This paper addresses the challenge of handling missing values in multivariate time series data, particularly in healthcare applications. The authors propose a novel deep learning model called GRU-D, which is based on Gated Recurrent Units (GRUs). GRU-D incorporates two representations of missing patterns: masking and time interval. Masking indicates which variables are observed or missing, while time intervals capture the temporal patterns of observations. The model effectively captures long-term temporal dependencies and utilizes missing patterns to improve prediction performance. Experiments on real-world clinical datasets (MIMIC-III, PhysioNet) and synthetic datasets demonstrate that GRU-D outperforms state-of-the-art methods, including those that use imputation techniques. The model provides valuable insights into the impact of missing values on prediction tasks and offers a promising approach for handling missing data in various time series applications.This paper addresses the challenge of handling missing values in multivariate time series data, particularly in healthcare applications. The authors propose a novel deep learning model called GRU-D, which is based on Gated Recurrent Units (GRUs). GRU-D incorporates two representations of missing patterns: masking and time interval. Masking indicates which variables are observed or missing, while time intervals capture the temporal patterns of observations. The model effectively captures long-term temporal dependencies and utilizes missing patterns to improve prediction performance. Experiments on real-world clinical datasets (MIMIC-III, PhysioNet) and synthetic datasets demonstrate that GRU-D outperforms state-of-the-art methods, including those that use imputation techniques. The model provides valuable insights into the impact of missing values on prediction tasks and offers a promising approach for handling missing data in various time series applications.