Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Mixed models repeated measures

    Hi,

    I'm trying to find the best model for answering a repeated measures problem. The research question is basically "do patients who respond to treatment have lower rates of disability?".

    There are two sets of data: One is self-report measures of symptoms taken at pre- and post-treatment. This gives the variable "responds to treatment" which can either be categorical or dichotomous.

    The second set of data is collected from a registry and spans over two years, i.e. one year pre- and one year post-treatment. This gives the variable "disability". This is a dichotomous variable, monthly measurements (Meaning the "time" variable has 26 different times). I've been advised to look into mixed models as these are repeated measures of the same patients. (Note: "disability" is basically working or not, which means it varies over time).

    The problem, as far as I understand it, is that "repeated measures" in these data only applies to one aspect, "disability". The "response" variable is only derived from measurements at treatment, and isn't repeatedly measured, strictly speaking, at any other time. Like so (abbreviated version):
    patient time disability response
    1 1 0 0
    1 2 1 0
    1 3 0 0
    1 4 (treatment) 1 0
    1 5 (post) 0 0
    1 6 0 0
    1 7 0 0
    1 8 etc... 1 0
    2 1 0 1
    2 2 1 1
    2 3 0 1
    2 4 (treatment) 1 1
    2 5 (post) 0 1
    2 6 0 1
    2 7 0 1
    2 8 etc... 1 1
    When restructuring the data from wide to long, the response variable gets copied to all time values, but like I said, it's derived from only the treatment time(s).

    "melogit disability response" or "xtlogit disability response" works fine, but I don't quite trust the results. (I include other variables like gender and age, but I suppose that's beside the point).

    Eyeballing the raw data suggests that "responders" are more disabled in the year before treatment, and less disabled in the year after treatment compared to "non-responders". But how to test this?

    Thank you very much for any help.

    (And apologies for the Very Ugly Formatting, but IT security won't let me install any plugins or add-ons..).
    Last edited by Carl Burke; 22 May 2019, 10:03.
Working...
X