Association of sonographic features and molecular subtypes in predicting breast cancer disease outcomes

Abstract Background Features in preoperative ultrasound could predict the prognosis of triple‐negative breast cancer (TNBC), while its prognostic value in other molecular subtypes of breast cancer (BC) was unknown. The study aimed to assess the prognostic value of preoperative sonographic features, including orientations, on long‐term outcomes in BC and its association with different molecular subtypes. Methods Women diagnosed with invasive BC > 5 mm who underwent surgery were retrospectively reviewed. Clinical, pathological, and sonographic profiles were collected and recurrence‐free survival (RFS) and breast cancer‐specific survival (BCSS) were reported. Interactions between clinicopathological features and tumor orientations in predicting RFS and BCSS were analyzed. Competing risk model was performed to estimate prognostic values of sonographic features for RFS and BCSS. Results A total of 2812 patients were included. With a median follow‐up of 60.0 months, 268 (9.5%) patients suffered from recurrences and 104 (3.7%) died of BC. The prognostic values of vertical orientation in predicting RFS (P = .001) and BCSS (P = .001) were strongly associated with molecular subtypes. Non‐TNBC tumors with vertical orientation had less recurrence events compared with parallel orientation (6.3% vs 8.7%, P = .035), whereas failed to predict disease outcomes in multivariate analysis (P > .05). Oppositely, in TNBC, vertical orientation was associated with worse RFS (HR = 3.50; 95% confidence interval [CI] 1.69‐7.24; P < .001) and BCSS (HR = 6.36; 95% CI 2.86‐14.14; P < .001) in multivariate analysis with a 5‐year RFS and BCSS of 73.4% and 74.6%. Meanwhile, vertical orientation was related with smaller tumor size (P < .001), human epidermal growth factor receptor 2 nonamplification (P < .001), and lower Ki‐67 expression (P = .001) among non‐TNBC population, whereas TNBC tumors with vertical orientation had a higher burden of axillary lymph node metastases (2.8 ± 1.0 vs 1.4 ± 0.2, P = .001). Conclusions Prognostic values of sonographic orientation in predicting BC disease outcomes were associated with molecular subtypes. Vertical orientation in preoperative sonogram may serve as a prognostic biomarker for TNBC patients.


| INTRODUCTION
Breast cancer (BC), as one of the major reasons of morbidity and mortality for women worldwide, is a heterogenous disease with variations in biological characteristics and clinical outcomes. 1,2 Researchers have introduced multiple prognostic biomarkers to predict recurrence risks and guide optimal treatment. [3][4][5][6][7] Clinicopathological features including larger tumor size, presence of axillary lymph node (ALN) metastases, younger age, and higher histological grade have been proved to be associated with higher risk of recurrence. [3][4][5] Classification of molecular subtypes based on the expression of estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and proliferation levels also helps predicting recurrence patterns and tailoring more personalized therapy for BC patients. 6,7 Aside from traditional clinicopathological traits, combined models of risk factors and gene panels have been established to further determine prognosis. For instance, Composite Risk model have been proved to predict long-term outcomes for patients receiving adjuvant treatment, whereas the scoring system incorporating pretreatment clinical stage and post-treatment pathologic stage as well as estrogen receptor status and tumor grade (CPS+EG system) showed predictive values in neoadjuvant settings. 8,9 Meanwhile, multiple gene arrays as 21-gene recurrence score (RS) could identify patients of higher risk and optimize the choice of therapy. 10 Furthermore, novel biomarkers have thrown fresh light on understanding biological behaviors of BC. 11,12 Expression level of tumor infiltrating lymphocytes and programmed death-ligand 1 showed capability to predict prognosis 11 and treatment response 12 in BC, especially in triple-negative breast cancer (TNBC). However, to further exploring the heterogeneous intrinsic profiles of BC, more biomarkers should be brought into practice.
Breast imaging may provide extra information of tumorous features for clinicians. Examinations including ultrasonography (US), mammography (MG), and magnetic resonance imaging (MRI), have been widely applied in screening and diagnosis of BC. 13,14 Compared with MG and MRI, advantages of US included well tolerance, no radiation, and cost-effectiveness. What's more, US harbored higher accessibility in daily practice, which offered researchers more data resource to investigate. Studies have been done to evaluate the prognostic value of preoperative sonography features. [15][16][17] It was reported that BCs classified as Breast Imaging Reporting and Data System (BI-RADS) 4A category in screening US had higher risk of recurrence because tumors of higher proliferation may mimic the features of benign lesion. 15 What's more, nonmass lesions with calcification 16 have been previously reported to be associated with worse clinical outcomes. Notably, in our previous study, vertical orientation in preoperative ultrasound was found to be independently associated with inferior outcomes and higher ALN burden in TNBC patients. 17 However, there was relatively limited data concerning whether sonographic features could predict disease outcomes in all subtypes of BC.
Thus, our study was to investigate the prognostic value of sonographic features in all subtypes of BC patients and further explore whether its prognostic value was associated with different molecular subtypes of BC.

| Patient cohort
Consecutive patients who received surgery and systematic adjuvant treatment in the breast health center in our university between January 2009 and December 2015 were retrospectively reviewed. Patients diagnosed pathologically as invasive BC larger than 5mm with preoperative ultrasound record in the hospital were eligible for inclusion. Exclusion criteria included male BC, simultaneous bilateral BC, neoadjuvant treatment, history of breast surgery, history of other malignancy, and diffusive or occult lesions in ultrasound because of their interference in image morphology and disease outcomes. The protocol was conducted under the terms of the Declaration of Helsinki and has been reviewed and approved by the Ethical Committee Review Board of Institution.

| Ultrasound imaging technique and analysis
Preoperative sonograms were conducted by two physicians in the Ultrasonography Department majored in breast imaging with more than 10 years of experience. Images were collected by MyLab60 (Esaote) or Philip HD15 (Philips) with probes of 5-12 MHz and then stored in the system of Digital Imaging and Communications in Medicine. Features including tumor orientation, shape, margin status, calcification, posterior acoustic patterns, architectural distortion, changes in Cooper's ligament, and color Doppler flow imaging were assessed and reported under the norm of 5th American College of Radiology (ACR) BI-RADS Atlas. 18

| Pathological evaluation
Breast tumors were removed by surgery and evaluated by the Pathology Department. Estrogen receptor and PR

| Data collection and follow-up
Clinical characteristics, pathological data, and follow-up information of study population was recorded and retrieved from the BC database of the university. Clinicopathological variables were collected as follows: patients' age, menstrual status, family history, co-morbidity, surgery types of breast and axillary, tumor size, histopathological types, ALN involvement, histological grade, ER status, PR status, HER2 status, lymphovascular invasion, Ki-67 index, and adjuvant treatments. In this study, histopathological types were classified as invasive ductal carcinoma (IDC) and non-IDC; histological grade was categorized as I-II vs III and Ki-67 index as ≤14% vs >14%.
Patients' follow-up was done by BC-specialized nurses and two endpoints including recurrence-free survival (RFS) and breast cancer-specific survival (BCSS) were selected into analysis. RFS was defined as the interval between the date of surgery and the date of locoregional recurrence, distant relapse, or contralateral BCs. BCSS was identified as the time period from BC surgery to the occurrence of BC-specific deaths.

| Statistical analyses
Analysis was conducted by IBM SPSS Statistics 25.0 (Windows version). Baseline characteristics were shown as numbers and percentages for categorical variables and as means and standard deviations for continuous variables. Comparison of categorical variables among subgroups was analyzed by Pearson's Chi-square test (or Fisher's exact test when necessary), whereas continuous variables by independent sample t test. Stratified Mantel-Haenszel test was conducted between sonographic orientation and clinicopathological features for RFS and BCSS and interaction P value was recorded. To avoid the influence of competing risk in survival analysis, we conducted competing risk analysis with "cmprsk" package in R (Windows 3.6.3 version). Gray's test was performed for univariate analysis and P value was recorded. Sonographic variables with P < .1 in univariate analysis and clinicopathological factors with potential prognostic value were then taken into multivariate analysis with Fine-Gray model. Multivariate results were presented as subdistribution hazard ratio with 95% confidence interval (CI) with corresponding P values. Kaplan-Meier curves was plotted for sonographic features in predicting patients' outcomes. All the tests were two-sided and P < .05 was considered as significantly important.

| Demographics and prognosis in the whole population
A total of 3477 BC patients underwent surgery without neoadjuvant treatment, among which 2812 were finally included in the study ( Figure 1). Characteristics of the patients and the tumors were listed in Table 1. The median age of the cohort was 55 years old (range 23-93 years). One thousand five hundred sixty-four (55.6%) of the patients had tumors ≤2.0 cm and 1002 (35.6%) had node positive diseases. There were 2481 (88.2%) patients had IDC and 1124 (40.0%) with grade III tumors. One thousand eight hundred ninety (67.2%) patients had hormone receptor-positive diseases, whereas TNBC and HER2-emplified BC were found in 416 (14.8%) and 293 (10.4%) cases, respectively. In total, 1988 (70.7%) patients underwent adjuvant chemotherapy, among which 1209 patients received regimens of anthracycline plus taxanes. The median time interval between preoperative sonography and breast surgery was 6 days and was similarly distributed among patients of different molecular subtypes and different sonographic orientations (P = .730).

| Sonographic features predict patients' outcomes in different subsets of BC
Univariate analyses of sonographic orientations and patient outcomes in total population and subgroups of BC were demonstrated in Table 2. Vertical orientation failed to predict RFS (P = .881), but was significantly associated with inferior BCSS (P = .032). When it came to subgroup analysis, in TNBC patients, vertical orientation was independently predictable for both worse RFS (P = .003) and BCSS (P < .001) with a 5-year RFS and BCSS of 73.4% and 74.7%, respectively, which was worse than TNBC patients with parallel orientation feature (5-year RFS 89.0%, BCSS 94.1%) ( Figure 3). Similar results were also found in ER-negative BC that vertical orientation was associated with unfavorable RFS (P = .015) and BCSS (P < .001) ( Figure S1). On the other hand, in non-TNBC patients, tumors with vertical orientation showed superior RFS compared with parallel orientation (5-year RFS 96.0% vs 92.9%, P = .035), whereas orientations were not predictable for BCSS (P = .207) ( Figure 3). Additionally, prognostic effects of sonographic orientation on RFS and BCSS were analyzed in Luminal-A like, Luminal-B like, and HER2-enriched BC relatively. Vertical orientation failed to predict disease outcomes in those subtype of BC (All P > .05) (Table S3; Figure S2).

| DISCUSSION
In this study, the prognostic value of detailed sonographic features was explored and compared in all molecular subtypes of BC. It was found that the effects of tumor orientation in ultrasound in predicting RFS and BCSS were significantly associated with molecular subtypes. In TNBC patients, vertical orientation was independently associated with unfavorable RFS (HR = 3.50; 95% CI 1.69-7.24; P < .001) and BCSS (HR = 6.36; 95% CI 2.86-14.14; P < .001); oppositely, vertical orientation showed no prognostic value for non-TNBC patients although it was strongly correlated with less proliferative pathological features. It was indicated that the prognostic value of sonographic orientation was associated with molecular subtypes and vertical orientation can be served as a prognostic biomarker for TNBC patients. Efforts have been made in exploring the predictive and prognostic value of preoperative sonography features in BC patients while few consensuses have been reached by now. Kim SY et al reviewed 501 BC patients detected at screening ultrasound and demonstrated that tumors classified as BI-RADS 4A category were associated with higher risk of recurrence compared with other categories (HR = 5.11, 95% CI 1.53-17.20; P = .008). 15 However, detailed features of ultrasound were not evaluated in this study. In our study, tumors with BI-RADS 4A category failed to predict disease outcomes in terms of RFS (P = .080) and BCSS (P = .134). The conflicting results between two studies may be explained by the different proportion of BI-RADS 4A tumors that 30% (129/425) of tumors in Kim's cohort were assessed as BI-RADS 4A, whereas only 4.6% (128/2812) were 4A category in our study. Oppositely, compared with BI-RADS 4 category (including 4A, 4B, and 4C), tumors with BI-RADS 5 indicated unfavorable RFS (P < .001) and BCSS (P < .001) in univariate analysis. In terms of specific sonographic features, patterns of blood flow after treatment including peak systolic velocity, pulsatility index, and resistive index have been reported to be a surrogate predictor for treatment response and disease-free survival in BC patients. 21,22 However, as one of the most important prognostic biomarkers, molecular subtypes were not put into analysis in these studies. Additionally, in a large cohort of 3112 patients from Korea, nonmass lesions with calcification at US were found to be independently related with inferior RFS (HR = 1.4, 95% CI 1.1-1.8; P = .01), 16 whereas its predictive value was not validated in different molecular subtypes of BC. Our previous study specifically focusing TNBC patients found that vertical orientation in preoperative ultrasound was associated with worse outcomes in terms of RFS and BCSS. 17 However, its prognostic role in other molecular subtypes was still unknown. T A B L E 2 Univariate analysis of orientation and patient outcomes in subgroups of BC patients Thus, to further explore the prognostic value of tumor orientation in BC and its association with molecular subtypes, we conducted this study in a larger cohort of 2812 BC patients. It was demonstrated that the prognostic effect of tumor orientation was significantly associated with molecular subtypes (TNBC vs non-TNBC) in predicting RFS (P interaction = 0.001) and BCSS (P interaction = 0.001). Vertical orientation in TNBC was independently associated with worse RFS and BCSS, which was consistent with our previous study. 21 Although similar results were found in ERnegative BC, most RFS (14/17), and BCSS (12/14) events in vertical subgroup were among TNBC patients (Table 2), indicating the unfavorable prognostic value of vertical orientation in ER-negative BC was mostly due to the TNBC cohort. Interestingly, for non-TNBC patients, vertical orientation was associated with favorable RFS (P = .035) in univariate analysis, whereas had no predictive value in multivariate analysis. Furthermore, vertical orientation did not show significant prognostic values for RFS and BCSS (All Vertical orientation, which meant the lesion was oriented taller than wide according to ACR Reporting system, was able to reflect the pathological features of BC. 23 It was normally interpreted as high proliferation which led to expansive growth against peripheral tissues 24 and thus was always considered as a risk sign for BC. 25,26 Meanwhile, studies have also manifested that vertical orientation was significantly associated with more invasive proportion of tumor 27 and higher level of clinical risk in BC. 28 However, evidence concerning vertical orientation and clinical outcomes in BC was inconsistent. 29 Chae et al investigated a 267 cohort of ER+/HER2-invasive BC and concluded that parallel orientation was an independent predictor for higher Oncotype DX RS (OR = 5.53; P = .02). 29 Consistent with this conclusion, parallel orientation was also associated with vicious tumorous behavior in univariate analysis in our non-TNBC cohort. What's more, vertical orientation had strong relevance with smaller tumor size, HER-2 negativity, and lower Ki-67 T A B L E 3 Univariate analysis of sonographic features and clinical outcomes in total population, non-TNBC and TNBC patients level among non-TNBC patients in our study, which indicated less proliferative behaviors and potentially favorable prognosis. Similarly, several studies have found out that tumors with higher proliferation level may present as regular shapes, parallel orientations, and circumscribed margins. It was hypothesized that those lesions mimicked the morphology of benign lesions because of their rapid cellular proliferation. [30][31][32] One of the most interesting findings of our study was the association between sonographic orientation and molecular subtypes in predicting BC outcomes. This may be explained by diverse imaging patterns among different molecular subtypes of BC. As shown in previous studies, TNBC had distinguishing ultrasound morphology including circumscribed margins, regular shapes, and posterior acoustic enhancement patterns compared with non-TNBC tumors because of its distinct biological behavior. [33][34][35] Furthermore, studies have found that radiomic phenotypes were associated with genomic pathways and protein expressions which contributed to tumor development. 36,37 Thus, certain sonographic features may reflect distinctive biological profiles in different subgroups of BC. In our study, vertical orientation represented favorable features including smaller size and lower level of Ki-67 in non-TNBC while was associated with a higher ALN burden among TNBC patients. Radiogenomic research may be warranted to further understand the clinical implications of sonographic orientation in different molecular subtypes of BC.
Regarding inclusion criteria, patients who underwent neoadjuvant therapy were excluded from our study population.
The reasons were that neoadjuvant therapy could bring alteration of pathological profiles to BC, including hormone receptor status, HER-2 status, and Ki-67 index, which caused the change in molecular subtypes and would further influence patients' long-term prognosis. [38][39][40] Additionally, neoadjuvant treatment would bring change to the sonographic morphology of breast tumors, which would cause bias to the preoperative imaging features. Thus, prognostic values of sonographic features in neoadjuvant settings should be analyzed separately from adjuvant settings.
There were certainly several limitations in this study. First, it was retrospectively designed and conducted in single institution, which may lead to selective and treatment bias among the study population. Second, 213 (7.6%) patients with IHC HER2 2+ lacked further FISH tests, causing a relatively large proportion of undetermined HER2 status and group of patients with unclassified molecular subtype in our population. Additionally, applications of novel features including quantitative ultrasound 41 and shear wave elastography 42 were unavailable in this study, which may better predict treatment response and survival in BC.

| CONCLUSIONS
In conclusion, our study demonstrated that the prognostic value of sonographic orientation in predicting disease T A B L E 4 Multivariate analysis of sonographic and clinicopathological features for RFS and BCSS in non-TNBC patients outcomes was associated with molecular subtypes. Vertical orientation was independently associated with inferior prognosis in TNBC. Hence, vertical orientation could be recognized as a candidate prognostic risk factor for TNBC, deserving further clinical evaluation.