CCMapper: An adaptive NLP-based free-text chief complaint mapping algorithm.

Area of Science:

• Emergency Medicine
• Natural Language Processing
• Health Informatics

Background:

• Chief complaints (CCs) are crucial initial data in emergency departments (EDs).
• Manual categorization of free-text CCs is time-consuming and prone to errors.
• Standardizing CC data is essential for clinical decision support and research.

Purpose of the Study:

• To develop and validate a heuristic methodology for automatically mapping free-text chief complaints (CCs) into a structured list.
• To create a natural language processing-based algorithm, CCMapper, for efficient and accurate CC categorization.
• To assess the performance of CCMapper in a real-world ED setting.

Main Methods:

• A comprehensive structured list of CCs was created by experienced Emergency Medicine (EM) physicians.
• A natural language processing (NLP) algorithm, CCMapper, was developed using this structured list.
• CCMapper was trained and validated on free-text CC data from the Mayo Clinic ED, employing a consensus-based validation approach.

Main Results:

• High inter-physician agreement (κ = 0.958) with minimal human error (<2%) was achieved during manual mapping.
• CCMapper demonstrated high performance on the validation set: 94.2% sensitivity, 99.8% specificity, and 94.7% F-score.
• For multi-CC entries, CCMapper achieved 82.3% sensitivity, 99.1% specificity, and an 82.3% F-score.

Conclusions:

• CCMapper offers a simple, high-performing solution for automatically categorizing ED chief complaints.
• The algorithm's ability to incorporate new data facilitates its adoption by other EDs for improved clinical decision-making.
• CCMapper can aid in developing predictive models for critical ED events, such as ICU admissions.