SAN FRANCISCO — It was the fall of 2015 when researchers from Google and the University of California, San Francisco, first sat down together to hash out a research collaboration in an area that has since exploded with activity: using artificial intelligence to make predictions in the hospital.
Those initial discussions resulted in a contract, signed by both parties a few months later, mapping out an agreement under which UCSF would freely share deidentified patient data with Google — and stipulating what the tech giant would be allowed to do with the information, which covered at least 1.4 million patients. The goal was to see whether Google’s algorithms could predict whether patients had died in the hospital or whether they’d been quickly readmitted after discharge.
STAT obtained a copy of that 2016 agreement, stamped on every page with a “CONFIDENTIAL” watermark and published here for the first time, via a public records request filed with UCSF. It offers an unprecedented look at one of Google’s patient data-sharing deals at a time when it and other tech companies are moving to strike similar arrangements — and generating controversy.