Resources for Mixed-Effect Models in Medical Contexts

My knowledge of mixed-effect models is somewhat limited to non-medical-related statistical books, such as “Statistical Rethinking” by Richard McElreath.

I want to delve into this topic but focusing on medical-related examples. Could anyone recommend textbooks, video lectures, etc… that would fit this criterion?


1 Like

i’ve always liked Brown and Prescott’s “applied mixed models in medicine” but they are sas users and i think all the code is in sas. They’ve been running a course on the topic for nearly 20 years, in Edinburgh, i think Helen Brown still runs it: The PSI have a course in october on Repeated Measured and Mixed Models but it’s GBP465 for non members PSI Training Course: Repeated Measured and Mixed Models. Im not sure about free video tutorials


I read quite a few books and articles on mixed effects regression but none of them really worked for me until I read ‘Applied Longitudinal Data Analysis: Modeling Change and Event Occurrence’ by Singer and Willett. It is very well written so is intelligible to even those with a modest stats background like myself. They come from a frequentist perspective but the book has been translated into Bayesian R package brms code here.


Great suggestion. I enjoy Solomon Kurz’s book translations.

1 Like

Hi Arthur,
I have also been struggling with this. I am a MD. I believe this 2 excellent resources are just great papers to understand the basics of LMEM:

  1. Brown VA: An Introduction to Linear Mixed-Effects Modeling in R. Advances in Methods and Practices in Psychological Science 4:2515245920960351, 2021.

  2. Gałecki AT, Burzykowski T: Linear mixed-effects models using R : a step-by-step approach. New York, NY: Springer, 2013


I have a summary table comparing various longitudinal data methods in the longitudinal modeling chapter of rms. To oversimplify:

  • use fully specified models (serial correlation models or mixed effects model as opposed to GEE)
  • if you have clustering in the data (clinical sites, country/state/county) think of random effects
  • if all you have is serial data collected within each participant, don’t think of random effects as the most logical way to model the data (increased computational problems, unlikely to fit the correlation pattern) but think more of serial correlation models
  • the most universal and flexible serial correlation model is a Markov process (works equally on binary, multinomial, ordinal, and continuous Y) but for continuous Y serial correlation models (e.g., generalized least squares aka growth curve analysis) are worth a look
  • never use last observation carried forward