Model-Based Algorithm for Regimen Identification in Treatment of T2D

Sarah Ellinor Engell: Fast and Affordable Optimization of Insulin Treatment. Data from Artificial Pancreas systems may hold the key to reaching treatment goals within Type 2 Diabetes.

Type 2 Diabetes (T2D) is a growing pandemic with severe consequences for the individual and society. Several treatment options, tablets, non-insulin injections and insulin exist, but still few people with T2D reach treatment targets. In the long run, poor treatment outcomes can lead to severe complications, e.g. amputations and blindness, and a significant socio-economic burden. To control blood glucose in late stage T2D, insulin often becomes necessary. With standard of care algorithms, initiation and personalization of insulin treatment, known as titration, is a lengthy process that may take months to years. Throughout the titration period, the patients are faced with the complexity of dose adjustments and the fear of insulin overdosing. Finding and implementing new standards of automated care are key to improving treatment by reducing the burden on the patient and making every health care professional (HCP) a specialist in diabetes care.

Over the years, closed loop (CL) systems, also referred to as Artificial Pancreas systems, have developed from an academic interest to a viable treatment option, mainly for people with type 1 diabetes. State of the art CL systems use sensor measurements from a continuous glucose monitor (CGM) as input to an adaptive control algorithm to optimize insulin doses and administer the selected dose to the individual through a connected insulin pump. This setup mimics how a healthy body regulates glucose levels. As a result, CL systems can partially or completely automate insulin treatment. A few inpatient studies have proven fully automated CL systems feasible for people with T2D, but the price of a CL system may make long-term use of CL treatment inaccessible to a large proportion of T2D patients. This calls for alternative approaches to optimizing the individuals’ treatment regimen in an affordable setup, e.g. CGMs and injection pen-based treatment. We hypothesize that data from short-term usage of a CL system can be used to map a personalized, injection pen-based drug and dose scheme for people with T2D.

In this project, we will investigate whether a correlation exists between the profile of fast-acting insulin delivered in a pump and the amount of long-acting insulin delivered with an injection pen. The investigations will be based on data from a clinical trial funded by Novo Nordisk A/S and the Danish Innovation Fund as part of the ADAPT-T2D innovation consortium. We will explore how the basal rate during CL treatment can be translated to a pen-based insulin treatment. For the translation, we will use dose-response models and system identification methods. Additionally, we will examine how CGM readings alone can be used to identify the user’s need for intensifying treatment with meal-time insulin and investigate whether the identification can be improved using CL data.

The outcome of this research project is an individualized model-based algorithm for reaching blood glucose targets with pen-based treatment, through automated selection of the drug, timing and dose.

PhD project

By: Sarah Ellinor Engell

SectionScientific Computing

Principal supervisor: Dimitri Boiroux

Co-supervisor: John Bagterp Jørgensen

Project title: Model-Based Algorithm for Regimen Identification in Treatment of T2D

Term: 01/08/2020 → 31/07/2023


Sarah Ellinor Engell
Industrial PhD
DTU Compute


John Bagterp Jørgensen
DTU Compute
+45 45 25 30 88