Of the 373 consecutive cases that were enrolled, five were excluded from subsequent analyses due to missing data. The current analysis was therefore performed in 368 patients. Patient and tumor characteristics are shown in Table 1. EndoPredict allocated 238 patients (64.7%) in the EPclin low-risk and 130 patients (35.3%) in the EPclin high-risk group. The proportion of premenopausal and postmenopausal patients within each risk category was similar. For the 362 female patients (98.4%), the distribution of EPclin score was as follows: 32.9% (n = 77) premenopausal patients and 67.1% (n = 157) postmenopausal patients in the EPclin low-risk group (n = 234), and 35.2% (n = 45) premenopausal and 64.8% (n = 83) postmenopausal patients in the EPclin high-risk group (n = 128).

The tumor board recommendations for adjuvant chemotherapy according to EndoPredict test results are shown in Fig. 1. Six of the 238 EPclin low-risk patients (2.5%) were recommended to undergo chemotherapy despite their low-risk results, due to young age, multicentric tumor, or contralateral breast cancer. On the other hand, 13 of the 130 EPclin high-risk patients (10.0%) were recommended not to undergo chemotherapy despite their high-risk results, based on their individual risk factors such as age or comorbidities. Patients who were not recommended to undergo chemotherapy despite being EPclin high-risk were more likely to be older, have G1 tumors, and less likely to have nodal involvement (Table 2).

Patients’ EndoPredict test results and final tumor board recommendations

The treatment compliance is shown in Fig. 2. Adjuvant endocrine therapy for at least 5 years was recommended to all patients (n = 368). At the time of the last follow-up, 232 (63%) patients were compliant, 31 (8%) declined, and information on compliance with endocrine therapy could not be obtained in 105 (29%) patients. Regarding the distribution of the type of endocrine treatment being used the following medications were documented: aromatase inhibitor (153 postmenopausal pts, 4 premenopausal pts.), Tamoxifen (23 postmenopausal pts., 81 premenopausal pts.), various endocrine medications (25 postmenopausal pts., 27 premenopausal pts.), none (29 postmenopausal pts., 7 premenopausal pts.), and no data (10 postmenopausal pts., 3 premenopausal pts.).

Treatment recommendations given by the tumor board and patients’ compliance

Of the 123 patients to whom adjuvant chemotherapy was recommended, 89 (72%) were compliant, whereas 34 (28%) refused. One out of the 245 patients (0.4%) to whom adjuvant chemotherapy was not recommended underwent chemotherapy without tumor board recommendation. Overall, of the 130 EPclin high-risk patients, 87 (66.9%) underwent chemotherapy, whereas 30 (23.1%) patients opposed the recommended chemotherapy and 13 (10.0%) did not receive chemotherapy following tumor board recommendation. Among the high-risk patients to whom chemotherapy was recommended, the non-compliant patients (n = 30) tended to be older, have less nodal involvement and more G1 tumors, compared to the compliant patients (n = 87), while the median EPclin score was identical between the two subgroups (Table 2). Of the 238 EPclin low-risk patients, three (1.3%) underwent chemotherapy, whereas 235 (98.7%) did not.

Median follow-up was 8.2 (range 0.6–10.2) years. Five-year DFS was 95.3% (95% CI 92.6–98.0%) in the EPclin low-risk group versus 82.4% (95% CI 75.9–89.3%) in the EPclin high-risk group. With a HR of 2.08 (95% CI 1.26–3.44; p = 0.004), risk for disease recurrence or death in EPclin high-risk patients was twofold higher than in the EPclin low-risk patients (Fig. 3a). The 5-year DMFS in the EPclin low-risk group was 96.6% (95% CI 94.3–98.9%) and 85.5% (95% CI 79.6–92.0%) in the EPclin high-risk group. With a HR of 2.21 (95% CI 1.27–3.88; p = 0.005), the risk for distant metastasis or death in EPclin high-risk patients was more than twofold higher in comparison with EPclin low-risk patients. When considering only cancer-related death or distant recurrence EPclin high-risk patients were at significant higher risk than EPclin low-risk patients (HR 4.55; 95% CI 2.00–11.41; p < 0.001) (Fig. 3b).

Survival analysis. a Disease-free survival by EPclin risk category. b Tumor-related metastasis-free survival by EPcln risk category. c Disease-free survival in EPclin high-risk patients according to receipt of chemotherapy. d Distant metastasis-free survival by EPclin risk category in premenopausal patients. e Distant metastasis-free survival by EPclin risk category in postmenopausal patients

The analysis of DFS in the EPclin high-risk patients according to receipt of adjuvant chemotherapy showed a significant benefit toward the patients receiving chemotherapy (HR 0.46; 95% CI 0.23–0.95; p = 0.036). The 5-year DFS for the high-risk patients who received chemotherapy was 89.1% (95% CI 82.7–96.1%) versus 68.9% (95% CI 56.2–84.5%) for those who did not (Fig. 3c). When focusing only on the high-risk patients who had a chemotherapy recommendation (n = 117), of whom 87 were compliant and 30 non-compliant, there was also a significant DFS benefit toward the patients receiving chemotherapy (HR 0.41; 95% CI 0.19–0.89; p = 0.025). Furthermore, in a Cox regression model corrected for age, nodal status, and tumor grade in this specific patient population (high-risk patients with chemotherapy recommendation), we also found DFS benefit toward patients receiving chemotherapy (HR 0.31; 95% CI 0.14–0.71; p = 0.007).

We also performed multivariable analyses of Epclin adjusted for age, nodal status, tumor grade, tumor size, and ki67. With these analyses the hazard ratio of the EPclin stratification is corrected for these factors showing a 2.5 fold higher risk of distant metastasis (MFS: HR = 2.5 (95% CI 0.29–327.61), p = 0.48) and a 2.6 fold higher risk of any breast cancer recurrence (DFS: HR = 2.6 (95% CI 0.32–338.8), p = 0.451) in the Epclin high-risk group. The given sample size was sufficient for consistent effect estimation but statistical significance could not be reached.

When considering subgroups according to menopausal status, EPclin high-risk patients were at significant higher risk of experiencing distant metastases or death than EPclin low-risk patients in both subgroups. In premenopausal patients, the 5-year DMFS was 98.6% (95% CI 96.0–100.0%) for the EPclin low-risk patients and 86.0% (95% CI 76.2–97.1%) for the EPclin high-risk patients (HR 3.55; 95% CI 1.17–12.32; p = 0.025) (Fig. 3d). In postmenopausal patients, 5-year DMFS was 95.5% for the EPclin low-risk patients (95% CI 92.3–98.8%) and 84.9% (95% CI 77.4–93.2%) for the EPclin high-risk patients (HR 1.92; 95% CI 0.99–3.70; p = 0.054) (Fig. 3e).

We analyzed the effect of the EPclin classification in the context of Ki-67 subtypes. The determination of Ki-67 was possible in 306 (83.2%) samples. Ki-67 levels were low (< 10%; luminal A) in 73 (23.9%) patients, intermediate (10–25%) in 170 (55.6%) patients, and high (> 25%; luminal B) in 63 (20.6%) patients. We evaluated the correlation between EPclin risk class and tumor grade, Ki-67 distribution, tumor size, and nodal status (Fig. 4). There was moderate correlation (r = 0.465) with nodal status, and weak correlation with tumor grade (r = 0.348), Ki-67 (r = 0.287), and tumor size (r = 0.381). The EPclin-based low-risk classification was significantly associated with improved DFS compared to high-risk classification in both Ki-67 subtypes (Ki-67 low: HR 4.00; 95% CI 1.25–12.04; p = 0.021 and Ki-67 high: HR 3.77; 95% CI 1.19–18.93; p = 0.022). Using EndoPredict test result, 33.3% (21/63 patients) of all tumor samples classified as luminal B were re-classified toward the low-risk group, thereby sparing chemotherapy recommendation. On the contrary, 19.2% (14/73 patients) of all luminal A would be categorized to EPclin high-risk.

Correlation between EPclin risk class and tumor grade, Ki-67 expression levels, tumor size, and nodal status