Mixed invasive ductal and lobular carcinoma has distinct clinical features and predicts worse prognosis when stratified by estrogen receptor status

In order to investigate clinicopathological characteristics and prognosis of mixed invasive ductal and lobular carcinoma (IDC-L), 209,109 primary breast cancer patients diagnosed with invasive ductal carcinoma (IDC), invasive lobular carcinoma (ILC) or IDC-L were included. It was found that IDC-L patients had lower tumor grade and higher hormone receptor positive proportions than IDC patients. Moreover, IDC-L patients were younger and had a similar hormone receptor status compared with ILC patients. Kaplan-Meier plots showed that the breast cancer-specific survival (BCSS) of IDC-L patients was significantly better than IDC patients (P < 0.001) and tended to be better than ILC patients (P = 0.166). However, after adjusting for clinicopathological factors, survival advantage of IDC-L disappeared. Subgroup analysis indicated that IDC-L had higher hazard ratios (HRs) than IDC in grade 1, grade 2, ER-positive and ER-negative subgroups. Survival analysis in ER-positive and ER-negative subgroups showed that IDC-L predicted a worse prognosis than IDC. In conclusion, IDC-L is a distinct histological subtype compared with IDC and ILC. Lower grade and higher ER-positive proportions mainly contribute to its better prognosis. In both ER-positive and ER-negative subgroups, IDC-L predicts worse prognosis than IDC, which suggested the inadequacy of IDC-based therapy and the need of escalated therapy.

have indicated that IDC-L may be divided into two groups: "ILC-like" and "IDC-like" 24 . The prognostic data of IDC-L are also conflicting. One study that included 261 IDC-L patients demonstrated a similar survival outcome between patients with IDC-L and those with ILC 20 , while another study that included 140 IDC-L patients indicated that the survival outcome of IDC-L patients was similar to that of IDC patients but was significantly worse than that of ILC patients 19 . Previous studies of IDC-L are limited and contradictory because of small sample sizes, inadequate follow-up periods, a lack of adjustment for confounding factors and a lack of subgroup analyses [19][20][21][22][25][26][27] . These limitations may lead to misunderstandings and inappropriate therapies for IDC-L. Therefore, it is important to clarify the clinicopathological features and prognostic factors of IDC-L within a large population, which may help doctors determine a more precise therapeutic strategy for IDC-L patients.
By using the Surveillance, Epidemiology and End Results (SEER) database, our study aimed to investigate the clinicopathological characteristics and prognosis of IDC-L by comparing it with IDC and ILC. We also conducted a subgroup analysis to compare the survival outcome of IDC-L patients with IDC and ILC in each subgroup.

Comparison of survival outcomes among patients with IDC-L, IDC and ILC.
We compared the prognosis of these three histological types within the 11-year median follow-up period. Kaplan-Meier plots were used to evaluate the BCSS and overall survival (OS) for IDC-L, IDC and ILC (Fig. 1). In comparison with IDC patients, IDC-L patients had a significantly better BCSS and OS (both log-rank test P < 0.001). Moreover, it was obvious that the HR of IDC-L versus that of IDC increased over time according to the Scaled Schoenfeld residuals plots of the BCSS (Supplementary Fig. 1a). Additionally, compared with ILC patients, IDC-L patients exhibited a similar BCSS (log-rank test P = 0.166) and a significantly better OS (log-rank test P < 0.001). Scaled Schoenfeld residuals plots of the BCSS indicated that the HR of IDC-L versus that of ILC decreased over time ( Supplementary Fig. 1b).
The Cox proportional hazards model was used to further investigate the effect of the baseline characteristics of the disease on BCSS and OS (Table 2 and Supplementary Table 1). According to the univariate analysis, several factors were significantly associated with the BCSS and OS (Supplementary Table 1). It was suggested that patients with IDC-L exhibited a better BCSS than patients with IDC (hazard ratio (HR) = 0.88, 95% confidence interval (CI): 0.83-0.92, P < 0.001), while no significant difference was observed in the BCSS between patients with IDC-L and those with ILC (HR = 0.96, 95% CI: 0.90-1.03, P = 0.262). These significant variables were then included in the multivariate analysis to confirm their prognostic effect. Most of the variables remained significant prognostic predictors in the multivariate analysis. However, after adjusting for other prognostic predictors, IDC-L was no longer an independent prognostic predictor of BCSS compared with IDC (HR = 1.00, 95% CI: 0.96-1.05, P = 0.837) or ILC (HR = 0.99, 95% CI: 0.93-1.06, P = 0.814). We concluded that although IDC-L predicted a better prognosis than IDC and ILC, it was not an independent predictor. Subgroup analysis. In order to investigate whether IDC-L predicted homogeneous prognosis when stratified by different clinical parameters, we conducted a subgroup analysis to compare the BCSS among IDC-L, IDC and ILC in each subgroup. Forest plots of HRs in the univariate Cox analysis summarized the exploratory subgroup analysis of the BCSS and are shown in Fig. 2. Compared with IDC, IDC-L no longer had lower HRs for BCSS in some subgroups. IDC-L had higher HRs for BCSS than IDC in the grade 1 (HR = 1.52, 95% CI: 1.28-1.82, P < 0.001) and grade 2 (HR = 1.11, 95% CI: 1.04-1.19, P = 0.002) subgroups. IDC-L was also found to be a risk factor for a low BCSS rate compared with IDC in both the ER-positive (HR = 1.06, 95% CI: 1.01-1.11, P = 0.02) and ER-negative (HR = 1.32, 95% CI: 1.19-1.47, P < 0.001) subgroups. In addition, consistent with the analysis of the entire study population, the HRs of IDC-L versus those of ILC for BCSS were not significant in most subgroups. However, IDC-L was found to predict a better BCSS than ILC in patients older than 50 years of age (HR = 0.86, 95% CI: 0.79-0.93, P < 0.001) and in those with grade 1 (HR = 0.55, 95% CI: 0.45-0.67, P < 0.001) and grade 2 tumors (HR = 0.88, 95% CI: 0.80-0.96, P = 0.005), while this effect was reversed in patients younger than 50 years of age (HR = 1.25, 95% CI: 1.09-1.41, P < 0.001) and in those with tumor sizes between 2-5 cm (HR = 1.17, 95% CI: 1.06-1.29, P = 0.002). These data suggested that some clinicopathological markers, such as tumor grade and ER status, were important confounders in determining the prognosis of IDC-L patients.

Survival analysis in the ER-positive and ER-negative subgroups. Although Kaplan-Meier plots and
the univariate Cox analysis indicated that IDC-L was associated with a better BCSS in entire cohort, a subgroup analysis suggested that the BCSS was not the same in some specific subgroups. As ER is an important therapeutic target in clinical practice, we conducted an additional survival analysis of the ER-positive and ER-negative subgroups.
Kaplan-Meier plots showed that IDC-L patients in the ER-positive subgroup had a moderate BCSS, which was significantly worse than that of IDC patients (log-rank test P = 0.045) and better than that of ILC patients (log-rank test P = 0.002) (Fig. 3a). The difference in survival among these patients in terms of BCSS disappeared after the adjustment for other clinicopathological data (Supplementary Table 2). Additionally, the OS analysis of patients in the ER-positive subgroup demonstrated that the OS of IDC-L patients was similar to that of IDC patients (log-rank test P = 0.621) and was significantly better than that of ILC patients (log-rank test P < 0.001) (Fig. 3b). Moreover, IDC-L patients in the ER-negative subgroup had a significantly worse BCSS and OS than IDC patients (both log-rank test P < 0.001), while they had a similar BCSS and OS compared with ILC patients (log-rank test P = 0.192, log-rank test P = 0.909, respectively) (Fig. 3c,d). The survival disadvantage of IDC-L still existed even after adjusting for other clinicopathological data (Supplementary Table 3). These results demonstrated distinct prognostic features between the entire study population and the ER-positive and ER-negative subgroups. In the ER-negative subgroup, IDC-L independently predicted worse prognosis than IDC while in the ER-positive subgroup, further research were needed to investigate whether there exists a smaller subgroup of IDC-L patients that independently predicted worse prognosis than IDC patients.

Discussion
In this study, we retrospectively investigated the clinicopathological characteristics and prognostic features of IDC-L through a comparison of IDC-L with IDC and ILC. The results suggested that IDC-L has baseline characteristics that are distinct from those of IDC and ILC. A survival analysis indicated that IDC-L was associated with a significantly better BCSS than IDC and tended to have a better BCSS than ILC. However, the difference in BCSS between IDC-L and IDC disappeared after adjusting for confounding factors. The subgroup analysis that followed revealed that a different distribution of tumor grade and ER status accounted for the better survival of IDC-L patients than that of IDC patients. Furthermore, we found that when stratified according to their ER status, patients with IDC-L had a worse BCSS than those with IDC, which was observed for patients in both the ER-positive and ER-negative subgroups.
As the largest analysis of IDC-L to date, our research took advantage of the high number of SEER datasets to further investigate the clinicopathological characteristics of IDC-L. The proportions of the three histological types in our study were 82.4% (IDC), 8.4% (ILC) and 9.2% (IDC-L). Although the proportion of IDC-L was slightly higher than in some other studies [19][20][21][22] , we confirmed the accuracy of the data in the selection of patients with each subtype in our study through other SEER datasets based on the research of Li et al. 5 . Through a comparison of the clinicopathological characteristics among the three histological types, we observed that IDC-L not only shared similarities but also shared differences with IDC and ILC. It is therefore more reasonable to consider that the clinicopathological characteristics of IDC-L are intermediate and between those of IDC and ILC.
According to the survival analysis of the entire study population, our study demonstrated that IDC-L patients exhibited a significantly better BCSS than IDC and tended to exhibit a better BCSS than ILC patients. We attributed the differences between our study and several others 19,21,22 to the small sample sizes, short follow-up periods and different definitions of IDC-L in those studies. Furthermore, we analyzed the change in the HRs over time for IDC-L versus IDC and for IDC-L versus ILC. Scaled Schoenfeld residuals plots, which indicated the increased HR of IDC-L versus IDC over time, reminded us to be cautious with the long-term risk of IDC-L despite its better prognosis. In addition, the significantly decreased HR of IDC-L versus ILC over time suggested that the long-term prognosis of IDC-L was significantly better than that of ILC.
Our study also focused on deeper subgroup analysis of IDC, ILC and IDC-L. A multivariate Cox analysis indicated that IDC-L was not an independent predictor of a better BCSS compared with IDC. The subgroup analysis that followed focused on the identification of the underlying factors that contribute to this phenomenon and revealed that tumor grade and ER status were important confounders. IDC-L patients were more likely to have lower grade tumors and ER-positive disease, and the survival advantage of low tumor grade and ER positivity themselves contributed to the better BCSS of IDC-L versus that of IDC. Additionally, the forest plot showed that compared with IDC, IDC-L no longer predicted lower HRs in the ER-positive and ER-negative subgroups. These findings indicated that it was too ambiguous to consider IDC-L as a histological type with a better prognosis than IDC, and thus offer these patients with similar or de-escalated therapies compared with IDC patients. In contrast, more individualized therapy for IDC-L should be considered based on the personal characteristics of each patient, such as ER status. Kaplan-Meier plots confirmed that the BCSS of IDC-L was worse than that of IDC for  patients in both the ER-positive and ER-negative subgroups. The survival disadvantage of IDC-L patients in the ER-negative subgroup was still present even after we adjusted for other clinicopathological data. Current treatment for IDC-L is based on evidence from studies of IDC 28 , but our analysis suggests that this treatment might be inadequate for IDC-L patients in both the ER-positive and ER-negative subgroup. For ER-negative IDC-L patients, escalated chemotherapy may be considered after comprehensive evaluation. For ER-positive IDC-L patients, we need further research to investigate whether there exists a smaller subgroup that may need extended endocrine therapy, chemotherapy or ovarian function inhibition, even evaluated as at low risk in IDC-based guidelines.
Our research also has several limitations. First, it is a retrospective study and may have some potential selection bias. In addition, previous studies [19][20][21][22] suggested that IDC-L and ILC were associated with a significantly lower percentage of HER2-positive cases compared with IDC, which might have affected the outcome, but HER2 information in the SEER datasets was not available until 2010. To evaluate a long follow-up period, we included patients who were diagnosed between 1998 and 2007, and therefore, we lacked information on their HER2 status. Furthermore, as the understanding of IDC-L in our study focused on the clinicopathological characteristics, further studies are needed to investigate its molecular features and to reveal the genetic features and biological nature of IDC-L.
In conclusion, IDC-L is a histological subtype that is distinct from IDC and ILC with respect to clinicopathological characteristics and prognostic features. Although IDC-L is associated with a survival advantage in the entire study population, it predicts a worse prognosis than IDC for patients both in the ER-positive and ER-negative subgroup. The high proportion of ER-positive status of IDC-L patients and the survival advantage of ER-positive status itself contribute to this phenomenon. Therefore, an escalated therapeutic strategy may be considered for both ER-positive and ER-negative IDC-L patients.  An analysis of the demographic and clinical characteristics of the IDC, ILC and IDC-L subtypes included the year of diagnosis, age at diagnosis, race, marital status, grade, ER and PR status, tumor size, number of positive nodes, radiation type and surgery type. We considered the year of diagnosis as a binary variable and classified it into two groups: 1998-2002 and 2003-2007. The age at diagnosis was also considered a binary variable: <50 and ≥50. Marital status was summarized, and patients were classified as married, not married or unknown. Grade 3 and undifferentiated grade were merged into a single group. Moreover, tumor size was classified into 4 groups: <2 cm, 2-5 cm, >5 cm or unknown, while the number of positive lymph nodes was categorized into 5 groups: 0, 1-3, 4-9, ≥10 or unknown. In addition, the types of radiation were summarized as yes, no or unknown; the types of surgery were classified as no surgery, lumpectomy, mastectomy or unknown. Detailed classification information is described in Table 1. All unknown data were excluded from the Cox analysis and subgroup analysis but were included in the generation of the Kaplan-Meier curves.
Statistical analysis. The demographic and clinical characteristics of the included cases were compared across groups by the Pearson Chi-square test or Fisher's exact test for categorical nominal data and by the