Analysis of variance in data with ordinal output

[DrJon]

Hi there. I am a new member with a query relating to the output from a clinical trial.

I have four groups of patients (30 in each group) who have been treated with one of four different medications to reduce side effects while they are on chemotherapy. The severity of symptoms is graded on an ordinal scale from 0 (no symptoms) to 5 (severe symptoms). I want to find out firstly whether there is an overall between-groups difference and then carry out multiple pairwise comparisons. If the symptoms were graded on a continuous scale, I would use ANOVA followed by Tukey HSD, but I don’t think I can do this with ordinal output.

Is the right strategy here to use Kruskal-Wallis? And if so, what is the right test to use for the subsequent multiple comparisons?

Thanks for your help

[ledzep]

Since you have ordinal data, I would use Proportional Odds Model (aka Ordinal Logistic Regression Model) to analyse this data with cumulative logit link function. I will then report the odds ratio for the last 3 groups relative to the reference category.

*In SAS;
proc genmod data=mydata; 
freq mycount; 
model severtiy.grade = group / dist=mult link=cumlogit; 
run;

If using R, I would look to use polr in the MASS package, which is very handy.