AEGIS: an operational infrastructure for post-market governance of adaptive medical AI under US and EU regulatory frameworks

Abstract

Abstract Medical AI requires governance preserving safety during continuous improvement. FDA Predetermined Change Control Plans (PCCP) and EU post-market surveillance address iterative updates but lack operational tooling. We present AEGIS, an internal governance and post-market-monitoring infrastructure designed to support the documentation of predetermined governance pathways for PCCP and Medical Devices Regulation alignment. AEGIS comprises dataset assimilation and retraining, model monitoring, and conditional decision-making modules. It utilizes a four-category taxonomy (APPROVE, CONDITIONAL APPROVAL, CLINICAL REVIEW, REJECT) and an independent ALARM signal. Instantiating AEGIS for sepsis prediction (PhysioNet) and brain-tumor segmentation (BraTS) exercised all decision categories, successfully routing failures to the critical REJECT + ALARM state (no deployable model; released model at risk). On real cross-site data, drift detection ROC-AUC rose from 0.82 to 1.00, reliably triggering near 500-patient batches with zero false alarms. This proof-of-concept utilizes illustrative clinical models; drift scoring quantifies covariate shift only, leaving concept drift for future work.

The authors' abstract, as published at the source. npj Digital Medicine, 2026 · DOI ↗