Abstract: To overcome the limitations that most of the existing knowledge transfer and fusion schemes are based on the horizontal federated learning algorithm and improve the training accuracy, and to fully explore the value of the massive patient data in medical institutions so that hospitals with different resource situations can benefit from it, this paper proposes a new vertical federated knowledge transfer framework. By utilizing the feature selection module based on information gain and the knowledge distillation module based on power iteration to assist in the vertical federated knowledge transfer, this framework can not only enhance the local sample learning performance and benefit hospitals with limited shared samples, but also guarantee the independence of the knowledge transfer process, enabling hospitals with scarce medical resources to cooperate with each other and effectively raise the quality of medical services. Simulation results show that compared with the LOCAL method, FTL method and VFedTrans method, the algorithm proposed in this paper can increase the disease prediction accuracy by about 10%.