A Nomogram for Identifying HR+/Her2− Breast Cancer Patients with Positive Sentinel Lymph Nodes and Omitted Axillary Lymph Node Dissection Who Need Abemaciclib Therapy

Background

Breast cancer is the most frequent malignancy in women, and axillary lymph node dissection (ALND) was the standard treatment for positive ALNs in patients. The ACOSOG Z0011 [1] and AMAROS [2] studies proved that ALND did not improve the overall survival rates in cT1-2 patients with clinically negative axilla and 1–2 SLN metastases. As regards locoregional lymph node recurrence, the Z0011 trial showed that the sentinel lymph node biopsy (SLNB) group had a cumulative incidence of axillary recurrence of 1.5% compared to 0.5% in the ALND group. Similarly, patients undergoing ALND had a 5-year axillary recurrence rate of 0.43% compared to 1.19% in patients undergoing radiotherapy in the axilla. The above data demonstrated that ALND presented no advantage in extending overall survival and controlling locoregional recurrence. Hence, most early-stage breast-conserving patients with negative lymph nodes avoid taking ALND. HR+/Her2− breast cancer is the most common subtype among early-stage patients [3], accounting for about 60% of all breast cancers. Despite the more favorable prognosis than Her2 and triple-negative breast cancer, patients still have a 40% risk of developing long-term recurrence, and the risk rises along with increased anatomical risk [4]. Patients with high recurrence risk should be treated as early as possible to prevent recurrence and metastasis. MonarchE [5] is a global, randomized, III-stage trial that proved that standard endocrine treatment + abemaciclib (an inhibitor of cyclin-dependent kinases 4 and 6) could improve the therapeutic outcomes of early-stage patients with high-risk HR+/Her2− breast cancer. The 4-year absolute invasive disease-free survival (iDFS) benefit was 6.4% (Abemaciclib group (85.8% [95% CI 84.2–87.3] vs 79.4% in the standard endocrine treatment group [95% CI 77.5–81.1]). The risk of an IDFS event was reduced by 33.6%. Patients included in MonarchE had positive ALNs ≥4 or had 1–3 positive ALNs and at least 1 high-risk clinical feature: tumor ≥5 cm or histological grade 3 or Ki-67 ≥20%. Since the Z0011 trial, omitting ALND has been extensively applied in breast-conserving patients with 1–2 positive SLNs. This leads to a new problem of how to decide if a patient needs abemaciclib therapy when they have 1–2 positive SLNs without high-risk clinical features? Is ALND essential to confirm axillary nodal metastatic burden? It was shown that 13.0–18.4% of patients who fulfilled the Z0011 criteria had more than 4 positive ALNs [1,2,6]. Therefore, we need to identify the risk of ALN metastases ≥4 in HR+/Her2− breast cancer patients with 1–2 positive SLNs and omitted ALND without high-risk tumor characteristics. These patients can probably benefit from abemaciclib therapy to escalate systemic therapy and de-escalate local surgical treatment. The present study aimed to analyze single-center retrospective data via univariate and multivariate regression analysis and construct a nomogram to select patients suitable for abemaciclib therapy.

Material and Methods

STATISTICAL ANALYSIS:

Patient clinical data were acquired, including age, BMI, menopausal status, unifocal/multifocal tumors, ALN ultrasound (US-ALN), mammographic calcifications, surgical method, pathological type, histological grade, tumor size, lymphovascular invasion (LVI), positive and negative SLN number, N classification, progesterone receptor, and Her2 expression levels. Clinicopathological factors were analyzed as categorical variables to explore their correlation with positive ALNs ≥4. Categorical variables were analyzed using univariate logistic regression analysis, from which the obtained significant variables were analyzed in multivariate logistic regression analysis to confirm independent predictors for positive ALNs ≥4. A nomogram was constructed using the “rms” package of R software. Discrimination and calibration were employed to evaluate the performance of the nomogram. Internal validation was performed using a 1000 bootstrap resampling method, with the C-index calculated and the calibration curve plotted. The discrimination ability was evaluated by the area under the ROC curve and C-index. Decision curve analysis (DCA) was used to evaluate the clinical efficacy and net benefit of the nomogram. All statistical analyses were conducted using SPSS 25.0 (IBM Corporation, Armonk, NY, USA) and R 4.1.1 (The R Foundation for Statistical Computing, Austria, Vienna). P<0.05 was interpreted as statistically significant.

Results

We recruited 444 breast cancer patients who fulfilled the inclusion criteria from January 2015 to June 2022. Patients’ baseline characteristics are displayed in Table 1. The median age was 50 years. There were 274 patients (61.7%) who had cT1 tumors, 318 patients (71.6%) had 1 positive SLN, and 126 patients (28.4%) had 2 positive SLNs. Among patients with 1 positive SLN, 9.43% had positive non-SLNs ≥3, and among those with 2 positive SLNs, 36.5% had positive non-SLNs ≥2. A total of 367 (82.7%) patients had 1–3 positive ALNs (pN1), and 77 (17.3%) patients had positive ALNs ≥4(pN2-3) (Table 2). Univariate and multivariate logistic regression analyses were performed to identify factors for positive ALNs ≥4. Factors with P<0.05 in the univariate analysis were analyzed in the multivariate analysis. Independent predictors for positive ALNs ≥4 included US-ALN (OR=2.289, 95% CI: 1.327–3.948, P=0.003), mammographic calcifications (OR=2.452, 95% CI: 1.297–4.638, P=0.006), T stage (OR=2.426, 95% CI: 1.414–4.161, P=0.001), and positive SLN number (OR=5.175, 95% CI: 3.007–8.904, P<0.001). A nomogram was plotted using multivariate logistic regression analysis data to predict positive ALNs ≥4 among cT1-2N0M0 HR+/Her2− patients with 1–2 positive SLNs (Figure 1). The possibility of positive ALNs ≥4 was calculated by adding scores of the 4 factors and referring to the total score and minimum risk scale. The calibration curve and ROC curve were used to evaluate the calibration and discrimination of the nomogram. The calibration curve indicated good agreement between predicted and observed outcomes for the nomogram (Figure 2A). The ROC curve of the nomogram is shown in Figure 2B. Sensitivity was 75.32%, specificity was 73.30%, and AUC was 0.777 (95% CI: 0.735–0.815, P<0.001), suggesting that the nomogram had good predictive performance. The nomogram exhibited good discrimination, with a C-index value of 0.802 (95% CI: 0.779–0.824). Decision curve analysis was used to determine the clinical applicability of decision-making based on the predictive model compared to the default strategy. When compared to treating all patients (slope line) or not treating any patients (horizontal line), the model exhibited high net benefits. The DCA curve (Figure 3) demonstrated that within the threshold risk range of 5% to 54%, intervention decisions based on the nomogram were clearly beneficial.

Discussion

In this study, a nomogram was constructed using clinicopathological factors to assess the risk of positive ALNs ≥4 among cT1-2N0M0 HR+/Her2− patients with 1–2 positive SLNs. The nomogram included US-ALN, mammographic calcifications, T stage, and positive SLN number. Internal validation suggested that the nomogram had good discrimination, calibration, and clinical efficacy.

At present, the main trends in breast cancer treatment are de-escalating chemotherapy for low-risk patients and intensive systemic therapy for high-risk patients. Multiple gene detection methods focusing on HR+/Her2− breast cancer (eg, Oncotype DX [9,10] and MammaPrint [11] have been developed to devise adjuvant chemotherapy protocols for early-stage breast cancer patients with HR+ and negative lymph nodes or 1–3 positive lymph nodes. Patients with high-risk clinicopathological characteristics (eg, subjects in the MonarchE study) had mortality rates similar those of patients with triple-negative breast cancer (TNBC) [12]. The MonarchE study [5] showed that 2-year administration of abemaciclib can improve clinical benefits. Currently, most patients who fulfilled the Z0011 criteria have omitted ALND, making it impossible to acquire complete ALN metastatic data, so partial patients do not have a basis for intensive treatment. The Ki-67 index can reflect cell proliferation degree, and patients with high Ki67 levels had worse prognoses [13]. TheMonarchE study population was allowed to be treated with abemaciclib when Ki-67 ≥20% as long as there was lymph node metastasis, with the exception of only ≥4 lymph node metastases when Ki-67 was <20%. Therefore, those with Ki-67 ≥20% were excluded from this study. In the ITT and cohort 1 of the MonarchE trial, the abemaciclib treatment significantly improved iDFS in patients with Ki-67 high-expression tumors. In cohort 1, the benefit of abemaciclib was consistently observed regardless of the Ki-67 expression, indicating that Ki-67 could not predict the therapeutic benefits of abemaciclib. Notably, patients with high Ki-67 expression had higher recurrence rates than those with low Ki-67 expression. Thus, Ki-67 expression is a prognostic indicator for recurrence, but cannot predict abemaciclib’s therapeutic benefits [14]. A previous study indicated that 13.0–18.4% of patients who met the Z0011 criteria had over 4 positive ALNs [1,2,6]. Our study showed that the possibility of positive ALNs ≥4 among cT1-2N0 HR+/Her2− patients with 1–2 positive SLNs was 17.1%, 9.43% for 1 positive SLN, and 38.1% for 2 positive SLNs.

Most predictive studies have focused on non-SLN metastasis presence, while studies concerning the exact lymph node metastasis number (eg, positive lymph nodes ≥4) are scarce. In addition, variables included in those studies were primarily SLN metastatic focus size, extramembranous infiltration of lymph nodes, and LVI [15], which were all based on postoperative pathological examinations. Gilles et al [16] established a preoperative clinical model and a postoperative pathological model predicting the risk of ALN metastases based on 12572 early-stage breast cancer cases, showing that the postoperative model had better discrimination (AUC=0.780) than the preoperative model (AUC=0.717). Intraoperative models are scarce. Shimazu et al [17] constructed an intraoperative predictive model for non-SLN metastasis using the one-step nucleic acid amplification method. The model included total tumor burden and tumor size, with an AUC of 0.70. The AUC of the model in this study was 0.777, and it was as effective as the postoperative model above. It is believed that tumor size, histological grade, LVI, age at diagnosis, positive SLN number, hormone receptor expression, Her-2 status, and molecular subtype are all significant risk factors for non-SLN metastasis [18]. This study revealed that US-ALN, mammographic calcifications, T stage, and positive SLN number are independent predictors for positive ALNs ≥4.

In the Z0011 trial [1], IBCSG 23-01 trial [19], and AMAROS trial [2], clinically negative lymph nodes were defined as the absence of enlarged lymph nodes in physical examination. Nevertheless, assessing ALN status via clinical examinations is inaccurate, and ultrasound is conducive to identifying patients with high axillary lymph node burden[20,21]. In this study, 146 patients had abnormal ALN on ultrasound, of which 39 patients (26.7%) had ≥4 positive lymph nodes, 260 patients had normal ANL ultrasound, and 38 (14.6%) of them had positive lymph nodes ≥4. It was shown that patients with abnormal ALN on ultrasound had higher axillary burden (OR=2.289, 95% CI: 1.327–3.948, P=0.003) versus patients with normal ultrasound results, consistent with Muneer et al’s study [21]. Lim et al [22] reported that ≥3 abnormal lymph nodes on ultrasound was a critical predictor for high axillary lymph node burden in patients who met the Z0011 criteria. Moreover, it was indicated that preoperative MRI could accurately detect ALN metastases to assess axillary tumor burden [23,24]. Radioimic variables can be included in future predictive models to further improve the model’s predictive performance.

The association between imaging manifestations of the primary tumor and ALN metastasis has not been adequately explored, especially with the commonly-used mammography. Zheng et al [25] found a significant association between ALN metastasis and mammographic calcifications among 7317 patients. Yan et al [26] reported that mammographic calcifications were significantly related to high ALN metastasis burden as an independent prognostic factor for breast cancer patients. Primary locus imaging manifestations were not included in studies like the Z0011 trial. However, we demonstrated a significant correlation between mammographic calcifications and positive ALNs ≥4 (OR=2.452, 95% CI: 1.297–4.638, P=0.006).

LVI is an independent predictor for lymph node metastasis and unfavorable outcomes [27,28]. It requires postoperative sections routinely stained by hematoxylin and eosin (HE), which is unsuitable for intraoperative models. LVI correlates significantly with age, T stage, histological grade, and hormone receptor expression [29]. In this study, LVI was found to be irrelevant to ALN metastasis (P=0.420), which might be because patients enrolled had positive hormone receptors, excluding the involvement of advanced T/N stage and histological tumor grade 3.

Her2 expression is significantly correlated with the metastatic potential of breast cancer cells. Ahmed et al [30] reported that Her2 expression is an independent predictor for ALN metastasis. It is currently believed that low Her2 expression may be a novel molecular subtype. Francesco [31] revealed that patients with HR-positive diseases (65.4%) had a higher proportion of low Her2 expression than patients with TNBC (36.6%). Low Her2 expression was irrelevant to the overall survival of patients with HR-positive breast cancers. Few studies have explored the relationship between Her2 expression and ALN metastasis. Our study showed that the proportion of low Her2 expression (85.6%) was higher than Her2 0 (14.4%) among HR+/Her2− patients, consistent with the Francesco study. However, no significant correlation was observed between low Her2 expression and HER2 0 status with the lymph node metastasis number (P=0.151).

The nomogram in this study exhibited a C-index of 0.802 (95%CI: 0.779–0.824), demonstrating good discrimination. Regarding calibration evaluation, the calibration curve displayed a strong concordance between the apparent predicted value distribution curve of the model and the distribution curve obtained after correcting for overfitting through resampling. It closely aligned with the optimal curve (standard curve), indicating good calibration of the model. To assess clinical effectiveness, decision curve analysis (DCA) was employed to calculate the net benefit and construct the DCA curve. The DCA curve illustrated that the model’s decision curve deviated from the 2 extreme scenario curves, signifying good clinical effectiveness.

The present study has some limitations. First, it was a single-center retrospective study with potential biases. Second, the nomogram was not externally validated. Thus, prospective multi-center studies are needed to evaluate and validate the findings.

Conclusions

We established an intraoperative predictive nomogram to predict the possibility of positive ALNs ≥4 among HR+/Her2− patients with positive SLNs and omitted ALND to assist clinical decision-making. The predictive model is accurate and can help oncologists identify patients who need abemaciclib therapy. Therefore, patients can be offered better management strategies, including de-escalation surgery and systemic escalation therapy in the SLNB and precision medicine era.