New AI Tool Identifies Additional Cases of Long COVID from Patient Health Records

Researchers at Mass General Brigham have unveiled an AI-powered tool designed to help clinicians identify undiagnosed cases of long COVID, a complex and often under-recognized condition. Leveraging electronic health records (EHRs), this tool can sift through patient data to identify persistent symptoms that could signal long COVID, such as fatigue, chronic cough, and cognitive difficulties. The AI’s enhanced capability to differentiate these symptoms from other conditions could provide critical support to clinicians faced with the diagnostic challenges of long COVID.

Precision Phenotyping Method Enhances Diagnosis

The tool applies a unique approach known as “precision phenotyping,” allowing it to track patient symptoms over time and isolate those likely stemming from a COVID-19 infection. Developed under the guidance of Dr. Hossein Estiri, head of AI Research at Mass General Brigham’s Center for AI and Biomedical Informatics, the AI algorithm was trained on the de-identified records of nearly 300,000 patients from 14 hospitals and 20 community health centers. By excluding symptoms linked to pre-existing conditions, the tool narrows in on cases where no other diagnosis explains the symptoms, thus improving long COVID detection accuracy by approximately 3% compared to conventional diagnostic codes.

Addressing Diagnostic Bias and Expanding Access to Care

This AI tool could significantly reduce the diagnostic bias associated with current methods, which often rely on specific codes or single clinical encounters, disproportionately affecting individuals with limited access to healthcare. By examining patient data comprehensively, the AI aims to create a more inclusive representation of long COVID cases. Preliminary findings suggest that nearly 23% of people with prior COVID-19 infections may experience long COVID—a much higher prevalence than earlier studies estimated. This broader, data-driven scope ensures a more accurate and inclusive view of the condition, capturing cases that would otherwise be missed in marginalized populations.

Implications for Future Research and Global Access

Beyond clinical use, the AI tool is positioned to support future research into the genetic and biochemical causes of long COVID, potentially illuminating the condition’s various subtypes. Researchers plan to make the algorithm publicly available, enabling healthcare systems worldwide to apply it across diverse patient populations, which could standardize and improve long COVID diagnostics globally. By identifying cases more equitably and accurately, this tool could set a new standard for long COVID care and open avenues for specialized studies in specific patient groups, such as those with chronic conditions like COPD or diabetes.

As the true impact of long COVID continues to emerge, this AI innovation provides a pathway to understanding and addressing the pandemic’s enduring effects.