Matthew J. Smith

🚨New paper🚨 Tutorial on causal inference in Stata, R, and Python bit.ly/3pMAKAK Code on GitHub! bit.ly/3nE3cCg with @Mohamma70696197 @WATZILEI @camaringe @PausalZ @nondogmatist @LeyClem @aurelien_belot @icon_lshtm @epi_twit @statstwitbot @CausalInferBot

📢 New paper in Statistics in Medicine! We compare TMLE vs CV-TMLE across tough simulation settings: ✔️ Misspecification ✔️ Limited overlap ✔️ High-dimensional data ✔️ Super Learner + cross-fitting Read it here: onlinelibrary.wiley.com/doi/10.1002/si… @epi_twit @statstwitbot

🚨 New #SystematicReview in @AnnalsOfEpi with @rachaelvp, @watzilei, @camaringe 🎉 How has Targeted Maximum Likelihood Estimation (#TMLE) been applied in public health and epidemiological studies?👉 bit.ly/3NmbVWM TL;DR #Thread 🧵 (@UKRI_News @The_MRC @NIHRresearch)

We argue that dissemination of #TMLE relies on 5 key factors: 1. Software availability (doi.org/10.1002/sim.92…) 🧑‍💻 2. Accessibility of available material 📄🔓 3. Number of experts in #TMLE 🤓🤓🤓 4. Teaching 🏫🧑‍🎓 5. Collaborations 🧑‍🎓🤝🧑‍💼🤝🧑‍💻🤝🧑‍🔬🤝🧑‍⚕️ ...... 💡...... 🤯🤯🤯

#TMLE has been methodologically adapted to multiple data settings: 1. Small sample size 2. (Nested) Case-control studies 3. Collaborative TMLE 4. Longitudinal data 5. Sequential randomised trials 6. Mediation analysis 7. Meta-analysis 8. Cross-validated TMLE 9. One-step TMLE etc

New #SystematicReview available in @AnnalsOfEpi - by @MJTSmith, @rachaelvp, @WATZILEI & @camaringe 👏 The team investigated the application of #TMLE in public health and epidemiological studies👇 sciencedirect.com/science/articl…

@rabois @MaartenvSmeden For a confounder with low prevalence (e.g., 1%), if you only sample the other 99% of the population, then you should get an unbiased estimate? So as you increase n from 50% to 99%, you decrease the bias of the confounder that you never captured in the first place…? Hmmmm….

@MaartenvSmeden you guys are so ridiculous. as large n (actually day) approaches y (total population) the risk of bias declines non-linearly to be negligible > superior to all the flaws of random control.

Still no equation

@kenkoonwong @epi_twit @statstwitbot @CausalInferBot @WATZILEI @camaringe Of course 🙂

@MJTSmith @epi_twit @statstwitbot @CausalInferBot @WATZILEI @camaringe Thanks for sharing! Will read. Will you as the first author of the tutorial paper be available for questions? 😁

If you’ve considered using Targeted Maximum Likelihood Estimation in the past for an observational study and decided not to use it, what was the main reason for not using it? @epi_twit @statstwitbot @CausalInferBot #epitwitter #statstwitter #causalinference #TMLE

@kenkoonwong @epi_twit @statstwitbot @CausalInferBot Certainly! Below are two tutorials. The first offers a comparison of the common estimators (including TMLE). The second is a tutorial of TMLE for a binary treatment. onlinelibrary.wiley.com/doi/10.1002/si… and onlinelibrary.wiley.com/doi/10.1002/si… @WATZILEI @camaringe

@MJTSmith @epi_twit @statstwitbot @CausalInferBot Have you come across a good tutorial and explanation of TMLE?

G-formula for causal inference via multiple imputation thestatsgeek.com/2023/01/31/g-f…

📢 New publication by @icon_lshtm researchers highlights importance of diagnostic route in survival of people with #lymphoma and comorbidities. ➡️mdpi.com/1889940 Further research is vital to ensure timely diagnoses to reduce the #inequalities in #cancer survival.

Save the dates! EuroCIM 2023 will take place in Oslo, April 19-21 (with short courses April 18). Keep an eye on eurocim.org for more information.

Coming soon...

@TheStatsGeek @LSHTM Welcome back!

Super excited to be (re)joining @LSHTM tomorrow!

@PausalZ Or Pauled logistic regression

@tmorris_mrc @kerryhood Do you mean a Mars bar flavoured ice cream? Or an ice cream flavoured Mars bar?