An HHS spokesperson did not respond to a request for comment.
The study used artificial intelligence to comb through electronic medical records of patients at 58 hospitals in four regions of the country: New England, Southeast Texas, Southern California, and Western Pennsylvania. The hospitals in New England included the two flagship hospitals of MGB, Massachusetts General Hospital and Brigham and Women’s Hospital.
Deploying a newly developed algorithm, the researchers hunted for chronic conditions that were likely related to long COVID but might not have led to that diagnosis. Those conditions, which had no other apparent explanation in patients’ medical histories, included sudden onset of prediabetes, heart problems, neurological disorders, persistent fatigue, and chronic pain.
“There’s a good possibility the patients themselves didn’t know this condition could be long COVID,” said Hossein Estiri, director of the Clinical Augmented Intelligence Group at MGH. He was the corresponding author on the study published Wednesday in the journal JAMA Network Open.
Health care systems have a federally approved code for a diagnosis of long COVID. But Estiri said that doctors use it in less than 7 percent of cases because symptoms of the disorder vary widely and long COVID has no treatment.
“If you go to the doctor with chronic pain, it doesn’t really matter to the doctor if the pain is COVID-induced,” said Estiri, an associate professor of medicine at Harvard Medical School. “They just try to treat the chronic pain.”
The Centers for Disease Control and Prevention, which is part of HHS, defines long COVID as a chronic condition that occurs after infection with the virus that causes COVID-19 and is present for at least three months. Long COVID includes a wide range of symptoms that may improve, worsen, or persist.
Although many people associate long COVID with the waves of infections that occurred early in the pandemic, the researchers found that cases of the chronic condition increased through mid-2025 in all four regions that they studied — and are likely still increasing.
Estiri was the senior author of another AI-powered study of long COVID, in 2024, that found an even higher prevalence of long COVID — one in four Americans. Estiri said that study used a broader definition of the disorder, when the scientific understanding of long COVID was less advanced, and only examined the electronic medical records of MGB patients.
That study drew skepticism from Dr. Eric Topol, executive vice president of Scripps Research in San Diego, a biomedical research center, as did the latest paper.
“The estimate does seem high to me,” Topol, a cardiologist, said in an email. He said many of the cases of long COVID that MGB researchers identified may have featured symptoms that were mild or didn’t last long. “I believe the real number of people with [long COVID], not recovered, is a single-digit percentage.”
However, Dr. Shawn Murphy, the coauthor of the latest study and a specialist in biomedical informatics who left MGB recently to become chief research information officer at University of Washington Medicine, said the research team was actually a “little conservative” in its calculations.
If the federal government ultimately agrees that long COVID is more widespread than thought, he said, it could prompt the Centers for Medicare and Medicaid Services to expand insurance coverage for treatment of the condition and spur private insurers to follow suit.
Jonathan Saltzman can be reached at jonathan.saltzman@globe.com.