Characterization of intravenous medication administration in an intensive care unit

Abstract

This project focuses on characterizing intravenous (IV) medication administration in an intensive care unit at a partner hospital. Information regarding IV medication dose was extracted from MIMIC II, a large database containing real patient data; this information was used to characterize the use of twelve hemodynamic drugs. Characterization was performed by extracting features such as maximum dose and overall shape from each trend plot. Additionally, because the administration of vasoactive drugs is generally accompanied by a change in blood pressure, several methods were explored of representing patient state by combining the mean blood pressure and drug dose trends to gain more information than can be obtained by each trend alone. The results of drug use characterization show that an adequate picture of drug use can be gained by examining the characteristic shape of the dose trend in addition to features such as maximum dose administered. The patterns of medication administration have been shown to be indicative of overall patient state. The development of algorithms which match drug use trends to underlying physiology may aid in the annotation of large databases such as MIMIC II, and may also prove useful in tracking the hemodynamic state of a patient during his or her stay in intensive care.