A Non-Endoscopic Dynamic Nomogram for Atrophic Gastritis Risk Stratification Based on Serum Indicators and Clinical Factors

Abstract

Background

Atrophic gastritis (AG) is a key precancerous lesion in gastric carcinogenesis, underscoring the need for non-endoscopic screening tools to improve early detection. This study aimed to develop a predictive model for AG based on serum indicators and clinical factors.

Methods

This single-center retrospective study enrolled 520 patients who underwent gastroscopy with biopsy at Shaoxing Second Hospital between January 2023 and December 2025. AG severity was classified via the Kimura-Takemoto system (no/mild: C0-C1, n=349; moderate/severe: C2-C3 and O-type, n=171). Multivariate analysis was performed to identify independent predictors. The performance of the resulting predictive model was evaluated, and decision curve analysis was used to evaluate the clinical utility across a range of threshold probabilities. An interactive dynamic nomogram was created for real-time risk calculation.

Results

Multivariate analysis identified four independent predictors of moderate/severe AG: older age (OR = 1.082, P < 0.001), elevated neutrophil count (NE) (OR = 2.036, P = 0.043), decreased fibrinogen (FIB) (OR = 0.56, P < 0.001), and elevated carcinoembryonic antigen (CEA) (OR = 1.171, P = 0.011). The predictive model demonstrated good discrimination in the development cohort (AUC = 0.780) and was further evaluated in an independent 94-patient inter-campus cohort (AUC = 0.705), confirming its discriminatory ability within the same hospital system.

Conclusion

This non-endoscopic model incorporating age, NE, FIB, and CEA provides a practical tool for AG risk stratification, As a single-center retrospective model, it requires further multi-center prospective validation before clinical implementation; nevertheless, it may help guide endoscopic surveillance decisions and contribute to the early prevention of gastric cancer.

Graphical Abstract

Introduction

Atrophic gastritis (AG), defined by the loss of gastric glands and their replacement by fibrous tissue or intestinal-type epithelium, is a critical precursor lesion in the well-established multistep pathway of gastric carcinogenesis.¹ The early detection and effective management of AG are essential for interrupting the progression to malignancy.² Currently, the diagnosis of AG relies primarily on endoscopic examination and confirmatory histopathological analysis, an invasive approach often hampered by suboptimal patient compliance and potential risks of complications.^3–5 Consequently, there is a pressing need for practical, non-endoscopic screening tools that leverage cheap, fast, and widely accessible hospital data to facilitate the identification of individuals at elevated risk, particularly in resource-limited settings where advanced serological panels or endoscopy may not be readily available.

Chronic inflammation is a central driver in the tumorigenic process, and systemic inflammatory markers derived from routine blood tests—such as the neutrophil count (NE), fibrinogen (FIB), and the neutrophil-to-lymphocyte ratio (NLR)—have shown promise as accessible and cost-effective indicators associated with GC risk.^6–9 Existing clinical models for predicting AG have largely incorporated factors like age, Helicobacter pylori infection status, and pepsinogen ratios.^10–12 However, while useful, these models often exhibit limitations in predictive accuracy and general applicability. Failing to integrate dynamic, physiological indicators of real-time inflammatory status, these models lack the capacity to accurately represent the ongoing pathological processes driving AG progression.¹⁰

Therefore, identifying accurate, non-endoscopic, and readily accessible approaches using routine hospital data holds significant practical importance for refining AG risk prediction. This study aimed to develop a practical, non-endoscopic risk assessment model for AG screening based on serum indicators and clinical factors—parameters that are inexpensive, rapidly obtained, and available in virtually all healthcare settings. By improving early detection at the precancerous stage, this work seeks to contribute to broader strategies for the prevention of GC.

Methods

Study Population and Data Collection

This retrospective study was conducted using data from patients who underwent upper gastrointestinal endoscopy at Shaoxing Second Hospital between January 2023 and December 2025. Electronic medical records were reviewed to collect demographic, clinical, endoscopic, and laboratory data. Fasting venous blood samples were collected on the morning of endoscopy. Routine laboratory parameters were measured using standard automated analyzers and corresponding assay methods at our institution’s central laboratory, with appropriate specimen types (eg, EDTA, citrate, and serum separator tubes) used according to each test’s requirements. The study protocol was approved by the hospital’s Ethics Committee (ethics approval number: 2026-011-001). Due to the retrospective nature of the study, the requirement for informed consent was waived.

Inclusion and Exclusion Criteria

Patients were included if they: (1) underwent gastroscopy with biopsy; (2) had complete baseline demographic and laboratory data, including complete blood count, serum inflammatory markers, and tumor markers; and (3) were aged 18 years or older. Exclusion criteria were: (1) history of gastric cancer or prior gastric surgery; (2) active gastrointestinal bleeding, acute infection, or severe systemic inflammatory disease at the time of endoscopy; (3) use of proton pump inhibitors, antibiotics, or non-steroidal anti-inflammatory drugs within four weeks prior to endoscopy; (4) diagnosis of other malignancies; and (5) incomplete or missing pathological results.

Endoscopic and Pathological Evaluation

All endoscopic procedures were performed by experienced gastroenterologists using high-definition white-light endoscopy. Gastric atrophy was visually assessed and classified according to the Kimura-Takemoto classification system, which categorizes atrophy as closed-type (C1, C2, C3) or open-type (O1, O2, O3).¹³ Unlike the OLGA/OLGIM staging systems, which require extensive histological sampling (at least five targeted biopsies), the Kimura-Takemoto classification relies on real-time endoscopic visual assessment. This system is straightforward, highly reproducible, and widely used in routine clinical practice, particularly in East Asia, with minimal additional cost or time. For analytical purposes, patients were grouped into those with no/mild atrophy (C0 and C1) and those with moderate/severe atrophy (C2, C3, and O-type). At least two biopsy specimens were obtained from the antrum and corpus per the updated Sydney System. All slides were reviewed by pathologists, and histopathological examination confirmed the diagnosis of chronic atrophic gastritis in patients classified as having atrophy by the Kimura-Takemoto system.

Model Feature Selection

The selection of predictor variables was performed through a systematic two-step process. First, univariate logistic regression analyses were conducted to evaluate the association between each potential predictor and the presence of moderate to severe atrophy. Variables demonstrating statistical significance with a threshold of P < 0.05 in these initial analyses were considered candidates for inclusion in the multivariate model. Subsequently, these candidate variables were entered into a multivariable logistic regression model using backward elimination method to identify independent predictors.

Model Development and Assessment

Based on the regression coefficients derived from the multivariate analysis, a predictive model was constructed. A static nomogram was developed to visualize the model. To enhance clinical utility, a dynamic web-based nomogram was created allowing real-time probability estimation via an interactive interface. Model discrimination was evaluated by the area under the receiver operating characteristic curve (AUC). Calibration was assessed using a calibration plot with 500 bootstrap resamples to calculate the bias-corrected curve and mean absolute error. Clinical usefulness was quantified by decision curve analysis to evaluate the net benefit across a range of threshold probabilities. An independent internal validation cohort of 94 patients from another branch of the hospital, enrolled from January 2025 to December 2025, was used to assess model generalizability.

Statistical Analysis

All statistical analyses were performed using R software (version 4.3.1). Continuous variables were expressed as mean ± standard deviation and compared using the Student’s t-test or Mann–Whitney U-test, as appropriate. Categorical variables were presented as frequencies and percentages and compared using the Chi-square test. Variables with a P < 0.05 in univariate logistic regression were entered into multivariate logistic regression to identify independent predictors of moderate to severe atrophy.

Results

Baseline Characteristics of the Study Population

After strictly applying the inclusion and exclusion criteria, among the 520 subjects included in the analysis, 263 (50.6%) were males and 257 (49.4%) were females. The subtypes of atrophic gastritis in the study subjects were as follows; C0 (n = 204, 39.2%), C1 (n = 145, 27.9%), C2 (n = 67, 12.9%), C3 (n = 42, 8.08%), and O (n = 62, 11.9%). The subgroups of atrophic gastritis were as follows: those with no or mild atrophy (C0 and C1; n = 349, 67.1%) and those with moderate to severe atrophy (C2, C3, and O; n = 171, 32.9%). The flowchart is shown in the Figure 1.

Figure 1.

The flowchart of this study.

The analysis of baseline characteristics in Table 1 revealed significant differences between individuals with moderate to severe atrophy (n = 171) and those with no or mild atrophy (n = 349). Patients in the moderate to severe atrophy group were significantly older (63.93 ± 8.59 vs 55.08 ± 12.35 years, P < 0.001) and had a higher proportion of males (57.3% vs 47.3%, P = 0.04). Key inflammatory markers were notably elevated in the atrophy group, including the absolute neutrophil count (NE: 3.53 ± 1.34 vs 3.09 ± 0.66, P < 0.001), neutrophil-to-lymphocyte ratio (NLR: 2.45 ± 1.84 vs 1.97 ± 0.84, P < 0.001), and systemic immune-inflammation index (SII: 515.38 ± 505.05 vs 421.13 ± 190.38, P = 0.002). Conversely, serum albumin (ALB) levels were significantly lower (41.05 ± 4.48 vs 42.31 ± 3.53 g/L, P = 0.001). The tumor marker CEA was also significantly higher in the atrophy group (3.01 ± 2.45 vs 2.15 ± 1.34 ng/mL, P < 0.001). In the internal validation cohort, a total of 94 patients were enrolled from another campus of the hospital. Among them, 36 patients were classified as having no or mild atrophy, and 58 patients were classified as having moderate to severe atrophy. The baseline characteristics of the validation cohort are presented in Supplementary Table 1.

Table 1.

Baseline Characteristics of the Study Subjects (N = 520)

Characteristics	No or Mild Atrophy (N = 349)	Moderate to Severe Atrophy (N = 171)	P Value
Gender (%)
Male	165 (47.3)	98 (57.3)	0.040
Female	184 (52.7)	73 (42.7)
Age	55.08 ± 12.35	63.93 ± 8.59	<0.001
Smoking (%)
No	110 (75.9)	306 (81.6)	0.179
Yes	35 (24.1)	69 (18.4)
Alcohol (%)
No	114 (78.6)	300 (80.0)	0.819
Yes	31 (21.4)	75 (20.0)
Hypertension (%)
No	99 (68.3)	254 (67.7)	0.989
Yes	46 (31.7)	121 (32.3)
WBC (*10^9/L)	5.63 ± 1.45	5.72 ± 1.65	0.529
NE (*10^9/L)	3.09 ± 0.66	3.53 ± 1.34	<0.001
LYM (*10^9/L)	1.75± 0.59	1.68 ± 0.59	0.207
Monocyte (*10^9/L)	0.35 ± 0.13	0.35 ± 0.14	0.774
PLT (*10^9/L)	216.41 ± 48.29	204.57 ± 67.65	0.022
CRP (mg/L)	2.00 ± 3.98	2.62 ± 5.25	0.134
ALB (g/L)	42.31 ± 3.53	41.05 ± 4.48	0.001
FIB (g/L)	3.05 ± 0.76	2.90 ± 0.91	0.045
SII	421.13 ± 190.38	515.38 ± 505.05	0.002
NLR	1.97 ± 0.84	2.45 ± 1.84	<0.001
PLR	135.40 ± 49.59	141.22 ± 105.04	0.391
LMR	5.45 ± 2.16	5.37 ± 3.11	0.729
NPR	0.01 ± 0.00	0.02 ± 0.01	<0.001
PAR	5.14 ± 1.21	5.06 ± 1.98	0.566
CAR	0.05 ± 0.11	0.07 ± 0.13	0.100
CALLY	11.12 ± 12.55	10.34 ± 10.78	0.482
CEA (ng/mL)	2.15 ± 1.34	3.01 ± 2.45	<0.001
CA199 (U/mL)	10.01 ± 10.91	8.31 ± 9.84	0.086
CA724 (IU/mL)	4.37 ± 11.80	3.59 ± 10.17	0.460

Notes: Data are presented as mean ± SD or median (IQR) as appropriate. * indicates multiplication sign.

Abbreviations: N, number of subjects; P, P value for comparison between groups; WBC, white blood cell; NE, neutrophilic granulocyte; LYM, lymphocyte; PLT, platelet; CRP, C-reactive protein; ALB, Albumin; FIB, fibrinogen; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; LMR, lymphocyte-to-monocyte ratio; NPR, neutrophil-to-platelet ratio; PAR, platelet-albumin ratio; CAR, C-reactive protein to albumin ratio; CALLY, C-reactive protein-albumin-lymphocyte; CEA, carcinoembryonic antigen; CA199, carbohydrate antigen199; CA724, carbohydrate antigen 724.

Model Variables Selection

Univariate logistic regression identified several significant factors associated with the extent of AG, including age, NE, FIB, and CEA. Subsequent multivariate analysis confirmed these as independent predictors (Table 2 and Table 3). Specifically, age (OR = 1.082, 95% CI: 0.012–6.307, P < 0.001) and CEA (OR = 1.171, 95% CI: 0.062–2.532, P = 0.011) were independent risk factors, while FIB was a significant protective factor (OR = 0.563, 95% CI: −3.900–0.147, P < 0.001). NE was retained in the model as an independent risk factor (OR = 2.036, 95% CI: 0.351–2.024, P = 0.043). Therefore, the final model features selected through the analysis were age, NE, FIB and CEA. Collinearity assessment showed no severe multicollinearity among predictors (VIF: age = 1.12, NE = 1.06, FIB = 1.12, CEA = 1.03; all < 1.13). The sample size was adequate, with 171 events and 4 predictors resulting in an Events Per Variable (EPV) of 42.8, which far exceeds the recommended minimum of 10.

Table 2.

The Results of Univariate Logistic Regression Analysis

Variables	OR	95% CI Low	95% CI High	P Value
Gender	1.497	1.037	2.169	0.032
Age	1.080	1.059	1.103	<0.001
Hypertension	1.667	1.135	2.449	0.009
Smoking	1.432	0.918	2.233	0.114
Alcohol	1.181	0.755	1.847	0.467
WBC	1.039	0.921	1.171	0.529
NE	1.612	1.325	1.981	<0.001
LYM	0.814	0.586	1.114	0.207
Monocyte	1.218	0.312	4.594	0.773
PLT	0.996	0.992	0.999	0.022
CRP	1.030	0.990	1.078	0.157
ALB	0.921	0.878	0.965	<0.001
FIB	0.785	0.616	0.991	0.046
SII	1.001	1.000	1.002	0.008
NLR	1.394	1.179	1.672	<0.001
PLR	1.001	0.999	1.004	0.400
LMR	0.987	0.911	1.061	0.729
NPR	1.798	1.474	2.220	<0.001
PAR	0.964	0.844	1.089	0.566
CAR	3.571	0.754	21.941	0.127
CALLY	0.994	0.978	1.010	0.482
CEA	1.302	1.167	1.465	<0.001
CA199	0.983	0.963	1.002	0.089
CA724	0.992	0.961	1.010	0.481

Abbreviations: OR, Odds Ratio; CI, Confidence Interval; P, P value from univariate logistic regression analysis; WBC, white blood cell; NE, neutrophilic granulocyte; LYM, lymphocyte; PLT, platelet; CRP, C-reactive protein; ALB, Albumin; FIB, fibrinogen; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; LMR, lymphocyte-to-monocyte ratio; NPR, neutrophil-to-platelet ratio; PAR, platelet-albumin ratio; CAR, C-reactive protein to albumin ratio; CALLY, C-reactive protein-albumin-lymphocyte; CEA, carcinoembryonic antigen; CA199, carbohydrate antigen199; CA724, carbohydrate antigen 724.

Table 3.

The Results of Multivariate Logistic Regression Analysis

Variables	OR	β Coefficients	95% CI Low	95% CI High	P Value
Age	1.082	0.078	0.012	6.307	<0.001
Gender	1.135	0.126	0.222	0.570	0.569
Hypertension	1.066	0.064	0.677	1.678	0.782
NE	2.036	0.711	0.351	2.024	0.043
PLT	0.992	−0.008	−1.154	0.007	0.249
ALB	0.986	−0.014	−0.469	0.029	0.639
FIB	0.563	−0.575	−3.900	0.147	<0.001
SII	1.002	0.002	0.002	1.120	0.263
NLR	0.714	−0.337	−1.094	0.308	0.274
NPR	0.000	−0.119	−0.313	0.381	0.754
CEA	1.171	0.158	0.062	2.532	0.011

Abbreviations: β, regression coefficient; OR, odds ratio; CI, confidence interval; P, P value from multivariate logistic regression analysis; WBC, white blood cell; NE, neutrophilic granulocyte; PLT, platelet; ALB, Albumin; FIB, fibrinogen; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; NPR, neutrophil-to-platelet ratio; CEA, carcinoembryonic antigen.

Model Construction and Performance Evaluation

The predictive model was constructed using multivariable logistic regression as follows: Model score = −6.266 + 0.084 × age + 0.590 × NE - 0.617 × FIB + 0.174 × CEA (Table 4). In the development cohort, the model demonstrated favorable discriminatory power, with a receiver operating characteristic (ROC) curve yielding an AUC of 0.780 (95% CI: 0.738–0.822; Figure 2A).

Table 4.

Final Multivariable Logistic Regression Model

Variables	OR	β Coefficients	95% CI Low	95% CI High	P Value
Age	1.087	0.084	1.064	1.113	<0.001
NE	1.805	0.590	1.442	2.286	<0.001
FIB	0.540	−0.617	0.407	0.706	<0.001
CEA	1.190	0.174	1.061	1.343	0.0036

Abbreviations: β, regression coefficient; OR, odds ratio; CI, confidence interval; P, P value from multivariate logistic regression analysis; NE, neutrophilic granulocyte; FIB, fibrinogen; CEA, carcinoembryonic antigen.

Figure 2.

Performance of the multivariable logistic regression model for atrophic gastritis prediction. (A) ROC curve illustrating the diagnostic performance of the model. (B) DCA evaluating the net benefit of the prediction model across different high-risk thresholds.

Abbreviations: NE, neutrophilic granulocyte; FIB, fibrinogen; CEA, carcinoembryonic antigen; AUC, area under the curve; ROC, receiver operating characteristic; DCA, decision curve analysis.

To evaluate the incremental value of NE, FIB, and CEA beyond age alone, we compared the full model against an age-only model. The full model achieved a significantly higher AUC (0.780 vs 0.712; DeLong test, Z = 3.690, P < 0.001; likelihood ratio test, χ² = 52.23, P < 0.001), confirming that the addition of these serum markers significantly improved predictive performance.

To evaluate whether lifestyle factors and comorbidity confounded the association between the four predictors and atrophic gastritis, we compared the final model with an extended model additionally adjusted for hypertension, smoking, and alcohol consumption. The regression coefficients for Age, NE, FIB, and CEA changed by +0.7%, −0.8%, −3.5%, and +0.2%, respectively, all well below the 10% threshold, and the AUC remained essentially unchanged (0.780 vs 0.780; AIC: 545.5 vs 550.4). These results confirm that the four-variable model is robust and that the included additional variables do not act as significant confounders; they were therefore not retained in the final model.

In the internal validation cohort, the model maintained good discrimination with an AUC of 0.705 (95% CI: 0.595–0.814; Supplementary Figure S1). Decision curve analysis (DCA; Figure 2B) demonstrated that the combined model (red line) provided a positive net benefit across a wide range of threshold probabilities, particularly between 0.0 and 0.6, consistently outperforming the “intervene on all” and “intervene on none” strategies as well as models based solely on age, NE, FIB, or CEA. These findings indicate that the integrated model offers clinically useful net benefit for guiding endoscopic surveillance decisions within this practical threshold range.

Construction and Validation of Dynamic Nomogram

Based on the four independent predictors (age, NE, FIB, and CEA) identified from the multivariate analysis, a predictive model was constructed and translated into user-friendly clinical tools, including both a static nomogram and an interactive dynamic nomogram. The static nomogram visually maps individual patient values for each predictor onto a points scale, where the total score corresponds to a predicted probability of moderate to severe atrophy (Figure 3A).

Figure 3.

Nomogram development and internal validation/calibration. (A) The static nomogram for predicting the probability of AG. (B) The dynamic nomogram application interface, demonstrating the calculation process for a specific patient case. (C) Calibration plot for the prediction model.

Abbreviations: NE, neutrophilic granulocyte; FIB, fibrinogen; CEA, carcinoembryonic antigen; AG, atrophic gastritis.

To enhance clinical applicability, a dynamic web-based nomogram was developed (https://modelpredictionzyj.shinyapps.io/DynamicNomogram/), allowing clinicians to input patient data via adjustable sliders and instantly obtain individualized risk predictions upon clicking the “Predict” button (Figure 3B). The calibration of the nomogram, evaluated using bootstrapping with 500 repetitions, demonstrated excellent agreement between predicted and observed outcomes, with a bias-corrected curve closely following the ideal line and a very low mean absolute error of 0.008, indicating highly reliable and well-calibrated predictions (Figure 3C).

Discussion

Atrophic gastritis, a well-established precursor to gastric cancer, represents a critical lesion for early detection and intervention in the gastric carcinogenesis cascade.¹⁴ The diagnosis of AG currently relies on histopathological examination of biopsies obtained during gastroscopy; however, results can be subject to sampling error and variability.¹⁵ Therefore, identifying readily available and non-invasive serum biomarkers to predict the likelihood of AG upon endoscopy holds significant promise for improving early detection rates and optimizing surveillance strategies.¹⁶ While several existing approaches, such as the OLGA and OLGIM staging systems that require extensive histological sampling, or complex molecular panels requiring specialized testing.^17,18 Our study innovates by focusing on a panel of routinely available, inexpensive, and rapidly obtainable serum indicators and clinical factors. Unlike advanced serological panels or complex molecular tests that may not be universally available, these parameters are measured in standard blood tests, making them cheap, fast, and widely accessible in virtually all hospital settings. The intended clinical application of this practical predictive tool is as a pre-endoscopic triage tool for patients with gastrointestinal symptoms or risk factors who have not yet undergone initial endoscopic evaluation, helping to prioritize high-risk individuals for timely endoscopic examination while allowing low-risk individuals to be managed with routine follow-up. This approach assists clinicians in triaging patients for more meticulous examination and biopsy during gastroscopy, especially in resource-limited areas where endoscopy or specialized testing may not be readily available.

The core finding of our study is the identification of age, NE, FIB and CEA as independent predictors for AG. The association of older age with increased AG risk is consistent with the natural history of the disease, which is often a chronic, progressive process frequently linked to Helicobacter pylori infection or autoimmune mechanisms.¹⁹ The positive association of NE with AG may reflect the systemic inflammatory burden associated with progressive gastric mucosal damage. As a hypothesis, chronic inflammation driven by persistent mucosal injury leads to neutrophil recruitment and activation, which may contribute to the oxidative damage and cellular proliferation that characterize the atrophic process.^20,21 However, we acknowledge that causal inferences cannot be drawn from a retrospective cross-sectional design, and this mechanistic interpretation should be considered hypothesis-generating rather than definitive.

A particularly intriguing and potentially novel finding from our analysis is the inverse relationship between FIB levels and the presence of AG. This contrasts with its more typical role as an acute-phase reactant that is often elevated in various inflammatory and neoplastic conditions.²² FIB is not only a coagulation factor but also a potent modulator of inflammation.²³ It is plausible that in the advanced stage of AG, the extensive loss of gastric mucosa and the associated chronic inflammatory milieu may lead to a state of low-grade, persistent consumption of fibrinogen, as a systemic consequence of long-standing chronic gastritis.²⁴ This hypothesis finds some support in emerging studies suggesting that alterations in coagulation parameters may be associated with the progression from precancerous lesions to gastric cancer, positioning hypofibrinogenemia as a potential marker of severe, long-standing mucosal damage rather than acute inflammation.²⁵ Alternatively, this finding may reflect a complex interplay between the coagulation system and the inflammatory response that warrants further investigation. Further longitudinal studies are needed to elucidate the precise role of fibrinogen in the AG cascade.

Furthermore, the inclusion of CEA as an independent predictor is clinically relevant. While CEA is a recognized tumor marker for advanced gastrointestinal cancers, its mild elevation in a non-malignant condition like AG underscores the concept of AG as a field of cancerization.²⁶ The chronic atrophic mucosa may harbor molecular and cellular alterations that lead to low-level CEA expression, making it a surrogate marker for the presence of a high-risk gastric environment.²⁷ Previous studies have reported elevated CEA levels in a subset of patients with benign gastric conditions, including AG, suggesting its utility for risk stratification.²⁸

The present findings should also be considered within the broader landscape of non-invasive biomarker-based risk prediction. Recent advances in cfDNA methylation-based cancer screening have demonstrated the potential of liquid biopsy approaches for early cancer detection.²⁹ In the context of gastric diseases, inflammation-related biomarkers have been increasingly recognized as valuable predictors of disease progression.³⁰ Furthermore, externally validated biomarker-based nomograms have shown promise in improving risk stratification for various gastrointestinal conditions.³¹ Our model complements these efforts by leveraging routinely available and inexpensive serum indicators, which may be particularly advantageous in resource-limited settings where advanced molecular testing is not readily accessible.

Despite these advances, several limitations must be acknowledged. The single-center design may affect generalizability, and the cross-sectional nature precludes assessment of temporal relationships between biomarker changes and disease progression. Additionally, while we included major confounding factors, unmeasured variables such as detailed H. pylori infection status or dietary factors may influence the results and should be integrated into future prospective studies. Therefore, future research should prioritize prospective multicenter validation studies to confirm our findings, as well as the integration of serum markers, including pepsinogen and gastrin, with endoscopic imaging features to develop comprehensive risk stratification algorithms. The development of dynamic monitoring protocols using these markers could further help track disease progression and treatment response.

Conclusion

In conclusion, our study provides preliminary evidence for using a panel of routine serum indicators and clinical factors to predict AG. The developed model, highlighting the roles of age, NE, FIB, and CEA, offers a promising, low-cost tool for risk stratification. Given the single-center retrospective design, prospective, multicenter external validation is required before clinical implementation; the model may help guide targeted endoscopic surveillance and support early gastric cancer prevention strategies.

Funding Statement

The authors declared that financial support was not received for this work.

Data Sharing Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ethics Approval and Consent to Participate

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of the Shaoxing Second Hospital (approval number: 2026-011-001). Given that this was a retrospective study utilizing de-identified data with minimal risk to participants, the Ethics Committee granted a waiver of the requirement to obtain informed consent from individual patients. The RECORD reporting guidelines were followed.

Author Contributions

Ziyi Fang: data curation, writing – original draft, conceptualization, methodology, investigation and supervision. Yunjie Zhang: formal analysis, validation and writing – review & editing. All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no conflicts of interest in this study.