Clinicopathological, immunohistochemical, and overall-survival associations of poorly cohesive gastric carcinoma stratified by signet ring cell percentage

Abstract

BACKGROUND

Poorly cohesive carcinoma (PCC) is a major histological type of gastric cancer characterized by decreased intercellular adhesion and manifests as isolated tumor cells or small, scattered clusters. PCC is morphologically heterogeneous and comprises two pathological components: Signet ring cell carcinoma (SRC) and PCC-not otherwise specified (PCC-NOS). According to the European Consensus, PCC is further classified into three subtypes based on the SRC proportion: SRC (≥ 90%), SRC/PCC-NOS (10% < SRC < 90%), and PCC-NOS (≤ 10%).

AIM

To comprehensively evaluate the clinicopathological, immunohistochemical, and prognostic profiles of PCC and its subtypes stratified by SRC proportion.

METHODS

This single-center retrospective cohort study included 1592 patients who underwent gastrectomy for gastric cancer between September 2016 and December 2018, with a loss-to-follow-up rate of 23.6%. Among them, 374 patients were diagnosed with PCC (70 SRC, 171 SRC/PCC-NOS, and 133 PCC-NOS) and 1218 had non-PCC carcinoma. Propensity score matching (PSM) was performed to minimize selection bias between the PCC and non-PCC groups and among the PCC subtypes. Survival outcomes, clinicopathological features, and immunohistochemical markers were analyzed.

RESULTS

Five-year overall survival (OS) did not differ significantly between the non-PCC and PCC groups (68.87% vs 69.11%, P = 0.4425), and this finding remained consistent after PSM (P = 0.9189). Among the PCC subtypes, the SRC group presented at an earlier tumor stage. Unadjusted 5-year OS was highest in the SRC group (80.77%), followed by the PCC-NOS (67.80%) and SRC/PCC-NOS (65.03%) groups (P = 0.0460). However, this survival advantage was no longer significant after PSM (P = 0.5820). Immunohistochemical analysis revealed inter-subtype heterogeneity in the expression of Ki67, mismatch repair (MMR) proteins, and programmed death protein ligand 1 (PD-L1). PD-L1 expression tended to be lower in the SRC group, whereas all cases with MMR protein loss were PCC-NOS, while the result remains exploratory and descriptive with the limitation of sample size.

CONCLUSION

This study demonstrated that PCC as a whole had a prognosis comparable to that of other gastric cancers. However, PCC subtypes exhibited distinct clinicopathological and immunohistochemical profiles, with SRC associated with an earlier stage and a better OS. These findings support the clinical relevance of PCC subclassification and warrant further validation in larger cohorts.

Key Words: Gastric adenocarcinoma; Poorly cohesive carcinoma; Signet ring cell carcinoma; Pathological classification; Prognosis; Retrospective cohort study

Core Tip: This single-center retrospective cohort study evaluated poorly cohesive gastric carcinoma and its subtypes, defined by the proportions of signet ring cell (SRC) and poorly cohesive carcinoma-not otherwise specified (PCC-NOS) components, and identified distinct clinicopathological and immunohistochemical profiles. Although the SRC-dominant group was associated with superior overall survival, this prognostic advantage disappeared after propensity score matching, whereas the PCC-NOS-dominant group was potentially associated with microsatellite instability and programmed death protein ligand 1 expression. These findings emphasize the need for further investigation and comprehensive evaluation of poorly cohesive carcinoma and its subtypes.

INTRODUCTION

Poorly cohesive carcinoma (PCC) is a type of gastric cancer characterized by decreased cell-to-cell adhesion, with tumor cells occurring individually or in small, scattered groups. It includes signet ring cell carcinoma (SRC) and PCC-not otherwise specified (PCC-NOS). In 2010, the fourth edition of the World Health Organization (WHO) classification of digestive system tumors listed PCC as one of the main pathological types of gastric adenocarcinoma for the first time[1]. This classification was retained in the fifth edition of the WHO classification, published in 2019[2]. Previously, SRC was considered an independent type of gastric cancer and was widely regarded as aggressive and associated with a poor prognosis. However, studies of SRC have yielded inconsistent results. For instance, Graziosi et al[3] reported that early gastric cancer with SRC components had a better prognosis than disease without SRC components, whereas the opposite was observed in advanced-stage cancer. In contrast, Kang et al[4] suggested that SRC was no more aggressive than other subtypes in early gastric cancer. These conflicting findings may reflect the lack of standardized classification[5].

The effects of PCC on tumor behavior and survival also remain controversial. Koseki et al[6] reported poorer survival in PCC than in non-PCC; while Nakazawa et al[7]and Jouini et al[8] did not support this association when considering other factors, reporting no significant difference in survival between patients with PCC and those with non-PCC gastric cancer. The two PCC subtypes, SRC and PCC-NOS, are pathologically heterogeneous and differ significantly in tumor behavior and prognosis. Therefore, based on the proportions of SRC and PCC-NOS components, the European Consensus subdivides PCC into three categories: Pure SRC (≥ 90% SRC), PCC with an SRC component (10% < SRC < 90%), and PCC-NOS (≤ 10% SRC)[9]. Differences in prognosis, tumor aggressiveness, and gene mutations among these subtypes have been demonstrated in several studies[10-17].

In this study, we collected data from patients with gastric cancer and compared clinicopathological and immunohistochemical characteristics and prognosis between PCC and non-PCC and among PCC subtypes. The primary outcome was overall survival (OS), with restricted mean survival time (RMST) as a secondary measure. We pre-specified the hypothesis that the SRC-dominant subtype would be associated with superior OS.

MATERIALS AND METHODS

Study design and patient selection

This was a single-center retrospective cohort study conducted in the Department of Gastrointestinal Surgery at Renji Hospital, Shanghai Jiao Tong University School of Medicine, a comprehensive tertiary academic medical center in Shanghai, China. Patients who underwent gastrectomy for gastric cancer in our department between September 2016 and December 2018 were retrospectively identified. Patient identifiers were removed before the analysis dataset was constructed to ensure de-identification.

The patients were identified from two sources: The pathological database and the departmental follow-up register. Both sources followed the same inclusion and exclusion criteria. Cases of potential PCC were identified from the pathology database using a combination of keywords, including: “poorly cohesive”, “signet-ring cell”, and “poorly differentiated”. The pathology reports of all identified cases were manually reviewed to confirm the diagnosis and determine the SRC proportion. The inclusion criteria were as follows: (1) Pathologically confirmed gastric adenocarcinoma; (2) Surgical resection; and (3) Complete clinicopathological and follow-up data. The exclusion criteria were as follows: (1) Remnant gastric cancer; (2) Neoadjuvant chemotherapy or radiotherapy (excluded because of altered histology and prognosis); (3) A history of other malignancies within 5 years; (4) Perioperative mortality; and (5) Incomplete medical records or loss to follow-up immediately after discharge. The study flow diagram is shown in Figure 1.

Figure 1 Study flow diagram. The patients were identified from two sources: (1) The pathological database, searched using poorly cohesive carcinoma (PCC)-related keywords; and (2) The departmental follow-up registration system for non-PCC cases. PCC was further classified into signet ring cell carcinoma (SRC) (≥ 90%), mixed SRC/PCC-not otherwise specified (NOS) (10% < SRC < 90%), and PCC-NOS (≤ 10%). Cross-validation was used to ensure that no PCC cases were wrongly classified as non-PCC cases. PCC: Poorly cohesive carcinoma; SRC: Signet ring cell carcinoma; NOS: Not otherwise specified.

Clinicopathological analysis

Two senior pathologists re-evaluated the specimens from all cases considered PCC and determined the proportions of the different pathological components. Both pathologists were blinded to clinical outcomes. The classification was based on well-defined cutoffs for the proportions of SRC and PCC-NOS components, and the final subtype for each case was determined by consensus between the two reviewers. Any discrepancies were resolved by discussion. According to the European Consensus, PCC was further classified into three subtypes: SRC (SRC ≥ 90%), SRC/PCC-NOS (10% < SRC < 90%) and PCC-NOS (SRC ≤ 10%). Finally, our cohort included 1218 patients with non-PCC and 374 with PCC: Of 70 patients in the SRC group, 171 in the SRC/PCC-NOS group, and 133 in the PCC-NOS group.

Other clinicopathological variables collected included age, sex, tumor location (upper/middle/Lower), invasion depth (T stage), lymph node metastasis (N stage), and pathological TNM stage. Pathological stage was reclassified according to the American Joint Committee on Cancer 9^th edition for consistency. The tumor site was determined by the predominant location of the tumor mass.

Immunohistochemistry

Immunohistochemistry (IHC) results for all markers [carcinoembryonic antigen (CEA), Ki67, HER2, P53, epidermal growth factor receptor (EGFR), mismatch repair (MMR) proteins: MutL homolog 1 (MLH1), MutS homolog (MSH) 2, MSH6, and postmeiotic segregation increased 2 (PMS2), programmed death protein 1 (PD-1), and programmed death protein ligand 1 (PD-L1)] were retrieved from original pathology reports. HER2 was scored per American Society of Clinical Oncology/College of American Pathologists guidelines; fluorescence in situ hybridization status was not collected. MMR status included MLH1, MSH2, MSH6, and PMS2. Loss of any marker was defined as deficient MMR (dMMR). PD-L1 was assessed using the combined positive score (CPS). Ki67 was recorded as the percentage of positive cases, and analyzed as a continuous variable. Percentages were calculated among evaluable cases. Antibody clones and platforms were not available for all markers. Findings for MMR, PD-L1, and PD-1 were exploratory and descriptive because of the limited number of evaluable cases.

Surgical procedures and lymph node evaluation

All patients underwent gastrectomy with dissection (D) 2 lymphadenectomy for pathologically confirmed gastric cancer, consistent with institutional practice and national guidelines. Lymph nodes were numbered according to Japanese Gastric Cancer Association guidelines, and the D1/D2 compartments were defined according to tumor location (upper/middle/Lower third).

Follow-up protocol

OS was defined as the time from the date of surgery to death from any cause, truncated at 60 months. Patients were censored at their last known follow-up date if they were alive at 60 months, lost to follow-up, or had not yet reached 60 months of follow-up at the censoring date. Follow-up data were obtained from the departmental registry, with follow-up administratively censored at December 2021. For the PCC subtypes, dedicated follow-up was extended to April 2024, ensuring a minimum follow-up of 60 months for all patients in this group.

Follow-up was conducted according to a standardized protocol. Patients were followed up every 3 months during the first year, every 6 months from the first through third years, and annually thereafter. Follow-up assessments included physical examination, laboratory tests, imaging studies [chest computed tomography (CT) and chest X-ray, abdominal CT/ultrasound], endoscopic surveillance, and positron emission tomography-CT when indicated. Follow-up was primarily conducted through outpatient clinic visits and was supplemented by telephone interviews. For patients who could not be reached after three telephone attempts, the most recent outpatient visit after the preceding follow-up, if available, was recorded as the last contact. These patients were censored at the date of last contact in the survival analysis.

Survival analysis

Survival was analyzed using the Kaplan-Meier method and compared using the log-rank test. Univariable Cox proportional hazards models were used to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). Survival was censored at 60 months for 5-year OS estimation. Subgroup-analysis interaction tests were performed in multivariable Cox models using likelihood ratio tests. RMST was estimated at 60 months as a complementary measure to the HR. Between-group differences in RMST and 95%CI were calculated using the standard normal approximation. The proportional hazards assumption was assessed using Schoenfeld residuals (Grambsch-Therneau test) for overall and propensity score matching (PSM) cohorts. For stage-stratified analyses, the assumption was considered satisfied given the exploratory nature of these comparisons and the absence of statistically significant survival difference across strata. Sensitivity analyses included worst- and best-case scenarios for loss to follow-up, and a restricted analysis excluding stage IV patients.

PSM

To reduce selection bias between the two main groups and among the three subtypes, PSM was performed. The estimand of interest was the average treatment effect in the treated. For PSM between two main groups, propensity scores were estimated using logistic regression with age, sex, T stage, N stage, and M stage as covariates; patients were matched at ratios of 1:1 using nearest-neighbor matching with a caliper of 0.1 and replacement. For PSM among three subgroups of PCC, propensity scores were estimated using logistic regression with age, sex, and stage as covariates, due to the limitation of sample size; patients were matched at ratios of 1:1:1. After matching, covariate balance was assessed using standardized mean differences (SMDs), with values < 0.1 considered adequate. After matching, Cox regression was performed with the treatment indicator as the only covariate, using robust standard errors. PSM was performed using the Mengte Cloud Statistics platform (https://mengte.pro/psm)[18].

Univariable and multivariable Cox regression analysis

Univariable Cox regression was performed for each candidate variable (sex, age, T stage, N stage, and M stage). All variables were then entered into a multivariable Cox model simultaneously, regardless of their univariable significance, to adjust for potential confounding. Results are presented as HRs with 95%CIs.

Statistical analysis

Data were analyzed using R version 4.3.2 and GraphPad Prism 8.0 software. Clinical and pathological features were analyzed using the “tableone” package; univariable and multivariable Cox regression, proportional-hazards assumption testing, and subgroup interaction analysis were performed using the “survival” packages; RMST was performed using the “survRM2” package. Continuous variables were summarized as the mean (SD) and compared using Student’s t-test. Categorical variables were summarized as n (%) and compared using Pearson’s χ² test or Fisher’s exact test, as appropriate. P < 0.05 was considered statistically significant. All tests were two-sided.

RESULTS

Clinical characteristics and pathological features

A total of 1218 non-PCC and 374 PCC patients were involved in this study (Table 1). The non-PCC patients were older than the PCC patients (P < 0.001), and the non-PCC patients included more males (P < 0.001). Tumor invasion did not differ significantly between the two groups, but more patients in the PCC group had > 6 metastatic lymph nodes (N3) (P = 0.006). The proportion of patients with stage I cancer was similar between the groups, whereas stages II and III were more frequent in the non-PCC group and stage IV was more frequent in the PCC group (P = 0.016). The primary reason patients with stage IV disease underwent palliative surgery was the occurrence of complications such as bleeding or obstruction, which adversely affected their prognosis.

Table 1 Clinicopathological features of poorly cohesive carcinoma and non-poorly cohesive carcinoma, n (%).

		Non-PCC	PCC	P value
n		1218	374
Age, mean (SD)		64.75 (10.19)	59.87 (11.84)	< 0.001
Sex	Female	323 (26.5)	176 (47.1)	< 0.001
	Male	895 (73.5)	198 (52.9)
Stage	I	337 (28.0)	104 (28.6)	0.016
	II	232 (19.3)	54 (14.8)
	III	606 (50.3)	187 (51.4)
	IV	29 (2.4)	19 (5.2)
T	T1	240 (19.9)	68 (18.7)	0.216
	T2	161 (13.3)	57 (15.7)
	T3	81 (6.7)	15 (4.1)
	T4	725 (60.1)	224 (61.5)
N	N0	492 (40.4)	140 (37.5)	0.006
	N1	187 (15.4)	44 (11.8)
	N2	225 (18.5)	59 (15.8)
	N3	314 (25.8)	130 (34.9)
M	M0	1175 (97.591)	353 (94.892)	0.016
	M1	29 (2.409)	19 (5.108)

Missing data: Age, 0; sex, 0; stage, 24 (1.51%). Percentages are based on the total cohort (n = 1592). PCC: Poorly cohesive carcinoma.

The PCC group was further classified into three subgroups: Of 70 patients in the SRC group, 171 in the SRC/PCC-NOS group, and 133 in the PCC-NOS group (Table 2). The SRC group was smaller than the other groups. Patients in the SRC/PCC-NOS group were the youngest overall (P = 0.048). Although men predominated in the PCC group overall, only the PCC-NOS subgroup had a higher proportion of men, whereas women predominated in the SRC and SRC/PCC-NOS groups (P < 0.001). SRC was more likely to occur in the lower third of the stomach, SRC/PCC-NOS in the middle third, and PCC-NOS in the upper third (P = 0.024). The SRC group had the least tumor invasion, lymph node metastasis, and distant metastasis. Tumor invasion was deepest in the PCC-NOS group. Patients in the SRC/PCC-NOS group had higher rates of lymph node and distant metastases than those in the PCC-NOS group.

Table 2 Clinicopathological features of defined subtypes within poorly cohesive carcinoma, n (%).

		SRC	SRC/PCC-NOS	PCC-NOS	P value
n		70	171	133
Age, mean (SD)		60.39 (11.09)	58.30 (12.76)	61.62 (10.75)	0.048
Sex	Female	37 (52.9)	95 (55.6)	44 (33.1)	< 0.001
	Male	33 (47.1)	76 (44.4)	89 (66.9)
Site	Upper	6 (8.8)	12 (7.2)	25 (19.1)	0.024
	Middle	30 (44.1)	79 (47.6)	49 (37.4)
	Lower	32 (47.1)	75 (45.2)	57 (43.5)
Stage	I	35 (51.5)	44 (26.3)	25 (19.4)	< 0.001
	II	12 (17.6)	24 (14.4)	18 (14.0)
	III	19 (27.9)	86 (51.5)	82 (63.6)
	IV	2 (2.9)	13 (7.8)	4 (3.1)
T	T1	28 (41.2)	28 (16.8)	12 (9.3)	< 0.001
	T2	9 (13.2)	27 (16.2)	21 (16.3)
	T3	2 (2.9)	8 (4.8)	5 (3.9)
	T4	29 (42.6)	104 (62.3)	91 (70.5)
N	N0	45 (64.3)	54 (31.8)	41 (30.8)	< 0.001
	N1	6 (8.6)	22 (12.9)	16 (12.0)
	N2	7 (10.0)	25 (14.7)	27 (20.3)
	N3	12 (17.1)	69 (40.6)	49 (36.8)
M	M0	67 (97.1)	155 (92.3)	123 (96.9)	0.135
	M1	2 (2.9)	13 (7.7)	4 (3.1)

Missing data: Age, 0; sex, 0; stage, 10 (2.7%). Percentages are based on the total cohort (n = 374). SRC: Signet ring carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified.

Lymph node status

Among the 374 patients in the PCC group, a total of 9385 lymph nodes (mean, 25.09 per case) were dissected and examined pathologically (Table 3). The number of metastatic lymph nodes was lower in the SRC group than in the SRC/PCC-NOS and PCC-NOS groups, and this pattern was consistently observed in both the D1, D2, and D3 compartments. Specifically, the SRC group had 192 metastatic lymph nodes (11.8%) compared with 1349 (31.2%) in the SRC/PCC-NOS group and 1053 (30.7%) in the PCC-NOS group (P = 0.0018). In the D1 compartment, the metastatic rates were 13.8%, 33.9% and 34.5%, respectively; in the D2 compartment, the rates were 5.1%, 21.6% and 23.3%, respectively; in the D3 compartment, the rates were 8.9%, 29.7% and 22.7%, respectively (P < 0.001). However, the SRC group had a higher proportion of early-stage disease, which may partly account for the lower nodal burden. When stratified by N stage, the number of metastatic lymph nodes differed significantly among the three subtypes in N3 cases, with the SRC group having fewer metastatic lymph nodes than the other two groups (SRC group: 10.58 per case; SRC/PCC-NOS group: 16.16 per case; PCC-NOS group: 16.00 per case). This finding suggests that even among patients with similarly advanced nodal disease, SRC tumors may exhibit a distinct pattern of lymphatic metastasis.

Table 3 Lymph node burden among defined subgroups of poorly cohesive carcinoma, n (%).

		SRC	SRC/PCC-NOS	PCC-NOS	P value
Total examined LNs, n		1624	4327	3434
Examined LNs per case, n		23.20	25.30	25.82
Examined LNs, n	D1	1147	2964	2293
	D2	253	736	660
	D3	224	627	481
Total Metastatic LNs		192 (11.8)	1349 (31.2)	1053 (30.7)
Metastatic LNs	D1	159 (13.9)	1004 (33.9)	790 (34.5)	0.0018
	D2	13 (5.1)	159 (21.6)	154 (23.3)
	D3	20 (8.9)	186 (29.7)	109 (22.7)
Metastatic LNs	N1	9 (7.4)	37 (7.4)	26 (7.0)	< 0.001
	N2	37 (20.9)	94 (16.4)	149 (22.8)
	N3	127 (46.2)	1115 (65.7)	784 (59.6)
Metastatic LNs per case, n	N1	1.50	1.68	1.63
	N2	5.28	3.76	5.52
	N3	10.58	16.16	16

SRC: Signet ring carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified; LNs: Lymph nodes.

IHC

IHC results were statistically analyzed for the markers CEA, Ki67, HER2, P53, EGFR, MLH1, MSH2, PMS2, PD-1, and PD-L1 (Table 4). Most markers were not expressed; however, Ki67, MMR proteins, and PD-L1 were expressed differently across subgroups. Ki67 expression was lowest in the SRC group and highest in the PCC-NOS group, indicating differences in tumor proliferative activity among the three subgroups. For gastric cancer, a PD-L1 CPS ≥ 1 is generally considered to indicate potential benefit from immunotherapy. The SRC group had lower PD-L1 expression than the other groups (P = 0.035). Regarding MMR protein expression, 63 cases were evaluated for MLH1, MSH2, MSH6, and PMS2; of these, four cases showed dMMR, and all belonged to the PCC-NOS group (Supplementary Table 1). To assess potential selection bias, baseline characteristics (sex, age, and stage) were compared between patients with and without evaluable MMR and PD-L1 results, respectively (Supplementary Table 2 and 3). For MMR, no significant differences were observed in sex, age, or stage. For PD-L1, sex and age were comparable; although the evaluable group had a higher proportion of stage III and a lower in stage IV (P = 0.018), the proportion in early and advanced stages was balanced between groups. Given the limited sample size of stage IV (19 cases), the absolute impact of this difference is limited. The proportion of cases with MMR and PD-L1 increased significantly over the study period (both P < 0.001), consistent with the gradual adoption of these assays. However, given the limited number of evaluable cases, these findings are exploratory and hypothesis-generating and require further validation in larger datasets.

Table 4 Immunohistochemical characteristics of defined subtypes within poorly cohesive carcinoma.

	Evaluable cases		SRC	SRC/PCC-NOS	PCC-NOS	P value
n			70	171	133
CEA, n/N (%)	209	-	6/42 (14.3)	5/89 (5.6)	6/78 (7.7)	0.416
		-/+	4/42 (9.5)	7/89 (7.9)	6/78 (7.7)
		+	26/42 (61.9)	72/89 (80.9)	61/78 (78.2)
		++	1/42 (2.4)	2/89 (2.2)	1/78 (1.3)
		+++	5/42 (11.9)	3/89 (3.4)	4/78 (5.1)
Ki67, mean (SD)	355		0.40 (0.24)	0.52 (0.22)	0.55 (0.20)	< 0.001
HER2, n/N (%)	350	-	44/62 (71.0)	106/162 (65.4)	82/126 (65.1)	0.122
		+	14/62 (22.6)	42/162 (25.9)	25/126 (19.8)
		++	2/62 (3.2)	11/162 (6.8)	18/126 (14.3)
		+++	2/62 (3.2)	3/162 (1.9)	1/126 (0.8)
P53, n/N (%)	260	-	12/47 (25.5)	33/113 (29.2)	26/100 (26.0)	0.244
		-/+	3/47 (6.4)	9/113 (8.0)	2/100 (2.0)
		+	16/47 (34.0)	33/113 (29.2)	23/100 (23.0)
		++	5/47 (10.6)	11/113 (9.7)	10/100 (10.0)
		+++	11/47 (23.4)	27/113 (23.9)	39/100 (39.0)
EGFR, n/N (%)	111	-	3/18 (16.7)	13/45 (28.9)	9/48 (18.8)	0.545
		-/+	1/18 (5.6)	4/45 (8.9)	2/48 (4.2)
		+	14/18 (77.8)	28/45 (62.2)	37/48 (77.1)
MLH1, n/N (%)	63	-	0/7 (0.0)	0/32 (0.0)	4/24 (16.7)	0.031
		+	7/7 (100.0)	32/32 (100.0)	20/24 (83.3)
MSH2, n/N (%)	63	+	7/7 (100.0)	32/32 (100.0)	24/24 (100.0)
MSH6, n/N (%)	63	-	0/7 (0.0)	0/32 (0.0)	1/24 (4.2)	0.438
		+	7/7 (100.0)	32/32 (100.0)	23/24 (95.8)
PMS2, n/N (%)	63	-	0/7 (0.0)	0/32 (0.0)	4/24 (16.7)	0.031
		+	7/7 (100.0)	32/32 (100.0)	20/24 (83.3)
PD-1, n/N (%)	38	-	6/6 (100.0)	13/15 (86.7)	14/17 (82.4)	0.546
		+	0/6 (0.0)	2/15 (13.3)	3/17 (17.6)
PD-L1, n/N (%)	171	CPS < 1	21/23 (91.3)	68/78 (87.2)	51/70 (72.9)	0.035
		CPS ≥ 1	2/23 (8.7)	10/78 (12.8)	19/7 (27.1)

Data are presented as n/N (%), where N represents the number of evaluable cases within each subtype group. Immunohistochemistry markers were not available for all patients; analyses based on evaluable cases were exploratory. SRC: Signet ring carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified; CEA: Carcinoembryonic antigen; EGFR: Epidermal growth factor receptor; MLH1: MutL homolog 1; MSH: MutS homolog; PMS2: Postmeiotic segregation increased 2; PD-1: Programmed death protein 1; PD-L1: Programmed death protein ligand 1; CPS: Combined positive score.

Survival analysis: Non-PCC vs PCC

In this cohort, the median follow-up period was 36 months (interquartile range: 25-49 months; range: 1-89 months), with 23.618% lost to follow-up. During the follow-up duration, 405 deaths and 1187 censored observations were documented. For non-PCC group, 306 deaths and 912 censored cases occurred in 1218 patients; for PCC group, 99 deaths and 275 censored cases occurred in 374 patients. Five-year survival did not differ significantly between the non-PCC and PCC groups (non-PCC: 68.869% vs PCC 69.108%, HR: 0.9158, 95%CI: 0.7337-1.143, P = 0.4425) (Figure 2A; detailed HRs for all group comparisons are summarized in Supplementary Tables 4 and 5). The proportional hazards assumption was not violated (global P = 0.36) (Supplementary Table 6). RMST at 60 months was 48.690 months (95%CI: 47.584-49.795) for non-PCC group and 49.704 months (95%CI: 47.849-51.558) for PCC patients, with a difference of 1.014 months (95%CI: -1.425 to 3.173, P = 0.357).

Figure 2 Overall survival. A: Overall survival (OS) of poorly cohesive carcinoma (PCC) and non-poorly cohesive carcinoma (non-PCC); B: OS of PCC and non-PCC after propensity score matching (PSM); C: OS of defined subtypes within PCC. Kaplan-Meier curves for the three subtypes of poorly cohesive carcinoma: Signet ring cell carcinoma (SRC), mixed SRC and poorly cohesive carcinoma-not otherwise specified (SRC/PCC-NOS), and pure PCC-NOS; D: OS of defined subtypes within PCC after PSM. PCC: Poorly cohesive carcinoma; PSM: Propensity score matching; SRC: Signet ring cell carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified.

After PSM, the survival difference remained nonsignificant (76.20% vs 75.05%, P = 0.9189). When stratified by stage, no significant difference was observed between the non-PCC and PCC groups for stage I (92.43% vs 95.05%, P = 0.2696), stage II (77.88% vs 84.73%, P = 0.2132), stage III (53.35% vs 52.51%, P = 0.5366), or stage IV (35.25% vs 32.07%, P = 0.3038) (Figure 2B, Supplementary Figure 1, Supplementary Table 7). The proportional hazards assumption was not violated (global P = 0.57) (Supplementary Table 6).

Survival analysis: PCC subtypes

During the follow-up duration, 11 deaths and 59 censored cases were documented in the SRC group; corresponding cases were 51 and 120 in the SRC/PCC-NOS group, and 37 and 96 cases in the PCC-NOS group. Five-year survival differed significantly among the PCC subgroups (SRC: 80.77%; SRC/PCC-NOS: 65.03%; PCC-NOS: 67.80%; P = 0.0460) (Figure 2C). On univariable Cox regression, SRC/PCC-NOS (HR: 2.212, 95%CI: 1.299-3.766, P = 0.0137) and PCC-NOS (HR: 2.078, 95%CI: 1.161-3.720, P = 0.0123) had similar survival, and both had significantly worse survival than the SRC group. The proportional hazards assumption was not violated (global P = 0.095) (Supplementary Table 6). Among the PCC subtypes, RMST at 60 months was 54.874 months (95%CI: 51.983-57.766) for SRC group, 48.555 months (95%CI: 45.760-51.350) for SRC/PCC-NOS group, and 48.377 months (95%CI: 44.977-51.777) for PCC-NOS group. Compared with SRC, the RMST difference was 6.320 months (95%CI: 2.298-10.347, P = 0.002) for SRC/PCC-NOS and 6.498 months (95%CI: 2.034-10.961, P = 0.004) for PCC-NOS group.

Because baseline characteristics differed among the groups, PSM was used for further comparison. PSM balanced age and sex among the three subgroups (Table 5). After PSM, 56 patients were matched in each subgroup. No statistically significant differences were observed among the groups in age, sex, or stage (age: P = 0.984, SMD = 0.094; sex: P = 0.909, SMD = 0.048; stage: P = 0.995, SMD = 0.080) The difference in OS also disappeared (79.538% vs 74.935% vs 86.322%, P = 0.5820). The SRC group showed no survival advantage over SRC/PCC-NOS (P = 0.7243) or PCC-NOS (P = 0.4912) (Figure 2D, Supplementary Figure 2). The proportional hazards assumption was not violated (global P = 0.74) (Supplementary Table 6).

Table 5 Clinicopathological features of signet ring carcinoma, signet ring carcinoma/poorly cohesive carcinoma-not otherwise specified, and poorly cohesive carcinoma-not otherwise specified after propensity score matching, n (%).

		Before PSM					After PSM
		SRC	SRC/PCC-NOS	PCC-NOS	P value	SMD	SRC	SRC/PCC-NOS	PCC-NOS	P value	SMD
n		70	171	133			56	56	56
Sex	Female	37 (52.857)	95 (55.556)	44 (33.083)	< 0.001	0.309	27 (48.214)	29 (51.786)	29 (51.786)	0.909	0.048
	Male	33 (47.143)	76 (44.444)	89 (66.917)			29 (51.786)	27 (48.214)	27 (48.214)
Age	< 60	28 (40.000)	82 (47.953)	53 (39.850)	0.581	0.144	26 (46.429)	25 (44.643)	25 (44.643)	0.984	0.094
	60-74	36 (51.429)	74 (43.275)	65 (48.872)			27 (48.214)	27 (48.214)	26 (46.429)
	≥ 75	6 (8.571)	15 (8.772)	15 (11.278)			3 (5.357)	4 (7.143)	5 (8.929)
Stage	I	35 (51.471)	44 (26.347)	25 (19.380)	< 0.001	0.580	27 (48.214)	26 (46.429)	25 (44.643)	0.995	0.080
	II	12 (17.647)	24 (14.371)	18 (13.953)			8 (14.286)	8 (14.286)	10 (17.857)
	III	19 (27.941)	86 (51.497)	82 (63.566)			19 (33.929)	20 (35.714)	19 (33.929)
	IV	2 (2.941)	13 (7.784)	4 (3.101)			2 (3.571)	2 (3.571)	2 (3.571)

PSM: Propensity score matching; SRC: Signet ring carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified; SMD: Standardized mean difference.

When the cohort was stratified by stage for survival analysis, the differences among the subgroups were no longer significant in stage I (P = 0.3509), stage II (P = 0.3336), stage III (P = 0.4277), or stage IV (P = 0.7807). This finding suggests that the superior prognosis observed in the SRC group was primarily attributable to the earlier disease stage at initial diagnosis (Supplementary Figure 2).

To address potential interactions between nodal burden and tumor location, a subgroup survival analysis stratified by N stage and tumor location was conducted among the PCC subtypes. Given the limited number of patients in some strata, N stage was categorized as pN0 or pN+. Subgroups with fewer than five cases were excluded from the analysis. Among pN+ patients, upper-third tumors were associated with a poor prognosis in the SRC/PCC-NOS (P < 0.0001) and PCC-NOS (P = 0.0002) groups, followed by middle-third tumors. In contrast, survival in the SRC group (P = 0.3549) was less affected by tumor location. Subgroup-analysis interaction tests showed significant heterogeneity in the prognostic effect of tumor site (subgroup × site: P = 0.002) or lymph node status (subgroup × N stage: P < 0.001) (Supplementary Figure 3, Supplementary Table 8).

In addition, eight patients with PCC had tubular/papillary components (Supplementary Figure 4). Their 5-year OS was 46.875%, which was worse than that of the other groups.

Sensitivity analysis

Among the cohort, 92 patients were classified as stage IV. This group included a mixture of curative-intent R0 resections for oligometastatic disease and palliative surgery, which could not be reliably distinguished retrospectively. When the survival analysis was restricted to stage I-III, differences in N stage and overall stage between the non-PCC and PCC groups were no longer significant, whereas the statistical significance of all other variables remained unchanged (Supplementary Table 9, Supplementary Figure 5). The survival difference between the non-PCC and PCC groups remained nonsignificant (70.267% vs 71.561%, P = 0.2399). Among the three subtypes, baseline characteristics remained consistent with the primary analysis, with SRC continuing to show more favorable OS (81.261% vs 68.540% vs 68.525%, P = 0.0801) (Supplementary Figure 6, Supplementary Table 10).

Since not all patients reached a follow-up duration of 60 months, OS was truncated at 36 months to ensure comparable follow-up across all patients in the overall cohort. The OS of the non-PCC and PCC groups was comparable without significant difference (75.420% vs 78.414%, P = 0.2131), consistent with the primary analysis.

Given the substantial proportion of patients lost to follow-up, a sensitivity analysis was performed. In the overall cohort, patients lost to follow-up had more advanced disease than those who completed follow-up (P < 0.001) (Supplementary Table 11). However, within the PCC group, baseline characteristics did not differ significantly between patients who completed follow-up and those who did not (Supplementary Table 12). Furthermore, in the worst-case scenario (assuming all patients lost to follow-up were dead at the last contact) and the best-case scenario (assuming all patients lost to follow-up were alive at the cutoff), the 5-year OS ranking between the non-PCC and PCC groups and among the three PCC subtypes remained unchanged. For the comparison between non-PCC and PCC groups, PCC demonstrated better OS than non-PCC in the worst-case scenario (P = 0.0270), whereas the difference was not significant in the best-case scenario (P = 0.7594). For the comparison among three subgroups, in the best-case scenario, SRC demonstrated significantly better OS than both SRC/PCC-NOS (P = 0.0164) and PCC-NOS (P = 0.0333) groups; in the worst-case scenario, SRC remained significantly superior to SRC/PCC-NOS (P = 0.0150) but not the PCC-NOS group (P = 0.1171) (Supplementary Figures 7-9, Supplementary Tables 13-15). These findings support the reliability of the primary survival analysis.

Prognostic factors

Univariable and multivariable analyses were performed to identify independent prognostic factors. In univariable analysis, the PCC subgroup was significantly associated with prognosis. Specifically, the non-SRC subgroups were associated with poorer survival [SRC vs SRC/PCC-NOS: HR: 2.21, 95%CI: 1.15-4.24, P = 0.0173; SRC vs PCC-NOS: HR: 2.09, 95%CI: 1.07-4.11, P = 0.0315]. However, after age, sex, and TNM stage were taken into account, no significant survival differences remained (SRC vs SRC/PCC-NOS: HR: 1.34, 95%CI: 0.693-2.60, P = 0.3831; SRC vs PCC-NOS: HR: 1.19, 95%CI: 0.60-2.37, P = 0.6145).

In univariable analysis, the following were prognostic factors: Age (≥ 60 years: HR: 1.53, 95%CI: 1.03-2.28, P = 0.0345); T3 and T4 stages (T3 vs T1: HR: 32.6, 95%CI: 3.924-270.8, P = 0.0013; T4 vs T1: HR: 38.75, 95%CI: 5.393-278.4, P < 0.001); N1, N2, and N3 stages (N1 vs N0: HR: 3.99, 95%CI: 1.62-9.82, P = 0.0026; N2 vs N0: HR: 5.88, 95%CI: 2.68-12.92, P < 0.001; N3 vs N0: HR: 10.85, 95%CI: 5.37-21.93, P < 0.001); and M stage (M1 vs M0: HR: 3.33, 95%CI: 1.81-6.11, P < 0.001). Multivariable analysis showed that only T stage (T3 vs T1: HR: 13.44, 95%CI: 1.508-119.73, P = 0.0199; T4 vs T1: HR: 14.41, 95%CI: 1.864-111.30, P = 0.0106) and N stage (N3 vs N0: HR: 3.87, 95%CI: 1.730-8.64, P < 0.001) were independent prognostic factors, whereas sex, age and M stage were not significantly associated with OS (Figure 3, Supplementary Table 16).

Figure 3 Univariable and multivariable hazard ratios of defined subgroups within poorly cohesive carcinoma. HR: Hazard ratio; CI: Confidence interval; SRC: Signet ring cell carcinoma; PCC-NOS: Poorly cohesive carcinoma-not otherwise specified.

DISCUSSION

In this study, we demonstrated that PCC is not a prognostically uniform entity and provided a refined characterization of PCC heterogeneity by classifying it based on SRC proportion. Consistent with our prespecified hypothesis, the SRC subtype demonstrated superior OS among the three subtypes, while stage emerged as the predominant prognostic factor in multivariable analysis. Using the European Consensus subclassification, the SRC subgroup initially conferred a survival advantage. However, this difference was confounded by clinicopathological factors and disappeared after PSM. Furthermore, the PCC-NOS group exhibited potentially distinct molecular features in MMR status and PD-L1 expression, although these findings require confirmation in larger cohorts. These observations suggest that integrating histological quantification with molecular profiling may warrant further investigation for risk stratification. Below, we compare our results with the prior literature and explore their clinical implications.

The PCC subclassification criteria used in this study were based on the European Consensus, currently the most widely recognized standard, and on growing evidence that the SRC proportion is inversely correlated with tumor aggressiveness. In the European cohort reported by Bencivenga et al[10], the proportions of patients with SRC ≥ 90%, 10% < SRC < 90% and SRC ≤ 10% were 18.50%, 56.65% and 24.86%, respectively. In the Japanese cohort reported by Nakamura et al[11], the corresponding proportions were 27.67%, 40.25%, and 32.08%, which were comparable to those at our center. This consistency across geographically distinct populations supports the reliability of the European Consensus-based classification. Notably, the Japanese cohort had a lower proportion of T1 tumors and a higher proportion of T3 tumors, which may reflect differences in pathological staging practices or screening intensity rather than true biological variation. In addition, the intergroup differences in sex distribution and tumor location observed in our cohort were not statistically significant in the Japanese cohort.

Regarding immunohistochemical features, we found that Ki67, PMS2 and PD-L1 expression differed significantly among the three subgroups. Ki67 is closely associated with cell proliferation. Low Ki67 expression in the SRC group and high expression in the PCC-NOS group were consistent with the observed differences in tumor stage and survival. Low PD-L1 expression in the SRC group may suggest limited benefit from PD-L1-based immunotherapy, although this finding requires validation in larger cohorts. MLH1, MSH2, MSH6, and PMS2 are established markers of MMR status. In our cohort, PMS2 and MLH1 expression differed significantly among the subgroups, and all cases with dMMR were confined to the PCC-NOS group, suggesting a possible enrichment of dMMR cases in this subtype; however, this observation should be interpreted with caution due to the small sample size. This finding is consistent with previous reports by Kim et al[19] that microsatellite instability-high tumors are less common in SRC than in other histological types; in line with this, Hirotsu et al[20] found all SRC cases in their cohort to be proficient MMR. The pattern may reflect an association between dMMR and PCC-NOS subtype. HER2 has been reported to be lowly expressed in PCC[21]. In our study, HER2 expression did not differ significantly among PCC subgroups. Mutual exclusivity between HER2 and dMMR has been reported in gastric cancer[22]. Consistent with this finding, all 4 dMMR cases in our cohort were negative for HER2 expression. However, given the limited number of evaluable cases, these observations are descriptive and exploratory, precluding formal survival comparisons or combined stratification.

The survival analysis further highlighted the prognostic heterogeneity among the three PCC subtypes. The multicenter retrospective study by Bencivenga et al[10] indicated that tumors with SRC ≥ 90% had superior 5-year OS, whereas the single-center retrospective study by Nakamura et al[11] found that the SRC-dominant and mixed groups had similar OS and recurrence-free survival, both significantly better than those of the group with PCC-dominant. Our findings are largely consistent with these observations, although, notably, we did not observe a significant difference between the SRC/PCC-NOS and PCC-NOS groups.

Alternative cutoff strategies have also been explored. Li et al[12] applied a 50% threshold within the 10%-90% range, showing a significant difference in pairwise comparison among four subgroups. Ultimately, they combined groups above and below the 50% boundary to confirm that the higher-proportion SRC group showed better survival. However, such classification criterion may lack generalizability. European cohorts reported that none of the cases had an SRC proportion exceeding 30%[14,15]. Subsequent regrouping using cutoffs of 10% and 20% still showed that SRC ≤ 10% had the worst prognosis, and redivided them into groups with SRC < 10%, 11% < SRC < 20%, and SRC ≥ 20%. These findings reinforce our approach of using the widely accepted thresholds of 10% and 90%, which balance applicability across diverse populations with prognostic discrimination.

In the overall cohort, patients lost to follow-up had more advanced disease, indicating informative censoring that may overestimate survival. This was also reflected in the worst-case scenario, where assumed death events resulted in poorer survival. This imbalance was not observed within the PCC group, supporting the robustness of the analysis among PCC subtypes. Sensitivity analyses under extreme censoring assumptions confirmed the survival advantage of SRC histology, whereas the PCC vs non-PCC comparison was sensitive to these assumptions and should be interpreted with caution.

The number of metastatic lymph nodes in the SRC group was lower than that in the other two subtypes, and this pattern was consistently observed in both the D1, D2 and D3 compartments. This finding should be interpreted with caution because the SRC group had a higher proportion of early-stage disease, which may partly explain the lower lymph node burden. Nevertheless, even when the analysis was limited to N3 patients, the SRC group still had fewer metastatic lymph nodes per case than the other two subtypes, suggesting that the lymphatic spread pattern of SRC tumors may not be explained solely by stage. Given the upper bounds of lymph node numbers in N1 and N2, differences in nodal burden are unlikely to be discernible within these strata. This observation might reflect a difference in lymphatic metastasis among subtypes with higher lymph node burden. The biological basis of this difference warrants further investigation.

In addition to classification based on SRC proportion, the presence of minor histological components in PCC may also have clinical implications. PCC may contain small amounts of other histological components that are insufficient for classification as a mixed type. Researchers have demonstrated that the presence of tubular components in SRC cancer is a risk factor for lymphatic metastasis in early submucosal PCC. Therefore, pathology reports should document the proportion of tubular components to guide subsequent treatment and follow-up[16]. A multicenter study in China reached a similar conclusion, showing that mixed tubular/papillary components in PCC are a risk factor for lymphatic metastasis[17]. Consistent with these observations, PCC with minor tubular/papillary components in our cohort also had a worse prognosis than the other groups or subgroups.

At the genomic level, PCC also exhibits considerable diversity. Several studies have reported frequent gene mutations in PCC, particularly involving TP53, CDH1, RHOA, and FGFR2[6,13,23-29]. However, the specific mutational landscape of PCC subtypes remains incompletely understood, particularly with respect to dMMR and PD-L1 status[30,31]. Further genomic studies are warranted to clarify the molecular drivers underlying the distinct clinicopathological profiles observed in our study.

Our study had several limitations. First, this was a single-center study; therefore, patient selection may have been biased and may not fully represent the general patient population. Second, several methodological limitations should be acknowledged: (1) Several immunohistochemical markers, including MMR proteins, PD-1, and PD-L1, were not routinely tested, resulting in limited evaluable sample sizes; all related findings should therefore be considered exploratory and descriptive; (2) The loss to follow-up rate was 23.6%, and patients lost to follow-up had more advanced disease, although worst-case scenario yielded a significant difference, likely reflecting differential follow-up intensity rather than a genuine survival difference; and (3) Stage IV patients were included because of heterogeneous surgical intent, but a sensitivity analysis excluding these patients yielded consistent results. Third, differences in postoperative treatment may have affected prognosis, but these were not analyzed, as many patients returned to local hospitals for adjuvant therapy. Fourth, follow-up data were insufficient for analysis of disease-free survival. Finally, we excluded patients who received preoperative adjuvant therapy to ensure consistent pathological staging.

CONCLUSION

In this retrospective cohort study, patients with PCC overall had survival comparable to that of patients with non-PCC gastric cancer. However, significant heterogeneity existed among the three PCC subtypes. The SRC subtype was characterized by younger age, female predominance, earlier stage at diagnosis, lower lymph node burden, and better OS, although this survival advantage was not independent of stage and measured covariates. The distinct clinicopathological profiles likely reflect underlying differences in tumor biology among the PCC subtypes. Large-scale, multicenter, prospective studies are warranted to further validate these findings and refine PCC subclassification.