Development of an early-warning system for high-risk patients for suicide attempt using deep learning and electronic health records