Leveraging the Power of the EMR: Using a Real Time Prediction Model to Decrease Inpatient Hypoglycemic Events

Our goal for this Learning Healthcare System Demonstration Project is to reduce the rate of inpatient hypoglycemia. Hypoglycemia can result in longer lengths of stay and increased morbidity and mortality (ie falls and cardiovascular or cerebral events).

The group at Washington University (WSL) developed a predictive hypoglycemia risk score. Using current glucose, body weight, creatinine clearance, insulin type and dosing, and oral diabetic therapy, they identified patients at high risk for hypoglycemia and then provided in-person education to the providers of these patients. This resulted in a 68% reduction in severe hypoglycemia (blood glucose < 40 mg/dL). This approach required significant personnel hours and is difficult to replicate in other systems.

The investigators will implement an EHR-based intervention at UCSF to predict which patients are at high risk of inpatient hypoglycemia and take action to prevent the hypoglycemic event. In real time, all adult (non OB) patients with a glucose < 90, and a high risk of future hypoglycemia (based on the WSL formula) will be identified. Patients will be randomly assigned to intervention or no intervention (current standard care). The intervention will consist of an automated provider alert with recommendations on what adjustments could be made to avoid a potentially serious hypoglycemic event.

The outcomes that will be measured include: 1) reductions in serious hypoglycemic events, 2) monitor the changes made by providers as a result of alerts in order to study provider behavior and identify future areas of intervention, and 3) provider satisfaction with the alert system.