Prognostic analysis of gastric signet ring cell carcinoma and mucinous carcinoma: a propensity score-matched study and competing risk analysis

Background: Limited evidence and contradictory results have been reported regarding the impact of signet ring cell carcinoma (SRC) and mucinous gastric cancer (MGC) classifications on the prognosis of gastric cancer (GC). Results: Information on 6017 patients and 266 patients was extracted from the SEER database and our hospital records, respectively. We found that patients with MGC had a better survival rate than those with SRC (P=0.012), but in the early stage, MGC was a risk factor for a poor prognosis. After PSM, for both patients from the SEER database and our hospital, the prognosis of patients with SRC was poorer than that of patients with MGC (P<0.05), but patients with MGC in early-stage GC showed poorer survival. Additionally, SRC was demonstrated to be a risk factor in the multivariate competing risk regression model for cancer-specific survival. Conclusion: Patients with SRC may have a worse prognosis than those with MGC, but for early-stage GC, patients with SRC have a better prognosis than those with MGC. Method: Patients from the SEER database and from our hospital diagnosed with SRC or MGC were included in a Cox regression analysis, multivariate competing risk model and propensity score matching (PSM) analysis.


INTRODUCTION
Gastric cancer (GC) is one of the most common types of cancer worldwide, and there were over 1,000,000 new cases of GC and an estimated 783,000 related deaths in 2018; GC is especially common in East Asia, such as Japan and Korea, compared with North America and Europe [1]. According to the WHO classification [2], GC is divided into four main types, including signet ring cell carcinoma (SRC) and mucinous gastric carcinoma (MGC). MGC accounts for a smaller proportion of GC cases than SRC and is defined as a tumor containing more than 50% extracellular mucin, while SRC is defined as a tumor with intracellular mucin pools causing the nucleus to be squeezed to the margin of the cell [2]. With respect to the prognosis of these two types of cancer, some small population-based studies have been performed. Although MGC is rare, it is usually detected at an advanced stage, leading to a poorer prognosis than common gastric adenocarcinoma. According to several studies, the 5-year survival rate is 30%-50%, which is obviously lower than that of patients without MGC [3,4]. Compared with MGC, SRC has been regarded as an independent predictive factor of survival [4]. As described in the latest large-population study [5], the 5year survival rate was 46.1%, while other studies reported a 5-year survival rate of 30%-40% [6,7]. To date, several studies have compared the differences in survival between MGC and SRC. Some studies have reported patients with MGC to have better survival than In our study, we extracted information for 6017 patients from the SEER database and 266 patients from our hospital to investigate the survival difference between MGC and SRC. Cancer-specific survival (CSS) and overall survival (OS) were regarded as the observation indices to evaluate the prognosis in the two groups. We describe the clinicopathological characteristics based on a large-population analysis and increase the knowledge of these two types of tumors with the use of propensity score matching (PSM) and competing risk analysis.

Basic clinical information of the patients
As shown in the flow chart, we finally extracted data for 6017 patients diagnosed with SRC or MGC between 2004 and 2015 from the SEER database, including 752 patients with MGC and 5265 patients with SRC (Supplementary Figure 1). Additionally, according to the inclusion and exclusion criteria (Supplementary Figure  2), we extracted data for 266 patients diagnosed with SRC or MGC between 2014 and 2020 from our hospital records. Then, according to the histological type of GC (SRC or MGC), we recorded the patient demographics, which are shown in Tables 1 and 2. As shown in Table 1, the incidence of early-onset GC in the SEER database was similar for SRC and MGC (10.54% vs 9.57%, P=0.416), while in our data, we found that early-onset GC was more frequent in SRC patients than in MGC patients (18.02% vs 5.16%, P=0.0008) ( Table 2). The proportion of male patients with MGC was higher than that with SRC (69.41% vs 53.37%, P<0.001) ( Table 1), which is in line with the results of our data ( Table 2). In addition, MGC occurred in the cardia more often than SRC (34.57% vs 18.82%), and the size of MGC tumors was larger at diagnosis than the size of SRC tumors (>3 cm, 76.19% vs 63.84%) ( Table 1), which is consistent with our data ( Table 2). The incidence of early-stage MGC was lower than that of early-stage SRC (13.56% vs 22.45%, P<0.05), which is also in agreement with our data (25.82% vs 31.53%). Although our data show that patients with MGC tended to have LNM (81.29% vs 53.15%, P<0.001) ( Table 2), there was no significant difference in LNM or distant metastasis between SRC and MGC (P>0.05) ( Table 1).

Survival differences between SRC and MGC
To investigate the survival differences between SRC and MGC, we created a K-M survival curve. The 1year, 3-year and 5-year OS rates of SRC were 63%, 37.5% and 27.5%, respectively, while those of MGC were 67.5%, 39.7% and 29%, respectively, with no significant differences ( Figure 1A). However, in terms of CSS, the 1-year, 3-year and 5-year survival rates of MGC were better than those of SRC (P=0.012) ( Figure  1B). In line with the results of the SEER database, our data revealed no significant difference in OS between the two histological types (P=0.77) (Supplementary Figure 3). We performed univariate and multivariate Cox regression analyses to determine the independent risk factors. As shown in Figures 2 and 3, both analyses indicated that distant metastasis, advanced tumor stage (T4) and SRC were independent risk factors for the survival of patients with SRC or MGC. Moreover, the results of the SEER database suggested that the number of examined LNs was a protective factor and that tumors located in the cardia predicted a worse prognosis ( Figure 2). Based on our data (Figure 3), we found that lymphatic invasion was an independent risk factor, while chemotherapy was beneficial for patients. Interestingly, in the competing risk model, in both the univariate analysis and the multivariate analysis, we found that SRC had a higher rate of GC-related death than MGC (HR=1.329, 95% CI, 1.12-1.783, P<0.05) (Supplementary Figure 4 and Table 3). Moreover, in addition to age, we found that advanced T stage, LNM, large tumor size and distant metastasis were risk factors for patient survival (Table 3). Furthermore, we extracted data for patients with early-stage GC to explore survival differences between SRC and MGC. Inconsistent with the results of the previous analysis, we found that patients with MGC histology in the early stage had poorer survival than those with SRC histology (Figures 4 and 5). The 5-year survival rate of SRC was 54.5%, while that of MGC was only 34.2% (P=0.002).

Comparison of survival between SRC and MGC after matching
To balance the confounding factors, we performed PSM at a 1:1 ratio. As shown in Table 4     AGING determined by logistic regression analysis ( Table 4). As we expected, all SMD values were lower than 0.1, and the P values were higher than 0.05, suggesting that the data were balanced. Then, we performed K-M survival analysis and found that patients with SRC had poorer survival than those with MGC (P<0.0001), consistent with the CSS results of the two groups (P<0.0001) ( Figure 6). Similarly, we performed PSM with our data by matching 45 MGC patients with 45 SRC patients ( Table 5) and found that patients with SRC had poorer survival than those with MGC (P=0.03) (Figure 7). However, in the survival analysis for patients with early-stage GC, we found that the survival of patients with MGC was poorer than that of patients with SRC after matching (Supplementary Figure 6 and Figure 8).

DISCUSSION
To the best of our knowledge, MGC is a rare subtype of GC, and few large-population studies have examined differences in the clinical features and prognosis of SRC and MGC. In our study, we included 6017 patients from the SEER database and comprehensively analyzed the differences in survival between SRC and MGC. Additionally, we extracted data for 266 patients from our hospital records to improve the reliability of the findings. Our study demonstrates that MGC does have a better prognosis than SRC via PSM and competing risk model analysis. However, in terms of early-stage disease, patients with MGC have a poorer prognosis than those with SRC.  MGC is rare and accounts for 3-10% of GC cases, while SRC accounts for 8-30% of GC cases [17][18][19].

AGING
Although their incidence rates are low, MGC and SRC are important tumor types because of their high malignancy and poor prognosis [11]. With respect to the clinical characteristics, SRC mainly occurs in younger patients, ranging from 50-60 years old. SRC appears to be more frequent in female patients and in Asian or other ethnic groups than MGC and is more frequently diagnosed in the early stage, which is in concordance AGING   AGING with our study [18,20]. In addition, the difference in prognosis between SRC and MGC is controversial. In line with our results, some studies have indicated that patients with SRC have a worse survival rate than those with MGC [21][22][23]. Some studies have suggested that patients with SRC have survival similar to those with MGC [24,25]. In our study, we used PSM to adjust for confounding factors, increasing the robustness of the results [26]. In the group of patients from the SEER database, we found that SRC could be an independent risk factor for predicting survival after PSM. Moreover, considering the limitations of the information provided by the SEER database, we extracted patients from our hospital records and adjusted for other confounding factors, such as chemotherapy, lymphatic vessel invasion and treatment method; we found that SRC was a predictor of a poor prognosis. Furthermore, since being proposed in 1972, competing risk models have been regarded as highly suitable for the clinical prediction of prognosis because these models consider the existence of a competitive risk relationship with an observable end point [27]. In both Gray's univariate and multivariate competing risk regression models, SRC was a risk factor compared to MGC. With regard to the possible causes, some studies have identified that SRC has a greater incidence of LNM, a greater risk of peritoneal metastasis, a greater risk of recurrence and lower chemosensitivity than MGC [23,28], and these differences are closely associated with the different molecular characterizations of SRC and MGC [29].
Tsenga et al. conducted a larger population-based study with 2637 patients and found that patients with MGC were inclined to have more poorly differentiated tumors and greater tumor infiltration than those with SRC [4], which could explain our results regarding the survival difference among patients with early-stage GC [4]. Moreover, considering information from both the SEER database and our hospital, the results demonstrated that SRC was more frequently considered early-onset GC, found in females and diagnosed as early-stage GC, which may have contributed to the better survival outcomes of SRC than those of other histological types of GC. Recent studies have shown that patients with early-stage SRC have a lower risk of LNM, resulting in early-stage SRC having a favorable prognosis [30,31]. Similarly, some studies have reported that patients with SRC have a significantly better survival rate than those with non-SRC [18,24]. Potential factors related to the better survival include a younger age at diagnosis and a lower incidence of lymph node invasion in earlystage SRC patients [18]. However, some studies have suggested that there is no difference in the survival of early GC patients between SRC and other types of GC [20], while SRC was found to be a risk factor for predicting survival in patients with advanced cancer [19,21]. The heterogeneous populations are an important reason for the inconsistent findings and differences in the results. With regard to the opposite survival trends of SRC in the early stage and late stage, some researchers have considered the SRC type of early GC to be characterized by a latent state with low aggressiveness and suggested that tumor invasiveness could significantly increase and accelerate when tumor cells invade the submuscular layer, resulting in a high risk of peritoneal metastasis [32]. Especially for those with CDH1 mutations, patients with SRC have poor chemosensitivity and a greater risk of metastasis [32,33].    AGING Finally, our study has some limitations that need to be discussed. First, we only focused on the OS and CSS of patients without considering cancer recurrence or disease-free survival, limiting our results in terms of clinical application. However, a competing risk model was developed to assess the value of histology in predicting survival by considering death unrelated to cancer. Second, the current analysis of the patient population could not exclude the possibility of selection bias. Therefore, interpretation of the survival differences between SRC and MGC requires caution.
In conclusion, our results indicate that compared to SRC, MGC is characterized by better survival. However, when considering early-stage GC, patients with MGC have a worse prognosis than those with SRC.

Patients
The data of all patients with GC were retrieved from the SEER database with the National Cancer Institute's SEER*Stat software (version 8.3.6). The patients did not provide informed consent because the SEER database is free for public use.  Supplementary Figure 2.

Clinicopathological factors
The patients extracted from the SEER database and our hospital for our study were divided into the SRC group and the MGC group. The patients from the SEER database were divided into two age groups: <=45 and >45 years; patients at our hospital were divided into two age groups: <=45 and >45 years. Race was classified into three types: white, black and other. Sex included male and female. Historic stage A was recorded as localized, regional or distant.

Statistical analysis
For basic statistical analysis, the patients were divided into two groups, namely, the SRC and MGC groups, and Pearson's chi-squared test was utilized to investigate the associations among categorical variables.
To explore the potential risk factors for CSS, we performed univariate and multivariate Cox regression, and the results are presented as the hazard ratio (HR) with 95% confidence interval (CI). With respect to the OS and CSS of patients with SRC and MGC, we created survival curves using R software. For the competing risk model, we constructed the model as described in a previous study [35]. Briefly, we selected CSS as the outcome of interest, whereas death due to other causes was considered a competing risk event, and a patient who was alive was regarded as a censored event. We created cumulative risk curves using Fine and Gray's competing risk regression analysis. In addition, a multivariate competing risk model was used to explore the potential risk factors for CSS by R software.

AGING
Regarding the imbalance between the SRC and MGC groups, we performed PSM and inverse probability of treatment weighting (IPTW) to obtain new data for analysis with the MatchIt package in R software. The caliper value was set as 0.02, and the effect was evaluated based on the standardized mean difference (SMD) and P value. The effect was balanced when the SMD was less than 0.1 or the P value was greater than 0.05 [36]. The detailed process was as follows. First, we calculated the propensity scores of each patient according to histological type (SRC and MGC) with the multivariate logistic regression model. Then, we matched patients between the two groups at a ratio of 1:1. Next, we analyzed the differences in all variables between the SRC and MGC groups with the chi-squared test. Finally, we explored the correlation between survival and histological type by performing a K-M survival analysis.
All statistical analyses were performed with R software (version 3.6.1, StataCorp LLC, College Station, Texas). For comparisons among different patient groups, the chisquared test was used for categorical variables, Student's t-test was used for continuous variables with a Gaussian distribution, and the nonparametric Kruskal-Wallis ranksum test was used for nonnormally distributed continuous variables or ordinal categorical variables. The chisquared test was carried out with SPSS (version 24.0). The results were considered statistically significant when the P value was less than 0.05.

Ethics statement
Ethics approval and consent was obtained from SEER database.

AUTHOR CONTRIBUTIONS
CTT: data collection, data analysis, and manuscript writing. CYZ and YXC: project development.