Factors associated with gastric adenocarcinoma and dysplasia in patients with chronic gastritis: a population-based study.

OBJECTIVE: Gastric cancer (GC) is one of the leading causes of death in China and other Asian countries. Recently, gastric endoscopy has become the main approach for GC screening, but the identification of high-risk individuals remains a challenge in GC screening programs.
METHODS: There were 7,302 patients with chronic gastritis involved in this study. Endoscopic examinations were performed, and their demographic characteristics and lifestyle data were collected. Each possible associated factor of GC/premalignant and precursor lesions was evaluated by univariate and multivariate logistic regressions. Nomograms were used for visualization of those models, and receiver operating characteristic (ROC) curve analysis was used to present the predictive accuracy.
RESULTS: We detected 8 (0.11%) gastric adenocarcinomas, 17 (0.23%) dysplasia cases, 14 (0.19%) hyperplasia cases, 52 (0.71%) intestinal metaplasia cases, 217 (2.97%) inflammatory lesions, 141 (1.93%) gastric ulcers, 10 (0.14%) atrophic gastritis cases, 1,365 (18.69%) erosive gastritis cases, and 5,957 (81.58%) superficial gastritis cases in 7,302 patients. The age (P<0.001), gender (P=0.086), labor intensity (P=0.018) and leek food intake (P=0.143) were identified as independent predictive factors of GC/premalignant lesions possibility. The corresponding nomogram exhibited an area under the curve (AUC) [95% confidence interval (95% CI)] of 0.82 (0.74-0.89) for the modeling group and 0.80 (0.75-0.85) for the validation group. The age (P=0.002), gender (P=0.024), smoking (P=0.002) and leek food intake (P=0.039) were independent predictive factors of precursor lesions possibility. The corresponding nomogram exhibited an AUC (95% CI) of 0.62 (0.60-0.65) for the modeling group and 0.61 (0.59-0.63) for the validation group.
CONCLUSIONS: We identified several potential associated factors and provided a preclinical nomogram with the potential to predict the possibility of GC/premalignant and precursor lesions.

Introduction

Gastric cancer (GC) is one of the most commonly diagnosed carcinomas worldwide with 680,000 new cases and 754,000 deaths in 2012 (1). GC has an especially high incidence and mortality in East Asia and China, accounting for approximately 40% of GC cases and accompanied by a poor nutrition status and unhealthy living habits of the residents in high-risk areas (2). Although gastric endoscopy has been used in the screening for GC and has a decreasing trend in mortality, GC remains the main cancer burden in some areas (3).

GC is highly aggressive and usually involves lymph nodes and distant organ metastasis (4). Most patients are diagnosed at a late stage and have a very poor prognosis (5). Many studies have shown that the overall 5-year survival of GC has a strong relationship with the stage at detection and treatment (6). Kim et al. reported that the 5-year survival rates of GC decreased dramatically when it was not diagnosed at a very early stage; therefore, proper screening management for GC is important (7). Increasing the early detection and treatment rate of GC would be very beneficial for GC patients (8,9).

Recently, for GC, premalignant and precursor lesions have been studied in detail. Risk factors, such as Helicobacter pylori (HP) infection, chronic atrophic gastritis (CAG) and intestinal metaplasia (IM), are widely recognized and adopted in GC screening guidelines (10). However, some studies have also showed that HP eradication did not completely reduce the incidence of GC (11); on the other hand, many other risk factors have been proposed, such as the age, gender, smoking, drinking, other dietary factors, etc., but there is no consensus (12). Due to the lack of clinical and epidemiological evidence to select eligible people with high risk, GC screening has nearly completely depended on endoscopy (13). Recent GC screening programs have only selected patients by HP and CAG, especially in China (14), resulting in low compliance with screening and waste of medical resources.

In this study, we established a cohort of 7,302 chronic gastritis patients in a single medical center who were screened for GC as well as its premalignant and precursor lesions by endoscopy; we then collected epidemiological data. Here, we identified several associated factors of GC/premalignant and precursor lesions and established a method to select eligible populations for endoscopic screening based on demographic data, lifestyle factors, eating habits, and psychological factors.

Materials and methods

Participant inclusion/exclusion criteria

A total of 7,497 chronic gastritis patients in Jizhong Energy Fengfeng Group Hospital from Handan City, Hebei Province, were enrolled in this study during October 2010 to September 2011, and all were 40–70 years old. All 7,497 patients provided informed consent and 7,484 completed the epidemiology questionnaire. General physical examination was conducted, and the reports showed that 182 people were excluded because of: 1) iodine allergy; 2) diagnosed with other tumors; 3) pregnancy; 4) hyperthyroidism; 5) lactation; or 6) psychosis. Based on the questionnaire and health examination, 7,302 participants were recruited, and they all successfully underwent a gastrointestinal endoscopy examination (). All procedures in this study had been approved by the Ethics Committee of Beijing Friendship Hospital.

Flow chart of patient inclusion and exclusion.

Performance of endoscopy and its quality control

All endoscopic procedures (examinations and therapies) were performed by experienced endoscopists. Before the endoscopic procedure, all participants were required to have an empty stomach for 8 h and stop anticoagulant drugs for 1 week. We used iodine staining to facilitate the discovery of lesions, and any lesions were biopsied, including carcinoma, hyperplasia, dysplasia, ulcers, inflammatory lesions, and others. Also, we examined 5% of normal gastric mucosa as negative controls. The diagnostic results were determined by two pathologists based on the diagnostic criteria of the World Health Organization.

Acquisition and processing of data

We acquired data through questionnaires of demographics, eating habits, lifestyle factors, psychological factors, and others. Smoking status was defined by four subgroups: “yes” means smoking for more than 6 months and ≥1 cigarettes per day; “occasionally” means <1 cigarette/d or less than 6 months; “refrained” means stopped smoking for more than 6 months; and “never” means never smoke. Eating habits (including leek food intake, fresh vegetable consumption, etc.) were defined by two subgroups: “yes” means ≥3 per month; and “rarely or never” means never or <3 per month. Data on different stomach lesions were collected by endoscopic examinations, and the consensus of two pathologists provided the pathological diagnosis data. We imported all questionnaires and relative information by two independent researchers into the database.

Statistical analysis and modeling

R Software (Version 3.3.1; Rocket Software, Inc. Waltham, MA, USA) was used to analyze all statistics with logistic regression models. Packages “dplyr”, “plyr”, and “reshape2” were used for data cleaning. Packages “rms” and “ordinal” were used for nomogram building. Package “ggplot2” was used for visualization. We conducted univariate analysis to identify the risk factors of gastric lesions, and displayed the odds ratio (OR) with 95% confidence interval (95% CI). We use multivariate regression analysis to identify the independent influence factors. Nomograms were established based on the results by half of the cases which sampled by the Monte Carlo method. The other half of the cases were used as a validation set. The receiver operating characteristic (ROC) curve was used to graphically display the predictive accuracy of nomograms. The area under the curve (AUC) for validation was used to assess the nomogram accuracy.

Results

Baseline data of chronic gastritis patient population

Overall, 7,302 chronic gastritis patients successfully underwent gastric endoscopy and completed epidemiology questionnaires. Detailed data of the demographics and lifestyle factors were obtained. We detected 8 (0.11%) gastric carcinomas, and all were adenocarcinomas. Seventeen (0.23%) dysplasia and 14 (0.19%) hyperplasia cases were also detected. There were 52 (0.71%) cases with intestinal metaplasia, 217 with inflammatory lesions (2.97%), 141 (1.93%) with gastric ulcers, 10 (0.14%) with atrophic gastritis, 1,365 (18.69%) with erosive gastritis, and 5,957 (81.58%) with superficial gastritis ().

Among all 7,302 patients, 6,574 (90.03%) were between 41 and 60 years old, 5,008 (68.58%) were male, 5,246 (71.84%) were workers, and 5,229 (71.61%) earned ¥2,001–3,000 ($286–428) per month. The demographic baseline data of this population are also displayed in .

Detection of associated factors for GC/premalignant lesions

To identity the associated factors for GC and premalignant lesions, we performed univariate logistic regression analyses of GC/premalignant lesions ( and ). We found that age (P<0.001), occupation (P=0.030), labor intensity (P=0.006), tea consumption (P=0.014), and water source (P=0.050) were all significantly associated with GC and premalignant lesion incidence, while gender (P=0.077), tumor history (P=0.093), high temperature food intake (P=0.063), and leek food intake (P=0.079) were at a marginal level of association.

However, in multivariate analysis, only the age (P<0.001), gender (P=0.086), labor intensity (P=0.018) and leek food intake (P=0.143) were independent predictive factors of GC/premalignant lesions possibility ().

Detection of associated factors of precursor lesions

To identity the associated factors of GC precursor lesions, we performed univariate logistic regression analyses of precursor lesions ( and ). We found that age (P=0.002), gender (P=0.016), income (P=0.018), and smoking (P<0.001) were significantly associated with the GC and premalignant lesion incidence, while leek food intake was at a marginal level (P=0.090).

However, in multivariate analysis, only age (P=0.002), gender (P=0.024), smoking (P=0.002) and leek food intake (P=0.039) were independent predictive factors of precursor lesions possibility ().

Nomograms for predicting gastric lesions

All independent GC/premalignant and precursor lesion associated factors (age, gender, etc.) were included to develop predictive nomograms for predicting those diseases, which would provide very helpful information in the clinical evaluation and selection of eligible people for GC screening.

Half of the cases were sampled by the Monte Carlo method to generate predictive nomograms (, ). For each case, we assigned a point for each factor, and then summed them to generate a total score. A corresponding predicted probability of certain lesions was calculated from the nomogram. ROC curves were used to evaluate the predictive accuracy of the nomograms. For prediction of GC/premalignant lesions, the AUC (95% CI) was 0.82 (0.74–0.89) for the modeling group () and 0.80 (0.75–0.85) for the validation group (). In the prediction of precursor lesions, the AUC (95% CI) was 0.62 (0.60–0.65) for the modeling group () and 0.61 (0.59–0.63) for the validation group ().

Nomograms for predicting gastric cancer (GC)/premalignant and precursor lesions possibility. (A) Predictive nomogram of GC/premalignant lesions (adenocarcinoma, dysplasia and intestinal metaplasia); (B) Receiver operating characteristic (ROC) curve of a nomogram model to predict the GC/premalignant lesions possibility (modeling data); (C) ROC curve of a nomogram model to predict the GC/premalignant lesions possibility (validation data); (D) Predictive nomogram of precursor lesions (erosive gastritis, atrophic gastritis and gastric ulcer); (E) ROC curve of the nomogram model to predict the precursor lesion possibility (modeling data); (F) ROC curve of the nomogram model to predict the precursor lesion possibility (validation data).

Discussion

Many independent studies have suggested that identification of GC at an early stage could reduce GC mortality and substantially decrease medical burden (15-17). Additionally, detection of precancerous and precursor lesions could also reduce both the incidence and mortality of GC (18,19). Gastric endoscopy is the main approach to screen for GC and its precancerous lesions. Recently, most GC endoscopy screening programs were conducted in a target population that was roughly defined by specific associated factors, such as age and family history (20-22).

Inefficient pre-selection of high-risk people demonstrated low cost-efficiency in most GC screening programs. The low true positive rate in screening was also influenced by very weak compliance according to our experience. Systematic evaluation of the possible associated factors of GC/premalignant and precursor lesions would further our understanding of GC and facilitate endoscopists’ decision making in terms of whether to perform endoscopy examination.

In this study, it was not surprising that GC/premalignant and precursor lesions largely shared independent associated factors, considering the natural historical origination of GC (23). Age was greatly associated with both GC/premalignant and precursor lesion incidence, which may be caused by DNA injury accumulation in the elderly (24). Decreased immunity in the elderly could also explain the general increase in most diseases (25). Additionally, we found that men were more likely to develop GC/premalignant lesions, while women were more likely to be affected by precursor lesions. This phenomenon could be explained by potential confounding factors, such as different lifestyles between men and women (men are more addicted to smoking and drinking, while women are more likely to go on a drastic diet) (26). Previous studies suggested that leek consumption could reduce the gastrointestinal cancer risk in the general population. However, we found that leek food intake was positively associated with GC/premalignant incidence, while it was negatively associated with the precursor lesion incidence. The exact role of leek food intake in GC cancer development requires more, higher level evidence.

Nomograms exhibited very promising potential in clinical trial design and interpretation; also, they were widely adopted in prognostic models (27-29). However, they were rarely used in primary health screening (30,31), and they were never used in GC screening programs. In this study, nomograms were used as a visualization tool to display the prediction model based on logistic regression. Here, we established a nomogram-based method to evaluate the possibility of GC/premalignant and precursor lesions in chronic gastritis patients, which is the first nomogram to predict the individual GC risk based on demographic and epidemiological data. The AUC of our model ranged from 0.61 to 0.82, suggesting that a nomogram would be a very promising tool in individual GC risk assessment, especially in primary health screening programs.

There were also several shortcomings in this study. First, limited by the condition in the local medical center, we did not examine the HP status, which may be a very important GC associated factor. Second, because only demographic and lifestyle information were collected to build disease associated models, no biochemistry or hematology tests were involved in this study. Third, our nomograms were based on a retrospective single centered cohort. Nevertheless, we did not aim to provide a perfect prediction of GC/premalignant and precursor lesions; instead, our goal was to deliver an easy to use tool for rough filtering of an eligible population for GC screening that is better than the currently available strategies. Additionally, there were only 8 carcinomas detected in our population, which is much lower than expected. The composition of participants and their age distribution might partially explain this outcome. Most of the patients from Jizhong Energy Fengfeng Group Hospital were workers in Jizhong Energy Fengfeng Group Co. Ltd., aged under 60 years (before retirement). Considering the elder people (>60 years old) have much higher GC risk than others, it is not surprising we obtained a low carcinoma detection rate. Thus, we could only investigate GC along with its premalignant lesions.

Conclusions

We provided a systematic evaluation of possible associated factors for GC/premalignant and precursor lesions based on a large population of chronic gastritis patients and then generated nomograms to predict the disease possibility of every individual. A multi-centered prospective study with a larger population would be expected to provide a more reliable estimation in the future.

Acknowledgements

This work was fully supported by the National Natural Science Foundation of China (Grant No. 81302160 and 81272447) and Beijing Natural Science Foundation Program and Scientific Research Key Program of Beijing Municipal Commission of Education (Grant No. KZ201410025024).

Footnote

Conflicts of Interest: The authors have no conflicts of interest to declare.

1

Gastric lesions in patients (N=7,302)

Lesions	n	%
Carcinoma	8	0.11
Dysplasia	17	0.23
Hyperplasia	14	0.19
Intestinal metaplasia	52	0.71
Inflammatory lesions	217	2.97
Gastric ulcer	141	1.93
Atrophic gastritis	10	0.14
Erosive gastritis	1,365	18.69
Superficial gastritis	5,957	81.58

2

Demographic baseline data of chronic gastritis patient population

Characteristics	Case No.	%
Age (year)
≤40	473	6.48
41–50	4,293	58.79
51–60	2,281	31.24
>60	255	3.49
Gender
Male	5,008	68.58
Female	2,294	31.42
Education
College and above	824	11.28
High school	2,630	36.02
Middle school	3,356	45.96
Primary school	492	6.74
Occupation
Worker	5,246	71.84
Peasant	102	1.40
Officer	968	13.26
Service	255	3.49
Retired	629	8.61
Others	102	1.40
Labor intensity
Light	3,747	51.31
Moderate	2,553	34.96
Heavy	1,002	13.72
Income (¥)
≤2,000	1,597	21.87
2,001–3,000	5,229	71.61
3,001–4,000	1,751	23.98
>4,000	1,425	19.52

3

Univariate and multivariate analyses of GC and premalignant lesions associated factors

Characteristics	Univariate*				Multivariate
	OR	95% CI	P		OR	95% CI	P
Age (year)			<0.001				<0.001
41–50 vs. ≤40	1.4E6	NA			1.2E6	NA
51–60 vs. ≤40	5.3E6	NA			3.9E6	NA
>60 vs. ≤40	1.4E7	NA			9.7E6	NA
Gender			0.077				0.086
Female vs. male	0.42	(0.12, 1.09)			0.38	(0.11, 1.04)
Occupation			0.030
Peasant vs. worker	8.72	(1.35, 32.5)
Officer vs. worker	1.36	(0.31, 42.8)
Service vs. worker	3.45	(0.53, 12.7)
Retired vs. worker	4.20	(1.46, 1.09)
Others vs. worker	3.8E–6	NA
Labor intensity			0.006				0.018
Moderate vs. light	0.15	(0.02, 0.53)			0.19	(0.03, 0.69)
Heavy vs. light	0.79	(0.23, 2.10)			1.02	(0.29, 2.89)
Income (¥)			0.390
2,001–3,000 vs. ≤2,000	0.42	(0.14, 1.17)
3,001–4,000 vs. ≤2,000	0.51	(0.15, 1.47)
>4,000 vs. ≤2,000	0.62	(0.19, 1.80)
Smoking			0.147
Occasionally vs. Yes	0.66	(0.10, 2.33)
Refrained vs. Yes	0.36	(0.02, 1.79)
Never vs. Yes	0.38	(0.14, 0.89)
Tumor history			0.093
No vs. Yes	2.32	(0.88, 7.95)
Other disease history			0.302
No vs. Yes	2.00	(0.59, 12.45)
Tea consumption			0.014
No vs. Yes	0.33	(0.15, 0.79)
Eating postures			0.629
Sit vs. squat	1.1E6	NA
Stand vs. squat	1.00	NA

S1

Univariate analysis of gastric lesions associated factors

Characteristics	Precursor lesions				Premalignant/malignant lesions
	OR	95% CI	P		OR	95% CI	P
Age (year)			<0.002				<0.001
41–50 vs. ≤40	2.10	(1.13, 4.45)			1.406	(6.17E–15, NA)
51–60 vs. ≤40	2.73	(1.46, 5.85)			5.286	(2.33E–14, NA)
>60 vs. ≤40	3.45	(1.53, 8.26)			1.367	(6.02E–14, NA)
Gender			<0.016				0.077
Female vs. male	0.73	(0.56, 0.95)			0.42	(0.12, 1.09)
Race			0.218				0.156
Others vs. Han	2.04	(0.61, 5.06)			6.55	(0.36, 32.06)
Married			0.624				0.475
Others vs. married	1.30	(0.39, 3.17)			2.49–6	(NA, 2.61)
Education			0.393				0.536
High school vs. college	1.10	(0.74, 1.69)			0.63	(0.16, 2.97)
Middle school vs. college	1.20	(0.82, 1.83)			1.06	(0.34, 4.65)
Primary school vs. college	1.54	(0.90, 2.62)			1.68	(0.31, 9.01)
Occupation			0.640				0.030
Peasant vs. worker	1.19	(0.42, 2.68)			8.72	(1.35, 32.5)
Officer vs. worker	1.03	(0.72, 1.43)			1.36	(0.31, 4.28)
Service vs. worker	0.95	(0.46, 1.72)			3.45	(0.53, 12.70)
Retired vs. worker	1.28	(0.87, 1.84)			4.20	(1.46, 1.09)
Others vs. worker	0.46	(0.08, 1.47)			3.77–6	(NA, 5.32E10)
Labor intensity			0.750				0.006
Moderate vs. light	1.03	(0.80, 1.31)			0.15	(0.02, 0.53)
Heavy vs. light	0.89	(0.61, 1.26)			0.79	(0.23, 2.10)
Income (¥)			0.018				0.390
2,001–3,000 vs. ≤2,000	0.70	(0.53, 0.94)			0.42	(0.14, 1.17)
3,001–4,000 vs. ≤2,000	0.64	(0.46, 0.88)			0.51	(0.15, 1.47)
>4,000 vs. ≤2,000	0.63	(0.44, 0.90)			0.62	(0.19, 1.80)
Mining exposure			<0.527				0.445
Yes vs. No	1.08	(0.86, 1.35)			0.73	(0.32, 1.62)
Mining exposure time (hours/week)			<0.482				0.573
1–40 vs. 0	1.11	(0.74, 1.61)			0.71	(0.11, 2.52)
41–60 vs. 0	1.18	(0.90, 1.53)			0.93	(0.35, 2.21)
>60 vs. 0	0.88	(0.58, 1.30)			0.31	(0.02, 1.51)
Dust exposure			0.982				0.192
Yes vs. No	1.00	(0.79, 1.25)			0.59	(0.26, 1.30)
Gas exposure			0.287				0.124
Yes vs. No	0.88	(0.70, 1.11)			0.52	(0.20, 1.19)
Smoking			<0.001				0.147
Occasionally vs. Yes	0.73	(0.45, 1.11)			0.66	(0.10, 2.33)
Refrained vs. Yes	0.59	(0.34, 0.95)			0.36	(0.02, 1.79)
Never vs. Yes	0.58	(0.46, 0.75)			0.38	(0.14, 0.89)

4

Univariate and multivariate analyses of precursor lesions associated factors

Characteristics	Univariate*				Multivariate
	OR	95% CI	P		OR	95% CI	P
Age (year)			0.002				0.002
41–50 vs. ≤40	2.10	(1.13, 4.45)			2.03	(1.09, 4.32)
51–60 vs. ≤40	2.73	(1.46, 5.85)			2.63	(1.39, 5.62)
>60 vs. ≤40	3.45	(1.53, 8.26)			3.73	(1.65, 8.96)
Gender			0.016				0.024
Female vs. male	0.73	(0.56, 0.95)			1.06	(0.72, 1.56)
Occupation			0.640
Peasant vs. worker	1.19	(0.42, 2.68)
Officer vs. worker	1.03	(0.72, 1.43)
Service vs. worker	0.95	(0.46, 1.72)
Retired vs. worker	1.28	(0.87, 1.84)
Others vs. worker	0.46	(0.08, 1.47)
Labor intensity			0.750
Moderate vs. light	1.03	(0.80, 1.31)
Heavy vs. light	0.89	(0.61, 1.26)