- Mar. 16, 2017, 8:07 AM
Wednesday, March 22, 2017
It Is Very Important To Make Sure Data Mining Patient Records Is Properly Managed.
Here is a saga that has just started to unwind and be revealed.
A data-sharing deal between Google DeepMind and the Royal Free London NHS Foundation Trust was riddled with "inexcusable" mistakes, according to an academic paper published on Thursday.
The "Google DeepMind and healthcare in an age of algorithms" paper — coauthored by Cambridge University's Julia Powles and The Economist's Hal Hodson — questions why DeepMind was given permission to process millions of NHS patient records so easily and without patient approval.
"There remain many ongoing issues and it was important to document how the deal was set up, how it played out in public, and to try to caution against another deal from happening in this way in the future," Powles told Business Insider in Berlin the day before the paper was published.
DeepMind and Royal Free say that the study "completely misrepresents the reality of how the NHS uses technology to process data" and that it contains "significant mistakes."
Powles and Hodson said the accusations of misrepresentation and factual inaccuracy were unsubstantiated, and invited the parties to respond on the record in an open forum.
DeepMind has used the data access it was given to create a mobile app called Streams, which was initially designed to help clinicians monitor patients with acute kidney injury (AKI).
Powles, a research associate in law and computer science at St John's College, Cambridge, and Hodson, a technology correspondent for The Economist, argue in the paper published in the Health and Technology journal that the terms of the initial deal (a subsequent one has been made) were "highly questionable" and that they lacked transparency.
DeepMind has tried to defend the deal by saying that it's providing something known in the healthcare industry as "direct care," which assumes that an identifiable individual has given implied consent for their information to be shared or uses that involve the prevention, investigation, or treatment of illness.
"The specific problems are they had access to the data of every patient in the hospital on the legal basis that they were providing direct care to every patient," said Powles. "We think that's problematic given that they only ever asserted that they were interested in helping patients with acute kidney injury. They've since pivoted to look at a bunch of conditions but they haven't addressed the gap in the initial deal between what the purpose was and why they had the access they did."
She added: "If they'd had a small sample set with appropriate consents that proved the results, that showed that this app was working, and then they engaged patients and said we're going to roll it out on these terms, for this period and with this amount of money passing hands, then it would be a totally different game."
Powles believes that the case should be considered a "cautionary tale" for the NHS and other public institutions that are look to work with innovative tech firms.
Vastly more is found here:
This is a hard one to me. Clearly the proponents of the program think they are doing valuable and useful work while the authors of the report have a range of concerns and issues.
It seems to me it is likely the friction is due to a failure of governance and planning before the work began and in that there is certainly a lesson. We are clearly going to have more discussions like this as AI and data mining come together - so we all need to be alert to the risk of problems and issues.
The article is well worth a read for the various nuances.
Posted by Dr David G More MB PhD at Wednesday, March 22, 2017