Federated Learning Authenticity Standard for Healthcare as derived from lessons in self-driving cars.

In this Perspective, we highlight a critical mislabeling problem in healthcare federated learning research. Although federated learning is widely promoted as a privacy-preserving approach for multi-institutional artificial intelligence development, most published studies are still single-institution simulations that do not cross real organizational boundaries. As such, we propose a six-level classification scale (0 to 5), modeled on the autonomous driving taxonomy used in the automotive industry, that ranges from no federation through purely mathematical simulations to fully autonomous, continuously learning systems operating across separate institutions. We term it Federated Learning Authenticity Standard for Healthcare. Further we propose that authors declare in their studies the achieved level in a single sentence within the abstract, making this standard an honest description of study design unavoidable. We discuss the standard’s potential role as a reporting framework analogous to established guidelines for clinical trials and prediction model studies.