Statistical approaches for systematically missing covariates in individual participant data meta-analysis: insights and applications using 5 large cardiovascular trials

Combining multiple studies for evidence synthesis can improve the precision of effect estimates and enable investigation of more granular clinical research questions. However, individual participant data (IPD) meta-analyses are often limited by the presence of systematically missing covariates that are only available in a subset of studies. Common solutions include removing missing covariates from the analyses or excluding studies that have not collected the missing covariates. Both approaches risk losing valuable information and introducing bias. Two additional approaches have been proposed: bivariate meta-analysis and multiple imputation of systematically missing data. The potential advantages and performances of these approaches, however, have only been discussed using simulation studies. In this analysis, we used data from five large cardiovascular clinical trials and illustrate the advantages and applicability of these two approaches in retaining missing covariates across various scenarios of increased complexity as compared to the more common approaches of excluding the covariates or the studies. Our results showed that both bivariate meta-analysis and multiple imputation of systematically missing data preserve information and improve the precision of the combined estimates. The findings underscore the importance of retaining information that is otherwise lost due to systematically missing values.