Lightnews — Scholar-powered news

Maria Glymour @mariaglymour.bsky.social · 6d

New @melodem.bsky.social working group presentation from Dr. Brian Downer at @utmbscoa.bsky.social UTMB on education trends in Mexico across birth cohorts and cognitive aging, using data from MHAS.
melodem.org
#EpiSky #DataSky #Demography #Education #SDOH

melodem.org

Maria Glymour @mariaglymour.bsky.social · 12d

One of the fun things about launching @timead.bsky.social is the amazing early career folks. We started a "scientist spotlight" website section. The 3 posted are just our first few - really gratifying to see the work of Master's & PhD students, and post-docs -more spotlights coming!

bit.ly/3VCa84g

1 3

Maria Glymour @mariaglymour.bsky.social · 16d

I wrote a LinkedIn post on advice for incoming Master's students in public health and epidemiology. tl;dr: do hard things, get help when needed, make friends and thank people, and have fun - it's amazing. #EpiSky #DataSciSky #PublicHealth #GradSchool
Full article: bit.ly/4neVYlE

Picture of feet on the edge of a diving board about to jump off backwards, with blue-painted toenails and a tattoo just above the ankle.

1 7

Maria Glymour @mariaglymour.bsky.social · 21d

Discussion led by Sam Schildroth www.bu.edu/sph/profile/...

1

Maria Glymour @mariaglymour.bsky.social · 21d

She found failure to account for healthy live births as *competing events* led to overestimation of risks...and often biased effect outcomes. She cites Bluhmki 2019 which discusses pregnancy as a multi-state outcome.

1 1

Maria Glymour @mariaglymour.bsky.social · 21d

For A-J, non-preeclmaptic live birth was a competing event. For the outcome of SGA live birth, she did KM and also AJ using fetal death as a competing event OR an alternative AJ where fetal death or non-SGA live birth were competing events.

1

Maria Glymour @mariaglymour.bsky.social · 21d

She did a simulation study comparing kaplan-meier vs aalen-johansen estimators for the outcome of composite fetal death-pre-delivery /pre-eclampsia. For K-M, no competing events but censoring at non-preeclamptic live birth or loss to follow-up.

1

Maria Glymour @mariaglymour.bsky.social · 21d

The debate centers on whether we consider competing events such as abortion or healthy early birth as censoring (person assumed still at risk but outcome unobserved) vs competing event (person is no longer at risk) and estimate the direct effect of exposure or the total effect of the exposure.

1

Maria Glymour @mariaglymour.bsky.social · 21d

She found surprising result in Figure S2 (phd students read the supplements...) and (I didn't understand this) but

1

Maria Glymour @mariaglymour.bsky.social · 21d

This was prompted by the Chronic Hypertension and Pregnancy Trial (CHAP) evaluating exposure was hypertension management preeclampsia/placental abruption/preterm birth/fetal or neonatal death as a primary outcome (RR=0.82). Primary safety outcome was small for gestational age (RR=1.04).

1

Maria Glymour @mariaglymour.bsky.social · 21d

The @busph.bsky.social PROGRESS Initiative journal club hosting Dr. Chase Latour from UNC Chapel Hill Epi for her paper on healthy births as censoring versus competing events in studies of prenatal medication use.
sites.bu.edu/progressinit...
onlinelibrary-wiley-com.ezproxy.bu.edu/doi/10.1111/...

PROGRESS Initiative

sites.bu.edu

1 2

Maria Glymour @mariaglymour.bsky.social · 22d

Potential paediatric tuberculosis incidence and deaths resulting from interruption in programmes supported by international health aid, 2025–34: a mathematical modelling study - The Lancet Child & Adolescent Health www.thelancet.com/journals/lan...

Potential paediatric tuberculosis incidence and deaths resulting from interruption in programmes supported by international health aid, 2025–34: a mathematical modelling study

Without actions to restore discontinued services, cuts to health aid for tuberculosis and HIV programmes could result in millions of additional childhood tuberculosis cases and deaths in low-income an...

www.thelancet.com

1 6

Maria Glymour @mariaglymour.bsky.social · 24d

Such a cool paper. #EpiSky @amjepi.bsky.social #DataScience #Reproducibility Led by @sfackley.bsky.social at
@brown-epi.bsky.social
Research on reproducible research funded by NIH/NIA - thank you!

Triangulation of Innovative Methods to End Alzheimer’s Disease @timead.bsky.social · 26d

🔎 Meta-analysis and triangulation depend on credible science.

📈 TIME-AD’s latest paper, led by Dr. Sarah Ackley, examines 20,000+ abstracts using an LLM to determine whether P value trends reflect changes in p-hacking, publication bias, or statistical power.

Read it: tinyurl.com/3f3nuxma

2 10

Reposted by Maria Glymour

Triangulation of Innovative Methods to End Alzheimer’s Disease @timead.bsky.social · Sep 8

🔁 ICYMI: TIME-AD introduces the AI-Y Checklist, a framework for evaluating the development and deployment of AI tools in population health through an ethical lens.

🔗 Check it out: link.springer.com/article/10.1...

1 1

Maria Glymour @mariaglymour.bsky.social · Sep 7

In this podcast w/ Dr. Salma Abdalla, we discussed dementia epi (including Piller's book "Doctored"), choosing health research topics, non-causal studies and... It was fun, tho without the video you can't appreciate the Flour Bakery product placement. bit.ly/485sC4x #publichealth #EpiSky #EndAlz

1 5

Maria Glymour @mariaglymour.bsky.social · Aug 28

P-hacking or power improvements? Sarah Ackley examined trends in p-value distributions in epi journals and compared to distributions expected under p-hacking vs changes in statistical power over 20 years. #EpiSky #DataScience #MetaSci #Reproducibility

Sarah Ackley @sfackley.bsky.social · Aug 28

Check out our recently published paper at AJE!

academic.oup.com/aje/advance-...

We analyzed 21,332 abstracts (2000-2024) from four major epidemiology journals, extracting 25,288 P values using ChatGPT’s 4o model with few-shot in-context prompting.

#episky #reproducibility #metasci #metascience

Validate User

academic.oup.com

4

Maria Glymour @mariaglymour.bsky.social · Aug 24

Nice defense of diversity from Oklahoma of all places: www.nytimes.com/2025/08/20/o...

Opinion | I’m a Red-State Mayor. Diversity Is Not Reverse Bigotry.

www.nytimes.com

2 7

Maria Glymour @mariaglymour.bsky.social · Aug 15

Compared univariate fidelity, bivariate fidelity, population fidelity, population fidelity (can an ML algorithm distinguish synthetic from real observations), and application fidelity. Note that 2 datasets may share a covariance matrix but be generated from different data-gen proceses.

2

Maria Glymour @mariaglymour.bsky.social · Aug 15

They compared synthpop (cran.r-project.org/web/packages...) and Q-net (www.science.org/doi/10.1126/...)

synthpop: Generating Synthetic Versions of Sensitive Microdata for Statistical Disclosure Control

A tool for producing synthetic versions of microdata containing confidential information so that they are safe to be released to users for exploratory analysis. The key objective of generating synthet...

cran.r-project.org

1 3

Maria Glymour @mariaglymour.bsky.social · Aug 15

Subgroup 2: comprehensively evaluating synthetic datasets: how do we know it's mimicking real data 'well enough' and can we be sure it does NOT identify folks w/ PHI?

1

Maria Glymour @mariaglymour.bsky.social · Aug 15

But they have to do a bunch of iterations to incorporate information on order. They're organizing everything an #OpenScience database.

1 1

Maria Glymour @mariaglymour.bsky.social · Aug 15

Can they apply confirmatory factor analysis? Compared to HRS 206 Core Measurement model. Then used a TOST, ie within a caliper of +/-0.10 SMD can they confirm equivalence between estimates in original and synthpop data within that caliper? Almost all parameters (34/37) fell within caliper.

1 1

Maria Glymour @mariaglymour.bsky.social · Aug 15

But order matters! Cannot start the chain in the middle or at the end. But what order (? causal ordering?) Distributions almost identical.

1

Maria Glymour @mariaglymour.bsky.social · Aug 15

If we can really use synthetic data, it will be incredibly valuable for reproducibility, data sharing.They compared SMD of many risk factors (in their simulation which was based on HCAP) estimated from the synthpop version of HCAP. They found strong consistency, almost always < 0.04 and none sig

1

Maria Glymour @mariaglymour.bsky.social · Aug 15

Final report-outs from #ΨMCA2025 -Workgroup 3 focused on using synthetic data: can we regenerate the features of the original data using either sythpop or Q-net (large science model). #EpiSky #DataSci #EndAlz #NIA

1 3