A fast, accurate, and generalisable heuristic-based negation detection algorithm for clinical text.

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Clinical informatics

Background:

• Negation detection is crucial in clinical settings to ascertain the presence or absence of medical findings.
• Existing rule-based negation detection systems, while utilizing dependency graphs, are often complex, time-consuming to develop, and lack generalizability.

Purpose of the Study:

• To investigate a heuristic approach for negation detection using grammatical distance in typed dependency graphs as a simpler alternative to complex rule-based systems.
• To implement and evaluate a novel algorithm for negation detection in clinical text.

Main Methods:

• Developed a heuristic algorithm based on grammatical distance from a negatory construct within a typed dependency graph.
• Created two testing corpora from MIMIC-III clinical data and hypertrophic cardiomyopathy patient documents.
• Established gold-standard validation datasets through human annotation and error analysis.
• Compared the algorithm's performance against four other rule-based systems.

Main Results:

• The heuristic algorithm achieved the best performance by f-measure on the MIMIC-III dataset.
• It demonstrated comparable performance to syntactic negation detection systems on the hypertrophic cardiomyopathy dataset.
• The algorithm was the fastest among the dependency-based negation systems evaluated.
• It proved to be a stable method requiring minimal adaptation across datasets.

Conclusions:

• A heuristic approach to dependency-based negation detection offers a powerful, stable, and efficient method for clinical text mining.
• This algorithm can serve as a drop-in replacement or augmentation for existing negation detection approaches, especially when rapid implementation and minimal adaptation are needed.