Schizophrenia is a persistent mental disorder manifested by significant abnormalities in perception, emotion, and behavior. Nevertheless, the neural mechanisms underlying this disorder are still not fully understood. In order to explore the differences in whole-brain causal connectivity between patients with schizophrenia and healthy controls in the resting state, a hierarchical degree (HD) index was proposed based on eigenmode method to overcome the inadequacy of node degree measured at a single level in traditional graph theory. It was found that the node degree of the whole-brain causal network of schizophrenia patients reduced. In addition, the most significant changes in in-degree were found in the motor system, whereas the most significant changes in out-degree were found in the default mode system. Higher-order node degree was further extracted and found to be superior to traditional graph theory degree in distinguishing schizophrenia patients from healthy controls based on a machine learning approach, and more accurately predicted positive and negative symptoms of schizophrenia, suggesting that higher-order network features can be used as biological indicators of schizophrenia. The findings of this paper reveal abnormal higher-order network features of schizophrenia and contribute to the advancement of objective diagnostic technologies for schizophrenia.