Novel Nomograms to Predict of Overall Survival and Cancer-Specic Survival of Patients of Metaplastic Breast cancer

Background Metaplastic breast cancer (MBC) is a rare type of breast cancer with an increasing incidence, we aim to develop clinical nomograms to predict the overall survival and cancer-specic survival for patients with MBC. Patients data were collected from the SEER database between 1973 and 2015. All included patients were randomly assigned into the training and validation sets. Univariate and multivariate Cox analysis were performed to identify independent prognostic factors of MBC. These essential prognostic variables were combined to construct nomogram models to predict overall survival (OS) and cancer-specic survival (CSS) in patients with MBC. Model performance was evaluated by concordance index (C-index) and calibration plots. A total of 1129 patients were collected and divided into the training (753) and validation (376) groups. The multivariate Cox model identied age, stage_ajcc, T stage, chemotherapy and radiotherapy as independent covariates associated with OS, while age, race, stage_ajcc, T stage, and radiotherapy were independent prognostic factors of CSS. The nomogram constructed based on these covariates demonstrated excellent accuracy in estimating 3-, and 5-year OS and CSS, with a C-index of 0.744 (95% CI, 0.701-0.787) for OS and 0.746 (95% CI, 0.695-0.797) for CSS in the training cohort. In the validation cohort, the nomogram-predicted C-index was in OS and 0.818 for OS (95% CI, 0.775-0.861) and 0.800 (95% CI, 0.747-0.853) for CSS. All calibration curves exhibited good consistency between predicted and actual survival.


Background
Metaplastic breast carcinoma (MBC) is a relatively rare form of breast cancer with worse clinical outcomes and resistance to neoadjuvant systemic chemotherapies [1], accounting for 0.2-5% of all breast cancers [1]. The incidence of MBC is increasing since it was recognized as a distinct pathological diagnosis in 2000 [3]. Histologically, MBC is classi ed into several subtypes, including spindle, squamous, chondroid, osseous and/or rhabdomyoid MBC [4]. MBC commonly shows a triple negative breast cancer (TNBC) phenotype, due to the lack of expression of the ER, PR and HER2 [5], and is managed with surgical resection in combination with radiotherapy and chemotherapy [1]. However, only radiotherapy shows improvement on overall survival (OS) in MBC patients [6]. Compared with invasive ductal carcinoma, the 5-year survival rate for MBC remains poor owing to its rapid tumor growth rate and chemoresistance [7,8]. The Surveillance, Epidemiology, and End Results database (SEER database) is a well-constructed database from multiple institutions in the United States (US), covering about 30% of the US population. It has become a distinctive resource to investigate special malignancies, such as MBC, by taking advantages of its wide range of data on cancer.
Nomograms have been proposed as a novel and dependable tool to incorporate demographic and clinicopathologic factors for accurate prognostic prediction of many cancers. They were generated from regression analysis and showed to compare favorably to the standard TNM staging systems. Currently, to our knowledge, there was no available nomograms for individual MBC patients derived from populationbased data. Herein, we aim to establish a novel nomogram to forecast individualized survival of MBC depended on the personalized demographic, pathologic and therapeutic information from the SEER database based on our previous works posted on Research Square as preprints [9,10].

Patient population
The SEER database supported by the National Cancer Institute (NCI) is a population-based cancer registry covering approximately 30% of the United States population [12]. Date of patients with MBC diagnosed between 1973 and 2015 was obtained from the SEER program of the National Cancer Institute. The variables of interest extracted from the SEER database included age at diagnosis, race, tumor size, grade, status of ER, PR and HER2, American Joint Committee on Cancer (AJCC) tumor stage, T stage, N stage, and treatment information including (record of chemotherapy and adjuvant radiotherapy). The following SEER ICD-0-3 codes, including 8052, 8070-8072, 8074, 8560, 8571, 8572, 8575, and 8980 were adopted to identify cases of MBC. Patients with unknown race and marital status, unavailable pathological or survival data were excluded. Figure 1 illustrates the detailed ow diagram for patients screening. Overall survival and cancer-speci c survival (CSS) were chosen as endpoints. OS was de ned as the time from original diagnosis to death from any cause or to the time of the last follow-up. CSS was de ned as the time between the date of diagnosis and cancer-related death.

Nomogram Construction And Conformation
Patients were randomly divided into the training set and validation set with a ratio of 2:1. X-tile tool was used to set the optimized cut-off points for continuous variable. Univariate and multivariate analyses were carried out by employing the Cox proportional hazard regression models to determine the hazard ratio (HR) along with corresponding 95% con dence interval (CI) for all possible risk factors. All independent risk factors were identi ed by the forward stepwise selection method using the multivariate Cox proportional hazards models. The nomogram was established by combing all independent risk factors prognostic factors for the prediction of the 3-year and 5-year OS and CSS using the "rms" R package (cran.rproject.org/web/packages/rms). The Harrell's concordance index (C-index) was used to access the discrimination, and calibration curves were applied to estimate the consistency between the actual prognosis and the nomogram-predicted survival probability of the model.

Statistical analysis
IBM SPSS statistics 22 software (SPSS Inc., Chicago, IL, USA) was used to conduct statistical analysis. R software v 3.6.1 (http://www.r-project.org) was adopted to construct nomograms based on the multivariate results and the "RMS" package was used to develop survival models. The two-tailed P-value < 0.05 was assumed statistically signi cant.

Patients characteristics
A total of 3927 MBC patients were obtained from the SEER database, and 1129 of these patients with available information required for analysis were screened out and used to perform subsequent analysis. The detailed clinicopathologic characteristics of included patients were presented in Table 1. All included patients were allocated randomly into two datasets, including 753 patients randomly assigned to the training set and 376 patients to the validation set. Of these patients, 620 (54.92% patients were diagnosed at the age more than 60 years, most patients (76.62%) were white. The cut-off value was set as 58 mm by X-tile software, and 14.26% patients had a tumor more then 58 mm diameter. For the degree of cancer differentiation, poorly differentiated (Grade III) was the most type 877 (77.68%). 883 (78.21%) 977 (86.54%), and 1065 (94.33%) patients were observed to negatively express ER, PR and HER2, respectively. According to the AJCC7 system, stage II was the most type (61.65%), follow by stage I (23.91%) and stage III (14.44%). Most patients (51.64%) were categorized as T2 stage. Meanwhile, most patients (78.12%) were categorized as N0 stage. More than half of patients (66.43%) had undergone chemotherapy, while 47.03% patients had undergone radiotherapy. identi ed as independent prognostic factors of CSS of MBC patients, as shown in Table 2 and Table 3.

Construction And Validation Of Os And Css
According to the results of multivariate analysis, all independent prognostic factors in the training set were incorporated to create the nomograms for estimating 3-and 5-year OS and CSS of patients with MBC. Figure 2(A) and Fig. 2(B) showed the prediction of the 3-and 5-year OS and CSS in the nomogram, respectively. Each factor was allocated a score on the points scale in the nomogram, and we can estimate the 3-and 5-year survival probability of patients with MBC by calculating the total score via adding up all points on the basis of personal patients features.
Nomograms were validated using C-index in the training set and validation set, respectively. The results showed su cient accuracy in forecasting the prognosis of MBC in training set and validation set. The Cindex of the nomogram for OS and CSS is 0.744 (95% CI, 0.701-0.787) and 0.746 (95% CI, 0.695-0.797) in the training set, respectively ( Table 4). The C-index calculated from the validation set is 0.818 (95% CI, 0.775-0.861) in OS and 0.800 (95% CI, 0.747-0.853) in CSS, respectively. The calibration plots showed good coordination between prediction by nomogram models and observed outcomes in the probability of 3-and 5-year OS and CSS of patients with MBC in both training cohort and validation set (Fig. 3).

Discussion
Metaplastic breast cancer is a kind of heterogenous breast cancer, which is relatively rare in clinical practice. Despite several studies have found risk factors related to the clinical outcomes of MBC patients [11][12][13], there is no recognized prognostic factors to predict the prognosis of MBC. Paul Wright et al. [14] found that for patients with positive or negative hormone receptors, there was no signi cant difference in the 5-year survival rate of MBC, which indicated that the status of hormone receptors could not be considered as a prognostic factor of MBC. Additionally, previous study demonstrated that the subtype of MBC could be an independent predictor of its prognosis [3]. Several studies revealed that the prognosis of MBC patients with larger tumor and lymph node metastasis are generally poor [11,15]. In recent years, some studies have also focused on the relationship between gene signatures and prognosis of MBC patients, such as the high expression of 60S ribosomal protein L39 (RPL39) [13] and the mutation of the colony stimulating factor 1 receptor (CSF1R) [16], all of which can indicate poor prognosis.
However, single prognostic factors play a limited role in predicting individual survival probability.
Nomograms are graphical display of mathematical models for predicting cancer risk, prevention and therapeutic outcomes, which becomes increasingly popular clinical decision aids thanks to their ability to deal with complex problems in a systematic and unbiased manner [17][18][19]. It has been revealed that nomograms exhibited more excellent prediction precision and prognostic value in diverse malignancies than the existing tumor system [20,21]. To construct a prognostic nomogram, we conducted univariate and multivariate analyses to nd clinical characteristics that correlated with the OS and CSS of MBC patients on the basis of a large data set from the SEER database. We demonstrated that several clinicopathological characteristics were independent prognostic factors for OS, including age, stage_ajcc, T stage, chemotherapy and radiotherapy. In addition, multivariate analysis con rmed that age, race, stage_ajcc, T stage, and radiotherapy were independent prognostic factors for CSS of MBC patients. The nomograms established in this study showed favorable discrimination and calibration for 3-year and 5-year OS and CSS of MBC patients and offered a more accurate and personalized clinical tool for prognosis evaluation of MBC patients.
Prognostic studies have given con icting results regarding factors associated with prognosis and survival [22][23][24][25][26][27][28][29][30][31][32]. In the present study, we critically evaluated the prognostic value of various factors based on a large cases of MBC recorded on the SEER. The clinical signi cance of age, stage_ajcc, T stage, chemotherapy, and radiotherapy in MBC patients were highlighted in nomogram models. The result demonstrated that half of patients were older then 60 years, who suffered worst survival and poor OS. Patients with older age generally accompanied a higher-risk histological phenotype [33], which has been considered as an independent risk factor and may eventually result in lower survival [34][35][36]. Of note, race was de ned as a independent prognostic factor for CSS but not for OS. In addition, we found that chemotherapy is an independent prognostic factor for OS in MBC patients.
Although it is correlated with CSS in univariate analysis, it is not an independent prognostic factor for CSS. It may result from the worse response to chemotherapy regimens in MBC [23,27,29,37,38].
Previous studies have concluded that radiotherapy was able to improve the survival of patients with MBC [8,28,38], and our data also demonstrated that radiation was independently prognostic factors associated with survival probability of patients with MBC [8,30,39]. Moreover, radiotherapy was revealed to be able to reduce the risk of local recurrence [40]. T stage is closely related prognosis of cancers and previous study based on US population found that T stage was a is a independent prognostic factor for the OS and CSS of MBC patients [8]. Similarly our results also identi ed T stage as a independent prognostic factor of MBC patients. Lymph nodes metastasis (LNM) has been identi ed as a key prognostic indicator for a variety of malignancies, and the number of LNM has been included into the N-staging. Previous studies reported that lymph node status was signi cantly correlated with survival endpoints in patients with MBC [41,42]. However, N-stage was not identi ed as a independent prognostic factor for MBC patients in this study.
There were several potential shortcomings in this study. First, retrospective data retrieved from the same database was used in the generation and validation of the nomogram models, which may lead to the risk of potential selection bias. Therefore, it would be more reliable to validate the nomograms in another dataset. Second, in this study, we only included two endpoints: 3-and 5-year survival.
However, the assessment of recurrence risk is considered as a more meaningful endpoint than OS or CSS because of the rare speci c mortality of MBC, which was not performed in this study owing to the lack of data with respect to recurrence in SEER database. Moreover, several other crucial prognostic factors, such as RET mutation status and calcitonin doubling times, were also unavailable in the SEER database.

Conclusion
In the present study, we identi ed the independent prognostic factors for the OS and CSS of MBC patients, and the nomograms reliably predicted 3-and 5-year OS and CSS of MBC patients were successfully established and well-validated on the basis of large population from SEER. The nomograms established in this study are expected to assist clinicians to conduct individual prognostic assessments and to formulate reasonable treatment strategies. However, further external validation using external data in the future is required to increase model reliability and generalize the applicability of these nomogram model in clinical practice.

Availability of data and materials
All data analyzed during this study are included in this published article.

Competing interests
The authors declare that they have no competing interests. Writing-review&editing: MC and HCJ.
All authors have read and approved the nal manuscript.