www.nejm.org/doi/full/10....
Sharing my last paper. It was an attempt to quantify /proof something suspected for a long time. Dengue can trigger Guillain-Barré syndrome (we have case reports from 1998).
1/n
#episky
Partial pooling would help
acute PE
How is the data regarding bleeding?
Congrats! Well deserved
My interpretation is: money was spent and more data is needed! @f2harrell.bsky.social
Oh, is that version different from the one you had in bookdown?
yes but I would still buy the book. Amazing team also deserves the compensation
Trackdown package?
I am grateful to Scott Berry for inviting me onto his podcast
55: A Visit with Stephen Senn: Time, Concurrent Controls, and the Bayesian Guidance
www.youtube.com/watch?v=Tnb0...
ORBITA’s trials basically want to know if PCI reduces angina
I guess only overestimating would be a big problem here
but yes, a prior predictive check is mandatory to fully interpret this complex model @f2harrell.bsky.social
Loved their estimand “difference in angina-free days”, very convincing
Nice example to explore type M and S errors
If we assume true effect is HR = 0.7:
- 11% chance that the estimate is in the wrong direction
- Magnitude of the effect is exaggerated by a factor of 5.2
If HR = 0.5
- S-type error= 1.5%
- M-type= exaggerated by a factor of 2.8
code: shorturl.at/Ba8nG
@solomonkurz.bsky.social you might like too
I don’t have access to IPD. I simulated some deaths and coded it as “29”. Then fitted a Bayesian constrained partial proportial odds model in Stan to separately estimate an OR for Y == 0, another for Y == 29 and a third for Y > 0 & < 29. Not bad!! Interesting estimands too @f2harrell.bsky.social
0 -> docs judged these patients didn’t have infection
0-3 -> docs initially though pt had infection, but changed their mind
3-6 -> probably had infection and improved quickly
7 -> most common suggested duration for antibiotics
>7 -> hard to treat infections or wasn’t getting better
Very interesting distributions from a RCT of antibiotic. Bimodal peaking at 0 and ~7, while also truncated at 0
What distribution would fit this best?
Can’t thank you enough. Open-source teaching at its best
Have you tried Claude Code extension within VSCode/Positron? It’s pretty good
@pwgtennant.bsky.social
✨ New open-access book✨
I’ve seen how powerful prediction models can be, but also how often they fall short. We wrote a book, covering not just development, but also when models are needed, and how to ensure real-world impact.
www.maastrichtuniversitypress.nl/cpm
@gscollins.bsky.social could you please send me a screenshot of your letter? thanks jamanetwork.com/journals/jam...
half-normal(0.5)
Here you go bsky.app/profile/rbly...
hbiostat.org/rmsc/genreg#...
