Multi-State Markov Model for Analysing Blood Glucose Changes

Abstract

A multi-state Markov model was adapted in order to model glycemic control, based on continuous glucose measurements (CGM). A library was implemented, written in Python, that allows for user-specific input in regards to modelling parameters and analysis. The CGM-readings were collected during two previous clinical trials, involving patients with type 1 diabetes and inadequate glycemic control. The clinical trials involved the administration of dapagliflozin which together with insulin improves glycemic control, compared with only administering insulin. The states of the Markov model were defined based on blood glucose levels, where increased time in the target range, normoglycemia, constituted better glycemic control. Based on the CGM-readings collected, this Markov model was used to analyse how the glycemic control of patients is affected by their kidney function as well as insulin reduction. Results show that improvement in glycemic control due to dapagliflozin is independent of kidney function in the range investigated. When modelling the insulin reduction in patients, it was seen that an increased insulin usage corresponded to increased glucose levels. There is a well established causal relationship between insulin and decreased blood glucose levels. The opposite relation seen in the modelling and data must mean that something is masking the effect. One explanation would be that this is due to the fact that insulin is only a proxy for eating unevenly, but the exact cause is unknown.