Proportional odds with repeated measurements

data analysis models
1

I have a data set of 284 patients suffering from depression and treated with either escitalopram or citalopram. The patients had genetic testing for CYP450 variants (poor metabolized, rapid metabolizers, ultra rapid and normal). They were followed for a maximum of 6 years, and had an appointment in the clinic every three months ( so possibly 4 observations a year for 6 years, total= 24 observations per patients). The main outcome is ordinal ( no side effects, mild, and severe, requiring hospitalizaion or specialist appointment) and we are planning to use 3 explanatory variables in the model.l ( genetic variant, age, socioeconomic status) with possibly 24 observations per patient. 82 patients had no side effects. There were 983 mild side effects and 84 severe side effects.
Is the proportional odds with repeated measurements appropriate ( using"mixor" in R)?

Other suggestions?
Many thanks.

2

Random effects proportional odds regression could be appropriate here although it’s unlikely that assuming compound symmetric correlation patterns will give you a very good fit to the correlation structure. You might also consider a Markov proportional odds model. Examples are given in the re-analysis of the ORCHID and VIOLET 2 studies that you will find here.

1 Like