Developing and Validating a Prediction Tool for Diabetic Ketoacidosis Among Diabetes Patients

ABSTRACT Diabetes affects an estimated 30.3 million people in the United States (9.4% of the population). Diabetic ketoacidosis (DKA) is a life-threatening complication of diabetes characterized by metabolic acidosis, increased ketone concentration and uncontrolled hyperglycemia. Though largely preventable, DKA is one the most common acute complications of diabetes. In the US, rates of DKA hospitalization are increasing at a rate of ~6% per year. In addition to the increasing incidence of DKA, the costs associated with DKA hospitalizations are increasing. From 2003-2014, the average hospital charge for a DKA hospitalization increased by 40% (to $26,566 per hospitalization) resulting in an annual aggregate charge to the US healthcare system of nearly $5 billion. This imposes a massive financial burden on the healthcare system. It also represents a significant source of morbidity for individual patients as DKA is associated with a range of negative health outcomes, including cerebral edema, acute renal failure, and increased risks of dementia, re-hospitalization, and death. Among those who experience recurrent hospitalizations for DKA, the two-year mortality rate is as high as 23%. A number of predictors of DKA have been identified across a range of studies. Strategies to reduce rates of DKA have also been identified. Because DKA is largely preventable and the risk factors are well-established, evidence-based prediction models are particularly well-suited to address this problem. Prediction models incorporate risk factors into algorithms to predict an individual’s future risk of an event. They can be used to stratify large populations of patients and provide individualized information to inform patient management. The overall objective of this application is to develop a DKA prediction tool to identify patients at high-risk for DKA-related hospitalizations by leveraging data from the Kentucky Statewide Inpatient Database and electronic health record (EHR) data from two healthcare systems. The proposed work will result in a developed and validated DKA prediction model that has the potential to improve long-term health for those at risk for DKA.