Previous studies have shown that combining clinical and genomic information allows for a more precise prognostic assessment than using either type of information alone in patients with breast cancer [12, 13]. This study focused on early recurrence within 3 years after surgery and revealed that clinical risk classification, when combined with a MammaPrint risk score, provided a more accurate assessment of the likelihood of early recurrence in HR +/HER2- early breast cancer patients. The study findings suggested that patients with both high clinical and genomic risk were at the highest risk for early recurrence, whereas those with both low clinical and genomic risk were at the lowest risk. Interestingly, patients with low clinical risk but high genomic risk appear to have an intermediate risk for early recurrence.

MammaPrint identifies genes related to distant metastasis through comprehensive genomic analysis rather than selecting genes known to be associated with prognosis [14, 15]. The genes in the MammaPrint assay are involved in early metastatic invasion processes, which consist of seven steps: growth and proliferation, angiogenesis, local invasion, intravasation, survival in circulation, extravasation, and microenvironmental adaptation at the metastatic site [16]. MammaPrint results had the greatest prognostic value for identifying patients at high risk for recurrence within 5 years after surgery in a validation study [16]. These findings aligned with the results of the current study, in which high-risk MammaPrint status was more frequently observed in patients with early recurrence.

In addition, a significantly greater percentage of patients in the early recurrence group than in the no recurrence group had Luminal B-type disease according to the Blueprint assay; basal-type disease was observed in only one patient in the no recurrence group, and the HER2-type disease was not observed in either group. In the MINDACT trial, prognostic outcomes varied by molecular subtype, with the Basal-type having the poorest prognosis, followed by the Luminal B, HER2, and Luminal A subtypes [5-year distant metastasis-free survival (DMFS): 90.4%, 92.7%, 94.3%, and 96.7%, respectively] [17]. In the NBRST trial, which aimed to demonstrate the utility of MammaPrint and BluePrint in the neoadjuvant setting, approximately 20% of the 426 cases pathologically classified as HR + HER2- were reclassified into non-luminal subtypes (85 cases as Basal-type and 2 cases as HER2-type) [18]. Furthermore, 80% of the cases classified as Basal-type experienced recurrence events within three years of diagnosis [18]. We hypothesized that the proportion of Basal-type cases would be the highest among patients with early recurrence. However, possibly due to the small sample size, Basal-type cases were rarely observed.

In this study, an association was observed between MammaPrint high risk results, Ki-67 levels, and nuclear grade. Previous reports have also shown a correlation between MammaPrint high risk results and a high Ki-67 index, as well as an association between a high Ki-67 index and the Nottingham histological grade [19]. On the other hand, when these clinicopathological factors were excluded, a MammaPrint high risk results was considered an independent prognostic marker.

In the present study, we did not evaluate whether clinical or genomic factors are predictive of systemic treatment outcomes. In the FLEX registry trial, patients with MammaPrint high-2–risk, BluePrint Luminal B, HR +/HER2- early breast cancer had a higher 3-year recurrence-free survival rate with anthracycline plus a taxane and cyclophosphamide than with a taxane and cyclophosphamide therapy in a nonrandomized cohort, and this tendency was not observed in patients with MammaPrint high-1–risk disease [20]. Although this study primarily evaluated prognostic factors, identifying predictive markers for the efficacy of individual therapies, including molecularly targeted agents, remains essential.

Although there are several commercially available gene expression profiling assays including Oncotype DX, we didn’t compare the performance of different assays in the present study. MammaPrint/BluePrint provides both prognostic risk stratification and molecular subtyping based on tumor biology, and we anticipate that the precise identification of tumors prone to early recurrence by MammaPrint/BluePrint may guide patient selection for intensified therapeutic strategies, including molecularly targeted agents.

Early recurrences were also observed among patients with both clinical low-risk and genomic low-risk disease, underscoring the challenge of identifying such cases and guiding appropriate treatment strategies. Further clinical studies are needed to optimize treatment, particularly for patients with clinical low-risk and genomic high-risk disease. More accurate identification of patients at elevated risk of early recurrence may enable better selection of candidates for adjuvant treatment intensification, such as CDK4/6 inhibitors. In addition, whether MammaPrint and BluePrint can serve not only as prognostic tools but also as predictive markers for the efficacy of specific therapies warrants evaluation in future prospective clinical trials. Such investigations will advance precision medicine and help refine therapeutic decision-making for patients with HR +/HER2 − breast cancer.

The primary limitations of this study are its retrospective nature and the small sample size. The availability of tissue specimens from the WJOG15721B trial was limited, and although 115 samples were submitted for MammaPrint and BluePrint testing, successful measurement was achieved in only 86 cases, resulting in an analysis success rate of approximately 75%. One possible reason for the low success rate is that the tissue samples used in this study were collected between 2012 and 2017, during which time the cold ischemia time and formalin fixation time were not as strictly controlled as they are today. As a result, the samples may not have been in an optimal condition for RNA analysis. Additionally, variations in the type and concentration of formalin used across different institutions and time points of sample collection may have influenced the results. Although a matched analysis was originally intended, only 86 samples were successfully measured, resulting in a number of unmatched cases and necessitating an unmatched analysis. This reduction in the number of evaluable samples represents sample attrition from the original study population and may have decreased the statistical power of the analysis. Furthermore, because the analysis was ultimately restricted to cases with successfully measured specimens, selection bias cannot be excluded, as these cases may not fully represent the overall study population. Therefore, the loss of the original matched design and the reduced sample size should be taken into account when interpreting the robustness and generalizability of our findings, and the results should be interpreted with appropriate caution. Despite these limitations, our study demonstrated meaningful results for current and future personalized medicine approaches.