My question contains a few variations; I apologize for this in advance.
I have two groups, and I have four outcome measures. These outcome measures are assessments of pain, function, depression level, and quality of life. I want to analyze the difference between the groups on the outcomes before and after a treatment without repeated measures. Normally, I would perform Mann-Whitney U test (the assumptions are not met) separately for each outcome measures. However, I read some articles, and I guess I am wrong. I have read a few posts and answers in CV related my question (here, here, here, and here), but they do not fully answer my questions.
My questions are as follows:
- If I am correct about analyzing the outcomes separately with Mann-Whitney U test, should I conduct an adjustment obtained p-values (e.g., Bonferroni with
p.adjustin R)? - In this, this, and this articles, the authors indicate that when analyzing of multiple outcomes in same individuals, it should be performed the simultaneous analysis of the outcomes using multivariate methods because these outcomes tend to be correlated. Normally, my outcome measures are independent, but for example, pain may affect function, or function may affect quality of life. Should I perform a correlation analysis on my outcomes, and if these are correlated with each other, should I perform a multivariate analysis? Or, should I analyze them with a multivariate analysis directly, assuming the outcomes can be correlated or can affect each other? If so, what analysis should I use, and should I adjust the p-value?
- Instead of two different groups, if I compare before and after treatment results in a single population (i.e., there is no group), would the answers to the questions above change (i.e., in this situation, Wilcoxon vs multivariate analysis)?
I am aware that it's a long question; sorry for this.
