NATURAL LANGUAGE PROCESSING IN NEUROLOGICAL CLINICAL DOCUMENTATION: AUTOMATED CODING, INFORMATION EXTRACTION, AND CLINICAL DECISION SUPPORT

Abstract

Neurological clinical documentation — encompassing admission notes, discharge summaries, neurology consultation reports, and operative records — represents a rich source of clinically relevant information that is largely inaccessible to computational analysis due to its unstructured natural language format. Natural language processing (NLP) — the computational analysis of human language — has advanced dramatically with the development of large language models (LLMs) including BERT, GPT-4, and domain-specific variants, enabling automated extraction of clinically relevant information from neurological text with accuracy approaching human expert performance. This review analyses 38 studies of NLP applied to neurological clinical documentation published between 2018 and 2024, evaluating tasks including automated ICD coding, clinical entity extraction (symptoms, diagnoses, medications), phenotyping of neurological conditions from electronic health records, and clinical decision support. NLP-based automated coding achieves F1 scores of 0.82–0.94 for neurological diagnoses, clinical entity extraction achieves F1 of 0.88–0.96, and phenotyping algorithms based on NLP demonstrate sensitivity of 85–95% for identifying neurological patient cohorts. Implications for neurology documentation efficiency and clinical informatics in Uzbekistan are discussed.