The study was conducted at Kamuzu Central Hospital (KCH) in Lilongwe, Malawi. The hospital is a tertiary care institution with 600 beds, although at times the total number of patients admitted can range up to 1000. It serves as a referral center for all district hospitals in the central region of Malawi, which has about 5 - 6 million inhabitants [7] . The hospital has a 24-hour Emergency Department, a Surgical High Dependency Unit with 5 beds and an Intensive Care Unit with 4 beds.

This was a retrospective study, reviewing admission charts of patients with a diagnosis of UGIB, who were admitted between 1st January and 31st December 2010. Patients over 18 years of age, that were admitted for vomiting blood or passing melena within 24 hours before admission were included in the study. Furthermore, patients with a cause of death related to UGIB, regardless of the initial cause for admission, were also included the study. Records of “Upper GI bleeding”, “Upper GIT bleeding”, gastric or duodenal ulcer bleeding, esophageal varices, vomiting blood or black lumps, hematemesis and peptic ulcer disease (PUD) was considered as UGIB as the cause of death. Furthermore “GI bleeding” or “GIT bleeding” was also included because of the probability of an upper localization in acute, fatal cases. Patients with known chronic liver diseases, cancer of the stomach or esophagus or gastritis were not considered UGIB unless a history of hematemesis was noted. Bleeding from oropharynx that could not be attributed to a pulmonary or upper airway cause was presumed to be gastrointestinal. Patients referred from other hospitals were included in the study if they matched the criteria at the time of admission at the initial hospitalization.

We reviewed approximately 5000 admission charts for patients admitted for UGIB from January to December 2010. A total number of 187 patients met the definition of UGIB for the retrospective review.

The data was collected from the charts from the male and female adult general surgery wards, the surgical high dependency unit and the intensive care unit archives. All available charts from the study period indicated were reviewed, although there is no guarantee that the archives were complete. From the eligible patients, date of admission, readmissions, date of death, age, gender, residency, heart rate, systolic blood pressure and comorbidities were extracted. Furthermore, endoscopic diagnosis, endoscopic bleeding were also recorded and final Rockall score was calculated by the study team.

A clinical diagnosis was assigned in all patients who did not have an endoscopically verified diagnosis. The diagnosis was set to “Upper gastrointestinal bleeding—no other diagnosis specified” unless the patient had signs of liver disease. Liver disease was defined as the presence of either one of the following: hepatomegaly, ascites or liver cirrhosis or at least two of the following clinical signs: splenomegaly, jaundice, finger clubbing or dilated veins on the abdominal wall. If the patient was found positive for liver disease, the diagnosis “Oesophageal varices” was assigned, due to the high prevalence of the condition in the setting of the study. If the cause of death for a patient was listed as an UGIB secondary to liver disease, the patient was assigned oesophageal varices as clinical diagnosis and got their comorbidity diagnosis modified accordingly.

The numerical risk scoring was done in accordance with the original Rockall study [8] , with some additions to comorbidity. Total score is calculated by simple addition. A score less than 3 carries good prognosis but total score more than 8 carries high risk of mortality. The Rockall score can be applied prior to diagnosis to determine patient with higher mortality risk and post-endoscopic diagnosis score includes the addition of the diagnosis score (Table 1).

Patients between 60 - 79 years of age were assigned 1, while patients over 80 got 2 points. Signs of shock were defined as a heart rate of over 100 beats per minute or a systolic blood pressure under 100 mmHg at presentation, which yielded 1 and 2 points respectively.

For this study, the concept of comorbidity was broadened to include HIV disease and active tuberculosis due to the high prevalence in Malawi, i.e. these were considered major comorbidities and given a score of 2. Further- more, since few patients had a confirmed diagnosis of liver disease, which yields 2 points on the comorbidity score, signs of ascites or hepatomegaly were considered signs of liver disease. Additionally, jaundice, finger club- bing, dilatation of abdominal veins and splenomegaly were considered signs of liver disease if at least two were present or together with ascites or hepatomegaly, as outlined above. Congestive heart failure and ischemic heart disease were assigned 2 points as a major comorbidity. 3 points were given to patients presenting with renal failure, liver cirrhosis and/or failure or metastatic cancer. Endoscopic diagnoses were divided into normal endoscopy, with no signs of bleeding, or Mallory-Weiss tear (0 point), gastrointestinal malignancy (2 points) and all other diagnoses (1 point), while signs of endoscopic bleeding scored 0 point if no bleeding or dark spots only were present and 2 points for the occurrence of spurting vessels, adherent clots or blood. The points were then added to produce a maximum pre-endoscopic Rockall score of 7 and a maximum score of 11 post-endoscopy.

Haemoglobin levels were only recorded if they were collected before the patient had a blood transfusion while other laboratory analyses could be collected at any time.

Study data were collected and managed using REDCap electronic data capture tools hosted at University of North Carolina. REDCap (Research Electronic Data Capture) is a secure, web-based application designed to support data capture for research studies [9] . The application is accessible online and this allowed all involved investigators to access relevant data, regardless of location. All data was initially entered into a Microsoft excel database template and then uploaded into REDCap during the study.

For data analysis, some general characteristics were first stratified by whether the participant was male or

female. 2-sample t-tests for continuous participant characteristics and Pearson’s chi-square test for categorical participant characteristics were used to examine which variables differed significantly between males and females. Bivariate relations between the various participant characteristics and mortality (the outcome of interest) were then examined. For dichotomous exposure variables, 2-sample t-tests were performed while Pearson’s chi- square test was performed for categorical participant characteristics. Multiple logistic regression were performed to estimate odds ratios for mortality based on specific patient characteristics. A multiple logistic regression model was created to estimate adjusted odds ratios, with only pre-endoscopic Rockall score and undergoing endoscopy remaining in the model. Variables without an effect on adjusted mean were removed from the model. Stata 12.0 (Stata Corp, College Station, TX) was used for the analyses.

An international Institutional Review Board (IRB) for this study through the University of North Carolina, Chapel Hill, USA has been approved (IRB Number IRB11-1649) in addition to a local IRB via the National Health Science Research Committee in Malawi (NHSRC #891). Since this was a retrospective study, the main risk to the patients was breach of confidentiality. All data was stored on a secure, password-protected computer with data backed up to the UNC Project secured servers.