Appropriate statistical model for prepost design with covariates?

02 February 2024 6 7K Report

Hi all,

I'm struggling in determining the most appropriate statistical model to use.

The experimental design is a single group pre-post design.

dependent variable=connectivity strength (functional MRI), dependent variable=anxiety total score, other time-invariant variables (age/sex).

I built in r the following model: lmer( y ~ condition + age + sex + anxiety + (1|subj), data=dat, REML=TRUE), where y = connectivity strength, condition=pre/post and anxiety=anxiety total score.

Is the model appropriate?

Best regards

A.G.

Chuck A Arize

Mixed-effects model: This model treats the pretest and posttests as a factor.

Chuck A Arize

Check this link.

https://www.theanalysisfactor.com/pre-post-data-repeated-measures/

Simone Di Plinio

Hi A.G.,

Your approach to modeling your fMRI data considering the change in connectivity strength as a function of a pre-post condition (intervention or time), age, sex, and anxiety score is a good starting point.

However, there are a few consideration:

In your model, condition represents the pre/post measurement. This is correct for capturing the main effect of time (pre vs. post). Ensure that condition is coded as a factor. If it's coded as a numeric variable, R might interpret it as a continuous variable, which wouldn't be appropriate in this case.

You have included random intercepts for subjects ((1|subj)), which accounts for the repeated measures within subjects. This is appropriate as it allows for individual baseline differences in connectivity strength. However, consider whether random slopes should also be included. For example, if you expect that the change from pre to post (time effect) might vary across individuals, you might want to include random slopes for condition by subject. This would be specified as (condition|subj).

Depending on your hypothesis, you might be interested in whether the effect of the intervention (pre vs. post) on connectivity strength is moderated by anxiety, age, or sex. If so, you should include interaction terms in your model (e.g., condition*anxiety, condition*age, condition*sex). This will allow you to assess if the change from pre to post differs across levels of anxiety, age, or sex.

Considering these points, you can explore the following additional models:

y ~ condition + anxiety + age + sex + (condition|subj)

y ~ condition * anxiety + age + sex + (condition|subj)

These models includes interaction between condition and anxiety (the second), and random slopes for condition by subject (both), assuming that's relevant for your hypothesis.

Exploring more models allow you to better understand your data. Don't forget to: run diagnostics and model assessment to help you decide on which model you should focus; to report all the models used when writing results; to draw interpretations based on all the models.

Hope this helps!

Simone

Alberto Galimberti

Thank you very much and how would you model this experimental design using GLM? Paired sample t-test with covariates?

Simone Di Plinio

Models such as "y ~ condition + anxiety + age + sex + (condition|subj)" are linear mixed effect models. You can refer as "age" and "anxiety" as continuous variables or covariates, and as "condition" and "sex" as categorical variables. "subj" is the random grouping factor and "Condition|subj" implies both a random intercept and a random slope for condition at the subject level. The type and quality of tests depends on your specific aims. You may want to take a look at the documentation of lme4 or lmer (https://cran.r-project.org/web/packages/lme4/vignettes/lmer.pdf)

Alberto Galimberti

Thank you again

Possible to infer a "treatment effect" from a design without control?

How can I prepare virus for a TEM or SEM imaging?

Can you connect an HPLC to a Mass Spec only at a certain time point?

Handling Missing Data and Building a Predictive Model with Incomplete Information ?

Is it possible to use the Fused Deposition Modeling (FDM) to additively manufacture interconnected porous structure generation of >100-200 micrometer?

How to define an anisotropic material with asymmetric elastic compliance/stiffness matrix in ANSYS APDL?

How can I apply boundary conditions in an orthotropic steel deck numerical model using ABAQUS software?

How to normalize and take the significance of the MTT OD values with 3 replicates for the same cell-line?

Can you suggest reliable sources defining "3D mesh" and "3D city models"?

Please explain how the plastic input value should be considered from the true stress-strain curve for the bilinear elastoplastic material model ?

How can major oxides determined by ICP-MS/ES be normalized to 100% volatile free? How does this connect with LOI?