Predicting the health status of power Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs) is of great significance for power electronic systems. It can enhance system reliability and reduce security risks effectively. Traditional health status prediction methods are usually offline or unconditionally continuously updated. It leads to low prediction accuracy or a large waste of computing resources. This paper proposes a digital twin-driven online health status prediction method. A surrogate model of the power MOSFETs is constructed in virtual space, and continuous drift detection and automatic updates are performed. Thus, the reliable online prediction of the entire life circle health status of power MOSFET can be realized. The proposed method's effectiveness is confirmed using NASA's public dataset. The surrogate model can continuously evaluate the difference with the real power MOSFETs and perform automatic updates, thus maintaining excellent prediction performance.