Summary: Mass General Brigham researchers have developed an innovative AI algorithm that identifies long COVID cases by analyzing electronic health records. The tool suggests that long COVID affects 22.8% of the population—significantly higher than previous estimates—and helps reduce diagnostic biases by taking a more comprehensive approach to symptom analysis.
Journal: Med, November 8, 2024, DOI: 10.1016/j.medj.2024.10.009 | Reading time: 5 minutes
Unraveling the Long COVID Mystery
For healthcare providers, identifying long COVID cases has been like searching for needles in a haystack of symptoms. Now, artificial intelligence might make that search considerably easier.
“Our AI tool could turn a foggy diagnostic process into something sharp and focused, giving clinicians the power to make sense of a challenging condition,” says Dr. Hossein Estiri, senior author and head of AI Research at Mass General Brigham’s Center for AI and Biomedical Informatics.
The research team analyzed health records from nearly 300,000 patients across 14 hospitals and 20 community health centers in Massachusetts, developing an algorithm that looks beyond simple diagnostic codes to understand the full picture of a patient’s health journey.
A More Accurate Picture
The new approach, called “precision phenotyping,” examines individual records to differentiate long COVID symptoms from other conditions. Rather than relying on a single diagnostic code, the AI considers a patient’s entire medical history to rule out alternative explanations for symptoms.
Dr. Alaleh Azhir, an internal medicine resident at Brigham Women’s Hospital, explains the practical impact: “Physicians are often faced with having to wade through a tangled web of symptoms and medical histories, unsure of which threads to pull, while balancing busy caseloads. Having a tool powered by AI that can methodically do it for them could be a game-changer.”
Addressing Healthcare Disparities
The research reveals that current diagnostic methods may significantly undercount long COVID cases. While previous studies suggested about 7% of the population experiences long COVID, the new AI tool indicates a much higher rate of 22.8%—a figure that aligns more closely with national trends.
Importantly, the tool shows promise in reducing healthcare disparities. The algorithm identified patients across demographic groups more evenly than traditional diagnostic methods, which often skew toward populations with better healthcare access.
Looking Ahead
The researchers acknowledge several limitations in their study. The algorithm may miss cases where COVID-19 worsened pre-existing conditions, and recent declines in COVID-19 testing make it harder to pinpoint initial infection dates. The study was also limited to Massachusetts patients.
Future research will explore the algorithm’s effectiveness in specific patient populations, such as those with COPD or diabetes. The team plans to make the algorithm publicly available for healthcare systems worldwide.
Key Terms
• PASC: Post-Acute Sequelae of SARS-CoV-2 infection, the medical term for long COVID
• Precision Phenotyping: A method that analyzes individual health records to identify patterns of symptoms and conditions
• Diagnosis of Exclusion: A diagnosis reached by ruling out all other possible explanations for symptoms
• Algorithm: A step-by-step procedure or set of rules followed by a computer to solve a problem
Test Your Knowledge
1. What percentage of the population was found to have long COVID using the new AI tool?
Answer: 22.8%
2. How many patients’ records were analyzed in the study?
Answer: Nearly 300,000 patients
3. What is the main advantage of the AI tool over traditional diagnostic methods?
Answer: It can analyze entire medical histories and rule out alternative explanations for symptoms
4. What is one limitation of the study?
Answer: It was limited to patients in Massachusetts and may miss cases where COVID-19 worsened pre-existing conditions
Enjoy this story? Get our newsletter! https://scienceblog.substack.com/
