Google’s artificial intelligence (AI) & machine learning division DeepMind, is going to get its hands on detailed health data from 1.6 million hospitalized patients through a partnership with a group of hospitals in the United Kingdom.
Google has made a number of moves into the big data movement in healthcare in recent years. Several years ago, they launched the Baseline project with Duke and Stanford to define “normal” physiology by collecting tons of data at high frequency on a few thousand people. They also recently won an NIH grant to launch a pilot study with 50,000 patients for the eventual 1-million patient Precision Medicine cohort.
It was announced earlier this year that DeepMind was partnering with hospitals in the UK’s National Health Service to develop a new generation of decision support tools. DeepMind is a division of Google that focuses on artificial intelligence (AI) and machine learning.
One area of early focus is on acute kidney injury (AKI) in hospitalized patients. They worked with a nephrologist to develop an app called Streams. In its current form, it seems like an app that provides active notification of an acutely elevated creatinine with graphs & charts showing prior data.
It was recently revealed that the data sharing agreement between DeepMind and the NHS will give Google access to detailed historical patient data on 1.6 million people going back five years. It’ll include data like labs, medical history, and even hospital bed tracking data. Protections including third-party data hosting, data use limits on Google, and an expiration date in 2017 are also included.
Another app called Hark was also acquired by Google as part of the deal. Hark is much more far-reaching, including everything from task management, communication (i.e. replacing the pager), and a variety of automated alerts for impending patient deterioration.
The data windfall that DeepMind is receiving is ostensibly in support of Streams but it’s been noted that Google is getting a lot more data than you might think relevant on first pass. They’ll also be getting access to some particularly sensitive data like HIV status. And that has raised a lot of concern in the United Kingdom among privacy advocates, particular given some issues the National Health Service has had in that arena recently.
On the other hand, it seems like this collaboration is geared at solving that core big data problem – how to actually drink from the firehose. The promise of machine learning, and AI as applied to big data, is in finding the unexpected signals and trends, things that clinicians would have otherwise never seen. To do that, it’s understandable that Google would want to get its hands on the biggest smorgasbord of messy real world clinical data it can.