Table 1.
Evaluated approaches to missing outcome data during follow-up when using data from an observational study as an external control arm to a clinical trial
| Approach | Missing outcome data at visitf | Intermittent missing visitsg | Drop-outsh |
|---|---|---|---|
| Complete follow-up case analysesa | Exclusion | Exclusion | Exclusion |
| Strict censoring + IPCWb | Censoring (set as drop-out) | Censoring (set as drop-out) | IPCW |
| MI + censoring + IPCWc | MI | Censoring (set as drop-out) | IPCW |
| MI + IPCWd | MI | MI | IPCW |
| MI for alle | MI | MI | MI |
IPCW Inverse probability of censoring weighting, MI Multiple imputation
aAssumptions: Patients with complete follow-up data are exchangeable with patients with missing data
b Assumptions: The IPCW model is correctly specified when modeling the missing mechanism given previous observations in individuals with missing data due to drop-out (naturally occurring or created due to artificial censoring)
c Assumptions: Both the MI model and the IPCW model are correctly specified. MI models missing outcome variables at visits given available information in the dataset. IPCW models the missing mechanism given the observed past for missing data due to drop-out (naturally occurring or created due to artificial censoring)
d Assumptions: Both the MI model and the IPCW model are correctly specified. MI models missing outcome variables at visits and intermittent missing visits given information in the dataset. IPCW models the missing mechanism given the observed past for missing data due to naturally occurring drop-out
eAssumptions: The MI model is correctly specified when modeling all missing outcome data given information in the dataset. Separate models were specified for each cohort and the imputed datasets were thereafter combined
fA visit is recorded at the required time point, but the disease activity measure is missing
gA visit is missing for the 6-month and/or 12-month follow-up, but there is a later visit within the timeframe of the study
hA follow-up visit and all subsequent visits are missing within the timeframe of the study