Brian HoltAug 3, 2023
mhoye
I heard a confession from a friend the other day who's working on ML in health care who's leaving that field as soon as he can because as best as he and his team can measure, the most important thing ML analytics can bring to hospitals is the mathematical certainty they should have taken the money they spent on ML analytics and hired nurses instead.
Aug 2, 2023 at 1:20amWeb
Show threadClifford AdamsAug 2, 2023
@mhoye
Well, *ANYONE* could tell you to hire more nurses! It takes a skilled executive (who deserves a huge bonus) to suggest that more ML analytics are needed. 🤪
Show threadMin HeadroomAug 2, 2023
@mhoye to be faaaiiir that also applies to a great many other things that we spend money on!
Show threadGraham KnappAug 2, 2023
@mhoye sounds like a business opportunity for a nursing agency! Set up a ML consulting bureau, study the hospital's request and then say: actually the best clinically proven solution for your specific situation is more nurses 🏥
Show threadFiadh (she/fae/it)Aug 2, 2023

@mhoye Oh no, it's this joke IRL

https://existentialcomics.com/comic/199

Philosophy News Network: Strike

A philosophy webcomic about the inevitable anguish of living a brief life in an absurd world. Also Jokes

Show threadthe gripe ibisAug 2, 2023
@mhoye ah consultancy by any other name...
Show threadBraggAug 2, 2023
@mhoye that’s disappointing I was hopeful ML could act as a backup for cases when something is missed as a scan to prompt a second look/scan.
Show threadJoe Cooper 🇺🇦 🍉Aug 2, 2023
@mhoye fwiw, they're having a hard time hiring enough nurses, and lab techs, and everything else. My aunt is a semi-retired critical care nurse and she keeps going back to work because they are desperate and paying well. COVID, and the way it was handled (or, more accurately, not handled) really did a number on the profession. I'm not saying ML helps with that problem, just that "hire more nurses" is actually a hard problem.
Show threadarclightAug 2, 2023
@swelljoe @mhoye Maybe that's the hard problem we should be putting our minds and resources toward solving :)
Show threadfreeformAug 2, 2023
@swelljoe @mhoye but also it seems like not so hard a problem if it consists in compensating care work fairly (above the current surges too), especially if scrap the nonsense of ML expenses and actually acknowledging a honorable profession - first by compensation (bc capitalism, ugh) then by actually caring for care (well.. revolution), so yes I guess a hard problem but also no
Show threadJoe Cooper 🇺🇦 🍉Aug 2, 2023
@freeformtragedy @mhoye I believe it's kinda like the problem with skilled trades being incredibly hard to hire for. The pipeline of people going into those fields for the past couple decades has not been large enough to supply for the demand. Partly because everybody in my generation onward was told they had to get a university degree to get a good job and be respected and compensated well. In medical fields, it's exacerbated by the large and aging boomer population needing more care.
Show threadJoe Cooper 🇺🇦 🍉Aug 2, 2023
@freeformtragedy @mhoye so, yes, paying nurses and other medical caregivers well is a prerequisite, but it takes several years after that change for graduating high school kids to start wanting to go into those fields and for trained workers to start showing up. Some workers who left the field could be brought back in with increased wage offers, and many have, but it's still worse than pre-pandemic. And, of course, our profit-motivated healthcare system doesn't want to pay workers more.
Show threadWalter van HolstAug 2, 2023
@mhoye Then there's this https://www.theguardian.com/society/2023/aug/02/ai-use-breast-cancer-screening-study-preliminary-results
AI use in breast cancer screening as good as two radiologists, study finds

Preliminary results from major trial show use of technology almost halves human workload

The Guardian
Show threadferricoxideAug 2, 2023
@mhoye

Meanwhile, nursing home consortiums are regularly running commercials on our local stations saying how bad an idea it would be for the federal government to dictate how many nurses per patient an organization should have.
Show threadmenisadi Aug 2, 2023
@mhoye
That's a bummer 😕
I've always dreamed of someday working in this exact field (to create some good in the world with my knowledge)
Show thread{{template.display_name}}Aug 2, 2023

@mhoye @memory Which leads to yet another round of:

- We're "data-driven!"
- Okay here's what the data says
- Not like that

Show threadHealth Data AbacusAug 2, 2023
@mhoye I recently saw a low 8 figure grant that did a study that found poor people in my city have trouble acquiring healthy food because they can't afford it. Which sounds like the kind of problem an eight figure grant could help with...
Show threadCat HicksAug 2, 2023
@health_data_abacus @mhoye my biggest heartbreak working as a researcher in questions exactly like this was always that we all fully knew and wanted to start past the laws of gravity of social problems (eg, "kids need food sleep and safety") but so many of those situations were about trying to create the most indisputable evidence to argue with the worst people you've ever imagined who FULLY do not believe (insert whichever, "kids need food" etc)
Show threadLisPiAug 3, 2023

@grimalkina @health_data_abacus @mhoye Do not believe or have profit to be had in not refusing to believe it?

The two are subtly different problems, and I think the latter is worse.

Show threadCat HicksAug 2, 2023
@mhoye read this and then immediately saw this: https://med-mastodon.com/@mloxton/110820467378546657
Matthew Loxton (@[email protected])

Three years ago, I put a research paper on MedRxiv, and started trying to get it published. After four different journals said it was not quite their thing, and I started to get inundated by predatory journals, I gave up. That is the fate of about half of all papers written, and is a serious problem. https://www.medrxiv.org/content/10.1101/2020.03.16.20030684v3

Med-Mastodon
Show threadLionelBAug 2, 2023

@mhoye

Bean counters measure the metrics which are easy to measure.

Natural law decrees that those things which are easiest to measure are the least useful things to know.

Show threadluddite 🇵🇸 computerAug 2, 2023
@mhoye tech is not always the solution to problems. One day we might learn that.
Show threadPaul HouleAug 2, 2023
@mhoye i worked with a doctor who’d gotten his MD but decided to start a startup instead of doing a residency and getting board certified; we were trying to develop foundation models for healthcare NLP before there was transformers; we disagreed about quite a few things, one of them was that I was convinced (as awful as it was) that the best market for our product would be automating recoding for docs to get more money out of insurance
Show threadSeán FenianAug 2, 2023
@mhoye I'm not convinced I have seen a single conclusion come out of big-data machine-learning that couldn't have been developed in less intrusive ways.
Show threadAnn Marie GambleAug 2, 2023
@mhoye ongoing willingness to collect moar data but never to hire the people to implement the practices the data suggest 😩
Show threadFlorian LawsAug 3, 2023
@mhoye Strictly ML analytics or other applications of ML as well?