AI will provide ‘treatment’ that doesn’t solve problems it cannot understand, because the data for the chaotic patients never gets into the research that grounds AI’s decision making.
Throughout history, there have been attempts to corral medicine into an easily defined system. Hippocrates (we think) famously said:
“Life is short,
and craft long,
opportunity fleeting,
experimentations perilous,
and judgment difficult.”
How reassuring.
Since then, there have been a variety of classification systems, organisational frameworks and funding systems to clean medicine up and make it more, well, organised.
As a medical student in the 1980s, we were still looking at the dregs of scientific classifications and super-specialties, which, broadly speaking, promised that if everyone focused on a tiny part of the body, eventually every single patient would have a medical home to go to.
I sympathise.
The paleontologists had a similar problem, trying to organise a taxonomy that included all the dinosaurs in a neat little table. It didn’t work. Eventually, they decided everyone needed a junk drawer in their classification system, to shove all the dinosaurs that just didn’t fit.
We GPs see a lot of that junk drawer – the patients who are rejected or ejected from every service and have little relevant evidence they can or should follow.
It will surprise nobody to know there is a business framework for this, from Wales, called the Cynefin framework. It talks about the habitats we all have, which are populated by people with four groups of problems.
The known knowns
The easiest are simple problems – theknown knowns – the ones the policymakers focus on.
A person comes in with a symptom, a diagnosis is made, the health professional follows an evidence-based protocol and there is a measurable outcome.
Everyone loves these patients. For registrars, I call this the “board walks” in the chaotic swamp of practice.
They are a relief, frankly. We polish off our evidence-based spiel for whichever condition we need to manage, and roll it out, and everyone is happy. It works for the little kid with asthma. Conjunctivitis. The healthy person with a broken hip.
At the moment, the country is throwing an extraordinary amount of money at these problems – pharmacy with their algorithm-driven treatment, urgent care clinics, single disease clinics.
These are nice problems – the ones with measurable outcomes, and evidence, and the possibility of large shiny clinics providing services with a multidisciplinary team.
Everyone knows what to do. Healthcare can be efficient and effective and audits work.
The unknown knowns
The second group are slightly complicated – unknown knowns.
Usually this involves people with multiple conditions, or complicated disease, or rare diseases. There is an algorithm, but it’s not well validated, or it is difficult to combine multiple algorithms for multiple diseases.
Again, there is quite a lot of support for these people and the health professionals who care for them. Audits help somewhat, particularly if the person has a few well-known problems, and there are many similar people in the population.
The country is throwing some money at these problems – women’s health clinics, outpatients departments with specialty clinics, aged care facilities.
They are difficult conundrums that can be sorted out, given time and resources. They attract some funding, but it is harder to argue for it, because they don’t look efficient.
The complex group
The third group is less obvious, because they are complex.
These problems evolve over time. In retrospect, these problems are relatively clear, if not obvious. They are often defined by what they are not, rather than what they are.
The classic case is a person with something autoimmune, with vague symptoms including fatigue, weird rashes, non-specific joint pain and frustratingly normal or borderline blood tests.
These poor patients can easily get stuck in what Balint called the “collusion of anonymity” – circulating around specialist clinics finding out what they don’t have. For years.
They will be misdiagnosed in pharmacy, because they will look like common diseases if you don’t dig deep enough, so they’ll be treated “as if” they are common – the opposite problem of students who hear hoof beats in the city, and assume it is a zebra.
Related
Algorithms treat all zebras like horses.
The investment in these patients has declined significantly. We need general physicians, general psychiatrists (the type who have time to think) and of course GPs.
Audits aren’t helpful, although debriefing usually is.
We need people who can think about rare diseases, and are curious about the “zebras” we see. Thinking of my last six months in practice, it includes genetic conditions, Marfan’s syndrome, neuroendocrine tumours, Sjogren’s disease and others that don’t fit any evidence-based mould.
The chaotic patients
But the ones we are really struggling with at the moment are the chaotic patients.
Medicine has no clue what to do with them. Now and then, we solve a piece of the puzzle, but then the problem evolves, and the situation just becomes worse.
These people are very likely to live with the financial, social, emotional, and physical impacts of marginalisation, discrimination and trauma. Nothing is “fixable”, all we can hope for is to convert them to complex if we are lucky.
These patients have personal challenges (e.g. disability, low literacy, gender diversity) unusual and ill-defined symptoms (e.g. fatigue, anxiety, nausea) and multimorbidity.
There are no guidelines. There is really no research. They are rejected or ejected from all the services, and nobody is funded to care for them. They don’t meet the criteria for disability support pensions – they will never be considered “fully diagnosed or fully treated” and there is no hope of them surviving NDIS processes.
Instead, they stay in general practice, with GPs and their colleagues trying to sew together a patchwork of services using the age-old technique of indiscriminate begging. None of it is eligible for funding, and the advocates are never paid.
The bottom line is this.
The hardest problems we face are situations where there is no relevant evidence, no funding and absolutely no place for quantifying whatever the “outcome” might be. Like the scientist in the 1950s, we could pretend they are simple, and propose an app, or a tool, or a digital solution that will treat these patients as if they benefit from evidence.
They clearly don’t.
Or we can just make healthcare so bureaucratically florid that these people give up and go away. It’s a terrible indictment on our attitude to care.
I don’t understand (except of course I do) why investments are always heavier at the simple end. I don’t mean the disease is simple, but the problem is straightforward.
The chaotic people are existing in the system, either being forced into an irrelevant series of simple care, or being exited out of everywhere.
In our current age, I note the same wishful thinking around technological innovation I saw when science was king. If we only had all the algorithms, and we could add them together we would get it to a point that every single person could be classified and costed, and get standardised care that works, is measurable and removes uncertainty.
I am a non-believer, I’m afraid. There will still be a junk drawer. There always is.
AI will augment inequality, providing “treatment” that doesn’t solve problems it cannot understand, because the data for the chaotic patients never gets into the research that grounds its decision making.
Frustrating, isn’t it?
Professor Louise Stone is a GP in Canberra and an academic at Adelaide University. A collection of her research, policy and teaching materials can be found at drlouisestone.com.
