Maria Glymour

How HIV/AIDS ended up in Trump's State of the Union speech Top public health officials didn’t expect idea to get high-profile endorsement

Which is why I think it's so valuable to juxtapose Trump's 2019 comments. www.science.org/content/arti...

This is not how we achieve this goal:
bsky.app/profile/pozm...

MSF: Plan to dismantle PEPFAR must be stopped The shuttering of PEPFAR will undoubtedly lead to millions of preventable deaths.

Neither is this: www.doctorswithoutborders.org/latest/msf-p...

woh- this surprised me. The NIH director thinks we can end the HIV epidemic by 2029 and notes this was President Trump's prediction some years ago, when it seemed wildly optimistic. youtu.be/Qtt80qfdGyU?...
I hope he means end it globally. #EpiSky #PublicHealth #NIH #ImplementationScience

Belatedly enjoying Kowall et al multi-analyst paper rdcu.be/ey1uI & Rohrer etal commentary: rdcu.be/ey1uH. A quibble: the trope "these analyses answer different questions". Rather, I'd say the analyses answer the question posed under very different assumptions. #EpiSky #DataSci #ReproducibleResearch

www.cancer.gov/grants-train...

I asked my mom to guess at the NCI payline - she thought 40th percentile? Maybe 60th? Right now they anticipate funding up to the 4th percentile. Can folks point me to their NCI-funded research (from past years) that had a worse score than the 4th %ile? What are we losing? #Cancer #EpiSky #NIH

What can pharma do *now* - with data they already have in their files but have not yet shared with other researchers- to learn how to treat AD? Sarah Ackley illustrates with data from the A4 trial (one of few w/ full data release). @fda.gov get the data. Use the data. #EndAlz

It’s like a 2 year old with perpetual interruptions but without any offsetting toddler charm and it never learns. Do you think the 358th time you ask me, I’ll change my mind? If you were such a great AI you’d have learned by now I probably don’t want your assistance.

Can only hide AI for one session at a time. Driving me bonkers.

Dear Adobe: Your AI assistant is possibly the most irritating invention since the frigging microsoft paperclip. Please stop pushing it. I bet it has collectively reduced efficiency of all Adobe users by 10% - in part because of the pure fury of having to try repeatedly to get rid of it. @adobe.com

This paper from Mary thoma @ucsf-bchsi.bsky.social phd student was lots of fun. Now out in @amjepi.bsky.social

🧠 TIME-AD is now on Bluesky!

#TIME-AD brings together results from different study designs and populations to find what really works to prevent and delay dementia.

Follow along as we share what we’re learning and check out our website to learn more: www.time-ad.org

#EndALZ #EpiSky #NeuroSky

Texas flood should not have happened. What went wrong? | Opinion The deaths of children at Camp Mystic show a heartbreaking failure of local, state and federal government to invest in people, prepare for disasters.

Thanks to the Oklahoman newspaper for publishing my Opinion piece about the devastating losses in Kerr County. bit.ly/44CFFZl

In case people have a hard time accessing the oklahoman, I read it aloud here: on.soundcloud.com/LyVOzvWTjOwv...

A good peer reviewer - who engages thoughtfully and constructively, teaches, notes what's good about a paper and what can be improved - is such a positive force. As a mentor (and as an editor and an author), I send out deep thanks to everyone who works hard to write kind and useful reviews.

I mean, you're apparently *right* empirically- but I would want to see performance of any particular implementation of the data synthesis in actual analyses. And I expect my skepticism (or stronger) would be the typical response of applied researchers.

It's kinda unfair to critique the paper b/c you don't like the "predictive performance", ie, the results. You may have a better model building approach, but would it deliver importantly better prediction given the same data? Maybe they would give you the synthetic data to demonstrate...

Wait why don't you like the synthetic data comparisons? That seemed quite useful to me but you know the literature better than I do.

Summary Table of the massive defunding of biomedical research in the United States
www.nytimes.com/interactive/...

Collider bias...
And agreed, explaining non-causal associations due to collider bias to the general public (or to scientists) can be challenging.

The prediction model will find factors that increase risk of cognitive impairment but do not affect accumulation of progressive pathology will be associated w/ lower transition to dementia. Key is distinguishing b/w the prediction model (useful) and the causal model (also useful, but different).

You mean b/c of no discussion of causality? Indeed. But they did up front say they were aiming to develop a predictive model, not a causal model. Much harder to evaluate causality conditional on *having* MCI. (An important question, but hard)

Predicting 5‐year dementia conversion in veterans with mild cognitive impairment INTRODUCTION Identifying mild cognitive impairment (MCI) patients at risk for dementia could facilitate early interventions. Using electronic health records (EHRs), we developed a model to predict M.....

Just read this nice paper on predicting conversion from MCI to dementia in real world data (from the VA and replicated outside the VA). Great illustration of using synthetic data. From #VA in Phoenix w/ #NSF funding #EpiSky #EndAlz #RealWorldData
alz-journals.onlinelibrary.wiley.com/doi/full/10....

Dear BMJ and Manuscript Central- I just spent half the time I had allocated to review the manuscript trying to log into your system and complete the required form with an opaque website that kept erasing my selections. Please update it. It is so irritating. @episky.bsky.social @bmj.com

"When we cut these programs, we're not trimming fat — we're cutting into the muscle that keeps America strong and competitive." Hailey Robertson re NSF graduate research training budget cuts.

Also strong kudos to A4 trial investigators for sharing their data! Lots of challenges in the trial but sharing the data shows real commitment to advancing towards a cure.

Strong call for better evaluation of amyloid as a surrogate marker from Sarah Ackley-she shows how to do it with individual level trial data. @fda.gov should require release of *all* individual level data from the anti-amyloid RCTs to evaluate the surrogate. Could do it today. #MELODEM2025 #EndAlz

But found *no* evidence of reverse causation. If anything the reverse: as people have very early cognitive changes, their family and friends may begin to offer more support. This is great news as it indirectly supports the potential benefits of social ties as preventing/delaying Alz dementia.

We all know that strong social support and connections is associated with lower risk of dementia later, but is it causal? Maybe people with subtle cognitive changes of early Alz withdraw from social life. This paper tried to evaluate reverse causation from genetic risk of AD to social connections.

The Association of Alzheimer’s Disease Genetic Risk with Social Connectedness in Middle- And Older-Ages Abstract. Observational evidence suggests that social connectedness protects against Alzheimer’s Disease (AD), but reverse causality has not been ruled out

Congrats to Scott Zimmerman, Jackie Torres & colleages on TIME-AD (www.time-ad.org) on their new pub:
"The Association of Alzheimer’s Disease Genetic Risk with Social Connectedness in Middle- And Older-Ages" doi.org/10.1093/aje/... #EpiSky #EndAlz