Prediction and Prevention of Hypoglycemia in Veterans with Diabetes

控制高血糖以预防或延缓血管并发症的发生是一个基本目标

Control of hyperglycemia to prevent or delay the onset of vascular complications is a fundamental goal of diabetes care. Intensive treatment is limited, however, by risk of hypoglycemia, a common and potentially hazardous metabolic complication of glucose-lowering treatment. Traditionally, this risk was considered unavoidable during treatment but new models of care focus on improving the safety of treatment while optimizing glycemic control. To monitor for safety and foster better care, work is needed to develop standardized methods and strategies for performance evaluation. The proposed research employs multiple methodologies to address the issue of hypoglycemia and improved safety of diabetes treatment. For identification of the condition in the Veteran patient population with diabetes, we propose to analyze national VA and non-VA structured data and claims, measure patient-reported experience through patient surveys collected from stratified random samples of patients, and develop accurate and efficient natural language processing (NLP) tools to search documentation in the medical records for evidence of hypoglycemia. This will include development of a valid case-finding algorithm. These measures will be combined and compared to obtain a unique and comprehensive evaluation of the condition in the patient population and to provide practical information on the accuracy and completeness of various methods. Patients with identified hypoglycemia will be followed forward to evaluate the risks of subsequent adverse outcomes associated with the condition, including repeat hypoglycemia, preventable hospitalizations, and death. We will combine all available and relevant information and model hypoglycemia to identify predictors in the sample of patients who completed the survey and in the whole VA diabetes population, limiting candidate predictors to factors available from structured medical data or from NLP extractions. We will identify those factors obtained from surveys or NLP extraction that add substantially to the predictive models. These models will inform the process of developing parsimonious predictive algorithms for the whole population and in relevant subgroups. Risk algorithms will include branching to classify patients by contextual factors that are useful in guiding clinical management. The best practical algorithms will be implemented in an integrated system for near real-time hypoglycemia case finding and assignment of diabetes patients by predicted hypoglycemia risks. This work will generate methods and tools for monitoring population health and safety among Veterans with diabetes and for improving care to reduce risks. The near real-time hypoglycemia case-finding and risk assignment system will be available for use by operations and research as it is implemented. This work could form the basis for new measures of care quality, providing technologies for risk adjustment, facility and provider profiling, and practice evaluations. It could be used in generating clinical alerts or as a point of care decision aid, suggesting approaches for tailored risk reduction. Ultimately, it would improve monitoring of population health and safety among Veterans and should facilitate precision medicine in diabetes care, with the emergence of optimal, tailored, and patient centered approaches for managing diabetes.