- cross-posted to:
- technology@beehaw.org
- cross-posted to:
- technology@beehaw.org
Inside a bustling unit at St. Michael’s Hospital in downtown Toronto, one of Shirley Bell’s patients was suffering from a cat bite and a fever, but otherwise appeared fine — until an alert from an AI-based early warning system showed he was sicker than he seemed.
While the nursing team usually checked blood work around noon, the technology flagged incoming results several hours beforehand. That warning showed the patient’s white blood cell count was “really, really high,” recalled Bell, the clinical nurse educator for the hospital’s general medicine program.
The cause turned out to be cellulitis, a bacterial skin infection. Without prompt treatment, it can lead to extensive tissue damage, amputations and even death. Bell said the patient was given antibiotics quickly to avoid those worst-case scenarios, in large part thanks to the team’s in-house AI technology, dubbed Chartwatch.
“There’s lots and lots of other scenarios where patients’ conditions are flagged earlier, and the nurse is alerted earlier, and interventions are put in earlier,” she said. “It’s not replacing the nurse at the bedside; it’s actually enhancing your nursing care.”
You must log in or # to comment.
This is exactly what we want machine learning to do, analyze existing data and quickly report to a human with what it found.
Generative LLM’s are garbage, analyzing with machine learning aids is useful.
Generative LLM’s are garbage
And they hallucinate uncontrollably. You literally cannot trust their output.
To be honest, I was expecting the AI to just predict deaths with some accuracy so that the deaths were no-longer “unexpected.” I’m glad ML can be used for something other than 10 second clickbait and revenge-porn.
That warning showed the patient’s white blood cell count was “really, really high,” recalled Bell, the clinical nurse educator for the hospital’s general medicine program.
I’m not a doctor, but even an idiot would know when a WBC is “really, really high” and assume infection. I mean, shit, "suffering from a cat bite and a fever, but otherwise appeared fine "… um, a cat bite AND A FEVER… red flag!
“It’s not replacing the nurse at the bedside; it’s actually enhancing your nursing care.”
I would argue that this would make nurses less important, and would make them “lazy” by not giving them opportunities to identify these simple things on a regular basis.
Would a nurse who doesn’t know what a very high WBC entails be paid less? I would think so.
I can see AI/machine learning used in very complex cases where a human HCP would simply not have the number-crunching capability to find a diagnosis, but this was not that case.
I think this is exactly the case for automation to be useful without negatively impacting the professional. It’s not a matter of nurses having the knowledge or expertise, but a tool that takes away the toil of monitoring - which is boring, easily skipped or performed badly by a tired brain, and is trivially interpretable. If a thingamabob beeps louder and makes the nurse pay attention to the blood cell count, the human is still in the loop of decision making.
but a tool that takes away the toil of monitoring
Ok, so Lifelabs posts patient lab results online for them to see. They CLEARLY mark “high” and “low” for items that are out of range (of the norm).
A nurse would quite literally crosscheck 50 blood markers in a matter of seconds, without the need for expensive AI or at a risk of them losing their job/qualifications.
In this specific case, the fever + high WBC would be more than enough for a nurse to know that something was up. It makes me think that adding AI just adds another step.
I’m not saying that the application of AI to detect abnormalities is wasteful, but I do think it’s unnecessary and possibly a negative in the context of basic lab work.
A nurse would quite literally crosscheck 50 blood markers in a matter of seconds
Yes, but also a nurse has bazillion other things to do. That’s probably why, as the CBC journalist reports, “the nursing team usually checked blood work around noon”. So even though it costs a second to do, it’s done was done once a day. Now it’s done continuously because it’s an alert system instead of something the nurse has keep an eye on.
In this specific case, the fever + high WBC would be more than enough for a nurse to know that something was up. It makes me think that adding AI just adds another step.
Sure, there’s another computation step. But that’s cheap. Nurse time is the bottleneck. From the POV of a nursing team, before, there was a step (check blood pressure at noon), now there are no steps. They replaced a process of checking some numbers with an automated metric-based alarm. This is textbook operations process optimization, great for everyone involved.
I understand the optimisation. The hospitals must be happy, but if I were a nurse (or doctor), this would make me nervous.
Any good healthcare professional would still want to look over the results, even if an obvious flag wasn’t raised.
To me, it’s just good practice (as a patient).
Or maybe they still do, and this system is simply a reducency safety check.
I would argue that this would make nurses less important, and would make them “lazy” by not giving them opportunities to identify these simple things on a regular basis.
Not just nurses, but doctors too. This exact problem was discussed at a conference I recently attended. Some doctors do better with AI assistance, some do worse. As far as we know, it seems to be dependent on how much they “believe in AI”. The more they do, the worse they perform when assisted.
I think it can be useful in predicting a diagnosis months/years before a doctor would be able to, since it can analyze data and look for patterns across millions of cases. This would be especially useful in rare diseases, or even something like dementia.
But using it to tell a nurse or doctor that their patient’s white blood counts are “really, really high” after being bitten by an animal is borderline insulting to healthcare professionals.
This is how you get worse racism in hospitals
How?
Black people are more likely to die (due to systemic racism), so AI says: save the white person.
We saw this a lot at the height of the pandemic, which is why many nurses argued that the best triage method was random selection.
As always the problem isn’t inherently that AI exists. The problem is that humans trust its output and use that to make decisions (and the laws still allow them to do it in many jurisdictions).
But this isn’t generative AI, where AI creates an outcome. It simply notified the staff OF the outcome of human-performed tests.
I get AI is scary. We should be wary of how much control we give it. But in this case it had no control over any outcome.
