Opens profile photo
Follow
Pierre Elias, MD
@PierreEliasMD
Assistant Professor of Cardiology & Biomedical Informatics || Medical Director of Artificial Intelligence
New York, USAmembers.dbmi.columbia.edu/CRADLE/Joined August 2011

Pierre Elias, MD’s Tweets

Pinned Tweet
My mother in law's bottle of atorvastatin is just filled with peanut M&M's... THIS IS WHY CARDIOLOGISTS HAVE TRUST ISSUES.
Image
112
6,342
Yesterday evening we got to welcome Luna Claire Elias to the world and our family. Biqi and Luna are doing great. Wishing you happy holidays as we enjoy our own little gift!
Image
Image
38
429
George taught me the motto "nice, bright, and hardworking" really is the foundation of how you build great culture. Thank you for recruiting me and supporting me through the past 7 years, George!
Quote Tweet
Nice. Bright. Hard-working. Those words described the DBMI culture under the leadership of George Hripcsak, whose 15 years as DBMI chair brought innovations, new connections within @Columbia, and the next generation of informatics leaders. @Columbiaps dbmi.columbia.edu/connections-in
Show this thread
5
(please retweet) I am taking CS PhD students! If you are interested in ML, data science, health, or inequality, please apply to Cornell :) We have an island campus in New York City and a fantastic group of students I feel lucky to collaborate with every day. Join us!
Image
Image
Image
Image
13
650
Great work by and team about scaling HTN/Cholesterol management for over 10k pts using digital health!
Quote Tweet
Very excited & proud to publish in @JAMACardio our study of remote hypertension and cholesterol management in more than 10,000 patients @BrighamWomens, bringing specialty care to diverse patients at scale & improving clinical care bit.ly/3zWmBW7, bit.ly/3WZd0b0
Show this thread
Image
1
My wife and I went to our first birthing class today. Our instructor informed us she'd just gone viral for filming a hawk carrying off an entire dead rat from a trash can. So that's my explanation of what it's like to live in NYC.
Quote Tweet
In New York City, the rats go viral for eating pizza – and the hawks go viral for eating rats. Watch this bird shock onlookers in the Upper West Side by descending into a trash can and pulling out a sizable "snack."
Embedded video
1:30
425.3K views
3
Fourth, you must think about positive predictive value to get doc buy-in. How many of my patients flagged as ++ must I send to find one new pt with disease? Challenge is that means asking what's underlying prevalence of unknown disease. Answer? Model many guesses!
Image
1
19
Show this thread
Third, shifting model populations showed us worse performance in our 4th hospital. I don't think this is overfitting. NYP Lawrence is much older population, and we knew model performance was worse in older cohort. We admit ignorance to how well it will perform elsewhere.
Image
2
13
Show this thread
Second, non-physiologic signals exist everywhere. To make sure ValveNet didn't cheat (label leakage), we excluded ECGs w/ V-pacing & baseline wander. Why? Because they're strong signals of whose sick, & we want to max learning on physiology. Painful to throw out a 3rd of our data
Image
1
19
Show this thread
First, you can't trust AUROC plots. If I want to make mine look better, I just have to include more healthy young pts that the model can be very confident are negative for disease. Without VERY clear inclusion/exclusion criteria, they are meaningless.
Image
2
22
Show this thread
Get hired for a new job in the place you've worked at for 7 years. Then find out they've had your birthdate wrong this whole time. Tell HR your correct birthdate to fix. And then get this email...
Image
3
17