Add us as a preferred source
From apple cider vinegar to unfamiliar prescriptions, Dr Tim Cooper, GP and Clinical Director for UK & I at Heidi, explains why clinicians need AI that draws on trusted evidence rather than the whole internet.
There’s a medical textbook that sits in my practice today. It taught me at the time much of what I needed to learn the art of medicine. That, and a lot of experiential learning on the job. It now occupies the role of raising my second screen. NHS IT. Don’t ask.
As I worked through my 80th online consultation of the day, I was struck by a question I had no idea of the answer to. “Can you prescribe me Itulazax?” Instinctively, I thought, let me just search it online. That experience will be recreated across the country in every GP practice, outpatient room, and ward.
Medical knowledge is increasing at a pace previously unseen. New technology, medicines, pathways – it is a constant source of change. And yet all of this was previously confined to a search bar. We’ve seen an explosion of tools that sit there in our consulting room. The days of paper forms, the days of handout leaflets – almost all of that now lives in a digital space. And whilst this helps democratise the availability of information, it presents the risk of the quality that sits behind it.
For example, all of us will have had the experience of searching in a browser bar and asking a simple question, whether that’s searching for a new microwave, a place for dinner, or actually a new treatment class. The challenge here is that when we use generalist search tools, we receive generalist knowledge back. And by that, I mean it pulls from the entire internet that sits out there, regardless of the quality of the source.
For example, many of you will have worked across healthcare over the past few years, and I’m sure many of you will have had the experience of apple cider vinegar and its link to acute gastritis. There was a time where apple cider vinegar, perhaps rightly, perhaps wrongly, was seen as a cure-all for many of the challenges we saw. This was more so prevalent at a height of understandable health anxiety during the pandemic.
What I saw in my clinic was more and more people coming to me with acute epigastric pain. Again, the skill of the history taking, particularly when you’re asking, “Has anything else changed?” revealed the fact that for many of these patients, they had read online that a shot of apple cider vinegar could help do everything from boost immunity, improve sleep, to even preventing COVID. And this is the heart of the challenge of generalist knowledge today. How do we know that the quality of the information that both us and our patients are finding in front of us actually lends itself to the future of high quality care?
As we increasingly use tools like this, we will need to have safeguards. We will need to have boundaries. We need to know that if we ask the question, we’ll get an appropriate, realistic, and importantly, true source. As I’ve commented previously, when we’re learning to train, we weight evidence differently. We know there’s a difference as to why a randomised controlled trial occupies a higher status in the hierarchy of evidence than a simple observational one. Observations and an “n=1” could be hugely helpful, but ultimately, we need to be able to compare evidence with a true evidence bank in order to make sure we make the right decisions. We need to weight the RCT more than the Reddit thread.
The future of AI in healthcare increasingly looks exciting and optimistic. Clinical decision support, where actually we’re supporting clinicians of any type at the time that they’re making their decisions, is increasingly likely. I, for one, am not afraid that this suddenly means my role as a GP is redundant. Rather, I think it enhances my role and actually ensures that I maintain my medical knowledge and depth and breadth across everything that will present to my clinic room. Indeed, one of the joys of being a GP has been precisely because you’re never quite sure what’s coming through the door. Although this has been helped a little bit by triage, which gives us a bit of a heads up before the patient walks through. Ultimately, that’s one of the joys of medicine – you never quite know what someone’s going to ask, what they’re going to say, or what they’re going to present with.
Where I see these tools being hugely valuable is acting as a support for me. Rather than having to reach for the book that sits under my monitor and frantically checking to see what this could possibly be, I’ll have decisions in real time, augmenting my ability to both manage patients but also navigate the system. I feel this is more important than ever, not least because of the pressures and challenges that we face across the entire system. We know that healthcare’s capacity is struggling. We know that the demographic challenges suggest that we’re going to be dealing with more patients who are more ill and who live longer. We know as well that there cannot be a traditional response to this of if only we could just train more people. The macroeconomic picture, evident to the majority of people who watch health policy, is that that is not plausible, so we must find alternative routes.
I also think that by making every consultation count in a way that matters both for the patient but also for the clinician, we can drastically reduce the need for recurrent follow-ups or even ending up in the wrong part of the system. For example, the ability to host knowledge and pathways in a system could be hugely transformative. All of us who’ve been through training will understand that feeling on the first day, when you’re being asked to look at a process or move a patient through a pathway. Where do I look? Who do I ask? How does it work? Do they need this test before they go? What do I tell the patient? Again, that’s not even medical knowledge – this is system knowledge, which again usually is held tightly, in the minds of those who work in it and is not often as visible as perhaps it should be.
The future of clinical tools here looks like a repository that will not only host this information, but be able to present it back to you almost immediately. No longer should there be a day where it’s unclear about what pathway to use or how to refer. No longer should there be a question of having to use a generic chat bar to ask a clinical or pathway question. No longer should there be a day where I feel uncomfortable because my training didn’t quite cover what I’m being asked in front of me, or I’m too new to the organisation to know what I should. Googles’ strapline has always been “to organise the world’s information and make it universally accessible and useful”. The future of clinical AI should be to organise the world’s clinical and system information and make it universally accessible and useful to all in healthcare. It won’t be Dr AI, but it may well be Dr and AI.
