Classifying the risk for myasthenic crisis using data-driven explainable machine learning with informative feature design and variance control – a pilot study