Abstract
In the past, healthcare practitioners believed that patient outcomes were dependent almost exclusively on: training, capability and skill of individual clinician; patient characteristics; specifics of the illness or procedure being performed. New insights from the study of complex work environments suggest that clinical outcomes may be strongly influenced by the structure or design of the system in which care is delivered. In this project, we will examine data collected in ICUs of large, hospital-based health systems and consider how to reduce two key adverse outcomes among such patients: death from sepsis, and agitation leading to self-extubation (accidental removal of a breathing tube by a patient, which leads to certain death, if not corrected very quickly).
We hypothesize that we will be able to predict, within suitable confidence limits, which patients will experience these outcomes in the context of current system design and conventional processes of care using empirical data obtained from the monitoring systems. In our project, models will be derived to perform both classification and prediction using neural networks and/or fuzzy logic. Our analyses should provide a robust quantitative basis for informing key clinical policies and practices that can be used for redesigning the overall system processes and functions. Further, our work should provide a reproducible framework for mining data sources to identify practices in other venues of hospital care that would be candidates for systems-based interventions to improve outcomes.