Logistic model with random effects

Hi everyone,

I ran a logistic mixed model, but I’m not sure about its random effects.

Let’s start by describing the data.
Each patient on this study underwent 4 examinations:

  • 2 examinations with the device A
  • 2 examinations with the device B

For each examination, there are about 10 measurements, and for each measurement there is a flag specifying if a specific feature/pattern is detected.
So, the measurement is nested within the examination, which is nested within the device, which is nested within the patient.

The objective of this model is to test if the occurrence of this feature is higher for one device.

I ran with lme4::glmer function the logistic model by taking:

  • Response: the occurrence of the feature as the response
  • Fixed effect: the device : DEVICE
  • Random effect: a “full nested” nested effect: (1 | PATIENT/DEVICE/EXAMINATION)

Do you think this model is relevant ?

Thanks a lot for your answers.

1 Like

is it a crossover design? ie randomly assigned to sequence A-B, and B-A?

Indeed, patients were randomized to control the effect of the sequence (A-A-B-B versus B-B-A-A) on the region of interest where the measurement is performed.
The 4 examinations were performed one after the other (ie. without any change in patient condition).

it’s sometimes useful to work backwards ie design the summary table for the results that your non-stat colleagues will find useful. Eg maybe the model can be simplified. Each measurement indicates a different feature, and you want estimates presented separately for each feature, or it makes sense to collapse the measurements ie summarise across them? Maybe then you just have an ordinal outcome (sum of flags) with random effects for patients…