Multiple imputation in the outcome evaluation of psychotherapies of variable duration

Participants who drop out of studies have on average a poorer outcome than completers and therefore should not be ignored in the evaluation of treatment outcome. Multiple imputation is a technique to obtain unbiased statistical results from incomplete data. We developed a strategy to impute complex longitudinal data from psychotherapies of different lengths using functions from the MICE package in R. Missing data were imputed within a multilevel framework with bayesian or bootstrap methods. Data analysis after imputation encompassed both the calculation of pre-post-differences and the comparison of groups after a propensity score matching. We used a cross-validation to successfully verify the capability of our procedure to discriminate sufficiently between improved patients and those with a poor outcome