Identification of misspelled words without a comprehensive dictionary using prevalence analysis

Area of Science:

• Medical informatics
• Natural language processing
• Clinical documentation

Background:

• Misspellings in medical documents hinder information retrieval.
• Accurate identification of misspellings is crucial for data integrity and clinical decision-making.

Purpose of the Study:

• To evaluate an algorithm for identifying misspelled words using prevalence analysis in medical text.
• To assess the algorithm's accuracy in detecting misspellings of anti-hypertensive medication names.

Main Methods:

• Algorithm evaluated on 2,000 potentially misspelled words from narrative medical documents.
• Prevalence ratios computed by software comparing potentially misspelled words to correct forms.
• Software results compared against manual review by an independent expert.

Main Results:

• Area under the ROC curve was 0.96, indicating high accuracy.
• Sensitivity (99.25%), specificity (89.72%), and positive predictive value (82.9%) achieved at a specific prevalence ratio threshold.
• Highest F-measure of 0.903 demonstrated the algorithm's effectiveness.

Conclusions:

• Prevalence analysis is a highly accurate method for identifying misspellings in medical documents.
• The developed algorithm can effectively detect and potentially correct errors in clinical notes.
• Improved misspelling detection enhances medical information retrieval and data quality.