Predictive Analytics: Closing the Loop in the Imperfect World
Advanced models based on patient physiology have started to be utilized across healthcare enterprises to, in theory, accelerate and optimize care. These predictive analytics have continued to escalate on the technology hype-cycle and are beginning to reach or exceed the “peak of inflated expectations” and head towards the “trough of disillusionment.” Many of the problems that occur with predictive analytics are a lack of understanding of exactly what the models are predicting and their performance. Additional concerns come from using the output of the models to deliver care. Frequently, such outputs are delivered directly to providers without education on what to do with the information. This lecture will attempt to create a construct to use predictive models successfully in organizations to help optimize care and return value on a model’s implementation.
James M. Blum, MD, FCCM
Associate Professor of Anesthesia and Chief Medical Information Officer
University of Iowa
Being trained as a computer scientist, I entered medicine with the thought of contributing to the improvement of healthcare through data analysis and machine learning. To date, most of my work has used informatics, large datasets, and computational methods to understand and optimize treatment of the critically ill. In particular, my interests have centered around the study of the acute respiratory distress syndrome (ARDS) and sepsis using these techniques, but the techniques are amenable to research in several different domains in the surgical and critically ill populations.
Objectives for CE:
After attending the webinar, participants should be able to:
- Understand existing state of predictive models based on EMR data
- Explain basic statistics to assess model performance
- Describe successful implementations of predicitive models in health systems