Statistical validity of using different "time" variables in longitudinal data

[Linda Zhang]

Hello,

I apologise in advance if the format I'm presenting here is incorrect.

I am using -mixed- to analyse atrophy rates of brain volumes between two genotypes in a longitudinal study. I have subjects of varying ages at the baseline visit. The model I am currently using is:

mixed rtot_long c.z_age##genotype sex education z_tiv conv || subjid: z_age

where rtot_long is the volume of interest, z_age is the standardised age of the participant at each visit, and genotype is dummy coded as 0 or 1. I have also tried this using visit (1-8), with baseline age as a fixed covariate, as below:

mixed rtot_long c.visit##genotype sex education z_tiv conv ageatbaseline || subjid: visit

The results are slightly different, but I don't know which one is more "valid". I understand this is probably a theoretical question, but any help would be greatly appreciated.

I am using Stata/IC 15.1 on a Windows 10 machine.

[Clyde Schechter]

It depends. If you are trying to observe a purely biologic process and nothing that happens at the visits is expected to alter the outcome variable, then age would be a more natural measure of time. But if the visits involve some kind of intervention that may affect the outcome, then visit number, representing something like a "dose" of intervention, might be more appropriate.

[Linda Zhang]

Thank you, Clyde!