|
Sam Finlayson
@
IAmSamFin
Boston, MA
|
|
MD-PhD Candidate, working on machine learning for (bio)medicine at @harvardmed + @MIT_CSAIL. Past: @Stanford (BA, MS)
|
|
|
4.288
Tweetovi
|
233
Pratim
|
3.517
Osobe koje vas prate
|
| Tweetovi |
|
Sam Finlayson
@IAmSamFin
|
12 h |
|
@JFutoma might be the closest yet? twitter.com/EricTopol/stat…
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
12 h |
|
To be clear: I don't mean this is a "gotcha," as I see the authors even pointed this out themselves on the other thread. It's a cool study and it's exciting to see AI enter the sham-controlled phase. Great progress.
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
12 h |
|
Great point, and I'm definitely not criticizing the authors for not doing the prohibitively expensive. That said, I still don't think we can say we've had our first proven case of AI benefiting patients yet. No outcome counterfactual for (small) otherwise missed lesions.
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Scott Gottlieb, MD
@ScottGottliebMD
|
17 h |
|
Thread: There are now 2 deaths in 146 patients diagnosed with #coronavirus outside China. It’s a case fatality rate of 1.37%. This may be our best estimate to date. The patients include mild disease as well those presenting with pneumonia. It tracks with 2% CFR reported in China.
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
20 h |
|
Yep, I immediately regretted saying "clinical" endpoint rather than "clinical outcome." Sorry
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
20 h |
|
Congrats to the team on a nice paper! And good thread describing it.
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
20 h |
|
Cool, thanks for applying. Checks the double-blind box (great!) but endpoint is not a clinical one, correct?
AI for endoscopy/colonoscopy is an interesting problem area in general, bc there is usually/often a procedural element baked into the scoping session.
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Tyler Berzin MD, FASGE
@tberzin
|
22 h |
|
1/9 Quick synopsis of our recent double-blind randomized study on computer-aided polyp detection in @LancetGastroHep, which can be found here: thelancet.com/journals/langa… pic.twitter.com/CsFWgjXaDQ
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
21 h |
|
Interesting (and painful) breakdown of what likely happened in Iowa last night.
Also, I don't know who came up with the name for "Shadow, Inc." but that's almost a self-parody-level naming move on the part of a contractor building an election tallying tool. twitter.com/megerman/statu…
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Derek Lowe
@Dereklowe
|
23 h |
|
More disputes on AI-style drug discovery, and hopes that the field can eventually get past this phase:
blogs.sciencemag.org/pipeline/archi…
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Saige Rutherford
@being_saige
|
24 h |
|
Definitely going to use this feature for my next talk! blog.google/outreach-initi…
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Jeremy Howard
@jeremyphoward
|
23 h |
|
How did I not know of this until now?
pytorch.org/docs/stable/au… pic.twitter.com/vgwvszgFEo
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
24 h |
|
This is a perfectly reasonable concern w.r.t. optics.
I was just making clear that Venky’s concerns about dichotomania etc were more about one instantiation of the lottery idea than the randomization principle itself. Yours is much more fundamental.
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
4. velj |
|
How about scoring all grants but then converting those directly into sampling probabilities? You could use a smooth function that strongly favors the upper quartile, but doesn’t create dichotomania. Then sample over the full range with those probs.
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
hardmaru
@hardmaru
|
4. velj |
|
A pneumonia outbreak associated with a new coronavirus of probable bat origin
(This is an unedited manuscript that has been accepted for publication. @Nature is providing this early version of the manuscript as a service to their readers.)
Open Access: nature.com/articles/s4158…
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
Vinay Prasad
@VPrasadMDMPH
|
3. velj |
|
I enjoyed reading this article about companies that are moving into the space of
A blood test to screen for cancer
They have an uphill battle that I fear they do not yet fully recognize
[Thread]
cnbc.com/2020/02/01/det…
|
||
|
|
||
| Sam Finlayson proslijedio/la je tweet | ||
|
\mathfrak{Michael "El Muy Muy" Betancourt}
@betanalpha
|
2. velj |
|
Intro physics experiments are more complicated than expected but that's what makes them perfect for demonstrating the utility of Bayesian modeling. In my most recent case study I try to infer gravity from falling ball data and find a few surprises: betanalpha.github.io/assets/case_st….
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
4. velj |
|
Great point. Whole body CT is definitely a better analogy than 23andMe, in my opinion.
I wonder if whole body CT was also held back because the potential harms from radiation was probably more obvious in a big noisy scanner than are the more epidemiological harms.
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
4. velj |
|
To be clear, I was commenting on the market, not the actual health value to that market!
|
||
|
|
||
|
Sam Finlayson
@IAmSamFin
|
4. velj |
|
These *feel* like different markets to me--or at least a very different pitch.
ctDNA would be serially ordered and (be sold as) immediately actionable health information/alt to colonoscopy, etc.
23andMe is 1 time, and even then for most users buy it either for geneology or fun.
|
||
|
|
||