Comparing the median difference between two groups

[MD Johan]

Hi,

I have a problem regarding the analysis of medians between two groups . I have data from a study (RCT) comparing a treatment group with a control group. The variable is ordinal data and is collected at baseline and at a follow-up. Since the data is ordinal I have chosen to present median values (with interquartile range).

The difference from baseline to follow-up is calculated for each group. This difference is then compared between the groups with Mann-Whitney U (ranksum). Ranksum only gives a p-value. Is it possible to get the difference with a 95% CI using another test?

I have tried using bpdifmed, but then I get a 95% CI that shows a significant change when the ranksum test doesn't.

Thanks for the help!

[Joseph Coveney]

I have tried using bpdifmed, but then I get a 95% CI that shows a significant change when the ranksum test doesn't.

ranksum doesn't test for equality of medians. Did you mean median?

Is there a reason why you don't fit a regression model to your data? There are estimation commands specifically for ordered-categorical outcomes, and a subset of these can handle repeated measurements.

[ericmelse]

Note also the Stata Journal paper that I referred to:

Dear Vivian,

Possibly, this paper offers good advice to compare and examine median values between group categories:
http://www.stata-journal.com/article...article=st0253

Best,
Eric

[Joseph Coveney]

The focus on nonparametric tests of change scores of before-and-after ordered categorical data strikes me as a little misplaced. I don't know what the protocol and statistical analysis plan call for, but I suspect that Frank E. Harrell, Jr. wasn't on the IRB / EC that approved the study.

[MD Johan]

Note also the Stata Journal paper that I referred to:

Thank you ericmelse, the article was really helpful.

[MD Johan]

The focus on nonparametric tests of change scores of before-and-after ordered categorical data strikes me as a little misplaced. I don't know what the protocol and statistical analysis plan call for, but I suspect that Frank E. Harrell, Jr. wasn't on the IRB / EC that approved the study.

Thank you Joseph for your answer. I agree with you that the focus on nonparametric tests of change scores of before-and-after ordered categorical data is a little bit misplaced. Since it is ordinal data a change in score is hard to interpret hence a change from 4 to 5 might not be the same as a change from 7 to 8. I would have preferred to just analyse the follow-up values with a Mann-Whitney-U test, to se if there was a significant difference between the groups. Since all patients are randomized and there were no differences at baseline I didn't think of using a regression model.

The reason I am analysing the difference in score is because other members of my group wishes so and that a journal asked for the difference between the groups with a 95% CI.