I tend to think there is no perfect solution that meets all criteria. For example, censoring may cause trouble for trial-level modeling. I suspect that different orders would have little impact most of the time. You may try various orders and see if they make big differences in the end at the population level.
Gang
Edited 1 time(s). Last edit at 01/10/2021 10:55AM by Gang.