Article Text

Original research
Risk stratifying gastric ulcers: development and validation of a scoring system
  1. William M Brindle1,
  2. Rebecca K Grant1,
  3. Marianne Smith1,
  4. Meghan Suddaby1,
  5. Angus Wallace1,
  6. Sarah-Louise Gillespie1,
  7. Nicholas I Church1,
  8. Colin L Noble2,
  9. Ian D Penman1,
  10. John N Plevris1,
  11. Alexander R Robertson2,
  12. Eleanor F Watson2,
  13. Christian P Selinger3,
  14. Rahul Kalla1,
  15. Gail S M Masterton1
  16. On behalf of EGAR (Edinburgh GI Audit and Research) Collaborative
  1. 1 The Centre for Liver and Digestive Disorders, Royal Infirmary of Edinburgh, Edinburgh, UK
  2. 2 The Edinburgh IBD Unit, Western General Hospital, Edinburgh, UK
  3. 3 Leeds Gastroenterology Institute, Leeds Teaching Hospitals NHS Trust, Leeds, UK
  1. Correspondence to Dr William M Brindle, The Centre for Liver and Digestive Disorders, Royal Infirmary of Edinburgh, Edinburgh EH16 4SA, UK; william.m.brindle{at}


Objective Debate is ongoing regarding the need for universal endoscopic follow-up to ensure gastric ulcer healing. We aimed to assess the value of follow-up oesophago-gastro-duodenoscopies (OGDs) for gastric ulcer healing and stratify patients according to risk of malignancy by developing a risk score.

Design/method All patients in National Health Service (NHS) Lothian with an index OGD and a diagnosis of gastric ulcer between 1 January 2014 and 31 December 2018 were identified. Data were analysed with logistic regression to identify factors significantly associated with a diagnosis of cancer; a risk score was derived and externally validated.

Results 778 patients were identified and 60.3% (469/778) of patients had a follow-up OGD. 8.6% (66/778) of patients were diagnosed with cancer. No cases of cancer were found on follow-up OGD of a benign appearing ulcer with negative biopsies. Macroscopic suspicion of malignancy was present at index OGD in 100% (3/3) of those diagnosed with cancer on subsequent OGDs. Older age (p=0.014), increased ulcer size (p<0.001) and non-antral location (p=0.030) were significantly associated with malignancy. A risk score (area under the curve (AUC) 0.868, p<0.001, minimum score=0, maximum score=6) was derived from these variables. 78.0% of patients with malignant ulcers scored ≥3, only 15.8% with benign ulcers scored ≥3 (negative predictive value (NPV) 97.4%). External validation yielded an AUC of 0.862 (p<0.001) and NPV of 98.6%; 84.0% of those with malignant ulcers scored ≥3.

Conclusion Ulcers with a combination of macroscopically benign appearances, at least six negative biopsies and a low risk score do not necessarily need endoscopic follow-up.

  • gastrointestinal cancer
  • gastric and duodenal ulcers
  • gastrointesinal endoscopy

Data availability statement

Deidentified data available on request via corresponding author.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Significance of this study

What is already known on this topic

  • A significant proportion of gastric ulcers diagnosed at endoscopy are in fact gastric cancers.

  • Current UK guidelines advise repeat endoscopy in all patients to check for ulcer healing despite some studies questioning the value of this approach in low-risk populations.

  • There are significant pressures on endoscopy services that have been exacerbated by the current COVID-19 pandemic.

What this study adds

  • 8.5% of endoscopically diagnosed ulcers were found to be malignant.

  • No cancers were diagnosed on designated follow-up oesophago-gastro-duodenoscopy of a benign appearing ulcer with negative index biopsies.

  • We propose the first risk stratification score for gastric ulcers (significant predictors of outcome: ulcer location, size and patient age).

How might it impact on clinical practice in the foreseeable future

  • We suggest that ulcers with a combination of macroscopically benign appearances, at least six negative biopsies and a low risk score do not necessarily need endoscopic follow-up.


The association between gastric ulcers and gastric malignancy is well known. Guidelines therefore suggest repeat oesophago-gastro-duodenoscopies (OGDs) to ensure ulcer healing in all gastric ulcers1–4 and a follow-up OGD after 6–8 weeks is the target within many UK centres.

However, there is debate regarding the value of repeat OGDs, especially in low-risk populations such as Scotland, where gastric cancer incidence has fallen by 33.0% in 10 years.5 Studies supporting ulcer follow-up are mostly historical,6 7 with the exception of one conducted between 1995 and 2003,8 which reported that following up 50 benign appearing ulcers would identify one malignancy. A larger 2016 UK study9 concluded that all gastric malignancies would have been diagnosed on initial OGD alone; a combination of benign index histology and no endoscopic suspicion of malignancy excluded cancer. Other studies10–12 have shown similar results and suggested a policy of universal follow-up is unlikely to be beneficial. The cost-effectiveness of follow-up OGDs has consequently been questioned, with one calculation suggesting that the prevalence of cancer would have to be 6% in apparently benign gastric ulcers to be cost-effective.13

There is a need to reduce demand on stretched endoscopy services to facilitate shorter waiting times for higher yield investigations; this has been heightened by the current COVID-19 pandemic. There is also an opportunity to reduce unnecessary procedure-related risk, patient discomfort/anxiety and associated healthcare costs.

The aim of this study was to assess the value of follow-up OGDs (OGD2) for gastric ulcer healing in our population (low to intermediate risk—incidence 10.8 per 100 000 population5) and to stratify patients according to risk of malignancy through the derivation and validation of a novel risk score.


Derivation cohort

A retrospective search (April–June 2020) of our endoscopy reporting software (Unisoft 14.66.00) for reports with the gastric ulcer option selected was conducted. All patients with an index OGD (OGD1) diagnosis of a gastric ulcer within National Heath Service (NHS) Lothian between 1 January 2014 and 31 December 2018 were identified. There are approximately 11 250 OGDs performed annually14 (inpatient and outpatient) across the health board. OGDs are carried out by consultant gastroenterologists, upper gastrointestinal (GI) surgeons, specialty registrars and nurse endoscopists.

All patients with a gastric ulcer present on OGD1 were included. Any patients who died within 1 year of OGD1 without a diagnosis of gastric cancer and any who were not normally resident within NHS Lothian were excluded.

Data gathered from the OGD reports included age, sex, indication, ulcer size (as estimated by the endoscopist), ulcer location, number of ulcers and whether the macroscopic appearance was concerning for malignancy. Data were collected on biopsies taken, histology and survival from the local healthcare information system (TrakCare). Identical information was collated for subsequent endoscopies. Helicobacter pylori status, non-steroidal anti-inflammatory drug (NSAID) or aspirin use were also recorded where available.


‘OGD1’ refers to the index OGD at which an ulcer was first diagnosed. ‘OGD2’ refers to the follow-up OGD.

If a patient required a second OGD because indeterminate pathology suggested an early repeat OGD or, if the OGD was done a few days after OGD1 to acquire biopsies of a previously bleeding ulcer, this is referred to as OGD1b.

A benign ulcer was defined as having non-malignant histology and/or the patient had not re-presented with a gastric malignancy (minimum follow-up time of 12 months for patients that had died, minimum of 18 months for all other patients).

Ulcers that were present within either the antrum or pre-pylorus were defined as ‘antral ulcers’.

Evidence of H. pylori infection was defined as Campylobacter-like organism test positive or positive on histology.

Statistical analysis

IBM SPSS Statistics Subscription (Build and MedCalc (2020) were used for all statistical analysis. Descriptive statistics are presented as median values and IQR for continuous variables and percentages with frequencies for categorical variables. The Mann-Whitney U test compared non-parametric continuous variables; the χ2 test compared categorical variables. Multivariable logistic regression identified predictors of malignancy. Variables with a p value <0.05 on univariable analysis were fitted and the enter method, in which all input variables were entered simultaneously, was used to identify significant predictors.

Risk score development

Receiver operating characteristic (ROC) curve analyses were performed on significant continuous variables and primary cut-off values were identified using the Youden Index. Categorised continuous variables were re-entered into the multivariable logistic regression; secondary cut-off values were identified by prioritising specificity and identifying values of clinical significance.15 16 A score was attributed to each variable of predictive significance based on the regression coefficient in the logistic regression. Antral ulcers were attributed a score of 0, non-antral ulcers a score of 1. Age <68, 68–79, ≥80 were attributed scores of 0, 1 and 2, respectively. Ulcer size <1.25 cm, 1.25–2.99 cm, ≥3.00 cm were attributed scores of 0, 2 and 3, respectively. Each patient was attributed a total score (minimum possible score=0; maximum possible score=6) and the result compared with outcome (benign or malignant).

A p value of <0.05 was considered significant in all statistical analysis.

Validation cohort

The validation cohort consisted of all patients with a first endoscopic diagnosis of gastric ulcer between 1 January 2012 and 30 September 2013 at Leeds Teaching Hospitals NHS Trust, Leeds, UK.9 The inclusion and exclusion criteria were identical to those in the derivation cohort.

The performance of the risk score was tested on the validation cohort using the same categories and cut-offs as those in the derivation cohort. Access to the validation cohort data was not gained until after the development of the risk score.


Derivation cohort

Patient demographics

In total, 925 patients were identified with a diagnosis of gastric ulcer; 778 patients satisfied the inclusion criteria (84.1%). A total of 110 patients died within 1 year of OGD1 without a pre-morbid diagnosis of gastric cancer and were excluded. Twenty-two patients were excluded as they were not resident within the area and 15 were excluded as, on review of the free text of the OGD reports, it was clear that the lesion was not an ulcer (online supplemental figure 1). Median age at OGD1 was 63 years (IQR 52–74) and 49.7% of patients were men (table 1).

Supplemental material

Table 1

Derivation cohort—background demographics and OGD results in 778 patients with a diagnosis of gastric ulcer

Clinical outcomes

A macroscopic suspicion of malignancy was reported in 7.6% of endoscopies, although only 15.2% explicitly reported that the ulcer looked benign. Ulcer size was documented in 55.7% (433/778). Of the 77.2% (601/778) patients who had H. pylori status documented, 31.4% (189/601) had evidence of H. pylori infection (table 1). 28.3% (220/778) of patients had no biopsies of their ulcer during OGD1. The OGD was performed to investigate possible bleeding in only 55.5% of these patients.

Only 60.3% (469/778) of patients had a follow-up OGD. Of those not receiving a second OGD, the majority had no reason documented in either the OGD report or in the medical records (table 1).

The median interval between OGD1 and OGD2 was 18 weeks (IQR 8.0–47.5 weeks). Sixty-six patients (8.5%) were found to have a cancer, of which the majority were adenocarcinoma (78.8%). The majority of cancers (95.5%) were diagnosed on OGD1 or OGD1b.

Predictors of malignancy

Univariate analysis was performed on continuous and categorical variables for all patients with complete data on ulcer size (n=433) (online supplemental figure 1). Three variables were significant in predicting malignancy: older age (p<0.001), increased ulcer size (p<0.001) and non-antral ulcers (p<0.001). H. pylori status, sex, gastric ulcer number, prior aspirin or NSAID treatment and indication for index OGD were not significant in predicting malignancy (table 2).

Table 2

Derivation cohort—statistical analysis: univariate analysis—predictors of malignancy

On multivariable logistic regression analysis, older age, increased ulcer size and non-antral ulcers were independently associated with malignancy (p=0.014, <0.001, 0.030, respectively) (table 3).

Table 3

Derivation cohort—statistical analysis: multivariable logistic regression analysis—predictors of malignancy

Risk score

ROC curves were generated for age (area under the curve (AUC) 0.701, 95% CI 0.624 to 0.778, p<0.001) and ulcer size (AUC 0.861, 95% CI 0.792 to 0.931, p<0.001). Cut-off values of age ≥68 years (sensitivity 70.7%, specificity 65.9%) and ulcer size ≥1.25 cm (sensitivity 82.9%, specificity 81.4%) were identified. Secondary cut-off values of age ≥80 years (sensitivity 29.3%, specificity 88.8%) and ulcer size ≥3.00 cm (sensitivity 48.5%, specificity 95.7%) were also identified. Age was categorised as <68 years, 68–79 years and ≥80 years and ulcer size as <1.25 cm, 1.25–2.99 cm and ≥3.00 cm (online supplemental table 1). The categories were re-entered into the multiple logistic regression model (table 3) and scores attributed. ROC curve analysis of this risk score produced an AUC of 0.868 (95% CI 0.800 to 0.936, p<0.001) (figure 1). A cut-off value of 3/6 (sensitivity 72.0%, specificity 89.3%) was selected by application of the Youden Index to identified patients at highest risk of malignancy. Seventy-eight per cent (32/41) of patients with malignant ulcers had a score ≥3; only 15.8% (62/392) of patients with benign ulcers had a score ≥3 (sensitivity 78.1%, specificity 84.2%, positive predictive value (PPV) 34.0%, negative predictive value (NPV) 97.4%, accuracy 83.6%) (figure 2, online supplemental table 2).

Supplemental material

Supplemental material

Figure 1

ROC curve analysis of novel risk score in derivation and validation cohorts. AUC, area under the curve; ROC, receiver operator characteristic.

Figure 2

Distribution of scoring.

Patients with cancer diagnosis at or after OGD2

Only three patients (4.5%) with gastric cancer were not diagnosed on OGD1 or OGD1b (online supplemental table 3). Macroscopic suspicion of malignancy was reported in two of these patients. A third patient did not have macroscopic concern reported on OGD1, however, on review of the endoscopic images by four consultant gastroenterologists, there was a consensus that the grossly thickened pylorus was suspicious for malignancy.

Supplemental material

Validation cohort

Three hundred and eighty-five patients were included in the validation cohort (360 benign ulcers, 25 malignant ulcers). Forty-seven patients were excluded due to missing data regarding ulcer size. The novel score was applied; ROC curve analysis produced an AUC of 0.862 (95% CI 0.777 to 0.948, p<0.001) (figure 1). Eighty-four per cent (21/25) of patients with malignant ulcers had a score ≥3; only 21.4% (77/360) of patients with benign ulcers had a score ≥3 (sensitivity 84.0%, specificity 78.6%, PPV 21.4%, NPV 98.6%, accuracy 79.0%) (figure 2, online supplemental table 2). 88.9% (8/9) of patients with cancer diagnosed at or after OGD2 scored ≥3 on OGD1.


We report a novel GI ulcer score to risk stratify patients at index OGD with a high NPV and validate our findings in an independent cohort from Leeds.9 Our data suggest that a repeat OGD for ulcer healing is not necessary for all gastric ulcer patients in our population. Of the 778 patients only two were diagnosed with a cancer on designated follow-up OGD and, in both cases, there was a suspicion of malignancy based on macroscopic appearances at OGD1. This finding is now based on follow-up of over 1200 patients across two separate regions of the UK.

These findings do, however, disagree with other studies such as that of Hopper et al.8 Their study included OGDs undertaken between 1995 and 2003; in 277 patients, 12 gastric cancers were diagnosed on OGD2 that all had had benign ulcer appearances and negative biopsies on OGD1. This disparity may stem from (a) their exclusion criteria (267 patients were excluded due to being elderly or suffering from suspected alternative aetiologies), (b) a higher risk population or (c) advances in endoscopic technologies. The number of biopsies or number of endoscopists in the study was also not clear.

Our proposed risk score is based on patient age, ulcer size and ulcer location. It offers endoscopists the opportunity to recommend follow-up for higher scoring patients (larger, non-antral ulcers in older patients), assuming macroscopically benign appearances and adequate biopsies that have been reported as benign (figure 3). We report a score of ≥3 in 78.0% of patients with malignant ulcers (NPV 97.4%) in our derivation cohort, underlining the significant potential that the score has to inform clinical practice. At present, there are no risk scores routinely used in this clinical context.

Figure 3

Proposed algorithm for application of gastric ulcer risk score. GI, gastrointestinal; OGD, oesophago-gastro-duodenoscopy.

If a policy of no ulcer follow-up had been adopted (providing benign appearances and pathology), 469 OGDs could have been avoided over the previous 5 years. Assuming a cost of £560 per OGD,17 this is equivalent to a saving of £262 640. This saving rises to £400 400 assuming an OGD2 rate of 100%. If a more cautious approach is taken in which repeat OGDs are requested for those with a score of ≥3, 5-year savings of approximately £133 280 (based on actual OGD2s performed), or £189 840 (assuming a 100% follow-up rate) may be gained. Follow-up OGDs should still be arranged if there are ongoing symptoms or any features that suggest rarer aetiologies, such as a family history of inflammatory bowel disease, or ulcers of atypical appearance.

Our reported experience of ulcer follow-up is poor. In the majority of cases there is no documented reason for the lack of follow-up. It is possible that undocumented pragmatism in elderly patients and/or scepticism about the value of ulcer follow-up may have influenced the decision. The long interval to OGD2 (median 18 weeks) reflects current waiting list pressures and may reduce the value of performing OGD2 due to the increased chance of tumour progression beyond treatable criteria. Local quality improvement work (including adding a drop-down menu on the electronic OGD report request form) has allowed better identification and prioritisation of designated follow-up procedures.

We had a ‘no-biopsy’ rate of 28.3% (220 patients) on OGD1, which includes 98 patients who were not suspected of having a GI bleed. In some cases of a benign appearing ulcer, biopsies may not have been taken on OGD1 due to low suspicion of malignancy and the reassurance of a follow-up OGD. This practice risks significant delays in diagnoses of cancer; 17.5% of our patients did not attend their OGD2, and benign appearances alone are not sufficient to exclude a cancer, as in 34.8% of cancer patients in this cohort there was no documented concern for malignancy. We therefore strongly advise that all gastric ulcers are biopsied at OGD1, and if not, a repeat OGD should be performed as soon as possible to obtain biopsies.

There are several strengths to this study. The risk score is novel and the very high NPV indicates that it has the potential to make an important contribution to clinical decision making. The score has been validated in an external cohort and yielded a comparably excellent AUC. Furthermore, it has the largest cohort size of any recent similar study and improves on previous studies9 by recording data on H. pylori and NSAID status and by having a longer follow-up time (maximum follow-up period 6.5 years vs 3.8 years, median follow-up was 3.8 years (IQR 2.6–5.0)). It is also involved multisite data acquisition capturing the practice of a range of units/endoscopists.

There are limitations to our study. It is retrospective which does not allow for the standardisation of variables. Due to restrictions during the COVID-19 pandemic, validation was also retrospective; prospective multicentre validation is being considered as endoscopy services resume. Only 60.3% of our derivation cohort patients had an OGD2 to prove ulcer healing, however, our median follow-up time of 3.8 years (IQR 2.6–5.0) was considered appropriate for adequate data capture. The subjectivity of endoscopic and histological reporting is hard to control for, in particular, there is likely to be variation in their classification of lesions as erosions or ulcers, benign versus malignant appearance and in the endoscopist’s estimate of ulcer size. Missing data in relation to ulcer size reduced the number of patients included in the statistical analysis contributing to the risk score. However, given the widely reported11 15 18 strong association between ulcer size and malignancy, we felt that it was appropriate to prioritise patients with these data present.

We conclude that in our population a high-quality index OGD (including an adequate description of a macroscopically benign ulcer) combined with negative biopsies (at least 6) may reduce the requirement for a follow-up OGD. If a more cautious approach is preferred, the risk score criteria presented could be used such that only ulcers that are macroscopically suspicious for malignancy or have a risk score ≥3 receive endoscopic follow-up.

Data availability statement

Deidentified data available on request via corresponding author.

Ethics statements

Patient consent for publication

Ethics approval

In accordance with NHS Health Research Authority guidelines, specific ethical review and approval was not considered necessary as this research is a retrospective audit using data already obtained as part of regular clinical care.


We are grateful to Rebecca Cochrane, Sangeetha Thanaraj, Anita Sainsbury, Venkat Subramanian and Simon Everett for their contributions to the Leeds study used for score validation.


Supplementary materials


  • RK and GSMM are joint senior authors.

  • Twitter @gastronautian, @rahul_kalla

  • Contributors WMB contributed to the conception of the work, data collection, interpretation of the data and drafting of the manuscript. RKG contributed to data collection, analysis and interpretation of the data and performed critical revision of the manuscript. MSm, MSu, AW and S-LG contributed to data collection and critical revision of the manuscript. NIC, CLN, IDP, JNP, ARR and CPS performed critical revision of the manuscript. EFW contributed to data collection and performed critical revision of the manuscript. RK and GSMM were senior authors and contributed to the conception of the work and performed the critical revision of the manuscript for important intellectual content. Guarantor of article: WMB.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.