Study participants and MRI ALN prediction rate
A total of 1560 patients with primary operable breast cancer who underwent pre-operative breast MRI evaluations and had detailed post-operative pathologic ALN biopsy results were enrolled in the current study (Fig 1). Among them, 490 (31.2%) patients had pathologically confirmed metastatic ALNs and 1074 (68.8%) had negative ALNs. According to intrinsic subtype classifications, there were 603 (42.4%) Luminal A, 370 (26%) Luminal B1, 206 (14.5%) Luminal B2, 117 (8.2%) HER-2, and 126 (8.9%) TNBC patients. The demographic data and histological results of the 1560 patients were summarized in Table 1. The diagnostic performance of breast MRI to predict ALN metastasis was calculated: the sensitivity was 61.4% (301/490), specificity 73.3% (784/1070), NPV 80.6% (784/973), PPV 51.3% (301/587), and the overall accuracy was 69.6% (1085/1560). The results were summarized and compared with other reported studies in Table 2.
MRI ALN prediction in different intrinsic subtypes
The accuracy of breast MRI for detecting metastatic ALNs in different intrinsic subtypes was evaluated (Table 3), and the overall accuracy in Luminal A, Luminal B1, Luminal B2, TNBC, and HER-2 was 74.1% (447/603), 66.5% (246/370), 67.5% (139/206), 67.5% (85/126), and 59.9% (70/117), respectively. The NPV was 85.6% (374/437) in Luminal A, 69.9% (146/209) in Luminal B1, 75.7% (78/103) in Luminal B2, 84.1% (58/69) in TNBC, and 71.2% (42/59) in HER-2. The sensitivity was highest in the TNBC group (71.1%, 27/38), and the highest PPV was in the Luminal B1 group (62.1%, 100/161). Accuracy, sensitivity, specificity, NPV, and PPV of MRI in detecting metastatic ALNs all significantly differed between different intrinsic subtypes (Table 3).
Prediction of ALN metastasis by clinicopathologic biomarkers
Using univariate analysis, we found that pathologic tumor size, MRI tumor size, histologic type, histologic grade, HER-2 status, and Ki-67 were statistically significant predictors of ALN metastasis. Multivariate analysis identified pathologic tumor size, MRI tumor size, and histologic type (invasive ductal carcinoma (IDC) versus non-IDC) as independent predictive factors of ALN metastasis (Table 4).
Prevalence of ALN metastasis and potential to omit SLNB based on clinicopathologic and MRI imaging factors
Based on the results of multivariate analysis, four individual risk groups of nodal involvement for each intrinsic subtype were generated, and the prevalence of ALN metastasis differed between subgroups (Table 5). According to our predefined condition of MRI lymph node prediction having a NPV ≥ 90% and no FN ALN metastatic disease worse than N1, we tried to identify some low risk groups of patients with invasive breast cancer that are potential candidates to omit SLNB (Fig 2).
Two definitions of measures of diagnostic performance were conducted in this study. By the definition of TNM staging, patients with pre-operative breast MRI tumor size ≤ 2 cm, no sign of ALN metastasis and pre-operative intrinsic subtypes Luminal A (metastatic ALN prevalence rate: 7.3%, MRI NPV: 95.4%, 125/131), Luminal B1 (metastatic ALN prevalence rate: 13.1%, MRI NPV: 91.8%, 45/49), or TNBC (metastatic ALN prevalence rate: 5.9%, MRI NPV: 93.8%, 15/16) were potential candidates to omit SLNB if they planned to receive BCS followed by whole breast radiotherapy. While by the definition of ACOSOG Z0011, patients with pre-operative breast MRI tumor size ≤ 2 cm, regardless of intrinsic subtypes, all reached NPV over 90% (range 93.8-100%, Table 5, Fig 2).
Validation analysis from the population-based Taiwan Cancer Registry (TCR)
According to our evaluation, patients with pre-operative MRI tumor size ≤ 2cm and no sign of ALN metastasis might be potential candidates to omit SLNB. To validate our hypothesis, breast cancer patients who had a clinical tumor size ≤ 2cm at diagnosis, had no sign of ALN metastasis (cT1N0M0) breast cancer, and underwent BCS and whole breast radiotherapy were identified from the TCR.
A total of 19,620 patients were enrolled according to inclusion and exclusion criteria, and these patients were stratified by whether surgical axillary staging was performed (performed n=17,275; not performed n=2345). The 5-year local recurrence free survival, distant metastasis free survival, and overall survival were not statistically different between these two groups (Fig 3 A-C). In subgroup analysis by molecular subtype, there were also no differences between patients who received or did not receive surgical axillary staging (Fig 3 D-F).