Enhancing Clinical Documentation with AI: Reducing Errors, Improving Interoperability, and Supporting Real-Time Note-Taking

The increasing administrative burden in healthcare, particularly in clinical documentation, has driven research into artificial intelligence (AI)-powered solutions to enhance transcription accuracy, improve interoperability in electronic health record (EHR) systems, and enable real-time clinical note generation. This study systematically reviewed ۱۴ relevant studies to evaluate the role of AI technologies, including natural language processing (NLP) and automatic speech recognition (ASR), in addressing these challenges. Results showed that AI significantly reduces transcription errors when combining ASR with domain-specific NLP models, such as ClinicalBERT, and fine-tuned large language models (LLMs) like GPT-۴, by improving context understanding and terminology accuracy. Real-time clinical note generation was commonly achieved using hybrid extractive-abstractive summarization techniques and structured templates, such as SOAP (Subjective, Objective, Assessment, Plan) notes, with enhanced usability and time savings demonstrated in clinical settings. Additionally, systems employing semantic knowledge graphs and ontologies (e.g., UMLS) facilitated greater standardization and interoperability between disparate EHR systems. However, critical challenges were noted with hallucination risks in text generation, data privacy concerns, and low clinician trust in automated tools. Evaluation metrics such as ROUGE, BERTScore, and domain-specific measures (e.g., DeepScore) revealed variability in the quality and factual consistency of AI-generated notes. This review highlights the potential of AI to alleviate documentation burdens, though further advances in real-time integration, accuracy, and user acceptability are required for widespread adoption in healthcare environments.

کلیدواژه ها:

Clinical Documentation ، Artificial intelligence ، Electronic health record ، Large Language Models ، Natural Language Processing