EU regulatory mapping

AEGIS is designed for dual-jurisdiction compliance, supporting both US FDA and EU regulatory frameworks. This page maps AEGIS elements to the EU AI Act, EU MDR, and relevant MDCG guidance documents. Together with the PCCP mapping page, it provides the complete regulatory alignment picture.

Regulatory landscape overview

Medical AI systems marketed in the EU must comply with three overlapping regulatory instruments:

AEGIS addresses requirements from all three instruments through its modular architecture.

AI Act Article 43(4) -- Substantial modification

Article 43(4) of the AI Act defines when a high-risk AI system has undergone a "substantial modification" requiring re-assessment. AEGIS helps determine whether an iterative update constitutes a substantial modification:

AEGIS's position is that pre-specified iterative updates within the AEGIS governance framework do not constitute substantial modifications when:

1. The modification protocol is pre-specified (CDM hierarchy)
2. The impact assessment methodology is pre-defined (MMM evaluation)
3. All updates remain within predetermined performance boundaries
4. The intended purpose is unchanged

When the CDM produces REJECT, the candidate model is not deployed, so no substantial modification occurs. When ALARM fires on the released model, this may signal a need for regulatory reporting under MDR vigilance rules rather than a new conformity assessment.

**Sepsis example:** Across 11 iterations, the sepsis prediction model underwent iterative retraining. Iterations 0--5 produced APPROVE decisions, all within the predetermined performance boundaries. Iteration 6 was CONDITIONAL APPROVAL (cross-site drift), and iteration 7 was CLINICAL REVIEW (regression from P^ref). Under Article 43(4), these iterations would not constitute substantial modifications because the modification protocol was pre-specified, the impact was assessed against locked thresholds, and the intended purpose remained unchanged. Iteration 10 (REJECT + ALARM) did not result in deployment of a modified model, and iteration 9 was an APPROVE (recovery). The governance framework prevented potentially harmful modifications from reaching clinical use.

Technical documentation requirements

The AI Act requires high-risk AI systems to maintain comprehensive technical documentation. AEGIS outputs map to these requirements:

EU MDR Article 83 and the ALARM signal

EU MDR Article 83 requires manufacturers of implantable and Class III devices to report on safety and performance through Periodic Safety Update Reports (PSURs). For other classes, post-market surveillance plans must define monitoring and reporting procedures.

The AEGIS alarm channel maps directly to MDR vigilance obligations:

ALARM as an early warning system

The ALARM channel provides an earlier signal than traditional PMS approaches because it monitors the released model proactively against pre-specified thresholds rather than waiting for adverse event reports:

Traditional PMS:  Adverse event occurs → Report → Investigate → Act
AEGIS ALARM:      Metric threshold breached → Immediate alert → Investigate → Act

This proactive approach aligns with the EU regulatory philosophy of risk prevention rather than reactive harm mitigation.

**Sepsis example:** At iteration 10, alarm A3 fired because the drift score reached 1.000 (exceeding theta_major = 0.90). In an EU MDR context, this would trigger a review under the post-market surveillance plan to determine whether the detected distribution shift constitutes a field safety concern. If the drift affected the currently deployed model's ability to correctly predict sepsis in the patient population, this could escalate to a field safety corrective action (FSCA) under MDR Art. 87. The AEGIS audit trail would provide the documentary evidence needed for the competent authority notification.

MDCG 2025-6 key points

MDCG 2025-6 provides guidance on the clinical evaluation and post-market clinical follow-up for AI/ML medical devices. Key points relevant to AEGIS implementers:

MDCG guidance emphasises that AI/ML devices require a "living" clinical evaluation plan that accounts for post-market performance changes. AEGIS provides the infrastructure for this through its iterative evaluation loop and structured audit trail.

PSUR reporting

For devices requiring PSURs under MDR Article 86, AEGIS outputs provide structured data for the safety and performance sections:

PSUR content from AEGIS

Risk management philosophy

Table I of the manuscript maps AEGIS to the broader regulatory risk management philosophy:

AEGIS implements risk management as a continuous, automated process rather than a point-in-time assessment. Every iteration evaluates the model against the complete risk control framework, ensuring that risk management is not a one-time activity but an ongoing operational practice.

Dual-jurisdiction considerations

When operating in both US and EU markets, be aware of key differences:

AEGIS addresses both jurisdictions through its modular design: the same CDM logic, audit trail, and threshold framework satisfy requirements from both regulatory systems.

**Caveat:** Regulatory landscapes for AI in healthcare are evolving rapidly. The mappings in this guide reflect the state of regulations as of early 2026. Confirm all regulatory interpretations with qualified regulatory counsel in both jurisdictions before relying on these mappings for submission or compliance purposes.

Source

Manuscript Section II-A (regulatory context and Table I mapping six frameworks), Section IV-F (dual-jurisdiction regulatory mapping), and Table I (regulatory framework comparison). EU AI Act Article 43(4) substantial modification criteria. EU MDR Articles 83, 86, and 87. MDCG 2025-6 guidance on AI/ML clinical evaluation. Supplementary S2.8 provides the regulatory pathway mapping configuration.