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

Background

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.

Methods

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.

Results

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.

Conclusion

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.

Comparison of risk scoring systems for upper gastrointestinal bleeding in patients after renal transplantation: a retrospective observational study in Hunan, China

Background

Upper gastrointestinal bleeding (UGIB) is a common complication in renal transplant recipients. However, the risk stratification value of bleeding scoring systems in these patients is unclear, and data regarding risk factors are limited.

Methods

Clinical data of renal transplant recipients in The Third Xiangya hospital were collected. The predictive ability of Glasgow Blatchford score (GBS), pre-endoscopy Rockall score (pRS), and AIMS65 score were assessed by the area under the receiver operating characteristic curve (AUROC). Risk factors of UGIB were analyzed using binary logistic regression analysis.

Results

A total of 220 patients were enrolled, of which 55 with UGIB. Endoscopy improved the overall survival rate of patients. Glasgow Blatchford score (AUROC 0.868) performed best at predicting UGIB patients who need intervention or death, with a threshold of 10, sensitivity and specificity were 82.4% and 70%, respectively. In terms of predicting mortality, the GBS score was comparable with AIMS65 score (p = 0.30) and pRS score (p = 0.42). Viral hepatitis, intravenous hormone usage, low platelet count, and low albumin level were significant factors associated with UGIB.

Conclusions

The Glasgow Blatchford score (AUROC 0.868) was best at predicting the need for intervention or death. However, their ability to predict mortality was limited, with AUROC less than 0.8. Our study also identified four independent risk factors for renal transplant recipients with UGIB.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12876-022-02426-3.

Comparison of Glasgow-Blatchford Score and Rockall Score in Patients with Upper Gastrointestinal Bleeding

Introduction:

Upper gastrointestinal bleeding can be a life-threatening condition and requires careful evaluation from the very first episode in order to reduce the risk of rebleeding, hemorrhagic shock and death. The outcome of a patient with upper gastrointestinal bleeding depends on resuscitation measures taken during admission to the hospital and an adequate assessment of the patient’s risk level.

Aim:

The aim of the study is to compare Glasgow Blatchford score and Rockall score and to identify the most accurate score used in predicting unfavorable outcomes and the need for intervention.

Methods:

This study involves 237 patients with upper gastrointestinal bleeding. The accuracy of the scoring systems was assessed by plotting receiver-operating characteristic curves (ROC curves) and was calculated for GBS and RS with 95% confidence interval (CI).

Results:

As for mortality prediction, RS was superior to GBS (AUC 0.806 vs. 0.750). The GBS had a higher accuracy in detecting patients who needed transfusion units and was superior to the RS (AUC 0.810 vs.0.675). In predicting the need for intervention, RS was superior to GBS (AUC 0.707 vs. 0.636.

Conclusion:

GBS and RS are developed to help clinicians to triage patients appropriately in order to assess endoscopic therapy within a suitable time frame, as well as identify low risk patients for possible outpatient management. High accuracy of the GBS in predicting a need for transfusion represents an important endpoint to assess. RS was superior to GBS in predicting a need for intervention as well as mortality. Currently, a combination of these scoring systems is the best way for proper assessment.

MAP(ASH): A new scoring system for the prediction of intervention and mortality in upper gastrointestinal bleeding

BACKGROUND AND AIM

Risk stratification for upper gastrointestinal bleeding (UGIB) is recommended. However, scoring system accuracy is suboptimal, and score calculation can be complex. Our aim was to develop a new score, the MAP(ASH) score, with information available in the emergency room and to validate it.

METHODS

The score was built from a prospective database of patients with UGIB and validated in an international database of 3012 patients from six hospitals. Outcomes were 30-day mortality, endoscopic intervention, any intervention (red blood transfusion, endoscopic treatment, interventional radiology, surgery, or death), and rebleeding. Accuracy to predict outcomes was assessed by the area under the receiver operating characteristic curve (AUROC).

RESULTS

Five hundred forty-seven patients were included in the development cohort. Impaired mental status, albumin < 2.5 g/dL, pulse > 100, American Society of Anesthesiologists score > 2, systolic blood pressure < 90 mmHg, and hemoglobin < 10 g/dL were included in the score. The model had a good predictive accuracy for intervention (AUROC = 0.83; 95% confidence interval [CI]: 0.79-0.88) and fair for mortality (AUROC = 0.74; 95% CI: 0.68-0.81). Regarding endoscopic intervention, AUROC was 0.61 (95% CI: 0.56-0.66) in the original cohort and 0.69 (95% CI: 0.66-0.71) in the validation cohort, showing a poor performance, similar to other scores. For rebleeding, the MAP(ASH) (AUROC 0.73; 95% CI: 0.69-0.77) was similar to Glasgow Blatchford score (AUROC = 0.72; 95% CI: 0.67-0.76) but superior to AIMS65 (AUROC = 0.64; 95% CI: 0.59-0.68).

CONCLUSION

MAP(ASH) is a simple pre-endoscopy risk score to predict intervention after UGIB, with fair discrimination at predicting mortality. Because of its applicability, it could be an option in clinical practice.

Clinical utility of pre-endoscopy risk scores in upper gastrointestinal bleeding

Introduction: Acute upper-gastrointestinal bleeding (AUGIB) is a common medical emergency, with an incidence of 103-172 per 100,000 in the United Kingdom (UK) and mortality of 2% to 10%. Early and accurate prediction of the severity of an AUGIB episode may help guide management, including in or outpatient management, level of care required, and timing of endoscopy. This article aims to address the clinical utility of the various pre-endoscopic risk assessment tools used in AUGIB.Areas covered: The authors undertook a literature review of the current evidence on the pre-endoscopic risk assessment scores. Additional the authors discuss the recently published novel risk assessment scores.Expert opinion: The evidence shows that GBS is the most clinically useful risk assessment score in correctly identifying very low-risk patients suitable for outpatient management. At present, research is ongoing to assess machine learning in the assessment of patients presenting with AUGIB. More research is needed but it shows promise for the future.

Glasgow Blatchford Score and risk stratifications in acute upper gastrointestinal bleed: can we extend this to 2 for urgent outpatient management?

Upper gastrointestinal (GI) bleeds are a common presentation to emergency departments in the UK. The Glasgow Blatchford score (GBS) predicts the outcome of patients at presentation. Current UK and European guidelines recommend outpatient management for a GBS of 0. In the current study, our aim was to assess whether extending the GBS allows for early discharge while maintaining patient safety. We also analysed whether pathologies could be missed by discharging patients too early. Data were retrospectively collected on patients admitted with symptoms of an upper GI bleed between 1 October 2013 and 10 June 2016. The GBS was calculated and gastroscopy reports were obtained for each patient. In total, 399 patients were identified, 63 of whom required therapy. The negative predictive value (NPV) for excluding the need for endoscopic intervention with a GBS score up to 1 was 100%. Extending the score to 2 and 3 reduced the NPV to 98.53% and 98.77%, respectively. The NPV of GBS in excluding any diagnosis at 0 was 43.55%. Two patients died as a result of GI bleeding, with a GBS score of 3. Therefore, we can conclude that, for non-variceal bleeds, the GBS can be extended to 2 for safe outpatient management, thereby reducing the number of bed days and pressure for urgent endoscopies.

Upper Gastrointestinal Bleeding in Patients with End Stage Renal Disease: Causes, Characteristics and Factors Associated with Need for Endoscopic Therapeutic Intervention

BACKGROUND AND OBJECTIVES

The risk of upper gastrointestinal bleeding (UGIB) is increased among the end-stage renal disease (ESRD) patients. The aim of the current study was to describe the causes and characteristics of UGIB in ESRD patients at our center and to assess the need for endoscopic therapeutic intervention (ETI) using Rockall (RS) and Glasgow Blatchford scores (GBS).

MATERIAL AND METHODS

All patients with ESRD and UGIB with age ≥14 years were included. Frequencies and percentages were computed for categorical variables. Chi square test or Fischer's exact test was used for statistical analysis.

RESULTS

A total of 59 subjects had a mean age of 47.25 ± 15 years.The most common endoscopic findings seen were erosions in 33 (55.9%) patients, followed by ulcers in 18 (30.3%) patients. ETI was required in 33 (55.9%) patients, which included adrenaline injection in 19 (32.3%), hemoclip in 9 (15.2%) and argon plasma coagulation in 5 (8.4%) patients. Factors associated with the need of ETI were identified as: a combined presentation of hematemesis and melena (P=0.033), ulcer (P=0.002) and associated chronic liver disease (P=0.015). Six (10.1%) patients died. Death was more common if ETI was not performed (P=0.018).

CONCLUSION

ETI was more commonly required in patients on maintenance hemodialysis with UGIB, who had presence of combined hematemesis and melena, ulcers and associated chronic liver disease. A Glasgow Blatchford score of >14 was helpful in assessing the need for ETI in these patients.

External validation of scoring systems in risk stratification of upper gastrointestinal bleeding

AIM

The aim of this study was to externally validate the four commonly used scoring systems in the risk stratification of patients with upper gastrointestinal bleed (UGIB).

METHODS

Patients of UGIB who underwent endoscopy within 24 h of presentation were stratified prospectively using the pre-endoscopy Rockall score (PRS) >0, complete Rockall score (CRS) >2, Glasgow Blatchford bleeding scores (GBS) >3, and modified GBS (m-GBS) >3 scores. Patients were followed up to 30 days. Prognostic accuracy of the scores was done by comparing areas under curve (AUC) in terms of overall risk stratification, re-bleeding, mortality, need for intervention, and length of hospitalization.

RESULTS

One hundred and seventy-five patients were studied. All four scores performed better in the overall risk stratification on AUC [PRS = 0.566 (CI: 0.481-0.651; p-0.043)/CRS = 0.712 (CI: 0.634-0.790); p<0.001)/GBS = 0.810 (CI: 0.744-0.877; p->0.001); m-GBS = 0.802 (CI: 0.734-0.871; p<0.001)], whereas only CRS achieved significance in identifying re-bleed [AUC-0.679 (CI: 0.579-0.780; p = 0.003)]. All the scoring systems except PRS were found to be significantly better in detecting 30-day mortality with a high AUC (CRS = 0.798; p-0.042)/GBS = 0.833; p-0.023); m-GBS = 0.816; p-0.031). All four scores demonstrated significant accuracy in the risk stratification of non-variceal patients; however, only GBS and m-GBS were significant in variceal etiology. Higher cutoff scores achieved better sensitivity/specificity [RS > 0 (50/60.8), CRS > 1 (87.5/50.6), GBS > 7 (88.5/63.3), m-GBS > 7(82.3/72.6)] in the risk stratification.

CONCLUSION

GBS and m-GBS appear to be more valid in risk stratification of UGIB patients in this region. Higher cutoff values achieved better predictive accuracy.

Risk assessment scores for patients with upper gastrointestinal bleeding and their use in clinical practice

Upper gastrointestinal bleeding (UGIB) is a common cause for emergency admission to hospital representing a significant clinical as well as economic burden. UGIB encompasses a wide range of severities from life-threatening exsanguination to minor bleeding that may not require hospital admission. Patients with UGIB are often initially assessed and managed by junior doctors and non-gastroenterologists. Several risk scores have been created for the assessment of these patients, some requiring endoscopic data for calculation and others that are calculable from clinical data alone. A key question in clinical practice is how to accurately identify patients with UGIB at high risk of adverse outcome. Patients considered high risk are more likely to experience adverse outcomes and will require urgent intervention. In contrast, those patients with UGIB who are considered to be low risk could potentially be managed on an outpatient basis. The Glasgow Blatchford Score (GBS) appears best at identifying patients at low risk of requiring intervention or death and therefore may be best for use in clinical practice, allowing outpatient management in low risk cases. There has been some debate as to the optimal GBS cut-off score for safely identifying this low-risk group. Many guidelines suggest that patients with a GBS of zero can be safely managed as outpatients, but more recent studies have suggested that this threshold could potentially be safely increased to ≤1. Most other patients require inpatient endoscopy within 24 h and the full Rockall score remains important for risk assessment following endoscopy, particularly as it includes the endoscopic diagnosis. A minority of patients will require emergency endoscopy following resuscitation, but at present there is no evidence that risk scores can accurately identify this very high-risk group. Studies have shown the latest risk assessment score, the AIMS65, looks promising in the prediction of mortality. However, to date there is no data on the use of the AIMS65 in identifying low risk patients for possible outpatient management.