Data-driven identification of post-acute SARS-CoV-2 infection subphenotypes.

Area of Science:

• Clinical Medicine
• Infectious Diseases
• Data Science & Machine Learning

Background:

• Post-acute sequelae of SARS-CoV-2 infection (PASC), or long COVID, encompasses a wide range of persistent or new symptoms after acute infection.
• Previous research often examined PASC conditions individually, lacking insight into co-occurring symptoms and patient subgroupings.

Purpose of the Study:

• To identify and characterize distinct subphenotypes of PASC using machine learning on electronic health record data.
• To investigate the heterogeneity of PASC and its association with patient demographics, pre-existing conditions, and acute infection severity.

Main Methods:

• Leveraged electronic health record data from two large cohorts (INSIGHT and OneFlorida+) comprising over 34,000 SARS-CoV-2 infected patients.
• Employed machine learning to analyze newly incident diagnoses (30-180 days post-infection) across over 137 symptoms and conditions.
• Identified and validated four reproducible PASC subphenotypes.

Main Results:

• Four distinct PASC subphenotypes were identified: cardiac and renal; respiratory, sleep, and anxiety; musculoskeletal and nervous system; and digestive and respiratory system.
• These subphenotypes represented significant proportions of patients in both development and validation cohorts (e.g., cardiac/renal: 33.75% and 25.43%).
• Subphenotypes showed associations with distinct patient demographics, prior health conditions, and acute SARS-CoV-2 infection severity.

Conclusions:

• PASC is heterogeneous, presenting as distinct, reproducible subphenotypes rather than a monolithic condition.
• These findings provide critical insights into the diverse manifestations of long COVID.
• The identified subphenotypes may facilitate stratified decision-making and personalized management strategies for PASC.