I am using Poisson regression with a log link, to estimate Relative Risk and need some guidance on selecting a correlation structure. My observations are courses of antidepressant treatment where a dichotomous variable of adhered or not adhered is decided based on if the individual had >=84 days supply (adhered) or <84 days supply (nonadhered) over a 115 day window.
Because the study window goes across 3 years, and the washout period is 105 days, individuals can represent more than one observation in the sample. Approximately 20% of the unique individuals in the sample have more than one course of treatment.
Any suggestions on which correlation structure I should explore? Does it matter that not all of the cases have a repeated measure?