Detecting false-positive disease references in veterinary clinical notes without manual annotations.

Area of Science:

• Veterinary medicine
• Clinical informatics
• Natural Language Processing

Background:

• Clinical notes frequently contain disease references not representing actual diagnoses, such as differential diagnoses or ruled-out conditions.
• Manual annotation of clinical notes for training disease-disambiguation models is time-consuming and labor-intensive.

Purpose of the Study:

• To develop a method for generating large-scale training datasets for disease reference disambiguation without manual annotation.
• To achieve state-of-the-art performance in classifying disease references in veterinary clinical notes.

Main Methods:

• Utilized a large dataset of veterinary clinical notes.
• Developed a novel approach to automatically create training data, avoiding manual annotation.
• Employed a bidirectional long short-term memory (BiLSTM) model for classification.

Main Results:

• Successfully generated very large training sets automatically.
• Achieved state-of-the-art classification performance.
• Demonstrated the model's ability to distinguish between actual disease diagnoses and other disease references in clinical notes.

Conclusions:

• Automated training data generation is feasible and effective for building disease-disambiguation classifiers.
• The proposed method significantly enhances the efficiency of creating training data for clinical NLP tasks.
• This approach can improve the accuracy of extracting diagnostic information from electronic health records.