Analytics for healthcare practice, planning and insight

In April 2012, eHealth Insider and InterSystems convened a small round table event to discuss a very big theme: analytics for healthcare practice, planning and insight.

We wanted to explore how data is being used across the NHS and how analytics can support this. What do healthcare professionals want from analytics? Are their needs different to those of finance teams, commissioners and planners?

How can analytics support the new relationship between patients and professionals? What insights can analytics give us into the effectiveness, productivity and quality of health services today – and how these services might develop to meet future demands?

Around the table were:

Join the debate >> 

What information and for whom?

The first big question of the day was about the analytics that the NHS needs most. Is it for patient care or finances? Is it for audit and regulation? What is the interplay between operational data and the data needed for planning?

Martin Egan was clear. “Ultimately, my interest is in providing information to clinicians that helps them to provide better immediate care to the patient in front of them,” he said.

A great starting point, to which Dr Mark Davies added: “Information has the ability to change conversations and change the narrative between citizens and services. We have a great opportunity to radically change the relationship using data.”

But, he argued, information has wider uses. There are strategic needs to consider as well as the use of information in commissioning and delivery. Most of all, he made the case for information to be actionable.

“We need stuff you can do something with,” he said, adding that such usefulness applies on several levels: the clinician making a decision about the patient in front of them; the planner looking at citywide services; and at a national level.

Dr Dan O’Donnell came at the subject from the operational perspective. A one-time neuropsychiatrist and a US Marine Corps veteran, he came into analytics 14 years ago after observing how medicine lacks the operational data needed to manage a complex system.

This was not about computers, he said, but about an approach to using information. “Medicine has never really managed information very well. “I have always seen a mismatch between the risks in what we do, and the ever increasing complexity of the knowledge we have to juggle, and the information we apply to the task.

“We are now at a point where we have advances in technology coupled with a recognition of the need to use information better. It is an exciting time,” he said. “Maybe we will see fewer people die from a lack of data than when I was at medical school.”

But he warned that this is a complex undertaking. “I don’t think anyone is going to come out with one thing that drops into a big computer and spits out the information we need, when we need it. Organisations need a new skills set.”

One of those skills sets must be communication, said Linda Davidson. She talked about a “light bulb moment” when she listened to Daniel Ray, director of IT at University Hospitals Birmingham, talking about mining cardio thoracic surgery data to identify patients with poor outcomes, and how his trust was working to address that. The same trust has integrated data into clinical practice, contributing to a significant fall in mortality.

“Until then, analytics had always seemed to me to be a rather dry and academic area,” she said. “We need to get across the message that this is not about variance from the mean and statistical analysis. It is about critical things that can save people’s lives.”

This was an important point, agreed Dr O’Donnell. “What are you analysing, for what purpose and for whom? Are the users statistically knowledgeable? Can they use dashboards or manipulate the data themselves? All these questions need to be answered.”

Dr Davies agreed. “We need to think ‘what are the real priorities of the NHS?’ What are the really big ticket items?

“That we do not kill people, that we have a system in place where we feel confident that there will not be another Mid Staffordshire, that we have visibility of financial control [so we do not go] bust.”

Debating points: who is driving demand for analytics in your organisation? Is it clinicians, planners, commissioner or finance? What are the competing needs? And how do you prioritise them?

Actionable analytics

Dr Davies went on to talk about “actionable data” or “actionable analytics”: the stuff we can do something with.

He argued that this implied two things: combining data from different source systems to provide novel insights and different levels of granularity for different users.

He said: “We make a constraint for ourselves if we only think about data coming from operational systems. We need also to look at patient generated data, such as patient experience, complaints and comments. Increasingly, this will need to become an important part of the tapestry we paint.”

But for historical reasons, data often sits in silos and has not been integrated. The data used for clinical audit, registers, for Hospital Episode Statistics and financial flows, and for regulators needs to be brought together.

Egan pointed out that the situation is slightly different in Scotland, where there has been better integration. The NHS in Scotland is structured differently to England, with primary, acute and community providers all within one organisation.

“Some of the boundaries do not exist, although some of the problems are the same. The issue we have is one of ‘this is my data and you’re not going to see it’,” he said.

“We have started a Janet and John approach to data sharing and we are only at the beginning of that journey. There is a real appetite to do it.” NHS Tayside in particular is leading some innovative projects, he said.

Dr O’Donnell then raised the spectre of data quality. Clinicians will always criticise data, especially if it makes them look bad, he contended. This is not necessarily a function of data integrity.

However, there are real questions about the quality, structure and volume of information and therefore about the conclusions that can be drawn from analysing it. There are questions too about whether we measure the right things.

Dr O’Donnell cited the 28-day readmission rule as a good example. Hospitals can measure it easily enough, but there is very little they can do about it as so much depends on post-discharge care in the community.

“We are focussing on the wrong thing,” he said. “And that is a barrier to actionable data.”

Dr Davies also felt it made little sense to hold hospitals to account for breaching the 28-day readmission rule. Rather, he argued, this should fall to commissioners. He also said his experience of providing practice level data on readmission rates is creating some “interesting conversations.”

This brought the debate round to the notion of granularity. Data must be relevant to be actionable and that means getting the right level of detail.

“The closer you are to a patient, the more accurate the information has to be,” said Dr O’Donnell. “But for population studies, the information does not require the same level of precision.”

This was an important point, said Dr Davies. Returning to University Hospitals Birmingham, he said: “The data there largely comes out of the electronic prescribing system and that makes it extremely relevant to clinicians.”

In general practice, one third of income depends on the Quality and Outcomes Framework, giving GPs a high level of interest in the quality of data going into the system.

“A locum coming into my practice who cannot make the grade in terms of data quality is not going to come back,” he said. “We can’t afford it.”

Increasingly analytics is playing a role in handover – from night staff to day staff, or between consultants within a hospital or between services at discharge. And here a high level of granularity was required to improve safety.

But equally, measuring the cost of patient care requires a high level of granularity and data integration, suggested Dr O’Donnell. “If you are measuring the cost of care you have to be able to include factors such as staff patterns, components of kit and so on.”

Dr Davies agreed. “This has to be done at a level of granularity that enables you to make a link between quality and resources and this is something we keep getting wrong.”

Debating points: is this concept of actionable data useful? Can it inform discussions with different users of information? What are your data quality issues locally? Can you provide information at differential levels of granularity for different users?

Capturing patient information

Returning to the question of including patient information in the record and in analytics, Dr O’Donnell pointed out that much of this data is currently free text.

For example, there are thought to be 800,000 blogs supporting cancer patients. “There is an enormous amount of patient entered experience about the effects of cancer on the family and on themselves. We need to find ways to look at that free text.”

It is just free text but, increasingly, video material as well, added Mike Fuller. Capturing and analysing free text in a meaningful way will be important as a means of shaping future services, suggested Davidson, while Dr Davies suggested that patient feedback may become important in relicensing doctors.

However, Fuller, questioned the quality of some patient generated material. “I agree patients are a source of vital information,” he said. “But there is an interesting dilemma. Do we end up being overwhelmed by the mass of information and misled by a general acceptance of ‘bad science’?”

Davidson wondered whether this was a new horizon for analytics: devising useful tools to make better sense of this sort of data.

Dr Davies raised another changing dynamic in doctor-patient relationships: giving patients access to their records. “This is a radical development and will lead to patients being active participants in the narrative around their care,” he said. “That will significantly change the dynamic.”

This raised a challenge around the standard of records, and how to keep the rich, human narrative of disease and wellbeing in that standardisation. “When patients access their own records, they will become more acute,” he said.

The forthcoming Francis inquiry report into Mid Staffordshire Hospital, now expected in October, will heighten this debate, he predicted.

“I was involved in a couple of workshops hosted by the inquiry. Listening to some of the stories of patients let down by the system was quite moving; but it also gave a sense of ‘if only we had known we could have prevented these things happening.’ The report will be full of powerful messages about the importance of narrative.”

Egan explained how NHS Lothian is already exploring this new dynamic, with 60 patients going live with access to their records via a portal in May.

“Clinicians are already having panic attacks,” he said. “How will patients interpret it? Will clinicians have a right to see information first? It is only over time that we will realise what we have let ourselves in for. It cannot all be bad but there will be some bad things.”

There are ethical dilemmas and information governance issues to address around such access, the roundtable agreed. These include what information patients should see; how to support people who are unable for one reason or another to access their record via a portal; and answering questions such as whether patients should have a right to see who has accessed their data?

Dr O’Donnell brought the debate around to the question of ambiguous data. Clinicians are often ambiguous in their records, sometimes with good cause because reaching a diagnosis can be complex.

So, a radiology report saying that the results of a scan were inconclusive was an ambiguity of diagnosis.

However, the same report saying a new scan was needed in three months, but which failed to specify who should arrange such a scan, was dangerously ambiguous: was the hospital going to arrange it or should the GP refer the patient? The patient risked falling through this hole.

“We need to be really clear about where ambiguity is not acceptable and that’s an analytics challenge,” said Dr Davis.

Debating points: Are analysts being involved in the discussions around capturing patient information and opening access to records – or are they seen as technicians? Is analysing free text a new horizon or a step too far?

The challenges ahead

One emerging theme is that there is much untapped data in the electronic patient record. Dr O’Donnell discussed being able to tap into repositories of data derived from EPRs for research studies, benchmarking departments and hospitals, or modelling changes in services.

Could the EPR be combined with the powerful information held in databases on congenital abnormalities or drug interactions, for example?

This raises the question of what to do with ‘big data’. “We are reaching a position where we have enough data to compare populations,” he said. This would require the NHS to partner with universities and to address some information governance issues.

Some of this sort of work is starting now, particularly in the US, but Dr Davies argued that it is not well developed here.

He said: “Modelling is not well developed in the UK. We talk about measuring quality and capturing outcomes and that is absolutely right and developing the methodology to do so is something we need to do.”

The system needs aligning to measuring and rewarding outcomes rather than processes and activities, he added.

Egan talked about “joined-upness”: analysing data to find patterns of behaviour. “You start to see patterns where patients are pushed around the system so that they become the responsibility of whoever is in charge of them at 5pm on a Friday evening,” he said. “This insight comes from people joining up different datasets.”

Dr Davies argued that in England this would be a job for commissioners to undertake. But they could only do the job if there is a more equitable distribution of power between commissioners and providers.

“I would argue that those who are planning care and commissioning it need to be more powerful,” he said. “Most of the sophisticated analytic experts are sitting in providers. The commissioning world needs a maturity of analytics.”

This would become ever clearer as the challenges of an ageing population and rising numbers of people with long term conditions become ever more pressing, he added.

“The key part of commissioning will be providing sophisticated baseline intelligence about pathways and how we track and measure the impact of changes to them,” he said.

Egan said the dynamics were different in Scotland, where the purchaser provider split did not exist. But analytics still needed to answer the core question of whether providing services in the community would improve outcomes and reduce costs. And if the answer was yes – then the political problem of closing hospital services would came into play.

Dr Davies said this was another reason that it was crucial for doctors to take a lead in the analytics. It would be much more difficult for an MP to oppose evidence based decisions reached by GPs, he argued.

Egan added that if there was hard data showing that doctors who carry out procedures regularly have better outcomes than those who do them intermittently, this should support concentrating services geographically.

“Most of the evidence shows that improvements occur in health services not because of patient choice but because clinicians do not like to be at the bottom of the curve when they know it will be measured,” Dr Davies argued.

Fuller argued that the future of analytics lies in joined up data sources that allow analysts to track individual patient experiences and journeys. Scotland’s implementation of a national patient management system that isn’t all about what happens inside the hospital but what happens outside it too, has achieved a significant alignment coding, processes, and practices, he said.

This gives the NHS in Scotland a view across care settings of the patient’s needs and journey,and a valid basis for making comparisons between health board services levels.

“That is fantastic from an operational and analytics point of view,” said Dr Davies. “There is a massive potential for linking primary, community and secondary care data. We need the ability to link and understand the health economics of particular conditions and diagnoses.”

He predicted that the community will start to emerge as one of the most important data sources.

The final challenge raised by the round table was how to pay for analytics. “This is not free goods,” said Dr Davies. “Every £1 we spend on analytics is £1 not spent on direct healthcare.

“How do we prioritise analytics, what capacity do we need and how do we measure its contribution to the system? How do we pay for it when the benefits are long term but the costs are short term?”

He suggested that small, in-house teams of analysts at trust level were not the most efficient and productive way of tackling the analytics challenges. “We need to work in partnership with others.”

There was also general agreement around the table that the industry needed to increase its level of analyticsexpertise.

Fuller claimed: “It is too easy to think that buying software licenses, defining (often narrow) KPIs,publishing reports, or providing portals for easy access is itself the destination.

“Underneath of all that pretty presentation we need a firm foundation, with informatics platforms that leverage the widest utility of the information without losing its context and provenance; otherwise we are just building castles on sand.”

Fuller concluded: “To afford this smarter way of working we need new health economies powered by embedded analytics. So that every transaction at the point of care contributes both to patient well-being and to the overall health and well-being of the care community, organisation, and nation.”

Debating points: do you agree with these challenges? Or are there others? Does analytics need to move from being a small scale, in house activity at trust level to something more sophisticated done in partnership across health communities and with industry? Or should analytics develop around use cases?