🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Alas, a trial is required - one cannot use observational data, without more assumptions :/

board exam study tip: keep a mini-differential and update with new info, like a filter. less memory required, and fun! wordpress.com/post/samweisen…

Good! o1 acknowledging the independence assumption arxiv.org/pdf/2311.12188

IPW can be thought of as a density transform samweisenthal.com/2025/03/19/inv…

Here’s our new preprint led by graduating URMC MSTP candidate Sam Weisenthal. There has been previously published work which shows that sensitivity and specificity are not independent of covariates. doi.org/10.48550/arXiv…

Humanity's fundamental DAG

samweisenthal.com/2025/01/27/dec…

Medicine is swimming in pre-test probability scores (eg Wells’). These scores + constant sens/spec assumption (see Dawid 1976, Moons and Harrell 2003, Guggenmoos 2000) = Information Mismatch See preprint for details: arxiv.org/pdf/2503.15382

Fantastic piece on many of the limitations of test characteristics for determining post-test probability (which are well-known, and I rant about frequently with my poor residents)

diagnostic tests are characterized by sensitivity and specificity, patients by covariates. when we condition one and not the other, we violate the assumptions of Bayes’ rule. this “information mismatch” is a strong mathematical argument for precision medicine 👇

Any medical decision involves risks and benefits. Research studies (trials, risk score development, meta-analysis, etc) help us characterize these risks and benefits.

When deciding whether to take a med, we think about benefit but also side effects. We need more risk models that predict whether someone will experience a side effect when taking a med.

If a trial is powered for a primary outcome such as blood pressure, take care when drawing conclusions about noisier secondary outcomes, such as time-to-dialysis. samweisenthal.com/2025/04/17/sec…

Kudos for analyzing management decisions.