Breast cancer (BC) is one of the most prevalent and life-threatening malignant tumors affecting women worldwide, with high morbidity and mortality rates (1). According to the data from International Agency for Research on Cancer of World Health Organization, there were 2.26 million new cases of BC globally in 2020, accounting for 11.7% of new cancer cases. This means that BC has surpassed lung cancer for the first time and has become the most prevalent cancer in the world (2).

Radiotherapy is a crucial component of multimodal treatment for BC. It is used for early-stage, locally advanced and metastatic BC, especially in patients undergoing breast-conserving therapy and those with high-risk factors after modified radical mastectomy (3, 4). The main form of radiotherapy in comprehensive BC treatment is adjuvant radiotherapy (ART), although some cases with large tumors may require neoadjuvant radiotherapy. Numerous randomized controlled studies and systematic reviews have demonstrated that ART significantly reduces the risk of locoregional and distant BC failure and improves overall survival (3, 5, 6). However, despite advances in radiotherapy over the past two decades, locoregional recurrence (LRR) after radiotherapy remains the most significant treatment failure for BC (7, 8). LRR is a common failure mode in most BC cases and can be caused by radioresistance (9). Thus, identifying high-risk factors for cancer recurrence before radiotherapy can aid in determining the appropriate treatment approach and minimize adverse effects on patients. Nevertheless, this is challenging due to the relatively limited clinical indicators available for predicting cancer recurrence after radiotherapy (9, 10).

The Nottingham prognostic index (NPI) was developed by Haybittle in 1982 as a risk assessment tool for patients with BC. It calculates a score based on histopathological factors such as tumor size, lymph node status, and histological grade. The NPI remains one of the most important biological predictors for BC today (8, 11). By comprehensively considering tumor size, lymph node status, and histological grade, the NPI aids clinicians categorize patients into different prognostic groups, allowing for a more accurate prognosis prediction (12). Several studies have examined the relationship between overall survival and NPI, showing that higher NPI scores effectively predict worse long-term survival in breast cancer patients (13–15).

While higher NPI scores have been shown to negatively impact survival in BC, it is still unclear whether they can predict the risk of LRR. Therefore, the aim of this study was to determine whether higher NPI scores play a significant role in predicting LRR in BC by constructing two predictive models: a TNM staging-based prognostic model and an NPI-based prognostic model.

In our study, we developed two effective prediction models for LRR in BC. An important discovery and innovation of our study is the introduction of the NPI as a predictor of the LRR of BC, which has achieved ideal results. Our study has demonstrated that higher NPI scores are associated with higher LRR in BC in our real-world retrospective cohort. Additionally, we have preliminarily proven that NPI, as a predictor, performs better than TNM staging, which is an important prognostic indicator in BC.

Radiotherapy plays a crucial role in the comprehensive treatment of cancer. It is typically applied after surgical resection and often administered concurrently with chemotherapy and/or immunotherapy to achieve optimal tumor control (24). Among the advancements in radiotherapy technology, intensity-modulated radiation therapy (IMRT), a radiation therapy technology that modulates radiation intensity, has been widely used in cancer treatment. IMRT can preserve organs at risk and increase the radiation dose to the tumor, allowing for high-precision radiotherapy (25, 26). Additionally, hypofractionated radiotherapy based on IMRT and volumetric-modulated arc therapy technology has greatly improved the precision of BC treatment (27, 28). However, it is important to acknowledge that cancer recurrence after radiotherapy remains a significant form of treatment failure in most cases (29).

The recurrence probability for patients with BC after radiotherapy is approximately 20% - 30%, with the specific value depending on the patient’s condition, treatment effectiveness, and individual constitution (30, 31). One major reason for cancer recurrence after radiotherapy is the development of primary or secondary tolerance to radiation by tumor cells, known as radiation resistance (32). Many studies have reported the significant impact of LRR on cancer prognosis (33, 34). Patients with radiation-resistant tumors generally have worse survival rates, higher recurrence rates, and even higher rates of distant metastasis compared to radiosensitive patients (35, 36). Tumor cells with radioresistance can evade cell death after radiotherapy, leading to treatment failure. This issue is compounded by the accelerated repopulation of tumors, which primarily involves a group of residual radioresistant cancer cells, significantly reducing the sensitivity of recurrent tumors to treatment and resulting in poor clinical outcomes (37). Consequently, it is essential to understand the mechanisms whereby anti-radiation cells contribute to tumor repopulation to improve the prognosis of patients with cancer (38).

The heterogeneity of tumors makes it very difficult to predict the LRR of cancer (39). While prediction models based on genetic alterations or biomarkers have played an important role in predicting LRR in BC, it is undeniable that these prediction models often require patients to undergo gene expression testing, which greatly limits their clinical application (38, 40). Another difficulty in predicting LRR of BC is that radiotherapy in BC is primarily administered postoperatively. In this scenario, the patient’s tumor is usually completely resected, making the prediction of LRR quite challenging. Considering the ongoing controversy surrounding LRR prediction in BC, our study addresses this gap by constructing prediction models for locoregional recurrence of BC using common clinicopathological indicators. The results of our study demonstrate the effectiveness of these models in predicting LRR.

Many clinicopathological factors can significantly impact LRR (35). It is generally believed that as tumor invasion increases, tumor size increases, the proportion of hypoxic cells in tumor tissue increases, and the probability of tumor resistance to radiotherapy increases, which may be a leading factor in most cases of locoregional recurrence (33). In our study, we found a significant correlation between LRR of BC and pre-treatment T stage, specifically tumor size. This correlation has been confirmed in numerous radiobiological experiments (41). In addition, we found a significant correlation between lymph node stage and LRR. Generally, as cancer depth of invasion (T stage) increases, the rate of lymph node metastasis also increases (42). Many studies have shown that tumor invasion depth and differentiation degree are independent factors affecting cancer lymph node metastasis (43). Consequently, our study further highlighted TNM stage as an important clinical factor affecting LRR ( Table 2 ). By incorporating differentiation grades and the presence of perineural invasion and vascular invasion as indicators of radioresistance, we conclude that tumor invasion ability is a major factor influencing LRR.

In the present study, we found that age, BMI, and estrogen receptor status differed among different LRR groups. Specifically, there was a positive correlation between patients’ age, BMI, and LRR, although the specific mechanism remains unclear. We speculate that hormone levels may play a significant role, as hormone status is associated with BC prognosis. Furthermore, age, BMI, and estrogen receptor status are highly correlated with female hormone levels and BC pathogenesis (44, 45). In addition, in most cases, the molecular subtype was found to be a significant predictor of BC. In our study, the proportion of triple-negative BC showed an increasing trend in the recurrence group, although this trend was not statistically significant (P = 0.062). This may be due to insufficient sample size. Specifically, the proportion of patients with BC and negative expression of estrogen and progesterone was significantly higher in the recurrence group, indicating that hormone status is an important factor affecting recurrence. However, in our univariate analysis, we found that hormone status did not affect locoregional recurrence of BC after radiotherapy. Additionally, whether patients received endocrine therapy or not did not affect radiotherapy recurrence. This suggests that the mechanism of the influence of hormone expression on radiotherapy recurrence requires further research. Furthermore, only 14.29% of patients in our study had triple-negative BC and did not receive any endocrine therapy. This limited sample size may have prevented us from observing the influence of triple-negative BC on radiotherapy recurrence. Consequently, due to the negative results regarding the impact of endocrine therapy on recurrence in our univariate analysis, we did not include it as a variable in our multivariate model.

In contrast to the TNM staging system, the NPI is calculated based on tumor size, number of lymph node metastases, and degree of tumor differentiation. Previous studies have shown that NPI scores are associated with poor prognosis in BC (12, 13). In our study, we compared the distribution of NPI scores in patients with BC and different recurrent statuses as well as explored the correlation between NPI scores and important pathological features. We found that NPI scores were highly correlated with tumor invasion and lymph node metastasis in BC. Furthermore, the recurrent BC population had a higher NPI score, which led us to consider building a prediction model based on NPI. During the modeling process, we encountered significant collinearity between NPI scores and TNM stages, and their predictive effects would be compromised if considered together. Therefore, we built separate prediction models and compared their efficiency. Through univariate logistic regression analysis, we successfully identified six important variables that influence prognosis, including age, BMI, TNM stage, NPI, vascular invasion, and perineural invasion. After conducting multivariate logistic regression analysis, we found that vascular invasion was not an independent predictor of radioresistance and, therefore, excluded it from the models. Consequently, we constructed two significant prediction models: TNM-stage- and NPI-based models. The participating variables in the TNM-stage-based model were age, BMI, TNM stage, and perineural invasion. In contrast, the NPI-based model included age, BMI, perineural invasion, and NPI. While the composition of the two prediction models was highly consistent, the NPI-based model demonstrated better prediction accuracy and robustness than the TNM-stage-based model.

In recent years, the application of nomogram model in the field of cancer has gradually increased (46, 47). It incorporates various clinicopathological or genetic factors that affect the onset, prognosis or recurrence of patients into the prediction model and visualizes them, quantifies the risk ratio into specific scores, and obtains the risk probability to predict disease recurrence, metastasis and prognosis through simple calculation, providing a convenient and beneficial tool for clinicians and researchers (48, 49). The results of model validation in our study suggest that NPI-based model has good consistency and discrimination in predicting the status of radioresistance. In addition, the decision curve analysis method showed that the use of nomogram assessment could bring higher clinical benefits under a certain risk threshold.

Our findings further broaden the application scope of the NPI in BC and improve its clinical value. However, there are some limitations to our study that should be mentioned. First, the retrospective nature of this study introduces potential selection bias. Additionally, the sample size was insufficient to meet the criterion of ≥10 patients per risk factor, although we attempted to include important positive cases within the study period. Second, as the data for this study were from a single center, they may not fully represent the broader population. Furthermore, our prediction model only underwent internal validation; therefore, the selection bias present in the training cohort may also exist in the validation cohort. Further external validation in a multicenter setting is needed to determine if this nomogram can be widely used in other populations. Lastly, Gunda et al. conducted a similar study on the locoregional recurrence risk predicted by NPI, which showed that the NPI can effectively predict the locoregional recurrence of BC. Furthermore, two published studies have confirmed that the NPI can effectively predict the distant metastasis of BC (13, 50). Notably, our study focused solely on locoregional recurrence and did not address distant metastasis. Therefore, it may be necessary to verify the predictive value of the NPI for distant metastasis in our cohort in the future.

In summary, a higher NPI score is an independent risk factor for predicting locoregional recurrence in BC. Two nomogram prediction models related to radioresistance were constructed in this study. The nomogram prediction model based on the NPI has undergone internal validation and has been found to have good discrimination and calibration. It has the potential to provide clinical benefits and merits widespread use and application.