Valx: A System for Extracting and Structuring Numeric Lab Test Comparison Statements from Text.

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Biomedical Data Science

Background:

• Clinical trial eligibility criteria often contain complex numeric lab test comparison statements.
• Manual extraction and structuring of this data is time-consuming and prone to errors.
• Automating this process can significantly enhance clinical trial data management and analysis.

Purpose of the Study:

• To develop and evaluate an automated method, named Valx, for extracting and structuring numeric lab test comparison statements from text.
• To assess Valx's performance using clinical trial eligibility criteria data.

Main Methods:

• Valx utilizes the Unified Medical Language System (UMLS) and internet-acquired domain knowledge.
• It involves seven steps: text preprocessing, extraction of numeric values, units, and comparison operators, variable identification, association, filtering, unit normalization, and verification.
• A reference standard was created through consensus annotation by three raters for HbA1c and glucose comparison statements in Type 1 and Type 2 diabetes trials.

Main Results:

• Valx achieved high precision, recall, and F-measure scores for structuring HbA1c comparison statements (e.g., 99.6% precision, 98.1% recall for Type 1 diabetes trials).
• Performance for glucose comparison statements was also strong, with F-measures of 96.1% for Type 1 and 92.3% for Type 2 diabetes trials.
• The method demonstrated robust accuracy in extracting and structuring lab test comparison data.

Conclusions:

• Valx is an effective tool for extracting and structuring free-text lab test comparison statements within clinical trial summaries.
• Further research is recommended to evaluate Valx's generalizability to other types of clinical text.
• The open-source nature of Valx encourages community-driven evaluation and improvement.