Predicting item difficulty parameters in high-stakes licensure examinations has the potential benefit to enhance test security, reduce the cost and time associated with item calibration and test development, and maintain the psychometric quality of the examination. This study predicted the item difficulty of multiple-choice questions from a high-stakes medical examination using supervised machine learning algorithms based on item characteristics, including the content blueprint, task, number of words in the vignette, average number of words in the options, average response time, and image type.