Artificial Intelligence-Assisted Clinical Decision Support for Preventing Hypoglycemia in Hospitalized Patients - PROJECT SUMMARY Dr. Aileen Wright, MD, MS, is an Instructor in the Department of Biomedical Informatics and Department of Medicine at Vanderbilt University Medical Center (VUMC). Her long-term goal is to become an independent physician scientist bridging the gap between artificial intelligence (AI) and healthcare. To this end, Dr. Wright seeks a mentored career development award to develop an AI-assisted clinical decision support (CDS) tool for diabetes care. Hypoglycemia is the most common complication of insulin therapy, and has been associated with increased risk of acute coronary syndrome, stroke, falls, length of stay, and mortality. Due to the severe consequences of hypoglycemia, insulin therapy is on the Institute for Safe Medication Practices’ list of high-risk medications and insulin-induced hypoglycemia has been designated a “never event”. Models to predict hypoglycemia in hospitalized patients have been developed, and could be implemented as CDS tools to improve the safety of insulin therapy. However, published models could benefit from improvements in accuracy. Furthermore, these models have not yet translated to tangible clinical outcomes as they have not been integrated into the electronic health record (EHR). For AI-assisted CDS tools to be effective, they must be developed with clinician input throughout the design process to ensure tools are utilized, fits into the clinician workflow, and reduce clinical workload rather than increasing cognitive burden. CDS which is not accurate and fits poorly into the clinician workflow, can contribute to ‘alert fatigue’, user dissatisfaction with the EHR, and clinician burnout. The objective of this proposed study is to use state-of-the-art machine learning methods and human-centered design processes to develop a high performing AI-assisted CDS tool for preventing hypoglycemia in hospitalized, non-critically ill adults. The specific aims of this proposal are to 1) validate and extend existing inpatient hypoglycemia models, 2) expand feature space and apply deep learning to hypoglycemia prediction, and 3) integrate and prospectively validate prediction models for diabetes care. Dr. Wright is a practicing general internal medicine physician who has completed a fellowship in biomedical informatics. During the award period, Dr. Wright’s research and career objectives include broadening her methodological foundation in machine learning to include deep learning techniques, gaining experience integrating clinical prediction models into the EHR, and forging new collaborations in informatics and clinical medicine. These objectives will be met through a combination of didactic coursework, mentored research, and career development activities. This award will facilitate Dr. Wright’s transition to an independent investigator who develops, implements, and evaluates AI-assisted CDS tools to transform healthcare, preventing harm for patients with diabetes and lifting burdens for clinicians.
