Identifying Structure in the MIMIC ICU Dataset

Zad Chin, Shivam Raval, Finale Doshi-Velez, Martin Wattenberg, Leo Anthony Celi

Keywords: healthcare, projection-based clusterings, structure, MIMIC, ICU, UMAP, t-SNE, clustering

TL;DR: We explored the structure of the MIMIC-III dataset from a bottom up approach. We found two dominant structure-sort by either by Fi02 or creatinine, which are important key features validated by our clinician co-authors.

Abstract: The MIMIC-III dataset, containing trajectories of 40,000 ICU patients, is one of the most popular datasets in machine learning for health space. However, there has been very little systematic exploration to understand what is the natural structure of these data---most analyses enforce some type of top-down clustering or embedding. We take a bottom-up approach, identifying consistent structures that are robust across a range of embedding choices. We identified two dominant structures sorted by either fraction-inspired oxygen or creatinine --- both of which were validated as the key features by our clinical co-author. Our bottom-up approach in studying the macro-structure of a dataset can also be adapted for other datasets.