Predictors of a variceal source among patients presenting with upper gastrointestinal bleeding

1Division of Gastroenterology; 2Division of Clinical Epidemiology, The McGill University Health Centre, Montreal General Hospital, McGill University, Montreal, Quebec. *Members of the the national REgistry of patients undergoing endoscopic and/or Acid Suppression therapy and Outcomes analysis for upper gastrointestinal bleediNg (REASON) are listed in Appendix 1 Correspondence: Dr Alan Barkun, Division of Gastroenterology, The McGill University Health Center, Montreal General Hospital site, 1650 Cedar Avenue, room D7-346, Montreal, Quebec H3G 1A4. Telephone 514-934-8309, fax 514-834-8531, e-mail alan.barkun@muhc.mcgill.ca Received for publication May 13, 2011. Accepted July 9, 2011 Variceal bleeding is associated with a high mortality rate. Over the past few decades, mortality has decreased (1) due to improvements in endoscopic interventions, antibiotic use and pharmacological therapies, with a drop in the six-week mortality rate from earlier rates of 42% (2) to approximately 16% (1,3). Nonetheless, the recurrence rate remains high at 13% to 29% (1,3). Early recognition of variceal bleeding is crucial because the initial management of patients with a variceal source is different from that of a nonvariceal source (4-7). The ability of bedside variables to predict the source of upper gastrointestinal bleeding (UGIB) remains controversial and has been poorly assessed in only a very few recent studies yielding predictors that have yet to be formally assessed in a North American population (8). Using clinical parameters and laboratory investigations obtained from a large registry database, the current study attempted to identify variables predictive of UGIB from a variceal source before endoscopic evaluation.


les variables prédictives de saignements oesogastroduodénaux d'origine variqueuse chez les patients
HISTORIQuE : Les patients présentant des saignements oesogastroduodénaux (SOGD) ont besoin d'une approche précoce et personnalisée orientée par la connaissance de la lésion hémorragique, notamment son origine variqueuse ou non variqueuse.OBJECTIF : En examinant un grand registre national, déterminer les variables prédictives de SOGD d'origine variqueuse au moyen de paramètres cliniques relevés avant l'évaluation endoscopique.MÉTHODOlOGIE : Les chercheurs ont procédé à une analyse prospective dans 21 hôpitaux canadiens entre janvier 2004 et la fin de mai 2005.Ils ont analysé les dossiers consécutifs de patients hospitalisés dont le diagnostic primaire ou secondaire au congé en était un de SOGD.Ils ont colligé des données au sujet des facteurs démographiques, y compris les antécédents, l'examen physique, les premiers examens de laboratoire, les traitements endoscopiques et pharmacologiques administrés et les issues cliniques.Ils ont effectué une modélisation de la régression logistique multivariable pour déterminer les variables prédictives cliniques de saignements d'origine variqueuse.RÉSulTATS : La population à l'étude se composait de 2 020 patients (âge moyen [±ÉT] de 66,3±16,4 ans; 38,4 % de femmes).Dans l'ensemble, 215 ( total of 21 hospitals across Canada participated.Patients initially assessed for the present episode of bleeding at an institution not part of the study and subsequently transferred to a participating site for further management, as well as patients presenting with UGIB to the emergency room and who were not admitted to hospital, were excluded from the study.Patients were identified through the diagnosis documented in hospital records using the International Classification of Diseases, 9th or 10th Revision (ICD-9 or ICD-10) codes.All patients who had a primary or secondary coded discharge diagnosis of UGIB were screened for eligibility.Successive patients fulfilling selection criteria were entered.

Study design
The charts of all hospitalized patients were retrospectively reviewed by a local, trained research nurse.Once a patient was identified as a potential candidate from medical record lists, the hospital chart was obtained for further review to ensure the patient met all eligibility criteria.If so, the patient was assigned an enrollment code, and information from the medical chart was entered into an electronic database by a trained research nurse.An Internet-based case report form, specifically designed for the present study with standardized definitions for variables, was used in addition to a centralized data validation process previously described in the literature (9); 10% of all records were also audited by an independent study nurse.The entered data were reviewed centrally, with validation performed by an independent reviewer with medical knowledge who audited the progress of care of a given patient according to the entered data and decided on whether the recorded information was internally consistent.Ranges for each variable were required to fall within a range of preset, biologically plausible values.

Recorded information
Demographic data, historical information, endoscopic and pharmacological therapies administered, as well as clinical outcome data were collected.
A variable -'stigmata of chronic liver disease' -was created to simplify the analysis while maintaining clinical relevance.This variable was a composite of the presence of any of the following for a given patient: hepatomegaly, splenomegaly, peripheral edema, jaundice, hepatic encephalopathy and ascites.The Model for End-stage Liver Disease (MELD) score was calculated with the available data (10).To ensure patient confidentiality, no personal identification information or other personal identifiers, such as address or hospital identification number, were recorded.

Statistical analysis
Descriptive variables are presented as percentages or mean ± SD.Inferential univariable analysis was only performed on clinically relevant variables determined a priori using the χ 2 test for categorical variables or a Wilcoxon nonparametric test for continuous variables.Multivariable analyses were performed using logistic regression modelling, with associated ORs.All analyses were performed using SAS

RESulTS
Patient population Demographics and historical information: A total of 2020 patients were included in the REASON registry.Overall, 215 (10.6%) patients were found to be bleeding from upper gastrointestinal varices.Varices were esophageal in 90.7% and gastric in 28.9% of patients.
The basic demographic and medical history information of all patients presenting with UGIB and, more specifically, among those with variceal bleeding are presented in Table 1.In the latter group, the mean age was 58.0±12.3years, including 64 (29.8%) females.A history of variceal UGIB was noted in 91 (42.3%) patients, liver disease in 166 (77.2%) and excessive alcohol use in 113 (52.6%).The mean number of comorbidities was 2.5 (range zero to eight) at the time of initial bleeding.Presenting symptoms are shown in Table 2.For patients with variceal UGIB, the number of patients who presented with a history of melena was 134 (62.3%), hematemesis 116 (54.0%) and syncope 15 (7.0%), and the number of patients with an American Society of Anesthesiologists (ASA) score of IV or V was 74 (34.4%).

Study population Physical examination findings:
Relevant physical examination findings for the overall population and, more specifically, for the patients with variceal bleeding, are summarized in Table 2.At presentation, the number of patients who had a variceal source of bleeding and initial hemodynamic instability was 67 (31.2%), splenomegaly 38 (17.7%), mild to moderate ascites 61 (28.4%) and severe ascites 22 (10.2%);jaundice was found in 47 (21.9%).
The MELD score among patients with variceal bleeding averaged 11.1±8.4(range six to 40 points).In the nonvariceal group, the mean MELD score was 6.4±7.9.
The 30-day mortality rate for patients with nonvariceal bleeding was 9.4%, while the rate for patients with variceal bleeding was 14.4%.
The multivariable model that was used demonstrated adequate discrimination, with a C statistic of 0.91.

DISCuSSION
UGIB is stratified into variceal and nonvariceal causes because such differentiation bears important clinical information when deciding on the most efficient and cost-effective subsequent approach based on published guidelines (11,12).Differences in management include the suggested optimal duration until endoscopy (12 h versus later), the judicious use of resuscitation fluids and target hemoglobin level (80 g/L versus greater) (13,14).
Bedside predictors of variceal UGIB have not been well studied.A recent literature search yielded only one study from Thailand assessing this issue (8).The predictive model reached a very high negative predictive value of 97% using an UGIB etiology score that included previous diagnosis of cirrhosis or the presence of signs of chronic liver disease, red vomitus and a red NGT aspirate.Factors limiting the universal applicability of these results include the small numbers of patients with variceal UGIB causes (47 in the initial cohort and 46 in the validation cohort) and the single-centre setting.The conclusions are, however, strengthened by validation of their initial findings in an independent population.These results may not be completely applicable to a North American population given differences in genetic backgrounds, causes of portal hypertension or cirrhosis.
Our study broadly represents patients from numerous centres that use a national Canadian database enrolling a large number of patients.Additional methodological strengths were the data quality procedures, such as an Internet-based case report form with standardized definitions for all variables and subsequent validation of 10% of all entered data by two independent nurses in addition to a centralized data validation process.Extensive pre-endoscopy information was collected for patients involved in the present study across multiple domains including presenting history, physical examination findings and laboratory data.
The present study demonstrated that patients with a variceal source of UGIB compared with nonvariceal causes were younger, more often men, and with risk factors that included a history of liver disease or bleeding disorder, excessive alcohol use, hematochezia, hematemesis, bright red blood on NGT aspiration, stigmata of chronic liver disease, lower serum albumin, lower platelet counts and more frequent abnormality in liver enzyme levels.On the other hand, patients with nonvariceal UGIB were more likely to have a history of abdominal surgery, had increased exposure to antithrombotics, calcium channel blockers or nonsteroidal anti-inflammatory drugs, and were less likely to have a high ASA score.Using multivariable logistic regression analysis, the only remaining significant independent predictors of a variceal source of bleeding included a documented history of liver disease (OR 6.36 [95% CI 3.59 to 11.3]), excess alcohol use (OR 2.28 [95% CI 1.37 to 3.77]), and the absence of use of antithrombotics (OR 0.44 [95% CI [0.35 to 0.78]) as well as a clinical presentation of hematochezia (OR 3.02 [95% CI 1.46 to 6.22]), hematemesis, (OR 2.65 [95% CI 1.61 to 4.36]) or stigmata of chronic liver disease (OR 2.49 [95% CI 1.46 to 4.25]).These findings include all criteria identified in the study by Pongprasobchai et al (8).The additional variables of excessive alcohol consumption, antiplatelet agents and hematochezia may relate to some of the aforementioned differences.
When simple laboratory test values, which can be obtained in a relatively short period from the time a patients presents with UGIB, were included in the multivariable model, they did not have an effect on the predictive probability of the model, but the study may have been underpowered in the analysis of these variables.Of note, the Thai study by Pongprasobchai et al (8) also did not find any predictive value attributable to laboratory data.

Symptoms on initial presentation with bleeding All patients (n=2020) bleed
To highlight the clinical impact of our findings, using our logistic regression model, we calculated the post-test predicted probabilities of a patient bleeding from a variceal source based on different possible clinical scenarios.From a baseline pretest (ie, prevalence) probability of 10.6%, use of antithrombotics in the absence of all other factors, dropped the predicted probability of a variceal source to 0%.In the case of a patient with a history of chronic liver disease and noted stigmata of chronic liver disease (the two most commonly used predictors in clinical settings), the predicted probability of a variceal source increased to 46%.If all significant predictors are present in a given patient, the predicted probability of a variceal source increases to 94% (Table 5).Such differences in predicted probabilities may be useful in refining a tailored management approach including determining the urgent need for an intravenous proton pump inhibitor or octreotide, or that of an endoscopy within 12 h versus 24 h.Additional research is needed to explore such implications.
The results of our study suggest that simple bedside parameters can be useful in the prediction of a variceal source among patients who present with UGIB.Significant variables include a history of liver disease, excessive alcohol use and the absence of antithrombotic medication use, as well as findings on physical examination of hematemesis, hematochezia and stigmata of chronic liver disease.The clinical implications of these findings warrant additional evaluative research in this patient population at high risk of negative outcomes.
FINANCIAl SuPPORT: The REASON study was funded by an atarms-length grant from Astra Zeneca Canada Inc.
DISClOSuRES: Alan Barkun is a consultant for AstraZeneca, Takeda Canada, Boston Scientific Inc, and Olympus Canada.Ahmad Alharbi, Majid Almadi and Myriam Martel report no conflicts of interest.

TAble 1 Demographic and baseline characteristics of patients with nonvariceal upper gastrointestinal bleeding (UGIb) or variceal bleeding Demographics and baseline characteristics All patients (n=2020) bleed P Variceal (n=215) Nonvariceal (n=1805)
Data presented as n (%); 95% CI unless otherwise indicated.Percentages shown are for the percentage of patients in the nonvariceal population (total n=1805) or in the variceal population (total n=215) or for the total population (n=2020) software version 9.2 (SAS Institute, USA).A statistical significance threshold of P=0.05 was adopted.No attempt at imputation was performed for missing data.

Primary investigators of participating hospitals in the national Registry of patients undergoing endoscopic and/ or Acid Suppression therapy and Outcomes analysis for upper gastrointestinal bleediNg (ReASON)
Continued on next page