Introduction

Breast cancer remains a significant global health burden and is the leading cause of cancer-related incidence among women worldwide and in Vietnam [1]. The cornerstone of effective management is accurate and timely diagnosis, which relies on the “triple assessment” approach: clinical examination, imaging, and pathological evaluation [2]. Within this framework, ultrasound is a critical imaging modality, and the Breast Imaging Reporting and Data System (BI-RADS) is universally used to standardize risk stratification [3]. Lesions classified as BI-RADS 4 (suspicious) and 5 (highly suggestive of malignancy) represent a particularly challenging cohort, requiring definitive pathological confirmation.

Fine-needle aspiration (FNA) cytology is a minimally invasive, rapid, and cost-effective procedure widely utilized for the initial pathological assessment of these suspicious lesions [4,5]. However, variability in cytological interpretation has historically been a limitation. To address this, the International Academy of Cytology (IAC) developed the Yokohama System, a standardized five-tiered reporting framework that categorizes breast cytology findings and links them to a specific risk of malignancy (ROM), thereby guiding clinical management [6].

While the diagnostic accuracy of the IAC Yokohama System has been validated in general populations [7], a critical knowledge gap remains. Its diagnostic precision within the challenging, pre-selected cohort of BI-RADS 4 and 5 lesions – where the pre-test probability of malignancy is already high – is less understood. This distinction is crucial, as a tool’s utility can differ significantly in a screening population versus a high-risk, diagnostic population. Furthermore, it is crucial to understand the interplay of other clinical and imaging factors that may independently predict malignancy in these patients, as this could help refine diagnostic pathways beyond cytology alone. Identifying robust, independent predictors could empower clinicians to better counsel patients on their individualized risk and to make more informed decisions-for instance, justifying an immediate excisional biopsy over a core needle biopsy in a patient with multiple high-risk features [8].

To address these gaps, this study aimed to: (1) evaluate the diagnostic performance of FNA cytology using the IAC Yokohama System in patients with breast lesions classified as BI-RADS 4 and 5; and (2) identify the independent clinical and imaging predictors of malignancy within this specific high-risk population.

Methods

Results

Patient and lesion characteristics

A total of 104 female patients with 104 breast lesions classified as BI-RADS 4 or 5 were included. The final histopathological diagnosis confirmed malignancy in 53 lesions (51.0%) and benign findings in 51 (49.0%).

As summarized in Table 1, patients with malignant lesions were significantly older (p = 0.011) and presented with larger tumors compared to those with benign lesions (p = 0.003 and p = 0.042, respectively). The upper-outer quadrant was the most common tumor location (46.2%). The specific histopathological diagnoses for all lesions, including individual benign entities and a distinction between in situ and invasive malignancies, are provided in S3 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of patients and lesions stratified by final histopathological diagnosis.

https://doi.org/10.1371/journal.pone.0352690.t001

Cytological diagnosis and diagnostic performance

Cytological analysis using the IAC Yokohama System classified 49.0% of lesions as either suspicious for malignancy (Category IV, 45.2%) or malignant (Category V, 3.8%). The distribution across all categories is detailed in Table 2, which also shows the risk of malignancy (ROM) for each category. The ROM for Category IV and V were 95.7% and 100%, respectively. Key cytological features observed in malignant cases included enlarged/pleomorphic nuclei (79.2%), absent or rare myoepithelial cells (75.5%), and prominent nucleoli (56.6%) (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Correlation between fine-needle aspiration cytology (IAC Yokohama System) and final histopathology.

https://doi.org/10.1371/journal.pone.0352690.t002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Representative cytological and corresponding histopathological images.

(A) Benign (Category II): A large, cohesive, branching sheet of ductal epithelial cells with scattered myoepithelial cells. The corresponding histology shows a fibroadenoma with compressed ducts and fibrous stroma. (B) Atypical (Category III): A moderately cellular smear showing discohesive ductal cells with mild nuclear pleomorphism and prominent nucleoli. The corresponding histology reveals an intraductal papilloma. (C) Suspicious for Malignancy (Category IV): A crowded, three-dimensional cluster of ductal cells with enlarged, overlapping nuclei and irregular nuclear contours. The corresponding histology shows ductal carcinoma in situ with a cribriform pattern. (D) Malignant (Category V): A highly cellular smear composed of discohesive, pleomorphic malignant cells with scant cytoplasm and coarse chromatin. The corresponding histology confirms invasive carcinoma of no special type, characterized by nests and cords of tumor cells infiltrating a desmoplastic stroma. Cytology panels (left column) are Diff-Quik stained; histopathology panels (right column) are H&E stained. Scale bar = 100 µm.

https://doi.org/10.1371/journal.pone.0352690.g001

For the purpose of calculating diagnostic performance, Categories IV and V were considered positive results. The overall sensitivity of FNA was 92.5% (95% CI: 82.1–97.9) and specificity was 96.1% (95% CI: 86.5–99.5). The positive and negative predictive values were 96.1% and 92.5%, respectively. The overall accuracy was 94.2% (Table 3). Results were consistent across two pre-specified sensitivity analyses; full details are provided in S4 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Diagnostic performance of fine-needle aspiration cytology for malignancy.

https://doi.org/10.1371/journal.pone.0352690.t003

Independent predictors of malignancy

While univariate analysis identified older age, larger tumor size, and a BI-RADS 5 classification as significant predictors of malignancy, only two factors retained statistical significance in the multivariate model. Older age (aOR 1.13, 95% CI: 1.07–1.19; p < 0.001) and larger tumor size (aOR 1.10, 95% CI: 1.04–1.16; p = 0.001) emerged as strong and independent predictors. The predictive value of a BI-RADS 5 classification was not statistically significant after adjusting for age and tumor size (p = 0.540) (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Univariate and multivariate logistic regression analysis of predictors for malignancy.

https://doi.org/10.1371/journal.pone.0352690.t004

Discussion

This study confirms the high diagnostic utility of fine-needle aspiration (FNA) cytology, when applying the IAC Yokohama System, for risk stratification of suspicious breast lesions classified as BI-RADS 4 and 5. Our key findings demonstrate that while FNA provides excellent sensitivity and specificity, the most robust independent predictors of malignancy in this high-risk cohort are older age and larger tumor size. Notably, the BI-RADS 5 classification-while highly predictive on its own-lost its independent predictive power after adjusting for these two fundamental clinical variables.

The overall diagnostic performance of FNA in our study (sensitivity 92.5%, specificity 96.1%) is consistent with the high accuracy reported in a growing body of international literature. A comprehensive review by Hoda et al. reported a pooled sensitivity and specificity of 96.3% and 98.8% [7], and our results closely mirror findings from large institutional studies in Italy, India, and the United States, confirming the system’s robust reproducibility across diverse populations [11–14].

The risk of malignancy (ROM) for each cytological category in our cohort also aligns with established benchmarks. The ROM for our ‘Suspicious for Malignancy’ (Category IV) and ‘Malignant’ (Category V) groups were 95.7% and 100%, respectively. These findings reinforce that such cytological results are highly indicative of cancer and warrant definitive histopathological examination, a conclusion supported by multiple validation studies [11,15]. This underscores the system’s reliability and applicability within the Vietnamese clinical context for guiding patient management.

The most significant finding of our study emerged from the multivariate analysis. While older age and larger tumor size are well-established risk factors for breast cancer [16,17], our model demonstrates they are powerful independent predictors that can overshadow the predictive value of BI-RADS classification in a pre-selected high-risk population.

The finding that BI-RADS 5 lost its statistical significance (p = 0.540) after adjusting for age and tumor size is particularly intriguing. This suggests that the morphological features on ultrasound that lead to a BI-RADS 5 classification (e.g., spiculation, irregular shape, posterior acoustic shadowing) may be biologically linked to the characteristics of tumors that develop in older women or that have grown to a larger size. Indeed, studies have shown that tumor size is a strong determinant of sonographic features, with larger tumors more likely to exhibit malignant characteristics [18–20].

Therefore, it is plausible that in a multivariate context, age and size act as more fundamental proxies for the underlying tumor biology. The sonographic features defining BI-RADS 5 may simply be the morphological manifestations of tumors that have had more time to grow (larger size) in a host environment conducive to progression (older age), thereby making age and size the more direct measures of malignant potential. This phenomenon, where clinical or pathological variables can attenuate the significance of imaging scores, highlights the importance of multi-faceted risk assessment [18,21,22].

Our study has several strengths. It is one of the first in Vietnam to rigorously evaluate the Yokohama system within a specific, clinically relevant cohort of BI-RADS 4 and 5 lesions. Furthermore, the use of multivariate logistic regression provides a deeper, more nuanced understanding of the interplay between clinical, imaging, and cytological factors.

However, we must also acknowledge certain limitations, as recommended by the STROBE guidelines [18,23]. First, the study was conducted at two major tertiary centers, which may limit the generalizability of our findings to provincial or community-level healthcare settings. Second, our sample size (n = 104), particularly for the BI-RADS 5 subgroup (n = 11), is relatively small. Retroactive expansion of the dataset was not feasible due to our strict inclusion criteria requiring matched high-quality FNA and definitive histopathology to ensure diagnostic accuracy. A post-hoc power analysis indicated that while the study achieved robust statistical power for tumor size (89.5%), the power for patient age was moderate (67.2%), and the BI-RADS 5 subgroup remained underpowered. This limitation resulted in wide confidence intervals for BI-RADS 5 in the multivariate model, raising the possibility of a Type II error for this specific variable [24]. Consequently, findings regarding the loss of independent predictive value for BI-RADS 5 should be interpreted as preliminary. Third, although cytological slides were reviewed by two pathologists with a consensus approach for discordant cases, potential interobserver variability remains an inherent limitation of morphological assessment. Fourth, subcategory data for BI-RADS 4 lesions (4A, 4B, and 4C) were unavailable, as structured subcategory reporting was not consistently documented in the institutional radiology records during the retrospective data collection period (2021–2023). This limits the granularity of the BI-RADS analysis, as the three BI-RADS 4 subcategories carry substantially different risks of malignancy (4A: ≥ 2% to <10%; 4B: ≥ 10% to ≤50%; 4C: > 50% to <95%). Future prospective studies should incorporate BI-RADS 4 subcategory classification to enable a more refined analysis of imaging-based malignancy predictors within this heterogeneous group.

Our findings have important clinical implications. FNA cytology based on the Yokohama system is a robust and reliable tool for the initial assessment of high-risk breast lesions in Vietnam. Clinicians and pathologists should recognize that for lesions already classified as BI-RADS 4 or 5, patient age and tumor size are potent, independent indicators of malignancy. This finding suggests that in resource-limited settings where further molecular testing may not be immediately available, these readily accessible clinical parameters can significantly aid in counseling patients and prioritizing cases for expedited biopsy or surgical consultation. This highlights the necessity of integrating clinical data with imaging and cytology for the most accurate risk assessment, a principle supported by major clinical guidelines such as those from the NCCN and ASCO [25,26]. Future research should include larger, multi-center prospective studies to validate our findings on the predictive roles of these variables and explore their molecular profiles [27].

Conclusion

In patients with suspicious breast lesions classified as BI-RADS 4 and 5, fine-needle aspiration cytology evaluated by the IAC Yokohama System is a highly accurate and reliable diagnostic tool. Importantly, our findings demonstrate that older age and larger tumor size are strong, independent predictors of malignancy in this high-risk cohort. These results suggest that in resource-limited settings like Vietnam, where molecular testing may not be immediately available, integrating these readily accessible clinical parameters with imaging and cytology is essential for optimizing risk stratification. Of note, the non-significant result for BI-RADS 5 in the multivariate model (p = 0.540; aOR 2.10, 95% CI: 0.20–22.73) should not be interpreted as evidence that BI-RADS 5 classification lacks clinical value; rather, it reflects the limited statistical power of the small BI-RADS 5 subgroup (n = 11), which rendered this estimate susceptible to Type II error. This approach can help clinicians prioritize patients for expedited biopsy or surgical intervention, ensuring timely and appropriate management.

Supporting information

Acknowledgments

The authors would like to thank the staff of the Departments of Pathology at Military Hospital 103 and 108 Military Central Hospital for their technical assistance and support during data collection.