Robin Blythe

You can understand plenty about the mechanisms of the citric acid cycle. Still got to memorize a huge amount of shit though lol

Greatest con the fossil fuel companies ever pulled was by telling everyone that emissions are just down to us.

Agreed. Does my head in that people think AI is going to change all that somehow, completely ignoring everything we've learned in the last decade(s)

Lmao this would have been priceless to see

There is no such thing as a validated prediction model - BMC Medicine Background Clinical prediction models should be validated before implementation in clinical practice. But is favorable performance at internal validation or one external validation sufficient to claim that a prediction model works well in the intended clinical context? Main body We argue to the contrary because (1) patient populations vary, (2) measurement procedures vary, and (3) populations and measurements change over time. Hence, we have to expect heterogeneity in model performance between locations and settings, and across time. It follows that prediction models are never truly validated. This does not imply that validation is not important. Rather, the current focus on developing new models should shift to a focus on more extensive, well-conducted, and well-reported validation studies of promising models. Conclusion Principled validation strategies are needed to understand and quantify heterogeneity, monitor performance over time, and update prediction models when appropriate. Such strategies will help to ensure that prediction models stay up-to-date and safe to support clinical decision-making.

Yeah, main issue seems to be that everyone can get their hands on some data and can publish a model. I mean, that was part of what I did for my PhD. I'm not entirely sure the problem is just validation though, I tend to follow van Calster et al's thinking on this: doi.org/10.1186/s129...

You're correct. Standard practice in economic and health economic research but lots of assumptions and not easy to explain to laypeople. Researchers will often elicit a 'willingness-to-pay' value for the utility gained.

I've only rarely interacted with psych research, but I got the feeling from this paper by Gelman and Brown that there's a big issue of people forming preconceived notions then doing silly analyses to support them

sites.stat.columbia.edu/gelman/resea...

> LinkedIn
Doctor, I've found the problem.

I think with regards to diagnosis more generally, there's a growing consensus away from that kind of thinking - I think @vickersbiostats.bsky.social has recently written about this. Positive signs things will improve eventually

I'm not even sure alert fatigue really captures the scope of the problem. It's not just alerts, it's the panoply of different decision aides that we seem to only add to in clinical systems. Plus, I think there's a disconnect between clinicians and informaticians re: perceived clinical utility.

That uncomfortable feeling where you upload a Quarto tutorial with your published paper, only to then realise that you left your stupid personal filepath in there when you rendered it ☠️ Absolutely gonna get roasted for this

Should probably clarify here if the bullet-hell comment didn't make it obvious, this is by far the better option lol

Lived in Australia for half a decade and never had a pie that even came close to your average NZ bakery offering.

This was from an executive agency, so I'm worried they actually thought it was.

I have already seen borderline fraudulent CEAs done on LLMs to "diagnose autism" with the payoff for positive predictions in childhood being the avoidance of unemployment and incarceration. Cost-benefit put the ROI at like $1m/dollar invested.

The joke about the reverse (Aussie visiting UK) writes itself, a la Jimmy Carr: "Returning to the scene of the crime?"
But we don't have any good jokes for Brits visiting Aus 🤔

LinkedInfluencers genuinely believe this shit though. Maybe the only weapon we have at this point is satire - make your pseudonym Wayne Kerr or Mike Hunt to drive the point home. Let us know when the Nature publication is ready

I can't tell if this is better or worse than if they crash into each other in space, creating a bullet-hell system of sharp particles travelling at near-orbital velocity

Yes, 100% agree. I didn't mean to imply that we shouldn't have informatics at all, rather exactly what you said.

At some point the field of medical informatics will have to come to terms with the fact that clinicians rarely even pay attention to prediction models in the EHR. "Decision support" is a nice fantasy but there are already too many competing demands on clinician time.
#statsky #medsky

It's like the paint-by-numbers version of medicine.

I turned off that feature for everyone. Turned it back on for a minute and immediately regretted it.

I haven't seen a whisker of this - link? Is it funny? Or just tragic? Or both?

I think the poor quality of many masking studies have given a lot of ammo to the anti-mask groups. Bit of a shame really, as IMO here what we're really studying is mask-wearing, not the masks themselves.

Yes certainly. Those are two different interventions really.

It seems as though the likelihood function is still useful for identifying different underlying theories for different models. Maybe the distinction is just that this ceases to be BMA, and is just using the likelihood function?

A Conceptual Introduction to Bayesian Model Averaging - Max Hinne, Quentin F. Gronau, Don van den Bergh, Eric-Jan Wagenmakers, 2020 Many statistical scenarios initially involve several candidate models that describe the data-generating process. Analysis often proceeds by first selecting the ...

Love this paper by Hinne doi.org/10.1177/2515... but curious why BMA isn't compatible with competing theories: "BMA is less useful when...each candidate model may represent a different theory of a physical process...the models are not a nuisance factor; they are the focus of the analysis." #statsky

All of my research is in medicine, so I was pretty surprised to learn RCTs are even used in other fields, much less held up as the best way to estimate generalisable effects.

Sure, I think most medical statisticians would agree with that. I can't say I've ever come across someone in medicine who thought that was the point of an RCT though!

Also FWIW I agree RCTs are held up as the solution in many complex systems when they're totally inappropriate - so perhaps we're talking past each other here. I was primarily referring to treatment efficacy in medical contexts.