Context
Within my field (neuropsychology), there is a well-known issue for some individuals to have very unstable diagnoses overtime. My area of interest is in dementia where the ideal diagnostic progression is from normal condition to mild cognitive impairment to dementia. While there are certainly people who do follow that trajectory over time, there are subgroups where people revert (e.g., going from mild cognitive impairment to normal cognition) rather than convert (e.g., going from mild cognitive impairment to dementia). My primary interest is in identifying what variables predict changing from each diagnostic category.
Research Goal
My initial thought was to use a multilevel multinomial regression; however, I don't think that this will give me the information that I want. I'm specifically interested in knowing what predicts movement from any one diagnosis to another. For example, ideally, the model would speak to what variable predict someone at time 1 with mild cognitive impairment having a diagnosis of normal cognition vs dementia at time 2. The data I'm looking at for this include upwards of 10 and 12 follow-up visits, so a model that can accommodate that timeframe would be ideal.
Question
What kinds of models should I be thinking about in this kind of question? My preference is something Bayesian using either brms or rstan more generally, but I'm open to other R based solutions.
