While de-escalation in local therapy of BC, including surgery and radiotherapy, has been the standard of care for many years, chemotherapy decisions are largely individualized based on the personalized risk of the patient [33]. Multigene assays have been used to avoid chemotherapy in patients with luminal-type BC, even in those with nodal involvement [6, 20]. The 70-GS assay, which includes 70 genes associated with tumor progression and metastasis, was approved by the FDA in 2007 for predicting the risk of distant recurrence at 5 or 10 years in early BC patients [34, 35]. When it was first applied among 295 consecutive patients with early BC, showing a 10-year DMFS rate of 54% for high-risk and 94% for low-risk women, differences were observed in the clinicopathological features and treatment of the 144 LN + patients and 151 LN- patients included in the study [36]. The predictive ability of 70-GS for the long-term prognosis of BC patients with 1–3 positive LNs was validated in an independent study [37]. Following this, the inclusion criteria for the MINDACT trial were revised, and ultimately, 21% of the patients included in MINDACT were LN+ [6]. Indeed, the evaluation method for the potential benefit of adjuvant chemotherapy differs according to the TAILORx and RxPONDER trials, particularly in terms of LN status and the 21-gene recurrence score (RS). LN status is also a key indicator for adjuvant CDK4/6 inhibitors, such as abemaciclib and ribociclib [12, 23, 24]. Therefore, in the current study, we developed updated nomograms for LN + and LN- patients separately.
The user-friendly, integrated multifactor nomograms typically facilitate individualized risk evaluation and assist in the prompt selection of patients. Several studies have established nomograms predicting the 21-gene RS [38–40], while efforts to establish nomograms predicting 70-GS risk have been limited [41]. Lee et al. established a nomogram to predict the probability of 70-GS low risk in women with clinically high-risk BC, incorporating age, grade, PR, and Ki-67, all of which were included in our previous nomogram models, except for age [19]. Based on our previous work, we conducted the current study with a doubled cohort size of 301 consecutive BC patients, with particular emphasis on the key parameter of LN status. We hypothesized that the distribution of risk factors between LN + and LN- subgroups of BC patients might differ, and the candidate parameters for nomograms for LN + versus LN- women might also be distinct. The 70-GS high-risk LN + patients had more children, fewer cardiovascular diseases, more BC with disordered blood flow on US, fewer T1 stage BC cases, BC with lower PR positivity, and a higher Ki67 index (all p < 0.05) (Table 1). These six parameters were all included in the LN + nomogram (Fig. 2). Conversely, the 70-GS high-risk LN- patients had more BC with microcalcifications, BC with lower PR positivity, and a higher Ki67 index (all p < 0.05) (Table 1). These three parameters were all included in the LN- nomogram (Fig. 4). With regard to the risk parameters, PR positivity and the Ki67 index were the common factors included in all three nomograms (Fig. 6). Interestingly, comorbidity was included in both our previous nomogram and the LN + nomogram as a 'protective' factor, suggesting that patients with cardiovascular diseases might be considered low-risk rather than high-risk (Fig. 2, 6).
To mitigate potential bias resulting from sample size discrepancies, we increased the sample size from 150 to 301 and conducted nomograms that included all patients, regardless of their lymph node status. For the binary categorized risk nomogram model, the AUC of the ROC improved from 0.826 to 0.853 in the training set and from 0.737 to 0.779 in the testing set. However, compared to the AUC of the ROC (training 0.826, testing 0.737) and C-index (training 0.903, testing 0.785) of binary risk prediction from our previous nomogram, the prediction performance significantly improved with the nomograms established for LN + and LN- populations separately. Among LN + patients, the AUC (training 0.948, testing 0.923), accuracy (training 0.907, testing 0.828), and C-index (training 0.948, testing 0.923) showed marked improvement. Similarly, among LN- patients, the AUC (training 0.917, testing 0.917), accuracy (training 0.870, testing 0.808), and C-index (training 0.917, testing 0.917) also improved (Table 3). We focused on the binary prediction of 70-GS risk and did not develop nomograms for quartile prediction due to the limited number of cases classified into quartiles among LN + and LN- patients.
Lymph node metastasis is a significant prognostic factor for early BC (EBC) patients. We established nomograms predicting 70-GS risk on an individualized basis with acceptable accuracy; however, the nomogram model did not effectively distinguish between patient cohorts with and without lymph node metastasis [19]. Upon further analysis, separating patients with LN + from LN-, we observed that the accuracy of AOL risk stratification significantly improved for LN + patients in our study. The AOL for Breast Cancer tool, a free web-based prognostication tool used globally to estimate 10-year survival probabilities and assess the benefits of adjuvant therapy, has been employed to aid in clinical decision-making [42]. Our prior study indicated that 43 (28.7%) patients classified as low-risk by AOL were subsequently evaluated by their physicians and underwent the 70-GS test due to the overly optimistic survival assessments provided by AOL in Asian patients.[42]. In our study, only 21 (12%) LN + patients with AOL low-risk received the 70-GS, with only 3 (14.3%) patients were ultimately identified as 70-GS high risk patients, which demonstrated that as for LN + patients, it might be safer to exempt from 70-GS to alleviate the patients’ economic burden. However, for LN- patients, there was no significant difference in the proportion of high-risk patients identified by AOL between the binary risk groups of patients evaluated by a 70-GS, indicating that LN- patients with AOL low-risk may not be safely excluded from a 70-GS assessment, and additional risk factors should be considered in clinical decision-making.
NPI had been confirmed and validated to stratify the prognosis of BC by incorporating of three prognostic factors: nodal status, tumor size and histological grade[43]. Compared with previously established classical risk models, we found that the risk stratification obtained by NPI model was most discordant with the 70-GS risk outcomes. For instance, the majority of 70-GS high-risk LN + patients (67, 91.8%) were moderate prognosis evaluated by NPI, while only 3 (4.1%) LN + patients classified as poor prognosis by NPI model were 70-GS high risk patients, which indicated that LN + patients classified as having a moderate prognosis by the NPI model may not be suitable candidates for exemption from adjuvant chemotherapy. Meanwhile, as for LN- patients, the NPI model did not exhibit the statistical difference in predicted prognosis between 70-GS high and low risk (p = 0.075) patients. In our study, we observed no statistically different number of positive nodes, tumor volume and histological grade between the 70-GS low and high risk both for LN + and LN- patients (Table 1). Additionally, the majority of patients included in our study had early-stage breast cancer, with a higher proportion of pT1 patients compared to pT2 and pT3 patients. Hence, it was hypothesized that the NPI may not be a dependable tool for predicting the 70-GS risk, though larger cohort was warranted in the future. Although NPI was considered as a robust and globally recognized system for stratifying EBC risk, a systematic review uncovered significant heterogeneity in studies examining the relationship between NPI categories and actual 5- and 10-year survival rates [44]. Additionally, previous studies have indicated that the predictive capacity of the NPI was less effective compared to other alternative prognostic tools [45, 46]. As observed in previous researches, NPI was a sub-optimal tool in predicting 10-year overall survival (OS) and disease-free survival (DFS) [47]. More valuable risk factors should be included in the NPI model to improve its predictive ability[48].
Our study was the first to compare the consistency between the MonarchE (FDA/NMPA) and 70-GS risk. Notably, MonarchE and NATALEE were not traditional risk calculation models. MonarchE was a phase III clinical trial which was designed to identify the efficiency of abemaciclib combined with endocrine therapy for adjuvant treatment of high-risk patients with HR+/HER2- EBC patients[49]. NATALEE was a phase III trial to evaluate the efficacy of reboxilide combined with endocrine therapy (ET) in patients with HR+/HER2- EBC at risk of recurrence[12]. Both abemaciclib in MonarchE and ribociclib in NATALEE were cyclin dependent kinase (CDK) 4/6 inhibitors that interrupted the proliferation of malignant cells through inhibiting the progression in cell cycle[50]. Based on the results from MonarchE, FDA and NMPA approved different indications for the application of abemaciclib in EBC patients. The main difference is that patients with a high level of Ki67 (≥ 20%) in the indications were also classified as high-risk patients and should receive adjuvant therapy in NMPA. As for LN + patients, our analysis revealed no statistically significant difference in the proportion of high-risk individuals based on the MonarchE (FDA) criterion between those classified as high and low risk by the 70-GS score (p = 0.066). However, we observed a significantly higher percentage of high-risk patients (66.2% vs 33.8%, p < 0.001) identified by the MonarchE (NMPA) criterion within the 70-GS high-risk group, and a greater proportion of low-risk patients (77.6% vs 22.4%, p < 0.001) identified by the MonarchE (NMPA) criterion within the 70-GS low-risk group. This may be attributed to the fact that all patients analyzed in our study were from China. The incorporation of a high Ki67 level (≥ 20%) as a marker of high risk in LN + individuals notably improved the consistency between the MonarchE (NMPA) and 70-GS risk stratification models. As for LN- patients, the high-risk stratification determined by the NATALEE criterion failed to demonstrate perfect concordance with the 70-GS high-risk stratification (p = 0.056). Our analysis suggested that Chinese LN + EBC patients classified as high risk by MonarchE (NMPA) should undergo 70-GS testing, with a 66.2% likelihood of falling into the 70-GS high-risk group. Conversely, those classified as low-risk by MonarchE (NMPA) may be confidently excluded from 70-GS testing, as there was a 77.6% probability of belonging to the 70-GS low-risk group.
To figure out the common significant variables constructing our nomogram models, we took the intersection of the variables included in our previous nomogram model [19], LN + nomogram model and LN- nomogram model (Fig. 6). Through this process, PR positivity and Ki67 were identified as significant variables. Similarly, we all excluded ER from the final nomogram models as all the patients exhibited high expression level of ER while PR positivity (%) varied between 0% and 100%. In accordance with prior studies, our nomogram demonstrated a negative correlation between PR positivity (%) and high-risk 70-GS. [51, 52]. Low or absent PR expression may serve as a potential indicator for patients who could derive greater benefit from adjuvant chemotherapy[53]. The Ki67 index, a well-established proliferation marker, has consistently shown a negative correlation with breast cancer survival. Therefore, the elevated Ki67 index may indicate increased 70-GS risk and a higher probability of benefiting from chemotherapy[54, 55]. Further endeavors were crucial to enhance the inadequate interlaboratory reproducibility and reconcile the discordance in cutoff selection for this biomarker[40]. For example, oncologists might be more confident in omitting adjuvant chemotherapy for Luminal A patients utilizing simplified nomogram models and more likely rely on 21-gene RS or 70-GS signature to decide the treatment decisions for Luminal B patients[28], and the lower cutoff of Ki67 would categorize more patients into Luminal B subtype.
There were also some limitations in our research. First, this was a single-institution and the external validation was absent. Second, selection bias was inevitable in this retrospective trail. Third, though we had incorporated in as many parameters as possible, some important parameters such as PET/CT or MRI were ignored due to the incomplete clinical information. Lastly, though our nomogram models achieved perfect diagnostic performance, other methods such as decision tree model, artificial intelligence might also work and further validation was warranted.