In a groundbreaking study, researchers at Mass General Brigham have developed an AI tool capable of detecting long COVID in approximately 22.8% of patients, a substantial increase from previous estimates of around 7%. By analyzing data from nearly 300,000 patients across 14 hospitals, the AI tool, based on “precision phenotyping,” was able to pinpoint cases of long COVID by excluding symptoms linked to other health conditions.

The AI tool sifts through patients' medical records to isolate symptoms directly associated with SARS-CoV-2, rather than pre-existing conditions like asthma or heart failure. Only when no other cause is identified does the algorithm diagnose long COVID. This targeted approach, according to the research team, improves diagnostic accuracy by about 3% over traditional ICD-10 codes, which can carry biases toward patients with better access to healthcare.

The AI-based algorithm reportedly offers a broader representation of the Massachusetts population, ensuring that diagnostic rates for long COVID reflect the state’s demographic makeup. Researchers note that conventional diagnostic codes often favor those with consistent healthcare access, while this tool may improve representation of underserved communities.

Dr. Hossein Estiri, a senior author of the study and AI research head at Harvard’s Center for AI and Biomedical Informatics, remarked that the tool could make diagnosing long COVID more precise. “Our AI tool could turn a foggy diagnostic process into something sharp and focused,” he said, adding that it might enable healthcare providers to identify and treat long COVID with greater accuracy.

Estiri’s team plans to release the algorithm for public use, which could potentially benefit healthcare systems globally. The tool's development also opens doors to more comprehensive research on long COVID’s underlying genetic and biochemical causes. Additionally, it may allow for targeted studies within specific populations, such as those with pre-existing conditions like diabetes or COPD.

While the tool represents a step forward, researchers note certain limitations. The data used may lack details present in physicians' clinical notes, and the algorithm might miss cases where COVID-19 worsened a prior condition, such as chronic obstructive pulmonary disease (COPD). Reduced COVID-19 testing in recent years also poses challenges in pinpointing when patients were initially infected.

The study, published in MedRxiv, received funding from the National Institutes of Health and other U.S. and international organizations, supporting the potential for this AI tool to improve long COVID diagnostics and help answer longstanding questions about its impact.