Oxford Neurological Society
Skip to main content

The Solaris of Artificial Intelligence in Clinical Research

The modern world collects an obscenely humongous amount of information about humans, their interactions, patterns of behaviour, health states and service use. Connected to super-cluster servers and cloud storages are endless streams of data coming from our phones, watches, cameras, fridges and toasters, most of which we submit voluntarily, with varying degree of awareness of what happens to them later on.

Huge clusters of hospital and healthcare data

No different are the modern hospitals: blood tests, discharge summaries, request sheets, communication minutes, imaging, clinical letters and functional studies are all documenting the state of individual’s health from cradle to grave. And this does not even include endless paperwork that’s not digitalised: heart traces, hourly nursing notes, checklists, printouts and, yes, Bristol stool charts.

Conspiracy theory specialists will let you believe that this avalanche of potentially priceless information is being sucked into a super computer and then carefully tracked and analysed by the Government, insurance companies, or your other favourite all-time villain.

 

In fact, nothing of a kind is happening.

All this information gets dumped right into the oblivion of databases and clinical information systems, and is covered by the dust of auto-filing functions and never looked at again unless by a lonely medical student completing their summer auditing project.

The answers for our clinical questions, efficiency savings or probability figures are out there.

There’s just not enough resources to swim deep through this ocean of information and make something out of the lengthy spreadsheets.

 

But this is about to change.

Very recently, the ocean of information started speaking to us. Out of the dark abyss of Marian Trenches came the sounds that were recognised by the clinical ear. Recommendations, lists of improvements, quantified tables and graphs, all obtainable in seconds and all produced by the modern cliché: the artificial intelligence.

Algorithms are now able to scan through dozens of documents in a split of a second, analyse complex associations and propose concrete ways of improving clinical practice, all on the scale that has never been seen before.

Yet the crazy and unrealistic expectations of the AI’s impact on modern medicine may be as a feeble and short-lived as the gene therapy and DNA sequencing obsession.

 

The elegance of disappointment

In the campuses of private companies, universities and government institutions, work on the most complex and ingenious neural networks, algorithms and analytical systems is on full steam ahead. Crowds of scientists with mathematical and computational abilities of undreamed magnitude produce amazing solutions to analyse the output of clinical work.

Google AKI can predict the probabilities complications for patients with kidney injury, Google Flu had a shot at anticipating the outbreaks of the disease and the IBM’s drug analytics embark on an arduous journey of preventing drug interactions.

 

However, the clinical relevance of these findings and indeed their potential implementation remains questionable.

Whilst the clever algorithm can deep-learn the kidney injury probability, it will not change the decision on whether to prescribe the drug or not. A couple of percentage points one way or the other is not going to determine a patient’s suitability for an operation.

In this case, the only thing that can indicate and meaningfully guide clinical practice is a large change over time. As the person progresses through the stages of chronic kidney disease, a computerised alert system could flag it up and even paint the screen in an outlandishly red frame.

 

But to achieve that, you need just a couple of lines of code, and all lab reporting packages are already doing that right now.

The flu prediction, however clever and intricate it may be, is not going to improve the take-up of flu jabs or change the usual infection control regimens. Whether a drug will interact with another medicine in 88.2 or 72.4 percent of cases, it will still be unlikely to be prescribed.

 

To risk every 8 out of 10 patients would be just unreasonable as risking 7 out of 10.

 

Least of all, the interactions are already built in the apps for doctors and they scream out loud every time you click on a button to prescribe it on the computer.

Is the case for accuracy really a worthwhile one whilst some GPs struggle with management of hypertension: a disease which requires a more-or-less straightforward sequence of three drugs?

The problem with AI in medicine is that the mathematical minds of the engineers are as out of touch with the clinical service, as the super-cluster computers are detached from a dusted, freezing NHS desktop PC.

What we need is a system that would be immediately effective and provide so much value that it will shoot us between the eyes: something that would be as seamless as to being invisible, as specific as to invoking concrete names, numbers and places, and as efficient as to being launched by a single press of a button.

I am not arguing for the ocean to give me a clever figure or an animated chart.

 

I want it to tell me that I have prescribed a wrong drug to Mr Jones

And, that I should have requested a scan for Mrs Smith. I want it to invoke the fond memories of an elderly patient I saw last Monday and propose concrete, pragmatic ways I could’ve improved their care. I want it to figure out why Theatre 16 is getting more infections than other facilities and what can I do to shorten the waiting times of patients waiting to see me in clinic.

This approach, granted, is deprived of all the amusements of purely academic form and shape. Yet in the world of limited resources, a finite amount of research money and time, I believe we should be investing in something that produces an immediate and tangible improvement for the life of our patients.

I want something that will give us, clinicians, more knowledge about concrete ways of improving practice and saving time and money. Something that will get us from the paperwork back to the clinic. Something that will allow us to do what we’ve all signed up for: spend more time with our patients and provide them with a compassionate, evidence-based, and well-informed care.

And I think that this vision is the most exciting.

 

In this trilogy, I will take you on the journey throughout the AI world. You’ll learn about my personal experience in creating the technology, its remarkably intriguing behaviour, and what I think about the future of the AI-powered NHS.

So, sit back, relax and enjoy the ride.

 

EPISODES:

(1) Landing: the development of the AI and its peculiar behaviours (you are here)

(2) Life on the station: how the AI can work within the current NHS system

(3) Farewell: the AI-powered brave new world [28th September ]

Don’t miss the release! Follow me on Twitter for live updates.

About the author


Max Brzezicki

Max Brzezicki

Passionate about evidence-based medicine and science, likes slicing meat, crushing rat brains, criminal & public law, foreign languages, rhetoric, history, classical studies and political thought. FNS since 2015.

Popular posts this week


Leave a Reply

Your email address will not be published.