Wouter van Amsterdam
@WvanAmsterdam
Followers
617
Following
2K
Media
86
Statuses
762
machine learning, causal inference, health care
Utrecht
Joined January 2014
very excited to have this pre-print online! some prediction models have good AUC pre- and post-deployment, but are (silently) harming patients when used for treatment decisions I'll present this at @SymposiumML4H Dec 10th and will attend #NeurIPS2023 after
Have you developed a prediction model to aid clinicians in a treatment decision? Then you may want to read our latest preprint titled *When accurate prediction models yield harmful self-fulfilling prophecies* See: https://t.co/nIzVNJ7gTx A π§΅
3
6
21
1/ The ππππππ of generations from an πππ is an important heuristic used during post-training to understand model behavior. π½ππ due to a πππππΏ ππππ πΎππππππ ππππΏππ, a large number of trajectories get truncated before ever reaching [ππ’π¦] token.
1
1
5
π Postdoc opportunity at UMC Utrecht on Natural Language Processing & Real-World Data π¬ Work on models for infections & post-COVID symptoms based on GP data. Apply by 30-Sep-2025. π https://t.co/dz37eL3tFl
#PhDJobs #NLP #Epidemiology
0
5
5
For those looking for natural experiments / instrumental variables in healthcare, UK is the place to look, not Saudi Arabia
If we are going to ask how AI aligns with doctor decisions, we have to first know what the doctor decisions are. As part of the Human Values Project, we have challenged doctors with triage decisions. Even though US doctors are 2/3 of respondents so far, Saudi clinicians appear to
0
0
0
TARGET: a reporting guideline for target trial emulation studies:
bmj.com
Importance When randomized trials are unavailable or not feasible, observational studies can be used to answer causal questions about the comparative effects of interventions by attempting to emulate...
0
1
2
work with @D_Schipaanboord , Floor B.H. van der Zalm, RenΓ© van Es, Melle Vessies, Rutger R. van de Leur, Klaske R. Siegersma, Pim van der Harst, Hester M. den Ruijter, N. Charlotte Onland-Moret , on behalf of the IMPRESS consortium
0
0
0
Conclusion: The convolutional neural networks in this study demonstrated resilience to simulated sex-imbalance in training ECG data. pre-print:
1
0
0
Discrimination remained stable across sexes; only calibration shifted in extreme scenarios when prevalence differed by sex, with similar patterns for women and men.
1
0
0
Using ~165k ECGs, we simulated sex-imbalances in representation (women-to-men ratio), outcome prevalence, and misclassification in the training data for LBBB, long QT syndrome, LVH, and physician-labeled βabnormalβ ECGs.
1
0
0
Pre-print alert: Many ECG-AI models have been developed to predict a wide range of cardiovascular outcomes. But, underrepresentation of women in cardiovascular studies raises the question: Are ECG-AI models equally predictive for women and men with sex-imbalanced training data?
2
1
1
He did ask for help in a cute physics with a βgeodesic to the understanding of your papersβ
Subrahmanyan Chandrasekhar, the great astrophysicist, showed humility and curiosity in his research. In 1967, at age ~57, he wrote to a 25-year-old Stephen Hawking, asking for guidance on the math behind Hawkingβs work on cosmological singularities. 1/
0
0
0
Live jamming with generative AI, pretty wild
Realtime interactive generative models FTW! Announcing a new π of details and features for Magenta RealTime, the open weights live music AI model from GDM! * Live Jamming with audio input π€πΈπ΅ * Personalize your own models π§ * Tech report π Links below in the π§΅...
0
0
1
BMS-ANed Spring Meeting on Thursday, June 19 Time: 13:00β18:00 (CEST) Location: Vredenburg 19, 3511 BB, Utrecht Details and registration: https://t.co/3IWPLOg285
0
3
3
Spend a week (7 - 11 July) learning about all things causal inference in Utrecht! Housing available through @utrechtsummer and discounts for those working in universities and non-profits! Sign up through the link in the post below!
Interested in answering causal research questions with non-experimental data? Mystified by DAGs & counterfactuals? Want to learn what Target Trial Emulation is all about? Join the 2nd edition of our summer school, 7-11 July in Utrecht @WvanAmsterdam
https://t.co/8d1Vy53mQ6
1
3
2
A question that remains is how these differences in environments may come about and what to do with this in practice? On this, I wrote a paper titled, available here: https://t.co/MsNGzTVXvC fin!
0
0
2
if the distribution of outcome given features remains the same (Y|X), calibration is preserved. If both are the same, the environments were not meaningfully different to begin with! a more lengthy explanation is in this blog post:
1
0
0
as promised (so all of you can breathe normally again), here's my TLDR answer: Environments must differ with respect to something. If the distribution of features given outcome remains the same (X|Y), discrimination is preserved,
1
0
0
e.g. 'epi-nephrine' -> "upon kidney" 'hypo-glaecemia' -> 'under sugar blood'
0
0
0
could be pretty useful for MD (students)
1
0
2
tagging some stats / prediction people @MaartenvSmeden @f2harrell @BenVanCalster @ESteyerberg @LucyStats @stratosinit
#MedStats #PredictionModels #Calibration #AUC #CausalInference #MedTwitter I'll share my answer (+ link to new framework) tomorrow
1
0
0
Which is stronger evidence for robustness? When evaluating predictive performance of one model in several different environments (e.g. regions / hospitals): A. stable discrimination (AUC) and calibration in all environments B. stable discrimination, varying calibration
1
0
2