Into the deep: proper utilisation of patient data10 April 2018
From automatically analysing medical scans to using digital records to predict illness, new technology is transforming how doctors care for their patients. But the rise of digital patient data comes with its own set of challenges, with clinicians and IT firms needing to work hard to keep data safe. Andrea Valentino catches up with Dr Dominic King, a clinical lead at DeepMind, about how his team gets round these problems, and how innovative use of patient data can improve life for doctors and patients alike.
Over the past decade, medical institutions have rushed to digitalise patient data. Between 2014 and 2015 alone, the number of hospitals worldwide that ‘went digital’ jumped from 54 to 65%. This trend is set to continue: in the five years to 2020, the digital health industry is expected to climb by 21% a year. The financial consequences are spectacular, too. The global electronic medical record (EMR) market is likely to reach $38 billion by 2025.
These changes are easy to understand. EMRs offer huge benefits, allowing doctors to track patient progress and compare them with their peers. EMRs are also far cheaper than the alternative, one US study finding that tech-savvy institutions save up to 9.3% on administrative costs compared with their paper-bound cousins. But the digital revolution is not without its difficulties. If splashing coffee over a cancer diagnosis is now more difficult, doctors have to work to keep patient data secure from leaks and other online mishaps.
For one London-based team, though, these challenges are worth the risk. Together with doctors and administrators from the NHS, they are working to transform how medical records are gathered and used. Apart from saving clinicians time, they have the chance to drastically improve patient care, saving lives along the way.
For 15 years until 2016, Dr Dominic King worked in the NHS as a doctor and clinical researcher. In that time, he did everything from lecturing on surgery at Imperial College to teaching art therapy. But as varied as his work was, outdated systems remained a constant irritation “Like many clinicians, I became frustrated with the state of technology in the health service,” he remembers. “We continued to use paper charts, fax machines and pagers.”
It would be bad enough if these drawbacks just wasted time and money. In the event, mismanaged data can have a serious impact on patient care. Though he stresses that NHS staff do an “incredible job” under pressure, King is aware of how problematic inaccessible records can be. “While most people get world-class care, nearly one in ten hospital in-patients still suffer some kind of avoidable harm. Many others fail to receive timely, evidence-based care.”
Faced with these difficulties, King thinks advanced technologies could play a huge role in addressing the challenges of poor data. By “helping clinicians analyse information much more efficiently”, thoughtful use of digital records could “ultimately help patients”, he believes. This optimism is easy to understand. After all, King himself has already encountered the power of digital data firsthand, writing papers on medical technology and co-founding Hark, a clinical task management platform.
Given these experiences, it is unsurprising King joined DeepMind, a London-based medical artificial intelligence (AI) company. Many of his colleagues had also worked in the NHS, making the two organisations natural partners. “We started working with the NHS two years ago,” he explains. “We are a team of people who grew up with and worked in [the service]. We really wanted to apply our expertise to help patients, nurses, doctors and the health service we believe in.”
Now a clinical lead at DeepMind, King has fronted several campaigns aimed at improving NHS use of patient data. One of the most intriguing is his work with the Moorfields Eye Hospital, in London. Though the hospital sees 600,000 patients a year and carries out more than 3,000 retinal scans a week, the results need to be interpreted by highly trained doctors. This can cause serious delays in treating patients, especially given how busy NHS staff are already. An aging population makes the problem even worse, with age-related illnesses such as glaucoma causing blindness in increasing numbers of people.
King and his team got round these difficulties by developing AI that can study thousands of retinal scans, gathered from patient data provided by Moorfields, and pick out signs of disease. Because the images provide such rich data – each one contains millions of pixels of information – they can catch patients with three of the most serious eye conditions: diabetic retinopathy, macular degeneration and glaucoma.
Even better, the whole system is automated. Rather than struggling through thousands of scans by hand or hunting down relevant images from the archives, doctors can leave the AI to do the legwork and focus on getting their patients better. Though the project is still at an early stage, with a report due out later this year, King is confident the scheme could soon transform patient prospects, and give clinicians a better and faster understanding of eye disease. “In turn, this could ensure the right patients are seen at the right time by the right clinician,” he says.
Given similar tests elsewhere, King is right to be positive. A team at the University of Nottingham recently found that AI could diagnose people at risk of heart attacks 4% better than doctors. Another survey noted that over 4,000 lung cancer patients a year could be diagnosed more quickly by training AI to scrutinise lungs cells, greatly improving their prospects.
Stream of information
DeepMind is also working with the NHS to develop Streams, a clinical app that uses medical records to get patient information to doctors. Like his AI work at Moorfields, Streams is not being rushed out everywhere, King explains. “Currently, the app is only deployed at the Royal Free London Hospital, where it is being used by clinicians in the treatment of acute kidney injury (AKI). This is a condition that involves the sudden deterioration of kidney function. The disease affects one in five patients in hospitals and leads to 40,000 deaths a year in the UK.”
By funnelling relevant blood test results into one place, Streams gives clinicians ‘breaking news’ alerts when patients might be suffering from AKI. Once staff receive a message, they can use the app to check other important patient information, including blood pressure and heart rates. All this helps them make accurate diagnoses and offer suitable care.
Though the app is in still in development, it is already helping patient care, King says. “We hold regular feedback sessions with nurses, doctors and patients, as we’re always keen to hear how we can improve Streams to better meet their needs,” he explains. “The feedback we get in these sessions has been fantastic. Nurses using Streams report it is saving them up to two hours a day, and deteriorating patients are having their results reviewed by specialists in minutes, rather than hours.”
Verifiable data audit
DeepMind clearly has the chance to enhance patient care, but its path to success has not always been smooth. Last year, the Information Safety Commission rebuked the Royal Free Hospital for handing over 1.6 million medical records to the company without fully informing patients. For its part, DeepMind used the data to test its Streams app, rather than directly provide care.
But if the scandal embarrassed DeepMind, it followed a familiar pattern. In 2016, the NHS wrapped up its ‘Care. data’ project after sharing patient records with private companies without properly explaining the details. The high profile nature of the scandal – by no means the only NHS data slip-up – speaks to the scale of the problem. Even several years after EMRs were first introduced to hospitals, medical providers are struggling to adapt.
King is clearly conscious of these challenges, noting that his company has a “great responsibility to keep patient data safe. We use world-leading security arrangements across all of the NHS trusts we work with.” In practice, the company is careful to fully encrypt data, training staff in how to use it. So-called 'data custodians' work to prevent any breaches. “The identifiable patient data we process on behalf of [the NHS] is likewise fully encrypted,” King adds. “It is stored in high-security facilities, separated from other datasets. Only those who need to access the data can [do so], and data is deleted when it is no longer required.”
At the same time, DeepMind is working on a system called Verifiable Data Audit (VDA). Using a form of cryptography similar to blockchain, itself invented to keep cryptocurrencies like bitcoin secure, VDA keeps a note of which records are accessed when. King feels this is a key step towards making people comfortable with how their data is shared. “We believe that VDA offers a way to establish real confidence about how data is being used in practice, and could bolster society’s trust and confidence in the vast amounts of data powering our most important institutions,” he says.
Hard at work on fixing patient security, King is also excited about how innovative use of EPRs could further sharpen patient care in future. “While mobile technology and AI is still in their early days in health, we’re incredibly optimistic about what they can achieve,” he says. “We believe that technology has real potential to support doctors and nurses with more accurate analyses, and the delivery of faster care to the patients who need it the most.”
Not that King imagines data will ever displace doctors entirely. “We see it playing a supporting role, not replacing qualified clinicians,” he points out. “Decisions about treatment will rightly continue to rest with these nurses and doctors, as with existing healthcare processes and pathways involving technology.” This is no simple task. Balancing the power of data with other concerns has already proved difficult. But get the formula right and healthcare firms could tip medical life towards a new age or at least help doctors save on paper costs.