Pathological underestimation and biomarkers concordance rates in breast cancer patients diagnosed with ductal carcinoma in situ at preoperative biopsy

Ductal carcinoma in situ (DCIS) often upgrade to invasive breast cancer at surgery. The current study aimed to identify factors associated with pathological underestimation and evaluate concordance rates of biomarkers between biopsy and surgery. Patients diagnosed with DCIS at needle biopsy from 2009 to 2020 were retrospectively reviewed. Univariate and multivariate analyses were performed to identify factors associated with pathological underestimation. Concordance rates between paired biopsy samples and surgical specimens were evaluated. A total of 735 patients with pure DCIS at biopsy were included, and 392 patients (53.3%) underwent pathological underestimation at surgery. Multivariate analysis demonstrated that tumor size > 5.0 cm [odds ratio (OR) 1.79], MRI BI-RADS ≥ 5 categories (OR 2.03), and high nuclear grade (OR 2.01) were significantly associated with pathological underestimation. Concordance rates of ER, PR, HER2 status and Ki-67 between biopsy and surgery were 89.6%, 91.9%, 94.8%, and 76.4% in lesions without pathological underestimation, and were 86.4%, 93.2%, 98.2% and 76.3% for in situ components in lesions with pathological underestimation. Meanwhile, in situ components and invasive components at surgery had concordance rates of 92.9%, 93.8%, 97.4%, and 86.5% for those biomarkers, respectively. In conclusion, lesions diagnosed as DCIS at biopsy have a high rate of pathological underestimation, which was associated with larger tumor size, higher MRI BI-RADS category, and higher nuclear grade. High concordances were found in terms of ER, PR, and HER2 status evaluation between biopsy and surgery, regardless of the pathological underestimation.

Photographs for discordant IHC results in terms of PR, HER2, and Ki-67 in patients without pathological underestimation were shown in Figure S1.  2B). And the overall agreements were also good for receptor status (ER: κ = 0.712, PR: κ = 0.864, HER2: κ = 0.957) and moderate for Ki-67 results (κ = 0.522, Table 4). The molecular subtype had concordance rate of 72.9%, with only good agreement (κ value = 0.611). HER2 positive (non-luminal) DCIS had a concordance rate of 86.5% (90 out of 104 patients) between biopsy and surgery, which ranked the highest among five molecular subtypes. Luminal A-like DCIS only had a concordance rate of 63.3% (19 out of 0 patients, Table S2).

Concordance between biopsy and surgery for biomarkers in patients with
Evaluation of ER, PR, HER2, and Ki-67 at biopsy had a concordance rate of 85.0%, 87.1%, 94.3%, 71.0% with the invasive component at surgery (Fig. 2C). The overall agreements were good for receptor status (κ = 0.699, κ = 0.730, and κ = 0.870, respectively) and moderate for Ki-67 results (κ = 0.412, Table 4). The molecular subtype had a good agreement with a κ value of 0.635, and the concordance rate between biopsy and surgery was 75.0%. Among the 11 patients diagnosed with luminal B-like (HER2-negative) DCIS at biopsy only 21 (63.6%) remained the same molecular subtype at surgery. HER2-positive (non-luminal) lesions showed the highest concordance rate (86.3%, 44 out of 51 patients) among five subtypes (Table S3).
In patients who underwent pathological underestimation, discordance in IHC results was shown in Figure S2.  (Table S4).

Concordance between in situ and invasive component for biomarkers in IDC + DCIS patients.
The Figure S3 showed the discordant IHC results in terms of ER in synchronous DCIS and IDC within a surgery sample.

Discussion
In the current study, we included 735 patients with DCIS at CNB and demonstrated that pathological underestimation was independently associated with tumor size, MRI BI-RADS category, and nuclear grade at CNB. Good concordances were observed between CNB and surgery in terms of ER, PR, and HER2 status irrespective of the presence of pathological underestimation or not. Whereas, there were more Ki-67 highly expressed lesions detected at surgery compared with at CNB. Regarding lesions with DCIS and IDC components, the overall agreement was also good for ER, PR, HER2, and Ki67 analysis between in situ and invasive components. The majority of the patients (94.0%) in our center were biopsied with 14-G CNB, and the rate of pathological underestimation was 53.3%. However, previous studies reported that the underestimation rate was approximately 15-20% by using the VAB 13,14 . The method of the biopsy would influence the presence of pathological underestimation. Kim reported that the underestimation rate was 49.8% for the CNB and 29.2% for the vacuum method (P < 0.001) 8 . And in Park's study, the underestimate rate was 50.0% and 18.8%, respectively (P < 0.001) 15 . Our result was consistent with previous data for the CNB cohort, and univariate analysis also showed that the underestimate rate was lower in patients who received VAB than CNB. Indeed, VAB by using Mammotome (8-gauge) and Mammotome elite (13-gauge) could provide more tissue samples, which can increase the diagnostic accuracy of biopsy. However, the Mammotome is not covered by the Chinese health insurance or government, and the Mammotome elite was just covered by the health care program at the end of the year 2018. Thus, when considering the treatment cost, both surgeons and patients may prefer 14-gauge CNB over VAB. Furthermore, using stereotactic biopsies instead of sonographic guidance may lower the risk of pathological underestimation (21.8% vs. 39.9%) 5 . It was acknowledged that DCIS is always associated with microcalcifications but not mass-like lesions. As was reported, mammography screening could increase the recall rate of suspicious microcalcifications, as well as the incidence of DCIS. However, there was lack of mammography national screening programs in China, thus the www.nature.com/scientificreports/ most common complaint was breast lump. In the current study, nearly 90% of the patients had mass-like lesions, thus we used the ultrasound-guided biopsy to diagnosis breast disease. Notably, even with an underestimation rate of 15%, it is still not appropriate to use the pathological results of biopsy specimen as a definitive diagnosis, as well as to safely guide the active surveillance or omit the axillary surgery. Therefore, identifying patients who are at high risk of pathological underestimation is of great importance. Several factors were reported to be associated with pathological underestimation in previous studies, including palpability, high nuclear grade, BI-RADS category 5, HER2 positivity, Ki-67 overexpression, suspicious invasion, mammographic mass finding, and radiological tumor size ≥ 2.0 cm 5,7-12 . In the current study, we found that MRI BI-RADS ≥ 5, maximum mass diameter > 5.0 cm, and high nuclear grade were independent predictors for pathological underestimation. The presence of high nuclear grade and at CNB have been shown to be associated with an invasive component at surgical specimen 5,16 . Meanwhile, it has also been reported that high-grade DCIS  www.nature.com/scientificreports/ developed to the invasive tumor more quickly than low-grade lesions 17 . This may indicate that high-grade DCIS was at a more advanced stage during progression, thus was more likely to harbor co-existing invasive component. Regarding the imaging feature, previous studies have suggested that breast MRI was highly sensitive for detecting and evaluating breast cancer of various types. DCIS often manifests as non-mass-like enhancements with segmental or ductal distribution, and presents clumped internal architecture in the MRI imaging 14,[18][19][20][21] . The common manifestation of DCIS included lump as well as segmental distribution in the current study. Kim et al. reported that tumor size > 2.0 cm was an independent predictor of underestimation, which was consistent with the present findings 8 . A plausible reason may be that the larger target area for sampling increases the possibility of sampling error 11 . This suggests that increasing the amount of sample collection by using a thicker needle or increasing the number of cores examined may help to improve the diagnostic accuracy for large tumors. Treatment patterns of breast cancer, including DCIS, relies on molecular subtype, which was determined by ER, PR, HER2 status, and Ki-67 index [22][23][24][25][26] . However, unlike invasive breast cancer, the IHC assessment of biomarker expression is less performed for DCIS at CNB. In the current study, in patients with a final diagnosis of DCIS at surgery, we observed good concordance rates regarding ER (89.6%), PR (91.9%), and HER2 (94.8%) status. Whereas, Ki-67 only showed a moderate agreement. Similar results had been constantly reported for invasive carcinoma. Tamaki et al. reported that CNB can provide reliable information on ER (κ = 0.82), PR (κ = 0.66), and HER2 (κ = 0.64) status of patients 26 . A meta-analysis study also demonstrated the high diagnostic accuracy of CNB in evaluating ER, PR, and HER2 status compared with open excision biopsy in breast cancer patients 27 . The poor agreement in terms of Ki-67 had also been recorded for invasive breast cancer 28 , which may due to the poorer fixation of surgical specimens compared with CNB ones and intra-tumor heterogeneity 29 . Meanwhile, the wound response of biopsy may also cause an increase in Ki-67 expression, since the invasive diagnostic procedure may accelerate the tumor growth 30 . Another point is the highly inconsistent measurement of Ki-67 among testing laboratories and pathologists, and further efforts were needed to set universally recognized cutoff for this biomarker.
In patients with a pathological underestimation, both the in situ components and the invasive components in surgery specimens showed good agreements (κ > 0.6) with CNB samples in terms of ER, PR, and HER2 status, which has rarely been evaluated in previous studies. Additionally, high concordance rates were also observed in terms of ER (92.9%), PR (93.8%), HER2 (97.4%), Ki-67 (86.5%) between the invasive component and in situ component at the surgery. Several studies investigated the genetic alteration between the in situ and invasive components of a tumor and found that they share a high degree of similarity [31][32][33][34] . Meanwhile, Schuetz et al. investigated the biology of transition from DCIS to IDC and identified progression-specific candidate genes 35 . Those to some extents indicated that the co-existing IDC was evolved from the in situ component, but not due to the intra-tumor heterogeneity. Thus, for tumors with micro-invasive foci that lack sufficient sample to complete the pathological analysis, IHC results from the in situ component can be used for tailoring subsequent treatment.
In the current study, we evaluated factors associated with pathological underestimation in a large cohort of patients diagnosed with pure DCIS at CNB and compared the concordance rate of IHC results between CNB and surgery. However, there were several potential limitations. First, selection bias may serve as an inevitable problem since this was a retrospective study, although we performed multivariate analysis to narrow this effect. Second, with data from a single institute, our results may not be appropriate when being extrapolated to other populations, and multicenter studies warrant consideration. Last but not least, the biopsy method, as well as the number of cores examined, were not considered in the current study. Thus, our results may only have meaning to estimate the upgrade of DCIS by using CNB, and further efforts were needed to evaluate the pathological underestimation by using VAB.
In conclusion, for patients diagnosed with DCIS at CNB, those who had MRI BI-RADS ≥ 5, high nuclear grade, and tumor size ≥ 5.0 cm were more likely to underwent pathological underestimation. ER, PR and HER2 status showed a high concordance rate between CNB and surgery, regardless of the presentation of pathological underestimation. www.nature.com/scientificreports/ to the ACR BI-RADS® Atlas Fourth Edition (before 2013) and ACR BI-RADS® Atlas Fifth Edition (since 2013) 36 . The maximum mass diameter was defined by using ultrasound imaging examination.

Methods
CNB, vacuum-assisted biopsy (VAB), and surgical specimen. The CNB were performed by using 14-gauge (G) automated biopsy guns (Magnum, BARD, Covington, U.S). The VAB were performed by using the Mammotome system with 8-G biopsy guns (Mammotome EX, Devicor Medical Products,Inc., USA) or 13-G biopsy guns (Mammotome elite, Devicor Medical Products,Inc., USA). The procedure was guided by sonographic device (sonosite S-Women's health). CNB samples were fixed immediately in adequate volume of 4% buffered formaldehyde for at least 6 h according to the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guidelines, and embedded in paraffin for histopathological analysis 37 . The surgical specimen was cut into 1-cm-thick slices, fixed within 1 h, and followed by paraffin embedding.
Pathological and IHC analysis. Pathological  Statistical analysis. Categorical variables between patients with and without underestimation were compared by using the Fisher's exact test. Univariate and multivariate logistic regression analyses were performed to identify clinicopathological characteristics related to pathological underestimation with odds ratio (OR) and 95% confidence interval (CI). Concordance analysis of receptor status and Ki-67 between CNB and surgical specimen was calculated by the kappa test. A κ value > 0.6 is considered as having with good agreement, value between 0.4 and 0.6 suggests moderate agreement, value < 0.4 indicates fair agreement, and value < 0.2 indicates poor agreement. Two-sided P values < 0.05 were considered statistically significant. Statistical analyses were carried out by using the IBM SPSS statistics software version 25.0 (SPSS, Inc., IL, USA) and graphing were conducted by GraphPad Prism version 7.0 (GraphPad Software, CA, USA).

Ethics approval. The study was approved by the Ethical Committees of Ruijin Hospital, Shanghai Jiao
Tong University School of Medicine. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.