EHR Audit Logs as Multi-Axial Traces for Foundation Model Training

A new paper from Eberhard, Apathy, and Johnson reframes electronic health record (EHR) audit logs as rich, multi-axial traces of care delivery. Instead of viewing them as mere operational metadata, the authors propose that each timestamped action simultaneously belongs to multiple clinically meaningful axes: a clinician's individual work, a patient's trajectory, a care team's collective activity, and a recurring workflow pattern. This multifaceted structure, they argue, is an untapped resource for training foundation models that can learn reusable representations directly from the raw event stream.

This perspective has concrete implications for both research and practice. By treating audit logs as multi-axial traces, developers can design models that preserve essential relationships across axes, enabling more accurate evaluation of clinician effort, patient outcomes, and team coordination. The paper also outlines governance requirements for such representations, ensuring they are used ethically and effectively. For healthcare IT professionals, this work points toward AI that can naturally understand the complex choreography of clinical work, leading to smarter EHR systems that reduce burnout and improve care.