Dear Statalist,

In the repeated-measures mixed-effects linear models, the fixed effects often includes main terms, time, and interaction with time. But when adjusting for confounders, do we need to add time interaction with confounders?

Here is the example I found from published paper (Janmaat et al, 2019 Nephrol Dial Transplant):

eGFR_epi are the eGFR values over time based on the CKD-EPI equation. The variable Time represents the time between the index date and each subsequent eGFR value. categoricalDBP represents the dichotomized diastolic blood pressure ≥80 mmHg and <80 mmHg at baseline. For the fixed effects, They included DBP at baseline and the time and the interaction between DBP at baseline and the time.
When defining the adjusted model for the association between categorical baseline DBP and subsequent kidney function decline, all categorical confounders should be placed behind the BY in the model. Furthermore, in addition to the baseline confounders, the interaction between baseline confounder and time is added in the fixed effects part. (let's suppose the confounders are baseline values, and remains same across the Time with individual studynumber).

Model A
mixed eGFR_epi categoricalDBP##Time sex##Time c.age##Time race##Time c.alcohol##Time || studynumber:

But, if we adjust only the confounders without additional interaction terms with time, the results must be different.

Model B
mixed eGFR_epi categoricalDBP##Time sex c.age race c.alcohol || studynumber:

My questions are:
1. Do you suggest add confounder*Time interaction when adjusting for confounders in repeated-measures mixed-effects linear models?
2. If so, how to interpret the confounder*Time interaction
3. How to explain to reviewers why we add this confounder*Time interaction?

Cheers!
Will