Google’s artificial intelligence (AI) focused company DeepMind has landed a five year deal with UK’s National Health Service to deploy its Streams health app in hospitals.
Streams is an AI-powered health app designed to predict and thereby help avert adverse patient outcomes, starting with acute kidney injury in hospitalized patients. To do that, the Royal Free London NHS Foundation Trust shared data on over a million patients with DeepMind to help it develop & train Streams.
Machine learning and other forms of AI offer tremendous opportunity to turn the enormous amounts of health data being generated by patients, hospitalized and ambulatory, into something actionable. IBM’s Watson Health is applying this technology to better tailor chemotherapy for cancer patients. General Electric (GE) is working with UCSF to use AI to “read” imaging studies and help direct radiologists more quickly to patients with critical findings.
In this partnership, DeepMind is building Streams to, among other things, identify hospitalized patients at risk of adverse events before they happen, giving us an opportunity to intervene. Superficially, that may seem like those annoying “This patient may have sepsis” alerts that many of us frequently see and ignore because they are too frequently wrong. The key difference, though, is that those alerts are often built on rigid rules – think SIRS criteria. Using different kinds of AI, Streams can be trained to develop and refine its own “rules” as it gets more data and feedback on when it was right or wrong. That is, it can get better over time and help analyze patient data at a scale that would just be impossible for clinical teams.
According to Mustafa Suleyman, Co-Founder & Head of Applied AI at DeepMind,
Over the course of the next five years, we’re going to expand Streams to cover other illness where early intervention is key and technology can ensure this happens. We think that Streams could also be used to help patients at risk from sepsis and other causes of organ failure, where signs of deterioration are often difficult for clinicians to spot, and where early intervention can be the difference between life and death. We also plan to build additional features that Royal Free clinicians have asked for, including messaging and clinical task management that will support better care.
We frequently talk about the information overload facing clinicians, which is precisely what these AI-powered systems are designed to tackle and at a far bigger scale than the information that makes it to our desks (or in baskets).