Epidemiologic information discovery from open-access COVID-19 case reports via pretrained language model

Area of Science:

• Epidemiology
• Computational Biology
• Natural Language Processing

Background:

• Open-access data is valuable for epidemiological research but requires extensive curation.
• COVID-19 case reports are often unstructured natural language text, hindering analysis.
• Automating information extraction from these reports is crucial for efficient research.

Purpose of the Study:

• To develop a computational framework for automatic extraction of epidemiological information from COVID-19 case reports.
• To improve the efficiency and accuracy of epidemiological data curation.
• To provide a real-time platform for estimating key epidemiological statistics.

Main Methods:

• Coupling a deep neural network language model with an optimized data annotation strategy.
• Training the model on COVID-19 case reports from mainland China.
• Developing an open-access online platform for disseminating the algorithm.

Main Results:

• The framework achieved an 80% matching rate with gold-standard manual coding.
• Outperformed existing state-of-the-art deep learning models in extracting epidemiological information.
• Demonstrated significant reduction in data curation effort.

Conclusions:

• The proposed computational framework effectively automates the extraction of epidemiological information from unstructured COVID-19 case reports.
• The developed online platform facilitates real-time estimation of epidemiological statistics, reducing manual curation burdens.
• This approach enhances the utility of open-access data for epidemiological research.