Development and validation of an extended Cox prognostic model for patients with ER/PR+ and HER2‒ breast cancer: a retrospective cohort study

doi:10.21203/rs.3.rs-1423615/v1

Background

The purpose of this study was to explore a new ER/PR + and HER2 − breast cancer prognostic model, called the extended Cox prognostic model, for determining the cutoff values for multiple continuous prognostic factors and their interaction via the new model concept and variable selection method.

Methods

A total of 335 patients with ER/PR + and HER2– breast cancer were enrolled for the final analysis. The primary endpoint was breast cancer-specific mortality. Prognostic factors (histological grade, histological type, stage, T, N, lymphovascular invasion, P53, Ki67, ER, PR, and age) were included in this study. The four continuous variables (Ki67, ER, PR, and age) were partitioned into a series of binary variables that were fitted in the multivariate Cox analysis. A smoothly clipped absolute deviation variable selection method was used. Model performance was expressed in discrimination and calibration.

Results

We developed an extended Cox model with a time threshold of 164 weeks (more than 3 years) postoperation and developed a user-friendly nomogram based on our extended Cox model to facilitate clinical application. We found that the cutoff values for PR, Ki67, and age were 20%, 60%, and 41–55 years, respectively. There was an interaction between age and PR for patients aged ≥ 41 years and PR ≥ 20% at 164 weeks postoperation: the older the patients with ER/PR+, HER2-, PR ≥ 20% were, the lower survival and more likely to recur and metastasize exceeding 164 weeks (more than 3 years) after surgery.

Conclusions

Our study offers guidance on the prognosis of patients with ER/PR + and HER2‒ breast cancer in China. The new concept can inform modeling and the determination of cutoff values of prognostic factors in the future.

Breast cancer

Estrogen receptor

Progesterone receptor

Human epidermal growth factor receptor 2

Clinicopathological prognostic factor

Prognostic model

Breast cancer is the malignant carcinoma with the highest incidence among Chinese women. In China, with an increase in breast cancer in recent years, research on prognostic models for breast cancer has become a growing concern [1]. Recently, prognostic gene signatures (Oncotype DX, Mammaprint, and so on) have been taken more seriously; however, the current prognostic gene signatures are not ready to be used in clinical practice due to a plethora of concerns regarding cost and technology, regardless of first- or second-generation gene signatures [2, 3]. In the United States, Oncotype DX is the most commonly used breast cancer genomic assay and costs around $4,000. It is used in only approximately one-third of patients with breast cancer in America and in less than 20% of patients in European countries [4]. In China, a 21-gene assay without a standardized test norm is commonly used and costs around $500-$1,000 (not covered by Chinese health care insurance). The test is performed in approximately 20–100% of patients with breast cancer at different hospitals in China; the instances where 100% of breast cancer patients were tested were mainly clinical trials. Furthermore, recently, the prognostic value of the classic clinicopathologic variables is being reconsidered [5–7]. Some evidence indicates that clinicopathologic variable models are excellent surrogates for prognostic gene signatures [4]. Hence, classic clinicopathologic variable models are highly valued due to their feasibility for clinical practice. From 2006 to the present, the National Comprehensive Cancer Network has classified invasive breast cancer into four subtypes [8]. The molecular subtype of breast cancer is a classification method similar to the intrinsic subtype and is more suitable for current clinical practice in China, also serving as an independent prognostic factor [9, 10]. Among the four molecular subtypes of breast cancer, ER and/or PR + and HER2– (estrogen receptor and/or progesterone receptor positive and human epidermal growth factor receptor 2 negative) occurr most commonly and account for approximately 60% of breast cancer patients [10, 11]. There are urgent demands and wider potential impacts for exploration of improved prognostic models for patients with ER/PR + and HER2‒ breast cancers based on classic clinicopathologic variables, particularly in China.

The current classic prognostic algorithms (PREDICT, Adjuvant! Online, and Nottingham Prognostic Index) are far from ideal [12–14]. These models were often based on datasets from non-Chinese or non-Asian patients. Specifically, it was assumed that these prognostic factors, including Ki67 (a nuclear marker of cell proliferation), estrogen receptor (ER), progesterone receptor (PR) and age, were continuous factors, or it was assumed that the cutoff values of prognostic factors, including Ki67, ER, PR and age, were determined merely based on univariate analysis, experience or speculation. In addition, existing models ignor the interaction effects between prognostic factors. Therefore, current models showed poor accuracy and were not suitable for clinical practice in China. It is critical to develop a novel improved prognostic algorithm to analyze clinical data from China.

For this reason, in this study we selected 335 patients with ER/PR + and HER2‒ breast cancer to explore our new ER/PR + and HER2‒ breast cancer prognostic model using classical clinicopathologic variables, called the extended Cox prognostic model. The cutoff values for multiple continuous prognostic factors were determined, and the interaction effect between the factors was elucidated via the new modeling concept and variable selection method.

1.1 Study population

All patients with invasive unilateral breast cancer admitted to the Department of Breast Surgery at Huangpu Branch, Shanghai Ninth People’s Hospital, Shanghai JiaoTong University School of Medicine from January 2009 to December 2009 were evaluated. Information obtained from medical records included age at diagnosis, number of lymph nodes sampled and number of positive lymph nodes (categorized as 0, 1 to 3, 4 to 9, and 10 + nodes positive [6]), lymphovascular invasion (categorized as positive or negative), tumor size (categorized as < 21 mm, 21 to 50 mm, 50 + mm [6]), histological grade (categorized as I, II, III [6]), pathological type, protein 53 (p53) status (categorized as positive or negative), proliferating cell nuclear antigen Ki67 (Ki67) status, ER status, PR status, HER2 status, information on local therapy (wide local excision, mastectomy, radiotherapy), and type of adjuvant systemic therapy (chemotherapy, endocrine therapy, or both). Patients with any one of the following conditions were excluded from the analyses: mucinous carcinoma, cribriform carcinoma, or tubular carcinoma [8]; those with incomplete information; who received chemotherapy or radiation before operation; who did not undergo surgery; who did not complete local treatment (local excision without radiotherapy); who had no axillary lymph node dissection; who did not complete adjuvant systemic therapy (chemotherapy and endocrine therapy); intraductal carcinomas; or bilateral breast cancers, leaving 568 individuals. Variables for each patient included age, TNM status, T stage, N stage, pathological subtype, histology grade, lymphovascular invasion, p53 status, Ki67 status, ER status, PR status, vital status, and survival time (Fig. 1).

1.2 Categorization of patients into ER/PR + and HER2‒ subtypes of breast cancer

All formalin-fixed paraffin-embedded tumor blocks were collected at the time of surgery prior to adjuvant therapy and were stored at room temperature. Approval was obtained from the Institutional Review Boards. Tumor sections of 4 to 10 µm were cut. One section was stained with hematoxylin/eosin to confirm the presence of invasive carcinoma, and the other sections were used for molecular analyses by two independent pathologists. ER, PR and HER2 were assayed by immunohistochemistry (IHC) and evaluated according to standard criteria [11, 15, 16].

The criteria for evaluating ER and PR in breast cancer cells by IHC [15] were as follows: ER or PR was positive if the cell nuclei showed a brown color. In one section, five high-power regions were selected randomly. The patient was assigned to subtype ER/PR + if the percentage of positive cells was ≥ 1% in these regions. The criteria for evaluating HER2 by IHC were as follows: the patients were categorized into four subtypes: 0, 1+, 2+, or 3+ [11, 16]. Subtypes HER2 IHC 0 and 1 + were HER2 negative (HER2‒). Subtype HER2 IHC 2 + was equivocal in HER2 status. Finally, subtype HER2 IHC 3 + was HER2 positive. Tumor cells of HER2 IHC 2 + subtype were further analyzed by fluorescence in situ hybridization (FISH). According to the HER2/chromosome 17 centromere ratio and the average HER2 gene copy number, FISH results were considered positive, equivocal, or negative [11]. Patients with equivocal FISH results were excluded from the study. Among the 568 patients, the HER2 status of 185 individuals was detected as HER2 IHC (++) [11, 16]. Tissue microarrays (TMAs) were constructed from the tissue core with a diameter of 1.5 mm to detect the HER2 status of these 185 patients using FISH methods. Finally, the 335 individuals among the 568 patients were determined to be ER/PR + and HER2 − type, comprising our study population (Fig. 1).

1.3 Treatment for the patients.

All patients underwent modified radical mastectomy/breast-conserving surgery and adjuvant chemotherapy. They were treated with six cycles of CEF (cyclophosphamide, epirubicin, and fluorouracil) chemotherapy, four cycles of CEF followed by four cycles of T (docetaxel) chemotherapy, or four to six cycles of TEC (docetaxel, epirubicin, and cyclophosphamide) chemotherapy after surgery. If necessary, patients received postoperative radiotherapy followed by endocrine therapy, but not trastuzumab treatment.

1.4 Follow-up

We defined breast cancer-specific mortality (BCSM) as mortality from breast cancer [17, 18]. The primary endpoint of this study was BCSM, which was determined by following up the survival of patients over a certain time period. From the first day after surgery until death or the end of the study (September 15 2016), we made telephone calls and/or outpatient visits every 3 months to follow up on the survival status of the patients and the cause of death. The follow-up records were double-checked with those from the Department of Cancer Control and Prevention, Shanghai Municipal Center for Disease Control and Prevention.

1.5 Statistical analysis

The extended Cox prognostic model was developed in all eligible patients as follows. First, to determine the cutoff values for each continuous variable (Ki67, ER, PR, and age), these four continuous variables were partitioned into a series of binary variables. Second, all variables were fitted in the multivariate Cox analysis. A smooth clipped absolute deviation (SCAD) variable selection method [19] was used to build a Cox prognostic model to determine the independent variables, cutoff values and interaction effect between different factors. Finally, when developing the model, we hypothesized that the model could be divided into two parts by a certain time point. We built a new model, named the extended Cox prognostic model, with a time threshold of 164 weeks (more than 3 years) based on the smallest Akaike information criterion (AIC) value.

We evaluated the predictive accuracy of the extended Cox prognostic model based on the parameters of discrimination and calibration. For model discrimination, receiver operating characteristic (ROC) curves were plotted for the data at 1 year, 3 years, and 5 years postoperation [20]. We also calculated the areas under the ROC curves (AUCs). Model calibration was assessed by a simplified goodness-of-fit (GOF) method [21]. We compared the number of deaths observed and calculated at 5 years postoperation. We grouped the risk scores into 10 sets and then calculated the GOF statistics for each set [22]. This provided a GOF chi-square test. All analyses were conducted using R software version 3.3.2.

2.1 Patient characteristics

All 335 patients were females. All clinical and pathological data of the patients were complete, including age, tumor size, axillary lymph node status, lymphovascular invasion status, histological grade, pathological type, ER/PR/HER2/Ki67/P53 status, and surgical and postoperative adjuvant treatment (specified as chemotherapy, targeted therapy, radiotherapy, and endocrine therapy). Patient characteristics and differences in characteristics by survival outcome are detailed in Table I. Among 335 patients, 22, 267, and 46 had histologic grade I, II, and III tumors, respectively [6]. According to the 2017 American Joint Committee on Cancer staging system [6, 7], 206, 126, 3, and 0 patients had stage T1, T2, T3, and T4 disease, respectively, while 180, 98, 41, and 16 patients had stage N0, N1, N2, and N3 disease, respectively.

2.2 Prognostic factors

First, the expression of Ki67, ER and PR in cancerous tissue areas was quantified as 1-100, which was the percentage of tumor nuclei positive (positive nuclei) over all the tumor nuclei (positive nuclei and negative nuclei) by manual counting. The prognostic factors included in this study were divided into categorical or continuous factors. Appendix I lists the prognostic factors included in this study. Second, to determine the cutoff values for these continuous variables (Ki67, ER, PR, and age), we transformed each continuous factor (Ki67, ER, PR, and age) into a series of binary variables using each observed value as a partition point. Appendix II provides the possible partition points for the four continuous factors. Each partition point converted the continuous variable into one binary variable. Taking Ki67 as an example, point 60% converted Ki67 into one binary variable named “Ki67 status (60%)”, which equals 1 if the immunochemical status of Ki67 was no less than 60% and equals 0 otherwise. Additionally, it was mandatory that each sample size of each binary variable was not less than 10. For each patient case, 95 variables were determined, including categorical factors (histological grade, histological type, stage, T, N, P53, and lymphovascular invasion), continuous variables (Ki67, ER, PR, and age) and a series of binary variables. Based on a previous clinic study and clinical practice [23], we introduced four potential interaction terms: T and N, ER and PR, age and ER, and age and PR. Therefore, we finally determined 1664 variables with the addition of interaction terms.

2.3 Analysis of patient survival times

On September 15, 2016, 311 of the 335 patients were still being followed, and 24 had been lost to follow-up (including one noncancer death). Among the 311 patients, 28 patients had died from breast cancer. The clinicopathologic characteristics of the 28 patients who died from breast cancer are presented in Appendix III. The median follow-up duration was 370 weeks, with a censoring rate of 91.6%. The longest survival time of patients who died was 341 weeks. The survival probability at 341 weeks (approximately six and a half years) of breast cancer patients was 91.2% (95% confidence interval 88.1%-94.3%) (Fig. 2).

2.4 Prognostic model development

To determine the cutoff values for continuous variables (Ki67, ER, PR, and age), we transformed each continuous factor (Ki67, ER, PR, and age) into a series of binary variables using each observed value as a partition point. We calculated the hazard ratios and their 95% confidence limits for these binary variables (see Appendix IV). It was meaningful to look for significant cutoff values for these four factors during modeling; we achieved this objective and focused on the specific prognostic models below.

2.4.1 Cox proportional hazards model without interaction terms

We first introduced 97 variables without interaction items into the model, and a SCAD variable selection method [19] was used to develop the Cox proportional hazard model (named Model 1).

$\text{h}\left(t\right)={h}_{0}\left(t\right)\text{e}\text{x}\text{p}\{1.38{x}_{1}+1.01{x}_{2}+1.643{x}_{3}-0.949{x}_{4}+1.04{x}_{5}$ } Eq. (1)

The prognostic factors, beta coefficients, hazard ratios for prognostic factors, and P values of Z tests are provided in Table II.

The multivariable Cox regression analysis for BCSM without interaction terms, Model 1, showed that histological grade, N status, Ki67 status, PR status, and age were statistically significant predictors of patient survival. The cutoff value for Ki67 was 60%, the cutoff value for PR was 20%, and the cutoff value for age was 55 years old.

2.4.2 Cox proportional hazards model with interaction terms

Second, 1664 variables including interaction terms were used in the model to evaluate the interaction effect between prognostic factors. A SCAD variable selection method [19] was also used to develop the Cox proportional hazard model (named Model 2).

$\text{h}\left(t\right)={h}_{0}\left(t\right)\text{exp}\left\{1.392{x}_{1}+0.995{x}_{2}+1.595{x}_{3}+1.202{x}_{5}-1.157{x}_{6}\right\}$ Eq. (2)

The prognostic factors, beta-coefficients, hazard ratios for prognostic factors, and P values of Z tests are provided in Table III.

Model 2 showed that there might be an interaction effect between PR (20%) and age (41 years). However, PR (20%) and age (41 years) were not statistically significant predictors of patient survival in Model 2. These results indicated that Model 2 may not be self-sufficient.

2.4.3 Extended Cox prognostic model

To further optimize the two above Cox models, we first included all variables in Model 1 and Model 2, which were histological grade, N status, Ki67 status (60%), PR status (20%), age (55 years), age (41 years) and PR-age (age ≥ 41years, PR ≥ 20%), in the Cox model. The outcome indicated that PR status (20%)、age (41 years old) and PR-age were not statistically significant. Furthermore, taking previous clinical experience into account, we were aware that the interaction between PR and age might vary according to the different time periods after surgery and that the model could be divided into two parts by a specific time point. To determine the optimal time point, we considered both death and censored time periods. We developed a new Cox Model 3 (named the “extended Cox prognostic model”) with a time threshold of 164 weeks after surgery, based on the smallest AIC value.

$\text{h}\left(\text{t}\right)=\left\{\begin{array}{c}{h}_{0}\left(t\right)\text{exp}\left\{A\right\}, t \le 164weeks\\ {h}_{0}\left(t\right)\text{exp}\left\{A+1.594{x}_{8}\right\}, t > 164weeks\end{array}\right.$ Eq. (3)

$$\text{A}=1.329{x}_{1}+0.972{x}_{2}+1.65{x}_{3}-1.802{x}_{4}+1.378{x}_{5}-1.563{\text{x}}_{7}$$

The prognostic factors, beta-coefficients, hazard ratios for prognostic factors, and P values of Z tests are provided in Table IV and Table V.

The extended Cox prognostic model determined the cutoff values for multiple continuous prognostic factors and the interaction effect between the factors. First, the cutoff values of the prognostic factors were determined by Cox prognostic model with the SCAD variable selection method [19]. Among 1730 predictors, histological grade and N status were considered categorical factors. However, the model showed that Ki67, PR and age were also categorical factors and the prognostic model automatically determined their reasonable cutoff values. For Ki67 expression, a cutoff value of 60% was selected to distinguish between low expression (< 60%) and high expression (≥ 60%). For PR expression, a cutoff value of 20% was selected to distinguish between low expression (< 20%) and high expression (≥ 20%). Regarding age, we categorized patients into three groups based on age at the time of surgery: the older group, who were 55 years or older; the younger group, who were less than 41 years old; and the middle-aged group, who were between 41 and 55 years old. Second, for the interaction effect, after 164 weeks postoperation, there was an interaction between age and PR: the older the patients with ER/PR+, HER2-, PR ≥ 20% were, the lower survival and more likely to recur and metastasize exceeding 164 weeks (more than 3 years) after surgery.

In our study, we found that the hazard ratio for BCSM increased 2.78 times as the histological grade increased by one level. If N statuses increased one level, the hazard ratio for BCSM increased 1.64 times. The patients with high Ki67 expression had a hazard ratio for BCSM that was 4.21 times higher than that of the patients with low Ki67 expression. Within 164 weekspostoperation, the hazard ratio for BCSM of the patients with low PR expression was 4.88 times higher than that of the patients with high PR expression. Within 164 weeks postoperation, patients aged < 41 had the highest hazard ratio for BCSM, followed by patients aged ≥ 55, while patients aged 41 to 55 showed the lowest hazard ratio for BCSM. After 164 weeks postoperation, there was an interaction effect between age and PR for patients aged ≥ 41 years and PR ≥ 20%. The hazard ratio for BCSM in patients aged ≥ 41 years and PR ≥ 20% was elevated after 164 weeks postoperation. For patients with high PR expression, age was positively correlated with mortality after 164 weeks postoperation. For patients with high PR expression after 164 weeks postoperation, the patients aged 41 to 55 had nearly the same hazard ratio for BCSM as those aged < 41, while the hazard ratio for BCSM of patients aged ≥ 55 years was 3.09 times higher than that of patients aged < 41 years.

2.5 Utilizing the nomogram

To improve the practicability of our model, we established a nomogram of 1-year, 3-year and 5-year survival probability based on our model (Fig. 3). The nomogram consisted of eleven rows. The first “points” row was the point assignment for each variable. For an individual patient, each variable (rows 2–7) was assigned a point value by drawing a vertical line between the exact variable value and the first “points” row. Subsequently, the total points could be obtained by summing all of the allotted points for the six variables (rows 2–7). The total point sum was found in the ‘‘total points’’ row. Finally, the 1-, 3-, and 5-year survival probability could be predicted by drawing a vertical line between the ‘‘total points’’ row and the “predicted probability” rows (rows 9–11, respectively).

2.6 Model discrimination and calibration

As expected, this extended Cox prognostic model showed good discrimination. Figure 4 shows the ROC curves for BCSM at 1 year, 3 years, and 5 years postoperatively. The AUCs for BCSM at 1 year, 3 years, and 5 years postoperatively are shown in Table VI. The AUC values for BCSM at 1 year, 3 years and 5 years postoperatively were all larger than 0.80. The AUC value for BCSM at 3 years postoperation was as high as 0.94, with a 95% CI from 0.89 to 0.99. Therefore, our extended Cox prognostic model showed good discrimination.

The extended Cox prognostic model was also well calibrated by the GOF test [21]. We grouped the risk scores into 10 groups and then calculated the GOF statistic 7.37, with a P-value of 0.6 (> 0.05) [22]. These results indicated that our extended Cox prognostic model fit was good.

In this study, we explored an extended Cox model for the prognosis of ER/PR + and HER2 − breast cancer, with calculating the cutoff points of prognostic factors and their interaction. The cutoff values of Ki67, PR and age and the interaction between age and PR status were generated from model calculations. The model was well calibrated and provided a high degree of discrimination.

Here, we found that the prognosis of patients was associated with histological grade, N stage, Ki67 and PR statuses, and age. We found that the cutoff values for PR, Ki67 and age were 20%, 60% and 41–55 years respectively. It is important to point out that the cutoff values of prognostic factors, including Ki67, PR and age, were determined only based on our extended Cox prognostic model (a multivariable analysis) and not based on univariate analysis, experience or speculation; this is different from previous studies [12–14, 24, 25]. The prognosis of patients was associated with histological grade, N stage, Ki67 and PR statuses, and age, which was consistent with previous reports [23]. Histological grade and N stage had a linear effect on the hazard ratio. The cutoff value (20%) for PR was consistent with the St. Gallen consensus of 2013 [26, 27], which indicated the high fidelity of our model. We believe that breast cancer with low PR expression is probably a different intrinsic subtype of breast cancer, that is, the luminal B subtype. Our prognostic model determined 60% as the cutoff value for Ki67 status, which was much higher than in other studies (e.g., 14% and 20%) [23, 26–29]. We speculated that this situation was because of a lack of a standardized procedure for Ki67 assessment and the controversial Ki67 assay interpretation [30] and could also be attributed to their determining the cutoff value for Ki67 with the ROC method, rather than the univariate analysis method used in a previous study [24, 28]. Therefore, the prognostic value of the Ki67 index in breast cancer needs to be further explored. The cutoff value for age (41 and 55 years) was consistent with the age range of perimenopausal and menopausal Chinese women, although it was different from those of other studies [23, 31]. Hence, our models showed high consistency to the currently available gold standard, such as the PR factor. The values are well correlated with the physiological conditions of our patients, such as age. Most importantly, our model generates the cutoff values based on algorithm calculation without empirical biases or univariate analysis.

In our study, algorithmic analysis with the extended Cox prognostic model showed that there was an interaction between age and PR for patients aged ≥ 41 years and PR ≥ 20% after 164 weeks postoperation. The interaction between age and PR caused the patients with age ≥ 41years and PR ≥ 20% to have relatively higher mortality after 164 weeks postoperation. The existence of the interaction between age and PR status has been previously reported [23]. We feel that our study of the interaction was deeper and clearer. We believe that the existence of the interaction may be related to multiple factors, including the subtype of breast cancer, the status of menopause and the health of the patients 164 weeks after surgery. However, the specific causes remain to be further studied. Meanwhile, we found that the older the patients with ER/PR+, HER2-, and PR ≥ 20% were, the lower the survival was and the more likely recurrence and metastasis were 164 weeks after surgery. Therefore, the clinicians should pay more attention to the long-term follow-up of ER/PR+, HER2-, PR ≥ 20% subtype breast cancer patients aged ≥ 41years(equivalent to the peri- and postmenopausal period in China)more than 3 years after surgery. Drug selection in adjuvant endocrine therapy requires careful consideration in these patients. Endocrine therapy should be sufficiently extended for these patients.

Our studies showed the following four advantages: 1. We adopted a new concept to determine the cutoff values based only on our extended Cox prognostic model, a multivariable analysis. The cutoff values for multiple prognostic factors, including age/PR /Ki67 status, were determined and were statistically significant for prognosis. 2. During the development of the model, we introduced potential interactions and surmised that the model could be divided into two parts by a specific time point. An extended Cox model with a time threshold of 164 weeks postoperation was built based on statistical analysis. We found that there was an interaction between age and PR at 164 weeks postoperation. The hazard ratio for BCSM in patients aged ≥ 41years and PR ≥ 20% was elevated after 164 weeks postoperation. Therefore, clinicians should pay more attention to the long-term follow-up of ER/PR+, HER2-, PR ≥ 20% subtype breast cancer patients aged ≥ 41years more than 3 years after surgery. 3. In actual clinical practice, we developed a user-friendly nomogram based on our extended Cox model to facilitate clinical application. The prediction factors in our model were all included in domestic routine testing covered by Chinese health care insurance with standardized test norms. There was no additional financial cost. Clinicians could predict the 1-, 3-, and 5-year survival probability in a specific patient by summing the scores of each variable based on our nomogram. 4. Our study was derived from Chinese clinical data, which could be the most relevant model for Chinese clinical practice. Due to gene similarity, the model could also apply to the prognosis of breast cancer in other Asian female populations. The cutoff value for age (41 and 55 years) was consistent with the age range of perimenopausal and menopausal Chinese women, although it was different from those of other studies.

One limitation of our study was that it was validated internally, but it lacked external validation. It would be better to perform external validation to validate the model for overfitting. Another limitation was that the overall sample size and the number of outcome events were relatively small. This might cause possible bias in the results. In addition, this was a single-center retrospective study and was thus vulnerable to selection bias. Therefore, whether our model and our cutoff values can be generalized to other populations needs further investigation.

In conclusion, using the new modeling concept and statistical method, an extended Cox prognostic model for the prognosis of ER/PR + and HER2 − breast cancer was explored while determining the cutoff points of prognostic factors and their interaction. The new conceptualization and statistical method in our study were different from those in previous studies, especially regarding the cutoff value of age and its interaction. The new modeling concept used in our study will likely become a research method for prognostic cutoff values. Moreover, the results of our study offer guidance for the prognosis and treatment of patients with ER/PR + and HER2 − breast cancer in China.

Acknowledgements

We grandly thank the Department of Cancer Control & Prevention, Shanghai Municipal Center for Disease Control & Prevention. And we would like to thank Dr. Fangwei Shao of Zhejiang University-University of Illinois at Urbana-Champaign Institute, Zhejiang University, Haining, China.

Funding

This study was supported by project of Science and Technology Commission of Shanghai Municipality (STCSM) (No. 14411972400), project of Shanghai Huangpu District Health Commission (No. 2019BJ04）, and project of Shanghai Huangpu District Health Commission (No. 2019GG05）.

Availability of data and materials

All data generated or analyzed during this study are included in this published article.

Author's contributions

YX, XL, and YW designed and supervised the project. YX, XL, and YW analyzed the results and drafted the manuscript. YX, WC financed the project. LY, YX conducted the computational analysis. XL, YW, and WC performed the experiments. YX, XL, YW, and WC prepared the figures and tables. All authors have read and approved the content of the manuscript.

Ethics approval and consent to participate

All protocols involving the use of humans were approved by the Ethics Committee of Huangpu Branch, Shanghai Ninth people’s Hospital, Shanghai JiaoTong University School of Medicine, China. Written informed consent was obtained from all subjects for the use of their tissue in this study.

Consent for publication

Not applicable

Competing interests

The authors declare that they have no competing interests

Fan L, Strasser-Weippl K, Li JJ, St Louis J, Finkelstein DM, Yu KD, Chen WQ, Shao ZM, Goss PE: Breast cancer in China. Lancet Oncol 2014, 15:e279-289.
Ribnikar D, Cardoso F: Tailoring Chemotherapy in Early-Stage Breast Cancer: Based on Tumor Biology or Tumor Burden? Am Soc Clin Oncol Educ Book 2016, 35:e31-38.
Goldstein DA, Mayer C, Shochat T, Reinhorn D, Moore A, Sarfaty M, Yerushalmi R, Goldvaser H: The concordance of treatment decision guided by OncotypeDX and the PREDICT tool in real-world early-stage breast cancer. Cancer medicine 2020, 9:4603-4612.
Orucevic A, Bell JL, McNabb AP, Heidel RE: Oncotype DX breast cancer recurrence score can be predicted with a novel nomogram using clinicopathologic data. Breast cancer research and treatment 2017, 163:51-61.
Cardoso F, van't Veer LJ, Bogaerts J, Slaets L, Viale G, Delaloge S, Pierga JY, Brain E, Causeret S, DeLorenzi M, et al: 70-Gene Signature as an Aid to Treatment Decisions in Early-Stage Breast Cancer. N Engl J Med 2016, 375:717-729.
Hortobagyi GN, Connolly JL, D’Orsi CJ, Edge SB, Mittendorf EA, Rugo HS, Solin LJ, Weaver DL, Winchester DJ, Giuliano A (Eds.): Breast. In: AJCC Cancer Staging Manual.8th ed. New York: Springer International Publishing; 2017.
Giuliano AE, Connolly JL, Edge SB, Mittendorf EA, Rugo HS, Solin LJ, Weaver DL, Winchester DJ, Hortobagyi GN: Breast Cancer-Major changes in the American Joint Committee on Cancer eighth edition cancer staging manual. CA Cancer J Clin 2017, 67:290-303.
Gradishar WJ, Anderson BO, Balassanian R, Blair SL, Burstein HJ, Cyr A, Elias AD, Farrar WB, Forero A, Giordano SH, et al: NCCN Guidelines Insights: Breast Cancer, Version 1.2017. Journal of the National Comprehensive Cancer Network : JNCCN 2017, 15:433-451.
Carey LA, Perou CM, Livasy CA, Dressler LG, Cowan D, Conway K, Karaca G, Troester MA, Tse CK, Edmiston S, et al: Race, breast cancer subtypes, and survival in the Carolina Breast Cancer Study. Jama 2006, 295:2492-2502.
Chen XS, Ma CD, Wu JY, Yang WT, Lu HF, Wu J, Lu JS, Shao ZM, Shen ZZ, Shen KW: Molecular subtype approximated by quantitative estrogen receptor, progesterone receptor and Her2 can predict the prognosis of breast cancer. Tumori 2010, 96:103-110.
Li J, Chen Z, Su K, Zeng J: Clinicopathological classification and traditional prognostic indicators of breast cancer. Int J Clin Exp Pathol 2015, 8:8500-8505.
Haybittle JL, Blamey RW, Elston CW, Johnson J, Doyle PJ, Campbell FC, Nicholson RI, Griffiths K: A prognostic index in primary breast cancer. Br J Cancer 1982, 45:361-366.
Olivotto IA, Bajdik CD, Ravdin PM, Speers CH, Coldman AJ, Norris BD, Davis GJ, Chia SK, Gelmon KA: Population-based validation of the prognostic model ADJUVANT! for early breast cancer. J Clin Oncol 2005, 23:2716-2725.
Wishart GC, Azzato EM, Greenberg DC, Rashbass J, Kearins O, Lawrence G, Caldas C, Pharoah PD: PREDICT: a new UK prognostic model that predicts survival following surgery for invasive breast cancer. Breast Cancer Res 2010, 12:R1.
Hammond ME, Hayes DF, Dowsett M, Allred DC, Hagerty KL, Badve S, Fitzgibbons PL, Francis G, Goldstein NS, Hayes M, et al: American Society of Clinical Oncology/College Of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer. J Clin Oncol 2010, 28:2784-2795.
Wolff AC, Hammond ME, Hicks DG, Dowsett M, McShane LM, Allison KH, Allred DC, Bartlett JM, Bilous M, Fitzgibbons P, et al: Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. Arch Pathol Lab Med 2014, 138:241-256.
Johnson HM, Irish W, Muzaffar M, Vohra NA, Wong JH: Quantifying the relationship between age at diagnosis and breast cancer-specific mortality. Breast cancer research and treatment 2019, 177:713-722.
Narod SA, Iqbal J, Giannakeas V, Sopik V, Sun P: Breast Cancer Mortality After a Diagnosis of Ductal Carcinoma In Situ. JAMA oncology 2015, 1:888-896.
Fan J, Li R: Variable selection for Cox’s proportional hazards model and frailty model. The Annals of Statistics 2002, 30:74-99.
Heagerty PJ, Lumley T, Pepe MS: Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics 2000, 56:337-344.
May S, Hosmer DW: A simplified method of calculating an overall goodness-of-fit test for the Cox proportional hazards model. Lifetime Data Anal 1998, 4:109-120.
Hosmer DW, Lemeshow S, Sturdivant RX: Applied Logistic Regression. 3rd edn. New York: Wiley; 2013.
Yao N, Song Z, Wang X, Yang S, Song H: Prognostic Impact of Progesterone Receptor Status in Chinese Estrogen Receptor Positive Invasive Breast Cancer Patients. J Breast Cancer 2017, 20:160-169.
Zhu X, Chen L, Huang B, Wang Y, Ji L, Wu J, Di G, Liu G, Yu K, Shao Z, Wang Z: The prognostic and predictive potential of Ki-67 in triple-negative breast cancer. Sci Rep 2020, 10:225.
Qin T, Zeng YD, Lu Q, Zhang X, Qin GE, Zheng Q, Xu F, Peng R, Yuan Z, Wang S: Nomogram Model of LNR Predicts Survival in Premenopausal Patients with Node-positive Luminal Breast Cancer. Anticancer Res 2017, 37:4575-4586.
Goldhirsch A, Winer EP, Coates AS, Gelber RD, Piccart-Gebhart M, Thurlimann B, Senn HJ: Personalizing the treatment of women with early breast cancer: highlights of the St Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2013. Ann Oncol 2013, 24:2206-2223.
Prat A, Cheang MC, Martin M, Parker JS, Carrasco E, Caballero R, Tyldesley S, Gelmon K, Bernard PS, Nielsen TO, Perou CM: Prognostic significance of progesterone receptor-positive tumor cells within immunohistochemically defined luminal A breast cancer. J Clin Oncol 2013, 31:203-209.
Cheang MC, Chia SK, Voduc D, Gao D, Leung S, Snider J, Watson M, Davies S, Bernard PS, Parker JS, et al: Ki67 index, HER2 status, and prognosis of patients with luminal B breast cancer. J Natl Cancer Inst 2009, 101:736-750.
Nishimura R, Osako T, Okumura Y, Hayashi M, Toyozumi Y, Arima N: Ki-67 as a prognostic marker according to breast cancer subtype and a predictor of recurrence time in primary breast cancer. Exp Ther Med 2010, 1:747-754.
Penault-Llorca F, Radosevic-Robin N: Ki67 assessment in breast cancer: an update. Pathology 2017, 49:166-171.
Liedtke C, Rody A, Gluz O, Baumann K, Beyer D, Kohls EB, Lausen K, Hanker L, Holtrich U, Becker S, Karn T: The prognostic impact of age in different molecular subtypes of breast cancer. Breast Cancer Res Treat 2015, 152:667-673.

Table I. Characteristics of the included patients

Characteristic	Full cohort (n=335)	%	Censored (n=307)	Died (n=28)
Age
Median age (range)	53 (26–89)		53 (26-89)	56 (34-80)
<41	27	8.1	23	4
41-55	159	47.5	153	6
≥55	149	44.5	131	18
Histologic Grade
I	22	6.6	22	0
II	267	79.7	251	16
III	46	13.7	34	12
Pathological Type
nonspecific invasive breast cancer	270	80.6	247	23
invasive lobular carcinoma	15	4.5	14	1
mixed carcinoma	14	4.2	14	0
micropapillary carcinoma	15	4.5	12	3
invasive papillary carcinoma	5	1.5	4	1
neuroendocrine carcinoma	5	1.5	5	0
metaplastic carcinoma	1	0.3	1	0
intraductal and microinvasive carcinoma	10	3	10	0
T stage
1	206	61.5	191	15
2	126	37.6	113	13
3	3	0.9	3	0
4	0	0	0	0
N stage
0	180	53.7	173	7
1	98	29.3	90	8
2	41	12.2	38	3
3	16	4.8	6	10
LVI status
-	220	65.7	205	15
+	115	34.3	102	13
TNM stage
I	133	39.7	127	6
II	142	42.4	133	9
III	60	17.9	47	13
Percentage of ER expression (%)
Mean (SD)	79.36 (23.53)		80.33 (22.42)	68.62 (32.05)
Percentage of PR expression (%)
Mean (SD)	56.69 (34.57)		56.28 (33.96)	39.43 (37.12)
<20	72	21.5	60	12
≥20	263	78.5	247	16
Ki67 status (%)
Mean (SD)	25.12 (22.02)		23.93 (21.14)	38.21 (27.19)
<60	292	87.2	274	18
≥60	43	12.8	33	10
P53 status
‒	176	52.5	163	13
+	159	47.5	144	15

Abbreviations: LVI, Lymphovascular invasion; ER, Estrogen receptor; PR, Progesterone

Table II. Prognostic factors, coefficients, hazard ratios, and P values of the Z tests in Model 1

Variable	Prognostic factor	Coefficient	HR (95% CI)	p-value
	Histological grade (I, II, III)	1.380	3.976 (1.851, 8.843)	<0.001
	N status (0, 1, 2, 3)	1.010	2.746 (1.959, 3.849)	<0.001
		1.643	5.172 (2.288, 11.691)	<0.001
		-0.949	0.387 (0.181, 0.826)	0.014
		1.040	2.828 (1.285, 6.222)	0.01

HR = hazard ratio, CI = confidence interval.

*Ki67 status (60%): a binary variable that is one if Ki67 is no less than 60% and zero otherwise.

*PR status (20%): a binary variable that is one if PR is no less than 20% and zero otherwise.

*Age (55 years): a binary variable that is one if age is no less than 55 years and zero otherwise.

Table III. Prognostic factors, coefficients, hazard ratios, and P values of the Z tests in Model 2

Variable	Prognostic factor	Coefficient	HR (95% CI)	p-value
	Histological grade (I, II, III)	1.392	4.024 (1.865, 8.684)	<0.001
	N status (0, 1, 2, 3)	0.995	2.706 (1.929, 3.795)	<0.001
		1.595	4.930 (2.181, 11.143)	<0.001
		1.202	3.325 (1.496, 7.389)	0.003
		-1.157	0.315 (0.148, 0.669)	0.003

HR = hazard ratio, CI = confidence interval.

*: a binary variable, an interaction item between age and PR, which is one if age is no less than 41 years and PR is no less than 20%, and zero otherwise.

Table IV. Prognostic factors, coefficients, hazard ratios, and P values of the Z tests in the extended Cox model

Variable	Prognostic factor	Coefficient	HR (95% CI)	p-value
	Histological grade (I, II, III)	1.329	3.777 (1.739, 8.204)	<0.001
	N status (0, 1, 2, 3)	0.972	2.644 (1.872, 3.734)	<0.001
		1.650	5.208 (2.311, 11.738)	<0.001
		-1.802	0.165 (0.050, 0.548)	0.003
		1.378	3.968 (1.548, 10.171)	0.004
		-1.563	0.210 (0.053, 0.829)	0.026
	after 164 weeks	1.594	4.922 (1.039, 23.325)	0.045

HR = hazard ratio, CI = confidence interval.

*Age (41 years): a binary variable that is one if age is no less than 41 years and zero otherwise.

*PR- age after 164 weeks (age ≥ 41 years, PR ≥ 20%): a binary variable, an interaction item between age and PR status after 164 weeks after surgery, which is one if age is no less than 41 years old and PR is no less than 20% and zero otherwise.

Table V. Coefficients, hazard ratios (95% CI) and P-values from the prognostic model

Prognostic factor	Coefficient	HR	95% CI	P-value
Histological grade
I	0	1
II	1.33	3.78	1.74 to 8.20	<0.001
III	2.66	14.29	3.03 to 67.24	<0.001
N status
0	0	1
1	0.97	2.64	1.87 to 3.73	<0.001
2	1.94	6.96	3.50 to 13.91	<0.001
3	2.91	18.36	6.54 to 51.90	<0.001
Ki67 status
low	0.00	1.00
high	1.65	5.21	2.31 to 11.74	<0.001
PR status
low	0.00	1.00
high	-1.80	0.17	0.05 to 0.55	0.003
Age*
young	0.00	1.00
middle-aged	-1.56	0.21	0.05 to 0.83	0.025
elderly	-0.18	0.83	0.25 to 2.72	0.761
PR-age after 164 weeks*
0	0.00	1.00
1	1.59	4.92	1.04 to 23.33	0.045

HR = hazard ratio, CI = confidence interval.

* Regarding age, the patients were divided into three groups: young (< 41 years old), middle-aged (41–55 years old), and elderly (55 and older).

* Indicates the interaction item between PR status and age after 164 weeks postoperatively.

Table VI. The AUC and 95% CI for our extended Cox prognostic model at 1 year, 3 years and 5 years postoperation.

	AUC	95% CI
1-year	0.85	0.78 to 0.93
3-year	0.94	0.89 to 0.99
5-year	0.81	0.68 to 0.94

CI = confidence interval.

No competing interests reported.

Appendix.docx

Development and validation of an extended Cox prognostic model for patients with ER/PR+ and HER2‒ breast cancer: a retrospective cohort study

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Patients And Methods

1.1 Study population

1.2 Categorization of patients into ER/PR + and HER2‒ subtypes of breast cancer

1.3 Treatment for the patients.

1.4 Follow-up

1.5 Statistical analysis

Results

2.1 Patient characteristics

2.2 Prognostic factors

2.3 Analysis of patient survival times

2.4 Prognostic model development

2.4.1 Cox proportional hazards model without interaction terms

2.4.2 Cox proportional hazards model with interaction terms

2.4.3 Extended Cox prognostic model

2.5 Utilizing the nomogram

2.6 Model discrimination and calibration

Discussion

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1