Logistic regression analysis for two treatment groups and two time points?

[Biochemist]

Hello,

I know how to perform an ordinary binomial logistic regression with two outcomes and several potential predictor variables. But my setup is a bit more complex.

There are two treatment groups in the experiment. For each treatment group, there is a binary outcome based on a response variable assessing the effectiveness of the treatment: treatment effective - yes or no.
In addition, there are two follow-up time points for each treatment: after 2 weeks and after 4 weeks.

So, in the end, there is:

• treatment A after 2 weeks: effective yes or no

• treatment A after 4 weeks: effective yes or no

• treatment B after 2 weeks: effective yes or no

• treatment B after 4 weeks: effective yes or no

How would I handle such an experimental setup with logistic regression in jamovi in order to analyze for the potential predictor variables?

[mcfanda@gmail.com]

You can use generalized mixed models, selecting a logistic model.

[Biochemist]

Hi, thanks for the answer. What would I select in jamovi for that? I assumed I would have to choose a type of "Regression" analysis, for example "Logistic Regression - N Outcomes/Multinomial". But how would I define or code the 8 different outcomes?

Or would I find the appropriate analysis type under "ANOVA"? But I could not find an option for selecting a logistic model there.

[Biochemist]

I think this publication here does a similar type of analysis as I would like to do one (in the statistical sense).

"Mixed effects logistic regression analysis of blood pressure among Ghanaians and associated risk factors"
https://www.nature.com/articles/s41598-023-34478-0

The response variable (blood pressure) is binary (high or normal). The response variable was determined across three measurement periods (in my case I only have two). Here, they, wanted to identify risk factors for high blood pressure and for that they used a mixed effects logistic regression model.

How can I do this mixed effects logistic regression model in jamovi?

[mcfanda@gmail.com]

You need to install GAMLj3 module. There you find generalized mixed models

[Biochemist]

Thank you. I noticed that the GAMLj3 is only available in the library with the most current version of jamovi and not the so-called solid version, which I had installed until now.

[mcfanda@gmail.com]

Current version is pretty ok, just go for it.

[Biochemist]

Alright, good to know.

Would it be wrong to analyze the two different treatments as two independent experiments and hence perform two separate logistic regression analyses?

Because what I want to know is which of the several potential predictor variables provide(s) a strong and statistically significant prediction of the response variable (treatment effective or not) after 2 weeks and/or 4 weeks. I am not interested in comparing treatment A and treatment B.

[Biochemist]

The research question is - more simply expressed: What predicts an effective response to the treatment at time point 1 and at time point 2, respectively?

[reason180]

If you're not interested in comparing treatment group A to treatment group B (and if you think your audience won't be interested in that comparison either), then you could simply code the data so that there's no distinction between A and B, and do just one analysis. (Doing two separate analysis halves the N and thus tends to drastically reduce the statistical power.)