Finding Long-COVID: Temporal Topic Modeling of Electronic Health Records from the N3C and RECOVER Programs

Area of Science:

• Epidemiology
• Computational Biology
• Clinical Medicine

Background:

• Post-Acute Sequelae of SARS-CoV-2 infection (PASC), or Long-COVID, presents diverse and complex symptoms following initial infection.
• The underlying mechanisms and clinical trajectories of PASC remain incompletely understood.
• Large-scale data analysis is crucial for deciphering the multifaceted nature of PASC.

Purpose of the Study:

• To identify clinical conditions and phenotypes significantly associated with SARS-CoV-2 infection.
• To explore patient-specific factors influencing PASC development and presentation.
• To enhance diagnostic capabilities and mechanistic understanding of Long-COVID.

Main Methods:

• Clustering of over 600 million condition diagnoses from 14 million patients within the National COVID Cohort Collaborative (N3C) database.
• Development of a novel method to associate patient clinical trajectories with identified clusters over time.
• Comparative analysis of clinical outcomes between COVID-19 patients and control groups.

Main Results:

• Hundreds of detailed clinical phenotypes were generated, revealing numerous conditions with increased incidence post-COVID-19.
• Identification of phenotypes specific to patient sex, age, infection wave, and PASC diagnosis status.
• Validation of known PASC symptoms and discovery of novel associations through high-resolution data analysis.

Conclusions:

• The large-scale analysis provides a granular view of conditions associated with PASC.
• Findings highlight the heterogeneity of Long-COVID and identify key demographic and temporal factors.
• This research offers potential pathways for improved Long-COVID diagnostics and a deeper mechanistic understanding.