Accounting for missing data caused by drug cessation in observational comparative effectiveness research: a simulation study