Model card
A short, standard document that travels with a model and says what it is for, who it was built for, what data it was trained and tested on, how it performed, including for subgroups, and what its known limits are. The format was published by a Google research team in 2019 and has since been adapted for healthcare, sometimes under the name model facts label or AI nutrition label. A model card is only as useful as its contents, so check that it covers what a buyer needs most: the population it was validated on, performance at the operating threshold, and the situations where it should not be used. Ask for one before a demo, and read the limitations section first.
A model card for a deterioration score lists its training data as adult medical inpatients at three US academic centers, and its limitations section says it was not tested on surgical patients. Your hospital plans to run it on every ward, surgical included. Start the local validation with surgical patients, before anyone books a demo.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →