Validating the therapy prediction model through a breakdown analysis on ICU patient medical records

Abstract

With the rapid advancement of computational data analysis tools, medical informatics has emerged as a discipline that explores the use of medical information in clinical practice. It searches for ways to effectively integrate as much information as is available to physicians when they make clinical decisions and represent the information in the most intelligent way possible. As part of an overall effort to develop a program that assists physicians in making clinical decisions on patients with heart disease, we developed a model for predicting therapy effects in heart disease using signal flow analysis that describes constraint relations among physiological parameters. In order to accurately describe and predict the therapy effects on a patient in heart failure, the model needs to be tested and analyzed with real-life patient data including any cardiovascular parameters measurable in the patient. This thesis will present methods for extracting hemodynamic relations and drug effects from patients in the intensive care unit. In this thesis, we propose to test our hypothesis that significant relationships between hemodynamic parameters can be derived from certain classifications of patients and sectioning of hospital stays, and explore the effects of drugs on patients with different sets of diseases.