Subgroup based on baseline of dependend variable?

Tess

New Member
#1
Hello everybody,

I am currently working on a study with heart patients. I have no data yet but am trying to write up my methodology and am currently thinking about how to do the statistics to evaluate my hypothesis. I am a bloody beginner in regards to statistic and would be thankful for any help! So here we go:

The Study Design:

I am performing two tests assessing the functional mobility in patients (outcome measure is the time the patients need to perform a certain task). I measure at baseline before they receive surgery and at four time points after their surgery (day 1, 3 and 5 after surgery, as well as hospital discharge). I want to assess the two tests seperately, because they tests slightly different parts of functional mobility but using the same statistical procedure.

I want to assess two research questions:
1. How does the procedure affect the functional mobility of elderly patients for the duration of their hospital stay and at discharge?
2. To what extent does a high / low functional status pre-surgery impact functional mobility post-surgery in elderly patients?

To answer the first question I wanted to perform a one-way repeated measures ANOVA. I was thinking the functional mobility (or rather the time they need for the test) is the dependend variable and time (the different measurement time points) is the independend variable. But I would appreciate if anybody could confirm this for me.

For the second question my first impulse was to split my sample into two equal sized groups utilizing the median. And then looking at the interaction of time*group. But I believe I have read that splitting at the median diminishes the power because of a regression towards the median (or something similar). I understood it that both groups then contain individuals that are fairly close to the median and are therefore very similar in their baseline (functional mobility) status resulting in probably very similar results of change over time and therefore reducing the interaction. But I could be totally wrong.
Secondly, I am unsure wether it is "okay" to use the baseline status of the dependend variable to differ between the two groups, since the baseline status is also used in calculating the change over time. Somehow I have in my head that I can not utilize variables twice like that. So I am pretty much asking: Am I allowed to use the baseline value of my dependend variable to split my sample into two equal sized groups at the median and then test the interaction of time*group?

Or am I not aware of a test that answers my second research question? I believe this kind of study were one wants to assess wether the baseline status of a patient has an impact on the treatment results is farily common. But all I ever find is studies that use a different variable to differ between subgroups (such as gender or a risk factor).

I appreciate any hints greatly! Thank you so much.

Kind regards,
Tess