IMDRF N88 Good Machine Learning Practice for Medical Device Development: Guiding Principles (2025) P3: Evaluate clinically with datasets representative of the intended patient population

What does P3 require?

Data collection protocols aim to represent, in samples of adequate size for training, testing and monitoring, the characteristics of the intended patient population that matter (age, gender, sex, race, ethnicity, geography, medical condition), the intended use environment and the measurement inputs, so results generalise to the intended population; this underpins clinical evaluation, manages unintended bias and dataset drift, supports usability assessment and reveals where, for which subgroups and under what circumstances, the model may fall short, including as time passes.