AI Tool Enhances Long COVID-19 Diagnosis, Identifies Gaps in Care

Researchers at Mass General Brigham have developed an AI tool that enhances long COVID-19 diagnosis accuracy by analyzing electronic health records, which uncovered a significant underdiagnosis compared to traditional methods. This advancement could improve care for patients and support further research into the condition’s subtypes and underlying causes.

The study highlights the effectiveness of precision phenotyping, a method that uses data analytics to more accurately diagnose long COVID-19 by analyzing de-identified clinical data from nearly 300,000 patients across 14 hospitals and 20 community health centers in the United States. The AI-driven tool uses this data to identify long COVID-19 symptoms — such as fatigue, chronic cough, and brain fog — and differentiate them from pre-existing conditions. According to Mass General Brigham News, symptoms must persist for at least two months after infection to qualify as long COVID-19.

Long COVID-19 impacts vital organ systems such as the heart, lungs, immune system, and neurological functions, as reported by MBN. 

"Physicians often face a complex array of symptoms and medical histories, making diagnosis challenging amidst their busy caseloads," says Alaleh Azhir, Co-lead author of the study. "An AI tool that methodically analyzes this data could transform patient care by streamlining the diagnostic process."

The AI tool reveals a significant gap in long COVID diagnoses, identifying the condition in 22.8% of cases, while traditional methods diagnosed 7% of cases. A separate study published in Nature Science estimates that around 400 million people worldwide have been affected by long COVID-19. 

Despite its prevalence, treatment options remain limited, and the absence of clinical trial data makes it difficult for healthcare providers to make evidence-based treatment decisions.

While an advancement, this tool has major limitations including its reliance on the quality of health records. The decline in COVID-19 testing rates further complicated the accurate identification of initial infections, posing a risk of misdiagnosis. Future research will focus on refining the AI algorithm and testing its accuracy with patient groups that have underlying conditions like COPD or diabetes.