Precision phenotyping of curating research cohorts of patients with unexplained post-acute sequelae of COVID-19 showed new insights. An AI-assisted review showed more than 1 in 5 Americans are likely to suffer from long COVID. The study published on November 8 in the journal Med suggested that nearly 23% of U.S. adults experienced symptoms of long COVID.
This estimate is significantly higher than other studies, one suggests that only 7% of people may experience long COVID according to researchers. Senior researcher Hossein Estiri, who led AI research at Mass General Brigham in Boston stated that the questions about the true burden of long COVID has been challenging to answer thus far and now seem to be more within reach.
For the study, researchers developed an AI tool capable of sorting mounds of electronic health records to identify the often subtle symptoms of long-term COVID-19. The symptoms can involve any body system and include fatigue, chronic cough, trouble with the heart, and brain fog. Usually, these are developed a week or two after a person recovers from their first bout with COVID-19 infection.
“Our AI tool could help take what might be a fuzzy diagnostic process and shine a laser beam on that condition, empowering clinicians to understand a difficult condition,” Estiri said in a Mass General news release. Researchers said the AI particularly looks for symptoms that can’t be accounted for by someone’s medical history, have lasted for at least two months, and started after a person tested positive for COVID.
Then, for example, the AI can rule out long COVID as a cause of shortness of breath-instead, it could be due to pre-existing heart failure or asthma and hence settle for a screening test result based on just that.
It means that often the physicians have to walk too through the world of a tangled web of symptoms and medical histories, not knowing which thread to pull while trying to maintain busy caseloads. “Whether this is done with a tool powered by AI that can methodically do it for them, will be a game changer, said lead researcher,” Dr. Alaleh Azhir, an Internal Medicine resident at Brigham and Women’s.
Using these parameters, the AI estimated that nearly 23% of Americans have long COVID aligning more closely with national trends. Doctors and healthcare systems will be able to use and test the AI, the researchers plan to release it publicly on open access.
The study included 85,364 COVID-19 cases, 170,497 post-pandemic controls (matched 1:2), and 39,817 pre-pandemic controls meeting the EHR longitudinal continuity threshold. The mean age was ~54 years and 62-63% of participants were female across groups. White participants predominated (~71-74%). PASC phenotyping identified 24,360 long-haulers (28.5% of cases) with systemic sequelae (41.2%) being the most common. Statistically significant odds of PASC development were identified across demographics, with p-values <0.05 for differences in systemic, neurological, and mental health PASC among women. Precision for the PASC algorithm was 79.9% and Cohen’s kappa for inter-rater reliability was 0.76 indicating substantial agreement.
Reference: Azhir A, Hügel J, Tian J, et al. Precision phenotyping for curating research cohorts of patients with unexplained post-acute sequelae of COVID-19. Med. Published online November 2024.
doi: https://doi.org/10.1016/j.medj.2024.10.009
