With an ordinal variable with four levels, you probably don't want to average the values.

One way to present the data is simply to report the percentage or ratings that are e.g., 1, 2, 3 or 4. In the following example, Likert.f is like your NCI-CTCAE ratings. And Patrick and Sponge Bob are like your two Treatments. If I understood, I guess you would have to simply pool the 16 values for each individual.

https://rcompanion.org/handbook/images/image051.png

It sounds like your design is more complicated than Friedman's test could handle. I imagine you have three independent variables that need to be taken into account: Individual, Treatment, and Time, since each individual was rated 16 times.

If so, this would be a task for mixed-effects ordinal regression. How easy this is may depend on what software you are using.

Thank you Dr. Sal Mangiafico for your reply.

This is a good way I can use to present the data. But would you please help me on how to pool the 16 values for each individual?

I use STATA for statistical analyses.

I did Mann-Whiteny U test to compare between the 2 groups at every time point (i.e. grades of adverse effect 1 at time point 1, grades of adverse effect 1 at time point 2, ...................then grades of adverse effect 1 at time point 16). But then I thought of Friedman's test followed by Bonferroni correction, to make use of repeated measurements and to include them in the analysis.

I guess regression analysis is used to examine the effect of multiple predictors (independent var.) on a single outcome (dependent var.). In my study, which is a Randomized controlled trial, the independent predictor is the treatment and the outcome is collected at multiple time points.

I hope I made it clearer that you can help me.

Thanks in advance.

Sal Mangiafico

I'm sorry I may have got you wrong. Is the mixed-effects ordinal regression the same as multilevel proportional odds model ?

Yes, you can fit a single regression model that takes into account all the independent variables. Yes, a proportional odds model can be used for ordinal regression. And, yes, I believe when you say multilevel that's the same as mixed-effects. ... By pool I just mean to include all the observations in a model, without, say, averaging them by individual across time points.