I have paired data pre and post-exposure to a drug. I would like to calculate the percentage change of my metric in response to the drug. I am getting a little confused and I am probably overthinking it. I think I should calculate the change per patient then the average of the individual changes to get the overall change. I would like to go on to see if the calculate p-values for the change.
$\begingroup$
$\endgroup$
1
-
$\begingroup$ The "average percent change" is also called the geometric mean change. To calculate that you would take the ratio of post to pre for each patient, then average that. To get the p-value, you actually have to log transform the response. Then you can do either a paired t test or (better) ANCOVA. Search "pre post designs" for more. $\endgroup$– AdamOCommented Mar 17, 2023 at 13:41
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Yes. You are probably overthinking it, although there could be some subtleties
Average percent change: Calculate the percent change for each individual, then average the results. This is probably what you are looking for
Percent change of the average: Calculate the "before" and "after" averages, then get the percent change.
-
$\begingroup$ This is thin, which would you say is right? Maybe a simple 4 observation example would illustrate. $\endgroup$– AdamOCommented Mar 17, 2023 at 13:42
-
$\begingroup$ @AdamO neither is right or wrong they're just calculating different things. For instance if sample before the drug is $5,5,1,1$ and sample after the drug is $15,10,3,1$, the first method would give you a $+125\%$ average change (the average of $200\%, 100\%, 200\%, 0\%$ while the second one would say that the average has changed by $141,67\%$ (the average has changed from $3$ to $7.25$) $\endgroup$– DavidCommented Mar 17, 2023 at 16:02