I am trying to analyze data from the longitudinal qualitative scoring of knee joints to measure differences in disease severity by a number of markers between various groups.

The subjects are classified by ID, Strain, Sex, and knee (L or R). At a series of time points surrounding an intervention, the knees of each subject are separately scored for damage to various structures (e.g. meniscus, patella, cartilage) using a qualitative scale (0 = no disease, 1 = mild, 2 = severe).

Normally, I would perform an ANOVA analysis to determine which of the classifiers are significantly associated with a given pathology, say a high meniscus score, and then perform a t-test between the different levels of the classifier (with multiple comparison corrections as necessary).

With the scores used in this study, I know that this is no longer appropriate and that I will need to use some other test, but I am a little lost as to which comparison to perform (non-parametric alternative to ANOVA such as Kruskal-Wallis, or logistic regression, or something else).

Any guidance would be greatly appreciated (references to the corresponding functions in R would be doubly so). Thanks.