Mortality prediction of nonalcoholic patients presenting with upper gastrointestinal bleeding using data mining.

BACKGROUND AND AIM: Patients with acute upper gastrointestinal (GI) bleeding commonly present with hematemesis and/or melena. More studies are needed to confirm the ability to predict mortality, length of stay, and cost. Alcohol abuse may worsen variceal bleeding or portal hypertensive gastropathy in a patient with a history of liver disease. Coexisting alcoholism may influence patient management in the setting of peptic ulceration or existing malignancy. Consequently, the overall morbidities and mortalities may differ in alcoholic and nonalcoholic groups accordingly. Mortality prediction using data mining programs is helpful for detection of significant mortality-related factors.

PATIENTS AND METHODS: We retrospectively reviewed 152 files of patients presenting with upper GI bleeding, because of nonalcoholic causes, 100 males and 52 females aged 16-77 years old. Causes of upper GI bleeding were esophageal and/or gastric varices (51), portal hypertensive congestive gastropathy (6), gastric and/or duodenal ulcers (39), gastroesophageal reflux disease (20), gastritis and duodenitis (19), cancer (8), gastric polyps (3), blood diseases (2), Dieulafoy's lesion (2), and no aberrant cause of bleeding in two patients.

RESULTS: The overall mortality was 29 patients (19.07%). The use of a descriptive model of the data mining program yielded the most significant mortality predictors. The overall accuracy was 92.08%.

CONCLUSION: Chronic hepatitis C virus infection and NSAID-associated splenomegaly because of portal hypertension are significant predictors of mortality in nonalcoholic patients presenting with upper GI bleeding.