Patient records are the lifeblood of the health service. Joshua Chambers examines the analytics techniques that can improve public health and the operation of the NHS – and the sensitivities around privacy and data protection.
It is the most sensitive information imaginable, of the utmost importance to its subjects and owners. Patients’ medical records contain every medical confidence, diagnosis, treatment and personal identifier known to the NHS. And that data must be kept safe, or the National Health Service will lose the trust of those it serves.
Equally, though, this same data can be used to improve the care that patients receive. When medical records information is collected, aggregated and analysed, it can provide important evidence for the value of a treatment – or for its failure – and enable patients to examine the successes and weaknesses of particular health providers.
This tension between privacy and potential is one with which the NHS must grapple as it seeks to apply complex analytical techniques to its data. CSW has examined current practice and spoken to experts in the field, picking out the transferable lessons for civil servants in other departments seeking to make the most of their datasets.
What’s happening here?
Tony Blair’s government first moved to increase transparency in NHS performance data – buying a stake in health and care information provider Dr Foster in 2006 – but it was Gordon Brown’s administration that gave things a hard push by publishing hospital performance data. The coalition introduced a drive to share information across health and care services, promising an “information revolution” in its 2010 white paper Liberating the NHS. This set out ideas to ensure that GP surgeries and hospitals share medical records in an electronic format; and the department’s 2012 information strategy fleshed out these plans, proposing a central database of patient records to be used by all parts of the health service.
That’s now begun in earnest, with the ‘care.data’ programme starting to link information about patient treatments with performance and outcome data. Patient records from GP practices are being sent to the Health and Social Care Information Centre (HSCIC), which has been charged with collating that data into useable datasets and making it available to businesses, charities, academic institutions and the general public.
The HSCIC became an executive agency in April this year, after the Health and Social Care Act handed it statutory duties that include “collecting, analysing and presenting national health and social care data” and “creating a register of all the information that we collect and produce, and publishing that information in a range of different formats so that it will be useful to as many people as possible while safeguarding the personal confidential data of individuals”. It must also create a library of all indicators used to measure care quality; publish a set of rules governing how patient records are used; and help health and care organisations to improve the quality of the information they collect.
Prescribing patient records
There’s now a “widespread acceptance” that publishing data is “beneficial to the system as a whole,” explains Roger Taylor, head of research at Dr Foster.
First, there’s the benefits to medical researchers of using patient data. “The scope of research in this country can broaden and deepen,” explains Professor Martin Severs, a practising geriatrician, university professor and chairman of the Information Standards Board – which approves all standards for health and social care data in England. The cost of medical research can also be greatly reduced by making this information freely available, he says.
It’s particularly useful for “epidemiological research”: research that links medical data to other information to provide new insights. For example, a recent study linked population data and medical records to show that incidents of kidney cancer are increasing. Other pieces of recent research have examined a link between weather conditions and admissions to hospital.
Patients’ genetic data can be used to create pioneering new treatments, explains Professor Peter Donnelly, director of the Wellcome Trust Centre for Human Genetics at the University of Oxford. In December last year, the Department of Health launched a world-first project to fund the ‘genomic sequencing’ of 100,000 patients, scanning them to understand their exact genetic makeup. “If you get genetic information from several people who are suffering from the same condition, and if you link their health information to their genetic information, you can see patterns that would help us understand why some people are more likely to develop a disease,” he says. He adds that, if those patterns reveal that a specific DNA characteristic is linked to an increased likelihood of getting a disease, that “gives us a whole new insight into the disease process, and we might understand a key part of the biology of getting sick which we had no idea about previously.”
This kind of data has recently been used to reduce heart disease. Donnelly explains that researchers looked at data on individuals who had very high cholesterol, and were able to spot a genetic cause. Once they discovered the specific gene responsible, they were able to choose drugs that would reduce cholesterol levels and ensure that it’s absorbed and recycled into waste products, rather than floating in the blood. “It had an amazing impact across, I would guess, hundreds of thousands of people in the developed and developing world,” he says. Currently, the same research methods are being used to examine different types of breast cancer, and to develop drugs that will tackle them.
Patient records can, when combined with other data, be used to improve the operation of health services. “There’s a very long history of this that goes back almost a century, when doctors in the UK noticed that in some parts of the country, only a tiny number of kids were getting tonsillectomies,” explains Dr Ben Goldacre, an academic at the London School of Hygiene and Tropical Medicine and the best-selling author of the book Bad Science. “Just over the county line, half of all kids were getting them. There was a clear variation that couldn’t reflect variation in the clinical need, and must reflect clinicians’ treatment decisions.”
“Fast forward a century, we can see the same [variation] in prescriptions,” he points out. With colleagues at the website prescribinganalytics.com, he was able to map the prescription patterns of a particular drug type – statins – for which prices vary widely. “We mapped, across the whole country, the areas in which people were prescribing unusually large numbers of the more expensive statins,” he says. “There’s no reason to believe there’s a variation in clinical need [that caused] that prescribing pattern. It’s likely that some, but not all, of the expensive statin prescribing will reflect irrational prescribing of expensive drugs.” These models can therefore be used to understand and improve commissioning in the NHS, saving money for more effective use elsewhere.
Patient outcomes data can also be analysed, explains Roger Taylor of Dr Foster, providing an understanding of “the degree of variation in the quality and effectiveness of care”. Before this data was published, he says, discussions on healthcare focused on “how much does it cost and what is the waiting list?” But now people also know how effective health services are: “We have a much richer public understanding of the strengths and weaknesses of our health system.”
How this works
A number of different techniques can be applied to get the most out of this data, Taylor says. “A lot of it is very simply raising awareness about the volume of different types of activity.” For example, he cites work to measure the level of alcohol addiction displayed in hospital in-patients.
The second type of information is “about trying to measure aspects of care, and that can be relatively simple metrics,” he says. For example, researchers could examine the number of procedures performed as a ‘day-case’ – often using keyhole techniques and lower levels of anasthetics – rather than an operation requiring an overnight stay. “Obviously there are a huge number of benefits when it’s a day case,” Taylor says: for a start, “it’s less disruptive to a patient’s life”; and it also places less of a drain on hospital capacity.
The more complicated work involves assessing the outcomes of public service delivery, he says. “Did the healthcare we provided make that patient better or not?” It’s more complex, he explains, because “there’s a risk of adjusting” the data erroneously: “Some patients are sicker than others, so that’s where the most complex statistical metrics come in.”
Dr Foster is partnering with Imperial College London to try to better understand outcomes, and it’s tricky because “there is no definition of what ‘good’ looks like,” Taylor says. Researchers have to pull together vast amounts of data to understand whether a particular service is relatively good or bad, taking into account a vast number of variations.
Risks of the work
Publishing this data is “politically very risky,” Taylor says, because the public sector loses “control over how you define what good and bad looks like in a public service.” He warns that “if you do try and control it, you will come up with inadequate definitions of good. They have to be both legitimate and sufficiently accurate to be useful.”
Accuracy can only be possible if the data provided is of a sufficient quality. Professor Severs says that it must conform to standards that truly reflect the quality of care, and therefore must be extracted directly from the patient record rather than being transcribed by humans – minimising the risk of errors. He was part of the Caldicott Review, an independent inquiry into information management in the health service, which examined whether patient data should be released.
The review concluded that it should, but set clear criteria around data releases to ensure the security of individual patient records. As Severs notes, “when you come and tell me as your doctor something really private, you give me that in confidence,” adding that “as a clinician, I am bound by the common-law duty of confidentiality, as well as by the Data Protection Act and the Human Rights Act.”
The key to making data-sharing work in the NHS, and allowing for complex analytical assessments, is ensuring that the data itself is anonymised. It should never be possible to identify a patient using data released by our health or care services.
The Department of Health therefore partnered with the Information Commissioner’s Office (ICO) to “provide a recipe for turning identifiable patient data into an anonymised, safe, releasable form,” explains Iain Bourne, ICO group manager for policy delivery. This work is now undertaken by the Health and Social Care Information Centre, extracting raw data securely and publishing it once it meets the agreed standard
Making sensitive data anonymous is difficult, Bourne says, because “you don’t know what other information is out there [that] it could be combined with to allow for identification... The test in data protection law is of actual identification or reasonable likelihood of identification, which is a fairly nebulous concept and difficult to assess in an empirical way.” There’s “a lot of judgement that goes into this,” he adds, and the growing amount of published data makes it more and more difficult to assess the risks. Nevertheless, the ICO has come up with a set of techniques that can strip out the personal information, while ensuring that the data retains its utility (see box for the specific techniques used).
The health department is a forerunner in Whitehall’s efforts to publish and utilise its data, and its experiences have produced a number of transferable lessons from which other departments can learn. The Department for Education, for example, also oversees highly sensitive records that mustn’t fall into the wrong hands.
Andrew Miller MP, chair of the Commons Science and Technology Committee, believes that independent oversight bodies such as the HSCIC will be needed in other departments “to properly handle the transfer of data across client groups, and my plea on that has always been: let’s create a structure that is focussed around the citizen’s needs, not around the state’s.” He adds that if other departments create oversight bodies that handle sensitive data, “you’ll get a much better buy-in from members of the public who are always anxious about how their data is being handled or mishandled.”
Miller’s committee recently examined the use of clinical trials data, and he says the Department of Health applies “a very good principle that ought to apply in other government departments.” That principle is that in order to open a clinical trials dataset, doctors must have to use NHS access cards with secure pin systems. “Even if he does nothing with that dataset, the fact that he’s entered it is recorded,” Miller says. Therefore, “if somebody improperly enters a dataset, they’ve left a thumbprint that is recorded elsewhere and they can’t remove that. As data becomes more rich, citizens can be given confidence that nobody is entering their dataset.” The education, local government and work and pensions departments could all adopt this principle, along with HMRC, he believes.
Another lesson concerns the anonymity of data: Bourne says that the DH has a good system in place, but organisations handling sensitive data must regularly reassess the risk of identification. “Imagine that the electoral register was published openly on the internet, as opposed to be being accessible through libraries,” he says. “How would that affect other forms of data release?” Given the explosion in social media, he argues, organisations must be “checking out what information people are putting about themselves on the internet that might allow re-identification to take place.”
A final lesson is on how departments interact with organisations that can turn their data into cutting-edge analysis. These can be small start-ups or research teams, but Goldacre says public sector commissioning structures aren’t used to working with this kind of organisation – or, indeed, this kind of service. “I could easily get funding to produce one academic paper that had an atlas of variation of one drug’s prescription,” he says, “but there is no obvious way to get a similar amount to produce a website that can produce an infinitely large number of these maps. That feels strange.”
He also believes that current consultancy contracts penalise small, innovative teams and favour the big players: “It would be really positive if Public Health England looked at how they could take some of the money that they spend on analysis of service provision and local needs through conventional channels, and see if that could be opened up to open competition.”
Taylor agrees, adding that the UK must foster its analytics industry. “One of the biggest weaknesses in the transparency agenda is the degree to which we [in the UK] have specific institutions able to make sense of data and the skills required,” he says: there aren’t enough organisations with the skills and desire to improve public services by using this data, and he wants to see more ways in which such organisations can receive funding for their work.
Cash-strapped civil servants might not immediately think they can help here, but perhaps there is a way that money splurged on traditional consultancy contracts could be redirected to ensure that those using analytics techniques are able to help public services find new ways to save money.
In the Department of Health, big steps have been made on this agenda, and there are clearly transferable lessons for other departments. The potential for using analytical techniques in the health sector is enormous, and demonstrates how valuable public sector data can be. While patient data is sensitive, it’s also incredibly useful. Indeed, it’s the lifeblood of the NHS, and the circulation of it through the health and care system is vital to ensure that the patient’s condition continues to improve.
Correction: This article originally said that Dr Ben Goldacre is an academic at King's College London. This is incorrect, he is an academic at the London School of Hygiene & Tropical Medicine.