JOPT2025
HEC Montreal, 12 — 14 May 2025
JOPT2025
HEC Montreal, 12 — 14 May 2025

AI-based approaches for healthcare management
May 13, 2025 03:30 PM – 05:10 PM
Location: EY (Blue)
Chaired by Monia Rekik
4 Presentations
-
03:30 PM - 03:55 PM
A comparative study of data-driven and personalized models trained to predict blood glucose levels of type-1 diabetes patients exercising in free-living conditions
People living with Type-1 Diabetes (T1D) require continuous insulin injections to prevent prolonged high blood glucose levels (hyperglycemia). They also frequently experience episodes of hypoglycemia (low blood glucose levels) when they are physically active. These episodes significantly impact their quality of life and can result in serious physical complications. Effective T1D management depends on a comprehensive understanding of each individual’s glycemic behavior, which is influenced by factors such as exercise, diet, stress, and lifestyle habits. The development of new technologies has generated a vast amount of data that can be analyzed to better understand and predict their glycemic behavior. Our work aims to assess whether a data-driven approach can accurately and safely predict the blood glucose levels of patients exercising in free-living conditions. To do so, we tested and compared multiple machine learning and deep learning regression models. Each deep-learning model was implemented twice: first, as a personalized model trained solely on the target patient’s data, and second, as a fine-tuned model of a population-based training model. We trained and tested our models on the data of 79 real patients from the Type 1 Diabetes Exercise Initiative (T1DEXI). In this talk, we will present and discuss the results obtained.
-
03:55 PM - 04:20 PM
Blood Glucose Prediction in Pediatric Type 1 Diabetes During Exercise Using ResLSTM in Free-Living Conditions.
Blood glucose prediction is critical for the proactive management of Type 1 Diabetes (T1D), particularly in children. This study utilizes both the pediatric and the adult Type 1 Diabetes Exercise Initiative (T1DEXIP and T1DEXI, respectively) datasets to develop and validate a robust deep learning model, ResLSTM, for accurate blood glucose prediction in real-life, free-living conditions for pediatric T1D patients. The developed model integrates essential variables known to influence blood glucose dynamics, including continuous glucose monitoring (CGM) data, insulin doses administered via insulin pumps, carbohydrate intake, and physical activity. This comprehensive approach ensures that the model accurately captures diverse factors affecting glucose fluctuations. Model performance was rigorously evaluated using standard metrics such as Root Mean Square Error (RMSE) and the percentage of predictions falling within clinically risky zones (zones D and E of the Clarke Error Grid). Results demonstrated strong predictive performance for forecasting glucose levels for 84 pediatric patients at horizons of 10, 20, and 30 minutes.
-
04:20 PM - 04:45 PM
Clustering Glycemic Profiles to Improve Glucose Forecasting in Type 1 Diabetes
This work introduces a clustering-based framework for improving glucose forecasting in individuals with type 1 diabetes (T1D). Multivariate time series segments comprising variables such as glucose levels, physical activity, and carbohydrate intake are encoded into fixed-length embeddings using a bi-directional GRU-based autoencoder with attention. The embeddings are then clustered. A dedicated prediction model is trained on each cluster to forecast future CGM (continuous glucose monitoring) values. Our dataset includes real data for 380 patients and 4 weeks. Our results show relatively small prediction errors and safe predictions for the majority of T1D patients.
-
04:45 PM - 05:10 PM
Blood glucose levels prediction for people living with type 1 diabetes and using multiple daily injections
In this research, we apply deep learning models to predict glycemic values in people living with Type 1 Diabetes (T1D) exercising in free living conditions and using Multiple Daily Injections (MDI). Predictions are made for three different prediction horizons of 10, 20, and 30 minutes. We use the T1DEXI dataset to train and test our models. Our results show that most of the T1D patients using MDI included in the study can be well predicted even during and after exercising. In this talk, we will present and discuss our results.