We have a challenging dataset that we would like some advice on
==research question Core aim is to characterise classes of variability in daily opioid dose in a sample of people tampering with prescription opioids. We anticipate that there would be some people with stable high and low doses; and some that have slightly or highly variable patterns of opioid dose over a week.
==nature of dataset - 450 cases - each case has a 7-day diary with dose of opioid (0-400mg), with preponderance of zeros
==challenges if we use day as a variable in the analysis, then a latent class/growth model type approach might characterise people that have high doses on days 2 and 3 and then low doses on all other days, as being distinct from people that have high doses on days 5 and 6 but low doses on all other days, but they are not. I can't remove this order by sorting by dose, as the variability across subsequent days is important to capture
We also can't convert the data to change scores as we would not be able to differentiate the people that have stable doses but of a large amount from people that have stable doses but of a low amount
==a bonus complication We also have this data repeated one year later, and we would like to test for stability of these patterns of use over the two study waves.
If anybody has any advice about how we could tackle this research question I would really appreciate your help!