Data Mining to Predict the Occurrence of Resistant Infection

Overview

In order to demonstrate the use of predictive modeling in SAS Enterprise Miner, the paper will examine the problem of resistant infection in the hospital using data from the National Inpatient Sample. The National Inpatient Sample contains approximately 8 million records and represents 37 states. A sample of 800,000 inpatient events or about 10% of the records will be used to investigate the problem. Of 800,000 events, 6000 or 0.75% of the records had a diagnosis of resistant infection. Using data visualization techniques and predictive modeling as well as SAS Text Miner, the paper found the infection treated with an infusion of antibiotics in only 197 patients, with an infusion of linezolid in 95 patients, indicating the use of antibiotics is under-reported, or under-utilized.