I test novel Hyperbolic embedding methods to learn representations of ICD-9 codes using co-occurrences in real hospital data.
I evaluate the performance of these embedding approaches in terms of their ability to group similar diagnoses together.
Finally, I use exploratory data analysis to understand the structure of the MIMIC-3 dataset and prepare the data for predicting mortality.
I briefly test whether features based on these ICD embeddings are useful predictors of in-hospital mortality.