Statistical Problems in Closed-Loop Diabetes Control

点击翻译按钮获取中文摘要

The project focuses on developing statistical models, methods and related theory for closed-loop diabetes control. An artificial pancreas with a closed-loop insulin delivery system, still in an early stage of its development, is expected to revolutionize the way diabetes is treated. As the PIs and many others recognized, a major impediment to the goal of developing an artificial pancreas is the unreliability of the glucose sensor. Sensor technology is not so new and it is also remarkably clever, but frequent recalibration is needed because of the physical processes underlying the way sensors work. The proposed project will solve the problem of proper recalibration by appropriate physical modeling of the metabolic processes involved. Sensors measure current in interstitial space (subcutaneous fat) at fixed time increments all around the clock. The current is nominally proportional to the amount of glucose in the space, but there are two problems with the measurements. Firstly, there is a delay in diffusion of glucose from the bloodstream into fat so that the glucose density in the fat lags the glucose density in the bloodstream. Secondly, there is a defense mechanism (white blood cells) which surround the electrode as a foreign body and attempt to get rid of it via biofouling. The white blood cells interfere with current flow and produce erroneous measurements unless accurate recalibration is performed. The proposed approach has not been tried and is expected to significantly outperform the current implementation of the sensor technology based on more straightforward regression without gaining the benefit and insight of the proposed physical modeling. A new differential equation approach will be used to deal with the delay problem related to glucose sensor. The differential equation is widely accepted and the rate for the diffusion of glucose from the blood into interstitial space governs the relationship between the glucose densities in the bloodstream and interstitial spaces. The diffusion rate and the effect of biofouling will be estimated from finger stick metered measurements taken a few times a day. A main innovation in the proposed approach is statistical models using the physics of the delay and biofouling problems.An artificial pancreas will be a godsend to the millions of Americans faced with the 24 hour a day tedium of having to decide when and how much insulin to inject. If this could be done automatically for them then the only task they would continue to have is to remember to replace insulin in the well of their insulin pump. An artificial pancreas would give them an effective treatment for their disease.