 values ≤ 10% and >10% in POETIC, according to Allred scores for ER levels and PgR, with the latter dichotomised at ≤5 vs ≥6. This dichotomy equates closely to 10% of cells PgR positive. Below, the PgR groups are termed PgR− and PgR+, according to this cut-off.
Cases with ER scores of ≤5 had a 65% incidence of Ki67 > 10% at 2 weeks. Tumors with an ER approximating to an Allred score of 6 but that were additionally PgR−, also had a high (47%) incidence of 2-week Ki67 > 10%. In contrast, tumors with an ER H-score of >175, approximating to an Allred score of 8, and that were also PgR+ had only a 15% incidence of 2-week Ki67 > 10%. In between these two extremes, the groups that formed Allred ER 7 or 8 but PgR− showed a 32% incidence of 2-week Ki67 > 10%, similar to the 27% incidence in those with Allred ER 6 or 7 and PgR+.
From POETIC, we have therefore defined three groups with Allred scores as shown in Table4. Data that emerge from POETIC using these categories are shown in Table5. Given that many centres use percentage (%age) ER/PgR scoring, we also developed groupings based on %ages based on data from POETIC shown in Table6. These groupings are also shown in Table4 and data that emerge from POETIC using these categories are shown in Table5.
It can be seen in Table5 that the proportions of patients in the groupings were very similar. We therefore tested the validity of the Allred subgroupings in the combined data from PALLET + IL1839/223 and from the ADAPT trial. For the latter, we treated the data from postmenopausal AI-treated patients separately from that of premenopausal patients, who mostly received tamoxifen (data shown in detail in Table7 according to Allred scores). Table8 show the data from PALLET + IL1839/223 and ADAPT postmenopausal patients, respectively. The proportion of patients with Ki67 > 10% or ≤10% in groups 1, 2, and 3 are very similar to those in POETIC. Of particular note, the proportions of patients with Ki67 > 10% in group 2 were 30% in POETIC, 32% in PALLET + IL1839/223 and 28% in ADAPT postmenopausal. In addition to percentages, the ADAPT trial also provided centrally performed Allred scores; the similarity between the study results supports the approximations to Allred scores made in POETIC, PALLET and IL1839/223.
Tables5 and 8 identify the patients who, on the basis of ER/PgR baseline status and on-treatment Ki67, should be considered for longer-term NeoET or conversely should not continue on NeoET. In PALLET + IL1839/223 and ADAPT postmenopausal patients, the proportions of patients who can remain on NeoET are very similar at 86% and 85%, respectively. From the POETIC data, the proportion was a little lower, which was expected, as this is a case-controlled cohort with 50% of the population experiencing a relapse. For POETIC, the overall proportion of patients recommended for on-treatment biopsy was 33% (110/338) if based on Allred scores and 42% if based on %age of cells staining; for PALLET/ IL1839/223 the proportion was 33% (85/257), and for ADAPT postmenopausal patients it was 41% (793/1925).
Use of baseline Ki67 for decreasing the proportion recommended for on-treatment biopsy
Baseline (i.e., pre-NeoET) Ki67 values are rarely below values obtained after short-term NeoET: in POETIC, 743/776 tumors had Ki67 ≤ 10% after 2 weeks among those that were ≤10% at diagnosis. To determine whether assessing (or when already analysed accessing) pre-NeoET Ki67 values could reduce the proportion of patients recommended for biopsy, we examined the relationship of pre-NeoET Ki67 values with on-treatment Ki67 specifically in Group 2. We applied cut-off values of 10%, 15%, and 20% to the pre-NeoET Ki67 values for three trial populations (Table9).
In POETIC, 43 (39%) of the Group 2 population had on-treatment Ki67 values ≤ 10% if they had values ≤ 15% at diagnosis. Four (9%) of the 43 had values above 10% at 2 weeks. In contrast, in the 27/67 (40%) patients with pre-NeoET Ki67 values >15% had on-NeoET Ki67 values of ≤10%.
In ADAPT post-menopausal patients, only 44/566 (7.8%) with baseline Ki67 ≤ 10% had an on-treatment Ki67 > 10%. Patients in Group 2 with baseline Ki67 ≤ 15% had a > 85% likelihood of obtaining on-treatment Ki67 ≤ 10%. This represent 44% of Group 2 patients (of note, the cutoff of 15% also applies if ER and PgR measurements were expressed according to Allred score).
Similar proportions of patients can be seen in each of the subdivisions of the Z1031B population with ER Allred scores of 6–8.
With regard to Group 3, post-menopausal ADAPT patients with baseline Ki67 measurements ≤ 30% had a > 85% likelihood of obtaining on-treatment Ki67 ≤ 10%. However, among patients with baseline Ki67 > 30% (who comprise only ~20% of Group 3), only about 57% obtained on-treatment Ki67 ≤ 10%. Thus, for this subset of Group 3, the authors recommend obtaining an on-treatment Ki67 measurement (if available) to ensure endocrine sensitivity before continuing NeoET.
Premenopausal patients
The safety concerns from COVID-19 infection are less in younger women and there are far fewer data on the long-term clinical efficacy of NeoET in premenopausal women. There are also fewer data on the short-term effects on Ki67 than in postmenopausal women. Consistent suppression of Ki67 was seen with both tamoxifen and a single 750 mg dose of fulvestrant in an Edinburgh study with tamoxifen showing a trend for greater suppression16. DeCensi reported that Ki67 was suppressed by tamoxifen after 4 weeks in premenopausal women, but the values were higher than those in postmenopausal women both before and at the end of treatment8. These findings are consistent with the higher proportion (60.5% vs. 24.4%) of pre- (88.6% treated with tamoxifen) vs. postmenopausal patients (91.9% treated with AI) in ADAPT having on-treatment Ki67 > 10%. The difference may result from Ki67 not being fully suppressed by tamoxifen within 3 weeks, due to the much higher competing estrogen levels in premenopausal women.
In Table8, it can be seen that there were no strong relationships between on-treatment Ki67 and ER or PgR levels in premenopausal patients in ADAPT, so these levels therefore cannot be recommended to reduce the number of patients who could be spared an on-treatment biopsy, even considering baseline Ki67.
Overall, it seems reasonable to use on-treatment Ki67 if there is a clinical need to prioritise NeoET for premenopausal patients, but there are few clinical data to directly support this so we are unable to recommend this. In ADAPT pre-menopausal patients, the percentages of patients with on-treatment Ki67 ≤ 10% were 25.8%, 37.0%, and 43.6% in Groups 1, 2, and 3, respectively.
Enhanced monitoring when on-treatment biopsy for Ki67 is unavailable
During the COVID-19 pandemic, resources for imaging-guided biopsies may be restricted and on-treatment biopsy for Ki67 unavailable for some patients. In the Edinburgh group’s audit of 456 postmenopausal patients treated with letrozole, a reduction of at least 15% in relative tumor volume early in treatment (mean = day 47) was significantly associated with continued neoadjuvant clinical response. Thus, for those patients in whom an on-treatment biopsy would otherwise be assessed, an interval assessment of maximum tumor size as measured clinically and by ultrasound or mammogram could be considered as an alternative. Failure to reduce maximum volume by ≥15% at 6 weeks and ≥20% at 3 months is associated with a significantly lower chance of long-term local control and a worse long-term survival. An on-treatment biopsy is preferred if resources and safety allow this.
Reliability of Ki67 analysis
It has been widely reported that Ki67 results can vary substantially between centres17. However, in an early study by the International Ki67 in Breast Cancer Working Group (KiBCWG)18, analysts at expert labs were found to have excellent within-lab consistency (ICC 0.94). Although the labs involved in the studies considered in the current report have not undertaken formal exchange of materials, there is evidence for their measurements being similar: the data relationships between steroid receptor subgroups and on-treatment Ki67 is very similar; and the ADAPT run-in phase almost exactly reproduced Ki67 assumptions based on the data from the Ellis and Dowsett labs19. Moreover, the analyses reported above for the ADAPT trial provided similar results using locally vs. centrally determined ER, PgR, and baseline Ki67.
Centres wishing to apply on-treatment Ki67 should confirm that their Ki67 analysis is operating to similar levels and should contact the authorship to access material if needed to do so. Differences in scoring of Ki67 were found by the International KiBCWG to be the most important source of between-laboratory variability. Consistency of scores is markedly enhanced by adherence to the methodology for “global scoring” which was validated and endorsed by the KiBCWG20 and is advocated here. Using this method, the median baseline Ki67 in ER + HER2- tumors in the whole POETIC trial (ie., not just the case-control series described here) was 14.3% (n = 3452). In ADAPT, median baseline Ki67 was 20% for both pre- and postmenopausal patients (n = 1954 and 2730, respectively). The median baseline Ki67 in ADAPT was expected to be higher than in POETIC since eligibility for ADAPT was confined to those that were candidates for (neo)adjuvant chemotherapy by conventional prognostic criteria.
Clinical factors
Clinical factors such as tumor size and nodal status do not impact on response to NeoET but do relate to the overall prognosis of patients. Their interaction with competing factors of advanced age/frailty and co-morbidity is helpful for decision-making in individual patients, and assists clinicians in prioritising the individual for surgery within a larger cohort of patients competing for appropriate care as in the COVID-19 pandemic. Thus, while the clinician should take these features into account when making decisions on appropriateness of NeoET the complexity of the interactive factors precludes our making precise recommendations.
The appropriateness or advantage of continued NeoET should be individualised and needs to re-appraised with frequent regularity as the situation changes in times of a pandemic such as the current one. While we do not recommend fixed durations for treatment, if the objective is to promote breast conserving surgery, 4–6 months is typical and longer can be helpful for larger tumors. In patients with smaller tumors who are already candidates for breast conservation, surgery could be as soon as the availability of resources returns.
The preoperative endocrine prognostic index (PEPI) was derived from patients in clinical trials where the length of NeoET varied between 12 and 18 weeks (12 weeks IMPACT/P024 and 16–18 weeks for Z10317). If a longer or shorter period of NeoET is used, the PEPI approach could still be applied on an individual basis, since the Ki67 is prognostic after very short periods of time (2 to 4 weeks). Thus, if a patient’s surgical sample was assigned PEPI-0 (Pathological stage T1 or T2, N0, Ki67 < 2.7%) and surgery was conducted when the patients was still on NeoET, post-surgical management without chemotherapy could be considered.
Genomic assays and use of NeoET
Genomic assays are in widespread use for the selection of those patients with ER+ HER2− node-negative breast cancer who should receive adjuvant chemotherapy as well as endocrine therapy. They estimate risk of distant recurrence on endocrine therapy alone largely by integrating information on baseline proliferation and features relating to the likelihood of endocrine responsiveness. Their main objective is to aid in counselling patients regarding their individual risk/benefit ratio for the use of chemotherapy. In contrast to the data presented here, there are no data on the relationship of genomic assays and local control with NeoET. Our proposal also has the advantage of undertaking an on-treatment measurement of the response to endocrine therapy (2–4 week Ki67). If a genomic assay is undertaken it must be conducted on a biopsy before starting NeoET since this has a profound but so far ill-described effect on the expression of many genes in the tests.
