RX Guard: A FHIR-Aware Intelligent Agent for Controlled Substance Prescription Safety Review

RX Guard is a FHIR-aware safety review agent for controlled substance prescription workflows, designed to assist clinicians in making faster, clearer, and defensible prescription decisions. Its core value lies in providing interpretable decision support, identifying missing documents and risk signals, adhering to design principles such as interoperability (FHIR standard), human-machine collaboration (human-in-the-loop), and interpretable outputs. It is positioned as an auxiliary tool rather than a system that replaces clinicians' independent decision-making.

Controlled substance prescription is a high-risk, high-friction workflow where clinicians need to make decisions under time pressure by integrating medical records, drug backgrounds, risk factors, and policy requirements. RX Guard is designed for this scenario, clearly positioned as a 'human-in-the-loop decision support system'. It emphasizes the use of synthetic/de-identified data to protect privacy while maintaining correspondence with real clinical workflows, avoiding replacement of doctors' professional judgments.

RX Guard supports the FHIR (Fast Healthcare Interoperability Resources) international standard and can exchange data with existing EHRs, pharmacy management systems, etc. Its core architecture adopts a hybrid model: the rule engine identifies clear compliance issues (e.g., missing document fields, drug interactions), while the AI explanation layer adds natural language explanations to rule findings, balancing certainty and flexibility. The output includes both structured data and human-readable descriptions.

RX Guard's MVP implements a hybrid review process. Inputs include synthetic patient context, visit summaries, medical record text, drug requests, etc. Outputs cover risk summaries, missing document lists, interpretable findings, suggested wording, and structured review results. The system provides decision support to clinicians by checking medical record text and structured clinical context, marking missing documents and risk signals.

RX Guard provides a complete local demonstration environment, including synthetic cases (e.g., the Sheila Bankston scenario), a local CLI adapter, and a static EHR-style UI. The demonstration process: receive EHR fields → parse into synthetic cases → send data via Prompt Opinion secure payload → render decision support JSON and PDMP line data. The static UI is based on the eCW style, displaying analysis results, PDMP data, and workflow buttons, which can be run locally for an intuitive experience.

RX Guard repeatedly emphasizes its auxiliary nature, with outputs labeled as 'clinician support guidance' rather than independent decisions. This reflects the core ethical principles of medical AI: the final decision-making power remains in the hands of human doctors, and AI only provides information, marks issues, and suggests improvements. The project focuses on compliance and transparency, with detailed documentation of design decisions, integration plans, and compliance considerations for easy auditing.

RX Guard represents a pragmatic approach to medical AI: using intelligent technology to enhance clinical workflows while retaining human decision-making power and responsibility. Combining the FHIR standard, hybrid architecture, and transparent release model, it provides a reference paradigm for controlled substance prescription review. As the regulatory framework for medical AI improves, such systems that focus on interpretability, auditability, and human-machine collaboration will become important references for the industry.