This paper presents filtering-based visual-inertial odometry (VIO) using an inertial measurement unit (IMU) and a dynamic vision sensor (DVS) in high-speed scenarios. A DVS is a bio-inspired sensor that captures asynchronous intensity change directions called an event instead of the absolute brightness as a conventional camera. We propose an IMU-aided event flow estimation method building upon the recent work of the contrast maximization in which event flow estimation is solved via maximizing the contrast composed of accumulated events given spatio-temporal window. Specifically, we initialize a gradient ascent method with a closed-form of optical flow computed from the current filter state. In the estimator, the estimated event flow updates the unscented Kalman filter in which an inertial navigation system (INS) propagates to yield prior information. The proposed method is validated through the challenging open-source dataset. The experimental results reveal that the proposed VIO notably decreases a position error accumulation of an INS when compared to the image-based VIO.