To account for missing data by MI, we identified the
baseline variables that were best predictors of whether a
subject completed follow-up and used these in a multiple
imputation to predict missing values of the outcomes. Based
on the literature we assumed that low GAF score, age,
duration of illness, compliance at baseline, and abuse status
were both related to whether a patient would fail to complete
follow-up and what the missing values would have been had
they not been missing. 10 sets of multiply imputed data were
generated and analyzed.
All analyses were conducted in Stata 10.