The origins of the BMJ’s ‘Gatekeeper goes digital’

Does the epistemology of digital information herald the end of general practice?

This paper eventually became ‘The Gatekeeper goes Digital’ published in the Christmas edition of the BMJ 2016.

General practitioners mediate the boundary between the expert and the lay. Over the last 30 years the information revolution has changed the nature of this boundary: patients are much better informed whilst the falling cost of information has led directly to the codification of much practice via evidence based medicine. Whilst these changes affect all branches of medicine the digital revolution poses particular questions about the foundations of general practice.

The epistemology of knowledge is profoundly affected by the medium which carries it. Books are written on paper and therefore have a beginning, middle and end. The age of the book necessarily involved leaving stuff out which reinforced the role of experts who curated and mediated knowledge for the rest of us. By contrast knowledge viewed via the web has no ‘end’. There is no paper’s edge, no limit to where your clicks can take you. In this process we have moved from ‘filter out’ to ‘filter forward’. Knowledge curated by experts into a defining canon has given way to a flatter knowledge landscape accessible to anyone [i]

Given that for every ‘fact’ there is now another equal and opposite ‘fact’ just a click away, this world can be very confusing and people still need help to navigate it. However it is not clear that such guides need to be doctors. Not only are doctors very expensive, the web is creating new forms of meta-knowledge that simply did not exist 20 years ago – for example knowing what ‘people like me’ chose when faced with similar choices. Who would you choose to guide you through a complex care pathway –  your GP? local consultant? Specialist nurse? or what people with the same problem attending the Mayo clinic decided to do? Probably all of the above, but whatever your choice, generalist claims to mediate between the known and the unknown now have to be argued against multiple legitimate contenders. Given the ease of finding information what is the justification for a profession that makes a fetish of not knowing the detailed stuff?

A world where all knowledge is just a click away also greatly extends the range of things that a generalist might reasonably be expected to know. ‘We are witnessing a steady increase in agents’ [i.e. GP’s] responsibilities. The more any bit of information is just a click away, the less we shall be forgiven for not checking it.’[ii] In a digital world common knowledge is becoming more common.

The challenge of having to know much more is one reason why we are seeing the rise of knowledge-handling intermediaries like NICE and care pathways. These help all clinicians but are especially important to generalists who otherwise have to live out their lives in the Badlands where both uncertainty and lawyers roam. But precisely to the extent that the regime of discipline exercised by NICE and its familiars gets bigger so the role of the generalist is diminished.

To sum all this up:  in a world of flat, low-friction, digital information, the value of being a professional generalist is falling whilst the risks are rising.

But what about referral? Surely GPs’ ability to know what to refer and – just as importantly – that sometimes the unknown is not worth knowing (‘I don’t know what it is, but it isn’t serious’) will always be valuable. At its heart skilled referral rest on the relationship between prevalence and predictive values and here the epistemology of digital knowledge may, in time, further weaken the GP role.

Being a good referrer depends on identifying symptom complexes that have a high predictive value of being associated with serious disease. Predictive value in turns depends on the prevalence of disease in the population you are used to dealing with. For GPs this population has historically been those turning up in surgery. The prevalence of disease in this population is significantly lower than the prevalence in the population presenting to hospital consultants.  Let those same patients you decided not to refer in the morning get to see a specialist in the afternoon and watch those false positives (and costs) rack up as the higher prevalence of serious disease in hospital populations increases the predictive value of the presenting symptoms[iii].  Since this is why healthcare systems with effective primary care gatekeepers are more cost effective than those without[iv] you might assume that GPs will always have a role.

But big data means that at least in theory the population from which the system derives predictive values can be matched to each individual. Suppose that you see a 33 year old man presenting with rectal bleeding of <2 weeks duration but no red flags. Thirty years ago management would have been based on clinical judgement derived from seeing many such symptom complexes in your low prevalence general practice population. Today your judgement might have been replaced by a local guideline derived from largely hospital-based research tempered by budgetary constraints. But what if the NHS routinely logged all symptom complexes against all outcomes for every one of the million people it sees each working day? What if in time, we added in people’s genomic predispositions? At this point we would know the prevalence (and hence predictive values) of symptom complexes within ever more closely defined populations. Now the conversation becomes ‘the chance of this pattern of rectal bleeding being associated with serious disease in people like you (i.e. all the white males aged 30-35 who presented with similar short duration rectal bleeding in the NHS in the last 2 years) is <X%. Here are some treatment plans. Let’s talk about what which ones suit you’

The story of general practice over the last 30 years is one where the predictive value of symptom complexes ceased to be physician-based (those presenting to general practitioners as opposed to those presenting to hospital clinicians) and came to be determined by the care pathway regime mandated in your health economy. Typically these place-based regimes wrongly assume that prevalence does not affect predictive value, that rectal bleeding presenting in a GP’s surgery in the morning ‘means the same’ (i.e. has the same predictive value) if the same patient with the same rectal bleeding presents to a consultant surgeon in the afternoon. This inability of place-based protocols to interpret clinical data in the light of accurate estimates of prevalence may in turn be one explanation for why the productivity of health systems worldwide has remained stubbornly low.

Over the next 10 years it should be possible to free ourselves from the tyranny of ‘non-prevalence-based’ pathways by constructing predictive values based on the prevalence of the patient’s symptom complex in populations that ever more closely match ‘people like you’ – in this case all white 33 year old men with rectal bleeding of <2 weeks presenting in England over the last 2 years.

Seventy years ago general practice began trying to systematically improve diagnosis by coding encounters in disease registers[v]. Thirty years ago the falling cost of collating research allowed us to move to evidence based care protocols. Today the epistemology of web-based information mean GPs add less and less value as either diagnosticians or guides to treatment. Tomorrow ever-more finely grained predictive values are likely to inform and in time replace all expert-based referral decisions. Along the way general practice will have completed a historic journey: starting from the heuristics of clinical judgement, moving through the one-size fits all tyranny of evidence-based (but not prevalence-based) pathways and finally arriving at a personalised diagnostic predictions mediated by the complex information systems and membranes of the emerging digital world. At which point the generalist medical practitioner may – at most – be a helpful discussant.

The changing epistemology of medical knowledge does not render all aspects of a generalist role redundant – there’s always that ‘Let’s discuss what suits you’ to execute skilfully – but exploring these present and future ontological threats is important since they are how disruptive change midwifes the real revolutions in medical practice. Along the way they cast new light on why so many practitioners feel their job is changing in ways that go beyond financial pressures or the constraints of whatever national or local regime of healthcare they may practice under.


Thanks to Kate Billingham, Margaret McCartney, Amar Rughani and Val Iles for comments and suggestions to this and subsequent drafts.

[i] Weinberger D ‘Too Big to Know. Rethinking knowledge now facts aren’t facts, experts are everywhere and the smartest person in the room is the room’ Basic Books, New York. 2012

[ii] The Fourth Revolution. How the infosphere is reshaping human reality. Floridi L. Oxford University Press 2014

[iii] The Gatekeeper and the Wizard: a fairly tale. BMJ  1989; 298: 172-3

[iv] Starfield B Primary care and health. A cross-national comparison. JAMA 1991; 266:2268-9

[v] Hodgkin K Towards Earlier Diagnosis, Livingstone, 1962

Scale is a bug not a feature for online health communities

This blog is derived from research on online health communities carried out with @benmetz. Full results can be seen at here.

‘When I finally discovered an online forum (for my condition) I cried every night for 3 weeks. Finally, here were people like me, who knew, really knew, what I was feeling’.

Brands would die for emotional engagement like this so why are there no really big, disruptive online sites in healthcare? Why is there no Facebook for health? In recent research interviewing a wide range of people in the online digital health space my colleague Ben Metz and I frequently came up against this question

Of course we do have big health websites like NHS Choices but that’s essentially a site that gives out information. And we have Patients UK and the forums run by Health Unlocked or the communities led by Patients Like Me. Who have all done great work. But let’s be honest there has not yet been a really disruptive platform to disturb the sleeping giants of healthcare anywhere in the world.

Health is different because for users of online sites scale is a bug not a feature. To understand this imagine the online behaviour of someone with a newly diagnosed serious condition. After the initial panicky search for information people quickly gravitate to sites where they can find others with the same condition. Here they find support that they treasure, support that is qualitatively different to that offered by clinicians. Of course the information can be of variable quality but people understand that. What matters is the sense of not being alone.

But as these forums grow in size the costs of contributing to them rises: different people hold different views, sometimes passionately, and the online world can be as cliquey as the real one. Sometimes people want more specialized groups such as a forum for people with both rheumatoid and renal failure. Lots of people lurk gaining benefit but contributing nothing. Key members move on, knowledge and expertise is lost. This gives rise to what we have called ‘clustered scaling’:

clustered scaling

Crucially the network dynamics in each bubble vary. For example the tone, trajectory, governance and technology needs of a palliative care site will be different to one concerned with Down’s syndrome. This in turn means that it is hard to scale, hard to build a generic platform that suits many different groups. This is much more than a technical problem – many patient groups use Facebook but the governance and privacy issues associated with Facebook mean that this too does not scale. In addition Facebook’s advertising based business model is itself limited by the sensitive and often private nature of the groups that do use it meaning that Facebook lacks strong commercial incentive to build a health brand.

Many digital start ups are tying to overcome these difficulties  but their business models presume that they will scale quickly because  the default position across the digital world is to aim for scale, with the apex predator in this space being the monopolistic unicorn organisation that just moments ago was a tech startup. Driven by the fear of someone else capturing the network effects that drive monopolistic growth they scarcely pause to ask whether scale in this field really works. In fact success in the world of online healthcare appears to require the opposite: a myriad of small entities who differentiate themselves through their granular expertise or focus and where a winner takes all outcome would mean that everyone loses.

To sum all this up: scale may be a feature for entrepreneurs, investors and patient organisations but it is a bug for patients, citizens and all those who want engagement and intimacy.

This largely unacknowledged tension is explored further in the Finance and Business Models Constrain section of our research


More about patient-led commissioning of ‘middle range’ conditions

This post gives more detail about how patient organisations could act as proxy commissioners of care pathways for their disease. It is linked to the BMJ blog ‘Is British  medicine ever corrupt?’ which briefly outlined how such a system could also help prevent abuses like Mid Staffs and Morecambe Bay. Both this post  and the BMJ blog attempt to increase patient agency as a systematic route to better care.

Patient-led commissioning of ‘middle range’ LTCs

Overcoming the current misaligned incentives preventing integrated car

This paper explores what could be done to really integrate care using those ‘middle range’ long-term conditions (LTCs) like multiple sclerosis (MS), Parkinson’s disease (PD), inflammatory bowel disease (IBD), rheumatoid arthritis (RA) and heamoglobinopathies. These are relatively unusual at the level of any single practice, fall outside practice incentives schemes, and tend to be the preserve of hospital ambulatory care.


A Clinical Commissioning Group (CCG) with 100,000 people will have around 1,000 people with RA, 400 with IBD, 300 with PD and about 120 with MS. In principle the CCG knows who these people are and how to contact them – i.e. they could construct accurate denominators for these conditions

These numbers are big enough to plan services around. For IBD for instance a CCG could employ a nurse to review them every 6 months, a GPwSI to oversee the service, and on-call gastroenterologists to discuss difficult cases with (not necessarily one from the local trust). This rarely happens because CCGs are already stretched and even if they had the capacity such a plan might easily result in redundancies at the local gastroenterology department and perhaps encouraging local consultants encouraging their patients to campaign against what the trust may perceives as a lower quality service proposed by the CCG.


These knotted incentives and accountabilities occur very frequently when trying to deliver integrated, flexible care. The only people with the motivation and moral authority to challenge all the interlocking providers to do better are patients themselves – but how can we do this without more organisational turmoil?

This paper explores how the patient organisations associated with these middle range conditions might act as proxy commissioners for their disease. This could look something like this:

  • The patient organisation would build generic, national pathways in conjunction with public health, pharma, NICE and others.
  • NHS England, Monitor and others would define tariff payments for both primary care and secondary care parts of these pathways. Tariff would be less than the estimated total current spend on these diseases in order to drive efficiency and encourage creativity. This end-to-end tariff would define the outer cost envelop within which everyone – including the patient organisation – had to agree care pathways.
  • Each participating CCG would give over an agreed % of care pathway tariff to the national patient organisation. Like a Visa fee this would be around 1-2% but would fund the patient organisation for its work.
  • To keep transaction costs low for both patient organisations and CCGs, CCGs would be encouraged to take responsibility for working with one or two conditions and patient groups. Their commissioning for other middle range conditions would be led by other pairs of CCGs and patient organisations. To choose an example at random a CCG might work with Crohn’s and Colitis UK to be the lead commissioner for IBD across 35 CCGs with a combined patient population of say 12 million people. This would mean that the CCG and C&C UK would be responsible for commissioning the care of 48,000 with IBD. This would allow the CCG and C&C UK to build up specialist knowledge and relationships combined with very considerable purchasing power.
  • Since the CCGs know the address of every single patient with IBD it becomes possible to have real involvement and accountability by mailing them or texting them about proposed changes to their care pathway. This would place the proposers of change in a powerful political position that was based on consultation and dialogue.
  • Each lead CCG would collect data about costs for each patient as now but with the addition of condition-specific primary care costs as agreed in the pathway and tariff.
  • Patients on integrated care pathways would be routinely encouraged to share their experience on platforms like NHS Choices or Patient Opinion. This would generate an on-going supply of service-specific feedback that ensure local staff keep the system tuned to how patients experience the results of their commissioning.
  • The patient organisation would run annual awards for those CCGs, care pathway and specialist units across the 35 health economies that have made the most changes and improvements as reported by their local members.

This system could provide a flexible, self-funding system of care that could be piloted and rolled out relatively quickly. It would also provide a measure of competition between different health economies and between lead commissioners.



This system is less easily subject to provider or commissioner capture but is still likely to get locked into historical patterns of expenditure. However the patient organisations also have the greatest incentive to ensure this does not happen.

So a further development would be:

  • Over time each practice in each CCG would try to get 100% of its patients to consent to being part of its database and to release their information on an anonymised basis. Personal data would continue to be held at practice or CCG level only. In some ways this would replicate what aims to do but with more local involvement and an explicit, opt-in drive to take part in clinical research related to their disease.
  • The patient organisations, via CCGs, would be able to access patients who had consented to take part in research and would work with academics and pharma to co-develop patient-led research. Since these databases would be large, accurate and contain very high numbers of people who had consented to participate in research they might be a source of income generation for patient organisations whilst increasing the UK’s clinical research capability.
  • Patient organisations would use income generated in this way to:
    • Run and evaluate local variants of the approved integrated pathway. These would deliver NICE approved guidance but in a variety of different ways – e.g. telehealth, specialist nurses employed by CCG vs. FTs vs. the patient organisaiton themselves – to test which was most effective.
    • Promote the successful variants to patients on the database so that their local groups could pressure both commissioners and providers to ensure that they are getting the best care.


There are many questions about the feasibility of this kind of patient-led commissioning but in principle such a system could generate a range of activities and benefits:

  • Provide the moral authority combined with a self-funding national and local infrastructure capable of lifting both CCGs and providers beyond self interest
  • Create self funded, patient-led pilots that endlessly tested best value/best satisfaction integrated care pathways
  • Ensure that successful local variants of particular integrated care were quickly rolled out into national progammes
  • Provide legitimacy for local changes (and any ‘cuts’) by being patient-led at every step of the way
  • Provide new sources of income for patient organisations
  • Relieve pressure on CCGs by outsourcing the commissioning of these ‘middle range’ conditions to patient groups but in ways that ensured that patient groups knew the total envelope of funds available to them in each health community.
  • Strengthen UK’s research capability at no cost
  • Provide a systematic way to improve the primary care treatment of middle range conditions in primary care at no cost.
  • Create a cadre of skilled nurses and GPwSI for these conditions across primary care.




The Four Staplers of the Apocalypse

whole human wrapped

This excerpt on the ‘interpretative labour’ that health systems force patients and carers to do is taken from Chris Hatton’s wonderful blog which in turn is summary of David Graeber’s excellent new book The Utopia of Rules: on technology, stupidity and the secret joys of bureaucracy

“One of my nieces is a teenager with learning disabilities with a line in quietly acerbic asides to accompany her general awesomeness. My sister spends an inordinate amount of time and intellectual/emotional energy engaged in exactly this interpretive labour with people in bureaucracies – how shall I present myself, my daughter and my family? Competent, with it, but not so much that people will say no support is needed? Wanting the best, but not pushy, demanding or unrealistic? Middle class enough to do business with, but not superior? Knowledgeable, but with due deference to professional expertise – and definitely not one of those internet mums? And of course, endlessly patient, as this interpretive labour only goes in one direction. Get to an appointment late as a parent and that can have serious repercussions; but if the bureaucracy forgets the appointment altogether you’ve got to suck it up, not complain, and try and reschedule.”


Testing Times

DNA testing

So we can get our DNA tested with no questions asked. Boots will do it for you £129. All sorts of tests are available at a price. if you want to know if your spouse, teenage son employees are taking drugs send off a hair sample and screen them to see what substances they’ve been consuming over the last 90 days.

Self-evidently the technology is running way ahead of the ethics here.

But ethics can get messy can’t it, what with all that ‘on the one hand this, on the other that’ stuff. So let’s cut to the science because right there, at the heart of testing, there is a show stopper that should make us all pause before reaching for our credit cards to order that online test.

Here is an example from Ebola. Suppose that there is a test for the infection that takes seconds to perform and gives an answer that is correct 97% of the time. And say you have 10,000 people per day from arriving in the UK from West Africa of whom 1% are infected with Ebola. Does it make sense to start testing these 10,000 people at the port of entry?

At 1% prevalence you will have 100 true positives out of that 10,000 who really do have the disease. Since your test is 97% accurate at detecting Ebola infections it will detect 97 of these people. There will be 3 false negative results meaning that 3 infected people get through by mistake but you’re still well ahead so what’s not to like? Well look at the false positive rate – that is people who in reality don’t have Ebola but who still test positive. Of the 9,900 people who don’t have the disease a random 3% are going to test positive in error – that’s 297 false positives compared to the 97 true positives. So for every one person in isolation who needs to be there you have to lock up 3 who don’t for more testing. Result: isolation facilities overwhelmed, 297 more people each day ending up scared and angry and your test rapidly getting a completely undeserved reputation for being inaccurate.

These figures are all made up but the point is that interpreting test results is counter intuitive. ‘Common sense’ may tell us that it is better to be safe than sorry, that any test is better than none but whilst such sentiments may be common they do not make sense. All tests, indeed all clinical information, has an error rate. Radiologists sometimes disagree about MRI scans, histologists disagree about histology grading and pathologists know that many of their tests have significant error rates. In technical terms the sensitivity and specificity of much clinical information is surprisingly low.

So what does this mean for an age when people can finally side-step all those pesky doctors and get their mitts on any test they want because ‘prevention is better than cure’. Or treat themselves to a whole body MRI scan (current London price ~£1,500) ‘just to be on the safe side’?

First off it is unlikely that the powers that be will be able to regulate themselves out of this problem. The nature of the web means that if people want stuff – information, copywrited material, drugs or self-testing kits – there will always be a Silk Road that will purvey it to them. And usually a pretty broad one.

Secondly the relationship between sensitivity, specificity and prevalence (which is the science at the root of all this) is hard for doctors to grasp let alone patients. As a result almost all clinicians tend to place undue certainty on the tests and information they have to hand. As for patients they often end up mistakenly thinking they can’t have their chosen test for financial reasons rather than because it will not yield any useful information.

What would help is to ensure that anyone ordering a test or a scan in the UK always gets the result via an app or an interface that has all this complex information already built into it: ‘Here is the result of your blood test. For someone of your age and sex this result means there is an X% chance that there may be something seriously wrong’. Or when Big Data really gets into its stride: ‘Your test is positive and you should go and see your doctor. But to put your results in contest it is worth comparing it against the last 10 results from people like you (in your case 45 year old white men living in postcodes similar to yours). Out of these 10 men like you who tested positive, 7 turned out to have nothing wrong with them.’ (See what I mean about it being counter-intuitive?). This kind of thing would actually be really helpful for clinicians and patients in normal clinical practice but for citizens venturing out into the Wild West of direct-to-consumer pathology testing it will be absolutely essential.

Many kinds of data that used to be expensive are now cheap or free. This shift is beginning to apply to our biology. From Fitbit to direct OTC testing and imaging, our biological self is being becoming transparent and measurable. Making sense of this will take good software as well as professional advice. But perhaps the most direct way that people will learn the pitfalls of over-eager testing will come from seeing friends and acquaintances exploring, regretting, celebrating, and scaring themselves witless accessing data they have never had before. Paternity test anyone?

Living through a revolution


In 1970 I was beginning medical school. The polished wooden lemechanical caluclator piccture theatres were steeply ramped, the benches deeply carved with graffiti from generations of medical students bored with looking down 30 feet into a distant dissection. Oil was $6.00 a barrel. The Beatles had yet to split up. But in retrospect the thing that brings all that change most sharply into focus was the way we learnt medical statistics. At the start of each statistics lecture we were given mechanical calculating machines. To multiply 1254 by 64 you set the input digits by hand on the counter, then turned the handle 64 times.

Progress was slow. Each instruction, each turn of the handle took a second. By contrast the laptop that I use can execute hundreds of million instructions per second. So in my professional life time I have seen the computing power that I can literally touch, the power available to me personally, increase by at least ten orders of magnitude.

21st Century Healthcare is about understanding why we have got the health system we have and what we can do to make it better. Why has our health system become so risk averse? Why do so many health professionals spend so little time face to face with patients? How is it that so many professionals feel so lost in modern health care? Why do patients feel so neglected by doctors and nurses, attended to but not cared for? What in short does it mean to ‘care’ in this intensely technical, bureaucratic, protocol driven system that we have constructed?

The answers to these questions all have roots in that tremendous increase in computing power but 21st Century Healthcare is not much about technology. It is about people, sick and well, what they need when sickness strikes and how we can keep touch with the heart whilst controlling the huge advantages that technology gives us.