For Type I diabetics, the challenge is to regulate exogenous insulin delivery so that it matches metabolic needs of the patient. The long-term goal is to develop a mechanical artificial pancreas that combines a glucose sensor, an insulin pump, and a controller to allow automated regulation of the insulin pump based on the sensor feedback. Appropriate physiologically justified algorithms for the controller are crucial for a successful mechanical artificial pancreas. To develop such algorithms, one must have accurate and validated methods of analyzing glucose sensors output and models for glucose/insulin dynamics. Such models may also improve prediction of hyper- and hypoglycemia in patients now manually controlling the insulin delivery rate depending on the current blood glucose level and expected consumption of carbohydrates. The overall goal of this application is to perform secondary analyses of the data collected in five clinical studies conducted in subjects with Type I and Type II diabetes as well as healthy volunteers at the Artificial Pancreas Center of Thomas Jefferson University, Philadelphia, PA. Statistical models for population-based rather than currently standard individual-based analysis of glucose/insulin dynamics will be developed in the framework of the functional data analysis and evaluated by comparing their fit to the real data. The existing models for single individuals describe glucose kinetics in terms of the systems of generally non- linear differential equations and incorporate numerous latent (immeasurable) variables describing the time-dependent levels of glucose, insulin and glucagon in internal physiological compartments such as heart or liver. The proposed studies will (1) develop subject-specific mixed effects models defined by the systems of physiologically meaningful differential equations for glucose and insulin kinetic;(2) extend statistical methodology to incorporate infinite dimensional regression terms and random effects into the functional models defined by the systems of non-linear differential equations;(3) develop computational algorithms and software necessary to estimate proposed functional models;(4) compare performance of various previously proposed models in terms of prediction accuracy of future blood glucose values and consider new models for glucose and insulin dynamics in diabetics and in healthy subjects.
The standard methods of blood glucose control, including multiple dose insulin therapy, insulin pump therapy, and frequent use of blood glucose meters are not sufficient tools to enable people with type I diabetes to achieve near-normal glucose control with a low risk for hypoglycemia. The ultimate treatment goal for type I diabetes is the creation of a mechanical artificial pancreas combining a glucose sensor, insulin pump, and controller. This project focuses on developing and validating statistical models for glucose/insulin dynamics, which would serve as a foundation for designing computational algorithms for regulating the insulin delivery by the mechanical artificial pancreas.
