In this study, we propose an approach to build a detection model for surveillance of healthcare-associated urinary tract infection (HA-UTI) based on the variables extracted from the electronic medical records (EMRs) in a 730-bed, tertiary-care teaching hospital in Taiwan. Firstly we mapped the CDC's HA-UTI case definitions to a set of variables, and identified the variables whose values could be derived from the EMRs of the hospital automatically. Then with these variables we performed discriminant analysis (DA) on a training set of the EMRs to construct a discriminant function (DF) for the classification of a patient with or without HA-UTI. Finally, we evaluated the sensitivity, specificity, and overall accuracy of the function using a testing set of EMRs. In this study, six surveillance variables (fever, urine culture, blood culture, routine urinalysis, antibiotic use, and invasive devices) were identified whose values could be derived from the EMRs of the hospital. The sensitivity, specificity and overall accuracy of the built DF were 100 %, 94.61 %, and 94.65 %, respectively. Since most hospitals may adopt their EMRs piece-by-piece to meet their functional requirements, the variables that are available in the EMRs may differ. Our approach can build a detection model with these variables to achieve a high sensitivity, specificity and accuracy for automatically detecting suspected HA-UTI cases. Therefore, our approach on one hand can reduce the efforts in building the model; on the other hand, can facilitate adoption of EMRs for HAI surveillance and control.
ASJC Scopus subject areas