0
$\begingroup$

I have a test data of a certain group of patients for their heart rate, blood pressure, BMI, and other demographics. These data were measure before and after they began their fitness sessions. So, we measured before they attended as “pre” data, and after each session they attended as “post” data. We also noted the number of sessions they have attended and used that measurements as post. What I want to do is see how blood pressure, BMI, and heart rate improved for the group of patients by number of sessions they attended. I wanna do it two ways: one keeping sessions as numeric values, and next analysis by categorizing it into four groups - <10, 10-20,20-30,>30. I’m using R.

Data looks like the following: This is only a sample. The actual data consists of a many other variables for 180 patients.

 Pre.hr <- c(70,60,90,99,92,87)
 Post.hr <- c(65,62,78,80,82,91)
 Pre.bp <- c(150,120,140,130,140,152)
 Post.bp <- c(140,120,138,124,132,128)
 sessions.attend <- c(27,5,18,14,12,2)

What kind of tests can I perform to show the relationship of fitness sessions attended with the improvement of the numbers. I want to be able to say, this is the number of sessions that should be taken to see improvement in blood pressure or BMI or something like that.

Improve this question
$\endgroup$
4
  • $\begingroup$ I am facing a similar issue but don't have an answer. I see a problem in the lack of a control group (in this case we could apply a diff-in-diff model or a Syntethic Control method). Without taking into account the number of sessions attended, I was thinking about a simple Regression Discontinuity Design to see whether there is a jump with the beginning of sessions, or a piece-wise regression to check for a change in slope. However, observations are few and we are not controlling for possible external factors. Things would be simpler by taking all the pre- and post- periods together $\endgroup$
    – Federico Tedeschi
    Commented Apr 20, 2018 at 12:29
  • $\begingroup$ @FedericoTedeschi, Thank you for the answer. The observation I show is only a small sample of the actual data. I have observations for 180 patients. $\endgroup$
    – user6401955
    Commented Apr 20, 2018 at 12:51
  • $\begingroup$ It is not clear whether you want to examine ( test) the relationship or just ascertain the magnitude of the relationship ? $\endgroup$
    – user10619
    Commented Apr 20, 2018 at 14:54
  • $\begingroup$ @subashcdavar, I don’t know the distinction. What I want to do is to be able to say the HR improved after this many sessions. One way I thought of doing is categorizing the sessions into 4 groups: <10,10-20,20-30,>30. And then do anova on the difference. Not sure how to interpret the data, or whether that is an good test. Any idea? $\endgroup$
    – user6401955
    Commented Apr 20, 2018 at 21:45

1 Answer 1

Reset to default
0
$\begingroup$

Sorry I misinterpreted then: please disregard my comment. Thus, you have 1 pre and 1 post observation for each patient.

1) To test for an overall effect of sessions: if you had a control group, you could perform difference-in-differences; otherwise, a paired t-test. In both cases, however, you wouldn't be using information on the number of sessions attended (and, in the case of t-test, you would be assuming that only changes due to chance would occurr in case of no sessions).

2) You could use the pre-post difference as your outcome and the number of sessions attended as your predictor variables. However, as you can see here: Is it valid to include a baseline measure as control variable when testing the effect of an independent variable on change scores?
it is preferable to model the post-value controlling for the pre-value and using the number of sessions as your dependent variable. In general I see it as a regression-context (so, where also other possible confounders may be controlled for, as well as non-linearities - like quadratic or interaction effects - introduced).

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.