An efficient prototype method to identify and correct misspellings in clinical text.

Area of Science:

• Clinical Natural Language Processing (NLP)
• Medical Informatics
• Computational Linguistics

Background:

• Misspellings in clinical free text pose significant challenges for accurate NLP.
• Effective identification and correction of these errors are crucial for reliable clinical data analysis.

Purpose of the Study:

• To develop and evaluate a prototype spelling analysis method for identifying and correcting misspellings in clinical free text.
• To assess the performance of the method on diverse clinical corpora, including surgical pathology reports and emergency department notes.

Main Methods:

• Implementation of a prototype method combining Word2Vec, Levenshtein edit distance, a lexical resource, and corpus term frequencies.
• Processing of two distinct corpora: surgical pathology reports and emergency department progress/visit notes.
• Performance evaluation using positive predictive value and detailed error analysis of false positives and spelling error types.

Main Results:

• The prototype method achieved high positive predictive values (0.9057 for pathology, 0.8979 for ED notes) in identifying and correcting misspellings.
• False positive rates varied between the corpora, with similar error types but higher error frequency in emergency department notes.
• Analysis revealed that emergency department notes contained over four times more spelling errors than pathology reports.

Conclusions:

• The developed spelling analysis method demonstrates strong performance in identifying and correcting misspellings within clinical notes.
• The findings suggest the method's potential for broad application across various clinical document types.
• This approach can enhance the accuracy and reliability of clinical NLP applications.