Google AI Has Access To Confidential NHS Patient Data

Google artificial intelligence (AI) has access to NHS patient data

A leaked data-sharing agreement between Google-owned artificial intelligence (AI) company DeepMind and the NHS reveals that Google has access to sensitive patient data belonging to 1.6 million citizens in the UK. 

The collaboration between Google and the UK’s National Health Service goes beyond what has publicly been admitted, according to the document obtained by New Scientist magazine. reports:

The agreement gives DeepMind access to a wide range of healthcare data on the 1.6 million patients who pass through three London hospitals run by the Royal Free NHS Trust – Barnet, Chase Farm and the Royal Free – each year. This will include information about people who are HIV-positive, for instance, as well as details of drug overdoses and abortions. The agreement also includes access to patient data from the last five years.

DeepMind announced in February that it was working with the NHS, saying it was building an app called Streams to help hospital staff monitor patients with kidney disease. But the agreement suggests that it has plans for a lot more.

This is the first we’ve heard of DeepMind getting access to historical medical records, says Sam Smith, who runs health data privacy group MedConfidential. “This is not just about kidney function. They’re getting the full data.”

The agreement clearly states that Google cannot use the data in any other part of its business. The data itself will be stored in the UK by a third party contracted by Google, not in DeepMind’s offices. DeepMind is also obliged to delete its copy of the data when the agreement expires at the end of September 2017.

All data needed

Google says that since there is no separate dataset for people with kidney conditions, it needs access to all of the data in order to run Streams effectively. In a statement, the Royal Free NHS Trust says that it “provides DeepMind with NHS patient data in accordance with strict information governance rules and for the purpose of direct clinical care only.”

Still, some are likely to be concerned by the amount of information being made available to Google. It includes logs of day-to-day hospital activity, such as records of the location and status of patients – as well as who visits them and when. The hospitals will also share the results of certain pathology and radiology tests.

As well as receiving this continuous stream of new data, DeepMind has access to the historical data that the Royal Free trust submits to the Secondary User Service (SUS) database – the NHS’s centralised record of all hospital treatments in the UK. This includes data from critical care and accident and emergency departments.

Royal Free did not respond to New Scientist’s questions about what opt-out mechanisms are available to its patients. There is already a way for patients to opt out of SUS data collection, but it is not straightforward and involves writing to your GP. But this does not cover live data on admission, discharge and transfer of patients.

The document also reveals that DeepMind is developing a platform called Patient Rescue, which will provide data analytics services to NHS hospital trusts. It states that Patient Rescue will use data streams from hospitals to build other tools, in addition to Streams, that could carry out real-time analysis of clinical data and support diagnostic decisions. One aim, the agreement says, is for these tools to help medical staff adhere to the UK’s National Institute for Health and Care Excellence guidelines.

DeepMind is not planning to automate clinical decisions – such as what treatments to give patients – but says it wants to support doctors by making predictions based on data that is too broad in scope for an individual to take in.

Comparing a new patient’s information with millions of other cases, Patient Rescue might be able to predict that they are in the early stages of a disease that has not yet become symptomatic, for example. Doctors could then run tests to see if the prediction is correct.Google declined to discuss what other kinds of tools it could build on the Patient Rescue platform – for instance by setting its artificial intelligences to work on huge volumes of data from millions of patients. However, it has previously stated that early detection of septicaemia – blood poisoning that kills 31,000 people in the UK every year – might be one future application.

The data that DeepMind is collecting will let it make predictions about any disease it wants, says Smith. “What DeepMind is trying to do is build a generic algorithm that can do this for anything – anything you can do a test for.”

We already know the kinds of things that are possible when machine learning is combined with large amounts of high quality medical data. David Clifton, who runs the Computational Health Informatics Lab at the University of Oxford, says his group has already deployed machine learning tools across the four hospitals that are part of the Oxford University Hospitals NHS Foundation Trust. “If you’re in one of these hospitals you’re being monitored with our stuff,” says Clifton.

Making predictions

“Healthcare is going digital in the 21st Century with huge benefits to patients,” says UK life sciences minister George Freeman. “But NHS patients need to know their data will be secure and not be sold or used inappropriately, which is why we have introduced tough new measures to ensure patient confidentiality.”

Google says it has no commercial plans for DeepMind’s work with Royal Free and that the current pilots are being done for free. But the data to which Royal Free is giving DeepMind access is hugely valuable. It may have to destroy its copy of the data when the agreement expires next year, but that gives ample time to mine it for health insights.

Data mining is the name of the game in the burgeoning field of machine learning and artificial intelligence, and there’s no company in the world better at that than Google.

As well as monitoring the health of individual patients, these systems also watch out for outbreaks of infectious disease. Clifton and his colleagues are also using machine learning on data from Oxford Health NHS Foundation Trust, which provides in-home social care. Clifton says they can use the data to predict things like when a person might develop bipolar disorder.

It is not about replacing doctors or nurses, says Clifton. “It’s about how can we bring the attention of medics to the right place,” he says. “We originally did this in jet engines. There the goal is the same – bringing human expertise to bear on the right part of the system.”

We need not be concerned about Google breaching patient privacy or misusing the data, says Ross Anderson of the University of Cambridge, who taught DeepMind co-founder Demis Hassabis as an undergraduate. In fact, Google has a good track record of keeping data secure and private. “If learning about adverse health outcomes helps them to predict patients at risk, then this is a perfectly reasonable way to make money,” he says.

For Anderson, the more important question is whether Google – already one of the world’s most powerful companies – should have so much control over health analytics. “If Google gets a monopoly on providing some kind of service to the NHS it will burn the NHS,” says Anderson.

Smith has a parallel concern about the knowledge DeepMind is getting via this agreement. “This is private knowledge,” he says. “It’s only going to be available through Google or the Royal Free.”