Application of Bayesian Statistics to Electronic Medical Records

• Activity Number: 229
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Song Zhang*+
• Companies: The University of Texas Southwestern Medical Center

Abstract

An increasing number of hospitals have implemented electronic medical record (EMR) systems. One important challenge faced by biostatisticians is to extract meaningful information from the ever expanding EMR databases and assist medical decision making. In this study we develop statistical methods to detect signs of clinical deterioration during patients' hospitalization, which enables health care providers to direct focused attention to such patients before an adverse event occurs. One difficulty in utilizing EMR data is the frequent occurrence of missing data. A patient's health status is characterized by the current values, as well as the temporal trends, of a large number of variables. It is inevitable that at any given time point, a significant portion of variables involved in the statistical model would be missing. We propose a Bayesian hierarchical method to impute missing data in EMR. It decompose the measurement of each variable into two elements, one representing a smooth temporal trend and the other the tendency to change concurrently with other health variables. We further described how to efficiently apply the statistical learning to support real-time computation.