All 103 patients in our cohort were evaluated for primary tumour response to NAC using three methods: standard pathology, MR score, and Chevallier score. Our results suggest that breast cancer treated with NAC followed by surgery and RT is associated with low rates of LRR (approximately 10% at 5 years) and relatively high OS (greater than 75% at 5 years). Rates of RFS and OS were significantly improved with increasing response to NAC, but LRR was not affected. Distant recurrence alone (15 patients) was twice as common as LRR alone (7 patients), similar to previously reported results of RT after NAC and surgery [
30]. Molecular subtype did not significantly affect recurrence or survival rates.
Rates of 5-year RFS for pCR, MR score 5, and Chevallier score 1 were 92%, 93%, and 86% respectively, while 5-year OS was 100% for all three groups. The similarity in outcomes suggests that all are valid methods for predicting outcomes after NAC, especially for those patients who achieve pCR. Patients grouped by MR score of 3 or 4 (denoting “partial responder”) or 1 or 2 (“non-responder”) tended to have worse outcomes than similar categories using standard pathology.
Tumour response assessment after neoadjuvant chemotherapy
In our cohort, the three response methods employed detected similar rates of pCR: 25.2% of patients achieved pCR by standard pathology, 30.1% achieved MR score of 5 and 21.4% had Chevallier score of 1. These rates are similar to the 26.1% pCR rate in NSABP B-27. Over 95% of patients received anthracycline and taxane-based chemotherapy, similar to that given to NSABP B-27 patients [
9].
All three treatment response methods showed similar predictive abilities for clinical outcomes. Using bivariate Cox modeling, all three methods were predictive of RFS, but none were predictive of LRR. Using Kaplan-Meier methodology, when patients were stratified by standard response assessment and MP, response to NAC did not predict for LRR but did predict both RFS and OS. Chevallier score did not predict LRR and did predict for OS; a borderline (p = 0.06) significant predictive ability was measured for RFS. Strong correlations between categorizations using the different methods were also demonstrated.
These results are reassuring, suggesting that the three methods (two pathologically derived and one based on radiographic findings) all serve as adequate predictors of response and ultimate clinical outcomes. The results agree with those of Romero et al. who reported that RCBI and RECIST-based assessments both predicted for RFS in a cohort of 151 patients [
21]. This cohort was similar to ours in proportion of tumours that were hormone-receptor positive (58% versus 64% in our cohort), although a higher proportion had Stage III disease (69% versus 49% in our cohort).
Recurrence and survival results in our cohort were similar to those reported in the initial publication by Ogston et al. that validated the Miller-Payne pathology scoring system [
19]. The initial cohort divided responses into three groups based on survival outcomes: Patients with Miller-Payne score 1 or 2 in the original cohort had 5-year RFS of 55–60%. For patients with Miller-Payne score 3 or 4, 5-year RFS in the initial cohort was 65–75%. For patients with Miller-Payne score 5 (indicating complete response), 5-year RFS was 95% in the initial cohort and 94% in our cohort. We were unable to conduct Miller-Payne scoring here but used the MR size-based method in order to assess potentially the same magnitudes of response in general terms. Our cohort had a lower proportion of node-negative patients (27%) compared with the initial cohort of Ogston et al. (57%). Lymph node involvement is a strong predictor of LRR and survival with breast cancer [
31,
32], which may explain why patients with similar magnitudes of response had lower survival rates in our cohort than in the initial cohort. Our results agree with existing literature, and suggest that multiple established pathologic and radiographic response measurements can provide reliable prognostic value for patients and clinicians.
Different pathologic subtypes of breast cancer may behave differently in response to NAC. Von Minckwitz et al. studied responses to NAC by different pathologic subtypes, in a cohort of over 6000 patients. They found that HER2-positive and triple-negative cancers achieved the highest pCR rates with Luminal A patients experiencing the lowest. If pCR was defined to allow residual in situ disease, pCR rates were 8.9, 51, and 35.8% for Luminal A, HER2-positive, and triple-negative, respectively [
29]. Our results closely match these findings, with pCR rates of 6.5, 42.9, and 41.7%, respectively, if in situ disease was allowed within the pCR definition.
Lee et al. studied over 500 patients and assessed response by both “relative” methods (i.e. those that compare size or cellularity of post-NAC samples with pre-NAC data) such as MR score and “absolute” methods (i.e. those that use only the post-NAC sample) such as RCBI. They concluded that for triple negative cancer, all response assessment methods can predict for disease-free survival. However, for hormone-receptor positive, HER2-receptor negative disease, only absolute methods had prognostic value [
33]. Our cohort was not large enough to study the prognostic value of response only among patients of a single pathologic subgroup.
All three response assessment methods we studied showed the highest rates of RFS and OS among patients who achieved pCR. The NSABP B-18 and B-27 studies compared outcomes after randomizing patients with operable breast cancer between NAC and post-operative chemotherapy. Patients in the NSABP B-18 trial received AC chemotherapy (doxorubicin and cyclophosphamide), while the B-27 study used AC plus docetaxel. Hazard ratios for recurrence and death for patients not achieving pCR compared with those who did not were 0.47 and 0.32, respectively, after 15-year follow-up in the B-18 trial and 0.49 and 0.36, respectively, after 8 year follow-up in B-27. Neither study demonstrated statistically significant differences between the NAC and adjuvant chemotherapy groups for either disease-free survival or OS [
8‐
10].
Our results support the hypothesis that pCR (as determined by all methods studied) is predictive of RFS and OS, even when RT is given after surgery (which was not given in the B-18 and B-27 studies). However, based on all three response methods assessed in this study, pCR does not appear to be predictive of LRR when RT is given. In the combined B-18 and B-27 cohort, predictors of LRR on multivariable analysis were: age at randomization, tumor size before NAC, clinical nodal status before NAC, and pCR after NAC. In our cohort, these same factors predicted for RFS (with pCR assessment by all three methods). However, we did not identify any factors that significantly predicted for LRR in our cohort. Our data support the NSABP findings that these factors are significant prognostic markers, but the RT that our patients received seems to have reduced these factors’ influence on LRR rates [
12].
Clinical outcomes after neoadjuvant chemotherapy, surgery, and radiotherapy
The role of adjuvant RT after NAC and surgery has not been evaluated in randomized trials but is generally recommended for patients with cT3-T4 and/or N2-N3 disease [
34,
35]. It may also be considered for patients with smaller, but still lymph node-positive, disease, especially those with high-risk factors such as young age or triple-negative biology [
36]. Retrospective analyses suggest that adjuvant RT benefits patients treated with NAC for advanced breast cancer. Huang et al. [
37] compared 542 patients who received NAC, surgery, and adjuvant RT with 134 patients who did not receive RT. The RT cohort had more advanced disease (73% pre-treatment stage III and 10% stage IV) than those who did not (46% stage III and 4% stage IV). Rates of LRR were nonetheless significantly lower for RT patients (11% versus 22%;
p = 0.0001), and RT significantly improved cause-specific survival for patients with stage IIIB or IV disease, clinical T4 tumours, and four or more positive lymph nodes. McGuire et al. [
10] studied 106 patients without inflammatory breast cancer, all of whom achieved pCR after NAC (92% included an anthracycline and 38% included a taxane in their NAC). Among the 67% of patients who had stage III disease at diagnosis, the 10-year LRR rate was 33.3% without RT and 7.3% with RT (
p = 0.04) and RT also improved OS (
p = 0.0017). No recurrence benefit was seen for patients with stage I and II breast cancer who achieved pCR however.
Our cohort had a similar stage distribution (51.5% stage II, 48.5% stage III) to these other retrospective reports. Our cohort's LRR rate (approximately 10% recurrence at 5 years) agrees closely with the results reported by Huang and McGuire for patients who received adjuvant RT. In the McGuire study, 5-year OS among patients who received RT was over 80%. Patients in our cohort who achieved a pCR had 5-year OS of 100% compared with 71% for partial responders and 65% for non-responders. The low rates of LRR seen in our cohort (particularly among patients who achieved pCR) mirrors the low rate of LRR in previous retrospective analyses. These retrospective cohorts together appear to support the routine use of adjuvant RT after NAC for clinical stage III disease, even after pCR. The question of RT after pCR is the subject of ongoing trials, including NSABP B-51 which is randomizing patients with T1–3, N1 disease between nodal RT and no nodal RT.
Buccholz et al. [
38] reported on 150 patients treated with NAC and mastectomy only. The plurality of patients (48%) had stage III disease (43% stage II, 7% stage IV). The reported 5-year LRR rate in this study was 27%, higher than the 8% in our cohort. Although our cohort did not include any stage IV patients, the relative proportion of patients with stage II and stage III disease was similar between our cohort and that of Buccholz et al. However, our patients all received post-operative RT, which may account for the lower LRR observed in our cohort. Buccholz et al. reported that pretreatment T stage and clinical stage group, lymph node involvement, and tamoxifen use predicted for LRR, while we did not identify identify any predictor for LRR on bivariate analysis, including nodal involvement or pretreatment stage. This difference may also reflect the effect of radiotherapy at reducing LRR risk.
In the NSABP B-27 trial, LRR at 10 years for patients who received AC-T chemotherapy was 9.5%. In the combined NSABP B-18 and B-27 dataset, the 10 year LRR rate was 11.1% (8.4% local and 2.7% regional recurrence). Local recurrences accounted for 71% of 10-year LRR in patients treated with mastectomy and for 79% of 10-year LRR in patients treated with lumpectomy plus breast XRT [
12]. Rates of LRR in our cohort were similar at 5 years to those reported at 10 years in NSABP B-27, although the patients in our cohort received postoperative radiotherapy and the NSABP patients did not. Patients in our cohort also had approximately twice the risk of distant recurrence as LRR, which is the opposite of the NSABP trials. These differences may arise because 70% of patients in our cohort having involved lymph nodes compared with around 30% in NSABP B-27, and patients with N2 staged disease being excluded from B-27 (13% of our patients had N2–3 disease). Thus, patients in our cohort likely had more aggressive cancer, on average, and a higher recurrence risk than the NSABP cohort, which may explain the elevated LRR in our cohort despite receiving adjuvant RT [
12].
Strengths and limitations
To our knowledge, this study represents the largest North American cohort to date with response to NAC assessed using multiple pathologic scoring systems, which represents a major strength. We observed strong correlations between all three response assessment methods and similar clinical outcomes between pathological pCR, MR-5 and Chevallier-1 (all indicating complete response). These results suggest that all three methods are valid predictors of clinical outcomes when assessing completeness of response. Our study also presents a cohort of relatively homogeneous patients who all received NAC followed by surgery and RT.
Limitations include the study’s retrospective nature, which depends on the completeness and accuracy of the information in clinical notes. Ki67 levels were generally not available, limiting our ability to distinguish between Luminal A and Luminal B (HER2-negative) cancers; however, this likely did not alter our conclusions.
Three patients who were otherwise eligible were excluded from the analysis due to lack of follow up. It is possible that these patients were lost to follow up due to poor outcomes (e.g. morbidity or death), which could bias our results. However, the number of such patients was small compared to the studied cohort size.
We did not have a sufficient number of deaths to study predictors of OS. Longer follow up may allow for more recurrences and deaths to be observed and late recurrence may be more likely in patients with certain outcomes (e.g. non-responders, aggressive molecular subtypes such as triple-negative). We also could not combine pathologic response assessment with molecular subtypes as was done by Lee et al. [
26]. A larger sample would allow for such subgroups to be studied, which could provide further data to improve breast cancer prognostics.