With the development of medical information, machine learning are applied on clinical aid to decision-making, for example disease prediction, mortality risk prediction, and so on. For the multi-dimensional, and multi-modality and sequential clinical data, there are being some challenges on feature fusion and representation which would impact the performance of prediction models. In addition, model interpretation attracts more attention in medical AI application field. According to the difference among the contribution from different examination indicators to predict the disease risk, we need consider the relation between indicators and disease as well as the time-dependence of indicators. In this work, we design a Multi-View and Multi-Channel (MVMC) method to construct patient record and propose an ensemble multi-view and multi-channel disease risk prediction model. In this work, we construct visit-view, feature-view and modality-view to represent the patient clinical features. Meanwhile multi-channel data processing mehod is used to extract temporal and spatial characteristics among features. We adopt GRU+CNN and Bi-GRU and transformer in our ensemble models for feature-view, visit-view and modality-view respectively. Attention mechanism is used for computation of correlation coefficient among each view to simulate physician diagnosis process. Experimental results show that our ensemble MVMC model achieves advanced performance in Precision, AUC, Recall and F1-score on the two data sets. Then we give the model interpretation by heatmap.