Efficient two-step multivariate random effects meta-analysis of individual participant data for longitudinal clinical trials using mixed effects models

Abstract Background Mixed effects models have been widely applied in clinical trials that involve longitudinal repeated measurements, which possibly contain missing outcome data. In meta-analysis of individual participant data (IPD) based on these longitudinal studies, joint synthesis of the regression coefficient parameters can improve efficiency, especially for explorations of effect modifiers that are useful to predict the response or lack of response to particular treatments. Methods In this article, we provide a valid and efficient two-step method for IPD meta-analyses using the mixed effects models that adequately addresses the between-studies heterogeneity using random effects models. The two-step method overcomes the practical difficulties of computations and modellings of the heterogeneity in the one-step method, and enables valid inference without loss of efficiency. We also show the two-step method can effectively circumvent the modellings of the between-studies heterogeneity of the variance-covariance parameters and provide valid and efficient estimators for the regression coefficient parameters, which are the primary objects of interests in the longitudinal studies. In addition, these methods can be easily implemented using standard statistical packages, and enable synthesis of IPD from different sources (e.g., from different platforms of clinical trial data sharing systems). Results To assess the proposed method, we conducted simulation studies and also applied the method to an IPD meta-analysis of clinical trials for new generation antidepressants. Through the numerical studies, the validity and efficiency of the proposed method were demonstrated. Conclusions The two-step approach is an effective method for IPD meta-analyses of longitudinal clinical trials using mixed effects models. It can also effectively circumvent the modellings of the between-studies heterogeneity of the variance-covariance parameters, and enable efficient inferences for the regression coefficient parameters.