A Symptom-Based Natural Language Processing Surveillance Pipeline for Post-COVID-19 Patients

Area of Science:

• Medical Informatics
• Computational Linguistics
• Public Health

Background:

• Post-acute sequelae of SARS CoV-2 (PASC) present a significant societal burden due to prolonged symptoms, healthcare costs, and disabilities.
• Accurate identification of PASC cases is essential for effective patient management and public health monitoring.

Purpose of the Study:

• To develop and evaluate a natural language processing (NLP) pipeline for identifying PASC symptoms.
• To estimate the proportion of suspected PASC cases using the developed NLP pipeline.
• To assess the usability of a dashboard designed for visualizing aggregated PASC symptom data.

Main Methods:

• Implementation of an NLP pipeline to process clinical data and identify PASC symptoms.
• Manual case review for validation and estimation of PASC incidence.
• Development and usability testing (System Usability Scale) of a PASC symptom dashboard.

Main Results:

• The NLP pipeline successfully identified PASC symptoms and estimated a sample incidence of 13%, consistent with population estimates.
• A significant number of cases were classified as indeterminate, highlighting diagnostic challenges even for experienced clinicians.
• The PASC symptom dashboard received positive feedback regarding its potential utility for clinical monitoring.

Conclusions:

• The developed NLP pipeline offers a valuable tool for monitoring post-COVID-19 patients and identifying PASC.
• Challenges remain in definitively classifying PASC cases, necessitating further refinement of diagnostic criteria and tools.
• The dashboard facilitates the visualization of PASC symptom data, supporting clinical decision-making and research.