AI-powered clinical trial eligibility and safety using Amazon Bedrock AgentCore

This article describes how AI agents built on Amazon Bedrock AgentCore can accelerate clinical trial enrollment by automating patient eligibility screening while maintaining clinician authority and regulatory compliance.

• AWS HealthLake ingests and normalizes patient data from fragmented EHR systems into FHIR-native queryable records
• Three specialized agents handle pre-screening, detailed clinical evaluation, and site matching with multi-step reasoning
• Amazon Bedrock Guardrails enforce PII/PHI filtering, content safety, and grounding in evidence
• LLM-as-Judge evaluators score decisions on clinical accuracy, operational effectiveness, and safety compliance
• Human-in-the-loop review gives clinicians final authority with tiered escalation for complex cases
• Immutable audit trails in DynamoDB support FDA 21 CFR Part 11 and regulatory compliance requirements
• Reduces screening time from days to minutes while maintaining consistency across trial sites

The architecture combines FHIR-native data foundations, multi-step agent reasoning, continuous evaluation, and human oversight to transform manual clinical trial screening into a scalable, compliant, AI-assisted workflow.