A Pattern Matching Algorithm for Self-Adjusting Basal Rates in Insulin Pump Systems

In a Type 1 Diabetic, Insulin can be administered in a pump system. There are two types of insulin that must be given: basal and bolus. Basal insulin is a long-acting form of insulin that works in the background while fasting, while Bolus insulin is rapid/short acting given in response to food to immediately begin working to lower blood sugar.

Modeling in Diabetes can be represented by algorithmic approaches ranging from simple autoregressive models of the Continuous Glucose Monitor time series to multivariate nonlinear regression techniques of machine learning. Other examples of modeling in Diabetes include prediction models of hypoglycemia and even non-linear models of glucose. Regardless of technique, modeling can be used to accurately predict vital trends regarding something as prominent as glucose levels. Data from predicting trends in glucose levels can then be used to potentially influence other parameters within technology like an insulin pump system.

With this being said, if a pattern-matching algorithm could be developed to allow for a “personalized” basal rate, then blood sugar could be more controlled and reflect overall normal levels. This project aims to satisfy this proposal by 1) Developing a pattern-matching method that utilizes multiple stored glucose readings and basal rates in order to predict a new, suggested basal rate. 2) The new Basal Rate and predicted Glucose Levels will be displayed on the user interface of a developed GUI and allow for the user to accept or decline any change; thus, allowing for more controlled glucose readings throughout the day.