Streams app needs access to identifiable patient data - Photo credit: Flickr, Medill DC, CC BY 2.0

The app, Streams, aims to provide doctors and nurses with real-time information on in-patients to help healthcare staff prioritise clinical work. DeepMind said that when it is fully built it will to cut the time it takes to alert staff of patients in need from hours to a few seconds.

The company first announced the work in February, in the form of a project between DeepMind and London’s three Royal Free NHS Trust hospitals that would share data to help the company develop an algorithm to create alerts for people at risk of kidney failure.

However, the company came under fire when in April New Scientist revealed that the data-sharing agreement went further than first thought, giving the company access to a range of identifiable healthcare data on the 1.6 million patients passing through the hospitals each year, as well as historic data from the last five years.

Related content

NHS grants Google access to 1.6 million patient records
Government told to establish effective regulation on AI
Earning public trust in the age of cyber threats

At the time the company said that it needed access to all data – which must be identifiable in order to direct clinicians to the patient in question – because there was not a separate dataset for people with kidney conditions.

Campaigners said that the initial agreement – despite stating that Google could not use the data anywhere else in its business, that the data would be stored by a third party and that data would be deleted when the agreement expired – was opaque on how data would be protected and how patients could opt out.

In addition, the company and trust failed to register the Streams app with the Medicines and Healthcare products Regulatory Agency as required.

Now, the company has moved to restart the work on Streams by signing a new data-sharing agreement that supersedes and extends the deal; it will now run for five years from now, rather than ending in September 2017.

At the same time, it has committed to being more open with the public, publishing the master services agreement and the information processing agreement that detail the workings of the partnership and how patient data will be stored, encrypted and used.

In a blogpost about the deal on its website, DeepMind said that the new deal would introduce “an unprecedented level of data security and audit”.

It said: “All data access is logged, and subject to review by the Royal Free as well as DeepMind Health’s nine independent reviewers. Our software and data centres will also undergo deep technical audits by experts commissioned by our independent reviewers.”

The reviewers – who were brought into the project at the outset in February – include former Government Digital Service boss Mike Bracken, former president of the Academy of Medical Sciences John Tooke and the editor-in-chief of The Lancet medical journal Richard Horton.

DeepMind has also indicated the app is now registered correctly with the MHRA and that the first version of Streams would be available to clinicians across the Royal Free hospital sites in early 2017.

Beyond this, DeepMind said that the work would expand to cover other illnesses that require early intervention, such as sepsis or organ failure, and that it would add in features that Royal Free clinicians have asked for, including messaging and clinical task management.

The Streams app does not currently use AI, but DeepMind said that there was potential for machine learning – which requires massive amounts of data to train computers – in the future.

“The reality is that current IT infrastructure in the NHS is way behind where it needs to be to make this kind of healthcare innovation possible,” DeepMind said.

“Working with hospital trusts to fix more fundamental problems means we can have an immediate impact and lay the foundations for more advanced technology in due course.”