Development and validation of a nomogram incorporating dietary factors for predicting Helicobacter pylori-negative early gastric cancer risk

Abstract

BACKGROUND

Gastric cancer (GC) is one of the most prevalent malignant tumors worldwide and poses a significant threat to human health. Helicobacter pylori (H. pylori)-negative early GC (HpN-EGC) often remains undetected because of its asymptomatic progression.

AIM

To accurately and efficiently identify high-risk HpN-EGC individuals and guide clinical diagnosis and treatment, we developed a clinical prediction model for HpN-EGC.

METHODS

This retrospective case-control study evaluated 593 confirmed H. pylori-negative cases at a hospital. Eligible patients were randomized into training (n = 416) and internal validation (n = 177) groups. Multivariate logistic regression analysis identified significant predictors, which were incorporated into the nomogram. Patients from a different hospital were included in the external validation group (n = 109). Subgroup analyses explored H. pylori eradication (> 1 year) in H. pylori-naive populations.

RESULTS

Risk factors for HpN-EGC were advanced age [odds ratio (OR): 1.13], digestive comorbidities (OR: 17.55), and frequent consumption of smoked and hot foods (OR: 19.00; OR: 4.19). Regular legume and nut intake had protective effects (OR: 0.30; OR: 0.14). The nomogram showed excellent discrimination [training area under the curve (AUC) = 0.904; internal validation AUC = 0.865; external validation AUC = 0.794], stable calibration, and predictive accuracy, with a C-index of 0.904 (95% confidence interval: 0.876-0.931). Good model fit was supported by a non-significant Hosmer-Lemeshow test result (χ² = 7.57, P = 0.477). Subgroup analysis revealed that smoking and alcohol consumption specifically increased the risk in H. pylori-naive patients, whereas legume and nut consumption consistently reduced the risk across subgroups.

CONCLUSION

The HpN-EGC risk prediction tool effectively identifies high-risk individuals based on age, digestive comorbidities, consumption of smoked and hot foods, and legume and nut intake.

Key Words: Early gastric cancer; Helicobacter pylori; Risk factor; Prediction model; Nomogram

Core Tip: Early gastric cancer (EGC), especially Helicobacter pylori (H. pylori)-negative EGC, often goes unnoticed because of the absence of obvious clinical symptoms in patients. Our study explored the risk and protective factors associated with the development of H. pylori-negative EGC, and established a risk prediction model. The model, established based on advanced age, digestive comorbidities, and the frequent consumption of smoked and hot foods, legumes and nuts, has superior accuracy and clinical application value.

INTRODUCTION

Gastric cancer (GC) is a malignant tumor that poses a significant threat to human health. According to the World Health Organization, approximately 968000 cases of GC were diagnosed worldwide in 2022, ranking fifth in cancer incidence and deaths[1]. Early GC (EGC) is limited to the mucosal or submucosal layers regardless of the presence of lymph node metastasis[2]. Most EGC cases can be managed through endoscopic treatment, with a 5-year survival rate of up to 95%, whereas advanced GC has a 5-year survival rate of 30%-70% after surgical treatment[3,4]. Therefore, timely diagnosis and treatment of GC are important factors that affect its prognosis.

Helicobacter pylori (H. pylori) infection is a major risk factor for GC[5]. Although eradication of H. pylori could effectively reduce the risk of GC, tumors can develop even after eradication[6]. Some reports have indicated that H. pylori-negative GC accounts for 0.42%-5.4% of all GC cases[7]. A previous study based on the Japanese population that conducted pathological analysis of the gastric mucosa after gastrectomy found that approximately 2% of GC cases occurred in H. pylori-negative (HpN) patients[8]. Recently, the prevalence of HpN-EGC was 0.66%-10.6% in Asian countries[9-11]. Notably, rising eradication rates of H. pylori have been accompanied by an increase in the proportion of HpN-EGC cases. This shift in epidemiology highlights the need for specific risk-prediction tools for HpN-EGC.

Currently, gastroscopy combined with a gastric mucosal biopsy is the most accurate method for diagnosing EGC. Other screening methods, including serological tests and imaging examinations, such as serum gastrin, gastrin 17, and upper gastrointestinal barium meal examinations, are limited by various factors, including high cost, invasive procedures, and low patient compliance. Epidemiological and statistical data has demonstrated that the occurrence of GC is the result of multiple factors, such as genetics, environment, and dietary habits[12]. Therefore, a convenient and non-invasive prediction model should be established based on the risk factors associated with EGC in H. pylori-negative patients.

Recently, nomograms have demonstrated great potential as risk assessment methods and are widely used as prognostic tools for clinical diseases[13]. With the ability to generate the individual probability of a clinical event by integrating various prognostic and determinant variables, nomograms may support the development of clinically integrated models and personalized medicine[14]. One study developed a nomogram based on the Kyoto Classification of Gastritis to predict the risk of GC[15]. Another innovative nomogram was developed to evaluate the cancer-specific survival of patients with EGC who underwent gastrectomy and provided prognostic evidence[16]. In addition, one study developed deep learning systems using patient pathological images to achieve an accurate diagnosis and prognosis prediction of GC[17]. However, the clinical application of this model is limited by its dependence on invasive tissue acquisition procedures.

To address this limitation, our study introduced a novel nomogram that advances the field in two key aspects. First, by integrating easily obtainable dietary factors as independent predictors, we obtained a more comprehensive view of the multifactorial etiology of EGC. Second, our tool delivered a non-invasive, cost-effective, and precise risk assessment specifically targeting the HpN-EGC sub-population. The study hopes to improve EGC detection rates and subsequently guide clinical decision-making.

MATERIALS AND METHODS

Study participants

A total of 874 patients who underwent gastroscopy and pathological examinations at the Beijing Friendship Hospital, Xicheng Campus between January 2018 and December 2021 were enrolled. A flow diagram of this study is shown in Supplementary Figure 1. The inclusion criteria for cases with HpN-EGC were as follows: (1) All patients aged ≥ 18 years; (2) Patients who underwent endoscopic submucosal dissection and postoperative pathological results confirmed EGC; and (3) Gastric biopsy, postoperative pathology, or the 13C-urea breath test confirmed negative results for H. pylori. The inclusion criteria cases without EGC were as follows: (1) All patients aged ≥ 18 years and agreed to undergo gastroscopy and pathological examination; (2) Pathological results revealed non-EGC; and (3) Gastric biopsy or the 13C-urea breath test confirmed negative results for H. pylori. Patients with a current H. pylori infection, H. pylori eradication within one year, severe comorbidities, mental illness, advanced GC, gastric stumps, metachronous GC, metastatic cancer, or incomplete clinical data were excluded. Patients with a history of gastric resection were also excluded. To mitigate potential methodological limitations, including model bias, evaluation distortion, and overfitting, we established a randomly selected HpN control group matched in sample size to the HpN-EGC cohort.

Following the inclusion and exclusion criteria, 593 eligible participants were enrolled in the study. The patients were randomly allocated in a 7:3 ratio between the training and internal validation datasets. The training set included 209 post-endoscopic submucosal dissection patients with EGC in the HpN-EGC group, and 207 patients with histologically confirmed non-EGC diagnoses in the HpN-control group. Additionally, 109 patients enrolled at the Beijing Friendship Hospital, Tongzhou Campus between March and December 2022 constituted the external validation cohort.

Patients with HpN-EGC were stratified into two groups: The H. pylori eradication EGC group, comprising patients who had successfully completed H. pylori eradication therapy more than 1 year prior, and the H. pylori-naive EGC group, comprising patients with no history of H. pylori infection. The 1-year interval following eradication therapy was established based on clinical evidence reported by Saka et al[18]. We screened the corresponding patients as controls and explored the risk factors for EGC across different backgrounds.

Lifestyle, dietary habits, and clinical indicators assessment

General information (age, sex, body mass index, and educational level) and lifestyle habits, including smoking, drinking, and dietary habits (hard, fried, spicy, smoked, hot, and moldy foods, as well as onion and garlic, legumes, nuts, coarse grains, etc.), were collected. “Frequent” consumption was defined as a dietary intake frequency of ≥ 3 times per week with a serving size of ≥ 50 g; otherwise, it was defined as “occasional”. The threshold setting of ≥ 3 times per week was established with reference to previous literature and was further refined based on the content of the food frequency questionnaire[19]. The threshold setting of > 50 g for dietary quality was based on the common options used in the simplified Food Frequency Questionnaire developed by Gao et al[20], which aimed to avoid significant deviations in both the frequency and quantity of patients’ dietary intake. Additionally, the nut intake threshold was slightly adjusted to 28 g/week[21].

Clinical symptoms (including dysphagia, acid reflux, heartburn, nausea, vomiting, abdominal pain, abdominal distension, hematemesis, melena, marasmus, fatigue, and anemia) and a family history of cancer were also recorded. In addition, we investigated whether the patients had comorbidities such as hypertension, diabetes, cardiac and cerebral diseases (including cerebral hemorrhage, cerebral infarction, and cerebral vascular stenosis), and digestive comorbidities [including chronic atrophic gastritis, gastric ulcers, and gastroesophageal reflux disease (GERD)]. The Charlson comorbidity index (CCI) was used to quantify the comorbidities[22]. Age and GC were not included in the total CCI score. Long-term medication history referred to medication use for > 1 year, including metformin, aspirin, corticosteroids, and proton pump inhibitors.

Statistical analysis

Statistical analysis of data and construction of the nomogram were performed using SPSS (26.0) and R (4.4.1). Continuous variables following a normal distribution were presented as mean ± SD and compared using Student’s t-test, whereas non-normally distributed variables were expressed as median (interquartile range) and analyzed using the Mann-Whitney U test. The assumption of linearity for the continuous variable “age” was assessed using restricted cubic splines with 4 knots. A likelihood ratio test showed no significant departure from linearity (P = 0.53); the spline curve is included in Supplementary Figure 2. Therefore, age was entered into the final model as a linear term. Categorical variables were analyzed using the χ² test or Fisher’s exact test. Logistic regression analysis (forward: Likelihood ratio method) was conducted to explore the risk factors associated with EGC. A P value < 0.05 was considered statistically significant. The nomogram was established using “rms”, “regplot”, and “DynNom” packages. A bootstrap test was conducted to validate the model. The model prediction efficacy was tested using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). The effectiveness and clinical application of the model were further validated using calibration curve, decision curve analysis, and clinical impact curve.

RESULTS

Comparison of characteristics between HpN-EGC and HpN-control groups

A total of 874 patients who underwent gastroscopy and pathological examinations agreed to participate in this study. Based on the inclusion and exclusion criteria, 593 patients were included in the study, with 416 and 177 patients in the training and internal validation sets, respectively (Figure 1). The clinical characteristics of the HpN-EGC and HpN-control groups are summarized in Table 1. Sex, age, smoking, drinking, and educational level were significantly different between the control and HpN-EGC groups (P < 0.001). The proportion of patients with clinical symptoms was significantly higher in the HpN-EGC group than in the HpN-control group (P = 0.001). The number of patients with heartburn (P < 0.001), abdominal pain (P < 0.001), and abdominal distension (P = 0.006) was higher in the HpN-EGC group than in the HpN-control group (Supplementary Table 1). Moreover, the HpN-EGC group exhibited a significantly higher prevalence of hypertension, diabetes, cardiac and cerebral diseases, and digestive comorbidities than the control group (P < 0.001). The proportion of patients consuming metformin or proton pump inhibitors for > 1 year was higher in the HpN-EGC group than in the HpN-control group (P < 0.05). Patients with EGC more frequently consumed foods that are hard, spicy, smoked and hot, as well as onions and garlic; the HpN-control group consumed legumes and nuts more frequently (P < 0.05).

Figure 1 Risk prediction and stratification analysis in Helicobacter pylori-negative early gastric cancer patients. H. pylori: Helicobacter pylori; BFH-XC: Beijing Friendship Hospital, Xicheng Campus; BFH-TZ: Beijing Friendship Hospital, Tongzhou Campus; EGC: Early gastric cancer; Acc: Accuracy.

Table 1 Comparison of clinical characteristics between the Helicobacter pylori-negative early gastric cancer and the Helicobacter pylori-negative-control groups in the training set, n (%).

	HpN-control group (n = 207)	HpN-EGC group (n = 209)	P value
Sex			< 0.001
Male	80 (38.6)	136 (65.1)
Female	127 (61.4)	73 (34.9)
Age, years	53.0 (42.0, 61.0)	63.0 (57.0, 68.0)	< 0.001
BMI (kg/m2)	24.0 (21.8, 26.3)	23.7 (21.5, 25.9)	0.499
Smoking1			< 0.001
Yes	21 (10.1)	101 (48.3)
No	186 (89.9)	108 (51.7)
Drinking1			< 0.001
Yes	33 (15.9)	88 (42.1)
No	174 (84.1)	121 (57.9)
Educational level			< 0.001
High school or above	178 (86.0)	127 (60.8)
Below high school	29 (14.0)	82 (39.2)
Clinical symptoms			0.001
Yes	130 (62.8)	163 (78.0)
No	77 (37.2)	46 (22.0)
Hypertension			< 0.001
Yes	38 (18.4)	77 (36.8)
No	169 (81.6)	132 (63.2)
Diabetes			< 0.001
Yes	9 (4.3)	37 (17.7)
No	198 (95.7)	172 (82.3)
Hyperlipidemia			0.626
Yes	32 (15.5)	36 (17.2)
No	175 (84.5)	173 (82.8)
Cardiac and cerebral diseases			< 0.001
Yes	12 (5.8)	50 (23.9)
No	195 (94.2)	159 (76.1)
Digestive comorbidities			< 0.001
Yes	83 (40.1)	179 (85.6)
No	124 (59.9)	30 (14.4)
Family history of tumors			0.214
Yes	53 (25.6)	65 (31.1)
No	154 (74.4)	144 (68.9)
CCI	0.39 ± 0.70	0.92 ± 1.11	< 0.001
Aspirin2			0.365
Yes	6 (7.9)	14 (12.0)
No	70 (92.1)	103 (88.0)
Clopidogrel2			0.170
Yes	0 (0.0)	5 (4.3)
No	76 (100.0)	111 (95.7)
Metformin2			0.003
Yes	1 (1.3)	16 (13.8)
No	75 (98.7)	100 (86.2)
Corticosteroids2			1.000
Yes	1 (1.3)	2 (1.7)
No	75 (98.7)	115 (98.3)
PPI2			0.020
Yes	10 (4.8)	23 (11.0)
No	197 (95.2)	186 (89.0)
Hard food			< 0.001
Occasional	204 (98.6)	166 (79.4)
Frequent	3 (1.4)	43 (20.6)
Fried food			0.055
Occasional	192 (92.8)	182 (87.1)
Frequent	15 (7.2)	27 (12.9)
Spicy food			< 0.001
Occasional	189 (91.3)	155 (74.2)
Frequent	18 (8.7)	54 (25.8)
Smoked food			< 0.001
Occasional	200 (96.6)	142 (67.9)
Frequent	7 (3.4)	67 (32.1)
Hot food			0.013
Occasional	156 (75.4)	134 (64.1)
Frequent	51 (24.6)	75 (35.9)
Moldy food			1.000
Occasional	207 (100.0)	208 (99.5)
Frequent	0 (0.0)	1 (0.5)
Onion and garlic			0.005
Occasional	46 (22.2)	25 (12.0)
Frequent	161 (77.8)	184 (88.0)
Legumes			< 0.001
Occasional	81 (39.1)	134 (64.1)
Frequent	126 (60.9)	75 (35.9)
Nuts			< 0.001
Occasional	138 (66.7)	174 (83.3)
Frequent	69 (33.3)	35 (16.7)
Coarse grains			0.522
Occasional	135 (65.2)	130 (62.2)
Frequent	72 (34.8)	79 (37.8)
Meat, egg, and milk			0.617
Occasional	16 (7.7)	19 (9.1)
Frequent	191 (92.3)	190 (90.9)
Fruits and vegetables			0.988
Occasional	5 (2.4)	4 (1.9)
Frequent	202 (97.6)	205 (98.1)

¹Smoking or drinking now or having been smoking or drinking for more than 1 year.

²Long term use of drugs refers to using for more than 1 year.

CCI: Charlson comorbidity index; BMI: Body mass index; PPI: Proton pump inhibitors; HpN: Helicobacter pylori-negative; EGC: Early gastric cancer.

Univariate and multivariate logistic regression analysis of the characteristics of HpN-EGC

Features with statistically significant differences were included in univariate and multivariate logistic regression analyses. The results demonstrated that age, digestive comorbidities, smoked and hot foods, legumes, and nuts were associated with the development of HpN-EGC (Table 2). Advanced age [odds ratio (OR): 1.13, 95% confidence interval (CI): 1.05-1.22], digestive comorbidities (OR: 17.55, 95%CI: 4.36-96.25), and frequent consumption of smoked and hot foods (OR: 19.00, 95%CI: 1.75-45.38; OR: 4.19, 95%CI: 1.10-18.41) were independent risk factors for HpN-EGC, whereas legumes and nuts intake (OR: 0.30, 95%CI: 0.09-0.89; OR: 0.14, 95%CI: 0.037-0.48) served as protective factors. Furthermore, bootstrap validation analyses for the significant factors were conducted, with the results shown in Supplementary Table 2. Additionally, the ROC curve for age was used to predict HpN-EGC (Supplementary Figure 3). Age exhibited an AUC of 0.761, and the optimal cutoff value of the age index was 54.5, corresponding to a sensitivity of 85.6% and a specificity of 53.6%, confirming that patients aged > 54.5 years had a high risk of HpN-EGC.

Table 2 Univariate and multivariate analyses of clinical characteristics between the Helicobacter pylori-negative early gastric cancer and the Helicobacter pylori-negative-control groups.

Characteristics	Univariate			Multivariate
	OR	95%CI	P value	OR	95%CI	P value
Sex	0.34	0.23-0.50	< 0.001	0.55	0.12-2.54	0.446
Female vs male
Age, years	1.11	1.08-1.13	< 0.001	1.13	1.05-1.22	0.001
Smoking1	8.28	4.98-14.33	< 0.001	3.79	0.74-21.96	0.119
Yes vs no
Drinking1	3.83	2.44-6.15	< 0.001	2.23	0.47-10.60	0.308
Yes vs no
Educational level	0.25	0.15-0.40	< 0.001	0.55	0.15-1.87	0.342
Above vs below high school
Clinical symptoms	2.10	1.37-3.25	< 0.001	1.76	0.53-6.12	0.359
Yes vs no
Hypertension	2.59	1.66-4.10	< 0.001	0.98	0.30-3.10	0.971
Yes vs no
Diabetes	4.73	2.32-10.70	< 0.001	1.93	0.35-11.82	0.456
Yes vs no
Cardiac and cerebral diseases	5.11	2.72-10.36	< 0.001	3.87	0.93-18.24	0.070
Yes vs no
Digestive comorbidities	8.91	5.60-14.55	< 0.001	17.55	4.36-96.25	< 0.001
Yes vs no
CCI	1.94	1.53-2.50	< 0.001	1.36	0.70-2.86	0.386
Metformin2	12.00	2.37-218.98	0.017	1.60	0.07-76.8	0.778
Yes vs no
PPI2	2.24	1.16-5.48	0.023	2.26	0.47-11.79	0.317
Yes vs no
Hard food	17.61	6.28-73.58	< 0.001	17.97	0.24-205.54	0.992
Frequent vs occasional
Spicy food	3.66	2.10-6.65	< 0.001	1.14	0.24-6.05	0.872
Frequent vs occasional
Smoked food	13.48	6.42-33.06	< 0.001	19.00	1.75-453.77	0.035
Frequent vs occasional
Hot food	1.71	1.12-2.63	0.013	4.19	1.10-18.41	0.043
Frequent vs occasional
Onion and garlic	2.10	1.25-3.62	0.006	4.31	0.95-23.22	0.068
Frequent vs occasional
Legumes	0.36	0.24-0.53	< 0.001	0.30	0.09-0.89	0.033
Frequent vs occasional
Nuts	0.40	0.25-0.64	<0.001	0.14	0.037-0.48	0.003
Frequent vs occasional

¹Smoking or drinking now or having been smoking or drinking for more than 1 year.

²Long term use of drugs refers to using for more than 1 year.

CCI: Charlson comorbidity index; PPI: Proton pump inhibitors; OR: Odds ratio; CI: Confidence interval.

Construction and application of the risk prediction model

Based on the results of the multivariate logistic regression analysis, six independent influencing factors (age, digestive comorbidities, smoked foods, hot foods, legumes, and nuts) were included in the risk prediction model. The assignments of the variables included in the model are presented in Supplementary Table 3. A nomogram was created to visualize the prediction model. Numerous variables included in the nomogram were incorporated to obtain total scores and, correspondingly, the risk probability of HpN-EGC (Figure 2A). The risk prediction cut-off for the model was 0.585 (Supplementary Figure 4). Patients exceeding this probability threshold have an elevated risk of developing GC and warrant further diagnostic evaluation.

Figure 2 The nomogram risk predictive model for Helicobacter pylori-negative early gastric cancer. A: The nomogram risk predictive model based on “regplot”; B: The online dynamic nomogram accessible at https://predictrt.shinyapps.io/DynNomapp/, depicting an example for predicting the probability of Helicobacter pylori-negative early gastric cancer. EGC: Early gastric cancer.

This nomogram serves as a clinical decision-support tool without invasive examinations and is specifically designed for HpN adult patients, enabling the quantitative assessment of HpN-EGC risk stratification. This model incorporates multiple risk and protective factors to generate individualized risk profiles, and gastroscopic examination is strongly recommended for patients with elevated risk scores. To illustrate the clinical application of our risk assessment model, we presented a representative case in the online dynamic nomogram accessible at https://predictrt.shinyapps.io/DynNomapp/ (Figure 2B). A 69-year-old male with a documented history of gastric ulcers reported frequent legume consumption along with occasional intake of nuts, hot food, and smoked products. Based on comprehensive evaluation, the patient’s cumulative risk score reached 356 points, corresponding to a 79.1% predicted probability of EGC. This elevated risk profile indicates the necessity for gastroscopic examination or the implementation of a rigorous follow-up protocol. This example is presented as a detailed nomogram (Supplementary Figure 5).

Verification of the risk prediction model

The C-index of the nomogram was 0.904 (95%CI: 0.876-0.931) and the AUC was 0.904 (Figure 3A), indicating that the nomogram had good discriminatory ability and predictive accuracy. The prediction curve of the model closely aligned with the actual calibration curve (Figure 3B). The Hosmer-Lemeshow χ² statistic was 7.57 with a P value of 0.477, indicating good model calibration. The decision and clinical impact curves demonstrated that the nomogram has a high clinical application value (Figure 3C and D). Furthermore, the nomogram model achieved good predictive performance, with an ROC value of 0.865 (95%CI: 0.813-0.917) in the internal validation set and 0.794 (95%CI: 0.813-0.917) in the external validation set (Supplementary Figure 6).

Figure 3 Evaluation of the nomogram for predicting Helicobacter pylori-negative early gastric cancer in the training set. A: Receiver operating characteristic curve. The area under the curve value of 0.904 indicates excellent discriminatory ability; B: Calibration curve. P value > 0.05 in the Hosmer-Lemeshow test suggested an agreement between the predicted probabilities and observed outcomes; C: Decision curve analysis. The black line represents the assumption of no patient having early gastric cancer (EGC), while the gray line assumes that all patients were diagnosed with EGC. The red line corresponds to the risk nomogram. The model demonstrates clinical utility where its curve exceeds both reference lines across a range of threshold probabilities; D: Clinical impact curve. The “Number High-Risk” curve shows the number of patients who are predicted to be at high risk of EGC by the model. The “Number high-risk with event” curve indicates the true positives among them. The gap between curves corresponds to false positives.

Additionally, we refitted the model by replacing the composite “digestive comorbidity” variable with separate indicators for each condition (chronic atrophic gastritis, gastric ulcer, and GERD). The results of this analysis, presented in Supplementary Table 4 and Supplementary Figure 7, show that the overall model performance remained largely unchanged (C-index: 0.895, 95%CI: 0.866-0.924). The regression coefficients for the individual conditions were directionally consistent, but underpowered because of the smaller subgroup sizes.

Subgroup analysis of HpN-EGC

Patients with HpN-EGC were further categorized into those who had successfully undergone H. pylori eradication > 1 year ago (H. pylori eradication group) and those who had never been infected with H. pylori (H. pylori-naive group). Table 3 presents a comparative analysis of the demographic and clinical characteristics of the H. pylori eradication control and EGC groups. The H. pylori eradication EGC group showed a significantly higher prevalence of patients with clinical symptoms and digestive comorbidities than the control group (P < 0.001). Additionally, patients in the H. pylori eradication EGC group had significantly higher CCI scores than those in the control group (P < 0.001). Dietary pattern analysis suggested that patients in the H. pylori eradication EGC group consumed smoked and hot foods more frequently, whereas those in the control group consumed fried foods, legumes, and nuts more frequently (P < 0.05). Multivariate logistic regression analysis identified legumes and nuts as significant protective factors against EGC after H. pylori eradication, and advanced age was an independent risk factor (Table 4). We conducted a similar comparative analysis between H. pylori-naive EGC and H. pylori-naive control groups (Table 3). Advanced age, smoking, alcohol consumption, and frequent intake of smoked food, hot food, onions, and garlic were identified as significant risk factors for H. pylori-naive EGC, whereas the intake of legumes and nuts demonstrated protective effects (Table 4).

Table 3 Characteristic analysis of the Helicobacter pylori eradication control vs early gastric cancer groups and Helicobacter pylori-naive control vs early gastric cancer groups.

Characteristics	H. pylori eradication control group, n = 47	H. pylori eradication EGC group, n = 51	P value	H. pylori-naive control group, n = 248	H. pylori-naive EGC group, n = 247	P value
Sex			< 0.001			< 0.001
Male	17 (36.2)	39 (76.5)		96 (38.7)	169 (68.4)
Female	30 (63.8)	12 (23.5)		152 (61.3)	78 (31.6)
Age, years	46.23 ± 11.27	62.41 ± 8.31	< 0.001	52.71 ± 12.49	63.85 ± 8.73	< 0.001
BMI (kg/m2)	23.74 ± 3.81	23.21 ± 2.78	0.432	24.34 ± 3.62	24.03 ± 3.58	0.339
Smoking1			< 0.001			< 0.001
Yes	7 (14.9)	30 (58.8)		24 (9.7)	125 (50.6)
No	40 (85.1)	21 (41.2)		224 (90.3)	122 (49.4)
Drinking1			< 0.001			< 0.001
Yes	6 (12.8)	27 (52.9)		43 (17.3)	106 (42.9)
No	41 (87.2)	24 (47.1)		205 (82.7)	141 (57.1)
Educational level			< 0.001			< 0.001
High school or above	45 (95.7)	31 (60.8)		209 (95.7)	146 (59.1)
Below high school	2 (4.3)	20 (39.2)		39 (4.3)	101 (40.9)
Clinical symptoms			< 0.001			0.032
Yes	20 (42.6)	45 (88.2)		165 (66.5)	186 (75.3)
No	27 (57.4)	6 (11.8)		83 (33.5)	61 (24.7)
Hypertension			0.099			< 0.001
Yes	8 (17.0)	35 (68.6)		47 (19.0)	98 (39.7)
No	39 (83.0)	16 (31.4)		201 (81.0)	149 (60.3)
Diabetes			0.114			< 0.001
Yes	1 (2.1)	6 (11.8)		11 (4.4)	49 (19.8)
No	46 (97.9)	45 (88.2)		237 (95.6)	198 (80.2)
Hyperlipidemia			0.889			0.984
Yes	6 (12.8)	7 (13.7)		41 (16.5)	41 (16.6)
No	41 (87.2)	44 (86.3)		207 (83.5)	206 (83.4)
Cardiac and cerebral diseases			0.094			< 0.001
Yes	2 (4.3)	8 (15.7)		10 (4.0)	67 (27.1)
No	45 (95.7)	43 (84.3)		238 (96.0)	180 (72.9)
Digestive comorbidities			< 0.001			< 0.001
Yes	17 (36.2)	44 (86.3)		106 (42.7)	181 (73.3)
No	30 (63.8)	7 (13.7)		142 (57.3)	66 (26.7)
Family history of GC			0.830			0.156
Yes	12 (25.5)	14 (27.5)		64 (25.8)	169 (68.4)
No	35 (74.5)	37 (72.8)		184 (74.2)	78 (31.6)
CCI	0.40 ± 0.68	0.90 ± 1.17	0.014	0.38 ± 0.73	0.95 ± 1.08	< 0.001
Aspirin2			0.672			0.015
Yes	3 (23.1)	4 (16.0)		4 (4.4)	21 (14.4)
No	10 (76.9)	21 (84.0)		87 (95.6)	125 (85.6)
Metformin2			0.538			0.001
Yes	0 (0.0)	2 (8.0)		1 (1.1)	19 (13.1)
No	13 (100.0)	23 (92.0)		90 (98.9)	126 (86.9)
Corticosteroids2			0.342			0.301
Yes	1 (7.7)	0 (0.0)		0 (0.0)	4 (2.7)
No	12 (92.3)	25 (100.0)		91 (100.0)	142 (97.3)
PPI2			0.094			0.129
Yes	45 (95.7)	43 (84.3)		11 (4.4)	228 (92.3)
No	2 (4.3)	8 (15.7)		237 (95.6)	19 (7.7)
Hard food			0.057			< 0.001
Occasional	47 (100.0)	46 (90.2)		244 (98.4)	204 (82.6)
Frequent	0 (0.0)	5 (9.8)		4 (1.6)	43 (17.4)
Fried food			0.025			0.066
Occasional	37 (78.7)	48 (94.1)		230 (92.7)	217 (87.9)
Frequent	10 (21.3)	3 (5.9)		18 (7.3)	30 (12.1)
Spicy food			0.111			< 0.001
Occasional	41 (87.2)	38 (74.5)		225 (90.7)	195 (78.9)
Frequent	6 (12.8)	13 (25.5)		23 (9.3)	52 (21.1)
Smoked food			0.019			< 0.001
Occasional	44 (93.6)	39 (76.5)		240 (96.8)	177 (71.7)
Frequent	3 (6.4)	12 (23.5)		8 (3.2)	70 (28.3)
Hot food			0.047			0.037
Occasional	38 (80.9)	32 (62.7)		184 (74.2)	162 (65.6)
Frequent	9 (19.1)	19 (37.3)		64 (25.8)	85 (34.4)
Onion and garlic			0.475			< 0.001
Occasional	10 (21.3)	8 (15.7)		61 (24.6)	28 (11.3)
Frequent	37 (78.7)	43 (84.3)		187 (75.4)	219 (88.7)
Legumes			0.002			< 0.001
Occasional	16 (34.0)	33 (64.7)		99 (39.9)	161 (65.4)
Frequent	31 (66.0)	18 (35.3)		149 (60.1)	85 (34.6)
Nuts			0.035			< 0.001
Occasional	31 (66.0	43 (84.3)		176 (71.0)	212 (85.8)
Frequent	16 (34.0)	8 (15.7)		72 (29.0)	35 (14.2)
Coarse grains			0.756			0.177
Occasional	30 (63.8)	31 (60.8)		168 (67.7)	153 (61.9)
Frequent	17 (36.2)	20 (39.2)		80 (32.3)	94 (38.1)
Meat, egg, and milk			0.679			0.510
Occasional	2 (4.3)	4 (7.8)		19 (7.7)	23 (9.3)
Frequent	45 (95.7)	47 (92.2)		229 (92.3)	224 (90.7)
Fruits and vegetables			0.496			0.724
Occasional	0 (0.0)	2 (3.9)		5 (2.0)	3 (1.2)
Frequent	47 (100.0)	49 (96.1)		243 (98.0)	244 (98.8)

¹Smoking or drinking now or having been smoking or drinking for more than 1 year.

²Long term use of drugs refers to using for more than 1 year.

H. pylori: Helicobacter pylori; EGC: Early gastric cancer; CCI: Charlson comorbidity index; PPI: Proton pump inhibitors; BMI: Body mass index; GC: Gastric cancer.

Table 4 Multivariate logistic regression analysis of the risk factors of Helicobacter pylori eradication early gastric cancer, Helicobacter pylori-naive early gastric cancer, and Helicobacter pylori infection early gastric cancer.

	H. pylori eradication EGC			H. pylori-naive EGC			H. pylori infection EGC
	OR	95%CI	P value	OR	95%CI	P value	OR	95%CI	P value
Age, years	1.20	1.05-1.37	0.008	1.13	1.06-1.19	< 0.001	1.12	1.02-1.22	0.014
Smoking	15.84	0.75-334.67	0.076	5.70	1.42-22.96	0.014	8.29	0.34-201.66	0.194
Yes vs no
Drinking	2.82	0.17-47.58	0.473	3.98	1.02-15.59	0.047	12.9	1.42-117.8	0.023
Yes vs no
Digestive comorbidities	3.57	0.35-36.60	0.284	5.32	1.57-18.03	0.003	6.63	1.02-43.34	0.048
Yes vs no
Smoked food	8.78	0.29-267.44	0.213	9.97	1.54-64.67	0.016
Frequent vs occasional
Hot food	11.28	0.70-182.98	0.088	3.59	1.07-12.11	0.039
Frequent vs occasional
Onion and garlic				5.26	1.21-22.88	0.027	3.67	0.54-25.08	0.186
Frequent vs occasional
Legumes	0.08	0.01-0.72	0.028	0.17	0.06-0.51	0.002	0.14	0.02-0.96	0.045
Frequent vs occasional
Nuts	0.04	0.01-0.55	0.017	0.24	0.08-0.77	0.015	0.11	0.02-0.86	0.035
Frequent vs occasional

H. pylori: Helicobacter pylori; EGC: Early gastric cancer; OR: Odds ratio; CI: Confidence interval.

To ensure a comprehensive assessment of EGC risk factors, we extended our analysis to include H. pylori-infected patients from previously excluded cases; the detailed results are presented in Supplementary Table 5. Notably, advanced age and alcohol consumption emerged as significant risk factors for EGC in H. pylori-infected patients, whereas the dietary intake of legumes and nuts demonstrated protective effects (Table 4). The results of the multivariate logistic regression analyses for the other variables are shown in Supplementary Table 6.

DISCUSSION

GC is a severe disease influenced by genetic and environmental factors, and therapeutic effects and prognosis are closely related to the timing of diagnosis and treatment[23]. However, most patients with EGC experience delayed diagnosis and treatment due to the absence of clinical symptoms or specific signs. Several studies have elucidated the characteristics and pathogenesis of GC related to H. pylori infection; however, studies on HpN-EGC were limited. This study explored the risk factors related to HpN-EGC and established a nomogram for the initial stratification of patients.

The results demonstrated that age, digestive comorbidities, the dietary habits of smoked food, hot food, legumes, and nuts were associated with the development of HpN-EGC. Digestive comorbidities, specifically chronic atrophic gastritis, gastric ulcers, and GERD, were associated with the risk of HpN-EGC and incorporated them into the prediction model. In addition, an alternative model was developed by replacing the composite “digestive comorbidity” variables with separate indicators for each condition. The regression coefficients for the individual comorbidities were directionally consistent with the composite measure, but were underpowered due to the smaller sample sizes in the subgroups. These results support the use of a composite variable in the main model to maintain parsimony. However, we emphasize the importance of interpreting this composite variable with caution.

The pathological features of chronic atrophic gastritis include reduction or loss of gastric glands accompanied by inflammatory infiltration, often indicating a significantly increased risk of GC[24]. In a pooled analysis of 11 case-control studies, researchers discovered a positive association between gastric ulcers and GC risk; however, no association was noted between duodenal ulcers and GC risk[25]. One study reported that patients with GERD had a two-four times higher risk of GC, possibly because of intestinal metaplasia of the cardia[26,27].

This study indicated that frequent consumption of smoked and hot foods was an independent risk factor for HpN-EGC, whereas frequent consumption of legumes and nuts was an independent protective factor for HpN-EGC. The World Cancer Research Fund/American Institute for Cancer Research concluded that smoked foods may be a primary cause of GC because they contain many carcinogens, including polycyclic aromatic hydrocarbons and N-nitroso compounds[28]. The consumption of hot foods and drinks is associated with an increased risk of esophageal cancer[29]. The habitual consumption of hot foods may contribute to persistent irritation of the esophageal mucosa and alterations in the esophageal microbiome. Such chronic irritation can provoke inflammatory responses, which in turn promote the endogenous formation of reactive nitrogen species such as nitrosamines. Additionally, compromised mucosal integrity may render the epithelial tissue more susceptible to exposure to other carcinogenic agents[30]. Furthermore, studies have identified a higher incidence of p53 mutations, specifically G:C to A:T transitions at CpG sites, in the esophageal tumors of individuals with prolonged exposure to high-temperature foods[31]. Nevertheless, the potential mechanisms linking hot food consumption to GC require further investigation.

Wang et al[32] discovered that soybean intake significantly decreased the risk of GC by 36%, suggesting that legumes are protective against GC, which is consistent with the results of our study. This may be attributed to the numerous beneficial phytochemicals present in leguminous foods, such as isoflavones, which have anti-cancer properties[33]. Soybeans improve the immune function of gastric cells and enhance lymphocyte proliferation, immune responses, thymocyte differentiation, and tumor immunity[33]. Some experts have suggested that a diet rich in legumes may increase the number of beneficial bacteria such as lactic acid bacteria in the microbiome, potentially reducing the risk of GC[34]. In addition, the biological effects of soy protein and other components in leguminous foods have been demonstrated to help prevent DNA damage and improve DNA repair ability[35].

Previous studies have suggested that aging is an important risk factor for GC, and its incidence gradually increases with age[36]. In the subgroup analysis, the incidence of EGC in older patients was significantly elevated, which is consistent with the results of previous studies, regardless of whether the patients were infected with H. pylori. This may be related to the decreased regenerative capacity of gastric mucosal cells and slow repair process after age-related damage.

In the subgroup analysis, we found that the association between smoking and alcohol consumption was confined to the H. pylori-naive population, suggesting that H. pylori-naive individuals should be encouraged to quit smoking and limit alcohol consumption. However, a study by Kim et al[10] found no statistically significant differences in smoking and alcohol consumption between resectable HpN and H. pylori-positive GC patients[10]. Consequently, future studies with larger sample sizes are required to elucidate the association between tobacco and alcohol use and H. pylori-naive EGC. Additionally, individuals without H. pylori infection should focus on maintaining a healthy diet. Compared to the H. pylori eradication EGC group, drinking and digestive comorbidities were independent risk factors for EGC in H. pylori infection group, suggesting that H. pylori-infected patients should avoid alcohol consumption and actively manage chronic atrophic gastritis, gastric ulcers, and GERD. The independent risk factors for EGC differed among populations with different H. pylori statuses, suggesting that stratified management and the development of personalized treatment for populations are necessary.

The prediction model for HpN-EGC had effective discriminatory ability, superior predictive accuracy, and good clinical application value, thereby assisting in clinical decision-making. A previous study developed a GC risk prediction rule comprising seven variables (age, sex, pepsinogen I/II ratio, gastrin-17 level, H. pylori infection, pickled food, and fried food) with an AUC of 0.757 but without visualization[37]. Compared to the established model, our nomogram demonstrated superior discriminative ability in our cohort, which is likely attributable to the incorporation of novel dietary and lifestyle-specific risk factors for HpN-EGC. Another nomogram based on the Kyoto Gastritis Classification was developed to predict the risk of GC, which still requires reference to gastroscopic examination results[15]. Compared with these models, our nomogram demonstrated improved accuracy, convenience, non-invasiveness, and acceptability.

Our study has several limitations. First, the nomogram was developed using a single-center cohort, which may limit its generalizability because of a potential selection bias. Furthermore, the inclusion criteria, particularly the requirement for complete dietary data, may have resulted in a specific patient subpopulation, potentially limiting the generalizability of our findings. Most notably, because dietary factors were integral to our model, significant regional and cultural variations could influence the external validity of the model when applied to populations with distinct dietary habits. Multi-center prospective studies across diverse geographical regions are essential to externally validate and potentially recalibrate our model. Second, the use of a Food Frequency Questionnaire is susceptible to recall bias and measurement errors when assessing dietary habits. Although this questionnaire is a widely used and validated tool, future studies may benefit from more objective dietary assessment methods. Third, our model focused on clinical and epidemiological variables and did not incorporate genetic factors that are known to play a role in gastric carcinogenesis. Integrating genetic markers could further enhance the model performance in the future. Although the model demonstrated good discriminative ability, we observed exceptionally high ORs for “digestive comorbidities” and “smoked foods”. While bootstrap validation suggests that these estimates are relatively statistically stable, their considerable magnitude may be influenced by relatively small subgroup sample sizes, unmeasured confounding factors, or over-optimism due to potential model overfitting. Therefore, these specific associations must be interpreted with extreme caution, and their exact risks require further validation in prospective cohort studies or studies with larger sample sizes. Moreover, owing to the limited sample size of the H. pylori eradication group, we were unable to perform a refined stratified analysis of the nomogram. Thus, the conclusions derived from this subgroup analysis should be considered merely exploratory findings and require further validation in future studies. External validation was performed using a relatively small sample size (n = 109), which may have affected the stability of the performance metrics and necessitated further validation in larger cohorts.

CONCLUSION

Our risk-prediction tool for HpN-EGC effectively identified high-risk individuals by leveraging six readily available clinical parameters: Age, digestive comorbidities, and consumption of smoked and hot foods, legumes, and nuts. Furthermore, the stratified analysis revealed distinct risk and protective factors across various population subgroups.