Prognostic stratification in gastric cancer with peritoneal metastasis: tumor biology and diagnostic modality as a surrogate for disease burden

Abstract

Background

Gastric adenocarcinoma with peritoneal metastasis (PM) has a poor prognosis, yet clinical outcomes vary significantly. This study aimed to identify independent prognostic determinants of PM-specific survival (PM-OS), focusing on tumor biology and disease burden surrogate.

Methods

We retrospectively analyzed 166 patients with gastric adenocarcinoma and PM treated at a single center between 2015 and 2024. Prognostic factors were evaluated using multivariable Cox proportional hazards models. To mitigate immortal time bias and confounding by indication, receipt of systemic therapy was excluded from the primary multivariable model.

Results

The median PM-OS was 8.2 months (95% CI, 6.63–9.79). Patients diagnosed via surgical exploration (radiologically occult) achieved a significantly longer median PM-OS compared to those diagnosed radiologically (13.6 vs. 7.4 months; p = 0.003). In multivariable analysis, HER2 positivity (HR0.312, 95% CI 0.164–0.593; p < 0.001) and surgical diagnosis (HR 0.555, 95% CI 0.338–0.913; p = 0.020), interpreted as a surrogate for radiologically occult/low tumor burden, were identified as independent predictors of improved survival. Additionally, an optimized CA 19 − 9 cutoff (> 175.4 U/mL), unlike the standard threshold, significantly stratified survival. Traditional factors, including signet-ring cell histology and age, did not retain independent significance.In sensitivity analysis including systemic therapy, treatment was strongly associated with survival and the prognostic significance of HER2 and diagnostic context remained.

Conclusion

Survival in gastric PM is fundamentally driven by HER2 status and the extent of PM. Surgical detection identifies a subgroup with limited, occult disease who achieve superior outcomes compared to those with radiologically overt metastases. Furthermore, the magnitude of biomarker elevation, rather than mere positivity, serves as a critical stratifier. These findings support a paradigm shift towards burden-based risk stratification to guide treatment intensity and clinical trial eligibility.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-15684-1.

Background

Gastric cancer remains one of the leading causes of cancer-related mortality worldwide [1]. The peritoneum is the most frequent site of dissemination in gastric adenocarcinoma, reported in approximately 15–30% of patients either at diagnosis or during the disease course [2]. This pattern reflects both the biological aggressiveness of diffuse-type tumors and the anatomical propensity of gastric cancer cells to spread through transcoelomic routes. Once peritoneal metastasis (PM) develops, prognosis is poor, and median overall survival is typically limited to a few months despite advances in systemic therapy [3].

Several clinicopathological factors have been associated with an increased risk of developing synchronous PM, including younger age, non-cardia tumor location, female sex, signet ring cell carcinoma, diffuse-type histology or linitis plastica, T4 stage, and the presence of multiple metastatic sites [4]. In addition, models incorporating tumor diameter, tumor type and location, clinical T stage, and serum tumor markers have shown promise for predicting PM in gastric cancer [5]. However, these efforts focus primarily on identifying patients at risk of peritoneal involvement rather than on characterizing prognosis once PM is established.

Understanding prognostic determinants in both de novo and recurrent PM is essential for refining risk stratification, anticipating clinical course, guiding systemic treatment choices, and designing future trials for this high-risk subgroup. To address these needs, we analyzed a single-center cohort of patients with gastric adenocarcinoma and PM managed in routine clinical practice. The study evaluates clinical, pathological, and biomarker characteristics; compares patterns of presentation and treatment between de novo and recurrent PM; and investigates determinants of survival. This evaluation aims to clarify prognostic factors relevant at the time of PM detection and to support more individualized management strategies in patients with peritoneal metastatic gastric cancer.

Methods

This retrospective cohort study included patients diagnosed with gastric adenocarcinoma with PM who were treated and followed at the Dokuz Eylül University Department of Medical Oncology between 2015 and 2024. Patients aged 18 years or older were eligible for inclusion. Clinical, demographic, pathological, and treatment-related data were retrieved from electronic medical records.

Eligibility criteria

Patients were included if they had histologically confirmed gastric cancer with either de novo or recurrent PM and if their clinical and survival data were accessible. For recurrent PM, patients were eligible only if more than six months had elapsed between completion of adjuvant therapy and the diagnosis of peritoneal recurrence. Patients who developed PM while still receiving palliative, perioperative or adjuvant treatment were excluded. Additional exclusion criteria were the presence of another synchronous primary malignancy, missing core variables, mixed tumor histology, or absence of any follow-up after the baseline date (Fig. 1).

Fig. 1.

Study flowchart depicting the patient selection process, exclusion criteria, and final study cohort

Definitions and variables

Clinical, demographic, and laboratory data were retrospectively extracted from institutional electronic medical records and the oncology database. The primary outcome was PM–specific overall survival (PM-OS), defined as the time from the date of definitive diagnosis of PM to death or last documented clinical visit. The date of PM diagnosis was established based on the first objective evidence: if PM was suspected on imaging and subsequently confirmed surgically, the date of surgical confirmation was used; if PM was not surgically evaluated, the index date was the date of the first imaging report documenting peritoneal metastasis.

Biomarker status was determined based on a retrospective review of original pathology reports from the time of initial diagnosis. No additional re-evaluation was performed. Results for human epidermal growth factor receptor 2 (HER2), programmed cell death ligand 1 (PD-L1), and microsatellite instability (MSI) were extracted directly from the pathology reports. HER2 expression followed the ASCO/CAP guidelines for gastric cancer, while PD-L1 was reported as a Combined Positive Score (CPS).

PM was defined by radiologic imaging, staging laparoscopy, or intraoperative identification during definitive or emergency surgery. Recurrent peritoneal metastasis was defined as the radiologic or surgical detection of PM following prior curative-intent therapy, whereas de novo PM referred to patients presenting with peritoneal involvement at the time of primary diagnosis. All surgically diagnosed PM cases, including those detected by staging laparoscopy, incidental intraoperative findings during definitive surgery, or emergency surgery, were pooled into a single ‘surgical diagnosis’ category for primary analyses. Modality-specific comparisons within surgical contexts were exploratory and descriptive.

Disease burden proxies were assessed through a review of formal radiology reports and operative notes. Macroscopic ascites was recorded as present if identified on baseline imaging or during surgical exploration. The extent of metastasis was categorized as isolated PM or PM with concomitant extra-peritoneal involvement. The number of metastatic sites was computed by considering the peritoneum as a single site, while each additional involved organ system or non-regional lymph node station was counted as an additional site.

Statistical analysis

Analyses were performed using SPSS version 29.0 (IBM Corp.) and R version 4.5 (R Foundation for Statistical Computing). Continuous variables were summarized as mean ± standard deviation if normally distributed and as median (minimum–maximum) otherwise. Categorical variables were summarized as frequencies and percentages. Group comparisons utilized Student’s t-test or the Mann–Whitney U test for continuous data, and the χ² or Fisher’s exact test for categorical data. Patients were analyzed overall and by age strata (18–70 vs. ≥ 70 years). Prognostic cut-off values for serum tumor markers were evaluated using a two-step approach. First, we applied standard clinical upper limits (5.0 ng/mL for CEA and 37 U/mL for CA 19 − 9) [6]. Because the standard CA 19 − 9 threshold lacked significant prognostic discrimination in this metastatic cohort, we subsequently identified optimal cut-off points using maximally selected log-rank statistics to determine the tumor burden threshold most strongly associated with PM-OS.

Survival distributions were estimated by the Kaplan–Meier method and compared using two-sided log-rank tests. Factors associated with PM-OS were evaluated via univariable and multivariable Cox proportional hazards models, with results reported as hazard ratios (HRs) and 95% confidence intervals (CIs). Variables with p < 0.10 in univariable analysis were considered for multivariable model, but selection was refined based on model stability and collinearity diagnostics. To address clinical heterogeneity, sensitivity analyses were performed excluding emergency or palliative surgeries.A two-sided p < 0.05 was considered statistically significant.

Results

Patient characteristics

A total of 166 patients with gastric adenocarcinoma and PM were included. The median age at diagnosis was 62 years (range, 29–91), and 56.6% of patients were male. At the time of PM diagnosis, 75.3% had ECOG-PS of 0–1, while 24.7% had ECOG-PS of 2–4. Most patients (83.7%) presented with de novo peritoneal metastasis, while 16.3% had recurrent PM (Table 1). Surgical and pathologic characteristics of the recurrent cohort are detailed in Supplementary Table S1.

Table 1.

Patient and tumor characteristics; data are presented as n (%) unless otherwise indicated

Characteristic	All patients (N = 166)	De novo PM (N = 139)	Recurrent PM (N = 27)	p
Demographics
Age at diagnosis, years	62 (29–91)	63 (29–91)	60 (32–78)	0.219
Sex – male, n (%)	94 (56.6)	79 (56.8)	15 (55.6)	0.902
Sex – female, n (%)	72 (43.4)	60 (43.2)	12 (44.4)
ECOG PS n (%)				0.001
0–1	125 (75.3)	98 (70.5)	27 (100.0)
2–4	41 (24.7)	41 (29.5)	0 (0.0)
Primary tumor characteristics
Primary tumor location				0.180
Proximal (Cardia/Fundus)	46 (27.7)	35 (25.2)	11 (40.7)
Distal (Antrum/Pylor)	50 (30.1)	45 (32.4)	5 (18.5)
Corpus/Diffuse	70 (42.2)	59 (42.4)	11 (40.7)
Lauren histology				0.028
Intestinal	18 (10.8)	11 (7.9)	7 (25.9)
Diffuse	78 (47.0)	67 (48.2)	11 (40.7)
Mixed	15 (9.0)	10 (7.2)	5 (18.5)
Unknown	55 (33.1)	51 (36.7)	4 (14.8)
Biomarkersa
HER2 positive (n: 145)	16/145 (11.0)	12/122 (8.6)	4/23 (14.8)	0.289
PM–related variables
Method of diagnosis				0.342
Radiologic	137 (82.5)	113 (81.3)	24 (88.9)
Surgical	29 (17.5)	26 (18.7)	3 (11.1)
Extent of metastasis				0.163
Isolated PM	72 (43.4)	57 (41.0)	15 (55.6)
PM + extraperitoneal	94 (56.6)	82 (59.0)	12 (44.4)
Systemic treatment characteristics
At least one line of therapy	137 (82.5)	115 (82.7)	22 (81.5)	0.875
Best supportive care	29 (17.5)	24 (17.3)	5 (18.5)

ECOG-PS Eastern Cooperative Oncology Group-Performance Score, GEJ Gastroesophageal junction, HER2 Human epidermal growth factor receptor 2, PM Peritoneal metastasis

^aBiomarker evaluation: Microsatellite instability (MSI) status was evaluated in 43 patients, with only one (2.3%) exhibiting MSI-high status. PD-L1 Combined positive score (CPS) was available for 29 patients, of whom 21 (72.4%) had a CPS ≥ 1.

Patterns of PM and diagnosis

PMwas diagnosed radiologically in 82.5% of patients, whereas 17.5% were identified surgically (staging laparoscopy, definitive surgery, or emergency surgery). Isolated PM was present in 43.4%, and 56.6% had peritoneal metastasis with concomitant extra-peritoneal metastatic sites. Detailed distributions of extra-peritoneal metastases are shown in Supplementary Table S2. The number of metastatic sites was ≥ 3 in 28.9% of the cohort.

The primary tumor was most commonly located in the corpus (30.7%) and antrum (26.5%). According to Lauren classification, diffuse-type histology predominated (47.0%). WHO 2019 histologic subtype was available for 90 patients; among these, 60.0% had signet-ring cell carcinoma and 34.4% had poorly cohesive carcinoma(Supplementary Table S2).

Biomarker profiles

HER2 testing was available in 145 patients; of whom 11.0% were HER2-positive. PD-L1-CPS was reported in 29 patients, and 72.4% had CPS ≥ 1. MSI status was assessed in 43 patients, with only one MSI-high tumor (2.3%). Overall, 24.7% of patients had undergone surgery for the primary tumor.

Clinical proxies for disease burden

To validate the interpretation of diagnostic modality as a surrogate for disease burden, baseline clinical proxies were compared between patients diagnosed radiologically and surgically (Table 2). Macroscopic ascites was significantly more prevalent in the radiologic group compared to the surgical group (43.1% vs. 6.9%; p < 0.001). Furthermore, patients diagnosed surgically were significantly more likely to present with isolated PM rather than concomitant extra-peritoneal disease (65.5% vs. 38.7%; p = 0.008). Importantly, no significant difference was observed in baseline albumin levels between the radiologic and surgical groups (median 3.67 g/dL vs. 3.74 g/dL; p = 0.696). Bowel obstruction or ileus were infrequent (n = 3) and were therefore not included in the formal comparative analysis.

Table 2.

Clinical proxies for disease burden according to diagnostic modality

Variable	Radiologic (n = 137)	Surgical (n = 29)	p-value
Ascites, n (%)	59 (43.1%)	2 (6.9%)	< 0.001
Extent of Disease n (%)			0.008
Isolated PM	53 (38.7%)	19 (65.5%)
PM + Extra-peritoneal	84 (61.3%)	10 (34.5%)
Albumin (median, IQR), g/dL	3.67 (0.68)	3.74 (0.97)	0.696

Survival analyses

A total of 155 of 166 patients (93.4%) died during follow-up. Using the reverse Kaplan–Meier method, the estimated mean follow-up was 60.7 months (95% CI, 43.6–77.8); the median follow-up time was not reached because fewer than 50% of patients had been censored at last contact. The median PM-OS for the entire cohort was 8.2 months (95% CI, 6.63–9.79).

Method of PM detection

When grouped by overall mode of detection, patients whose PM was diagnosed radiologically had a median PM-OS of 7.4 months (95% CI, 5.23–9.55), whereas those with radiologically occult and diagnosed during surgery had a median PM-OS of 13.6 months (95% CI, 8.11–19.16; p = 0.003) (Fig. 2A).

Fig. 2.

Kaplan–Meier estimates of peritoneal metastasis-specific overall survival (PM-OS) according to key prognostic factors. The solid lines show the estimated survival curves, and shaded areas indicate 95% confidence intervals. a Survival stratification by method of PM diagnosis (radiologic vs. surgical detection). Patients with surgically diagnosed (radiologically occult) metastasis exhibited significantly superior survival compared to those with radiologically overt metastasis (p: 0.003) (b) Survival outcomes based on the receipt of metastatic-line systemic therapy. Patients receiving at least one line of therapy had significantly longer survival compared to those receiving best supportive care only (p < 0.001) (c) Impact of HER2 status on survival. HER2-positive patients demonstrated significantly improved survival outcomes compared to HER2-negative patients (p: 0.004). PM-OS, peritoneal metastasis-specific overall survival; HER2, human epidermal growth factor receptor 2; CI, confidence interval

Exploratory descriptive breakdown of surgically diagnosed PM cases showed marked heterogeneity in outcomes. PM first identified at staging laparoscopy had a median PM-OS of 13.44 months (95% CI, 4.83–22.05), while unexpected PM discovered during definitive surgery was associated with the longest median PM-OS of 19.61 months (95% CI, 11.74–27.48). In contrast, PM detected during emergency or palliative surgery had a median PM-OS of 4.17 months (95% CI, 0.12–8.22; global log-rank p = 0.003).

Performance status and disease extent

Baseline performance status strongly correlated with outcome. Patients with ECOG-PS 0–1 had a median PM-OS of 10.32 months (95% CI, 8.79–11.84), whereas those with ECOG-PS 2–4 had a median PM-OS of 5.13 months (95% CI, 4.13–6.12; p = 0.009) (Supplementary Figure S1).

Systemic treatment for metastatic disease

Systemic treatment for metastatic disease was associated with a substantial survival advantage. Among patients who did not receive any metastatic-line therapy, median PM-OS was 2.37 months (95% CI, 1.50–3.23). In contrast, those who received at least one line of metastatic systemic therapy had a median PM-OS of 10.48 months (95% CI, 8.95–12.01; p < 0.001) (Fig. 2B). Detailed chemotherapy regimens and first-line treatment choices are summarized in Supplementary Table S2.

Biomarkers (HER2, CEA, CA19-9)

HER2 status was prognostic in this cohort. Median PM-OS was 8.84 months (95% CI, 6.79–10.88) in HER2-negative patients and 13.63 months (95% CI, 0.00–35.90) in HER2-positive patients (p = 0.004), although the wide confidence interval in the HER2-positive group reflects the small sample size (Fig. 2C).

Serum tumor markers were also associated with PM-OS. Patients with low CEA (< 5.0 ng/mL) had a median PM-OS of 9.1 months (95% CI: 6.83–11.56), whereas those with high CEA (≥ 5.0 ng/mL) had a median PM-OS of 7.06 months (95% CI: 4.17–9.95; p = 0.002). (Supplementary Fig. 2). Using the standard clinical cut-off of 37 U/mL, CA 19 − 9 levels were not significantly associated with PM-OS (p = 0.673). Consequently, an optimized cut-off of 175.4 U/mL was identified, which significantly stratified prognosis (median PM-OS: 10.25 vs. 7.75 months; p = 0.026) (Supplementary Figure S3). This cut-off is exploratory and requires external validation.

Univariable and multivariable Cox regression for PM-OS

On univariable analysis, ECOG-PS 2–4 (vs. 0–1), radiologic (vs. surgical) detection of PM, HER2 negativity, elevated CEA (≥ 5.0 ng/mL), elevated CA19-9 (≥ 175.4 U/mL), and absence of metastatic-line systemic therapy were each significantly associated with worse PM-OS. Age, sex, de novo versus recurrent PM, signet-ring cell morphology, and extent of disease were not significant predictors (all p > 0.05) (Table 3).

Table 3.

Univariate regression analyses for PM-OS

Variable	Overall Survival
	HR	95% CI	p-value
Age (≥ 70 vs. < 70)	1.392	0.961–2.016	0.080
Sex (Female vs. Male)	0.814	0.591–1.123	0.210
Metastatic pattern (Recurrent vs. De novo)	1.345	0.871–2.076	0.181
ECOG-PS (2–4 vs. 0–1)	1.627	1.127–2.349	0.009
Signet-ring component (Absent vs. Present)	0.917	0.652–1.292	0.621
Method of PM diagnosis (Surgical vs. Radiologic)	0.523	0.339–0.808	0.003
Extent of metastasis (PM + extraperitoneal vs. Isolated PM)	1.184	0.859–1.633	0.302
Liver metastasis (Present vs. absent)	1.287	0.893–1.856	0.176
HER2 expression (Positive vs. negative)	0.419	0.228–0.772	0.005
Metastatic-line therapy (Any vs. best supportive care)	0.130	0.080–0.211	< 0.001
Albumin (per 1 g/dL decrease)	1.887	1.393–2.551	< 0.001
CEA (≥ 5.0 vs. < 5.0)	1.516	1.056–2.177	0.024
CA 19 − 9 (≥ 175.4 vs. < 175.4)	1.553	1.049–2.301	0.028

CA 19 − 9 Carbohydrate antigen 19 − 9, CEA Carcinoembryonic antigen, ECOG-PS Eastern Cooperative Oncology Group-Performance Score, HER2 Human epidermal growth factor receptor 2, PM Peritoneal metastasis

In the primary multivariable model, receipt of metastatic-line systemic therapy was excluded to mitigate immortal time bias and confounding by indication. In this model (n = 136), three factors remained independently associated with PM-OS: HER2 positivity (HR 0.312, p < 0.001), surgical diagnosis of PM (HR 0.555, p = 0.020), and elevated CA19-9 levels (CA19-9 ≥ 175.4: HR 1.705, p = 0.024). Age and ECOG-PS were no longer statistically significant after adjustment (Model 1, Table 4). While serum albumin was a significant prognostic factor in univariable analysis (p < 0.001), it was excluded from the multivariable model to maintain parsimony and reduce potential collinearity. This decision was supported by the comparable albumin levels between diagnostic groups (p = 0.696).

Table 4.

Multivariable cox proportional hazards models for PM-OS

Variable	Model 1			Model 2
	HR	95% CI	p-value	HR	95% CI	p-value
ECOG-PS (2–4 vs. 0–1)	1.030	0.645–1.645	0.902	0.785	0.482–1.280	0.332
PM diagnosis method (surgical vs. radiologic)	0.555	0.338–0.913	0.020	0.603	0.366–0.994	0.047
HER2 expression (positive vs. negative)	0.312	0.164–0.593	< 0.001	0.315	0.167–0.596	< 0.001
CEA (≥ 5.0 vs. < 5.0)	1.415	0.929–2.157	0.106	1.356	0.888–2.072	0.159
CA 19 − 9 (≥ 175.4 vs. < 175.4)	1.705	1.074–2.705	0.024	1.582	0.991–2.523	0.054
Age (≥ 70 vs. < 70)	1.429	0.899–2.274	0.131	1.222	0.760–1.964	0.408
Metastatic Treatment (any vs. best supportive care)	-	-	-	0.114	0.057–0.227	< 0.001

The table presents two multivariable models to evaluate independent prognostic factors. Model 1 (Primary Model) focuses on baseline clinicopathologic and biologic drivers, excluding metastatic-line systemic therapy to mitigate immortal time bias and confounding by indication. Model 2 (Sensitivity Analysis) incorporates the receipt of systemic therapy to assess the independence of tumor biology (HER2 status) and disease burden (diagnostic modality) from treatment allocation. The multivariable analysis included 136 patients with complete data for all variables; 30 patients were excluded due to missing data

To address whether these biological and burden-related effects persist under realistic clinical adjustments, a secondary multivariable model was developed by adding receipt of systemic therapy (Model 2, Table 4). While systemic therapy was the strongest predictor of survival (HR 0.114, p < 0.001), HER2 positivity (HR 0.315, p < 0.001) and surgical diagnosis (HR 0.603, p = 0.047) maintained their independent significance.

Given the clinical heterogeneity within the surgical diagnosis group, a further sensitivity analysis was performed by excluding patients diagnosed during emergency or palliative procedures.When the “surgical” group was restricted to staging laparoscopy and definitive surgery, surgical detection remained associated with improved PM-OS compared with radiologic detection (HR 0.486, 95% CI 0.288–0.818; p = 0.007).

Discussion

This retrospective study of 166 patients with gastric adenocarcinoma and PM identified a median PM-OS of 8.2 months. Our multivariable analysis highlights HER2 positivity and surgical detection (a surrogate for low tumor burden) as the primary independent drivers of superior survival. Crucially, the prognostic weight of these factors remained independent of therapeutic intensity and outperformed traditional clinicopathological markers such as age or histology. We established surgical detection of PM as a robust surrogate for low-volume disease, an interpretation substantiated by its association with minimal ascites rather than superior physiological reserve. This suggests that identifying biological drivers and quantifying the occult peritoneal load are essential for precise risk stratification beyond conventional clinical assessments.

The observed median OS of 8.2 months aligns with international reports ranges from 2 to 11 months [4, 7], likely supported by the cohort’s favorable performance status (75.3% ECOG-PS 0–1) and high systemic therapy receipt (82.5%). While these outcomes capture the impact of trastuzumab-based standards, they primarily represent a survival reference for the chemotherapy-only era due to restricted access to immune checkpoint inhibitors (ICIs) necessitated by local reimbursement policies during the study period. Contemporary trials such as CheckMate-649 [8], KEYNOTE-811 [9] and KEYNOTE-859 [10] have since redefined the global standard of care; notably, with 72.4% of our tested patients exhibiting a PD-L1 CPS ≥ 1, the majority of this subgroup might have achieved superior outcomes in settings with broad ICI access. However, while absolute survival values may be higher in modern practice, the fundamental prognostic significance of the biological and burden-based drivers identified in this study is expected to persist as a critical framework for risk stratification in the evolving therapeutic landscape.

We observed no significant difference in PM-OS between de novo and recurrent cases, aligning with population-based data that demonstrate comparable outcomes for synchronous and metachronous peritoneal involvement [4, 11]. While recurrent tumors might theoretically harbor chemo-resistant clones selected by prior adjuvant therapy [12], they exhibited significantly better histological differentiation in our cohort compared to de novo cases. However, this favorable histological feature did not translate into a survival benefit. This disconnect suggests that once the peritoneum is involved, the acquisition of chemotherapy resistance and the functional impact of peritoneal carcinomatosis dominate the clinical course, overriding potential benefits associated with less aggressive histological grading.

Poor performance status (ECOG-PS 2–4) was associated with inferior survival in univariable analysis but lost significance in the multivariable model. This finding suggests that in peritoneal metastasis, disease burden and tumor biology are more dominant drivers of survival than baseline physiological reserve. Since fitter patients were more likely to undergo surgical exploration, the statistical weight of performance status was potentially diluted by this selection bias. Notably, age ≥ 70 years did not independently predict worse outcomes, suggesting that chronological age alone should not be the primary factor for withholding systemic therapy in patients with adequate functional status [13].

As anticipated, metastatic-line systemic therapy was strongly associated with improved survival, confirming its essential role despite the plasma–peritoneum barrier [14].Our findings mirror population-based data showing that active treatment extends survival from 3–4 months to nearly 10 months [15, 16]. Similar trends have been reported in population-based data from the Netherlands, in which an increasing proportion of treated patients was paralleled by improved survival (9.4 months for treated vs 2.1 months for untreated patients) [17]. However, the observed benefit in our cohort (HR 0.130) is likely inflated by immortal time bias and confounding by indication, as fitter patients are naturally more likely to receive therapy. Consequently, we excluded ‘receipt of therapy’ from our primary multivariable model to better isolate the independent prognostic value of tumor biology and disease burden. Importantly, a treatment-adjusted sensitivity analysis (Model 2) confirmed that HER2 status and surgical detection remain robust, intrinsic drivers of survival rather than mere proxies for treatment eligibility.

A notable finding in our study was the survival advantage observed in patients whose PM was radiologically occult and diagnosed surgically compared to those diagnosed radiologically (HR 0.555, p = 0.020). Because computed tomography typically requires high tumor burden or overt ascites [18], surgical exploration identifies radiologically occult, low-volume disease [19]. Consequently, the ‘surgical diagnosis’ group represents a subset with biologically less advanced disease [16, 20, 21]. This underscores the enduring value of staging laparoscopy and identifies potential candidates for intensified regional strategies, such as cytoreductive surgery or intraperitoneal chemotherapy [22–24]. Our data support this burden-based interpretation, while the radiologic group had significantly higher rates of macroscopic ascites and extra-peritoneal spread. This confirms that diagnostic modality serves as a surrogate for tumor volume rather than baseline physiological reserve. Sensitivity analyses excluding emergency/palliative surgery further solidified this surrogate role (HR 0.486, p = 0.007). While incidental PM exhibited the best survival, this likely reflects a highly selected subgroup with minimal disease load rather than a direct therapeutic effect of resection. Ultimately, diagnostic context reflects patient selection and clinical pathways, although residual confounding cannot be fully excluded.

Our study also confirms the critical prognostic and predictive value of HER2 specifically within the setting of peritoneal metastasis. HER2 positivity was independently associated with improved survival, aligned with the ToGA trial and subsequent real-world data demonstrating the efficacy of trastuzumab-based therapy [25–27]. In our cohort, HER2-positive patients had markedly longer PM-OS with 87.5% receiving anti-HER2 therapy. The survival benefit is likely driven more by sensitivity to trastuzumab-based regimens than by inherently indolent tumor biology. However, the HER2-positive subgroup was relatively small, which likely contributed to wider confidence intervals and limited the precision of the effect estimate. Accordingly, this finding should be interpreted in the context of the expected real-world prevalence of HER2 positivity in gastric cancer with peritoneal metastasis. These findings reinforce the necessity of routine HER2 testing and access to targeted therapies in all patients with peritoneal metastatic disease.

Elevated pre-treatment CEA and CA 19 − 9 levels were associated with poorer outcomes in univariable analysis, consistent with their established role in peritoneal dissemination [28–30]. However, in the multivariable model, CA 19 − 9 retained only borderline significance. The standard 37 U/mL cut-off failed to stratify survival (p = 0.673),whereas a data-driven threshold of 175.4 U/mL revealed a significant difference. This observation suggests that in established PM, the magnitude of elevation, which serves as a surrogate for high peritoneal tumor burden, is a more decisive prognostic driver than the marker’s mere presence. These findings align with evidence that markedly elevated CA 19 − 9 levels (especially > 100 U/mL) are required to predict unresectability and poor outcomes in gastric cancer [31, 32]. We emphasize that our optimized cut-off remains exploratory and hypothesis-generating, requiring validation in independent cohorts before routine clinical application.

Traditionally high-risk variables, including signet-ring cell morphology and extra-peritoneal metastases, did not significantly correlate with PM-OS in our cohort. This lack of association likely reflects that by the time PM is clinically evident, the substantial tumor burden and disease aggressiveness across all histologic subtypes obscure potential survival differences. In this setting, the peritoneal burdenappears to be the dominant determinant of the clinical course, effectively neutralizing the prognostic weight of visceral metastatic sites.

Our study has several limitations. Its retrospective, single-center design may limit generalizability and introduce unmeasured confounding. The lack of routine Peritoneal Carcinomatosis Index (PCI) scores precluded a more granular burden analysis, requiring the use of diagnostic modality as a surrogate. Furthermore, incomplete PD-L1 and MSI testing reflects the local reimbursement landscape during the study period. While we utilized a treatment-adjusted sensitivity model, residual confounding regarding patient fitness and functional reserve may persist. Despite these constraints, this cohort provides significant real-world evidence and offers valuable insights into the prognostic stratification of patients with gastric peritoneal metastases.

Conclusion

While gastric adenocarcinoma with PM carries a poor prognosis, our findings reveal significant clinical heterogeneity driven by tumor biology and disease burden. HER2 positivity and surgical detection, a surrogate for lower tumor volume, emerged as the primary independent determinants of survival. Conversely, radiologic detection and treatment ineligibility identify a subset with the poorest life expectancy. Given the critical role of neoadjuvant therapy, improving non-invasive detection of early PM is essential to prevent unnecessary surgeries and accelerate systemic treatment initiation by bypassing post-operative recovery. Integrating these clinical and molecular features into multidisciplinary care is vital to personalize treatment and prioritize patients for clinical trials involving novel regional strategies. Future prospective studies incorporating PCI scores, comprehensive molecular profiling, and contemporary chemo-immunotherapy are needed to validate these prognostic stratifiers.

Abbreviations

ASCO/CAP

American Society of Clinical Oncology / College of American Pathologists

CA 19 − 9

Cancer antigen 19 − 9

CEA

Carcinoembryonic antigen

CI

Confidence intervals

CPS

Combined Positive Score

ECOG-PS

Eastern Cooperative Oncology Group Performance Status

HER2

Human epidermal growth factor receptor 2

HR

Hazard ratio

ICI

Immune checkpoint inhibitor

IQR

Interquartile range

GEJ

Gastroesophageal junction

MSI

Microsatellite instability

OS

Overall survival

PCI

Peritoneal Carcinomatosis Index

PM

Peritoneal metastasis

PM-OS

Peritoneal metastasis - specific survival

PD-L1

Programmed cell death ligand 1

WHO

World Health Organization

χ²

Chi-square test

Authors’ contributions

KC: Investigation, Methodology, Data Curation, Formal analysis, Visualization, Writing - Original DraftED: Data Curation, Writing - Original DraftSÇ: Data CurationKA: Writing - Review & EditingEA: Writing - Review & EditingİÖ: Conceptualization, Methodology, Writing - Review & Editing, Supervision.

Funding

The authors received no funding for this research.

Data availability

The datasets used and analyzed during the current study are not publicly available due to institutional privacy regulations. Data may be made available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Dokuz Eylül University Ethics Committee, with a waiver of informed consent due to the retrospective design and was conducted in accordance with the Declaration of Helsinki (decision number 2024/42 − 13, 18.12.2024).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.