Description
Severe hypoglycemia (blood sugar < 40 mg/dL) is a rare but important adverse event in hospitalized patients associated with morbidity, mortality, and increased length of stay. Current practice is largely reactionary rather than proactive. We built two predictive models (multivariate logistic regression and decision tree) of severe hypoglycemia using a large multi-center inpatient cohort with commonly-measured electronic health record variables. The models displayed good performance and hold promise for deployment into EHR workflows.
Learning Objective: - Understand the importance of inpatient hypoglycemia and the current clinical practices of hypoglycemia prediction
- Learn the methodology behind data extraction from the EHR, data cleaning, as well as model construction from big data
- Learn how to approach model evaluation and how to navigate the transition from model development to model deployment
- Form an understanding of logistical challenges in model deployment in clinical settings
Authors:
Michael Simonov (Presenter)
Yale School of Medicine
Silvio Izucchi, Yale School of Medicine
Francis Wilson, Yale School of Medicine
Presentation Materials: