Justin Healy

Doctor. Bad chess player, worse swimmer.

[Justin Healy @justintomhealy]

@davidstutz92 The incentives are a nightmare though. Ethics committees are almost entirely incentivised to be extraordinarily conservative.

I remember once being involved in a trial involving aspirin post admission for pneumonia. I had to give people a 19 page consent form. For aspirin.

[Justin Healy @justintomhealy]

@davidstutz92 Much of that could be sped up. I've got a small paper using MedGemma on an NHS work-flow.

Research/paper done in 3 months. Ethics discussions 7 months. Peer-review 5 months (and counting).

And I'm aware this has been relatively quick.

https://t.co/2FJTXI76HX

[Justin Healy @justintomhealy]

@AdamRodmanMD Because historical medical trials have felt like testing different inks that come out of the same ballpoint pen. But this era feels like we are changing both the pen and the hand that holds it.

[Justin Healy @justintomhealy]

@AdamRodmanMD Would be very keen to hear those opinions! I guess there are overlaps with the ‘ai as normal technology debate’ - to what extent are the evaluation paradigms of the 20th century up to this task?

[Justin Healy @justintomhealy]

@tyhouch @nucleusgenomics Cool stuff. But it doesn't bode well for future healthcare costs if all of that results in some bog-standard lab recommendations with a suggestion to keep an eye on BMI + BP.

[Justin Healy @justintomhealy]

@AdamRodmanMD @krithikvish @ekoermann @METR_Evals Problem with standards and benchmarks is that they pre-suppose the existence of stable/objective ground truths.

But I worry that approach will doom us to years of watching eachother try to nail n-dimensional metaphysical concepts of ‘good clinical care’ to a wall.

[Justin Healy @justintomhealy]

@AdamRodmanMD @LiamGMcCoy @AnilMakam Exactly. Or we bring it all the way back to promoting eudaimonia. Aristotelian virtues as LLM benchmarks.

[Justin Healy @justintomhealy]

@AdamRodmanMD @LiamGMcCoy @AnilMakam In certain cases I suppose you could leave it entirely up to the patient to define 'good'.

[Justin Healy @justintomhealy]

@LiamGMcCoy @AnilMakam What do you think the next step looks like? If the inter-human variability is intolerably high then where does the gold standard come from?

[Justin Healy @justintomhealy]

@LiamGMcCoy @fl_saloni @emollick It’ll also be important to see if these actually translate into tangible better outcomes (for patients or systems) - otherwise you could imagine this making healthcare less efficient (especially if risk tolerance is dialled down).

[Justin Healy @justintomhealy]

@HMRCcustomers Yes, they say this is a HMRC issue.

[Justin Healy @justintomhealy]

@HMRCcustomers. There is a discrepancy between the tax code on my HMRC account and my payslip this month. My payslip is incorrect and has left me significantly out of pocket this month. Neither the telephone nor online chat can connect me to a human. How can I get this sorted?

[Justin Healy @justintomhealy]

@HMRCcustomers I have two assignments with the same employer. On HMRC online one tax code is K790 and the other BR. This month both have had K790 applied.

[Justin Healy @justintomhealy]

@AydinDikerdem @rcolvile Yeah, the whole discussion was way better than what you see on twitter.

[Justin Healy @justintomhealy]

@AydinDikerdem @rcolvile I’m not a huge fan of Colvile’s politics but he literally says that he doesn’t think the ideal level is zero at 19:01.

[Justin Healy @justintomhealy]

@BenRamanauskas I had my civil service interview rescheduled 5 times. I withdrew the application in sheer exasperation.

[Justin Healy @justintomhealy]

@jfdwolff @angularboxoid 10 ischaemic strokes compared to 5 on rivaroxaban. 2 haemorrhagic strokes on rivaroxaban, 0 with abelacimab. Small numbers but non-inferiority may be touch and go.

[Justin Healy @justintomhealy]

@AdamRodmanMD An interesting thing about this is the sheer variability in human performance. If we can squish the bottom of the distribution towards the top then the gains are potentially huge.

[Justin Healy @justintomhealy]

@AdamRodmanMD It may go up on a per-visit basis but it should really come down over the system as a whole. There're a lot of low-value speciality consults that could be eradicated by a team of souped-up generalists.