Prognostic model for patients with hepatoid adenocarcinoma of the stomach Authors: Shan Tong, Boqi Xu, Xiang Ye, Jun Yin, Hao Hu Published: 2026 ABSTRACT IntroductionHepatoid adenocarcinoma of the stomach (HAS) is a rare form of gastric cancer. This study assessed the impact of tumor node metastasis (TNM) stage on survival in HAS compared with non-HAS patients, identified prognostic factors, and developed a predictive nomogram.MethodsPatients diagnosed with HAS between September 2012 and August 2022 were retrospectively analyzed. Clinical characteristics and survival outcomes were compared between HAS and non-HAS groups before and after TNM stage matching. Prognostic factors were determined using Kaplan–Meier and Cox regression analyses, and nomograms were constructed for survival prediction.ResultsWithout stage matching, HAS patients showed significantly worse survival than non-HAS patients (HR = 3.09, 95% CI: 2.05–4.67, p < 0.001). After TNM stage matching, no significant difference was observed (HR = 1.14, 95% CI: 0.79–1.64, p = 0.476). The nomogram model incorporating Sal-like protein 4 (SALL4), serum iron (Fe), and free prostate-specific antigen (fPSA) accurately predicted overall survival (AUC = 0.946, 95% CI: 0.897–0.994).ConclusionsThese findings suggest that poorer outcomes in HAS are primarily due to advanced TNM stages, and that SALL4, serum Fe, and fPSA serve as key prognostic indicators.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12876-026-04933-z. What is known? Hepatoid adenocarcinoma of the stomach (HAS) is a rare and aggressive gastric cancer subtype associated with poor prognosis, but the specific reasons for its worse outcomes remain unclear. What new information does this article contribute? This study demonstrates that the poorer survival in HAS patients is primarily attributable to advanced TNM stages rather than intrinsic tumor aggressiveness, and identifies SALL4, serum iron, and free PSA as novel prognostic biomarkers for constructing a highly accurate predictive nomogram (AUC = 0.946). Introduction Gastric cancer (GC) is a heterogeneous, multifactorial malignancy arising from the gastric mucosal epithelium. It poses a serious threat to human health and represents a major focus of global public health prevention and control [1]. According to epidemiological data, GC was the fifth most commonly diagnosed cancer worldwide in 2018 and the third leading cause of cancer-related mortality [2]. In the Chinese population, the etiology of GC is attributed to a combination of biological and behavioral factors. Biological factors primarily include Helicobacter pylori (HP) infection, whereas behavioral factors include smoking, alcohol consumption, and dietary patterns [3, 4]. Notably, HP infection alone is insufficient to cause GC; even after bacterial eradication, persistent lifestyle factors continue to influence carcinogenesis [5]. Therefore, a comprehensive prevention strategy combining HP eradication with lifestyle modifications is essential for reducing GC risk [6]. In recent years, with improvements in living standards and medical conditions, the prevalence of GC has trend to decrease. However, emerging challenges—including population aging, increased consumption of processed foods, and sedentary lifestyles—have offset some of the gains in prevention and control, thereby slowing the decline in the incidence of GC [7, 8]. From a long-term perspective, the incidence and mortality of GC in China declined between 1990 and 2021, with corresponding improvements in life expectancy [9]; nevertheless, GC continues to pose a substantial threat to public health. Hepatoid adenocarcinoma of the stomach (HAS), a rare subtype of GC, accounts for only 0.3%–1.0% of all cases. The entity was first described by Ishikura in 1985 [10, 11]. This subtype has distinct clinical and pathological characteristics, including elevated serum alpha-fetoprotein (AFP) levels and a histological morphology resembling hepatocellular carcinoma (HCC) [12]. Previous studies have demonstrated that HAS patients exhibit more aggressive biological behavior than non-HAS patients do, characterized by poorer differentiation, more frequent neural and vascular invasion, a higher incidence of lymph node and liver metastasis, and a worse overall prognosis [13–15]. However, the survival outcomes of HAS patients compared with non-HAS patients remain controversial: while some studies have reported significantly poorer outcomes in HAS patients, other has reported no statistically significant differences between the two groups [16]. HAS represents a primary gastric adenocarcinoma subtype characterized by hepatoid differentiation, rather than a secondary metastatic condition. According to the 2019 World Health Organization Classification of Digestive System Tumors, the diagnosis is established by histopathological identification of hepatoid components within the gastric tumor, irrespective of the proportion of such components or serum AFP levels [14]. This entity exhibits inherently aggressive biological behavior, which is clinically characterized by rapid disease progression, a highlikelihood of liver metastasis, and poor prognosis. However, notably, a large proportion of HAS patients present with lymphatic or distant metastases, whereas this proportion is considerably lower in non-HAS populations. This discrepancy may partly explain the differences in reported survival outcomes between HAS patients and non-HAS patients [17]. Therefore, the conflicting findings may be attributed to the greater propensity of HAS for lymphatic and distant metastases. When propensity score matching (PSM) was used to balance the proportion of patients with lymphatic metastasis between groups, no significant difference in survival was observed. Conversely, if lymphatic and distant metastases are not matched—for example, after excluding patients with these features—the survival rate of non-HAS patients is significantly higher [18, 19]. Therefore, the main objective of this study was to investigate survival outcome differences between HAS patients and non-HAS patients matched or unmatched by TNM stage in a single-center cohort, identify risk factors influencing the three-year survival of HAS patients, and construct nomogram predictive models to evaluate their prognostic accuracy. Methods Sample source This study adopted a single-center retrospective cohort design, enrolling HAS patients admitted to the Department of General Surgery, ***, between September 2012 and August 2022, along with non-HAS patients admitted during the same period. Ethical approval for this study was waived by the Medical Ethics Committee of the First Affiliated Hospital of Soochow University. Given the retrospective nature of the study and the use of anonymized clinical data without direct patient intervention, the Medical Ethics Committee waived the requirement for informed consent. All patient identifiers were anonymized in strict accordance with ethical standards for medical research. Patient selection Patients were included if they met the following eligibility criteria; otherwise, they were excluded. The inclusion criteria were as follows:1) Adults aged ≥ 18 years, regardless of sex; 2) underwent radical surgery (R0 resection), defined as no residual microscopic tumor after resection; 3) pathological confirmation: HAS patients were diagnosed by biopsy as having gastric hepatoid adenocarcinoma, whereas non-HAS patients were diagnosed with nonhepatoid gastric adenocarcinoma subtypes; and 4) availability of complete clinical records (demographic data, preoperative tests, surgical records, pathology reports, and follow-up data) were available. The exclusion criteria were as follows: (1) severe comorbidities such as hematological diseases, cardiac insufficiency, chronic obstructive pulmonary disease, liver cirrhosis, renal failure, or immunodeficiency; (2) the presence of a primary malignancy in another organ (active or inactive); (3) palliative rather than curative (R0) surgery; and (4) incomplete clinical records (e.g., unclear pathological stage or treatment data) or interrupted postoperative follow-up (loss to follow-up). Surgery and postoperative treatment All enrolled patients underwent radical surgery for GC. The surgical procedures included primary tumor resection (distal, proximal, or total gastrectomy depending on the tumor location) and standard lymph node dissection, performed in accordance with clinical guidelines for tumor stage and location. When definite liver metastasis was detected preoperatively or intraoperatively and deemed resectable, simultaneous liver resection was performed. TNM stage was determined according to the 8th edition of the American Joint Committee on Cancer (AJCC) Cancer Staging Manual. Patients meeting the indications for adjuvant therapy received postoperative chemotherapy on the basis of their TNM stage. The main agents included fluoropyrimidines (e.g., capecitabine, S-1), cisplatin, and doxorubicin, which were administered either as monotherapy or in doublet regimens. Chemotherapy was typically administered for 6–8 cycles, with regimens adjusted according to individual tolerance. General data collection Clinical data were extracted from the hospital’s case information management system and independently collected by two trained researchers. The data collected included demographic characteristics, clinical treatment information, tumor characteristics, laboratory test results, and specific pathological data. The demographic variables included age and sex. Clinical treatment information primarily included the receipt of postoperative adjuvant chemotherapy. The tumor characteristics included the TNM stage. The laboratory indicators included tumor markers (e.g., AFP, carcinoembryonic antigen, carbohydrate antigen 199), coagulation parameters, routine hematological tests (e.g., white blood cell count, hemoglobin level), and biochemical tests (e.g., liver and renal function parameters). For HAS patients, immunohistochemical staining results from postoperative pathological specimens were collected, and the expression of relevant markers was recorded in detail. Pathological diagnosis and immunohistochemistry The pathological diagnosis and immunohistochemical confirmation of HAS were required to meet both histopathological features and marker expression criteria, namely, histology demonstrating hepatoid differentiation (large polygonal regions composed of eosinophilic tumor cells) and immunohistochemistry showing AFP positivity. Tumor samples from HAS patients were fixed in 10% neutral formalin immediately after surgery, subjected to graded dehydration, xylene clearing, and paraffin embedding, and sectioned into 4-µm slices for hematoxylin–eosin (HE) and immunohistochemical staining (Fig. 1). The primary antibodies used for immunohistochemistry included those against cytokeratin pan (CK-pan), CK7, CK18, AFP, hepatocyte-specific antigen 1 (Hepa-1), glypican-3, Sal-like protein 4 (SALL4), and chromogranin A (CgA). All HE- and immunohistochemically stained sections were independently reviewed by two experienced pathologists in a blinded manner. Immunohistochemical interpretation was performed by evaluating at least 10 high-power fields (400× magnification) per section. For SALL4, only nuclear staining (brown-yellow particles in cell nuclei) was considered positive, with cytoplasmic staining excluded; positive expression was defined as > 10% of tumor cell nuclei showing positive staining. For glypican-3, staining was considered positive when brown-yellow particles were observed in the cell membrane and/or cytoplasm, with pure nuclear staining excluded; positive expression was defined as ≥ 10% of tumor cells demonstrating positive membranous/cytoplasmic staining. In cases of diagnostic discrepancy, a consensus was reached through departmental consultation to ensure diagnostic reliability. Fig. 1Representative histopathological images of HAS and non-HAS GC tumors. A HAS tumor, ×10 magnification; B HAS tumor, ×20 magnification; C non-HAS GC tumor, ×10 magnification; D non-HAS GC tumor, ×20 magnification Representative histopathological images of HAS and non-HAS GC tumors. A HAS tumor, ×10 magnification; B HAS tumor, ×20 magnification; C non-HAS GC tumor, ×10 magnification; D non-HAS GC tumor, ×20 magnification Definition of outcome indicators and follow-up Patients were followed up through a combination of outpatient visits and telephone interviews, with survival status recorded monthly. The follow-up deadline was August 2025. Overall survival (OS) was the primary outcome and was defined as the time from radical surgery to death from any cause. Patients who were alive or lost to follow-up at the study endpoint were censored at the date of last contact, with survival time calculated from the date of surgery to the date of last follow-up. Statistical methods All the statistical analyses were performed via R software (version 4.4.2). Data processing and model construction employed specialized statistical packages, including rms and pROC. A two-sided significance level was set at α = 0.05, with p < 0.05 considered statistically significant. Propensity score matching (PSM) was performed, with HAS patients in the experimental group and non-HAS patients in the control group, at a matching ratio of 1:4. The caliper width was set at 0.05, with matching factors including age, sex, and tumor markers (excluding AFP). The selection of these variables was determined on the basis of the established gastric cancer literature and clinical rationale. Age and sex were included because they represent fundamental demographic confounders affecting tumor biological behavior, treatment tolerance, and prognosis. Tumor markers (carcinoembryonic antigen and carbohydrate antigen 19 − 9) were selected to reflect tumor burden, differentiation status, and metastatic potential [20]. AFP was excluded because elevated serum AFP is a pathognomonic feature of HAS rather than a confounding variable; furthermore, given the limited sample size of HAS patients (n = 67), including AFP as a matching covariate would result in insufficient matching pairs and compromise statistical power. Subgroup analyses were conducted for “unmatched TNM staging” and “matched TNM staging” to control for staging-related confounders. For descriptive statistics, the Shapiro–Wilk test was applied to assess the normality of continuous variables. Normally distributed data are expressed as the means ± standard deviations (x̄ ± s) and were compared via independent-samples t-tests. Nonnormally distributed data are expressed as medians (interquartile ranges) [M (Q1–Q3)] and were compared via the Mann–Whitney U test. Categorical variables are expressed as counts and percentages [n (%)]. Group comparisons were performed via Pearson’s χ² test or Fisher’s exact test when the expected cell counts were < 5. Survival curves were generated via the Kaplan–Meier method, and OS was estimated for different groups (e.g., HAS vs. non-HAS, and risk factor subgroups). Survival differences were assessed via the log-rank test. For continuous risk factors (e.g., tumor diameter and the AFP level), optimal cutoff values were determined via Youden’s index from receiver operating characteristic (ROC) curves, and variables were dichotomized accordingly. For risk factor analysis and model construction, all potential predictors (e.g., age, tumor stage, vascular invasion, and serum AFP) were first assessed via univariate Cox proportional hazards regression, and variables with p < 0.05 were retained. To ensure model stability, variance inflation factor (VIF) analysis was performed on these candidate variables to assess multicollinearity; variables with a VIF > 10 were excluded from further analysis. Significant variables without multicollinearity were subsequently entered into a multivariate Cox proportional hazards model, with age and sex adjusted as potential confounders, to identify independent prognostic factors for OS in HAS patients. Finally, a nomogram prediction model was constructed on the basis of the independent prognostic factors identified via multivariate Cox analysis. ROC curves [area under curve (AUC) values] were used to evaluate the predictive accuracy for OS in HAS patients. Internal validation of the nomogram was performed using the bootstrap resampling method with 300 replications to estimate and correct for model optimism. In each bootstrap iteration, a sample of the same size as the original dataset was drawn with replacement, and the model was refitted. The optimism for discrimination (Harrell’s C-index) and calibration was calculated as the difference between the apparent performance in the bootstrap sample and the performance when applied to the original dataset. Decision curve analysis (DCA) was performed to evaluate the clinical net benefit of the nomogram across threshold probabilities. Results Basic characteristics A total of 831 patients with GC who underwent radical surgery between September 2012 and August 2022 were included, comprising 67 patients with HAS and 764 with non-HAS. PSM analysis without TNM staging PSM was performed at a ratio of 1:4 (matching factors: age, sex, and tumor markers excluding AFP; caliper width = 0.05), yielding 67 HAS patients and 268 matched non-HAS patients. Serum AFP levels were significantly greater in the HAS group than in the non-HAS group (p < 0.001). Significant differences in TNM-related indices, including T1b, T3, and T4a subtypes of T stage, N3 subtype of N stage, and M stage (distant metastasis), were observed. The proportion of patients with T3, T4a, N3, and M disease was greater in the HAS group (all p < 0.05) (Table 1). Survival analysis revealed significantly worse OS in the HAS group than in the non-HAS group (HR = 3.09, 95% CI: 2.05–4.67, p < 0.001) (Fig. 2, A). Table 1Comparison of baseline characteristics between HAS patients and non-HAS patients before TNM-stage matchingVariableNo-HASHASZ/chi2 p N26867Age67(61-74.5)68(63.5–72.5)-0.9230.356Sex Female51(19.03%)12(17.91%)0.0440.834 Male217(80.97%)55(82.09%)SerAFP15.38(17.11–33.73)164.1(11.66-555.38)-5.896< 0.001CEA5.36(2.16–20.2)6.91(2.16–37.38)-0.760.447CA1998.84(3.84–24.66)9.39(5.02–37.62)-1.0280.304Ferritin72.63(22.74–97.19)72.63(23.55–97.19)-0.3970.691CA12514.7(8.5–22.5)16.5(9.55–38.65)-1.0890.276CA1539.3(6-12.54)9.73(6.4-12.54)-1.0390.299CA7242.29(1.12–6.44)2.64(1.17–8.4)-0.9860.324CYFRA2112.59(1.73–3.45)2.59(2.1–3.45)-1.4150.157NSE3.14(2.5–3.22)3.14(3.04–3.22)-1.2190.223SCCA0.97(0.7–1.06)0.97(0.7–1.06)-0.520.603T stage 1b41(15.30%)2(2.99%)0.004 257(21.27%)12(17.91%)0.370.543 362(23.13%)8(11.94%)4.0630.044 4a96(35.82%)39(58.21%)11.1670.001 4b12(4.48%)6(8.96%)0.22N stage 099(36.94%)18(26.87%)2.3940.122 172(26.87%)14(20.90%)1.0010.317 248(17.91%)13(19.40%)0.080.777 349(18.28%)22(32.84%)6.7960.009M stage 0235(87.69%)65(97.01%)0.025 133(12.31%)2(2.99%)Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Table 1Comparison of baseline characteristics between HAS patients and non-HAS patients before TNM-stage matchingVariableNo-HASHASZ/chi2 p N26867Age67(61-74.5)68(63.5–72.5)-0.9230.356Sex Female51(19.03%)12(17.91%)0.0440.834 Male217(80.97%)55(82.09%)SerAFP15.38(17.11–33.73)164.1(11.66-555.38)-5.896< 0.001CEA5.36(2.16–20.2)6.91(2.16–37.38)-0.760.447CA1998.84(3.84–24.66)9.39(5.02–37.62)-1.0280.304Ferritin72.63(22.74–97.19)72.63(23.55–97.19)-0.3970.691CA12514.7(8.5–22.5)16.5(9.55–38.65)-1.0890.276CA1539.3(6-12.54)9.73(6.4-12.54)-1.0390.299CA7242.29(1.12–6.44)2.64(1.17–8.4)-0.9860.324CYFRA2112.59(1.73–3.45)2.59(2.1–3.45)-1.4150.157NSE3.14(2.5–3.22)3.14(3.04–3.22)-1.2190.223SCCA0.97(0.7–1.06)0.97(0.7–1.06)-0.520.603T stage 1b41(15.30%)2(2.99%)0.004 257(21.27%)12(17.91%)0.370.543 362(23.13%)8(11.94%)4.0630.044 4a96(35.82%)39(58.21%)11.1670.001 4b12(4.48%)6(8.96%)0.22N stage 099(36.94%)18(26.87%)2.3940.122 172(26.87%)14(20.90%)1.0010.317 248(17.91%)13(19.40%)0.080.777 349(18.28%)22(32.84%)6.7960.009M stage 0235(87.69%)65(97.01%)0.025 133(12.31%)2(2.99%)Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Comparison of baseline characteristics between HAS patients and non-HAS patients before TNM-stage matching Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Fig. 2Kaplan-Meier overall-survival curves for TNM-stage-unmatched (A) and TNM-stage-matched (B) cohorts (generated by 1:4 PSM) of HAS and non-HAS GC patients Kaplan-Meier overall-survival curves for TNM-stage-unmatched (A) and TNM-stage-matched (B) cohorts (generated by 1:4 PSM) of HAS and non-HAS GC patients PSM analysis with TNM staging TNM stage was subsequently added as a matching factor, and 1:4 PSM was repeated. After matching, the clinical and pathological characteristics of the patients in the two groups were well balanced. After adjustment, only the serum AFP level remained significantly higher in the HAS group (p < 0.001). No significant differences were observed in the TNM stage indices (T, N, M), other laboratory parameters, or clinical treatment characteristics (all p > 0.05) (Table 2). Survival analysis after TNM-adjusted matching revealed no significant difference in OS between HAS patients and non-HAS patients (HR = 1.14, 95% CI: 0.79–1.64, p = 0.476) (Fig. 2, B). Table 2Comparison of baseline characteristics between HAS patients and non-HAS patients in the TNM-matched cohortVariableNo-HASHASZ/Chi2 p N26867Age67(59.5–73)68(63.5–72.5)-1.2210.222Sex Female62(23.13%)12(17.91%)0.850.357 Male206(76.87%)55(82.09%)SerAFP18.66(16.07–33.67)164.1(11.66-555.38)-4.459< 0.001CEA6.67(2.16–20.2)6.91(2.16–37.38)-0.4730.636CA1998.9(4.11–21.96)9.39(5.02–37.62)-0.9270.354Ferritin72.63(21.07–97.19)72.63(23.55–97.19)-0.9310.352CA12515.2(8.45–25.25)16.5(9.55–38.65)-1.0490.294CA1539.3(6-12.54)9.73(6.4-12.54)-1.0470.295CA7242.35(1.1–6.67)2.64(1.17–8.4)-0.9270.354CYFRA2112.59(1.67–3.45)2.59(2.1–3.45)-1.6450.1NSE3.14(2.5–3.22)3.14(3.04–3.22)-1.4050.16SCCA0.97(0.7–1.06)0.97(0.7–1.06)-1.190.234T stage 1b9(3.36%)2(2.99%)> 0.999 244(16.42%)12(17.91%)0.0860.77 333(12.31%)8(11.94%)0.0070.934 4a157(58.58%)39(58.21%)0.0030.956 4b25(9.33%)6(8.96%)0.0090.925N 078(29.10%)18(26.87%)0.1310.717 160(22.39%)14(20.90%)0.0690.792 249(18.28%)13(19.40%)0.0450.833 381(30.22%)22(32.84%)0.1720.679M 0259(96.64%)65(97.01%)> 0.999 19(3.36%)2(2.99%)Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Table 2Comparison of baseline characteristics between HAS patients and non-HAS patients in the TNM-matched cohortVariableNo-HASHASZ/Chi2 p N26867Age67(59.5–73)68(63.5–72.5)-1.2210.222Sex Female62(23.13%)12(17.91%)0.850.357 Male206(76.87%)55(82.09%)SerAFP18.66(16.07–33.67)164.1(11.66-555.38)-4.459< 0.001CEA6.67(2.16–20.2)6.91(2.16–37.38)-0.4730.636CA1998.9(4.11–21.96)9.39(5.02–37.62)-0.9270.354Ferritin72.63(21.07–97.19)72.63(23.55–97.19)-0.9310.352CA12515.2(8.45–25.25)16.5(9.55–38.65)-1.0490.294CA1539.3(6-12.54)9.73(6.4-12.54)-1.0470.295CA7242.35(1.1–6.67)2.64(1.17–8.4)-0.9270.354CYFRA2112.59(1.67–3.45)2.59(2.1–3.45)-1.6450.1NSE3.14(2.5–3.22)3.14(3.04–3.22)-1.4050.16SCCA0.97(0.7–1.06)0.97(0.7–1.06)-1.190.234T stage 1b9(3.36%)2(2.99%)> 0.999 244(16.42%)12(17.91%)0.0860.77 333(12.31%)8(11.94%)0.0070.934 4a157(58.58%)39(58.21%)0.0030.956 4b25(9.33%)6(8.96%)0.0090.925N 078(29.10%)18(26.87%)0.1310.717 160(22.39%)14(20.90%)0.0690.792 249(18.28%)13(19.40%)0.0450.833 381(30.22%)22(32.84%)0.1720.679M 0259(96.64%)65(97.01%)> 0.999 19(3.36%)2(2.99%)Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Comparison of baseline characteristics between HAS patients and non-HAS patients in the TNM-matched cohort Abbreviations: CEA Carcinoembryonic Antigen, CA125 Cancer Antigen 125, CA153 Cancer Antigen 153, CA199 Carbohydrate Antigen 199, CA724 Cancer Antigen 724, CYFRA211 Cytokeratin Fragment 21 − 1, NSE Neuron-Specific Enolase, SerAFP Serum Alpha-Fetoprotein, SCCA Squamous Cell Carcinoma Antigen Comparison of prognostic factors in the HAS population The 67 HAS patients were stratified into a death group (n = 32) and a survivor group (n = 35) on the basis of 3-year survival outcomes. A comparison of the two groups (Table 3) revealed significant differences in several laboratory parameters, including carbohydrate antigen 199 (CA19-9), ferritin, tumor specific growth factor (TSGF), free prostate-specific antigen (fPSA), cytokeratin fragment 21 − 1 (CYFRA21-1), NSE, the free-to-total prostate-specific antigen (F/TPSA) ratio, the serum iron (Fe), and the neutrophil percentage (NEUT%) (all p < 0.05). In terms of clinical treatment, a significantly greater proportion of patients in the survivor group than in the nonsurvivor group received chemotherapy (p < 0.05). For the immunohistochemical markers, the positive expression rates of Glypican-3 and SALL4 were significantly greater in the nonsurvivor group than in the survivor group (p < 0.05). Because multiple laboratory variables were explored, the findings should be interpreted as exploratory and hypothesis-generating. Table 3Comparison of clinical characteristics between survivors and nonsurvivors among HAS patients during 3 years of follow-upVariableSurvival (n = 30)Death (n = 37)Z/t/Chi2/Fisherp valueAge (year)67.4 ± 7.2667.73 ± 8.93-0.1670.868Serum AFP (ug/L)210.94(32.22-336.68)118.38(3.53-695.24)-0.0310.975CEA (ng/mL)9.77(2.4-35.16)4.51(1.9-37.38)-0.9520.341CA199 (IU/mL)15.9(7.17–77.11)7.49(3.79–19.94)-2.1830.029Ferritin (ng/mL)97.19(41.59–97.19)72.63(18.24–72.63)-2.1510.032CA125 (U/mL)14.95(9-22.5)19.1(9.6-43.36)-1.5630.118CA153 (U/mL)9.73(7.05–9.73)9.7(6.1-12.54)-0.0750.94TSGF (U/mL)66.72(66.22–66.93)70.28(68.6-70.28)-3.3280.001fPSA(ng/mL)0.27(0.27–0.27)0.45(0.35–0.45)-3.9880.000CA724 (U/mL)2.91(1.03–27.83)2.3(1.21–7.35)-0.8850.376CYFRA211 (ng/mL)2.59(1.8–2.59)3.45(2.43–3.45)-2.1250.034NSE (ng/mL)3.22(3.05–3.26)3.14(3.04–3.14)-2.2680.023SCCA (ng/mL)0.97(0.83–0.97)1.06(0.7–1.06)-1.5940.111F/TPSA0.36(0.36–0.36)0.31(0.31–0.31)-4.8660.000AP (sec)11.4(10.93–12.65)11.7(11.1–12.5)-0.9840.325APTT (sec)26.85(25.7-29.52)28.2(26.5–28.6)-0.9720.331TT (sec)16.3(15.83–17.92)17.4(16.5–18.6)-1.9560.051FIB (g/L)3.45(3.08–4.11)3.87(3.09–4.4)-1.0340.301Fe(µmol/L)8.42(5.37–8.42)8.03(8.03–8.03)-2.7980.005PA(mg/L)195.73 ± 51.99172.97 ± 62.751.6230.109LDLC(mmol/L)2.3 ± 0.742.38 ± 0.76-0.440.662HDLC(mmol/L)0.97 ± 0.231.04 ± 0.3-1.1660.248TG(mmol/L)1.25(1.04–1.81)1.24(1.07-2)-0.4340.664TC(mmol/L)3.62(3.15–4.34)3.95(2.15–4.35)-0.1830.855GLU(mmol/L)5.49(4.67–5.9)5.21(4.73–5.8)-0.0560.955UA(µmol/L)293.5 ± 89.44280.84 ± 77.030.6130.542Cr(µmol/L)65.5(60.3-72.65)71.9(64-79.4)-1.8910.059Urea(mmol/L)5.54(4.62–6.18)5.3(4.6–6.4)-0.6190.536Agr(ratio)1.43 ± 0.321.38 ± 0.320.6450.522Glb(g/L)26.07 ± 4.8826.9 ± 4.54-0.7130.479ALB(g/L)35.9 ± 4.6335.88 ± 5.010.0160.987TP (g/L)61.95 ± 7.1662.77 ± 6.11-0.4990.619ALP(U/L)75.31(61.9-88.83)81.77(69.1–91.3)-0.9460.344GGT(U/L)21(15-25.6)21(15-30.4)-0.9580.338AST(U/L)16.66 ± 4.8918.96 ± 5.09-1.8780.065ALT(U/L)13.05(9.48–15.9)12.7(10-16.5)-0.0820.935IBIL(µmol/L)6.74(4.93–7.97)6.7(5–9)-0.4410.659DBIL(µmol/L)3.48(2.5–4.27)3.2(2.5–5.6)-0.5620.574TBA (µmol/L)10.95(7.17–12.28)9.8(7-13.8)-0.5420.588WBC (*10^9/L)6.04(4.85–7.28)5.84(4.42–6.82)-0.6170.537LYM (*10^9/L)1.26 ± 0.451.36 ± 0.43-0.9440.349MC(*10^9/L)0.45 ± 0.150.44 ± 0.170.2990.766NEUT(*10^9/L)4.1(2.83–5.09)3.66(2.62–4.66)-0.7260.468EGC(*10^9/L)0.08(0.04–0.16)0.11(0.05–0.16)-0.5240.600BAS(*10^9/L)0.03(0.02–0.04)0.03(0.01–0.04)-0.1670.867LP(%)20.75(12.15–27.78)20.4(2.1–27.8)-0.3030.762MCP(%)7.35(5.53–8.9)6.1(0.4–8.5)-1.0710.284NEUTP (%)67.35(57.55–72.48)58(37.4–66.4)-2.080.038EGCP(%)1.3(0.52–2.4)1.4(0.04–2.3)-0.2390.811BASP(%)0.4(0.23–0.6)0.4(0.1–0.6)-0.6780.498RBC(*10^12/L)3.88 ± 0.743.63 ± 0.911.2480.217HB (g/L)108.67 ± 28.05101.06 ± 28.231.1010.275PLT(*10^9/L)243.5(206–278)259(214–325)-0.6430.520MTD (cm)5(3.2-6)5.5(4-6.8)-1.4610.144Sex Female5(16.70%)7(18.90%)0.0570.811 Male25(83.30%)30(81.10%)VTT Negative16(53.30%)16(43.20%)0.6760.411 Positive14(46.70%)21(56.80%)PNI Negative23(76.70%)29(78.40%)0.0280.867 Positive7(23.30%)8(21.60%)Chemotherapy No2(6.70%)16(43.20%)11.2810.001 Yes28(93.30%)21(56.80%)CK(Pan) Negative14(46.70%)10(27.00%)2.780.095 Positive16(53.30%)27(73.00%)CK7 Negative22(73.30%)23(62.20%)0.9370.333 Positive8(26.70%)14(37.80%)CK18 Negative23(76.70%)20(54.10%)3.6850.055 Positive7(23.30%)17(45.90%)AFP Negative3(10.00%)11(29.70%)0.070 Positive27(90.00%)26(70.30%)Hepa1 Negative22(73.30%)23(62.20%)0.9370.333 Positive8(26.70%)14(37.80%)Glypican3 Negative22(73.30%)18(48.60%)4.1960.041 Positive8(26.70%)19(51.40%)SALL4 Negative30(100.00%)8(21.60%)0.000 Positive0(0.00%)29(78.40%)CgA Negative23(76.70%)32(86.50%)1.0870.297 Positive7(23.30%)5(13.50%)CD56 Negative28(93.30%)31(83.80%)0.281 Positive2(6.70%)6(16.20%)Syn Negative22(73.30%)26(70.30%)0.0770.782 Positive8(26.70%)11(29.70%)HER2 Negative20(66.70%)25(67.60%)0.0060.938 Positive10(33.30%)12(32.40%)T Stage 1b2(6.70%)0(0.00%)0.197 23(10.00%)9(24.30%)2.3120.128 33(10.00%)5(13.50%)0.722 4a20(66.70%)19(51.40%)1.5970.206 4b2(6.70%)4(10.80%)0.684N stage 09(30.00%)9(24.30%)0.2720.602 16(20.00%)8(21.60%)0.0260.871 25(16.70%)8(21.60%)0.2600.610 310(33.30%)12(32.40%)0.0060.938M stage 028(93.30%)37(100.00%)0.197 12(6.70%)0(0.00%) Table 3Comparison of clinical characteristics between survivors and nonsurvivors among HAS patients during 3 years of follow-upVariableSurvival (n = 30)Death (n = 37)Z/t/Chi2/Fisherp valueAge (year)67.4 ± 7.2667.73 ± 8.93-0.1670.868Serum AFP (ug/L)210.94(32.22-336.68)118.38(3.53-695.24)-0.0310.975CEA (ng/mL)9.77(2.4-35.16)4.51(1.9-37.38)-0.9520.341CA199 (IU/mL)15.9(7.17–77.11)7.49(3.79–19.94)-2.1830.029Ferritin (ng/mL)97.19(41.59–97.19)72.63(18.24–72.63)-2.1510.032CA125 (U/mL)14.95(9-22.5)19.1(9.6-43.36)-1.5630.118CA153 (U/mL)9.73(7.05–9.73)9.7(6.1-12.54)-0.0750.94TSGF (U/mL)66.72(66.22–66.93)70.28(68.6-70.28)-3.3280.001fPSA(ng/mL)0.27(0.27–0.27)0.45(0.35–0.45)-3.9880.000CA724 (U/mL)2.91(1.03–27.83)2.3(1.21–7.35)-0.8850.376CYFRA211 (ng/mL)2.59(1.8–2.59)3.45(2.43–3.45)-2.1250.034NSE (ng/mL)3.22(3.05–3.26)3.14(3.04–3.14)-2.2680.023SCCA (ng/mL)0.97(0.83–0.97)1.06(0.7–1.06)-1.5940.111F/TPSA0.36(0.36–0.36)0.31(0.31–0.31)-4.8660.000AP (sec)11.4(10.93–12.65)11.7(11.1–12.5)-0.9840.325APTT (sec)26.85(25.7-29.52)28.2(26.5–28.6)-0.9720.331TT (sec)16.3(15.83–17.92)17.4(16.5–18.6)-1.9560.051FIB (g/L)3.45(3.08–4.11)3.87(3.09–4.4)-1.0340.301Fe(µmol/L)8.42(5.37–8.42)8.03(8.03–8.03)-2.7980.005PA(mg/L)195.73 ± 51.99172.97 ± 62.751.6230.109LDLC(mmol/L)2.3 ± 0.742.38 ± 0.76-0.440.662HDLC(mmol/L)0.97 ± 0.231.04 ± 0.3-1.1660.248TG(mmol/L)1.25(1.04–1.81)1.24(1.07-2)-0.4340.664TC(mmol/L)3.62(3.15–4.34)3.95(2.15–4.35)-0.1830.855GLU(mmol/L)5.49(4.67–5.9)5.21(4.73–5.8)-0.0560.955UA(µmol/L)293.5 ± 89.44280.84 ± 77.030.6130.542Cr(µmol/L)65.5(60.3-72.65)71.9(64-79.4)-1.8910.059Urea(mmol/L)5.54(4.62–6.18)5.3(4.6–6.4)-0.6190.536Agr(ratio)1.43 ± 0.321.38 ± 0.320.6450.522Glb(g/L)26.07 ± 4.8826.9 ± 4.54-0.7130.479ALB(g/L)35.9 ± 4.6335.88 ± 5.010.0160.987TP (g/L)61.95 ± 7.1662.77 ± 6.11-0.4990.619ALP(U/L)75.31(61.9-88.83)81.77(69.1–91.3)-0.9460.344GGT(U/L)21(15-25.6)21(15-30.4)-0.9580.338AST(U/L)16.66 ± 4.8918.96 ± 5.09-1.8780.065ALT(U/L)13.05(9.48–15.9)12.7(10-16.5)-0.0820.935IBIL(µmol/L)6.74(4.93–7.97)6.7(5–9)-0.4410.659DBIL(µmol/L)3.48(2.5–4.27)3.2(2.5–5.6)-0.5620.574TBA (µmol/L)10.95(7.17–12.28)9.8(7-13.8)-0.5420.588WBC (*10^9/L)6.04(4.85–7.28)5.84(4.42–6.82)-0.6170.537LYM (*10^9/L)1.26 ± 0.451.36 ± 0.43-0.9440.349MC(*10^9/L)0.45 ± 0.150.44 ± 0.170.2990.766NEUT(*10^9/L)4.1(2.83–5.09)3.66(2.62–4.66)-0.7260.468EGC(*10^9/L)0.08(0.04–0.16)0.11(0.05–0.16)-0.5240.600BAS(*10^9/L)0.03(0.02–0.04)0.03(0.01–0.04)-0.1670.867LP(%)20.75(12.15–27.78)20.4(2.1–27.8)-0.3030.762MCP(%)7.35(5.53–8.9)6.1(0.4–8.5)-1.0710.284NEUTP (%)67.35(57.55–72.48)58(37.4–66.4)-2.080.038EGCP(%)1.3(0.52–2.4)1.4(0.04–2.3)-0.2390.811BASP(%)0.4(0.23–0.6)0.4(0.1–0.6)-0.6780.498RBC(*10^12/L)3.88 ± 0.743.63 ± 0.911.2480.217HB (g/L)108.67 ± 28.05101.06 ± 28.231.1010.275PLT(*10^9/L)243.5(206–278)259(214–325)-0.6430.520MTD (cm)5(3.2-6)5.5(4-6.8)-1.4610.144Sex Female5(16.70%)7(18.90%)0.0570.811 Male25(83.30%)30(81.10%)VTT Negative16(53.30%)16(43.20%)0.6760.411 Positive14(46.70%)21(56.80%)PNI Negative23(76.70%)29(78.40%)0.0280.867 Positive7(23.30%)8(21.60%)Chemotherapy No2(6.70%)16(43.20%)11.2810.001 Yes28(93.30%)21(56.80%)CK(Pan) Negative14(46.70%)10(27.00%)2.780.095 Positive16(53.30%)27(73.00%)CK7 Negative22(73.30%)23(62.20%)0.9370.333 Positive8(26.70%)14(37.80%)CK18 Negative23(76.70%)20(54.10%)3.6850.055 Positive7(23.30%)17(45.90%)AFP Negative3(10.00%)11(29.70%)0.070 Positive27(90.00%)26(70.30%)Hepa1 Negative22(73.30%)23(62.20%)0.9370.333 Positive8(26.70%)14(37.80%)Glypican3 Negative22(73.30%)18(48.60%)4.1960.041 Positive8(26.70%)19(51.40%)SALL4 Negative30(100.00%)8(21.60%)0.000 Positive0(0.00%)29(78.40%)CgA Negative23(76.70%)32(86.50%)1.0870.297 Positive7(23.30%)5(13.50%)CD56 Negative28(93.30%)31(83.80%)0.281 Positive2(6.70%)6(16.20%)Syn Negative22(73.30%)26(70.30%)0.0770.782 Positive8(26.70%)11(29.70%)HER2 Negative20(66.70%)25(67.60%)0.0060.938 Positive10(33.30%)12(32.40%)T Stage 1b2(6.70%)0(0.00%)0.197 23(10.00%)9(24.30%)2.3120.128 33(10.00%)5(13.50%)0.722 4a20(66.70%)19(51.40%)1.5970.206 4b2(6.70%)4(10.80%)0.684N stage 09(30.00%)9(24.30%)0.2720.602 16(20.00%)8(21.60%)0.0260.871 25(16.70%)8(21.60%)0.2600.610 310(33.30%)12(32.40%)0.0060.938M stage 028(93.30%)37(100.00%)0.197 12(6.70%)0(0.00%) Comparison of clinical characteristics between survivors and nonsurvivors among HAS patients during 3 years of follow-up Kaplan-Meier survival analysis Kaplan–Meier survival analysis was performed for variables with significant intergroup differences, stratified by binary classification (e.g., chemotherapy vs. no chemotherapy; Glypican-3 positive vs. negative) or by optimal cutoff values determined by the Youden index of ROC curves (e.g., fPSA, Fe). Survival curves were generated, and differences were tested via the log-rank test. Significant survival differences were observed across all subgroup analyses (p < 0.05) (Fig. 3). For example, patients who received chemotherapy had a significantly higher 3-year survival rate than those who did not receive chemotherapy did; SALL4-negative patients had better survival than SALL4-positive patients did; and patients whose fPSA levels were below the optimal cutoff had higher survival than those whose fPSA levels were above the cutoff. Fig. 3Kaplan–Meier overall-survival curves for HAS patients stratified by risk factors dichotomized either by binary classification or by the optimal cutoff derived from receiver operating characteristic (ROC) analysis. A: Chemotherapy (received/none); B: CA199(Cutoff value: 67.05); C: ferritin (Cutoff value: 87.66); D: Fe (8.07); E: TSGF (Cutoff value: 67.6); F: fPSA (Cutoff value:0.42); G: CYFRA21-1 (Cutoff value: 2.61); H: NSE (Cutoff value: 3.2); I: FTPSA(Cutoff value:0.32); J: NEUTP (Cutoff value: 66.55); K: glypican-3 (Positive/Negative); L: SALL4 (Positive/Negative) Kaplan–Meier overall-survival curves for HAS patients stratified by risk factors dichotomized either by binary classification or by the optimal cutoff derived from receiver operating characteristic (ROC) analysis. A: Chemotherapy (received/none); B: CA199(Cutoff value: 67.05); C: ferritin (Cutoff value: 87.66); D: Fe (8.07); E: TSGF (Cutoff value: 67.6); F: fPSA (Cutoff value:0.42); G: CYFRA21-1 (Cutoff value: 2.61); H: NSE (Cutoff value: 3.2); I: FTPSA(Cutoff value:0.32); J: NEUTP (Cutoff value: 66.55); K: glypican-3 (Positive/Negative); L: SALL4 (Positive/Negative) Cox regression risk factor analysis Cox regression analysis was performed by including the above significant variables in the Cox proportional hazards model, which were entered as either continuous or binary variables. Univariate analysis revealed that chemotherapy, Glypican-3, SALL4, CA19-9, ferritin, TSGF, fPSA, CYFRA21-1, NSE, F/TPSA, Fe, and NEUT% were significantly associated with 3-year mortality in HAS patients (all p < 0.05)(Fig. 4, A). After adjustment for age, sex, and TNM stage, multivariate Cox regression revealed that chemotherapy, SALL4, TSGF, and fPSA (or FTPSA) remained significantly associated with 3-year mortality (all p < 0.05) (Fig. 4, B and C). For KM curves, continuous variables were dichotomized using optimal cutoffs to facilitate visual group comparisons. In Cox regression, they were analyzed as continuous variables to preserve statistical power and avoid information loss. Sensitivity analyses using continuous-variable Cox models confirmed consistent results. The results of the VIF analysis confirmed that there was no significant multicollinearity among the included variables (all the VIFs were < 2.0) (Supplementary Table 1). Fig. 4Forest plots showing Cox regression results for death-related risk factors in HAS patients. A: Unadjusted; B: Adjusted for age and sex; C: Adjusted for age, sex, and TNM stage Forest plots showing Cox regression results for death-related risk factors in HAS patients. A: Unadjusted; B: Adjusted for age and sex; C: Adjusted for age, sex, and TNM stage Construction and optimization of the nomogram prediction model On the basis of these independent risk factors, a nomogram prediction model was developed to estimate the 3-year survival probability of HAS patients. In the cohort of 67 HAS patients, the model demonstrated excellent predictive accuracy, with an AUC of 0.829 (95% CI: 0.723–0.935) in 1-Year, 0.948 (95% CI:0.899–0.997) in 2-Year, and 0.954 (95% CI: 0.902–1.000) in 3-Year (Fig. 5, A and B). The C-index was 0.726, the calibration and DCA plots for years 1–3 are shown in Supplementary Fig. 1. Fig. 5Predictive performance of nomograms constructed with different numbers of risk factors, as evaluated by ROC curves. Nomogram (A) and ROC curve (B) based on all statistically significant risk factors. Nomogram (C) and ROC curve (D) based on variables with p < 0.01. Nomogram (E) and ROC curve (F) based on the three most important factors selected by the relative score in the nomogram Predictive performance of nomograms constructed with different numbers of risk factors, as evaluated by ROC curves. Nomogram (A) and ROC curve (B) based on all statistically significant risk factors. Nomogram (C) and ROC curve (D) based on variables with p < 0.01. Nomogram (E) and ROC curve (F) based on the three most important factors selected by the relative score in the nomogram To enhance clinical applicability, a simplified nomogram was constructed using six risk factors with p < 0.01, yielding an AUC of 0.747 (95%CI: 0.614–0.879) in 1-Year, 0.866 (95%CI: 0.777–0.954) in 2-Year, and 0.945 (95% CI: 0.894–0.996) in 3-Year, with no significant loss of predictive power (Fig. 5, C and D). The C-index was 0.628, the calibration and DCA plots for years 1–3 are shown in Supplementary Fig. 2. Further reduction to three core factors (SALL4, fPSA, and Fe) produced a minimal nomogram, which retained strong predictive accuracy (1-Year: AUC = 0.774, 95%CI: 0.646–0.901; 2-Year: AUC = 0.869, 95%CI: 0.784–0.954; 3-Year: AUC = 0.946, 95% CI: 0.897–0.994) and offered greater practicality for rapid clinical assessment (Fig. 5, E and F). The C-index was 0.649, calibration and DCA plots for years 1–3 are shown in Supplementary Fig. 3. Discussion This study systematically investigated the underlying mechanisms driving differences in OS between patients with HAS and those with non-HAS GC. Our results indicate that differences in TNM stage distribution between the two groups may represent the major confounding factor underlying survival disparities. Since HAS exhibits more aggressive biological behavior, the greater proportion of patients with stage III–IV disease at diagnosis partly explains the poorer prognosis observed in this population. Within the HAS cohort, further survival analysis demonstrated that SALL4 expression, serum Fe, and fPSA were significantly correlated with OS. A Cox proportional hazards model and a nomogram incorporating these independent risk factors achieved high predictive accuracy, offering a quantitative tool for individualized prognostic assessment in HAS patients. Importantly, this study addressed the ongoing controversy regarding survival differences between HAS patients and non-HAS patients, highlighting that matching for TNM stage is a key determinant of the discrepant findings reported in previous studies. Huang et al. reported significant differences in TNM stage distribution between HAS patients and non-HAS patients, particularly in the N and M categories, where HAS patients accounted for a markedly greater proportion of advanced cases (N2-3 stage: 72.3% [34/47] vs. 40.4% [57/141], p < 0.001; M1 stage: 27.7% [13/47] vs. 1.4% [2/141], p < 0.001). According to their multivariate analysis, both M stage (p = 0.027) and HAS histology were independent prognostic factors, suggesting that the poorer outcomes of HAS patients were largely attributable to more advanced TNM staging at diagnosis [21]. Lin et al. in a retrospective analysis of 315 HAS patients who underwent radical resection, further emphasized the prognostic importance of lymph node metastasis, identifying perineural invasion, preoperative carcinoembryonic antigen ≥ 5 ng/mL, and pN3b stage as independent risk factors for OS [22]. Notably, they reported that synchronous liver metastasis occurred more frequently in pure HAS than in mixed HAS, implying that the adverse effect of M stage on prognosis may be mediated primarily through hepatic metastasis. Similarly, Xiang et al. identified preoperative serum AFP levels of 200–500 ng/mL and stage III/IV disease as independent risk factors for OS [23]. However, our study was limited by its retrospective design, small sample size (n = 68), and lack of stratification by HAS subtype. Furthermore, the very small number of M1 cases (n = 2) reduced the statistical power, preventing definitive conclusions about the prognostic role of TNM stage in HAS. Although serum AFP is a hallmark biomarker of HAS and was significantly elevated in our HAS cohort, it was not retained as an independent prognostic factor in the multivariate Cox regression model (p = 0.975 in univariate analysis for 3-year mortality; Table 3). This finding is consistent with previous reports suggesting that AFP levels, while diagnostically characteristic, may not independently predict survival after adjustment for TNM stage and other clinicopathological factors [23]. The prognostic value of AFP in HAS remains controversial, with some studies identifying high preoperative AFP as a risk factor [23], while others report no significant association [24]. In our cohort, the lack of independent prognostic significance may reflect the wide inter-individual variability in AFP levels and the dominant effect of TNM stage on survival outcomes. This study further examined the prognostic impact of immunohistochemical markers on survival outcomes in patients with HAS. Our results demonstrated that Glypican-3 and SALL4 expression was significantly associated with OS in HAS patients. Glypican-3 is a heparan sulfate proteoglycan located on the cell surface that contributes to cell proliferation and malignant transformation. In HCC, glypican-3 has been extensively investigated because its serum expression and immunogenic properties render it both a specific biomarker and a potential target for immunotherapy. In GC, glypican-3-positive tumors represent a distinct molecular subtype that is often associated with intestinal differentiation. Ma et al. reported that glypican-3 positivity is more frequently observed in adenocarcinomas with hepatoid or enteric differentiation, advanced T3–T4 invasion, and stage III–IV disease. The Kaplan–Meier survival analysis further demonstrated significantly worse OS in the glypican-3-positive group than in the glypican-3-negative group (p < 0.001) [25], which is consistent with our findings. In addition, our study revealed the significant impact of SALL4 expression on the survival outcomes of HAS patients. SALL4 is a protein-coding gene whose transcription factor product plays a crucial role in embryonic development and the self-renewal of hematopoietic stem cells. In early research, Ushiku et al. identified SALL4 as a stem cell marker and suggested its significance as a carcinoembryonic protein that is comparable to that of AFP and glypican-3. They proposed that SALL4 reflects fetal intestinal differentiation in GC and serves as a sensitive marker of AFP-producing tumors [26]. Yang et al. conducted immunohistochemical analysis in 139 HAS patients and reported a high SALL4 positivity rate (85.9%). However, Kaplan–Meier analysis revealed no significant association between SALL4 expression and OS (p = 0.368) [24]. In contrast, our findings demonstrated that positive SALL4 expression was significantly associated with poorer OS in HAS patients. Similarly, Wang et al. analyzed HAS-phenotype gastric adenocarcinomas and concluded that immunohistochemical SALL4 positivity was associated with poor prognosis [27]. These results are consistent with our findings, reinforcing the potential role of SALL4 as an important prognostic biomarker in HAS patients. To date, no studies have directly investigated the relationship between serum iron levels and prognosis in patients with HAS. Our study provides the first evidence suggesting a potential association between serum iron levels and OS in this population. In patients with GC, preoperative iron deficiency anemia and postoperative malnutrition have been reported to adversely affect prognosis. Serum iron levels are often significantly lower in patients with gastrointestinal tumors than in healthy controls, possibly due to duodenal bypass, impaired digestive function, chronic blood loss, or malabsorption secondary to hypochlorhydria [28]. Enteral nutritional support has been shown to improve outcomes in GC patients [29]. However, findings remain inconsistent: Sawayama et al. reported that low serum iron levels (cutoff of 60 µg/dL) were not significantly associated with disease-free survival (DFS) or OS [30], whereas Yang et al. reported that baseline serum iron was an independent predictor of OS in GC patients treated with PD-1 therapy (p = 0.001) [31]. Interestingly, Qin et al. demonstrated that alcohol dehydrogenase iron containing 1 (ADHFE1) was upregulated in GC and associated with lymph node metastasis, liver metastasis, and poor prognosis [32]. As ADHFE1 encodes an iron-dependent enzyme involved in energy metabolism, its dysregulation may provide a mechanistic link between altered iron metabolism and poor prognosis in HAS patients. In summary, our findings suggest that serum iron levels, together with ferritin levels, may serve as important prognostic indicators in patients with HAS. Serum iron is influenced by multiple physiological and pathological factors, including nutritional status, chronic inflammation (via hepcidin-mediated regulation), occult gastrointestinal bleeding, and tumor-related metabolic reprogramming [29, 33]. Although we adjusted for key confounders including hemoglobin, albumin, and inflammatory markers in multivariate analysis, residual confounding cannot be entirely excluded. The observed association between lower serum iron and poorer survival may reflect either a direct role of iron metabolism in tumor progression or an indirect marker of advanced disease burden and systemic inflammation. Further mechanistic studies are needed to clarify this relationship. At present, no studies have investigated the relationship between fPSA and survival outcomes in patients with HAS. Although PSA is traditionally regarded as exclusive to prostatic tissue, highly sensitive detection methods have demonstrated its expression in various normal and pathological female tissues and body fluids, including healthy breast tissue, breast tumors, nipple aspirate, and thyroid tissue [34]. Furthermore, studies have identified specific PSA expression in the tumor-related neovasculature of patients with HCC, where it serves as a prognostic marker and potential therapeutic vascular target [35, 36], suggesting its biological relevance to hepatoid differentiation. In prostate cancer, fPSA has been linked to tumor detection, therapeutic resistance, disease staging, and survival prognosis [37–40]. Elevated fPSA levels have also been reported in female patients with breast cancer, where it may serve as a marker for monitoring treatment response [41]. In Chinese populations, fPSA has shown diagnostic value in gastrointestinal tumors, advanced colorectal cancer, and lung cancer [42–44]. Our findings suggest that fPSA may hold prognostic significance in HAS patients, a relationship that warrants validation in larger cohorts. fPSA levels merit consideration, particularly given the inclusion of female patients in this cohort. Among the 67 patients with HAS, 12 (17.9%) were female, and fPSA was detectable in all of them. A statistically significant difference in fPSA levels between sexes was observed both in the general population and within the HAS cohort (t = − 14.862, p < 0.001). Importantly, the distribution of sex was similar between the survivor and deceased groups, which likely reduced the confounding effect of sex on the association between fPSA and OS. Future studies should also explore potential sex-related differences in the predictive value of fPSA. There are several limitations in the design and implementation of this study that may influence the generalizability and accuracy of the findings. First, the HAS cohort comprised only 67 patients, and certain subgroup analyses (e.g., specific TNM substages or biomarker stratifications) involved even smaller sample sizes, potentially resulting in insufficient statistical power and an increased risk of type II errors. This limitation compromises the robustness of our findings and restricts the ability to conduct detailed subgroup analyses, such as sex-specific differences. Second, this was a single-center study based on a Chinese population, which may restrict the generalizability of the findings. Differences in pathological characteristics, treatment responses, and prognoses across regions and ethnic groups suggest that extrapolation should be undertaken with caution. Third, the limited pathological data prevented a detailed distinction between simple and mixed HAS subtypes. Given their potentially distinct biological behaviors and prognostic implications, the lack of stratification may have obscured important subgroup effects. To address these limitations, future studies should adopt multicenter, large-sample designs to enhance robustness and external validity. Moreover, high-level evidence from meta-analyses is essential for synthesizing existing data and providing more comprehensive and reliable conclusions. In conclusion, this study demonstrated that the predominance of advanced TNM stages in HAS patients was significantly associated with poor survival outcomes. Survival outcomes did not differ significantly between HAS patients and non-HAS patients with comparable TNM stage. Independent prognostic factors identified in HAS patients included SALL4, serum Fe, and fPSA. Predictive models incorporating these factors demonstrated high accuracy in estimating 3-year overall survival. These findings offer new insights into prognostic evaluation in HAS; however, validation in larger, multicenter, and more diverse cohorts is warranted. Supplementary Information Supplementary Material 1. Supplementary Figure 1. Calibration curves (A) and Decision curve analysis (B) of the nomogram incorporating all statistically significant risk factors for predicting overall survival at 1, 2, and 3 years. The x-axis indicates the predicted survival probability, and the y-axis indicates the actual observed survival probability estimated by the Kaplan-Meier method. Supplementary Material 1. Supplementary Figure 1. Calibration curves (A) and Decision curve analysis (B) of the nomogram incorporating all statistically significant risk factors for predicting overall survival at 1, 2, and 3 years. The x-axis indicates the predicted survival probability, and the y-axis indicates the actual observed survival probability estimated by the Kaplan-Meier method. Supplementary Material 2. Supplementary Figure 2. Calibration curves (A) and Decision curve analysis (B) of the nomogram based on variables with p < 0.01 for predicting overall survival at 1, 2, and 3 years. Supplementary Material 2. Supplementary Figure 2. Calibration curves (A) and Decision curve analysis (B) of the nomogram based on variables with p < 0.01 for predicting overall survival at 1, 2, and 3 years. Supplementary Material 3. Supplementary Figure 3. Calibration curves (A) and Decision curve analysis (B) of the nomogram constructed using the top three factors ranked by relative importance scores for predicting overall survival at 1, 2, and 3 years. Supplementary Material 3. Supplementary Figure 3. Calibration curves (A) and Decision curve analysis (B) of the nomogram constructed using the top three factors ranked by relative importance scores for predicting overall survival at 1, 2, and 3 years. Supplementary Material 4. Supplementary table 1. Assessment of multicollinearity using the variance inflation factor (VIF) for covariates in the Cox regression model. Supplementary Material 4. Supplementary table 1. Assessment of multicollinearity using the variance inflation factor (VIF) for covariates in the Cox regression model.