Admission Rockall Research Articles

Predicting mortality in patients with nonvariceal upper gastrointestinal bleeding using machine-learning.

Non-endoscopic risk scores, Glasgow Blatchford (GBS) and admission Rockall (Rock), are limited by poor specificity. The aim of this study was to develop an Artificial Neural Network (ANN) for the non-endoscopic triage of nonvariceal upper gastrointestinal bleeding (NVUGIB), with mortality as a primary outcome. Four machine learning algorithms, namely, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), logistic regression (LR), K-Nearest Neighbor (K-NN), were performed with GBS, Rock, Beylor Bleeding score (BBS), AIM65, and T-score. A total of 1,096 NVUGIB hospitalized in the Gastroenterology Department of the County Clinical Emergency Hospital of Craiova, Romania, randomly divided into training and testing groups, were included retrospectively in our study. The machine learning models were more accurate at identifying patients who met the endpoint of mortality than any of the existing risk scores. AIM65 was the most important score in the detection of whether a NVUGIB would die or not, whereas BBS had no influence on this. Also, the greater AIM65 and GBS, and the lower Rock and T-score, the higher mortality will be. The best accuracy was obtained by the hyperparameter-tuned K-NN classifier (98%), giving the highest precision and recall on the training and testing datasets among all developed models, showing that machine learning can accurately predict mortality in patients with NVUGIB.

AIMS65 predicts prognosis of patients with duodenal ulcer bleeding; a comparison with other risk-scoring systems.

Duodenal ulcer bleeding has a higher risk of mortality than bleeding from other portions of the gastrointestinal tract. AIMS65 is an effective risk-scoring system to predict prognosis of upper gastrointestinal bleeding and can be easily calculated without endoscopic findings. In this study, we investigate the usefulness of AIMS65 to predict prognosis of patients with duodenal ulcer bleeding. Two hundred and fifty-five patients with endoscopically diagnosed duodenal ulcer bleeding at Kurashiki Central hospital from July 2007 to June 2017 were studied. We compared AIMS65, Glasgow Blatchford score (GBS), admission Rockall, and full Rockall scoring systems for predicting in-hospital mortality by calculating area under the receiver operating characteristic curve (AUROC). In-hospital mortality due to duodenal ulcer bleeding occurred in 17 (6.7%). Scores of all scoring systems were significantly higher in patients with in-hospital mortality than in patients without it. AUROC values for predicting in-hospital mortality was 0.83 in AIMS65, 0.74 in GBS, 0.76 in admission Rockall score, and 0.82 in full Rockall score, a statistically insignificant difference among the systems. In AIMS65, score more than or equal to 2 was an optimal value to predict in-hospital mortality, with sensitivities of 88.2% and specificities of 59.7%, respectively. AIMS65 predicted in-hospital mortality of patients with duodenal ulcer bleeding as accurately as did other scoring systems. Given its simplicity of calculation, AIMS65 may be a more clinically practical system in the management of bleeding duodenal ulcer patients.

Tu1496 THE ABC SCORE ACCURATELY PREDICTS MORTALITY IN HOSPITALIZED PATIENTS THAT DEVELOP UPPER GASTROINTESTINAL BLEEDING

Hospitalized patients that develop upper gastrointestinal bleeding (UGIB) have different patient characteristics, etiology and worse outcome than patients admitted from the community with UGIB. The utility of risk scoring systems has not been validated in this patient group. The aim of this study is to compare the accuracy of different risk scoring systems in hospitalized patients that develop UGIB. Consecutive hospitalized patients that developed UGIB in Canterbury, New Zealand, between June 2015 and February 2019 were included. Patients who had onset of UGIB less than 24 hours from the time of admission were excluded. UGIB risk assessment scores (Glasgow Blatchford, AIM65, ABC, full Rockall, admission Rockall and PNED scores) were calculated and their abilities to predict predefined clinical endpoints: 30 day mortality, endoscopic intervention and a composite endpoint (30 day mortality or endoscopic intervention) were compared using area under the receiver operating curve (AUROC). By convention AUROC>0.8 is considered accurate. A total of 229 patients were included. The median patient age was 74 years (range 31-94), 63% were male, and median duration from hospitalization to onset of bleeding was 5 days (range 1-71). Forty-six (20%) required endoscopic intervention, 35 (15%) died within 30 days and median duration from onset of bleeding to death was 13 days (range 0-30). Baseline characteristics, endoscopy findings, interventions, outcomes, and mean risk assessment scores are shown in Table 1. The ABC score accurately predicted 30 day mortality (AUROC 0.85), compared with PNED score (AUROC 0.80, p=0.22), full Rockall score (AUROC 0.75, p<0.05), Glasgow Blatchford score (AUROC 0.71, p<0.05), and AIM65 score (AUROC 0.70, p<0.05) (Figure 1). Patients with an ABC score ≤3 had a 30 day mortality rate of 1.6%, compared to 7.5% and 42% for scores of 4-7 and ≥8, respectively. The need for endoscopic intervention and the composite endpoint (30 day mortality or need for endoscopic intervention) were not accurately predicted by the Glasgow Blatchford score (AUROC 0.76 and 0.76, respectively), ABC score (AUROC 0.68 and 0.76), AIM65 score (AUROC 0.61 and 0.65), or admission Rockall score (AUROC 0.58 and 0.67). No patients with a Glasgow Blatchford score of ≤1 required endoscopic intervention or died within 30 days of bleeding onset. The ABC score accurately predicts mortality in hospitalised patients that develop UGIB. A Glasgow Blatchford Score of ≤1 can be used to identify low risk patients that may not require urgent endoscopy. These results show that UGIB risk assessment scores are a useful tool in risk stratifying hospitalised patients that develop UGIB.View Large Image Figure ViewerDownload Hi-res image Download (PPT)

Comparison of the performance of risk scoring systems for tumor bleeding in patients with inoperable gastric cancer.

The Glasgow - Blatchford bleeding score (GBS) and admission and full Rockall scores are widely used risk scoring systems to stratify risk and determine the need for intervention in patients with upper gastrointestinal bleeding. We evaluated the performance of these risk scoring systems in patients with gastric cancer and tumor bleeding. This retrospective study included patients with inoperable gastric cancer who presented with tumor bleeding at the National Cancer Center, Korea, between 2001 and 2015. The GBS, and admission and full Rockall scores were calculated. Primary outcome was the performance of the risk scoring systems in predicting the need for urgent interventions (endoscopic therapy, transarterial embolization, and surgery). The risk factors associated with urgent intervention were analyzed. Of 357 patients with tumor bleeding, 118 (33.1 %; 116 endoscopic therapy, 2 trans-arterial embolization) required urgent intervention. The full Rockall score was better at predicting the need for urgent intervention (area under the receiver operating characteristic curve = 0.78; P < 0.001) than the GBS (0.56) and admission Rockall score (0.56). Hemostatic intervention was not performed in patients with a full Rockall score ≤ 6 (103 patients, 28.9 %). On multivariate analysis, endoscopic stigmata of recent hemorrhage of Forrest classes Ia - IIb were significant factors associated with urgent intervention. The full Rockall score was superior to the GBS and admission Rockall score in predicting the need for urgent intervention for tumor bleeding. Thus, endoscopic evaluation is required to determine the need for urgent intervention in patients with gastric cancer and tumor bleeding.

The Shock Index is not accurate at predicting outcomes in patients with upper gastrointestinal bleeding.

Acute upper gastrointestinal bleeding (UGIB) remains a major cause of hospital admission worldwide. The recent UK National Confidential Enquiry into Patient Outcome and Death (NCEPOD) report on severe gastrointestinal bleeding used the Shock Index to assess bleeding severity and found an association between Shock Index and mortality. However, this has never been prospectively validated as a predictor of outcome in UGIB. To compare the Shock Index with existing pre-endoscopic UGIB risk scores in predicting outcomes after UGIB METHODS: In an international, prospective study of 3012 consecutive patients with UGIB, we compared the Shock Index with existing scores including the Glasgow Blatchford score (GBS), admission Rockall score, AIMS65, and the newly described "ABC" score. Pre-determined endpoints were need for major (≥4 units red cells) transfusion, need for endoscopic therapy and 30-day mortality. The Shock Index was inferior to the GBS in predicting need for major transfusion (area under the receiver operator characteristic curve [AUROC] 0.655 vs 0.836, P<0.001) and need for endotherapy (AUROC 0.606 vs 0.747, P<0.001). The Shock Index was inferior to all other scores for 30-day mortality: for example, AUROC 0.611 vs 0.863 for ABC score (P<0.001). Adding the Shock Index to the ABC score did not improve accuracy of the ABC score in predicting mortality (AUROC 0.864 vs 0.863, P=0.95). The Shock Index performed poorly with AUROCs <0.66 and was inferior to existing pre-endoscopy scores at predicting major clinical endpoints after UGIB. We found no clear evidence that the Shock Index is clinically useful at predicting outcomes in UGIB. [Correction added on 20 December 2019, after first online publication: Summary section has been changed for clarification.].

Validation of a Machine Learning Model That Outperforms Clinical Risk Scoring Systems for Upper Gastrointestinal Bleeding

Scoring systems are suboptimal for determining risk in patients with upper gastrointestinal bleeding (UGIB); these might be improved by a machine learning model. We used machine learning to develop a model to calculate the risk of hospital-based intervention or death in patients with UGIB and compared its performance with other scoring systems. We analyzed data collected from consecutive unselected patients with UGIB from medical centers in 4 countries (the United States, Scotland, England, and Denmark; n= 1958) from March 2014 through March 2015. We used the data to derive and internally validate a gradient-boosting machine learning model to identify patients who met a composite endpoint of hospital-based intervention (transfusion or hemostatic intervention) or death within 30 days. We compared the performance of the machine learning prediction model with validated pre-endoscopic clinical risk scoring systems (the Glasgow-Blatchford score [GBS], admission Rockall score, and AIMS65). We externally validated the machine learning model using data from 2 Asia-Pacific sites (Singapore and New Zealand; n= 399). Performance was measured by area under receiver operating characteristic curve (AUC) analysis. The machine learning model identified patients who met the composite endpoint with an AUC of 0.91 in the internal validation set; the clinical scoring systems identified patients who met the composite endpoint with AUC values of 0.88 for the GBS (P= .001), 0.73 for Rockall score (P < .001), and 0.78 for AIMS65 score (P < .001). In the external validation cohort, the machine learning model identified patients who met the composite endpoint with an AUC of 0.90, the GBS with an AUC of 0.87 (P= .004), the Rockall score with an AUC of 0.66 (P < .001), and the AIMS65 with an AUC of 0.64 (P < .001). At cutoff scores at which the machine learning model and GBS identified patients who met the composite endpoint with 100% sensitivity, the specificity values were 26% with the machine learning model versus 12% with GBS (P < .001). We developed a machine learning model that identifies patients with UGIB who met a composite endpoint of hospital-based intervention or death within 30 days with a greater AUC and higher levels of specificity, at 100% sensitivity, than validated clinical risk scoring systems. This model could increase identification of low-risk patients who can be safely discharged from the emergency department for outpatient management.

Comparison of risk scoring systems for patients presenting with upper gastrointestinal bleeding: international multicentre prospective study

Objective To compare the predictive accuracy and clinical utility of five risk scoring systems in the assessment of patients with upper gastrointestinal bleeding.Design International multicentre prospective study.Setting Six large hospitals...

OC-036 International Assessment of Outcome of Upper GI Haemorrhage at Weekends: Abstract OC-036 Table 1

Introduction Weekend admissions have been associated with higher mortality. For upper gastrointestinal haemorrhage (UGIH) some studies show significantly increased mortality and less early endoscopy while the UK UGIH audit reported no difference. We studied whether out of hours (OOH) admissions were sicker and/or had higher mortality. Methods Prospective study over 12 months (from March 2014) from 2 UK and 2 international centres. Admission time, demographics, pulse, BP, lab results, endoscopy findings, further procedures and 30 d mortality were recorded. 3 pre-endoscopy scores (Glasgow Blatchford (GBS), AIMS65 and admission Rockall scores) and 2 post-endoscopy scores (PNED and full Rockall scores) were determined. Chi-squared, Fisher’s exact and Kruskal-Wallis tests were used as appropriate. A two-tailed significance level of 5% was used. Results 2118 pconsecutive atients, 60% male, median age 66 years were seen. There were no significant differences in mortality, need for endoscopic therapy, surgery/embolisation or rebleeding in both UK and non-UK centres. There were no differences in comorbidity, mean ASA 2.3, pulse or BP although weekday admissions had a lower Hb (110 g/l vs 118 g/l (weeknight) vs 117 g/l (weekend) p 2118 consecutive patients admitted March 2014-March 2015 from Glasgow (600), Truro (544), Odense (541) and Singapore (433). Data shown are mean [95% CI] or number (%). Conclusion There is no difference in mortality in patients admitted with UGIH OOH compared to weekday admissions. The severity of UGIH was not related to time of admission. Similar findings were noted in the 2 UK centres and internationally. Disclosure of Interest None Declared

Is the Glasgow Blatchford score useful in the risk assessment of patients presenting with variceal haemorrhage?

The Glasgow Blatchford score (GBS) is a pre-endoscopic risk assessment tool for patients presenting with upper gastrointestinal haemorrhage. There are few data regarding use in patients with variceal bleeding, who are generally accepted as being at high risk. The aim of the study was to assess GBS in correctly identifying patients with subsequently proven variceal bleeding as 'high risk' and to compare GBS, admission and full Rockall scores in predicting clinical endpoints in this group. Data on consecutive patients with upper gastrointestinal haemorrhage presenting to four UK hospitals were collected. The GBS, admission and full Rockall scores were calculated and compared for the subgroup subsequently shown to have variceal bleeding. Area under the receiver operating curve (AUROC) was used to assess the scores ability to predict clinical endpoints within this variceal bleeding subgroup. A total of 1432 patients presented during the study period. Seventy-one (5%) had a final diagnosis of variceal bleeding. At presentation, none of this group had GBS less than 2, but six had an admission Rockall score of 0. In predicting need for blood transfusion, AUROC scores for GBS, full and admission Rockall scores were 0.68, 0.65 and 0.68, respectively. For endoscopic/surgical intervention the scores were 0.34, 0.51 and 0.55, respectively, and for predicting death the scores were 0.56, 0.72 and 0.70, respectively. None of these AUROC score comparisons were significant. At presentation, GBS correctly identifies patients with variceal bleeding as high risk and appears superior to the admission Rockall score. However, GBS and both Rockall scores are poor at predicting clinical outcome within this group.

OC-143 Artificial neural network for the risk stratification of acute upper gastrointestinal bleeding: multicentre comparative analysis vs the Glasgow Blatchford and rockall scores

IntroductionMost patients presenting with acute upper GI bleeding (AUGIB) are at low risk of requiring clinical intervention or death. Nevertheless, risk assessment conventionally involves inpatient upper GI endoscopy which increases...

Multicentre comparison of the Glasgow Blatchford and Rockall scores in the prediction of clinical end-points after upper gastrointestinal haemorrhage

The Glasgow Blatchford Score (GBS) is increasingly being used to predict intervention and outcome following upper gastrointestinal haemorrhage (UGIH). To compare the GBS with both the admission and full Rockall scores in predicting specific clinical end-points following UGIH. Data on consecutive patients presenting to four UK hospitals were collected. Admission history, clinical and laboratory data, endoscopic findings, treatment and clinical follow-up were recorded. Using ROC curves, we compared the three scores in the prediction of death, endoscopic or surgical intervention and transfusion. Results A total of 1555 patients (mean age 56.7years) presented with UGIH during the study period. Seventy-four (4.8%) died, 223 (14.3%) had endoscopic or surgical intervention and 363 (23.3%) required transfusion. The GBS was similar at predicting death compared with both the admission Rockall (area under ROC curve 0.804 vs. 0.801) and full Rockall score (AUROC 0.741 vs. 0.790). In predicting endo-surgical intervention, the GBS was superior to the admission Rockall (AUROC 0.858 vs. 0.705; P<0.00005) and similar to the full Rockall score (AUROC 0.822 vs. 0.797). The GBS was superior to both admission Rockall (AUROC 0.944 vs. 0.756; P<0.00005) and full Rockall scores (AUROC 0.935 vs. 0.792; P<0.00005) in predicting need for transfusion. Despite not incorporating age, the GBS is as effective as the admission and full Rockall scores in predicting death after UGIH. It is superior to both the admission and full Rockall scores in predicting need for transfusion, and superior to the admission Rockall score in predicting endoscopic or surgical intervention.

Is the Glasgow Blatchford Score useful in the risk assessment of patients presenting with variceal haemorrhage?

IntroductionThe Glasgow Blatchford Score (GBS) is a pre-endoscopic risk assessment tool for patients presenting with upper gastrointestinal haemorrhage (UGIH). It can predict need for intervention or death and identifies low...

OC-022 Multicentre comparison of the Glasgow Blatchford and Rockall scores in the prediction of clinical endpoints in upper gastrointestinal haemorrhage

Introduction The pre-endoscopic Glasgow Blatchford Score (GBS) can identify low-risk patients with upper gastrointestinal haemorrhage (UGIH) who may be suitable for out-patient management. Although it does not include the patient9s age, the GBS appears to have high accuracy in predicting clinically relevant endpoints. Our aim was to compare the GBS with both the pre-endoscopy (admission) and post-endoscopy (full) Rockall scores in predicting need for clinical intervention and mortality. Methods Data on consecutive patients presenting to four UK hospitals (Glasgow Royal Infirmary, Royal Cornwall Hospital Truro, University Hospital of North Tees and Ninewells Hospital Dundee) were collected. Admission history, clinical and laboratory data, endoscopic findings and treatment, and clinical follow-up were recorded. We used ROC curves to compare the three scores in the separate prediction of death, endoscopic or surgical intervention and transfusion. Results 1556 patients (mean age 56.7 years; 62% male) presented with UGIH to the four hospitals during the study period. 74 (4.8%) died, 223 (14.3%) had endoscopic or surgical intervention and 363 (23.3%) required transfusion. The GBS was equally effective at predicting death compared with both the admission Rockall score (area under ROC curve 0.804 vs 0.801) and the full Rockall score (AUROC 0.741 vs 0.790). In predicting endo/surgical intervention, the GBS was superior to the admission Rockall score (AUROC 0.858 vs 0.705, p Conclusion Despite not incorporating age, the GBS is as effective as the admission and full Rockall scores in predicting death after UGIH. It is superior to both the admission and full Rockall scores in predicting need for transfusion and superior to the admission Rockall score in predicting endoscopic or surgical intervention.

Outpatient management of patients with low-risk upper-gastrointestinal haemorrhage: multicentre validation and prospective evaluation

Upper-gastrointestinal haemorrhage is a frequent reason for hospital admission. Although most risk scoring systems for this disorder incorporate endoscopic findings, the Glasgow-Blatchford bleeding score (GBS) is based on simple clinical and laboratory variables; a score of 0 identifies low-risk patients who might be suitable for outpatient management. We aimed to evaluate the GBS then assess the effect of a protocol based on this score for non-admission of low-risk individuals. Our study was undertaken at four hospitals in the UK. We calculated GBS and admission (pre-endoscopy) and full (post-endoscopy) Rockall scores for consecutive patients presenting with upper-gastrointestinal haemorrhage. With receiver-operating characteristic (ROC) curves, we compared the ability of these scores to predict either need for clinical intervention or death. We then prospectively assessed at two hospitals the introduction of GBS scoring to avoid admission of low-risk patients. Of 676 people presenting with upper-gastrointestinal haemorrhage, we identified 105 (16%) who scored 0 on the GBS. For prediction of need for intervention or death, GBS (area under ROC curve 0.90 [95% CI 0.88-0.93]) was superior to full Rockall score (0.81 [0.77-0.84]), which in turn was better than the admission Rockall score (0.70 [0.65-0.75]). When introduced into clinical practice, 123 patients (22%) with upper-gastrointestinal haemorrhage were classified as low risk, of whom 84 (68%) were managed as outpatients without adverse events. The proportion of individuals with this condition admitted to hospital also fell (96% to 71%, p<0.00001). The GBS identifies many patients presenting to general hospitals with upper-gastrointestinal haemorrhage who can be managed safely as outpatients. This score reduces admissions for this condition, allowing more appropriate use of in-patient resources.

Management of patients with nonvariceal upper gastrointestinal hemorrhage before and after the adoption of the Rockall score, in the Italian Gastroenterology Units

Nonvariceal upper gastrointestinal hemorrhage is a frequent reason for ordinary hospital admission. In Italy the use of prognostic scores to stratify the risk has not been adequately validated: the impact on clinical management of a rating system like the Rockall score remains to be established. RING is a 'register' that has been collecting hospital discharge files from hospital gastroenterology units, giving a broad picture of the patients admitted for this pathology. We analyzed the hospital discharge files collected between 2001 and 2005 from 12 gastroenterology units, which issued more than 26,000 hospital discharge files for ordinary hospital admission and have been using the Rockall score for defining nonvariceal upper gastrointestinal hemorrhage since 2003. There were 2832 hospital discharge files with a main diagnosis of nonvariceal upper gastrointestinal hemorrhage: 1335 'before' the Rockall score was introduced, 1497 'after' the introduction. Patients' mean age was 67.7+/-16.7 years, with a male/female ratio of 1.7 and no significant changes over the years. There were no differences in the distribution of diagnoses in nonvariceal upper gastrointestinal hemorrhage patients before/after the introduction of the Rockall score, though the mean hospital stay became shorter (7.1+/-5.0 vs. 6.3+/-4.5 days), and mortality declined (2.8 vs. 2.3%), in parallel with the caselist as a whole. For 1102 ordinary hospital admission Rockall score was calculated. Diagnoses were more accurate: significantly fewer undefined causes and an increase in peptic ulcer. The mean Rockall score was 4.6+/-2.2: 17.8% low (0-2), 48.7% intermediate (3-5), and 33.5% high (>or=6). Mean hospital stay, rebleeding, and mortality were correlated with the severity of the score. The Rockall score enables the clinician to formulate a more precise diagnosis and substantially shortens the time in hospital, especially for patients at low-risk of rebleeding and death, so more resources can be dedicated to critically ill patients.