UTP: A Unified Term Presentation Tool for Clinical Textual Data Using Pattern-Matching Rules and Dictionary-Based Ontologies

Abstract

Clinical textual data such as discharge summaries and chief complaints summarize the patient’s medical history and treatment plan. These unstructured complex data include ambiguous medical terms, abbreviations, diagnostic investigation values and dates which pose significant challenges for human and machine learning tasks to process them. This paper proposes a novel approach that transforms clinical text with different writing styles into a uniform and standard presentation using pattern-matching rules and JSON dictionary-based ontologies. The main goal of the proposed approach is to improve the communication between healthcare parties or professionals by improving the quality of the clinical textual data and reducing its heterogeneity and ambiguity. In addition, this data quality improvement enhances the performance of machine learning downstream tasks. Our approach identifies the abbreviations, medical terms, negations, dates, and investigation values from the unstructured textual data. Then, it replaces the detected entities with their corresponding unified and normalized presentation based on pattern-matching rules that relies on the linguistic features, pattern-matching rules, and JSON dictionaries. The inductive content analysis method was followed to generate the pattern-matching rules with the help of a medical team. Its role is to validate the accuracy of the detected entities. Finally, the proposed approach was applied to a massive real-world dataset in order to evaluate its impact on the performance of various machine learning models. The results show a significant improvement in performance after preprocessing the clinical textual data using our approach.