Free during COVID-19 pandemic

Is your healthcare record too complex? ‘The Tyranny of Structured Data’

Posted 23rd December 2020

A friend of mine visited his specialist the other day for a follow-up visit:

“What pills are you taking?” said the doctor

“You should know!”, my friend replied

“What makes you say that?”

“Because you prescribed them!” 

It’s common for doctors to ask questions to which they may already have an answer. It’s a form of ‘patient-triangulation’. Asking open questions can check you are speaking to the right person, test your patient’s understanding, ensure the quality of your own information and, if needed, clarify the patient’s current state of mind.

To stop the patient assuming I’m dim, I personally try to preface my questions in such a way that we both know why I’m asking. A couple of my pet preambles are:

“Your doctor has written me a letter, but I’d love to hear  about it in your own words …”

(Not “why are you here?”)

“It says here you’re on various medications, can you confirm our list is right?”

(Not “what pills are you on?”)

Now, maybe if this had been a first appointment with a consultant,  the doctor would probably have had a referral letter and he was just triangulating.

A good referral letter will contain all the elements of a complete medical history – in a standard order –  which means it can be scanned in a matter of seconds (the Drug History should come right after the Past Medical History …)

But given that this was a follow-up, I’m going to surmise that the doctor in this case didn’t have a whole load of time. So, my suspicion is that the doctor really didn’t know the patient’s list of medications because this information wasn’t readily available. The doctor was probably unsighted …

The Electronic Patient Record (EPR) is less easily scanned, less easily used, than a well-structured piece of paper written by a colleague. 

Why has this happened?

I’m going to blame ‘The Tyranny of Structured Data’ – for which read:

The point at which the organisational desire for detail exceeds the efficacy of patient care

or put it more bluntly:

When ‘management’ gets in the way 

We have all been subjected to this. How many hours have I spent on online referral systems – which were clearly designed by UX* sadists? I often don’t know if the trauma patient was a smoker. Ticking ‘Don’t know’ is surely a waste of time. Trying to line-up my mouse-pointer with a tick box only 5 pixels square adds unnecessary frustration to a fraught situation.

How many online forms and modal responses have we had to deal with, thrown up by EPR systems? How often has the computer insisted on assessing DVT risk, even when you know from the start that the tick-boxes will be “No, No, No” … no risk)?

The thirst for structured data can cause havoc to clinicians motivations. Atul Gawande has made this point brilliantly in an oft-cited article for the New Yorker.

But we must understand the causes here. Structured data serves several purposes:

  • Completion
  • Control
  • Compliance
  • Coding
  • Conduct

Completion is born of the need to ensure that the record does not contain any important holes. The problem is that most patient records are ‘holey’ by nature. And the holes are situational. Smoking history may be important in the context of heart attack or cancer. Less so in a trauma patient.

Control and Compliance ensure that everything that should be done, is done – and in the right order. Every woman of child-bearing age, with abdominal pain, presenting at an ER should have a pregnancy test. We can argue over the definitions – but failing to do so is negligent. Structured data can tell us whether this has been done.

Coding is largely about how we get paid. But it is also about “conduct” for which read “performance analysis” or – in old-fashioned terms – “audit” and retrospective analysis. Much of the motivation here can be related to personal rewards, some of which are monetary. We use structured data to assess the decision-making, outcomes of care and pathways. When aggregated to populations and cohorts, such data can often be useful, even when not relevant at the patient level.

But it’s clear that structured data entry and – as in our example – its presentation, can cause problems for clinicians. Human factors and user interface design are fundamental to the usability of systems, but they are not always given the priority that makes systems efficient and safe.

What’s more, structured data often poses a barrier to communication with the patient. You ca send a patient a copy of your referral letter, but not of their Electronic Patient Record. You can’t print it out. Often, clinicians have to interpret and then re-write the medical record to create discharge summaries. This is a chore – and a time-consuming one at that.

So, in many circumstances, structured data can get in the way of patient-care – in a way that ‘free text’ might not.  It is also clear that free text alone, especially on paper, can’t deliver the benefits of IT systems for safety, decision support and performance management.

The question is whether structured data and free text are mutually exclusive. Can we use free text to deliver structured data?

There are clearly some remarkable innovations happening here. The most impressive example that I’ve seen is a ‘prompt as you type’ AI engine, developed by a company in Beijing. It populates an otherwise free-text field with the evidence based suggestions and highlights areas that appear to run against good care – or where information is missing.

Imagine (if you’re old enough) the Microsoft Paperclip suddenly knows more medicine than you do.

The other possibility is using post-writing analysis of free text to create structured data after the fact. For instance, taking the phrase “Low DVT risk, stockings only” to create the structured fields that show the DVT assessment took place and an appropriate clinical decision was taken. The problem, as is easily apparent, is that the error rate needs to be low enough to compete with manual structured data entry.

This is a problem that only gets worse when records from different organisations are collected together. Organising the transmission of structured health data using HL7 or its close cousin FYRE can make grown adults weep. It’s just too complicated.

Currently there’s no solution. In our own work at, we have taken the strategic decision to stick with as much free-text as much as possible – but we know the pressures on both users and IT providers will weigh against that decision. The problems is that one person’s structured data is another person’s garbage – and we want to keep our data as clean as possible for as long as possible. 

The structured-data debate will rage. But who knows, perhaps one day we will have a ‘killer app’ that will make the debate disappear.

Until then, I would urge you, take pity on your doctor if they ask you a silly question. They’re probably not being dim but just dealing with your structured  healthcare record which is, most likely, far too complicated.