I have a longitudinal data set like this. A number of categorical health measures, along with individual characteristics such as age and sex, were taken for 100 adults at N=4 waves of interview in 1980s. Then starting in 2000 (every year until 2006) we collected the annual Medicare reimbursement records for those eligible survivors from this group. One of the research questions is how health status measured earlier affects the Medicare expenses later on. Any suggestions for appropriate modeling strategy? Thanks.

LC