Generative Artificial Intelligence Applications in Electronic Medical Records and Health Information Management-An Updated Review

Abstract

Background: Electronic Medical Records (EMRs) are central to modern healthcare delivery but generate large volumes of complex, unstructured clinical data that are difficult to process and utilize effectively. Generative Artificial Intelligence (GenAI), particularly large language models, has emerged as a promising technology for enhancing EMR functionality across multiple clinical and administrative domains. Aim: This review aims to explore and categorize the applications of GenAI within EMR systems, assess its performance across different healthcare tasks, and identify its benefits, limitations, and implications for clinical practice. Methods: A scoping review approach was used to analyze studies examining GenAI applications in EMRs. A total of 55 studies were reviewed and categorized into key thematic areas including data manipulation, patient communication, clinical decision-making, clinical prediction, summarization, and other emerging applications. Findings were synthesized narratively to identify patterns of use, performance outcomes, and reported limitations. Results: GenAI demonstrated strong performance in data manipulation, clinical summarization, and patient communication, improving efficiency and reducing clinician workload. In clinical prediction and certain structured tasks, GenAI showed performance comparable to or exceeding traditional machine learning models. However, clinical decision-making applications revealed limitations, including inaccurate recommendations, hallucinations, and unsafe outputs in high-risk settings. Concerns regarding bias, interpretability, data privacy, and legal accountability were consistently reported across studies. Conclusion: GenAI shows significant potential to enhance EMR systems by improving efficiency, accessibility, and clinical support functions. However, its current limitations restrict its use as an autonomous clinical tool. Safe integration requires robust governance, continuous validation, and human oversight.