Invisible Women: diagnosing heart attacks with AI again...

...but this time it might not be terrible?? Plus, default male sports science reporting, also again.

Good morning GFPs! There seemed to be a bit of confusion following the last newsletter about the changes I’m trialling, so just to be very extremely clear:

Everyone can still get the newsletter for free!

It’s just that paying GFPs will get the full fat version on Mondays, while everyone else will get a preview on Monday followed by the full fat version on Tuesdays. And after two weeks, the free online version will go behind a paywall, although obviously you will be able to access the free version in your inbox for as long as you choose.

In other GFP member news, many of you have written to me to say that while you understand why I closed it, you miss the old forum we had and the space it created for you to meet and plot with other GFPs.

SO, I have a cunning plan! I really didn’t have time to give that forum the love it needed to thrive, but I think it’s possible that the Substack chat function might be able to provide a bit of a morphine hit for those suffering from GFP withdrawal — so, I’m going to try that too!

SO: if you’re a paying GFP please hit the comments at the bottom of this post to let me know what topics you’d like to be able to chat about and I’ll have a go at setting up some threads! What could possibly go wrong? 🤪

---

Gender data gap of the week

GFPs, I know it’s traditional for me to fill this newsletter with snark and complaints about how terrible everything is, but this week I find myself having to deviate from said tradition because I read a paper about an algorithm that might actually be…promising???

Let me take you back to 12th May when I was casually scrolling through The Times app when I came across this article:

Oh yeah, I thought, both sceptically and wearily because I am a hard-bitten gender detective who’s seen it all before. Excuse me if I’m dubious, lads, but this all sounds awfully familiar. Similar claims have been made for algorithms diagnosing kidney disease (hashtag tbt the VERY FIRST EDITION of this humble newsletter — are any of you first timers still hanging around?? I hope so!), lung disease, and yes heart attacks, and every single time the new shiny wonder-algo that is supposedly going to replace, variously, doctors, radiologists and lab techs, turns out to have been trained on male dominated data and its performance has not been sex-analysed. When/if that analysis is subsequently done it turns out, to the shock of apparently everyone except hard-bitten gender detectives, not to work that well in the 50% of the population that is female.

Anyway, so I clicked on the article, expecting to be entirely unsurprised and yet strangely still disappointed because I may be hard-bitten but somehow I am also endlessly optimistic.

The article started with the standard spiel. A trial found that a newly developed AI was “more effective than current testing,” it was “99.6% accurate” blah blah yeah but in WHOM, eh? A particular red flag for me was the mention of troponin, the biomarker that is an important part of heart attack diagnosis and which readers of Invisible Women will remember has been shown to differ by sex, not that we seem to pay much attention to this when testing for it because sure why would we (IW, p.239). Anyway, the article didn’t initially mention these sex differences, just referring to levels in gender neutral “patients” so I was all ready to dismiss the whole thing as your standard default male fare.

BUT THEN.