Abstract:
In recent times, continuous glucose monitoring (CGM) devices are becoming very popular, especially among clinicians and patients suffering from diabetes (both T1D and, T2D). They help in management of patients BG levels, in the formulation of the patient’s medication regime, etc. These CGM devices measure the concentration of glucose in the interstitial fluid periodically (usually every 5 min to 15 min depending on the device) for several days (usually from 7 days to 14 days, depending on the device) and provide the user with a sequence of time stamped ISF measurements, commonly known as the CGM trace. These devices are gradually becoming cheaper and more robust and in recent times are quite indispensable to many diabetes patients and clinicians. Our goal is to use the CGM trace to estimate all other important metrics like,
HbA1c, FBG, etc. This would not only reduce the total cost of diagnosis and management but also would remove the need for repeated blood-draws from the patient. In this thesis we show that the standard model used for estimating HbA1c from CGM traces, the Nathan model, cannot provide statistically reliable HbA1c estimates for Indian CGM traces at a population level. We then introduce two new models, which are able to provide statistically reliable HbA1c estimates for the Indian T2D population. We then show that the Direct Model is also able to provide reliable HbA1c estimates for the Indian T1D population. Next, we provide a method, based on trajectory optimization, for estimating the continuous BG and ISF traces from any given discrete CGM trace.
