EHRSummarizer: A Privacy-Aware, FHIR-Native Architecture for Structured Clinical Summarization of Electronic Health Records

Houman Kazemzadeh 1, Nima Minaifar 2, Kamyar Naderi 3, Sho Tabibzadeh 4 1 MedLedger365, Dubai, United Arab Emirates 2 MedConnect365, Dubai, United Arab Emirates 3 Xylemed, Dubai, United Arab Emirates 4 Kypath Associates Inc., Greater Toronto Area, Canada

Abstract Clinicians routinely navigate fragmented electronic health record (EHR) interfaces to assemble a coherent picture of a patient’s problems, medications, recent encounters, and longitudinal trends. This work describes EHRSummarizer, a privacy-aware, FHIR-native reference architecture that retrieves a targeted set of high-yield FHIR R4 resources, normalizes them into a consistent clinical context package, and produces structured summaries intended to support structured chart review. The system can be configured for data minimization, stateless processing, and flexible deployment, including local inference within an organization’s trust boundary. To mitigate the risk of unsupported or unsafe behavior, the summarization stage is constrained to evidence present in the retrieved context package, is intended to indicate missing or unavailable domains where feasible, and avoids diagnostic or treatment recommendations. Prototype demonstrations on synthetic and test FHIR environments illustrate end-to-end behavior and output formats; however, this manuscript does not report clinical outcomes or controlled workflow studies. We outline an evaluation plan centered on faithfulness, omission risk, temporal correctness, usability, and operational monitoring to guide future institutional assessments.

Keywords: Electronic health records; HL7 FHIR; Clinical summarization; Privacy-aware systems; Clinical NLP; Health informatics; Model guardrails

1. Introduction EHR platforms have increased the availability of patient data, but they often require clinicians to traverse numerous screens, tabs, and documents to assemble a coherent clinical narrative (1). This fragmentation can slow decision-making in time-limited workflows and contribute to documentation burden and cognitive overload (2). While modern EHRs provide dashboards and filters, they frequently optimize for data entry, billing, and resource-specific viewing rather than rapid synthesis of a patient’s current status and longitudinal trajectory. As a result, chart review often becomes an exercise in manual information retrieval: identifying active problems, linking medications to indications, reconstructing timelines, and interpreting trends across laboratory and vital signs (3). Additionally, recent increases in longitudinal data volume, cross-site care, and clinician mobility may contribute to increased difficulty in maintaining consistent chart review when relying solely on vendor-specific interfaces. Recent advances in generative modeling have made it possible to explore the conversion of structured clinical data into concise, clinician-oriented summaries (4). However, EHR summarization in practice requires more than text generation: careful resource selection, normalization, deployment choices that align with organizational privacy requirements, and guardrails that limit unsupported inference (5, 6).
   This paper describes a FHIR-native architecture intended to support structured summarization while minimizing data retention and enabling deployment within a healthcare organization’s trust boundary. The goal is not to provide diagnoses or treatment directives, but to assist chart review by presenting relevant information in a consistent format. This manuscript makes four architectural considerations: • A resource-targeted retrieval strategy for summarization that prioritizes clinically high- yield FHIR R4 resources and degrades gracefully when resources are missing. • A normalization step that constructs a clinical context package as a stable intermediate representation between EHR retrieval and summarization. • A privacy-aware deployment design that supports stateless processing, data minimization, and configurable trust-boundary placement (hosted versus local inference). • A safety posture and evaluation blueprint emphasizing faithfulness, omission risk, temporal correctness, and operational monitoring rather than clinical outcome claims. Beyond summarization, the architecture is designed to provide a consistent representation of clinical context across heterogeneous EHR implementations, supporting standardized chart review workflows independent of vendor-specific interfaces.

2. Problem Statement and Motivation 2.1 Clinical burden and cognitive overload Clinicians often spend substantial time navigating EHR interfaces and searching across notes, laboratory and vital signs, and medication lists to answer basic clinical questions such as: • What are the key active problems? • What medications is the patient curr