Multiple imputation for recovering missing values when data cannot be shared

Multisite studies are increasingly used to study human health across different populations and countries. However, a common challenge in using data from multiple studies is the presence of systematically missing values – when some studies have not recorded information on certain variables. Although it is possible to use data from sites with recorded observations to impute the missing values, this process becomes challenging when data pooling is not feasible because of logistic or legal constraints. We address this by introducing a framework for multiple imputation across study sites without the need of sharing individual data. In this talk, we present some motivating examples alongside a new command mi impute from that can handle the imputation of binary, discrete, and continuous variables. Given the increasing importance of multisite studies in medical and epidemiological research, mi impute from can offer a practical approach for imputing variables that have not been recorded in some study sites.