An NLP-Based System for Automatic Summarization of Electronic Health Records to Support Clinical Decision-Making

Abstract

The rapid growth of Electronic Health Records (EHRs) has created significant challenges in clinical information management and timely decision-making. This study presents the design and prototype-level evaluation of a Natural Language Processing (NLP)-based system for automatic summarization of EHRs to support Clinical Decision-Making. The framework integrates text preprocessing, Named Entity Recognition (NER), and hybrid summarization techniques, combining extractive methods (TF-IDF and TextRank) with transformer-based abstractive models (BART and T5). De-identified clinical notes from the publicly available MIMIC-IV dataset were used for experimental validation. A subset of 1,000 clinical notes was selected and divided into training (80%) and test (20%) sets. Performance was evaluated using ROUGE and BLEU metrics and compared with baseline extractive methods to assess improvements achieved by transformer-based models. The system achieved an average ROUGE-1 score of 0.78 and a BLEU score of 0.69 on the experimental subset. These 
findings represent proof-of-concept results rather than large-scale performance benchmarks. The system successfully extracted clinically relevant information, including diagnoses, medications, and treatment 
outcomes, and generated coherent summaries. Implemented in Python using deep learning libraries, the prototype demonstrates the potential of NLP-driven summarization to reduce information overload and improve 
clinical workflow efficiency, although further large-scale validation is required.