Study population
This prospective cohort study was performed at a university-affiliated teaching hospital in central Taiwan that includes an emergency department (ED) with approximately 80,000 patient visits per year. All adult patients (over 18 years of age) who were admitted to the ED at Chiayi Chang Gung Memorial Hospital between July 1, 2016 and June 30, 2019 with evidence of UGIB were eligible for inclusion in the study. Diagnoses of UGIB included hematemesis, melena, bloody or coffee ground vomitus, and bloody or coffee ground material in nasogastric aspirate. The study was approved by Chang Gung Medical Foundation Institutional Review Board (IRB No: 201600157B0). Informed consent was obtained from all study participants. Patients who were unable to provide informed consent were excluded.
Data collection
Demographic data, including age, sex, vital signs on triage in the ED, chief complaints, present illness, laboratory test results, comorbid medical conditions, current medications, and types of treatment (for example, blood transfusion, therapeutic procedures, operations, and medications) were recorded.
Definitions
Massive transfusion was defined as transfusion of three or more units of red blood cells within one hour during the ED stay(16, 17). Therapeutic procedures were defined as procedures to control or stop bleeding, including therapeutic endoscopy, angiographic embolization, and surgery. Shock was defined as a systolic blood pressure <100 mmHg at the time of ED triage. Further bleeding was defined as any of the following: (1) repeated esophagogastroduodenoscopy, angiographic embolization or operation to control bleeding within 3 days, and (2) continuous blood transfusion for more than 3 days.
BISHE score
The BISHE score has been recently developed to predict MT in patients with UGIB and was calculated for all patients. A BISHE score >7.5 indicates a high risk of in-hospital mortality(9).
Treatment protocol
All clinical management decisions were left to the discretion of the main attending physician. Standard management for all patients with UGIB in our ED was administration of an intravenous proton pump inhibitor before endoscopy for non-variceal bleeding. The choice of intermittent or continuous infusion of proton pump inhibitor was left to the discretion of the lead attending clinician. Intravenous terlipressin with an intravenous third-generation cephalosporin was administered before endoscopy for variceal bleeding. Blood transfusion was indicated for patients with a hemoglobin level <8 g/dL or signs of hemodynamic instability despite fluid resuscitation. The decision to administer a transfusion was made by the main attending physician.
Statistical analysis
Categorical variables are presented as the percentage and continuous variables as the median with the interquartile range. Associated factors were entered as categorical variables to construct a diagnostic scoring system. The cut-off points for all the laboratory variables were determined based on a combination of clinical experience, review of previous reports, and the median values in studies that included MT and non-MT groups(18-20). Univariate and multivariate logistic regression analyses were used in the current study. The initial model inclusion criterion was a p-value of <0.05, with predictive factors entered using backward-stepwise selection and retained when the p-value was <0.01. Assignment of score points was based on the corresponding regression coefficients. By dividing the coefficients by the smallest coefficients in the model and then rounding to the nearest 0.5, the regression coefficients were transformed to obtain item scores(21).
The risk score for each patient was calculated by totaling the scores for each independent variable. The Hosmer-Lemeshow goodness-of-fit test was used to evaluate the calibration of the model(22). The area under the receiver-operating characteristic (AUROC) curve was calculated to assess the predictive accuracy of the risk score(23). All analyses were conducted using MedCalc Statistical Software version 17.9.2 (MedCalc Software bvba, Ostend, Belgium; http://www.medcalc.org; 2017).