Abstract
Background:
Clinical opportunistic screening is a cost-effective cancer screening modality. This study aimed to establish an easy-to-use diagnostic model serving as a risk stratification tool for identification of individuals with malignant gastric lesions for opportunistic screening.
Methods:
We developed a questionnaire-based diagnostic model using a joint dataset including two clinical cohorts from northern and southern China. The cohorts consisted of 17,360 outpatients who had undergone upper gastrointestinal endoscopic examination in endoscopic clinics. The final model was derived based on unconditional logistic regression, and predictors were selected according to the Akaike information criterion. External validation was carried out with 32,614 participants from a community-based randomized controlled trial.
Results:
This questionnaire-based diagnostic model for malignant gastric lesions had eight predictors, including advanced age, male gender, family history of gastric cancer, low body mass index, unexplained weight loss, consumption of leftover food, consumption of preserved food, and epigastric pain. This model showed high discriminative power in the development set with an area under the receiver operating characteristic curve (AUC) of 0.791 (95% confidence interval [CI]: 0.750–0.831). External validation of the model in the general population generated an AUC of 0.696 (95% CI: 0.570–0.822). This model showed an ideal ability for enriching prevalent malignant gastric lesions when applied to various scenarios.
Conclusion:
This easy-to-use questionnaire-based model for diagnosis of prevalent malignant gastric lesions may serve as an effective prescreening tool in clinical opportunistic screening for gastric cancer.
Keywords: Early detection of cancer, Cancer early diagnosis, Gastric cancer, Diagnostic model, Opportunistic screening, Feeding behavior, Weight loss, Sex characteristics
Introduction
Gastric cancer (GC) ranks fifth among the most common cancers, and is the third leading cause of cancer-related deaths globally. There are substantial regional differences in the incidence of GC, and cases of GC occur mainly in Eastern Asia, Central and Eastern Europe, and Central and Southern America.[1] China harbors a heavy burden of approximately half of the GC cases and deaths worldwide,[2] but most GC patients are diagnosed at an advanced stage with low 5-year survival.[3]
Debate regarding widespread eradication of Helicobacter pylori (H. pylori) among asymptomatic populations,[4] that is, secondary prevention involving screening of malignant lesions in the stomach has become an important strategy for improving the detection of early-stage cases and reduction of mortality. Gastrointestinal (GI) endoscopic examination with biopsy is currently the gold standard for the identification of malignant gastric lesions, and has been recommended in some nationwide screening programs in South Korea and Japan.[5–7] In recent years, marked increases in age-standardized 5-year net survival for GC have been found in South Korea (48.6–68.9%) and Japan (50.5–60.3%).[8] However, GC survival has remained at a relatively low level in regions with limited screening coverage. More recently, clinical opportunistic screening, which aims at screening patients who present to healthcare professionals for any given complaint, has attracted attention.[9] This is naturally a more cost-effective and preferred screening modality for expanding GC screening coverage as compared to organized community-based screening efforts, as such a modality may enroll subjects who are more proactive in screening decisions, achieve a higher detection rate of malignant lesions, and reduce the cost of screening.[10,11]
An essential precondition of opportunistic screening is the preparation of an appropriate risk-stratification tool for distinguishing the high-risk individuals, which will facilitate precise referral to endoscopy and substantially improve the cost-effectiveness of screening.[10,12] Several methods or indicators, such as the "ABC method", which combined H. pylori serology with measurement of serum pepsinogen levels, have been reported to be potential prescreening tools for opportunistic screening.[13] However, most of these methods require extra blood tests or provide qualitative assessments only, which reduce their generalizability in real-world opportunistic screening. To generate a quick and low-cost risk evaluation prior to screening for GC, developing a questionnaire-based diagnostic model based on large-scale real-world clinical cohorts is urgently needed for the establishment and promotion of opportunistic endoscopic screening for GC in China and worldwide.
In this study, we constructed a new questionnaire-based quantitative diagnostic model for identifying prevalent malignant gastric lesions based on two large-scale outpatient cohorts from endoscopic clinics, and validated it in an independent community-based screening cohort.
Methods
Ethical approval
This study was approved by the Institutional Review Board of the Peking University School of Oncology, China (Nos. 2011KT27 and 2017KT51). Written informed consent was obtained from each participant.
Study participants
Development set
Two outpatient cohorts (China Cohort Consortium: No. CCC2020010301; https://chinacohort.bjmu.edu.cn) from endoscopic clinics were used to develop a diagnostic model for malignant gastric lesions. Details regarding these two cohorts have been described previously.[12] In brief, northern outpatient cohort was based on the Hua County People's Hospital (Henan, China) and recruited outpatients receiving upper GI endoscopy from March 1, 2017 to December 31, 2021. Hua County is a rural region with a high incidence of esophageal squamous cell carcinoma in the Taihang Mountain area.[14,15] Southern outpatient cohort was based on the Peking University Shenzhen Hospital and recruited outpatients undergoing endoscopy from June 19, 2017 to November 18, 2021. Shenzhen is an economically dynamic city with a huge migrant population in southern China. The two cohorts were combined as the full development set, and internal validation was carried out in each separate cohort.
Validation set
A community-based screening cohort was used to assess the applicability of our model. This cohort was based on the Efficacy of endoscopic Screening for Esophageal Cancer in China (ESECC) trial (No. NCT01688908; https://www.clinicaltrials.gov), which was a randomized controlled trial conducted in Hua County to evaluate the efficacy and cost-effectiveness of endoscopic screening for esophageal cancer. As described previously a total of 668 villages were randomly selected from the 846 villages with population sizes ranging from 500 to 3000 in Hua County and allocated equally into a screening group and a control group, using a blocked randomization procedure.[15] Endoscopic screening was conducted in the screening group but not in the control group. The two groups of the ESECC trial were combined as the validation set.
Inclusion criteria
For the development and validation sets, the inclusion criteria were the following: (1) Aged 45–69 years; (2) No history of cancer, no mental disorder, and no contraindications for endoscopy; and (3) Completion of an adequate upper GI endoscopic examination (not applicable for the control group of the ESECC trial).
Data collection and outcome definition
All participants in the outpatient cohorts from endoscopic clinics and the community-based screening cohort completed a one-on-one computer-aided questionnaire, which was standardized for the collection of demographic variables and information regarding potential predictors of upper GI cancer. Candidate predictors included age, gender, socioeconomic status (level of education, household income, and type of work), living habits (cigarette smoking, alcohol consumption, dietary habits, and source of drinking water), body mass index (BMI), disease history, family history of upper GI cancer, and upper GI symptoms.
The outcome was defined as detection of malignant lesions in the stomach (not including malignant lesions in the cardia), which included high-grade intraepithelial neoplasia (or severe dysplasia) of gastric mucosa, carcinoma in situ, and GC. For the two outpatient cohorts from endoscopic clinics, all participants received an upper GI endoscopic examination (chromoendoscopy if needed), with biopsies taken from all focal lesions to generate a pathologic diagnosis. For the community-based ESECC cohort, participants in the screening group received an upper GI endoscopic examination at enrollment and were assigned a diagnosis. Annual follow-up was conducted to collect outcome events using a combination of active door-to-door interviews and passive linkage with local health insurance claims data.[15–17] Malignant gastric lesions diagnosed at baseline screening or within 1-year's follow-up in both groups were defined as outcome events in the ESECC cohort in this study.
Statistical analysis
The Chi-squared and rank-sum tests were used to compare characteristics of the participants in the development and validation sets for categorical and continuous variables, respectively.
We used a two-step approach to derive the statistical prediction model. The correlation of each potential predictor and the presence of malignant gastric lesions were first assessed using univariable logistic regression in the development set. Predictors with P <0.05 or P <0.5 and odds ratio (OR) >1.3 were then subjected to multivariable logistic regression. Backward elimination under the Akaike information criterion (AIC) was used to determine the statistical prediction model. Well-recognized risk factors of GC, age, gender, and family history of GC were constrained into the final model. Patients with missing values or other upper GI cancers were excluded from the analysis.
Receiver operating characteristic (ROC) curves were employed to visually assess discrimination, and the area under the curve (AUC) was calculated according to the observed and predicted values.
We hypothesized specific screening coverages to evaluate the application performance of our model under different workloads of opportunistic screening. The highest predicted probability for achieving the desired screening coverage in the development set was selected as the cut-off value for determination of the high-risk population. Sensitivity, the average number of endoscopies needed to detect one malignant gastric lesion, the detection rate, and the detection rate ratio in each screening population were calculated.
All analysis in this study was conducted in R version 4.0.2 (R Development Core Team, Vienna, Austria). All tests were two-sided and P values <0.05 were defined as statistically significant.
Results
Characteristics of study participants
In the development set, we ultimately detected 38 (0.4%) cases with malignant gastric lesions in 8582 subjects from the northern outpatient cohort and 84 (1.0%) cases with malignant gastric lesions in 8778 subjects from the southern outpatient cohort. In the validation set, a total of 18 subjects were diagnosed with malignant gastric lesions at baseline or within 1-year's follow-up of the ESECC cohort. As shown in Table 1, outpatients recruited from endoscopic clinics (development set) were more likely to report a family history of GC, consumption of leftover food, unexplained weight loss, and epigastric pain compared to the ESECC population (validation set). The two outpatient cohorts also differed in terms of BMI, unexplained weight loss, consumption of leftover food, consumption of preserved food, and epigastric pain.
Table 1.
Characteristics of individuals in the development and validation sets.
| Variables* | Development set, n (%) |
Validation set, n (%) Community-based screening cohort |
Statistical value | P values§ | |
|---|---|---|---|---|---|
| Northern outpatient cohort | Southern outpatient cohort | ||||
| N | 8582 | 8778 | 32,614 | – | – |
| Age (years), median (Q1, Q3) | 55 (50, 62) | 55 (50, 60) | 57 (50, 62) | 225.8 | <0.001 |
| Gender | 5124 (59.7) | 4849 (55.2) | 16,684 (51.2) | 215.0 | <0.001 |
| Female | – | – | – | – | – |
| Male | 3458 (40.3) | 3929 (44.8) | 15,930 (48.8) | – | – |
| Family history of GC | 8154 (95.0) | 8349 (95.1) | 31,759 (97.4) | 183.6 | <0.001 |
| No | – | – | – | – | – |
| Yes | 428 (5.0) | 429 (4.9) | 855 (2.6) | – | – |
| BMI † (kg/m2) | 8439 (98.3) | 8335 (95.0) | 31,968 (98.5) | 426.9 | <0.001 |
| ≥18.5 | – | – | – | – | – |
| <18.5 | 143 (1.7) | 443 (5.0) | 484 (1.5) | – | – |
| Unexplained weight loss‡ | 6714 (78.2) | 5024 (57.2) | 31,225 (95.7) | 9000.0 | <0.001 |
| No | – | – | – | – | – |
| Yes | 1868 (21.8) | 3754 (42.8) | 1389 (4.3) | – | – |
| Consumption of leftover food (times/week) | 4456 (51.9) | 3014 (34.3) | 19,965 (61.2) | 2100.0 | <0.001 |
| <1 | – | – | – | – | – |
| ≥1 | 4126 (48.1) | 5764 (65.7) | 12,649 (38.8) | – | – |
| Consumption of preserved food (times/week) | 5582 (65.0) | 4299 (49.0) | 21,227 (65.1) | 798.3 | <0.001 |
| <1 | – | – | – | – | – |
| ≥1 | 3000 (35.0) | 4479 (51.0) | 11,387 (34.9) | – | – |
| Epigastric pain | 5042 (58.8) | 3341 (38.1) | 29,660 (90.9) | 12,000.0 | <0.001 |
| No | – | – | – | – | – |
| Yes | 3540 (41.2) | 5437 (61.9) | 2954 (9.1) | – | – |
*Variables were selected by a two-step selection method where all candidate predictors were first evaluated in univariable logistic regression models, and variables with P <0.05 or P <0.5 and OR >1.3 were subjected to multivariable logistic regression analysis where the AIC was used to determine the final predictor pattern. †BMI was not available for 162 (0.5%) patients in the validation set. ‡ Unexplained weight loss referred to a weight loss of over 5% within 1 year without any subjective intent weight loss. §P values were obtained from the Chi-squared and rank-sum tests for categorical and continuous variables, respectively. AIC: Akaike information criterion; BMI: Body mass index; GC: Gastric cancer; OR: Odds ratio; –: Not available.
Model development and internal validation
Final diagnostic model for malignant gastric lesions consisted of eight predictors, namely advanced age, male gender, family history of GC, low BMI, unexplained weight loss, consumption of leftover food, consumption of preserved food, and epigastric pain, as shown in Table 2.
Table 2.
Predictors and regression coefficients for the diagnostic model for malignant gastric lesions generated from the development set.
| Predictors* | Controls/Cases | Univariable OR (95% CI) | Multivariable OR‡ (95% CI) | Multivariable coefficients (95% CI) |
|---|---|---|---|---|
| Age † | – | 1.10 (1.07–1.13) | 1.10 (1.07–1.13) | 0.09 (0.06–0.12) |
| Gender † | ||||
| Female | 9943/30 | Reference | Reference | Reference |
| Male | 7295/92 | 4.18 (2.77–6.32) | 4.24 (2.79–6.43) | 1.44 (1.03–1.86) |
| Family history of GC † | ||||
| No | 16,388 /115 | Reference | Reference | Reference |
| Yes | 850/7 | 1.17 (0.55–2.52) | 1.24 (0.57–2.69) | 0.22 (–0.56 to 0.99) |
| BMI (kg/m2) | ||||
| ≥18.5 | 16,664 /110 | Reference | Reference | Reference |
| <18.5 | 574/12 | 3.17 (1.74–5.78) | 2.27 (1.22–4.23) | 0.82 (0.20–1.44) |
| Unexplained weight loss | ||||
| No | 11,679 /59 | Reference | Reference | Reference |
| Yes | 5559/63 | 2.24 (1.57–3.20) | 1.97 (1.36–2.84) | 0.68 (0.31–1.05) |
| Consumption of leftover food | ||||
| No | 7434/36 | Reference | Reference | Reference |
| Yes | 9804/86 | 1.81 (1.23–2.68) | 1.46 (0.98–2.18) | 0.38 (–0.02 to 0.78) |
| Consumption of preserved food (times/week) | ||||
| <1 | 9835/46 | Reference | Reference | Reference |
| ≥1 | 7403/76 | 2.19 (1.52–3.17) | 1.82 (1.25–2.65) | 0.60 (0.22–0.97) |
| Epigastric pain | ||||
| No | 8337/46 | Reference | Reference | Reference |
| Yes | 8901/76 | 1.55 (1.07–2.23) | 1.42 (0.98–2.07) | 0.35 (–0.02 to 0.73) |
| Intercept | – | – | – | –12.21 (–13.93 to –10.49) |
*Predictors were selected by a two-step selection method in which all candidate predictors were first evaluated in univariable logistic regression models, and variables with P <0.05 or P <0.5 and OR >1.3 were subjected to multivariable logistic regression analysis where the AIC was used to determine the final predictor pattern. †Age, gender, and family history of GC were constrained within the final model. ‡ OR = exp (coefficient). AIC: Akaike information criterion; BMI: Body mass index; CI: Confidence interval; GC: Gastric cancer; OR: Odds ratio.
Our model achieved an AUC of 0.791 (95% confidence interval [CI]: 0.750–0.831) for prediction of malignant gastric lesions in the full development set [Figure 1A]. Internal validation showed that this model worked well in both the northern outpatient cohort (AUC: 0.799, 95% CI: 0.718–0.881) and the southern outpatient cohort (AUC: 0.768, 95% CI: 0.718–0.817).
Figure 1.
ROC curves of the prediction model for malignant gastric lesions in (A) the development set, and (B) the validation set. AUC: Area under the curve; CI: Confidence interval; ROC: Receiver operating characteristic.
External validation
External validation of our model in a general population suggested that our model also has an ideal accuracy to diagnose prevalent cases (malignant gastric lesions diagnosed within 1-year's follow-up, AUC: 0.696, 95% CI: 0.570–0.822) [Figure 1B]. When used to predict incident cases in which malignant gastric lesions developed after 1-year follow-up, the performance of the model declined (AUC: 0.614, 95% CI: 0.526–0.701).
Application performance
To test the application performance of our model for precise referral, we hypothesized different screening coverages to select "high-risk" individuals and calculated the increase in the detection rate [Table 3]. There were similar trends in both the outpatient cohort and community-based screening cohort, and as the cut-off value rose, the detection rate increased notably [Figure 2], and the number of endoscopies needed to detect one case decreased significantly. When we expected to cover greater numbers of cases, such as 80% of all cases, we needed only to screen the top 40% in the development set (2-fold increase in detection rate, from 0.70% to 1.43%) or the top 60% in the validation set (1.4-fold increase in detection rate, from 0.06% to 0.08%). If only the top 5% of all individuals received endoscopy, the detection rate would be 5–6 times higher than universal screening and the number of endoscopies required to detect one case would be reduced from 142 to 22 in the development set.
Table 3.
Application performance of the diagnostic model for different screening coverages to detect malignant gastric lesions.
| Cut-offs* | Development set (outpatient cohort, n = 17,360 ) | Validation set† (community-based screening cohort, n = 32,452 ) | |||||
|---|---|---|---|---|---|---|---|
| Screening coverage, % (sensitivity in the screening, %) | Average N of endoscopies to detect one case | Detection rate (ratio) in the screening, % (compared to universal screening) | Screening coverage, % (sensitivity in the screening, %) | Average N of endoscopies to detect one case | Detection rate (ratio) in the screening, % (compared to universal screening) | ||
| 0.0244586 | 5.0 (32.8) | 22 | 4.61 (6.6) | 1.5 (11.1) | 239 | 0.42 (7.5) | |
| 0.0165160 | 10.0 (45.1) | 32 | 3.16 (4.5) | 4.5 (16.7) | 484 | 0.21 (3.7) | |
| 0.0098350 | 20.0 (63.9) | 45 | 2.23 (3.2) | 13.6 (33.3) | 737 | 0.14 (2.4) | |
| 0.0066909 | 30.0 (72.1) | 59 | 1.69 (2.4) | 22.9 (55.6) | 744 | 0.13 (2.4) | |
| 0.0048232 | 40.0 (81.1) | 70 | 1.43 (2.0) | 31.4 (61.1) | 927 | 0.11 (1.9) | |
| 0.0035823 | 50.0 (85.2) | 84 | 1.20 (1.7) | 41.2 (72.2) | 1029 | 0.10 (1.8) | |
| 0.0026978 | 60.0 (90.2) | 95 | 1.06 (1.5) | 50.1 (77.8) | 1160 | 0.09 (1.6) | |
| 0.0019726 | 70.0 (94.3) | 106 | 0.94 (1.3) | 60.9 (83.3) | 1317 | 0.08 (1.4) | |
| 0.0013741 | 80.0 (96.7) | 118 | 0.85 (1.2) | 72.8 (88.9) | 1476 | 0.07 (1.2) | |
| 0.0008817 | 90.0 (98.4) | 130 | 0.77 (1.1) | 83.9 (94.4) | 1602 | 0.06 (1.1) | |
| 0.0003106 | 100.0 (100) | 142 | 0.70 (Reference) | 100.0 (100) | 1803 | 0.06 (Reference) | |
*Cut-offs were selected at the highest predicted probabilities that ensured corresponding screening coverage in the development set. †Individuals with missing values were excluded. These were the cases that developed GC within a 1-year's follow-up. GC: Gastric cancer.
Figure 2.
Detection rate for malignant gastric lesions in the development and validation sets.
Discussion
In this study, we developed a simple risk stratification model for the diagnosis of malignant gastric lesions based on multicenter real-world outpatient cohorts, and successfully validated the predictive power of this model in a community-based screening cohort. This questionnaire-based model, which served as an easy-to-use and low-cost tool for the opportunistic screening of GC and achieved good diagnostic accuracy and generalizability in multiple source samples, might contribute to risk stratification, precise endoscopy referral at the prescreening stage.
In the past decades, a number of biomarkers have been found to be associated with GC. Among these, H. pylori has been considered the most important one.[5] However, for opportunistic screening carried out in outpatient clinics, it is almost impractical for a non-gastroenterologist to prescribe blood tests for all patients before making a referral decision. The economic and time costs of biomarker tests greatly limited their generalizability as a prescreening tool in opportunistic screening. Therefore, a stratification tool which is quicker, lower-cost, and easier-to-use may be ideal for opportunistic screening of GC.
In recent years, there have been an increasing number of prediction models developed for GC.[18–23] But most of these models have been prognostic models which estimate the risk of developing GC in the long term. To our knowledge, there have been only four models available which predict the prevalent risk of GC.[19–22] However, the requirement of these models for extra serum examination, or the small sample size of these studies, has largely limited their applicability and generalizability in large-scale screening programs. For example, the model established by Cai et al[19] requires blood testing for H. pylori, pepsinogen, and gastrin, and the only model that included questionnaire-based predictors was derived from a case-control design with only 382 patients.[20] In contrast, our questionnaire-based model, which is based on large multicenter cohorts, would have greater potential for employment in real-world opportunistic screening practice.
A total of eight predictors were preserved in the final model for malignant gastric lesions. Among these predictors, advanced age, male gender, and family history of GC were constrained into the final model because concrete evidence has been provided concerning their associations of these predictors with GC.[24,25] Other predictors, including low BMI, consumption of preserved food, and consumption of leftover food were also known risk factors for GC and the effect sizes we observed were all similar to those in previous studies.[26,27]
In addition to the above common risk factors, two upper GI symptoms were selected as predictors, namely unexplained weight loss and epigastric pain. These factors have also been previously reported as having a predictive effect on GC.[28,29] Incorporating symptoms into a screening tool seemed inconsistent with the traditional concept of screening, where the purpose of a screening test is to detect early-stage diseases among asymptomatic individuals. However, patients in the preclinical stage were not always asymptomatic for specific populations and specific diseases. Residents may not pay close attention to their symptoms, and gastroenterologists therefore often fail to refer patients to endoscopy in a timely manner because symptoms are not always specific at an early stage.[11,30] Therefore, we believe that incorporation of two main symptom indicators would better fit the opportunistic screening and would also contribute to a "downstaging" effect by alerting the patients and physicians to conduct an endoscopic examination as early as possible.[10,12]
In the present study, the model achieved good discrimination in the development set, which was composed of samples from two regions that are non-high-risk regions for GC with a certain degree of heterogeneity. We conducted internal validation in each sub-cohort, and considerable robustness and consistency were observed in the two datasets. In view of the fact that compared with the general population, outpatients from endoscopic clinics may have different characteristics. We further validated the model externally in the ESECC population, which was a community-based sampling cohort. It turned out that our model also worked well in the general population, especially for the diagnosis of screening detected GC cases, and cases which occurred within 1 year after enrollment. We then also tried to use our model to predict the risk of malignant cardia lesions, as GC and cancer of the cardia are in close proximity anatomically and demonstrate similar pathological subtypes. The results suggested that our model achieved a high degree of accuracy in both the development and validation sets [Supplementary Figure 1, http://links.lww.com/CM9/B783]. Endoscopic examination can be used to screen the esophagus and stomach (including the cardia of the stomach) in the same examination. Thus, by integrating the prediction model for esophageal cancer, it may be possible to develop a multicancer screening strategy, and this approach has attracted a great deal of interest in recent years.[12,31]
Although opportunistic screening is naturally a more cost-effective screening modality, risk enrichment prior to endoscopy is still important for increasing the cancer detection rate and avoiding excessive use of medical resources. To determine the criteria for identifying high-risk patients, several factors should be considered carefully. These factors include resource availability, population coverage, and the volume of endoscopic examination in given local health institutions. Therefore, we hypothesized different screening coverages to test the application performance of our model. When the aim was to achieve higher public health benefits using as few resources as possible, we would recommend a higher risk cut-off value, for example, 0.02 to screen 5% of the whole population. In this situation, our model achieved a detection rate of 4.61% (vs. 0.70% for universal screening, an increase of 6.6-fold), and reduced the number of endoscopies needed to detect one malignant gastric lesion from 142 to 22 in the development set. In contrast, if resources were adequate to cover a greater number of cases and achieved higher sensitivity, a lower cut-off value might be adopted, for example, 0.005 to screen 40% of the population in the development set. As such, over 80% of all cases would be detected with a detection rate of 1.43% (vs. 0.70% for universal screening, an increase of 2-fold). Thus, our model is adaptable to various situations with excellent application performance for the opportunistic screening of GC.
In a previous study, we constructed a GC risk scoring system (GC-RSS) to identify individuals at high risk.[23] The items in the GC-RSS were risk factors that had been reported in at least two studies, and the weighting in GC-RSS was determined according to the corresponding coefficients in previous reports. GC-RSS is a relatively rough but simple tool, which requires no complex calculations. In contrast, the model in the present study is based on statistical analysis and aims at more accurate and personalized risk estimation, and more precise risk stratification. Therefore, these two sets of models are appropriate for two different sets of circumstances. One is for simple evaluations at any time or in any place with limited resources to obtain an instant decision for endoscopic referral. For this set, GC-RSS would be a more feasible and practical tool. The other set is to obtain a more accurate risk estimation with the assistance of a computer to stratify the population, and the model in the present study would play a better role.
There are some limitations in the present study. First, all subjects in this study were recruited from non-high-incidence areas of GC. The generalizability of our model should therefore be further validated in high-risk regions. Second, in theory, ideal opportunistic screening means targeting all patients seeking medical service in a health care institution. However, it is impractical to carry out such studies in a 100% pure "opportunistic screening population" in the real world. Further validation is thus needed in non-gastroenterology outpatients.
In summary, we have successfully developed and validated a questionnaire-based diagnostic model for malignant gastric lesions. This model may serve as an effective prescreening tool for the opportunistic screening of GC to assist in rapid decision-making regarding endoscopy referral. With the advent of the 5G era, we look forward to integrating our model with smart portable terminals and social media platforms to increase the convenience of its application, and facilitate clinical or community-based opportunistic screening for GC.
Funding
This work was supported by grants from the National Science and Technology Fundamental Resources Investigation Program of China (No. 2019FY101102), the National Natural Science Foundation of China (No. 82073626), the National Key Research and Development Program of China (No. 2021YFC2500405), and the Sanming Project of Shenzhen (No. SZSM201612061).
Conflicts of interest
None.
Supplementary Material
Footnotes
Hongchen Zheng, Zhen Liu, Yun Chen, Ping Ji, Zhengyu Fang, and Yujie He contributed equally to this work.
How to cite this article: Zheng HC, Liu Z, Chen Y, Ji P, Fang ZY, He YJ, Guo CH, Xiao P, Wang CW, Yin WH, Li FL, Chen XJ, Liu MF, Pan YQ, Liu FF, Liu Y, He ZH, Ke Y. Development and external validation of a quantitative diagnostic model for malignant gastric lesions in clinical opportunistic screening: A multicenter real-world study. Chin Med J 2024;137:2343–2350. doi: 10.1097/CM9.0000000000002903
References
- 1.Xia C, Dong X, Li H, Cao M, Sun D, He S, Yang F, Yan X, Zhang S, Li N, Chen W. Cancer statistics in China and United States, 2022: profiles, trends, and determinants. Chin Med J 2022;135: 584–590. doi: 10.1097/CM9.0000000000002108. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Zheng R Zhang S Zeng H Wang S Sun K Chen R, et al. Cancer incidence and mortality in China, 2016. J Natl Cancer Cent 2022;2: 1–9. doi: 10.1016/j.jncc.2022.02.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Zeng H Chen W Zheng R Zhang S Ji JS Zou X, et al. Changing cancer survival in China during 2003-15: A pooled analysis of 17 population-based cancer registries. Lancet Glob Health 2018;6: e555–e567. doi: 10.1016/s2214-109x(18)30127-x. [DOI] [PubMed] [Google Scholar]
- 4.Savoldi A, Carrara E, Graham DY, Conti M, Tacconelli E. Prevalence of antibiotic resistance in Helicobacter pylori: A systematic review and meta-analysis in World Health Organization regions. Gastroenterology 2018;155: 1372–1382.e17. doi: 10.1053/j.gastro.2018.07.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kim Y, Jun JK, Choi KS, Lee HY, Park EC. Overview of the National Cancer screening programme and the cancer screening status in Korea. Asian Pac J Cancer Prev 2011;12: 725–730. [PubMed] [Google Scholar]
- 6.Leung WK Wu MS Kakugawa Y Kim JJ Yeoh KG Goh KL, et al. Screening for gastric cancer in Asia: Current evidence and practice. Lancet Oncol 2008;9: 279–287. doi: 10.1016/s1470-2045(08)70072-x. [DOI] [PubMed] [Google Scholar]
- 7.Hamashima C; Systematic Review Group and Guideline Development Group for Gastric Cancer Screening Guidelines . Update version of the Japanese guidelines for gastric cancer screening. Jpn J Clin Oncol 2018;48: 673–683. doi: 10.1093/jjco/hyy077. [DOI] [PubMed] [Google Scholar]
- 8.Allemani C Matsuda T Di Carlo V Harewood R Matz M Nikšić M, et al. Global surveillance of trends in cancer survival 2000-14 (CONCORD-3): Analysis of individual records for 37 513 025 patients diagnosed with one of 18 cancers from 322 population-based registries in 71 countries. Lancet 2018;391: 1023–1075. doi: 10.1016/s0140-6736(17)33326-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wang GQ, Wei WW. A new transition of the screening, early diagnosis and early treatment project of the upper gastrointestinal cancer: Opportunistic screening (in Chinese). Chin J Prev Med 2019;53: 1084–1087. doi: 10.3760/cma.j.issn.0253-9624.2019.11.002. [DOI] [PubMed] [Google Scholar]
- 10.He Z, Ke Y. Precision screening for esophageal squamous cell carcinoma in China. Chin J Cancer Res 2020;32: 673–682. doi: 10.21147/j.issn.1000-9604.2020.06.01. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.He Z, Ke Y. Response. Gastrointest Endosc 2020;92: 1137–1138. doi: 10.1016/j.gie.2020.06.071. [DOI] [PubMed] [Google Scholar]
- 12.Liu Z Guo C He Y Chen Y Ji P Fang Z, et al. A clinical model predicting the risk of esophageal high-grade lesions in opportunistic screening: A multicenter real-world study in China. Gastrointest Endosc 2020;91: 1253–1260.e3. doi: 10.1016/j.gie.2019.12.038. [DOI] [PubMed] [Google Scholar]
- 13.Yamaguchi Y Nagata Y Hiratsuka R Kawase Y Tominaga T Takeuchi S, et al. Gastric cancer screening by combined assay for serum anti-Helicobacter pylori IgG antibody and serum pepsinogen levels – The ABC method. Digestion 2016;93: 13–18. doi: 10.1159/000441742. [DOI] [PubMed] [Google Scholar]
- 14.He Z Liu Z Liu M Guo C Xu R Li F, et al. Efficacy of endoscopic screening for esophageal cancer in China (ESECC): Design and preliminary results of a population-based randomised controlled trial. Gut 2019;68: 198–206. doi: 10.1136/gutjnl-2017-315520. [DOI] [PubMed] [Google Scholar]
- 15.Tian H Yang W Hu Y Liu Z Chen L Lei L, et al. Estimating cancer incidence based on claims data from medical insurance systems in two areas lacking cancer registries in China. EClinicalMedicine 2020;20: 100312. doi: 10.1016/j.eclinm.2020.100312. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Tian H Xu R Li F Guo C Zhang L Liu Z, et al. Identification of cancer patients using claims data from health insurance systems: A real-world comparative study. Chin J Cancer Res 2019;31: 699–706. doi: 10.21147/j.issn.1000-9604.2019.04.13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Shi C Liu M Liu Z Guo C Li F Xu R, et al. Using health insurance reimbursement data to identify incident cancer cases. J Clin Epidemiol 2019;114: 141–149. doi: 10.1016/j.jclinepi.2019.06.009. [DOI] [PubMed] [Google Scholar]
- 18.Gu J Chen R Wang SM Li M Fan Z Li X, et al. Prediction models for gastric cancer risk in the general population: A systematic review. Cancer Prev Res (Phila) 2022;15: 309–318. doi: 10.1158/1940-6207.Capr-21-0426. [DOI] [PubMed] [Google Scholar]
- 19.Cai Q Zhu C Yuan Y Feng Q Feng Y Hao Y, et al. Development and validation of a prediction rule for estimating gastric cancer risk in the Chinese high-risk population: A nationwide multicentre study. Gut 2019;68: 1576–1587. doi: 10.1136/gutjnl-2018-317556. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Lee DS Yang HK Kim JW Yook JW Jeon SH Kang SH, et al. Identifying the risk factors through the development of a predictive model for gastric cancer in South Korea. Cancer Nurs 2009;32: 135–142. doi: 10.1097/NCC.0b013e3181982c2e. [DOI] [PubMed] [Google Scholar]
- 21.Qiu L Qu X He J Cheng L Zhang R Sun M, et al. Predictive model for risk of gastric cancer using genetic variants from genome-wide association studies and high-evidence meta-analysis. Cancer Med 2020;9: 7310–7316. doi: 10.1002/cam4.3354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Tu H Sun L Dong X Gong Y Xu Q Jing J, et al. A serological biopsy using five stomach-specific circulating biomarkers for gastric cancer risk assessment: A multi-phase study. Am J Gastroenterol 2017;112: 704–715. doi: 10.1038/ajg.2017.55. [DOI] [PubMed] [Google Scholar]
- 23.Zhou R Zheng H Liu M Liu Z Guo C Tian H, et al. Development and validation of a questionnaire-based risk scoring system to identify individuals at high risk for gastric cancer in Chinese populations. Chin J Cancer Res 2021;33: 649–658. doi: 10.21147/j.issn.1000-9604.2021.06.02. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Choi IJ Kim CG Lee JY Kim YI Kook MC Park B, et al. Family history of gastric cancer and Helicobacter pylori treatment. N Engl J Med 2020;382: 427–436. doi: 10.1056/NEJMoa1909666. [DOI] [PubMed] [Google Scholar]
- 25.Oliveira C, Pinheiro H, Figueiredo J, Seruca R, Carneiro F. Familial gastric cancer: Genetic susceptibility, pathology, and implications for management. Lancet Oncol 2015;16: e60–e70. doi: 10.1016/s1470-2045(14)71016-2. [DOI] [PubMed] [Google Scholar]
- 26.Karimi P, Islami F, Anandasabapathy S, Freedman ND, Kamangar F. Gastric cancer: Descriptive epidemiology, risk factors, screening, and prevention. Cancer Epidemiol Biomarkers Prev 2014;23: 700–713. doi: 10.1158/1055-9965.Epi-13-1057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ren JS, Kamangar F, Forman D, Islami F. Pickled food and risk of gastric cancer – A systematic review and meta-analysis of English and Chinese literature. Cancer Epidemiol Biomarkers Prev 2012;21: 905–915. doi: 10.1158/1055-9965.Epi-12-0202. [DOI] [PubMed] [Google Scholar]
- 28.Mansoor W Roeland EJ Chaudhry A Liepa AM Wei R Knoderer H, et al. Early weight loss as a prognostic factor in patients with advanced gastric cancer: Analyses from REGARD, RAINBOW, and RAINFALL phase III studies. Oncologist 2021;26: e1538–e1547. doi: 10.1002/onco.13836. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Park YS Park DJ Lee Y Park KB Min SH Ahn SH, et al. Prognostic roles of perioperative body mass index and weight loss in the long-term survival of gastric cancer patients. Cancer Epidemiol Biomarkers Prev 2018;27: 955–962. doi: 10.1158/1055-9965.Epi-18-0122. [DOI] [PubMed] [Google Scholar]
- 30.Wang J Liu F Gao H Wei W Zhang X Liang Y, et al. The symptom-to-treatment delay and stage at the time of treatment in cancer of esophagus. Jpn J Clin Oncol 2008;38: 87–91. doi: 10.1093/jjco/hym169. [DOI] [PubMed] [Google Scholar]
- 31.Doubeni CA, Lau YK, Lin JS, Pennello GA, Carlson RW. Development and evaluation of safety and effectiveness of novel cancer screening tests for routine clinical use with applications to multicancer detection technologies. Cancer 2022;128(Suppl 4): 883–891. doi: 10.1002/cncr.33954. [DOI] [PMC free article] [PubMed] [Google Scholar]