Abstract: Existing clinical prediction models are difficult to effectively utilize medical record data with many missing values and often only consider the time-series information of a single patient, ignoring the potential connections between similar patients. A spatial-like temporal graph convolutional model BSim-STGCN is proposed to address the above problems. The model designed a global missing information capture mechanism for obtaining the current missing representation of missing values in the entire time series. A spatial-like graph convolution (Spatial-like GCN) based on patient similarity was proposed to model dependencies between similar patients. Experiments on two real datasets show that the prediction accuracy of the BSim-STGCN model outperforms other clinical prediction models.