healthcare data

Clinical handwritten notes suffer from inconsistent handwriting, pervasive abbreviations, nonstandard terminology, grammatical and spelling errors, and disorganized formatting—severely hindering information extraction and interoperability in electronic health records (EHRs). To address this, we propose the first end-to-end LLM-based framework for clinical note standardization. It integrates terminology mapping, medical ontology alignment, and rule-augmented generative text reconstruction to jointly perform grammar/spelling correction, normalization of nonstandard clinical terms, abbreviation expansion, and structural reformatting—natively supporting interoperability standards such as FHIR. Evaluated on 1,618 real-world clinical notes, our method corrects on average 4.9 grammatical errors, 3.3 spelling errors, 3.1 nonstandard terms, and 15.8 abbreviations per note. Expert evaluation confirms high semantic fidelity and negligible information loss, significantly improving readability and downstream task performance.

Electronic health records (EHRs) frequently exhibit semantic inconsistencies between structured tabular data and unstructured clinical notes, posing risks to clinical decision-making safety. To address this challenge, we introduce EHRCon—the first benchmark dataset for cross-modal consistency verification in EHRs—comprising 105 clinical notes and 4,101 annotated entities, fully compatible with both MIMIC-III and OMOP Common Data Model standards, and rigorously validated by domain experts. We formalize a novel paradigm for EHR cross-modal consistency validation and propose CheckEHR, an eight-stage large language model–driven framework supporting zero-shot and few-shot inference. Extensive experiments demonstrate that CheckEHR significantly outperforms existing baselines. EHRCon exhibits high annotation confidence and strong cross-system reproducibility. Both the dataset and implementation code are publicly released to foster reproducible research.

This study addresses three critical challenges in electronic health record (EHR) analytics: (1) the limited utility of unstructured clinical text for high-quality clinical decision support; (2) cross-institutional semantic heterogeneity among EHR data; and (3) insufficient generalizability and fairness of medical AI models. To tackle these, we propose the first systematic, large language model (LLM)-driven framework that integrates heterogeneous EHR modalities—including free-text notes, structured laboratory values, and clinical codes. Our method introduces an ontology-guided, cross-institutional semantic alignment mechanism, coupled with interpretable fine-tuning and bias-correction strategies, to enable text-augmented multimodal representation learning. Evaluated on multicenter clinical prediction tasks, our framework achieves a mean AUC improvement of 5.2%, demonstrating enhanced model robustness. Furthermore, it exhibits superior predictive fairness across diverse demographic subgroups, validating its equitable performance in real-world heterogeneous healthcare settings.

This work addresses the challenge of constructing interoperable patient digital twins from unstructured electronic health records (EHRs), which is hindered by clinical text heterogeneity and the lack of standardized mappings. The authors propose the first end-to-end semantic natural language processing (NLP) pipeline that tightly integrates with the Fast Healthcare Interoperability Resources (FHIR) standard. By combining named entity recognition, concept normalization to SNOMED-CT and ICD-10 terminologies, and relation extraction, the pipeline automatically transforms free-text clinical notes into structured FHIR resources. Evaluated on the MIMIC-IV Clinical Database Demo, the approach significantly improves F1 scores for both entity and relation extraction, outperforms baseline methods in schema completeness and system interoperability, and enables the automated construction of patient digital twins with high semantic consistency.

Inconsistent diagnostic coding and poor interoperability across institutions and species hinder effective utilization of veterinary electronic health records (EHRs). Method: We propose an automated SNOMED-CT diagnostic coding framework leveraging large language models (LLMs), fine-tuning ten open-source Transformer architectures on 246,000 manually annotated clinical notes from the Colorado State University Veterinary Teaching Hospital. Contribution/Results: To our knowledge, this is the first approach achieving full coverage mapping to all 7,739 SNOMED-CT diagnosis codes used clinically in that institution. The best-performing model achieves an F1-score of 0.82—significantly outperforming baselines such as DeepTag and VetTag. Notably, even non-clinically pre-trained LLMs attain F1 > 0.78 under limited annotation budgets, demonstrating robust generalizability and feasibility in resource-constrained settings. This work establishes a scalable, low-cost paradigm for interoperable, cross-institutional integration of veterinary health data.

This work addresses the challenge of cross-institutional incompatibility in electronic health records (EHRs) caused by the absence of standardized data representations, which hinders large-scale machine learning applications. To overcome this limitation, the authors propose an open-source metadata repository grounded in the ISO/IEC 11179-3 standard, employing a “middle-out” standardization strategy and a microservices architecture. The system automatically catalogs EHR data elements and their value domains, supports both local Linux deployment and cloud hosting, and integrates modern authentication mechanisms. It features a user-friendly interface that enables error-free metadata registration and facilitates the visual discovery of interoperable features across heterogeneous databases. Validation through use cases such as rare disease patient identification demonstrates the system’s effectiveness in enhancing metadata management and enabling cross-institutional interoperability.

This work addresses the challenges clinicians face when using electronic health records (EHRs), particularly fragmented information and the absence of structured summaries. To mitigate these issues, the authors propose a privacy-preserving, FHIR-native, stateless architecture that retrieves key FHIR R4 resources, normalizes them into a unified clinical context bundle, and generates evidence-based, structured summaries devoid of diagnostic suggestions while explicitly flagging missing information domains. Designed to support data minimization and local deployment, the approach has undergone end-to-end validation in both synthetic and test FHIR environments. The resulting outputs align with clinical structuring requirements; however, formal evaluation of clinical impact remains pending.

In clinical research, manual extraction of structured clinical features from unstructured electronic health records (EHRs) is time-consuming, inefficient, and error-prone. To address this, we propose a privacy-preserving, modular, on-premises large language model (LLM) framework that integrates retrieval-augmented generation (RAG) with structured-output prompt engineering, enabling secure, scalable, containerized deployment in HIPAA-compliant environments. Our approach uniquely synergizes RAG with deterministic structured-response mechanisms for clinical text parsing—balancing domain adaptability and strict data privacy. Evaluated across multiple medical feature extraction tasks, the framework achieves high accuracy, substantially reducing manual annotation effort and improving data consistency. Notably, its systematic evaluation uncovered previously undetected systematic errors in prior human annotations, thereby validating both its reliability and its capacity for quality assurance and error discovery.

Existing clinical reasoning evaluation benchmarks predominantly rely on unstructured or static data, failing to capture the structured and interoperable nature of real-world electronic health records (EHRs). To address this gap, this work proposes a novel pipeline that integrates staged large language model (LLM) generation with terminology-anchored validation and repair, yielding MedCase-Structured—the first HL7 FHIR R4–compliant structured dataset for clinical reasoning assessment. Built upon the MedCaseReasoning benchmark, the pipeline successfully generates valid FHIR bundles for 82.5% of cases. Experimental results demonstrate that LLMs exhibit significantly lower diagnostic accuracy when provided with structured FHIR inputs compared to plain text, underscoring both the necessity of aligning evaluations with authentic clinical workflows and the innovative contribution of this dataset.

This work addresses the challenge of leveraging unstructured clinical notes in electronic health records (EHR) during training to enhance the performance of models that rely solely on structured data at deployment. The authors propose a multimodal learning framework that integrates clinical notes—encoded via BioClinicalBERT—with structured features such as demographics and medical codes during training. By combining a teacher–student architecture, contrastive learning, and contrastive knowledge distillation, the approach enables the final deployed model to achieve high inference efficiency using only structured inputs, without requiring access to textual data at test time. To the best of the authors’ knowledge, this is the first method to effectively augment structured EHR representations with unstructured clinical notes while maintaining deployment constraints. Evaluated on a cohort of 3,466 children with late language emergence, the model achieves an AUROC of 0.705, significantly outperforming baseline methods (AUROC = 0.656).