Google recently published a study online (not peer reviewed) of its AI system that predicts medical outcomes such as in patient mortality, readmission, etc
There is a lot of technical jargon in the paper, but here is a brief summary:
-Previously, predictive models required selection of variables and standardization of data, which is generally very labor intensive
-Google’s deep learning approach used raw EHR data including free text notes, eliminating the bottle neck of data standardization
-To validate this system, Google used raw EHR data from over 200,000 adult patients who were hospitalized for more than 24 hours at either UC San Francisco Medical Center or University of Chicago Medicine
-High accuracy was achieved for predicting in-hospital mortality, 30-day unplanned readmissions, prolonged length of stay and all of a patient’s final discharge diagnoses
-Per the study authors: “We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios, complete with explanations that directly highlight evidence in the patient’s chart”
