Closest Similar Subset Imputation for Missing Data Analysis

Classifying patients based on stated reasons for missing outcome from different intercurrent events induces patients’ subsets in data from clinical trials. Often, data imputation disregards these patients’ subsets. We discuss a non-parametric data imputation method that reflects reasons stated for missing data and hence patients’ subsets. This subset imputation method is based on a similarity measure between baseline covariates of patients’ subset with missing data and a random closest subset without missing data. An illustration using imputation of gadolinium enhancing lesions in multiple sclerosis is provided