Multimodal-fusion-severe-hypo-data

Datasets (raw and preprocessed) for reproducibility results of the paper "Interpretable and multimodal fusion methods to identify severe hypoglycemia in adults with T1D". We also added the reports and results obtained in the multiple experiments. The file "Multimodal-fusion-severe-hypo-data" contains the last version and results. This study aims to identify severe hypoglycemia (SH) in Type 1 diabetes patients, leveraging machine learning and explainable artificial intelligence (XAI) techniques. Using clinical data from the Jaeb Center for Health Research, including tabular, text, and time series data, multimodal fusion models outperform single-modality models by 5.8%, achieving an AUCROC of 0.779. XAI reveals a higher incidence of cardiovascular diseases and associated drugs in individuals with SH, along with a correlation between consistently low blood glucose levels and SH risk. These findings hold promise for early detection and intervention, benefiting both clinicians and patients.