Handling Repeated Measures Data

[Tiara Rosemond]

Hi all,

I have a dataset where patients' blood pressure is taken at each visit. The visits fall either before the intervention period (pre) or after the intervention period (post). Each visit is listed in the dataset as a separate observation. I need to look at pre-post changes, but I am not sure how to approach this - should I create a pre/post set of variables (how should I do that) or can I keep the blood pressure variables as-is and use some variable to indicate which values are pre or post? Here's an example of the data:

ID	Encounter#	SBP	DBP	Pre/Post Intervention
1	1	131	85	Pre
1	2	142	85	Pre
1	3	125	75	Post
1	4	120	70	Post

:

Thanks!

[Erik Ruzek]

Does it matter to you that you have multiple pre visits and post visits per patient? In other words, are you interested in using all that data in your model?

[Tiara Rosemond]

Does it matter to you that you have multiple pre visits and post visits per patient? In other words, are you interested in using all that data in your model?

As of now, I will use the most recent value for each phase (pre/post), but need to be prepared to include all values if asked.

[Erik Ruzek]

See https://www.statalist.org/forums/for...-in-each-group

[Leonardo Guizzetti]

I don't see a justification for selecting just some of your data, if those pre/post encounters occur at about the same nominal time. You'll need to justify why and how you arrived at your selection rule.

What you do for analysis depends on your experimental design, so you should tell us about that.

Points you may consider in order of simplest (discards the most data) to most complexity (discards the least data):
1) select just one of each pre/post data (as you asked about in #1)
2) average pre/post data to reduce multiple measures to one value per time point. This
3) use a mixed model to account for multiple visits/repeated measures for each patient for each time point.

[Erik Ruzek]

I agree with Leonardo Guizzetti. These are all the options I considered, but then I thought better of it and just asked the Tiara Rosemond what she wanted to do.

[Tiara Rosemond]

I don't see a justification for selecting just some of your data, if those pre/post encounters occur at about the same nominal time. You'll need to justify why and how you arrived at your selection rule.

What you do for analysis depends on your experimental design, so you should tell us about that.

Points you may consider in order of simplest (discards the most data) to most complexity (discards the least data):
1) select just one of each pre/post data (as you asked about in #1)
2) average pre/post data to reduce multiple measures to one value per time point. This
3) use a mixed model to account for multiple visits/repeated measures for each patient for each time point.

So, we are only doing these analyses for the treatment group. We are awaiting more details from the partner hospital about a possible comparison group. I definitely lean towards option #3, using a mixed model. Since the only time indicator is the variable designating pre/post intervention, should I just include in that "time" variable in the model?

Thanks!

[Carlo Lazzaro]

Tiara:
why not considering a panel data regression with fixed effect, using a notion of mean BP as a continuous regressand (As far as I can recall, the mean BP formula is: 2/3 of SBP + 1/3DBP)?

xtset patientid
xtreg bp i.prepost <otherpredictors> <controls>, fe vce(cluster patientid)

As you have repeated time values within the same panel due to -prepost-, you should avoid to include the -timevar- in -xtset-.