Construction of competing-risk nomograms and identification of optimal candidates for aggressive therapy in gastric cancer with peritoneal metastasis: a population-based study

Introduction

Gastric cancer (GC) remains a major global health burden, ranking as the fifth most common malignancy and the third leading cause of cancer-related mortality worldwide, accounting for approximately 800,000 deaths annually (1). Projections indicate that by 2040, the number of new cases will rise to 1.8 million and related deaths to 1.3 million, representing increases of 63% and 66%, respectively, compared with 2020 (2). Among the various metastatic patterns, the peritoneum is one of the most common distant metastatic sites, with 14–43% of patients presenting with peritoneal metastasis (PM) at initial diagnosis (3). PM progresses rapidly and severely, is often accompanied by severe complications such as ascites and bowel obstruction, severely impairs quality of life, and represents one of the major causes of mortality in advanced disease (4). Studies have shown that the median survival of patients with PM is only about 4 months, compared with approximately 14 months in those without PM (5). Despite recent advances in systemic therapies, the median survival of this population remains less than one year, which emphasizes the need to implement more effective management strategies (6).

Historically, the value of aggressive local treatments (such as palliative gastrectomy or metastasectomy) for patients with gastric cancer with peritoneal metastasis (GCPM) has been highly controversial, largely stemming from the landmark phase III REGATTA trial. This pivotal study demonstrated that cytoreductive gastrectomy combined with chemotherapy did not improve survival compared with chemotherapy alone in metastatic GC (7). However, the introduction of the “oligometastasis” concept and accumulating evidence have increasingly challenged this conclusion. Recent studies (e.g., the AIO-FLOT3 trial) suggest that, among selected patients with limited metastatic burden and favorable response to chemotherapy, conversion surgery or enhanced local control may lead to substantial survival benefits (8,9). Thus, the current clinical challenge is no longer whether aggressive treatment should be performed, but rather how to accurately identify patients who are likely to benefit.

Effective identification of heterogeneity among GCPM patients requires more precise prognostic assessment tools. While the current American Joint Committee on Cancer (AJCC) tumor-node-metastasis (TNM) staging system has some guiding significance, it is primarily based on anatomical factors and struggles to further differentiate the risk profile of individuals in the M1 stage (10,11). Moreover, most existing prognostic models for GC rely on traditional Cox proportional hazards regression, which treats non-cancer-related deaths as censored events and often overestimates cancer-specific mortality, particularly in elderly populations (12). Additionally, variable selection strategies based solely on univariate significance are vulnerable to multicollinearity, leading to reduced model stability. Fine-Gray competing-risk models combined with rigorous variable-selection approaches such as least absolute shrinkage and selection operator (LASSO) and Akaike information criterion (AIC) offer more strong prognostic evaluation, yet, their application in GCPM patients remains relatively limited.

In this study, we performed a systematic analysis of a large cohort of GCPM patients derived from the Surveillance, Epidemiology, and End Results (SEER) database. We constructed independent prognostic models utilizing a dual screening strategy (LASSO and stepwise AIC) and developed nomograms based on a competing risk approach to more accurately predict overall survival (OS) and cancer-specific survival (CSS). Furthermore, we established a novel risk-stratification system to evaluate the potential survival benefits of aggressive therapy across different risk groups. This study aims to provide a practical and clinically translatable tool to aid in more precise prognostic prediction and to inform individualized treatment decision-making, enabling identification of patients who may benefit from aggressive interventions while avoiding ineffective and potentially harmful surgical treatments for high-risk individuals. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2692/rc).

Methods

Data sources and patient selection

Managed by the National Cancer Institute, the SEER database serves as a key resource for cancer statistics, encompassing around 26.5% of the U.S. population across various racial, gender, age, and geographic demographics (13). This dataset is derived from multiple regional cancer registries and includes comprehensive information on demographics, clinicopathological characteristics, and treatment modalities. Data for patients diagnosed with GCPM between 2010 and 2015 were extracted from SEER*Stat 9.0.42 {SEER Research Data, 17 Registries, Nov 2024 Sub [2000–2022]}. The filter criteria were as follows: (I) primary site labeled as C16.0–C16.9; (II) year of diagnosis between 2010 and 2015; (III) CS Mets at dx (2010–2015) =40. The inclusion criteria were as follows: (I) patients diagnosed with GCPM between 2010 and 2015; (II) cases with a single primary tumor; and (III) patients with positive histology (the diagnosis was confirmed by microscopic examination of tissue samples). The exclusion criteria were as follows: (I) age <18 years; (II) race unknown; (III) tumor size was 0 or could not be assessed; (IV) missing or unknown surgical information; (V) rural-urban continuum code unknown, missing, or unmatched; (VI) survival time less than 1 month or unknown; (VII) unknown distant metastasis status; and (VIII) unknown AJCC T or N category. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study did not require ethical approval or informed consent due to the absence of interventional procedures or identifiable patient data. The criteria for selecting GCPM patients are depicted in Figure 1.

Figure 1 Flow chart of GCPM patients’ inclusion and exclusion criteria in the SEER database. AJCC, American Joint Committee on Cancer; GCPM, gastric cancer with peritoneal metastasis; N, node; SEER, Surveillance, Epidemiology, and End Results; T, tumor.

Study variables

This study extracted multidimensional patient information from the SEER database, including demographic characteristics (age at diagnosis, sex, race, marital status, median household income, and area of residence), tumor characteristics (tumor size, grade, primary site, histological type, other distant metastases, and AJCC 7th edition T or N stages), and treatment-related variables (surgery, chemotherapy, and radiotherapy), along with survival time and vital status. Residential areas were classified according to the United States Department of Agriculture (USDA) Economic Research Service (ERS) rural-urban continuum codes as large metropolitan areas, other metropolitan areas, and nonmetropolitan areas. Median household income was categorized into four quartiles: <$60,000, $60,000–79,999, $80,000–99,999, and ≥$100,000. Primary sites were classified anatomically as cardia, fundus, body, antrum, pylorus, lesser curvature, greater curvature, overlapping lesion, and stomach, not otherwise specified (NOS). Histological types were grouped according to the World Health Organization (WHO) digestive system tumor classification into adenocarcinoma, signet-ring cell carcinoma, and others. Distant metastases were defined as involvement of bone, liver, brain, or lung beyond the peritoneum. “No” indicated isolated PM without other extra-peritoneal involvement. The number of distant metastatic sites recorded in SEER was recategorized as no extraperitoneal metastasis, single, double, or multiple. Treatment strategies were classified based on SEER primary site surgery codes (Rx Summ-Surg Prim Site 1998+) and chemotherapy records into four categories: (I) supportive care: no surgery and no chemotherapy were received; (II) chemotherapy-based: chemotherapy was received only; (III) local control: surgery was received only, mainly referring to palliative resection; (IV) aggressive therapy: surgery combined with chemotherapy. Specifically, “aggressive therapy” was defined as having received any surgical procedure on the primary gastric tumor (SEER codes 10–90) in combination with systemic chemotherapy. This categorization aims to capture the clinical strategy of intensifying local tumor control in addition to systemic therapy, representing an intensified multimodal treatment approach. Notably, radiotherapy was not included in the treatment strategy classification. According to National Comprehensive Cancer Network (NCCN) and European Society for Medical Oncology (ESMO) guidelines (11,14), in advanced GCPM patients, radiotherapy is primarily used for palliative purposes (e.g., to control bleeding or obstruction) rather than curative intent. Therefore, it was included as a covariate to avoid indication bias (15). The primary endpoints of the study were OS and CSS. OS was defined as the interval from diagnosis to death from any cause, whereas CSS was defined as the interval from diagnosis to death attributable to GC, treating non-cancer deaths as competing events.

Statistical analysis

All statistical analyses were performed using R software (version 4.5.2). Continuous variables were summarized as mean ± standard deviation (SD) or median with interquartile range (IQR) and compared using the t-test or Mann-Whitney U test. Categorical variables were presented as counts (percentages) and compared using the chi-square test or Fisher’s exact test. The Optimal cut-off values for tumor size were determined using X-tile software (version 3.6.1). Patients were randomly split into training and validation sets (7:3 ratio).

To minimize selection bias, propensity score matching (PSM) was applied to balance baseline characteristics across different treatment groups using a 1:1 nearest-neighbor matching algorithm with a caliper width of 0.2. Covariate balance was assessed using standardized mean differences (SMDs), with SMD <0.1 indicating adequate balance.

To ensure model robustness and minimize overfitting, a dual-variable screening strategy was applied in the training set. Candidate variables with non-zero coefficients were first identified using 10-fold cross-validated LASSO regression, followed by backward stepwise multivariate Cox regression based on AIC to determine independent prognostic factors. Based on these factors, nomograms for OS and CSS were constructed using Cox proportional hazards and Fine-Gray competing-risk models (treating non-cancer death as a competing event), respectively. Multicollinearity was assessed using the variance inflation factor (VIF). A VIF value of <5 was considered to indicate no significant multicollinearity.

Nomogram performance was evaluated using the concordance index (C-index), time-dependent receiver operating characteristic (ROC) curves, and calibration plots (1,000 bootstrap resamples). Decision curve analysis (DCA) was conducted to quantify the net clinical benefit and to compare the clinical utility of the nomogram versus the AJCC staging system. Finally, a risk-stratification system was established based on total nomogram scores (tertiles), and treatment benefits were analyzed within each risk group using stratified Cox regression. Survival differences among the different risk groups were assessed using Kaplan-Meier analysis and the log-rank test.

All statistical tests were two-sided, and P<0.05 was considered statistically significant.

Results

Patient characteristics

This study included 4,280 patients with PM of GC. Table 1 summarizes the baseline characteristics of the entire population. The median age of the cohort was 63 years (IQR, 53–73 years), with a predominance of males (62.0%). Most patients were White (71.0%), married (57.0%), and diagnosed with poorly differentiated tumors (grade III: 54%). Regarding treatment strategies, patients were categorized into four groups: chemotherapy-based (n=2,405, 56%), supportive care (n=1,217, 28%), aggressive therapy (n=416, 10%), and local control (n=242, 6%). Significant differences existed among the treatment groups in terms of demographic characteristics, tumor biological behavior, and treatment patterns (all P<0.001). Notably, patients receiving aggressive therapy were younger (median age: 57 years) compared with the supportive care group (median age: 68 years), reflecting a greater tendency in clinical practice to adopt aggressive intervention strategies for younger patients. Regarding tumor characteristics, 49.0% had single distant metastases, 66.0% had adenocarcinoma histology, and 41.0% had undetermined T stage (TX). AJCC T and N stages were unevenly distributed across groups, with a higher proportion of T4 stage patients in the local control group (57.0%). SMD analysis further quantified these differences, with age imbalance being the most significant (SMD =0.752), while the SMD values for other variables were mostly greater than 0.1, suggesting selection bias between groups. PSM was subsequently applied to balance baseline characteristics across different treatment groups. For prognostic model development and validation, the entire cohort was randomly divided into a training set (2,596 patients, 61.8%) and a validation set (1,224 patients, 38.2%). As shown in Table 2, the training set and validation set showed good balance across all baseline features (P>0.05), indicating that the two groups were comparable in terms of basic patient characteristics, thus ensuring the reliability of the subsequent prediction model establishment and validation process.

PSM analysis

Given significant baseline differences among treatment groups, we performed 1:1 PSM to balance covariates between the aggressive therapy cohort (n=273) and the chemotherapy-based cohort (n=273). As shown in Figure 2A and detailed in Table 3, all covariate distributions, including age, tumor size, and AJCC stage, were well balanced after matching, with absolute SMD strictly below 0.1. In the matched cohort, aggressive therapy significantly improved survival outcomes. Kaplan-Meier curves demonstrated that OS was significantly better in the aggressive therapy group compared with the chemotherapy-based group (Figure 2B, P<0.001). Consistent with the OS results, cumulative incidence function curves for CSS also indicated that aggressive therapy was associated with a lower risk of cancer-specific death (Figure 2C, P<0.001). These findings suggest that, in selected patients with PM, combining surgery or other local control measures with chemotherapy provides a survival benefit exceeding that of chemotherapy alone.

Figure 2 Evaluation of treatment efficacy after PSM. (A) Love plot of SMDs before (red) and after (blue) matching. The vertical line at SMD =0.1 indicates adequate balance. (B) Kaplan-Meier curves for OS comparing aggressive treatment vs. chemotherapy. (C) CIF curves for CSS accounting for competing risks. CIF, cumulative incidence function; CSS, cancer-specific survival; OS, overall survival; PSM, propensity score matching; SMD, standardized mean difference.

Subgroup analyses of treatment effects

To evaluate whether the benefits of aggressive therapy were consistent across different populations, we compared OS hazard ratios (HRs) and CSS subdistribution hazard ratios (SHRs) for chemotherapy-based versus aggressive therapy within key clinicopathologic subgroups. In most subgroups, aggressive therapy was associated with better OS and CSS (Figure 3, all HR >1, P<0.05). For example, in grade III patients, chemotherapy alone was associated with a 1.84-fold increase in overall mortality risk [HR =1.84; 95% confidence interval (CI): 1.60–2.12; P<0.001]. However, notable heterogeneity was observed regarding the extent of distant metastases. Patients with a single distant metastatic site derived significant benefit from aggressive therapy (OS HR =2.17; P<0.001), whereas in patients with two metastatic sites, survival differences between the two treatment strategies were not significant (OS HR =0.98, 95% CI: 0.46–2.08, P=0.96; CSS SHR =1.32, 95% CI: 0.60–2.92, P=0.49). This finding indicates that although aggressive therapy is generally beneficial, its advantage may diminish in patients with a higher metastatic burden.

Figure 3 Subgroup analyses comparing aggressive therapy vs. chemotherapy-based. The aggressive therapy group serves as the reference. HR or SHR >1 indicates a higher risk of death with chemotherapy-based (favoring aggressive therapy). The x-axis is plotted on a log scale. AJCC, American Joint Committee on Cancer; CI, confidence interval; CSS, cancer-specific survival; HR, hazard ratio; N, node; OS, overall survival; SHR, subdistribution hazard ratio; T, tumor.

Variable selection and independent prognostic factors

To systematically identify satisfactory prognostic factors and reduce multicollinearity, we implemented a rigorous statistical workflow. Initial univariate analyses were conducted to assess the individual association of each clinicopathologic characteristic with survival outcomes, demonstrating that the majority of baseline variables were significantly associated with OS or CSS (Tables 3,4, left column). Candidate variables were then selected via LASSO regression (OS: Figure 4A; CSS: Figure 4B), followed by backward stepwise selection based on the AIC to obtain the optimal multivariate model. The VIF values for all predictors in the final model were <1.3, ruling out significant multicollinearity.

Figure 4 Variable selection using LASSO. (A) LASSO coefficient trajectories and 10-fold cross-validation curve for OS. (B) Corresponding plots for CSS. CSS, cancer-specific survival; LASSO, least absolute shrinkage and selection operator; OS, overall survival.

For OS (Table 4), the final model included age, race, marital status, grade, tumor size, household income, histology, primary site, AJCC T and N stages, distant metastases, and treatment strategy. Advanced age (HR =1.01), grade IV (HR =1.95; 95% CI: 1.36–2.81; P<0.001), N3 stage (HR =1.23; P=0.04), and multiple metastases (HR =1.85; 95% CI: 1.29–2.65; P<0.001) were significantly associated with worse prognosis. In contrast, higher household income (≥$100,000) was a protective factor (HR =0.78; P<0.001). Importantly, treatment strategy remained the strongest independent predictor after adjusting for these 11 covariates. Compared with supportive care, aggressive therapy markedly reduced mortality risk (HR =0.21; 95% CI: 0.18–0.25; P<0.001), outperforming local control (HR =0.38) and chemotherapy-based regimens (HR =0.37).

For CSS (Table 5), the final model incorporated the same 12 variables. High malignancy grade (grade IV vs. grade I: SHR =1.94; P<0.001) and significant metastatic burden (multiple metastases: SHR =1.64; P=0.03) were strongly associated with increased cancer-specific mortality. Consistent with OS findings, treatment strategy remained a strong independent predictor. Aggressive treatment showed the lowest subdistributed HR compared to supportive care (SHR =0.29; 95% CI: 0.25–0.35; P<0.001), further confirming its survival advantage after adjustment for confounders.

Construction and validation of nomograms

Based on independent prognostic factors identified through multivariate analysis, nomograms predicting 6, 12, and 24 months OS and CSS were constructed for GCPM patients (Figure 5A,5B). Each variable was assigned an integral value based on its regression coefficient, and the total score, obtained by summing all integrals, was mapped to the bottom scale to estimate the individual's survival probability. The length of each variable’s scale reflected its relative contribution to prognosis. The nomograms revealed that the treatment strategy exhibited the widest score range (approximately 30 to 100 points) in both the OS and CSS nomograms. Patients receiving aggressive therapy were assigned the lowest risk scores (31 points for OS; 33 points for CSS), while those receiving supportive care were assigned the highest risk score of 100 points. This highlights the dominant role of aggressive therapy in improving survival outcomes, outweighing the influence of many intrinsic tumor characteristics. Other factors, including higher tumor grade (grade IV) and extensive metastases, also substantially increased risk scores.

Figure 5 Nomograms for predicting 6, 12, and 24 months survival in patients with GCPM. (A) OS nomogram. (B) CSS nomogram. *, P<0.05; **, P<0.01; ***, P<0.001. AJCC, American Joint Committee on Cancer; CSS, cancer-specific survival; GCPM, gastric cancer with peritoneal metastasis; N, node; OS, overall survival; T, tumor.

Nomogram performance was validated in terms of discrimination, calibration, and clinical utility. Discrimination was assessed via C-index and ROC analysis. The C-index for the OS nomogram was 0.709 (95% CI: 0.698–0.719) in the training set and 0.683 (95% CI: 0.666–0.701) in the validation set. The C-index for CSS was 0.707 (95% CI: 0.695–0.718) in the training set and 0.679 (95% CI: 0.661–0.698) in the validation set. The time-dependent ROC curves further confirmed these findings. In the training set, areas under the curve (AUCs) for OS at 6, 12, and 24 months were 0.770 (95% CI: 0.753–0.786), 0.763 (95% CI: 0.746–0.781), and 0.792 (95% CI: 0.769–0.815), demonstrating satisfactory discriminative ability (Figure 6). Similar AUCs were observed for CSS and in the validation cohort, indicating model stability across datasets. The calibration curves for OS (Figure 7) and CSS (Figure 8) exhibited excellent agreement between predicted and observed survival probabilities, with the 6, 12, and 24 months curves closely aligned with the 45-degree ideal line. DCA indicated that the nomograms offered a potential for higher net benefit than the traditional AJCC staging system across a wide range of threshold probabilities (Figures 9,10).

Figure 6 Time-dependent ROC curves for 6, 12, and 24 months OS and CSS in the (A,C) training and (B,D) validation cohorts. CI, confidence interval; CSS, cancer-specific survival; OS, overall survival; ROC, receiver operating characteristic.

Figure 7 Calibration curves for OS at 6, 12, and 24 months in the (A-C) training and (D-F) validation cohorts. The x-axis represents the survival probability predicted by the nomogram, while the y-axis represents the actual observed survival probability. The gray dashed diagonal line represents the ideal prediction model. The red and blue solid lines represent the performance of our nomogram. A curve that closely aligns with the gray diagonal line indicates excellent calibration, whereas deviations above or below the diagonal suggest underestimation or overestimation of survival risk, respectively. OS, overall survival.

Figure 8 Calibration curves for CSS at 6, 12, and 24 months in the (A-C) training and (D-F) validation cohorts. The x-axis represents the survival probability predicted by the nomogram, while the y-axis represents the actual observed survival probability. The gray dashed diagonal line represents the ideal prediction model. The red and blue solid lines represent the performance of our nomogram. A curve that closely aligns with the gray diagonal line indicates excellent calibration, whereas deviations above or below the diagonal suggest underestimation or overestimation of survival risk, respectively. CSS, cancer-specific survival.

Figure 9 DCA curves for OS at 6, 12, and 24 months in the (A-C) training and (D-F) validation cohorts. The x-axis represents the threshold probability, while the y-axis represents the net benefit. The red solid line represents the performance of our proposed nomogram. The blue dashed line and green long-dashed line represent the AJCC T stage and N stage, respectively. The gray and black solid lines represent the assumptions of treating all patients (“Treat All”) and treating no patients (“Treat None”), respectively. A curve that positions higher across a wider range of threshold probabilities indicates superior clinical net benefit compared to other strategies. AJCC, American Joint Committee on Cancer; DCA, decision curve analysis; N, node; OS, overall survival; T, tumor.

Figure 10 DCA curves for CSS at 6, 12, and 24 months in the (A-C) training and (D-F) validation cohorts. The x-axis represents the threshold probability, while the y-axis represents the net benefit. The red solid line represents the performance of our proposed nomogram. The blue dashed line and green long-dashed line represent the AJCC T stage and N stage, respectively. The gray and black solid lines represent the assumptions of treating all patients (“Treat All”) and treating no patients (“Treat None”), respectively. A curve that positions higher across a wider range of threshold probabilities indicates superior clinical net benefit compared to other strategies. AJCC, American Joint Committee on Cancer; DCA, decision curve analysis; N, node; OS, overall survival; T, tumor.

Risk stratification system

To enhance the clinical operability of the models, we constructed a risk stratification system based on the total nomogram score, using ternary digits as the cutoff value to divide patients into low-risk, intermediate-risk, and high-risk groups. Kaplan-Meier curves for OS demonstrated clear separation among the three groups in both the training (Figure 11A, P<0.001) and validation (Figure 11B, P<0.001) cohorts. Low-risk patients exhibited favorable survival, whereas high-risk patients had poor outcomes. This discriminatory ability was also consistently observed in CSS analyses (Figure 11C,11D), confirming the robustness of the risk stratification system in differentiating patient prognoses.

Figure 11 Risk stratification based on nomogram-derived total scores. (A,B) OS Kaplan-Meier curves. (C,D) CSS CIF curves. CIF, cumulative incidence function; CSS, cancer-specific survival; OS, overall survival.

Risk-stratified treatment decision guidance

To facilitate clinical decision-making and explore the biological basis of treatment heterogeneity, clinical characteristics across risk groups were compared (Figure 12). The low-risk group consisted primarily of younger patients (<65 years old) with early disease characteristics (grade I/II, T1/T2, N0/N1) and single distant metastasis. In contrast, the high-risk group was dominated by advanced age (≥65 years old), high malignancy grade (grade III/IV), and extensive tumor burden (T4, N3, multiple metastases). This significant difference in baseline characteristics provided a basis for developing differentiated treatment strategies.

Figure 12 Clinicopathological characteristics of patients across low, medium, and high-risk groups. N, node; T, tumor.

Building on this, we further explored the relative effects of treatment strategies within different risk strata to provide preliminary clinical insights. Kaplan-Meier curves (Figure 13) and forest plots (Figure 14) illustrate the heterogeneity of treatment response, with chemotherapy-based treatment as the reference (HR =1). Survival analysis in the low-risk group revealed significant differences among treatment modalities (Figure 13A, P<0.001). The forest plots (Figure 14, left panel) confirmed that aggressive therapy (HR =0.6; P<0.001) and local control (HR =0.6; P<0.001) both conferred significant survival benefits, whereas supportive care was associated with poorer prognosis (P=0.03). In the medium-risk group, no significant differences were observed among treatments (Figure 13B, P=0.24), with forest plots showing that neither aggressive therapy (P=0.30) nor local control (P=0.08) provided an additional survival benefit. In the high-risk group, OS differed significantly among groups (Figure 13C, P<0.001), but forest plots revealed that aggressive therapy was associated with significantly increased the risk of death (HR >1; P<0.001). Supportive care also showed poor outcomes (P<0.001), whereas local control did not significantly differ from chemotherapy (P=0.67). These results suggest that for patients with extensive tumor burden and limited physiological reserve, the surgical burden of aggressive therapy may outweigh any potential oncologic benefit.

Figure 13 Kaplan-Meier survival curves for treatment strategies stratified by (A-C) risk groups.

Figure 14 Forest plots showing HRs for OS comparing supportive care, local control, and aggressive therapy with chemotherapy-based treatment within each risk category. CI, confidence interval; HR, hazard ratio; OS, overall survival.

In summary, our models support a risk-adapted treatment strategy: low-risk patients are ideal candidates for aggressive therapy and can achieve maximal survival benefit; medium-risk patients may not benefit from treatment escalation beyond standard chemotherapy; high-risk patients may be harmed by aggressive therapy, and chemotherapy-based or palliative treatment should be prioritized to avoid overtreatment.

Discussion

In this large population-based study utilizing the SEER database, we developed and validated competing-risk nomogram models to predict OS and CSS in GCPM patients. To our knowledge, this is among the first studies to integrate a dual variable selection strategy (LASSO regression combined with stepwise AIC) with Fine-Gray competing-risk modeling to construct a prognostic tool for this specific metastatic population. The models demonstrated improved discrimination, calibration, and potential clinical net benefit compared with the traditional AJCC staging system. However, it is important to note that predictive superiority does not automatically equate to clinical endorsement, which requires prospective validation. Furthermore, we established a practical risk-stratification system that provides a novel decision-making framework for metastatic GC, a setting where treatment strategies have long been controversial. The study indicated that aggressive treatment was associated with a clear survival benefit only in low-risk patients, whereas it was linked to a potentially increased risk of mortality in high-risk patients, highlighting the need for more refined, risk-adapted treatment strategies.

From a methodological perspective, accurate prognostic assessment serves as the cornerstone of precision medicine. Conventional Cox proportional hazards models often overstate cancer-specific mortality when competing events, such as cardiovascular death, are present, especially in elderly populations. The application of Fine-Gray competing-risk models yielded more excellent CSS estimates. Furthermore, unlike previous studies that typically selected variables based on univariate significance—an approach susceptible to multicollinearity and redundancy—our study employed a combined strategy utilizing LASSO regularization and stepwise AIC. This methodology harmonized model simplicity with predictive efficacy, preserving essential biological and socioeconomic attributes in the final 12-variable model (16).

The models included demographic, socioeconomic, pathological, and therapeutic factors, identifying multiple independent prognostic variables. Age became a significant risk factor. The majority of patients with peritoneal metastases are old, and the risk of mortality escalates markedly with age, in accordance with previous study findings (17,18). These findings may pertain to the biological characteristics of peritoneal metastases, which are challenging to identify in the early stages (19), as well as the existence of comorbidities and compromised immune function in elderly patients, hence diminishing treatment tolerance (20). As a result, aged patients exhibit diminished capacity to endure surgery or chemotherapy, leading to increased mortality rates (21). Previous investigations consistently identified tumor grade as an independent risk factor, indicative of tumor aggressiveness and prognosis (22,23). Poorly differentiated tumors exhibit higher malignancy, characterized by increased invasiveness and metastatic potential, leading to inferior prognoses (24). This observation is consistent with our findings. The prognostic significance of tumor size in GC remains controversial. Numerous studies indicate that bigger tumors correlate with elevated tumor burden, increased invasiveness, and a heightened chance of distant metastases, resulting in a worse prognosis (25,26). Our investigation noted a diminishing risk correlated with larger tumor size. However, given the high rate of missing data for this variable in SEER, this finding may reflect selection bias (e.g., measurable tumors are more likely to be resected) rather than a true biological phenomenon. Therefore, tumor size in this specific metastatic setting may not be a reliable independent prognosticator compared to other significant factors like tumor grade or metastatic burden. Our nomogram shows a certain association between median household income and survival, but this indicator is a regional-level ecological socioeconomic status (SES) variable and does not represent the true socioeconomic status at the individual level (27). The observed protective effect may partially reflect differences in regional healthcare resource availability or health service utilization, and the specific mechanisms still require verification with more detailed individual-level SES data. In addition, broader health-system and logistic influences may also influence access to care and outcomes, such as organizational disruptions due to coronavirus disease 2019 (COVID-19), which can significantly affect surgical candidates, perioperative morbidity, and multidisciplinary assessment (28). Signet ring cell carcinoma (SRC) did not confer a significant survival disadvantage in our multivariate analysis, a finding that diverges from traditional views associating SRC with poor prognosis. This discrepancy may be due to the limitations of the SEER database in the inability to quantify the proportion of SRC on histological classification. Emerging evidence suggests that the prognostic impact of SRC is driven by the proportion of signet-ring cells (e.g., >10%) rather than the qualitative WHO label alone (29). Since SEER codes histology based on the dominant pattern without specifying the percentage of SRCs, our “SRC” category likely represents a heterogeneous group, potentially diluting the negative prognostic signal observed in cohorts with pure or high-burden SRC histology. Studies have indicated that the place of residence has a potential impact on cancer survival. Specifically, rural patients typically experience poorer survival outcomes compared with their urban counterparts (30). In our study, however, place of residence was not retained in the final model; this exclusion is likely attributable to the underrepresentation of rural patients (9%), which limited statistical power. Radiotherapy was not identified as an independent prognostic factor, consistent with prior reports (17,31). This is likely because radiotherapy, as a locoregional modality, offers limited control over diffuse peritoneal dissemination (32). This supports our decision not to include radiotherapy as a treatment-strategy variable. Additionally, patients with multiple distant metastases had worse outcomes than those with a single metastasis, consistent with the concept of oligometastasis and supporting the notion that metastatic burden exists on a continuum rather than as a binary state (33).

The most clinically relevant finding of this study was the risk-dependent effect of treatment modality. In low-risk patients, aggressive therapy (surgery or local control combined with chemotherapy) was associated with a significant survival benefit (HR =0.6). This observed benefit likely reflects distinct biological behaviors of peritoneal metastases. Peritoneal dissemination exists on a biological spectrum, encompassing superficially disseminated, chemotherapy-responsive deposits and deeply infiltrative, treatment-refractory progression (34). It is plausible that the ‘low-risk’ profile defined by our nomogram (characterized by younger age, lower tumor grade, and limited metastatic burden) corresponds to a disease phenotype that is more localized and biologically less aggressive. In such patients, aggressive local cytoreduction may effectively reduce tumor burden to a level where it can be controlled by systemic therapy, thereby translating the surgical effort into a survival advantage (35). This is consistent with the findings of related studies in which cytoreductive integrity (CC-0/CC-1) and low Peritoneal Cancer Index (PCI) were key prognostic factors for surgical benefit (36). Conversely, in high-risk patients, aggressive therapy was associated with a worse prognosis (HR >1; P<0.01). It is critical to emphasize that this observational finding does not establish a causal, biologically detrimental effect of the therapy itself. Instead, the observed “harmful” association may reflect a confluence of factors inherent to this vulnerable population, including greater physiological frailty and comorbid burden not fully captured in SEER, a higher risk of perioperative morbidity and mortality, and a potentially reduced capacity to tolerate or derive benefit from multimodal treatment. As shown in the population profiles (Figure 12), the high-risk group primarily comprised older patients (≥65 years) with high-grade tumors (grade III/IV) and extensive metastatic burden (T4, N3, multiple metastases). For patients with poor physiological reserve and advanced-stage disease, surgical stress-induced immunosuppression, postoperative delays in systemic therapy, or complications leading to deteriorated quality of life may well negate or outweigh the marginal benefits of cytoreduction (37-39). Therefore, this association reveals that the net clinical benefit of increased surgical intervention in this subgroup is negative, rather than surgery directly accelerating disease progression. Patient selection, metastasis patterns, and intrinsic biological characteristics are key factors influencing treatment outcomes. This finding is crucial, as it reconciles discrepancies in prior trial results. Failures in earlier trials, such as REGATTA, may have been due to the inclusion of unselected high-risk patients. This reinforces the caution against prioritizing surgical intervention over systemic or palliative treatment in these patients. Thus, our risk stratification system serves as a valuable prognostic tool that identifies patients for whom aggressive therapy is strongly associated with benefit or harm, thereby highlighting subgroups for which this approach needs to be seriously considered or used with caution in clinical practice. Certainly, prospective validation in the context of clinical trials is required before any such tool can be used to dictate treatment escalation or de-escalation.

Several limitations of this study should be acknowledged. First, an intrinsic limitation of SEER-based analyses is the absence of specific peritoneal disease quantification, such as the PCI score and peritoneal lavage cytology status. Contemporary clinical decision-making regarding tumor resection depends on the extent of peritoneal distribution and the response to neoadjuvant therapy. Due to the lack of these detailed tumor burden indicators, the “low-risk” group identified in our study likely includes patients with smaller, localized peritoneal disease (low PCI score), while the “high-risk” group likely corresponds to patients with diffuse carcinomatosis. Future studies could incorporate these quantitative indicators for further analysis. Second, the retrospective design is inherently susceptible to selection bias. Although PSM and multivariable adjustment were employed to mitigate this concern, the modest size of the matched cohort for comparing treatment strategies warrants cautious interpretation of the estimates derived from this analysis. The PSM analysis should be interpreted primarily as a sensitivity analysis to validate an overall association. Third, SEER is deficient in comprehensive data regarding chemotherapy regimens (such as immunotherapy and targeted therapy), performance status [Eastern Cooperative Oncology Group (ECOG)], and comorbidities. Fourth, the absence of an external validation cohort restricts generalizability. Furthermore, a substantial proportion of patients (57%) had missing data on tumor size. Although we categorized ‘Unknown’ as a separate group to maximize sample size, this high rate of missingness may introduce unpredictable statistical artifacts. Consequently, the observed inverse association between tumor size and prognosis should be interpreted with caution and requires validation in datasets with more complete records. Lastly, the study variables lacked information on specific molecular biomarkers. Future research using multi-omics biomarkers [e.g., programmed death-ligand 1 (PD-L1), human epidermal growth factor receptor 2 (HER2) status] and dynamic predictive methodologies (e.g., landmark analysis) may clarify temporal patterns of treatment response and prognosis, thereby enhancing precision medicine for GC with peritoneal metastases.

Conclusions

We developed and validated a competing-risk nomogram to assess OS and CSS in GCPM patients. The model showed improved predictive performance and potential clinical net benefit with the use of a dual-variable selection technique (LASSO and stepwise AIC) relative to the traditional AJCC staging scheme. This model’s risk classification has shown significant variability in treatment efficacy and provides a foundation for a risk-adapted therapeutic approach, which could contribute to more personalized prognostic assessment and treatment discussions in clinical practice.

Acknowledgments

We are grateful to the SEER team for their valuable assistance in the collection of clinical data. The SEER database is publicly accessible, and we have obtained the necessary data usage protocol for this database.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2692/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2692/prf

Funding: This research was funded by the 2024 Medical Science and Technology Innovation Project of Xuzhou Health Commission (No. 2024103332).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2692/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Morgan E, Arnold M, Camargo MC, et al. The current and future incidence and mortality of gastric cancer in 185 countries, 2020-40: A population-based modelling study. EClinicalMedicine 2022;47:101404. [Crossref] [PubMed]
3. Szor DJ, Pereira MA, Ramos MFKP, et al. Peritoneal recurrence in gastric cancer after curative gastrectomy: risk factors and predictive score model. J Gastrointest Surg 2025;29:101850. [Crossref] [PubMed]
4. Shinkai M, Imano M, Kohda M, et al. Efficacy of palliative surgery for gastric cancer patients with peritoneal metastasis who still have residual peritoneal dissemination after chemotherapy. Langenbecks Arch Surg 2023;408:291. [Crossref] [PubMed]
5. Langellotti L, Fiorillo C, D'Annibale G, et al. Efficacy of Cytoreductive Surgery (CRS) + HIPEC in Gastric Cancer with Peritoneal Metastasis: Systematic Review and Meta-Analysis. Cancers (Basel) 2024;16:1929. [Crossref] [PubMed]
6. Guchelaar NAD, de Neijs MJ, Noordman BJ, et al. The prognostic value of peritoneal metastases in patients with gastric cancer: a nationwide population-based study. EClinicalMedicine 2025;81:103109. [Crossref] [PubMed]
7. Fujitani K, Yang HK, Mizusawa J, et al. Gastrectomy plus chemotherapy versus chemotherapy alone for advanced gastric cancer with a single non-curable factor (REGATTA): a phase 3, randomised controlled trial. Lancet Oncol 2016;17:309-18. [Crossref] [PubMed]
8. Yasufuku I, Tsuchiya H, Fujibayashi S, et al. Oligometastasis of Gastric Cancer: A Review. Cancers (Basel) 2024;16:673. [Crossref] [PubMed]
9. Kim HI, Badgwell BD. Peritoneal Oligometastasis in Gastric Cancer: Diagnostic Strategies, Patient Selection, and Emerging Therapeutic Approaches. J Gastric Cancer 2025;25:409-23. [Crossref] [PubMed]
10. Yu J, Yao R, Han N, et al. Evaluating the prognostic significance of tumor deposits in gastric cancer and strategies for their integration into the TNM staging system: a single-center retrospective study. Cell Oncol (Dordr) 2025;48:761-73. [Crossref] [PubMed]
11. Ajani JA, D'Amico TA, Bentrem DJ, et al. Gastric Cancer, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2022;20:167-92. [Crossref] [PubMed]
12. Su H, Xie S, Wang S, et al. New findings in prognostic factor assessment for adenocarcinoma of transverse colon: a comparison study between competing-risk and COX regression analysis. Front Med (Lausanne) 2024;11:1301487. [Crossref] [PubMed]
13. Che WQ, Li YJ, Tsang CK, et al. How to use the Surveillance, Epidemiology, and End Results (SEER) data: research design and methodology. Mil Med Res 2023;10:50. [Crossref] [PubMed]
14. Lordick F, Carneiro F, Cascinu S, et al. Gastric cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Ann Oncol 2022;33:1005-20. [Crossref] [PubMed]
15. Luo XF, Luo YH, Zhao XY, et al. Application and progress of palliative therapy in advanced gastric carcinomas. Front Oncol 2023;13:1104447. [Crossref] [PubMed]
16. Yang Q, Luo B, Yu C, et al. A Two-Step Variable Selection Strategy for Multiply Imputed Survival Data Using Penalized Cox Models. Bioengineering (Basel) 2025;12:1278. [Crossref] [PubMed]
17. Liu X, Ren Y, Wang F, et al. Development and validation of prognostic nomogram for patients with metastatic gastric adenocarcinoma based on the SEER database. Medicine (Baltimore) 2023;102:e33019. [Crossref] [PubMed]
18. Sun Y, Li Z, Tian Y, et al. Development and validation of nomograms for predicting overall survival and cancer-specific survival in elderly patients with locally advanced gastric cancer: a population-based study. BMC Gastroenterol 2023;23:117. [Crossref] [PubMed]
19. Ng D, Cyr D, Khan S, et al. Molecular mechanisms of metastatic peritoneal dissemination in gastric adenocarcinoma. Cancer Metastasis Rev 2025;44:50. [Crossref] [PubMed]
20. Chen W, Altshuler RD, Daschner P, et al. Older adults with cancer and common comorbidities-challenges and opportunities in improving their cancer treatment outcomes. J Natl Cancer Inst 2024;116:1730-8. [Crossref] [PubMed]
21. Bastiaannet E, Pilleron S. Epidemiology of cancer in older adults: a systematic review of age-related differences in solid malignancies treatment. Curr Oncol Rep 2025;27:290-311. [Crossref] [PubMed]
22. Liang Y, Sheng G, Guo Y, et al. Prognostic significance of grade of malignancy based on histopathological differentiation and Ki-67 in pancreatic ductal adenocarcinoma. Cancer Biol Med 2024;21:416-32. [Crossref] [PubMed]
23. ElKordy MA, Soliman RM, ElTohamy MI, et al. Predictors of peritoneal metastasis of gastric origin. J Egypt Natl Canc Inst 2022;34:53. [Crossref] [PubMed]
24. Sorrentino L, De Ruvo N, Serra F, et al. Role of poorly differentiated cluster in gastric cancer: is it a new prognosis factor? Scand J Gastroenterol 2022;57:44-9. [Crossref] [PubMed]
25. Wang Q, Shen K, Fei B, et al. Development and validation of a nomogram to predict cancer-specific survival of elderly patients with unresected gastric cancer who received chemotherapy. Sci Rep 2024;14:9008. [Crossref] [PubMed]
26. Zhou L, Li W, Cai S, et al. Large tumor size is a poor prognostic factor of gastric cancer with signet ring cell: Results from the surveillance, epidemiology, and end results database. Medicine (Baltimore) 2019;98:e17367. [Crossref] [PubMed]
27. Conley CC, Derry-Vick HM, Ahn J, et al. Relationship between area-level socioeconomic status and health-related quality of life among cancer survivors. JNCI Cancer Spectr 2024;8:pkad109. [Crossref] [PubMed]
28. Cavaliere D, Parini D, Marano L, et al. Surgical management of oncologic patient during and after the COVID-19 outbreak: practical recommendations from the Italian society of Surgical Oncology. Updates Surg 2021;73:321-9. [Crossref] [PubMed]
29. Marano L, Ambrosio MR, Resca L, et al. The Percentage of Signet Ring Cells Is Inversely Related to Aggressive Behavior and Poor Prognosis in Mixed-Type Gastric Cancer. Front Oncol 2022;12:897218. [Crossref] [PubMed]
30. Lewis-Thames MW, Langston ME, Khan S, et al. Racial and Ethnic Differences in Rural-Urban Trends in 5-Year Survival of Patients With Lung, Prostate, Breast, and Colorectal Cancers: 1975-2011 Surveillance, Epidemiology, and End Results (SEER). JAMA Netw Open 2022;5:e2212246. [Crossref] [PubMed]
31. Shi H, Yang H, Yan S, et al. Development and validation of nomograms based on the SEER database for the risk factors and prognosis of distant metastasis in gastric signet ring cell carcinoma. Medicine (Baltimore) 2024;103:e40382. [Crossref] [PubMed]
32. Xu P, Liu D, Zhou J, et al. Survival analysis of patients with metastatic head and neck squamous cell carcinoma treated with metastasis-directed radiotherapy and immunotherapy. Radiat Oncol 2025;20:31. [Crossref] [PubMed]
33. Radigan R, Kao CS, Krainock M, et al. Long-term survival and undetectable circulating tumor DNA following comprehensive involved site radiotherapy for oligometastases. Sci Rep 2025;15:6126. [Crossref] [PubMed]
34. Zhao JJ, Ong CJ, Srivastava S, et al. Spatially Resolved Niche and Tumor Microenvironmental Alterations in Gastric Cancer Peritoneal Metastases. Gastroenterology 2024;167:1384-1398.e4. [Crossref] [PubMed]
35. Chowdhury S, Aggarwal A, Goel S, et al. Gastric cancer with limited peritoneal metastasis: Role of cytoreductive surgery and hyperthermic intraperitoneal chemotherapy. Indian J Gastroenterol 2025;44:692-9. [Crossref] [PubMed]
36. Marano L, Marrelli D, Sammartino P, et al. Cytoreductive Surgery and Hyperthermic Intraperitoneal Chemotherapy for Gastric Cancer with Synchronous Peritoneal Metastases: Multicenter Study of 'Italian Peritoneal Surface Malignancies Oncoteam-S.I.C.O.'. Ann Surg Oncol 2021;28:9060-70. [Crossref] [PubMed]
37. Nakanishi K, Kanda M, Ito S, et al. Delay in initiation of postoperative adjuvant chemotherapy with S-1 monotherapy and prognosis for gastric cancer patients: analysis of a multi-institutional dataset. Gastric Cancer 2019;22:1215-25. [Crossref] [PubMed]
38. Tsujimoto H, Kouzu K, Sugasawa H, et al. Impact of postoperative infectious complications on adjuvant chemotherapy administration after gastrectomy for advanced gastric cancer. Jpn J Clin Oncol 2021;51:379-86. [Crossref] [PubMed]
39. Bezu L, Akçal Öksüz D, Bell M, et al. Perioperative Immunosuppressive Factors during Cancer Surgery: An Updated Review. Cancers (Basel) 2024;16:2304. [Crossref] [PubMed]