clinical informatics system implementation

Critical clinical information—such as risk factors and treatment responses—in mental health electronic health records (EHRs) is predominantly unstructured, hindering automated risk identification and precision interventions. Method: We developed VIEWER, a clinical informatics platform that achieves end-to-end integration of EHR semantic parsing, SNOMED CT/ICD-10 terminology standardization, knowledge graph modeling, and real-time decision support—deployed for the first time across a large UK National Health Service (NHS) mental health trust. Its implementation follows a novel cross-institutional, multidisciplinary co-design and deployment paradigm. Contribution/Results: A proof-of-concept evaluation demonstrated significant improvements in timeliness of high-risk patient identification and intervention coverage. VIEWER enables three-tiered decision optimization: individualized care delivery, interdisciplinary team coordination, and organizational-level population health management—thereby advancing scalable, evidence-informed mental healthcare.

Real-time, dynamic, and interpretable assessment of clinical decompensation and delirium risk in ICU settings remains challenging. Method: This study proposes a clinician–caregiver co-designed multimodal AI clinical decision support system (CDS) integrating electronic health records, wearable sensors, video analytics, and environmental data to enable real-time risk prediction and generate actionable, non-pharmacological intervention alerts. Contribution/Results: The work introduces (1) the first clinician-led CDS interface development paradigm; (2) the first deep integration of multimodal real-time sensing with evidence-based non-pharmacological delirium prevention strategies; and (3) intuitive information visualization embedded within clinical workflows. Qualitative human factors research—including focus groups and interviews—identified five key implementation themes and demonstrated significant improvements in early delirium detection, timeliness of clinical response, and decision actionability. The system received strong endorsement from ten frontline ICU clinicians and nurses.

This work addresses the common lack of continuous evaluation and governance mechanisms in deployed clinical AI systems, which hinders dynamic performance optimization. The authors propose the first end-to-end continuous governance framework tailored for clinical AI, integrating standards-driven validation, A/B testing for controlled version updates, real-time performance monitoring, fault tolerance, and deep integration with electronic health records (EHRs) to establish a closed-loop synergy between engineering iteration and clinical feedback. Applied to Hyperscribe—a speech-to-structured-clinical-note system—the framework achieved substantial improvements over seven iterative cycles: median clinician rating increased from 84% to 95%, negative user feedback decreased from 79% to 30%, median audio processing latency was 8.1 seconds, and task completion rate reached 99.6%, collectively enhancing system reliability and user satisfaction.

Despite their promise, large language models (LLMs) face a critical “algorithm-to-application” gap in clinical deployment, hindering real-world integration into electronic health record (EHR) systems. Method: Drawing on empirical EHR deployment experience, we propose the first systematic framework for implementing generative AI agents in clinical settings—centered on sociotechnical implementation tasks, which constitute over 80% of deployment effort. The framework addresses five core challenges: EHR data integration, clinical validation of model trustworthiness, economic sustainability, adaptive management of model and system drift, and multi-stakeholder governance. It synergistically integrates LLMs, prompt engineering, FHIR-compliant EHR interfaces, continuous performance monitoring, and governance protocols. Contribution/Results: We deployed irAE-Agent—a clinical AI agent for automated identification of immune-related adverse events—demonstrating feasibility and robustness. Evaluation by 20 clinical and technical experts confirms the framework significantly enhances translatability from pilot to routine clinical service.

Clinical decision support systems (CDSS) face significant challenges in development and validation due to stringent patient data privacy regulations, which restrict access to real-world electronic health records (EHRs) for integration, verification, and cross-platform migration testing. To address this, we propose SyntHIR: the first end-to-end CDSS development framework integrating the FHIR interoperability standard, Gretel’s synthetic data modeling, and the SMART on FHIR portable application architecture—enabling generation of privacy-compliant, FHIR-conformant synthetic EHR datasets. Leveraging Norway’s National Patient Register (NPR/NorPD), we successfully built and migrated a machine learning–driven CDSS into the Open DIPS clinical environment. All system components are open-sourced. SyntHIR overcomes the critical bottleneck of real EHR access, substantially accelerating early-stage CDSS development, regulatory compliance validation, and interoperability testing across heterogeneous clinical platforms—without compromising data privacy or semantic fidelity.

This study addresses the frequent violations of clinical coding standards—such as ICD-10, CPT, and HL7 FHIR—by large language models when generating structured medical data, which impedes integration with electronic health record systems. To mitigate this, the authors propose and validate a closed-loop verification-and-repair framework that automatically detects and iteratively corrects formatting errors. The approach is evaluated using three open-source models—Qwen2.5-7B, Llama3.1-8B, and Gemma2-9B—deployed locally across 320 clinical scenarios. Results demonstrate a substantial improvement in schema compliance across all models, achieving an overall adherence rate of 99.0% and increasing individual model performance by 7.8 to 12.5 percentage points. Notably, 96% of detected errors were attributable to repairable representation-layer issues, with most resolved within one or two correction rounds, effectively compensating for the models’ limited understanding of healthcare IT standards.

This study addresses the challenge of unreliable AI models and diminished clinical trust stemming from opaque data quality reporting in the secondary use of electronic health records (EHRs). To this end, the authors propose the first comprehensive framework for transparent data quality reporting across the entire EHR lifecycle. The framework innovatively distinguishes between data producers and consumers, explicitly defines five critical phases, and maps established data quality dimensions to specific workflow stages. Through iterative stakeholder and process analysis, a structured reporting mechanism is developed and validated on real-world datasets, demonstrating its ability to effectively trace the origins of data quality issues. The approach significantly enhances data interpretability, fitness-for-use, and governance efficacy, thereby providing a robust foundation for trustworthy AI development and clinical research.

This study addresses the challenge that existing healthcare IT systems face in extracting and structuring patient-specific clinical intent from natural language. The authors propose a three-layer framework—comprising documentation, clinical state, and clinical intent—and introduce the “Actionable Clinical Record” (ACR) as the fundamental computable unit at the clinical intent layer. They formalize, for the first time, the concept of “computable clinical intent” and develop a readiness maturity ladder alongside a framework for evaluating executable correctness. By integrating clinical information modeling, natural language processing, and standards such as FHIR, the work enables computable representation and validation of clinical intent. Feasibility of ACRs is demonstrated in specific subtasks, providing reusable foundational components for future research in this direction.

This study addresses the lack of standardization and explicit semantic representation in clinical data, which hinders interoperability and reproducibility in machine learning. To overcome this limitation, the authors propose the first integration of the MEDS clinical event model with Semantic Web technologies, resulting in MEDS-OWL—a lightweight OWL ontology comprising 13 classes, 10 object properties, 20 data properties, and 24 axioms. They further develop the meds2rdf tool to automatically transform MEDS data into FAIR-aligned RDF graphs. The approach leverages SHACL constraints for validation and RDF graph representations to provide a reusable semantic layer that enables semantic enrichment, cross-system interoperability, and graph-based analytics of clinical data. The methodology is successfully validated on a synthetic dataset capturing care pathways of patients with ruptured aneurysms.

Current medical AI systems often operate as isolated models, lacking accountability and the capacity for continuous evolution. This work proposes Clinical Harness, a runtime governance architecture that introduces the novel concept of “clinical AI skills” to establish a governable ecosystem unifying knowledge-driven, data-driven, and physics-enhanced AI capabilities. The framework enables registration, orchestration, safeguarding, and monitoring of these skills, facilitating coordinated clinical support across the entire care continuum. Using osteoporosis as a case study, the authors demonstrate that diverse types of AI skills can effectively collaborate under runtime governance to deliver comprehensive, end-to-end patient care.