Martin Posch

Did you know the FDA has gone full Bayesian?

Better take in this blogpost, including intro to the context of guidance documents

The guidance is quite good, but as stated in the blogpost, it leaves out where the FDA finds Bayesian methods acceptable.

statmodeling.stat.columbia.edu/2026/01/15/f...

Graph Based, Adaptive, Multiarm, Multiple Endpoint, Two‐Stage Designs The graph-based approach to multiple testing is an intuitive method that enables a study team to represent clearly, through a directed graph, its priorities for hierarchical testing of multiple hypot...

New open-access paper: we discuss graph-based adaptive multiarm, multiple-endpoint two-stage designs that allow early stopping, sample size re-estimation & dropping treatments/endpoints while preserving strong FWER control.

onlinelibrary.wiley.com/doi/10.1002/...

#multipletesting #adaptivedesign

#ISCB46 very nice talk by Michael Kammer on COVID-19 prognostic models, which involved a ‘deep dive’ into the pond life of modelling.
Bottom line: “If looking for reliable models, don’t hold your breath.”

Pleased to report on two papers on #ANVOVA with @poschm.bsky.social and Franz König
The first lnkd.in/eKmjiQdk
looks at median stratification for a single covariate.
The second lnkd.in/e3DH96G8
considers adjustment for many covariates.
Variance inflation factors are key.

We invite applications for a 1-year postdoctoral/postgraduate researcher position at the Institute of Medical Statistics, MedUni Vienna working in a collaborative project with regulatory authorities on causal inference for treatment effect estimation in RCTs cloudius.meduniwien.ac.at/index.php/s/...

NEWS New article "The validation of prediction models deserves more recognition" published in BMC Medicine 2025.03.24 Begüm Irmak Ön and Ulrich M…

Welcome to the official Bluesky account of SHARE-CTD!
We're a European research network dedicated to enhancing clinical trial data sharing to advance medical research and the betterment of society.
Learn more: share-ctd.eu
#SHARECTD #OpenScience #DataSharing

Congratulations to the #ENAR2025 Student Paper Award Winners! Checkout these eight winners (more to come). Catch their presentations at the 2025 ENAR Spring Conference. #ENAR #StudentAwards #NewOrleans

A revised version of the paper with @poschm.bsky.social and Franz Koenig on stratification and covariate adjustment in clinical trials is now up arxiv.org/pdf/2408.06760
We thank Robin Ristl and @timpmorris.bsky.social for helpful comments.

This past month has seen the publication of a couple of new helpful papers providing further insights into the thorny issue of #MultipleTesting in trials 1/7
#MethodologyMonday #112

A Comparison of Statistical Methods for Time‐To‐Event Analyses in Randomized Controlled Trials Under Non‐Proportional Hazards While well-established methods for time-to-event data are available when the proportional hazards assumption holds, there is no consensus on the best inferential approach under non-proportional hazar...

doi.org/10.1002/sim....

Our paper "A Comparison of Statistical Methods for Time-To-Event Analyses in Randomized Controlled Trials Under Non-Proportional Hazards" got published today 🎉

We describe commonly used methods, and compare their performance in a simulation study across different scenarios.

NEWS First schooling at Wissenschaftszentrum Schloss Reisensburg 2025.02.01 The 1st schooling of the SHARE-CTD project took place in person with internatio…

Some news of SHARE-CTD. Good news.

How should the advancement of large language models affect the practice of science? | PNAS Large language models (LLMs) are being increasingly incorporated into scientific workflows. However, we have yet to fully grasp the implications of...

4 perspectives on the use of LLM in the practice of science www.pnas.org/doi/10.1073/... . Should LLM be seen as collaborators?

Graph Based, Adaptive, Multi Arm, Multiple Endpoint, Two Stage Design The graph based approach to multiple testing is an intuitive method that enables a study team to represent clearly, through a directed graph, its priorities for hierarchical testing of multiple hypoth...

Together with Cyrus Mehta and Ajoy Mukhopadhyay, we present a unified framework for graph-based adaptive designs in multi-arm, multi-endpoint, two-stage studies. This approach integrates multiple strategies to enable flexible confirmatory trial designs. Preprint: arxiv.org/abs/2501.03197

A Conversation With Amy Racine‐Poon Professor Dr. Amy Racine-Poon is best known for her interdisciplinary contributions as an applied Bayesian statistician in the pharmaceutical industry and healthcare. She was born in Hong Kong and ob...

A beautifully written interview by Alex Sverdlov featuring my colleague Amy Racine-Poon, published in the International Statistical Review.

doi.org/10.1111/insr...

TRIPOD-LLM is out! Check out our consensus guidelines for reporting #LLM research in biomedicine. TRIPOD-LLM is intended to be a living guideline to keep up with the rapid advances in LLMs. Kudos to lead author
Dr. Jack Gallifant

Priors and decision thresholds in phase 2 and phase 3 randomized controlled trials evaluating drug efficacy using Bayesian methods: a systematic review - @jclinepi.bsky.social www.jclinepi.com/article/S089...

"It's easy to get a false positive risk of 40% withOUT any cheating at all"

Amazing collective problem solving in ants:

Great essay by David Spiegelhalter!

Sample size for developing a prediction model with a binary outcome: targeting precise individual risk estimates to improve clinical decisions and fairness When developing a clinical prediction model, the sample size of the development dataset is a key consideration. Small sample sizes lead to greater concerns of overfitting, instability, poor performanc...

Developing a prediction model?

Ever wondered how to target sample size to improve model fairness and precision of risk estimates?

Check out our new pre-print & software package pmstabilityss
arxiv.org/abs/2407.09293

Great to see this R package: stephenrho.github.io/pminternal/

Implements the (in)stability plots and checks recommended by myself & @gscollins.bsky.social after developing a prediction model

Our paper is here: onlinelibrary.wiley.com/doi/full/10....

The Journal of Clinical Epidemiology is, as of today, present at Bluesky 🎉

And, only here, you can have a sneak preview of our new journal and website design.

But of course, our contents remains key, with many thanks to our authors, reviewers, editorial board and editorial team

Last year I published research proving (yes, proving 😉 ) that new episodes of Doctor Who on Christmas Day reduce death rates in UK

Since then, the BBC have shown brand new episodes on Christmas Day! Including this year!

Who said my research has no impact? ❤️❤️ #DoctorWho

www.bmj.com/content/383/...

YouTube video by Richard_D_Riley IPD Meta-Analysis: Two-stage or Not Two-Stage? That is the question ...

Christmas 2024 Gift no. 4

NEW VIDEO:
"IPD Meta-Analysis: Two-stage or Not Two-Stage?
That is the question..."

youtu.be/WFSAfouhB-8

- guidance for when to choose a one-stage rather than a two-stage approach, & vice versa.
- ideas showcased using real examples (no equations)

I hope useful!

YouTube video by Bennett Institute for Applied Data Science OpenSAFELY in a nutshell

If you've been wondering where I went: my team in Oxford (60 of us!) build beautiful tools for NHS data. We help researchers work on the whole nation's GP records while protecting - provably - everyone's privacy. It's big potatoes!

www.youtube.com/watch?v=GRjR...

Efficiency of nonparametric superiority tests based on restricted mean survival time versus the log-rank test under proportional hazards Background: For RCTs with time-to-event endpoints, proportional hazard (PH) models are typically used to estimate treatment effects and logrank tests are commonly used for hypothesis testing. There is...

Hot off the preprint server:
‘Efficiency of nonparametric superiority tests based on restricted mean survival time versus the log-rank test under proportional hazards’
Short thread 👇
1/
arxiv.org/abs/2412.06442

Title slide: Will your code run again? Tips for making code reproducible in R

If you missed my talk but still want to learn how to make your R code more reproducible, my slides are here 🙂:

daxkellie.quarto.pub/will-your-co...

All the links to packages and resources I mentioned are there, so hopefully this can be a nice reference, too!

#ESAus2024 #rstats #quartopub 🧪🌏

‘A truly remarkable breakthrough’: Google’s new quantum chip achieves accuracy milestone Error-correction feat shows quantum computers will get more accurate as they grow larger.

It’s time to prepare for the next revolution in computing www.nature.com/articles/d41...

'...suppose that Sir R.A. Fisher ... had not taken over from ordinary English such evocative words as "sufficient," "efficient," & "consistent" & made them into precisely defined terms of statistical theory...Would his work have had the same smashing influence that it did?' Kruskal, 1978