Lead Investigator: Lisa Koch, University of Bern
Title of Proposal Research: Foundational machine learning models for large scale robust modelling of continuous glucose monitoring and activity data
Vivli Data Request: 10054
Funding Source: The proposed research will be funded by core funding from the University of Bern.
Potential Conflicts of Interest: None
Summary of the Proposed Research:
Improvements in continuous glucose monitoring (CGM) and their wide dissemination have improved diabetes management substantially. However, glycaemic control remains challenging. For example, a myriad of complex factors influence the individual absorption dynamics of meals and administered insulin. A crucial factor is exercise, but other lifestyle factors and demographic characteristics such as age play an important role as well. Artificial intelligence can play an important role for modelling these interactions, but often, the limited size of training data may not allow the development of robust and effective algorithms.
Recent advances in machine learning have led to the emergence of foundational models. Foundational models are powerful general-purpose machine learning models which can be trained on large unlabelled datasets by solving so-called pretext tasks such as predicting masked out parts of the data. Foundation models can then be finetuned to solve specialised tasks on much smaller labelled datasets. They are therefore ideally suited to study clinically relevant research questions, where typically only smaller cohorts are available.
In this project, we aim to develop self-supervised foundational CGM models on the T1-DEXI study data. We will then apply the CGM foundational models to develop downstream models for specialised clinically relevant tasks in T1D management, with the ultimate goal of improving treatment recommendations. In the scope of this project, the specific tasks we tackle are the ML-based detection of exercise and nutrition events from CGM data and the prediction of future glucose levels. To test how well our models can generalise to different populations, we will validate them on the pediatric cohort of the T1-DEXIP study.
Requested Studies:
Type 1 Diabetes EXercise Initiative: The Effect of Exercise on Glycemic Control in Type 1 Diabetes Study
Data Contributor: Jaeb Center for Health Research Foundation, Inc.
Study ID: T1-DEXI
Sponsor ID: T1-DEXI
Type 1 Diabetes EXercise Initiative Pediatric Study (T1DexiP): The Effect of Exercise on Glycemic Control in Youth with Type 1 Diabetes
Data Contributor: Jaeb Center for Health Research Foundation, Inc.
Study ID: T1-DEXIP
Sponsor ID: T1-DEXIP