Julie Simmons Ivy, Ph.D.
Professor in the Edward P. Fitts Department of Industrial and Systems Engineering
North Carolina State University
Abstract: Optimizing the First Response to Sepsis: An Electronic Health Record-based Markov Decision Process Model for Personalizing Acute Care for Deteriorating Patients
Sepsis is considered a medical emergency where delays in initial treatment are associated with increased morbidity and mortality. It is a life-threatening organ dysfunction caused by a dysregulated host response to infection. Sepsis affects more than 1.7 million Americans each year, causing approximately 270,000 deaths annually. One in three hospitalized patient deaths are associated with sepsis. In 2019, the total cost of sepsis care for inpatient admission and skilled nursing facility admission was estimated at more than $62 billion. Sepsis is a significant healthcare challenge, where the lack of a gold standard for diagnosis causes inconsistencies in categorizing sepsis phenotypes and accurately capturing patients’ trajectories, which evolve stochastically over time. This makes treatment decision making and early intervention difficult.
We integrate electronic health record (EHR) data with clinical expertise to develop a continuous-time Markov decision process model of the natural history of sepsis. We use this model to better understand the stochastic nature of patients’ health trajectories and determine the optimal treatment policy to minimize mortality and morbidity. Specifically, the optimal health states for first anti-infective and first fluid are identified. We formulate this as a stopping problem in which the patient leaves the system when he or she receives the first treatment (intervention) and receives a lump sum reward. Our objective is to find the optimal first intervention for health states to minimize expected mortality and morbidity. We explore the effect of the complex trade-offs associated with the intervention costs and patient disposition costs which are subjective and difficult to estimate. Our model captures the natural progression along sepsis trajectory using a clinically defined treatment delayed population. The model translates observations of patient health as defined by vitals and laboratory results recorded during hospitalization in the EHR to capture the complex evolution of sepsis within a patient population. This framework provides key insights into sepsis patients’ stochastic trajectories and informs clinical decision making associated with caring for these patients as their health dynamically evolves.
Tune in via Zoom:
Meeting ID: 993 4735 4264
Department of Industrial & Systems Engineering