Prognostic analysis of three forms of Ki‐67 in patients with breast cancer with non‐pathological complete response before and after neoadjuvant systemic treatment

Abstract Background Patients who do not achieve a pathological complete response (pCR) after neoadjuvant systemic treatment (NST) have a significantly worse prognosis. A reliable predictor of prognosis is required to further subdivide non‐pCR patients. To date, the prognostic role in terms of disease‐free survival (DFS) between the terminal index of Ki‐67 after surgery (Ki‐67T) and the combination of the baseline Ki‐67 at biopsy before NST (Ki‐67B) and the percentage change in Ki‐67 before and after NST (Ki‐67C) has not been compared. Aim This study aimed to explore the most useful form or combination of Ki‐67 that can provide prognostic information to non‐pCR patients. Patients and Methods We retrospectively reviewed 499 patients who were diagnosed with inoperable breast cancer between August 2013 and December 2020 and received NST with anthracycline plus taxane. Results Among all the patients, 335 did not achieve pCR (with a follow‐up period of ≥1 year). The median follow‐up duration was 36 months. The optimal cutoff value of Ki‐67C to predict a DFS was 30%. A significantly worse DFS was observed in patients with a low Ki‐67C (p < 0.001). In addition, the exploratory subgroup analysis showed relatively good internal consistency. Ki‐67C and Ki‐67T were considered as independent risk factors for DFS (both p < 0.001). The forecasting model combining Ki‐67B and Ki‐67C showed a significantly higher area under the curve at years 3 and 5 than Ki‐67T (p = 0.029 and p = 0.022, respectively). Conclusions Ki‐67C and Ki‐67T were good independent predictors of DFS, whereas Ki‐67B was a slightly inferior predictor. The combination of Ki‐67B and Ki‐67C is superior to Ki‐67T for predicting DFS, especially at longer follow‐ups. Regarding clinical application, this combination could be used as a novel indicator for predicting DFS to more clearly identify high‐risk patients.


| INTRODUCTION
Neoadjuvant systemic treatment (NST) is becoming increasingly popular for the treatment of breast cancer because of its superiority in evaluating drug susceptibility and creating opportunities for operation or even breastconserving surgery. 1,2 Achievement of pathological complete response (pCR) after NST is a good surrogate endpoint for better survival. 3,4 However, patients who achieve pCR after NST remain a minority, and more than half still have cancer in the breast or axillary lymph nodes (ALNs) after NST. 5 Interestingly, survival among these patients varies widely. 6 Thus, the identification of prognostic indicators is crucial for this patient population. Consequently, much effort is being invested in searching for a reliable predictor of prognosis to further subdivide non-pCR patients. 7 Ki-67 is a non-histone nuclear protein in proliferating cells that is present in all phases of the cell cycle, except for G0. 8 It is an important indicator of the degree of tumor malignancy and is used to classify breast cancer, predict the chemosensitivity of tumors, and evaluate prognoses. [9][10][11] For the non-pCR patient population, three factors related to Ki-67 can be determined: the baseline Ki-67 at biopsy before NST (Ki-67 B ), the percentage change in Ki-67 before and after NST (Ki-67 C ), and the terminal index of Ki-67 after surgery (Ki-67 T ).
Ki-67 B has been reported to predict the efficacy of NST. 12,13 Patients whose Ki-67 B levels are initially high are more sensitive to chemotherapy and are more likely to achieve pCR. 12 However, the role of Ki-67 B in predicting the prognosis remains controversial. [14][15][16] In contrast, the ability of Ki-67 C or Ki-67 T to predict survival is relatively clear, and patients with a low level of Ki-67 T or high Ki-67 C are less likely to develop recurrence and metastasis. 11,15,17 Theoretically, Ki-67 T reflects terminal proliferation of tumors after receiving NST. 18 This value is complicated because it is affected by several factors. Unlike Ki-67 T , Ki-67 B reflects the baseline proliferation of the tumor tissue before receiving any treatment, and Ki-67 C reflects the sensitivity of the tumor to NST, together with therapeutic efficacy. 19 Thus, the malignancy of the tumor itself and the treatment efficacy can be analyzed using Ki-67 B and Ki-67 C , respectively. However, no studies have discussed whether Ki-67 B and Ki-67 C can predict the prognosis of patients more accurately than Ki-67 T .
To date, the prognostic role in terms of disease-free survival (DFS) between Ki-67 T and the combination of Ki-67 B and Ki-67 C have not been systematically compared. Thus, we designed this study to search for an appropriate cutoff value for Ki-67 C and to explore the most useful combination form of Ki-67 that can provide prognostic information to patients who have completed NST but do not achieve pCR.

| Patients
We retrospectively reviewed the medical records of 499 patients who were diagnosed with inoperable breast cancer and received an NST with anthracycline plus taxane at the First Affiliated Hospital of Nanjing Medical University (Nanjing, China) between August 2013 and December 2020. We excluded patients with a history of malignancy or chemotherapy, bilateral breast cancer, incomplete NST, distant metastasis found before or during NST, unavailable information regarding tumor characteristics, and follow-up of <1 year.
This study was conducted in accordance with the Declaration of Helsinki of the World Medical Association (revised 2013).

| METHODS
All breast cancer cases were diagnosed using core needle biopsy. Then, the specimen was further examined using immunohistochemistry (IHC) and considered hormone receptor (HR)-positive if >1% of cells showed the expression of estrogen receptors or progesterone receptors. In addition, it was defined as HER2-positive if in situ hybridization revealed HER2 gene amplification or IHC staining of HER2 was 3+.
Breast cancer cases were further differentiated into four molecular subtypes according to their HR and HER status: HR+/HER2−, HR+/HER2+, HR−/HER2+, or HR−/ HER2− (triple-negative, TN). Clinical tumor (cT) size and clinical lymph node (cN) classification were determined using the TNM Staging System for Breast Cancer (8th edition), as per the American Joint Committee on Cancer.
All patients received chemotherapy regimens of four cycles of epirubicin/cyclophosphamide followed by four cycles of paclitaxel or docetaxel. HER2-positive patients received targeted therapy with trastuzumab ± pertuzumab, breast cancer, cell cycle, neoadjuvant chemotherapy, prognosis beginning concurrently with paclitaxel or docetaxel. After completing the systemic NST, patients were scheduled for surgery within 1 month. The decision to perform mastectomy or breast-conserving surgery was made according to the tumor burden of the breast and the patient's willingness. ALN dissection was performed in patients considered to have ALN metastasis before undergoing NST. Otherwise, a sentinel lymph node biopsy was performed before the NST. All the surgically resected specimens were sent to an experienced pathologist for further examination. A pCR was defined as the absence of invasive cancer in the breast as well as ALNs, regardless of the presence of ductal carcinoma in situ of the breast. 3 Patients were then grouped according to whether pCR was achieved postoperatively.
All specimens were tested for Ki-67 (clone MX006, MAB-0672; Fuzhou Maixin Biotech Co., Ltd.) before and after undergoing the NST. The percentage of positively stained cells, defined as any degree of brown staining in the nucleus, is reported. According to the Guidelines of the Chinese Society of Clinical Oncology (CSCO) and the 2021 St. Gallen expert Consensus, a Ki-67 B or Ki-67 T index >30% was considered high, and an index <30% was considered low. 20 According to the analysis of the receiver operating characteristic (ROC) curve, an increased or ≤ 30% reduction in Ki-67 C was considered a low Ki-67 C , and a reduction >30% was considered a high Ki-67 C . All pathologies identified using core needle biopsy before NST and the pathological responses of the primary site and ALN after NST were assessed microscopically by a specialist pathologist and reviewed by another professional pathologist.
In the postoperative intensive adjuvant phase of non-pCR patients, those with HER2+ disease were proposed to receive dual-targeted therapy or T-DM1 for 1 year. 21 HR+ patients were proposed to receive adjuvant endocrine therapy (aromatase inhibitor or tamoxifen with or without ovarian function suppression) for at least 5 years. 22,23 Patients with TN have been proposed to receive orally administered capecitabine for 1 year. 24 The decision to administer radiotherapy at a specific site was made by a radiologist.
Our study protocol included regularly scheduled follow-ups, and the last follow-up was conducted in February 2022. DFS was considered as the primary outcome measure. It was defined as the interval between the time of surgery and tumor recurrence, metastasis, tumorrelated disease, or the date of censoring. Patients with a follow-up period of <1 year were excluded.

| Statistical analyses
The probability of DFS was estimated using the Kaplan-Meier method and the log-rank test. We used univariate and multivariate Cox regression models to analyze the associations of DFS with the listed variables and the likelihood ratio test to establish the multivariate Cox model. In addition, a time-dependent ROC curve was drawn at years 1, 3, and 5 to compare the sensitivity and specificity of the prognostic ability between different Ki-67 categories. The area under the curve (AUC) was calculated using the ROC curve. The Youden J statistic was used to confirm the optimal cutoff value. Statistical significance was set at P < 0.05. All statistical analyses were performed using IBM SPSS Statistics v26.0 for Windows (IBM Corporation, NY, USA) and R v4.1.3 software (the R Foundation for Statistical Computing, Vienna, Austria).

| Patient and tumor baseline characteristics among non-pCR patients
We screened 499 patients who underwent an integral NST ( Figure 1). Of these, 32 were excluded because of incomplete data on indispensable characteristics. Postoperative analysis of surgically excised specimens revealed that pCR was not achieved in 388 patients. Furthermore, 53 patients were excluded because their follow-up period was <1 year. Finally, we included a total of 335 patients who had not yet achieved pCR after NST.

| Subgroup analysis of DFS according to Ki-67 C level
A significantly worse DFS (hazard ratio 3.10, 95% confidence interval [CI]: 1.86-5.19; p < 0.001) was observed in patients with low Ki-67 C ( Table 2). In the subgroup analysis of age at diagnosis, HR status, HER-2 status, and Ki-67 B level, the results for each subgroup were similar; patients with a high level of Ki-67 C were correlated with better DFS (Figure 2). In contrast, the reduction in Ki-67 C only played a satisfactory prognostic role in DFS in the HR+/HER2− and HR+/HER2+ subtypes and in the cN-positive subgroup. There was no significant interaction between every subgroup and the level of Ki-67 C (all p > 0.05).

| Univariate and multivariate analyses of DFS among non-pCR patients
The median follow-up period of the non-pCR patients was 36 months (range, 12-98 months). Univariate Cox regression analysis of non-pCR patients revealed that cN, subtype, Ki-67 B , Ki-67 C , and Ki-67 T were significantly associated with DFS (Table 2). Figure 3 shows the Kaplan-Meier plots for DFS grouped by Ki67 B and Ki67 C .
We used two methods to build regression models ( Figure 1 and Table 3) to compare the difference in the prediction of prognosis between Ki-67 T and the combination of Ki-67 B and Ki-67 C . In method A, cN stage and Ki-67 T were considered independent risk factors for DFS. Patients in the high Ki-67 T group had worse DFS than those in the low Ki-67 T group (adjusted hazard ratio: 2.77, CI: 1.68-4.56, p < 0.001) ( Table 3). In method B, cN and Ki-67 C were also considered as independent risk factors for DFS. The low Ki-67 C group was more prone to recurrence and metastasis than the high Ki-67 C group (adjusted hazard ratio: 2.90, CI: 1.73-4.86, p < 0.001) ( Table 3). Furthermore, we used the Kaplan-Meier method to plot DFS curves for patients in the Ki-67 B subgroup ( Figure 3). In the low Ki-67 B subgroup, low Ki-67 C was associated with worse DFS (hazard ratio: 7.55, CI: 2.11-26.94). In the high Ki67 B subgroup, the results were similar to those mentioned above (hazard ratio: 2.47, CI: 1.39-4.39).  Figure 4), and the AUC value was further estimated. There was a statistically significant difference in the AUC values at year 3 and year 5 between the two forecasting models (70.5% vs. 63.8%, p = 0.029 and 73.3% vs. 63.5%, p = 0.022, respectively).

| DISCUSSION
This study found that, individually, both Ki-67 C and Ki-67 T were independent predictors of DFS, whereas Ki-67 B was slightly inferior. Patients with high levels of Ki-67 C had a better prognosis, and this result showed relatively good internal consistency in the subgroup analysis. In this study, the prognostic value of Ki-67 C in the HR-positive subgroup was better than that in the HRnegative subgroup. Similarly, one clinical trial showed that the Ki-67 index after 2 weeks of perioperative endocrine therapy among HR-positive patients was a better prognostic indicator than Ki67 B . 19 Ki-67 B had a certain effect on predicting prognosis in univariate Cox regression analysis in this study. However, after adjusting for other confounding factors (cN and Ki-67 C ), the prediction of DFS using Ki-67 B was considered unsatisfactory. These results were similar to those of other clinical studies. 11,15 So far, no other relevant study has discussed further subgroup analysis of Ki-67 C according to Ki-67 B in DFS among non-pCR patient populations. In this study, despite of whether the patients were classified into the low or high Ki-67 B subgroup, those who showed a high reduction in Ki-67 C had better DFS than those who did not. These results suggest a relatively poor predictive ability for Ki-67 B . As the DFS of patients with low Ki-67 B but low Ki-67 C was not very satisfactory in this study, this patient population must not be neglected in clinical follow-up and treatment. Subsequent adjuvant therapy for this patient group should be emphasized and studied further; for example, the use of CDK4/6 inhibitors in HR-positive patients or antibody-drug conjugates in HER2-positive patients. 25,26 In addition, this study revealed that the combination of Ki-67 C and Ki-67 B is superior to Ki-67 T in predicting DFS in patients with breast cancer. At present, most studies only consider the respective relationship between K-67 B , Ki67 C , and Ki-67 T , but do not propose combining the first two to produce a new prediction indicator, as in this study. 12,16 In the DFS predictions at year 1, both groups had high AUCs of approximately 78%. Interestingly, analysis of the prediction effect of a longer follow-up time (3 and 5 years) revealed that the Ki-67 T has deficiencies than the combination of Ki-67 B and Ki-67 C in predicting the long-term DFS. This indicates that the combination of Ki-67 B and Ki-67 C has advantages in predicting long-term prognosis and should be given more attention by clinicians in the enhanced treatment in the future. It is feasible to use 30% as the cutoff value for Ki-67 (whether Ki67 B , Ki67 T , or Ki67 C ) to predict the prognosis among non-pCR patients with breast cancer in this study. Since the determined values of Ki-67 B and Ki-67 T may differ among pathological experimental centers, a median cutoff value of 20%-30% or the median value across testing laboratories can be used as the cutoff value. 27 A cutoff value of 30% was used in this study, according to the CSCO and 2021 St. Gallen expert consensus guidelines. 20 In addition, the majority of studies focus on the absolute value of Ki-67 changes in predicting prognosis, while few discuss the percentage change. It is worth noting that, we analyzed the percentage of change in Ki-67 values instead of the commonly used absolute change. 28,29 The percentage change can better reflect the overall change in Ki-67 and not just the stacking of absolute values. Because of the different detection methods and experimental centers, there are deviations in the measured absolute value of Ki-67, 30 and the introduction of the concept of percentage change can make it universal for clinical applications.
Furthermore, it removes interference from the initial Ki-67 B and allows the evaluation of drug susceptibility in a more independent and systematic manner. Since research on the definition of Ki-67 C is scarce, we thereby determined the optimal cutoff value of 30% by performing a time-dependent ROC curve analysis at year 3 and employing the Youden index. One study by Matsubara et al. also focused on the percentage change in Ki-67 decline. The difference between their study and ours is that they divided the patients into three groups: high-reduction, low-reduction, and increase groups, so there were two cutoff values: one was 80% and another is 0%. 29 However, in Matsubara's study, high reduction in Ki-67 comprised only 14.6% of all non-pCR patients. And most patients (63%) were concentrated in the low-reduction group, resulting in no good discrimination for this group of the patients. In contrast, setting a 30% cutoff value in this study is more scientific and intuitive.
There are few things to explain here. More than half of the non-pCR patient population (183/335) were HR+/ HER2−. This might be because these molecularly subtyped patients were not sufficiently sensitive to NST. 5 The proportion of HER2-positive patients in this study was low (<30%). Owing to the accessibility of dual-target therapy in China, only about 11.2% (11/98) of all HER2positive patients were treated with double-target therapy in this study. Therefore, this issue has not been discussed separately here. This study has several strengths. First, we propose a novel concept of combining Ki-67 C and Ki-67 B to predict prognosis, which has never been involved in previous studies. Second, we recruited only patients who did not achieve a pCR in this study. This is because non-pCR patients have a significantly worse prognosis than those with pCR, and this specific patient group is the focus of our research. 3,31 Third, the chemotherapy regimens all patients received epirubicin/cyclophosphamide followed by taxanes; thus, the overall treatment is relatively consistent.
This study had some limitations. First, it was a retrospective single-center study with a relatively small sample size and a long time span, which might have led to some bias. Some patients were lost to follow-up. Second, standardization of the technology for detecting the proportion of Ki-67-positive cells using IHC and the representativeness and reproducibility of overall Ki-67 expression by core needle biopsies before NST remains uncertain. Breast cancer is highly heterogeneous, and we can only evaluate the overall tumor of patients by puncturing part of the tissue. The inability to accurately judge the initial numerical state of Ki-67 B is indeed a limitation of NST. Third, since DFS is only a surrogate endpoint for overall survival, longer follow-up of non-pCR patients is crucial to further verify the long-term clinical benefit and differences between the two methods.
In conclusion, this investigation revealed that the combination of Ki-67 B and Ki-67 C is superior to Ki-67 T in predicting DFS, especially in a longer follow-up period. Regarding clinical application, this combination could be used as a novel indicator to predict DFS to more clearly identify those F I G U R E 2 Subgroup analysis of disease-free survival related to Ki-67 C among non-pCR patients. B , baseline; C , percentage change; CI, confidence interval; HR, hormone receptor; Non-pCR, not achieved pathological complete response; TN, triple-negative (HR−/HER2−).   patients at high risk and to better determine whether escalation strategies in terms of adjuvant treatment are necessary.