Extracting Medical Information from Paper COVID-19 Assessment Forms

Area of Science:

• Medical Informatics
• Data Science
• Public Health

Background:

• Healthcare workers faced challenges interacting with electronic health records during the COVID-19 pandemic due to personal protective equipment.
• Digitizing paper-based medical forms is crucial for data management and analysis, especially during public health crises.

Purpose of the Study:

• To evaluate the effectiveness of optical mark recognition (OMR) combined with a novel user interface (UI) and crowdsourced data validation for digitizing COVID-19 assessment forms.
• To assess the accuracy, efficiency, and reliability of this automated data extraction method.

Main Methods:

• Developed an OMR/optical character recognition (OCR) system to process 2,814 paper COVID-19 assessment forms.
• Implemented a novel UI for efficient data validation, displaying scanned forms alongside editable web forms.
• Utilized crowdsourced participants for data validation and calculated error rates and validation times.

Main Results:

• The OMR/OCR system achieved an average accuracy of 70% and a median accuracy of 78% when compared to crowd validation.
• Crowdsourced validation occurred at a mean rate of 157 seconds per document, processing approximately 108 documents daily.
• High interobserver agreement was observed (97% with narrative fields, 98% without).

Conclusions:

• The integration of OMR/OCR technology, a novel UI, and crowdsourcing provides an efficient solution for extracting data from large volumes of paper medical documents.
• This method is particularly valuable for managing data during public health emergencies like the COVID-19 pandemic, overcoming challenges posed by PPE use.
• The study demonstrates the potential of these technologies to streamline medical data digitization and validation processes.