Nomogram-based prognostic stratification for resectable gastric signet-ring cell carcinoma and adenocarcinoma: A retrospective cohort study

Abstract

BACKGROUND

Gastric signet-ring cell carcinoma (GSRCC) is a more aggressive subtype of gastric cancer compared to gastric adenocarcinoma (GA), with an increasing incidence. However, the prognostic differences between these subtypes, particularly in resectable cases, remain unclear.

AIM

To evaluate prognostic factors and develop a predictive model for GA and GSRCC patients undergoing curative resection.

METHODS

This retrospective cohort study included patients with GA and GSRCC who underwent curative surgery at the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences, from 2011 to 2018. Propensity score matching (PSM) (1:1) balanced the baseline characteristics. Prognostic factors were identified using univariate and multivariate Cox and least absolute shrinkage and selection operator (LASSO) regression analyses. Model performance was evaluated through calibration curves, decision curve analysis (DCA), and time-dependent receiver operating characteristic curves. Subgroup analysis and Kaplan-Meier survival curves were generated.

RESULTS

In a cohort of 3027 patients, the GSRCC group was characterized by a significantly higher prevalence of individuals under 60 years of age, females, cases with poor differentiation, and early-stage (stage I) disease (all P < 0.001). After PSM, the baseline was balanced and 761 patients were retained in each group. Variables identified through univariate Cox regression were included in the LASSO regression analysis. Multivariate Cox regression analysis identified age, tumor differentiation, tumor size, vascular invasion, and post-treatment nodal margin staging as independent prognostic factors. Subgroup analysis indicated a notably poorer prognosis for GSRCC in patients aged 60 and above (hazard ratio = 1.36, P = 0.025). The nomogram (C-index = 0.755) exhibited greater predictive accuracy than tumor node metastasis (TNM) staging for 1-, 3-, and 5-year overall survival (all P < 0.001), and provided a higher clinical net benefit according to DCA.

CONCLUSION

This study systematically compared resectable GA and GSRCC, revealing no overall survival difference. However, GSRCC demonstrated a significantly elevated mortality risk in subgroups stratified by age and tumor size. Multivariate analysis identified age, differentiation, tumor size, vascular invasion, and TNM stage as independent prognostic factors. The nomogram integrates clinicopathological features for precise risk stratification, surpassing traditional TNM staging.

Key Words: Gastric cancer; Adenocarcinoma; Signet-ring cell carcinoma; Nomogram; Prognostic model; Retrospective cohort study

Core Tip: This large-scale, retrospective cohort study compared prognostic outcomes of gastric adenocarcinoma and gastric signet-ring cell carcinoma (GSRCC) and revealed heterogeneous survival risks. While no overall survival difference was observed, GSRCC demonstrated significantly worse prognosis in age ≥ 60 years, size ≥ 4 cm, and poor differentiation. A new nomogram incorporating seven clinicopathological variables demonstrated greater accuracy than traditional tumor node metastasis staging in predicting 1-, 3-, and 5-year overall survival (P < 0.001). Key innovations include the application of least absolute shrinkage and selection operator regression to address multicollinearity and propensity score matching to minimize baseline bias. Integration of the model into clinical workflows could enable real-time risk stratification and guide personalized therapy for resectable gastric cancer.

INTRODUCTION

Gastric cancer remains one of the leading causes of cancer-related mortality worldwide. By 2022, recent studies indicate 968000 new cases and 659000 deaths worldwide, placing it fifth in both incidence and mortality[1]. Despite progress in diagnostic and therapeutic methods, the 5-year survival rate for patients with gastric cancer remains under 30%[2], largely because of its profound histological heterogeneity. According to the 2019 World Health Organization classification, gastric cancers are categorized into adenocarcinomas and non-adenocarcinomas[3]. This study focuses on gastric adenocarcinoma (GA) and gastric signet-ring cell carcinoma (GSRCC), a subtype of poorly cohesive carcinoma. GA, the most common subtype (60%-70% of cases), exhibits prognosis closely associated with tumor differentiation, molecular subtypes, staging, and treatment modalities[4]. In contrast, GSRCC, a distinct subtype characterized by intracellular mucin accumulation and signet-ring morphology, accounts for 25%-45% of newly diagnosed gastric cancers, with an increasing incidence[5,6]. Emerging evidence highlights the unique molecular features of GSRCC, such as higher mutation rates in CDH1 and ERBB2 than in other subtypes[7], which may disrupt cell adhesion and activate epithelial-mesenchymal transition, thereby promoting metastasis[8].

However, prognostic comparisons of GA and GSRCC remain controversial. While some studies suggest that GSRCC has a worse overall prognosis than GA, others report comparable or even superior outcomes for early stage GSRCC, and a subset of studies denies its role as an independent prognostic factor[9-12]. These discrepancies highlight the need to reassess the independent prognostic value of histological subtypes by controlling for confounders, such as tumor stage and treatment strategies. Current studies comparing GA and GSRCC face three major limitations: (1) Sample bias: Most single-center studies are underpowered (GSRCC cohorts often < 400 cases) and lack stratified analyses of prognostic differences across subgroups[13-15]; (2) Inadequate control for confounders: Variations in treatment protocols (e.g., surgical extent, adjuvant chemotherapy) may obscure the true impact of histological subtypes; (3) Limited generalizability of prognostic models: Recent studies highlight the limitations of tumor node metastasis (TNM) staging in GSRCC. Chen et al[16] developed a computed tomography-based radiomics nomogram for preoperative GSRCC diagnosis, while Puccini et al[17] identified CDH1/ERBB2 mutations as potential therapeutic targets. Existing TNM staging systems are not optimized for histological subtypes, leading to imprecise risk stratification for GSRCC patients. These limitations hinder individualized clinical decision making.

To address these gaps, we aimed to clarify the prognostic differences and independent risk factors between GA and GSRCC in patients undergoing curative resection using a large-scale retrospective cohort analysis. We developed a prognostic model integrating clinicopathological and histological features, and validated its performance against traditional staging systems, providing critical evidence for personalized therapeutic strategies for gastric cancer.

MATERIALS AND METHODS

Study design and population

This high-volume retrospective cohort study aimed to compare the prognostic differences between GA and GSRCC and to develop a predictive model. The study included patients who received surgical treatment for gastric cancer at the Cancer Hospital, Chinese Academy of Medical Sciences, from January 2011 to June 2018. The inclusion criteria were: (1) Age 18 years or older; (2) Postoperative pathological confirmation of GA or GSRCC by two independent pathologists; (3) All patients underwent curative resection (R0) with no evidence of distant metastasis; and (4) At least one postoperative follow-up completed, lasting 6 months or more. The exclusion criteria included: (1) Gastric stump cancer or concurrent malignancies; (2) Mixed histology, such as neuroendocrine carcinoma; (3) Evidence of distant metastasis (M1) from preoperative imaging or intraoperative confirmation; (4) Previous neoadjuvant chemotherapy or radiotherapy; (5) Insufficient lymph node dissection with fewer than 16 nodes examined or non-R0 resection; and (6) Incomplete clinical data, including missing key pathological parameters or follow-up records. The patient screening process is illustrated in Figure 1. The Institutional Review Board of the Cancer Hospital, Chinese Academy of Medical Sciences (No. 14-067/857) granted ethical approval, waiving the need for informed consent.

Figure 1 Flow diagram showing the study design and population. NCC: National Cancer Cener; GA: Gastric adenocarcinoma; GSRCC: Gastric signet ring cell carcinoma; PSM: Propensity score matching.

Data collection and variable definitions

Data were obtained from hospital electronic medical records and pathology databases. Baseline characteristics comprised age, sex, body mass index (BMI), smoking history, alcohol use, and comorbidities such as hypertension, cardiovascular disease, and diabetes. Tumor characteristics included histological type, tumor location (upper: Cardia/fundus; middle: Body; lower: Antrum/pylorus; multifocal/whole stomach), size (maximum diameter), differentiation, perineural invasion, vascular invasion, and TNM stage [8^th edition American Joint Committee on Cancer (AJCC) criteria]. The treatment details included the number of lymph nodes dissected and postoperative adjuvant chemotherapy. Overall survival (OS) refers to the time period extending from the point of diagnosis until death from any cause or until the latest follow-up assessment. Patients were categorized based on age (under 60 vs 60 and above), BMI (less than 24 vs 24 and higher), and tumor size (smaller than 4 cm vs 4 cm and larger).

Follow-up

During the initial two years, patients underwent follow-up assessments every three months. Subsequently, follow-ups occurred every six months for the next three years, followed by annual evaluations thereafter. Instances of loss to follow-up were treated as censored data. The median follow-up was 50 months.

Statistical analysis

Categorical variables were displayed as counts and percentages, and analyzed using χ² tests or Fisher’s exact tests. To minimize baseline bias between the GA and GSRCC groups, 1:1 nearest neighbor matching with a caliper width of 0.05 was conducted to balance covariates such as sex, age, tumor location, size, differentiation, and post-treatment nodal margin (pTNM) stage[18]. Standardized mean differences < 0.1 indicated adequate balance. After propensity score matching (PSM), univariate and multivariate Cox proportional hazards models were employed to determine prognostic factors. Least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation identified optimal predictors (λminimum). A nomogram was created to predict 1-, 3-, and 5-year OS using independent prognostic factors identified through multivariate Cox analysis. Model performance was assessed through calibration curves with 1000 bootstrap resamples, time-dependent receiver operating characteristic (ROC) curves comparing the nomogram to TNM staging, and decision curve analysis (DCA) for evaluating clinical utility. Univariate Cox regression and interaction tests were used to evaluate prognostic differences between GA and GSRCC across subgroups, and Kaplan-Meier curves and log-rank tests were applied for survival comparisons. A P value threshold of 0.05 was used to determine statistical significance. All analyses were performed using the R software (version 4.3.1).

RESULTS

Baseline characteristics and PSM

A total of 3027 patients were enrolled (GA: 2151; GSRCC: 886) after applying the inclusion and exclusion criteria. Prior to matching, the GSRCC group had a significantly higher percentage of patients under 60 years old (60.95% vs 43.65%, P < 0.001), female patients (34.99% vs 20.18%, P < 0.001), tumors located in the lower stomach (50% vs 39.1%, P < 0.001), poorly differentiated tumors (80.93% vs 36.45%, P < 0.001), and tumors smaller than 4 cm (49.1% vs 42.03%, P < 0.001). Patients with GSRCC exhibited a greater incidence of stage I disease (35.78% compared to 29.38%, P < 0.001). Following 1:1 PSM, each group comprised 761 patients, with no significant differences in baseline variables (P > 0.05; Table 1), indicating balanced intergroup characteristics.

Table 1 Patients and tumors’ clinical and pathological characteristics before and after propensity score matching, n (%).

Variable	All patients		P value	Matched patients		P value
	GA (n = 2151)	GSRCC (n = 886)		GA (n = 761)	GSRCC (n = 761)
Age (year)			< 0.001			0.382
< 60	939 (43.65)	540 (60.95)		404 (53.09)	422 (55.45)
≥ 60	1212 (56.35)	346 (39.05)		357 (46.91)	339 (44.55)
Gender			< 0.001			0.510
Male	1717 (79.82)	576 (65.01)		524 (68.86)	511 (67.15)
Female	434 (20.18)	310 (34.99)		237 (31.14)	250 (32.85)
BMI (kg/m2)			0.347			0.644
< 24	1099 (51.09)	470 (53.05)		407 (53.48)	397 (52.17)
≥ 24	1052 (48.91)	416 (46.95)		354 (46.52)	364 (47.83)
Comorbidity (yes)	906 (42.12)	341 (38.49)	0.070	332 (43.63)	306 (40.21)	0.194
Smoking history (yes)	1099 (51.09)	369 (41.65)	< 0.001	342 (44.94)	322 (42.31)	0.326
Drinking history (yes)	1010 (46.95)	383 (43.23)	0.067	334 (43.89)	335 (44.02)	1.000
Tumor location			< 0.001			0.449
Upper	859 (39.93)	171 (19.30)		190 (24.97)	171 (22.47)
Middle	411 (19.11)	231 (26.07)		183 (24.05)	208 (27.33)
Lower	841 (39.10)	443 (50.00)		357 (46.91)	351 (46.12)
More than two position or total	40 (1.86)	41 (4.63)		31 (4.07)	31 (4.07)
Differentiation			< 0.001			0.932
Poor	784 (36.45)	717 (80.93)		598 (78.58)	592 (77.79)
Moderate	1228 (57.09)	167 (18.85)		161 (21.16)	167 (21.94)
Well	139 (6.46)	2 (0.23)		2 (0.26)	2 (0.26)
Size (cm)			< 0.001			0.196
1 < 4	904 (42.03)	435 (49.10)		319 (41.92)	345 (45.34)
2 ≥ 4	1247 (57.97)	451 (50.90)		442 (58.08)	416 (54.66)
Perineural invasion (yes)	1039 (48.30)	424 (47.86)	0.854	418 (54.93)	412 (54.14)	0.797
Vascular invasion (yes)	867 (40.31)	328 (37.02)	0.100	335 (44.02)	315 (41.39)	0.325
pT stage			< 0.001			0.955
T1	510 (23.71)	303 (34.20)		198 (26.02)	198 (26.02)
T2	309 (14.37)	107 (12.08)		99 (13.01)	100 (13.14)
T3	728 (33.84)	182 (20.54)		173 (22.73)	181 (23.78)
T4a/4b	604 (28.08)	294 (33.18)		291 (38.24)	282 (37.06)
pN stage			< 0.001			0.723
N0	848 (39.42)	377 (42.55)		264 (34.69)	275 (36.14)
N1	367 (17.06)	100 (11.29)		86 (11.30)	96 (12.61)
N2	375 (17.43)	145 (16.37)		144 (18.92)	136 (17.87)
N3	561 (26.08)	264 (29.80)		267 (35.09)	254 (33.38)
pTNM stage			< 0.001			0.115
I	632 (29.38)	317 (35.78)		219 (28.78)	213 (27.99)
II	558 (25.94)	175 (19.75)		135 (17.74)	167 (21.94)
III	961 (44.68)	394 (44.47)		407 (53.48)	381 (50.07)
Adjuvant chemotherapy (yes)	747 (34.73)	304 (34.31)	0.859	310 (40.74)	291 (38.24)	0.345

BMI: Body mass index; GA: Gastric adenocarcinoma; GSRCC: Gastric signet ring cell carcinoma; pTNM: Post-treatment nodal margin.

Prognostic factor screening and multivariate model construction

Univariate Cox regression analysis revealed nine variables significantly associated with OS (P < 0.1): Subtype, age, tumor location, size, poor differentiation, perineural invasion, vascular invasion, pTNM stage, and adjuvant chemotherapy (Table 2). To mitigate collinearity, LASSO regression with 10-fold cross-validation (optimal λminimum) identified seven crucial variables: Subtype, age, size, poor differentiation, vascular invasion, pTNM stage, and adjuvant chemotherapy (Figure 2 and Table 3). Multivariate Cox analysis identified age ≥ 60 years, poor differentiation, tumor size 4 cm, vascular invasion, and pTNM stage as independent prognostic factors. Patients in stage II [hazard ratio (HR) = 2.50, 95% confidence interval (CI): 1.499-4.164, P < 0.001] and stage III (HR = 8.00, 95%CI: 5.065-12.648, P < 0.001) exhibited 2.5- and 8-fold increased mortality risks compared to those in stage I, respectively (Table 2).

Figure 2 Variable selection via least absolute shrinkage and selection operator Cox regression. A: Coefficient profiles of clinicopathological variables in the least absolute shrinkage and selection operator model. The upper x-axis indicates the number of non-zero coefficients, while the y-axis represents coefficient magnitudes. Nine variables were evaluated, including subtype, age, tumor location, tumor size, poor differentiation, perineural invasion, vascular invasion, post-treatment nodal margin stage, and adjuvant chemotherapy; B: Optimal lambda (λ) selection using 10-fold cross-validation (minimum criterion). The vertical dashed lines correspond to λminimum (left, minimizing cross-validated error) and λ1se (right, within one standard error of the minimum). The upper x-axis denotes retained variables after shrinkage. pTNM: Post-treatment nodal margin.

Table 2 Univariate and multivariate analysis of overall survival after propensity score matching.

Variable	Univariate analyses			Multivariate analysis
	HR	95%CI	P value	HR	95%CI	P value
Subtype
GA	Ref.			Ref.
GSRCC	1.601	0.973-2.635	0.064	1.112	0.916-1.349	0.282
Age (year)
< 60	Ref.			Ref.
≥ 60	1.391	1.148-1.685	0.001	1.330	1.094-1.617	0.004
Gender
Male	Ref.
Female	0.94	0.764-1.156	0.556
BMI (kg/m2)
< 24	Ref.
≥ 24	0.715	0.733-1.078	0.232
Smoking history
Yes	Ref.
No	0.938	0.733-1.137	0.514
Drinking history
Yes	Ref.
No	1.039	0.856-1.261	0.698
Tumor location
Upper	Ref.
Middle	0.559	0.427-0.732	< 0.001
Lower	0.585	0.464-0.737	< 0.001
More than two position or total	1.964	1.363-2.83	< 0.001
Poor differentiation
Yes	Ref.			Ref.
No	0.628	0.482-0.818	0.001	0.766	0.584-0.999	0.049
Size (cm)
< 4	Ref.			Ref.
≥ 4	2.714	2.172-3.390	< 0.001	1.376	1.088-1.739	0.008
Perineural invasion
Yes	Ref.
No	0.441	0.358-0.542	< 0.001
Vascular invasion
Yes	Ref.			Ref.
No	0.402	0.330-0.489	< 0.001	0.744	0.606-0.915	0.005
pTNM stage
I	Ref.		< 0.001	Ref.
II	2.805	1.701-4.628		2.498	1.499-4.164	< 0.001
III	10.631	6.97-16.215		8.004	5.065-12.648	< 0.001
Adjuvant chemotherapy
Yes	Ref.			Ref.
No	1.169	1.022-1.338	0.023	1.123	0.922-1.368	0.250

BMI: Body mass index; GA: Gastric adenocarcinoma; GSRCC: Gastric signet ring cell carcinoma; pTNM: Post-treatment nodal margin; HR: Hazard ratio; CI: Confidence interval.

Table 3 Coefficients for least absolute shrinkage and selection operator survival regression with lambda values.

Variable	Lambda
Subtype	0.0609
Age	0.2521
Poor differentiation	-0.2036
Size	0.2898
Vascular invasion	-0.2725
pTNM stage	0.9118
Adjuvant chemotherapy	0.0756
Perineural invasion	0
Tumor location	0

pTNM: Post-treatment nodal margin.

Subgroup analysis and interaction effects

Although no significant survival difference was observed between GSRCC and GA in the overall cohort (HR = 1.112, 95%CI: 0.916-1.349, P = 0.282), subgroup interaction analysis revealed three significant interaction factors (P for interaction < 0.05): Age, tumor size, and differentiation (visualized in Figure 3). Compared with GA subgroup, GSRCC was associated with a poorer prognosis in patients aged 60 and above (HR = 1.36, 95%CI: 1.04-1.78, P = 0.025) and in those with tumors measuring 4 cm or larger (HR = 1.28, 95%CI: 1.02-1.60, P = 0.033). For patients < 60 years, GSRCC exhibited a non-significant survival advantage (HR = 0.80, 95%CI: 0.61-1.06, P = 0.126). In the poorly differentiated subgroup, GSRCC was associated with a trend towards worse outcomes (HR = 1.18, 95%CI: 0.96-1.46, P = 0.124).

Figure 3 Forest plot of hazard ratios from univariate Cox subgroup analysis. Horizontal lines represent 95% confidence intervals. Diamonds indicate pooled hazard ratios for significant interaction subgroups. GA: Gastric adenocarcinoma; GSRCC: Gastric signet ring cell carcinoma; BMI: Body mass index; pTNM: Post-treatment nodal margin; HR: Hazard ratio; CI: Confidence interval.

Nomogram development and risk stratification

A nomogram using seven variables selected by LASSO was created to predict 1-, 3-, and 5-year OS, achieving a C-index of 0.755 (Figure 4A). Patients were stratified into low-, intermediate-, and high-risk groups based on nomogram scores. Kaplan-Meier curves showed significantly better OS in the low-risk group than in the intermediate- and high-risk groups (P < 0.001; Figure 4B).

Figure 4 Nomogram development and risk stratification. A: The nomogram of predicting overall survival in patients. Survival nomogram for the prediction of 1-year, 3-year, and 5-year overall survival; B: Kaplan-Meier curve to test the stratification system of the nomogram. GA: Gastric adenocarcinoma; GSRCC: Gastric signet ring cell carcinoma; pTNM: Post-treatment nodal margin.

Model validation

Calibration curves demonstrated strong concordance between the nomogram-predicted and actual 5-year OS rates (Figure 5). Time-dependent ROC analysis demonstrated that the nomogram significantly surpassed the 8^th edition AJCC TNM staging system in predicting OS across all time points: 1-year area under the curve (AUC) of 0.796 compared to 0.709 (P < 0.001), 3-year AUC of 0.775 vs 0.722 (P < 0.001), and 5-year AUC of 0.790 against 0.736 (P < 0.001) (Figure 6). DCA demonstrated that the nomogram provides a greater net clinical benefit than TNM staging for predicting 1-, 3-, and 5-year OS (Figure 7).

Figure 5 Calibration plot of the multivariate Cox regression model for 5-year overall survival. The solid line represents the model’s predicted probabilities, while the dashed diagonal line indicates ideal calibration. Bootstrap-derived 95% confidence intervals (shaded area) demonstrate robust agreement between predicted and observed outcomes.

Figure 6 Time-dependent receiver operating characteristic curve analysis comparing the nomogram and tumor node metastasis staging systems for predicting overall survival. A: 1-year; B: 3-year; C: 5-year. The nomogram achieved significantly higher area under the curve values at all timepoints (P < 0.001), underscoring its clinical utility for survival prediction. OS: Overall survival; AJCC: American Joint Committee on Cancer.

Figure 7 Decision curve analysis comparing the nomogram and tumor node metastasis staging systems for overall survival. A: 1-year; B: 3-year; C: 5-year. The nomogram provided greater net benefit than tumor node metastasis staging across clinically relevant threshold probabilities (10%-90%), validating its utility for guiding therapeutic decisions. OS: Overall survival; AJCC: American Joint Committee on Cancer.

DISCUSSION

GSRCC is a heterogeneous malignancy with distinct prognostic risks compared to other gastric cancer subtypes, such as GA, influenced by various factors. Previous studies have remained controversial, and there is no definitive consensus on its prognostic implications. Moreover, effective therapies for recurrent or metastatic GSRCC are lacking, underscoring the critical need for high-performance prognostic models for the early identification of high-risk patients. Utilizing a comprehensive retrospective cohort from the National Cancer Center, we developed a nomogram that incorporates clinicopathological factors such as histological subtype, age, tumor size, differentiation, vascular invasion, pTNM stage, and adjuvant chemotherapy for resectable GA and GSRCC. This model integrates readily available clinical variables for quick risk evaluation and exhibits strong predictive capability, achieving a C-index above 0.7. It significantly outperforms the 8^th edition AJCC TNM staging system in forecasting 1-, 3-, and 5-year OS (P < 0.001).

The clinical profiles of the patients with GSRCC and GA differed markedly. Baseline data revealed substantial pre-PSM imbalances, with GSRCC patients being younger, predominantly female, and having higher rates of poorly differentiated tumors, which is consistent with a meta-analysis reporting a lower male prevalence and younger age in GSRCC cohorts[19]. Emerging evidence links estrogen receptor β (ERβ) to GSRCC progression, suggesting ERβ-mediated inhibition of proliferation via the mammalian target of rapamycin-ARPC1B/EVL pathway[20,21]. Theuer et al[22] observed a higher incidence of GSRCC in younger patients (< 40 years, 28% vs 15%, P < 0.001), which was potentially linked to CDH1 germline variants[23]. Such baseline imbalances (e.g., higher smoking rates in GA vs GSRCC, 51.09% vs 41.65%, P < 0.001) could profoundly bias prognostic comparisons, necessitating rigorous PSM to mitigate the confounding effects.

While univariate Cox regression suggested a non-significant survival disadvantage for GSRCC (P = 0.064), multivariate analysis revealed no independent prognostic impact of the histological subtype, a finding echoing the contentious literature. Some studies have reported worse outcomes for GSRCC[24-26], while others have challenged its prognostic relevance, emphasizing its inherent heterogeneity. Notably, our subgroup analysis identified a significantly worse prognosis for GSRCC in older patients (≥ 60 years), larger tumors (≥ 4 cm), and poorly differentiated subgroups, likely due to its aggressive biological behavior. GSRCC is associated with higher rates of peritoneal/lymph node metastasis[27], and its limited sensitivity to conventional chemotherapy (e.g., fluorouracil/platinum regimens)[24] underscores the need for novel strategies, such as hyperthermic intraperitoneal chemotherapy or immunotherapy[28,29]. Intriguingly, no survival difference was observed between GA and GSRCC in the early or advanced stages, consistent with meta-analyses suggesting comparable outcomes in early GSRCC[30] and studies refuting its prognostic impact[31]. Our cohort, which was restricted to curatively resected patients with adequate lymphadenectomy (≥ 16 nodes), ensured accurate TNM staging, as supported by Yang et al[32].

Methodologically, this study innovatively applied LASSO regression to address high-dimensional variable collinearity, which is a common pitfall in gastric cancer prognostic research. The LASSO identified seven core predictors, circumventing the overfitting risks inherent in stepwise regression. Although Jiang et al[33] and Wu et al[34] employed LASSO for immune signature selection and TNM staging optimization, respectively, our study is the first to apply this approach for GA/GSRCC comparison. The resulting nomogram (C-index = 0.755) outperformed the AJCC TNM staging system (C-index = 0.702), corroborating recent efforts emphasizing histological subtype integration[35] and imaging-based GSRCC diagnostics[16].

The study had several limitations. First, its single-center retrospective design may have introduced a selection bias. Although PSM and LASSO regression enhanced internal validity, the conclusions require validation in multicenter cohorts, particularly in Western populations, given the documented racial disparities in GSRCC incidence, which may affect model generalizability[36]. Second, the prognostic assessment was based exclusively on OS, potentially introducing confounding factors from non-cancer-related mortality, as it did not consider cancer-specific survival or disease-free survival. However, OS has been prioritized because of its objectivity and feasibility in retrospective studies. Third, the lack of detailed treatment data (e.g., specific chemotherapy regimens, targeted therapies, or immunotherapies) limits the comprehensive assessment of treatment-related prognostic factors. Despite these limitations, the core contribution of this study lies in elucidating the subgroup-specific prognostic vulnerability of GSRCC and providing a clinically translatable risk-stratification tool. In the future, we recommend prioritizing intensified therapies for high-risk patients with GSRCC (e.g., advanced age and large tumors) and integrating the nomogram into electronic medical record systems to enable real-time prognostic assessment. Additionally, clinical trials targeting SRCC-specific molecular markers should be conducted to improve the overall prognosis of patients with GC.

CONCLUSION

This extensive retrospective cohort study systematically evaluated prognostic differences between GA and GSRCC to create a prognostic model incorporating clinicopathological features. While no OS difference was observed between GA and GSRCC, GSRCC exhibited significantly elevated mortality risk in patients ≥ 60 years, with tumors ≥ 4 cm, or poorly differentiated subgroups. The Cox regression-based nomogram, incorporating seven variables-subtype, age, size, poor differentiation, vascular invasion, pTNM stage, and adjuvant chemotherapy status exhibited greater accuracy than conventional TNM staging, offering a practical tool for personalized risk stratification in resectable gastric cancer.