Health informatics needs better data sharing

Abstract

Data storage, processing, and understanding have had immense leaps forward over the last 20 years. While computer vision models are able to segment natural images with high resolution, the most popular medical algorithms involve simple addition of one to three variables.
In this talk, I argue the disconnected progress of applied computer science in other fields compared to that of medicine is driven by the lack of publicly available datasets. I describe our work on the Medical Information Mart for Intensive Care (MIMIC) databases, which are completely free to any researcher in the world and form the foundation of numerous research studies. Deidentification of data for inclusion in MIMIC, in particular free-text clinical notes, is challenging but I will highlight our work in the area and emphasize that these challenges are entirely surmountable. I’ll conclude with the impact of MIMIC broadly, and raise some thoughts around accelerating medical research as a whole.

Speaker’s bio

Alistair got a PhD at University of Oxford (UK) in 2014 in Healthcare Innovation Oxford with a thesis titled: Mortality prediction and acuity of illness in critical care. He got a Master in 2009 in Biomedical & Electrical Engineering at Hamilton, ON, Canada.
He is the ideator of MIMIC-CXR, the largest database of chest x-rays with deidentified free-text radiology reports and eICU-CRD, the largest publicly available critical care dataset with over 200,000 stays.
He teaches machine learning at Harvard Medical School at medical students and Data Science in Medicine at the Massachusetts Institute of Technology.
He is co-author of more than 30 papers in prestigious journals.