Electronic health records (EHR) consist of longitudinal clinical observations\nportrayed with sparsity, irregularity, and high-dimensionality, which become\nmajor obstacles in drawing reliable downstream clinical outcomes. Although\nthere exist great numbers of imputation methods to tackle these issues, most of\nthem ignore correlated features, temporal dynamics and entirely set aside the\nuncertainty. Since the missing value estimates involve the risk of being\ninaccurate, it is appropriate for the method to handle the less certain\ninformation differently than the reliable data. In that regard, we can use the\nuncertainties in estimating the missing values as the fidelity score to be\nfurther utilized to alleviate the risk of biased missing value estimates. In\nthis work, we propose a novel variational-recurrent imputation network, which\nunifies an imputation and a prediction network by taking into account the\ncorrelated features, temporal dynamics, as well as the uncertainty.\nSpecifically, we leverage the deep generative model in the imputation, which is\nbased on the distribution among variables, and a recurrent imputation network\nto exploit the temporal relations, in conjunction with utilization of the\nuncertainty. We validated the effectiveness of our proposed model on two\npublicly available real-world EHR datasets: PhysioNet Challenge 2012 and\nMIMIC-III, and compared the results with other competing state-of-the-art\nmethods in the literature.\n