Breast cancer remains the most commonly diagnosed cancer in women worldwide and is a major contributor to cancer-related deaths. In 2020, it was responsible for approximately 685,000 deaths, accounting for 6.9% of all cancer deaths worldwide (1, 2). Breast conserving surgery (BCS) is a standard treatment for early breast cancer. It is recognized for being less invasive, offering cosmetic benefits, high patient satisfaction, and contributing to improved quality of life (3). Successful BCS is characterized by the achievement of “negative margins” (4), which indicates the absence of cancer cells at the edges of the tissue. When negative margins are achieved, BCS is comparable to mastectomy in terms of local recurrence and survival rates (5, 6), Consequently, the detection of “positive margins,” which indicate residual cancer cells at the tissue margins, after BCS necessitates reoperation (7).

In regions where reoperation is less favored due to various factors, the use of intraoperative frozen section analysis (FSA) for margin assessment in BCS has gained popularity (8). FSA provides high sensitivity and specificity for real-time margin assessment during surgery (9, 10), enabling on-the-spot re-excision of margins when positive margins are identified. This method significantly reduces the incidence of subsequent re-operations (11). Additionally, in many developing countries, including various regions in China, the procedure of conducting FSA-assisted diagnostic excisional biopsy before BCS within the same surgical session is still commonly practiced for economic and traditional reasons (8, 12, 13).

However, an FSA assessment often takes more than 30 minutes and requires the expertise of a trained histotechnologist for sample preparation and an experienced pathologist for interpretation (14). If FSA identifies positive margins during BCS, additional, sometimes multiple, re-excisions and subsequent FSAs may be required. This can result in prolonged surgery, longer anesthesia times, and increased operating room occupancy. Together, these factors can increase the risk of surgical complications and lead to higher healthcare costs. To reduce the incidence of recurrent positive margins in BCS with FSA-assisted excisional biopsy and margin assessment, and thereby reduce the frequency of multiple re-excisions and FSAs, predicting the risk of recurrent positive margins is critical. By identifying high risk cases, we can implement preemptive measures, such as increasing margin width, to reduce the likelihood of margin recurrence.

Existing models for assessing the risk of positive margins in BCS primarily rely on preoperative paraffin pathology and/or immunohistochemistry data (15–17). However, these data are often not available for BCS with FSA-assisted excisional biopsy and margin assessment. Therefore, it is essential to explore how to effectively utilize pathological information obtained by FSA for predictive purposes. In addition, these models generally focus on predicting initial margin positivity (16, 18–20). However, for BCS with FSA-assisted excisional biopsy and margin assessment, initial surgical margins are typically assessed at the time of tumor biopsy prior to BCS. Therefore, in this specific workflow, it is becoming increasingly important to predict the risk of positive margins at secondary margin excision in BCS.

The primary objective of this study is to develop a predictive model focused on predicting the risk of secondary margin positivity in re-excision procedures following the detection of initial positive margins. This model incorporates pathology data that can be obtained through intraoperative FSA along with clinical data that can be collected preoperatively. Therefore, the applicability of the model extends to BCS with FSA-assisted excisional biopsy and margin assessment within the same surgical session.

The patient selection process, including inclusion and exclusion, for this study is shown in Figure 1 . From January 1, 2016 to October 1, 2023, a total of 2,315 patients underwent breast cancer surgery in our department. Among them, 1,332 women initially underwent BCS for breast cancer. After applying our inclusion and exclusion criteria, 348 patients were identified who had a positive initial margin examination and subsequently underwent a secondary margin examination for study inclusion. Of these patients, 96 had positive secondary margins and 252 had negative secondary margins. The study cohort was divided into two groups for analysis: the training set, consisting of 256 patients admitted from January 1, 2016 to December 31, 2021, and the testing set, consisting of 92 patients admitted from January 1, 2022 to October 1, 2023. As shown in Table 1 , an analysis of variable distributions and comparative differences within the training and testing sets revealed no statistically significant differences in baseline characteristics between the sets (all p-values > 0.05), indicating a homogeneous distribution of variables across both sets.

During cross-validation of the LASSO over a range of λ values, the optimal λ was found to be 0.0447 (logλ = -3.1), corresponding to the point where the binomial deviation reached its minimum, as shown in Figure 2 . At this specific λ value, six variables with non-zero coefficients were retained, as detailed in Figure 2 . These variables, namely mammographic breast density, obscured mammographic mass margin, initial positive margin count, pathological tumor type, pathological multifocal carcinoma, and ultrasound mass margins, were subsequently included in the final logistic regression predictive model, as shown in Table 2 . The model demonstrates significant predictive ability, as indicated by a chi-squared statistic (χ²) of 61.09 and a p-value < 0.01. In addition, its pseudo-R² value of 0.31 supports this assessment. For practical application, a nomogram based on this logistic regression model has been constructed as shown in Figure 3 .

The mean ROC curves of the model for both training and testing sets, along with their 95% CI bands, are shown in Figure 4 . The training set achieved a mean AUC of 0.79 (0.72-0.85), while the testing set achieved a mean AUC of 0.83 (0.71-0.92), indicating high discriminative ability.

In the training set, the model achieved its maximum Youden index of 0.45 at a threshold of 0.3, corresponding to a nomogram score of 97 points, which was adopted as the optimal threshold for discriminating between high and low risk of secondary margin positivity. The confusion matrix and performance metrics for this threshold are detailed in Table 3 . In the training set, the sensitivity, specificity, PPV, and NPV were 0.66 (0.54-0.77), 0.80 (0.74-0.85), 0. 56 (0.47-0.64), and 0.86 (0.82-0.90), respectively, while in the testing set these metrics were 0.65 (0.46-0.84), 0.88 (0.79-0.95), 0.68 (0.53-0.85), and 0.87 (0.81-0.93), respectively. These results indicate the model’s robust predictive accuracy in both sets.

The calibration curves for the model in both the training and testing sets are shown in Figure 5 and demonstrate a close match between the predicted probabilities and the actual results. The Hosmer-Lemeshow test results support this, with a p-value of 0.214 (0.204-0.224) for the training set and 0.167 (0.158-0.176) for the testing set, indicating good calibration.

Finally, the DCA for both the training and testing sets is illustrated in Figure 6 , which shows the mean DCA curves along with their 95% CI bands. Notably, the mean net benefit of the model consistently exceeds that of the “intervention for all” and “intervention for none” strategies across a range of threshold probabilities. This is particularly evident at the optimal threshold, where the model’s mean net benefit and its 95% CI are significantly higher than these strategies in both sets. Such results highlight the model’s potential to improve patient outcomes, providing robust evidence for its clinical utility.

This study successfully developed a novel predictive model for secondary margin positivity applicable to BCS with FSA-assisted excisional biopsy and margin assessment.

During BCS, when the initial margin FSA result is positive, surgeons typically perform a margin re-excision and re-submitte for FSA within the same operative session. At this critical juncture, the surgeon must decide whether to perform a conservative re-excision with minimal tissue removal or to perform a more extensive re-excision or even escalate to a total mastectomy. However, this decision-making process is challenging due to the paucity of information regarding the risk of secondary margin positivity.

Our predictive model addresses this gap by integrating preoperative and intraoperative data available at the decision point to provide critical information about the risk of secondary margin positivity in subsequent re-excisions. This information allows surgeons to make more informed decisions, potentially reducing the likelihood of persistently positive margins. For example, in cases that are considered low risk (risk score below 97 points), surgeons can confidently opt for more conservative re-excisions, preserving healthy breast tissue and potentially improving cosmetic outcomes without compromising oncologic safety. Conversely, in high-risk cases (risk score above 97 points), surgeons may consider more extensive re-excision or even total mastectomy to minimize the likelihood of persistently positive margins and the need for further re-excision.

The implementation of this predictive nomogram in clinical practice has the potential to provide multiple benefits across multiple aspects of breast cancer surgery. By facilitating more informed decision-making during the critical intraoperative period following an initial positive margin, this model could contribute to a reduction in operative time. This temporal efficiency is particularly valuable in the context of breast-conserving surgery, where prolonged anesthesia time may increase patient risk. In addition, the model’s ability to stratify risk for secondary margin positivity could lead to a reduction in unnecessary re-excisions, thereby reducing pathologist workload. This is particularly important given the time-sensitive nature of frozen section analysis and the potential for resource constraints in pathology departments. From a patient-centered perspective, the use of the model may result in a reduction in surgical risks associated with prolonged surgery or multiple re-excisions. In addition, by potentially reducing the need for repeat surgeries, the model could contribute to an overall reduction in the cost of breast cancer care. Taken together, these factors result in an improved surgical decision-making process that enables more precise and personalized care. This alignment with the principles of patient-centered care represents a significant advancement in the field of surgical oncology, potentially improving both clinical outcomes and patient satisfaction.

This study represents a significant advancement in the application of predictive modeling for margin status in BCS. Previous research in this area has primarily focused on developing models based on preoperative paraffin pathology and/or immunohistochemistry data (15–17). In addition, these earlier models were generally designed to predict initial margin positivity (16, 18–20). However, such approaches have limited applicability in surgical settings where FSA is routinely used intraoperatively to determine tumor type and assess margins. In these scenarios, preoperative pathological and immunohistochemical information about the tumor is often not available. Furthermore, existing models are not adapted to predict the risk of secondary margin positivity in cases where re-excision and subsequent FSA are performed during the same surgical session following an initial positive margin. Our study addresses this critical gap by developing a nomogram specifically tailored to this surgical workflow. To our knowledge, this is the first model applicable to such a surgical workflow in BCS. By utilizing both preoperative and intraoperative data, including initial FSA results, our model provides a novel tool for real-time risk stratification of secondary margin positivity. This extension of predictive modeling to intraoperative decision making represents a significant step forward in the field of surgical oncology, potentially improving the accuracy and efficiency of BCS procedures.

The model demonstrates robust performance across multiple statistical measures. Its discriminative ability is evidenced by mean area under the curve (AUC) values of 0.79 (95% CI: 0.72-0.85) in the training set and 0.83 (95% CI: 0.71-0.92) in the test set, indicating strong predictive accuracy. Good calibration is supported by good agreement between predicted and actual outcomes, as demonstrated by calibration curves and Hosmer-Lemeshow test p-values of 0.214 and 0.167 in the training and test sets, respectively. The clinical utility of the model is demonstrated by DCA, which showed consistent superiority over “treat all” and “treat none” strategies. At the optimal threshold of 97 points on the nomogram, the model achieved balanced sensitivity and specificity in both sets (training set: 0.66 and 0.80; test set: 0.65 and 0.88, respectively). These consistent results across training and test sets, supported by bootstrap resampling for confidence interval estimation, provide compelling evidence of the model’s reliability and potential clinical value in identifying patients at high risk for secondary margin positivity during BCS procedures.

Our study found a significant correlation between invasive lobular carcinoma (ILC) and secondary margin positivity (p < 0.05). This association confirms previous findings that ILC is associated with an increased risk of initial positive margins in BCS (42–46). The peculiar biology of ILC, in particular the loss of E-cadherin leading to a discohesive cell structure, makes it difficult to achieve clear surgical margins (47). In addition, our study uncovers a novel finding: the correlation between ultrasound margins with indistinct microlobulated features and secondary margin positivity (p < 0.05). This novel observation highlights the need for further investigation. Furthermore, we found a potential but not statistically significant correlation between obscured mammographic margins and secondary margin positivity. Typically, obscured mammographic margins, which are often associated with dense breast tissue, can compromise the accuracy of mammographic imaging (48) and challenge the accurate localization and sizing of tumors (49). Our results highlight the critical role of considering the masking effect of dense breast tissue in preoperative planning. This suggests that patients with dense breasts may benefit from additional caution or alternative imaging modalities to reduce the risk of positive surgical margins.

Our study has several limitations:

1. Retrospective nature: As a retrospective study, there is an inherent potential for selection bias. For example, some patients with positive initial margins may have undergone total mastectomy or refused further re-excision. This study excludes their data from the analysis of secondary margin status, thus introducing selection bias. However, the number of such patients in our study was relatively small, mitigating the effect of this bias.

2. The relatively small sample size of our study poses a potential risk of model overfitting. To minimize this problem, we implemented LASSO regression to reduce the likelihood of overfitting in model construction. In addition, we used the bootstrap method to increase the accuracy of assessing model performance with a reduced bias in model validation. However, despite these precautions, the risk of overfitting and overestimation remains due to the small sample size. This concern calls for a cautious interpretation of our results. The limitation of the small sample size highlights the need for future research with larger and more diverse datasets.

3. Lack of external validation: This study did not include an external validation set from other medical institutions. Although we attempted to reduce assessment bias by dividing our dataset into training and testing sets based on different admission periods, the model’s performance in different clinical settings is still limited due to the lack of external validation. Thus, the nomogram requires validation in multiple clinical settings before it can be integrated into clinical practice.

4. This study is based on the assumption that surgeons can effectively intervene in cases where patients are identified as high risk to prevent the occurrence of secondary positive margins. However, the actual effectiveness of the model in reducing the incidence of secondary positive margins in real-world surgical settings has yet to be empirically validated. This underlying assumption is critical to the application of the model and underscores the need for further, more comprehensive investigation. Future research should focus on standardized preventive measures and rigorous clinical validation of the model’s effectiveness in reducing secondary positive margins. The implementation of randomized controlled trials is particularly important, as they would provide robust and definitive evidence of the utility and efficacy of the model in clinical practice.

In conclusion, this study presents an innovative nomogram for predicting secondary margin positivity in BCS with FSA-assisted excisional biopsy and margin assessment. This predictive tool addresses a critical gap in existing models by seamlessly integrating preoperative clinical data with intraoperative pathology findings, enabling real-time risk stratification at a critical decision point during surgery.

Implementation of this nomogram could have multiple benefits, including reduced operative times, reduced pathologist workload, and reduced surgical risks and costs for patients. The model’s potential to optimize surgical planning, reduce re-excision rates, and improve resource utilization represents a significant advancement toward more personalized and efficient breast cancer care.

However, we recognize that further validation and research is essential. Future investigations should prioritize external validation through multi-center studies, and cost-effectiveness analyses will help quantify the economic benefits of implementation of this predictive tool in clinical practice. Furthermore, prospective studies evaluating the practical integration of the nomogram into existing clinical workflows will be essential to ensure its seamless adoption and maximize its utility in real-world surgical scenarios.

While these additional steps are necessary before advocating for widespread clinical adoption, this nomogram offers a promising avenue for improving decision making in breast cancer surgery. By providing surgeons with more accurate, patient-specific risk information, it has the potential to significantly improve both clinical outcomes and patient satisfaction in breast cancer management. This tool represents a tangible step toward the realization of personalized medicine in surgical oncology, potentially changing the landscape of breast cancer treatment and setting a new standard for intraoperative decision-making tools in cancer care.