Atherosclerosis (AS) is the main root cause of cardiovascular disease. In order to make full use of the information contained in different modals data for auxiliary diagnosis, this paper proposes a risk prediction method for AS based on multimodal fusion of three types of modal data (Risk factors, Chief complaints and Electrocardiogram), which are low-cost, non-invasive and easy to obtain. The three types of modal data are respectively input into the classical classifiers, Bi-LSTM and 1D-ResNet, and the corresponding preliminary prediction results can be obtained. Then, based on the Choquet integral, a decision-making mechanism is proposed to effectively fuse the information contained in the three types of modal data and obtain the final prediction results. The experimental results show that, compared with the existing multimodal fusion methods, the proposed method can significantly improve the index of recall and accuracy (Recall: 0.85, Accuracy: 0.88).