AI tool spots long COVID in electronic health records
Researchers in the US have developed an artificial intelligence algorithm that can sift through electronic health records (EHRs) and help physicians to detect undiagnosed cases of long COVID.
Along with identifying people who should be receiving care for the potentially debilitating condition, the algorithm could also be used to try to find the genetic and biochemical factors behind the still-mysterious condition, which causes a range of symptoms including extreme tiredness, shortness of breath, chest pain, problems with memory, difficulty sleeping, heart palpitations, and dizziness.
According to the US Centers for Disease Control and Prevention (CDC), approximately 7.5% of the adult population of the US have symptoms of long COVID, which works out at 24.75 million individuals.
The new algorithm – developed by investigators at Mass General Brigham and published in the journal Med – was trained on de-identified data from EHRs of nearly 300,000 patients across 14 hospitals and 20 community health centres.
It uses an approach known as 'precision phenotyping', which goes through individual records to identify symptoms and conditions linked to COVID-19 and tracks them over time to differentiate them from other illnesses like asthma or heart failure. The algorithm identified a cohort of over 24,000 patients with 79.9% precision, according to the paper, and also suggested that the risk of long COVID increases with subsequent infections.
"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," said senior author Hossein Estiri, who is head of AI research at the Center for AI and Biomedical Informatics of the Learning Healthcare System (CAIBILS) at Mass General Brigham, as well as an associate professor of medicine at Harvard Medical School.
"With this work, we may finally be able to see long COVID for what it truly is – and more importantly, how to treat it," he continued, noting that the AI seems to be about 3% more accurate than current diagnostic approaches based on the International Classification of Diseases code for long COVID (ICD-10), but more importantly is less prone to bias.
In particular, diagnoses of patients using ICD-10 tend to favour individuals with better access to healthcare, putting less fortunate people at a disadvantage, so the AI tool could help reduce inequities in care.
"This broader scope ensures that marginalised communities, often sidelined in clinical studies, are no longer invisible," said Estiri.
Long COVID – or post-acute sequelae of COVID-19 (PASC) to give it the scientific term – may also be a lot more common than estimated by the CDC, according to the researchers. Their work suggests that figure could be 22.8%, not too far off the National Center for Health Statistics estimate of 24% for Massachusetts, which is based on 2022-23 data.
Future studies may explore the algorithm in cohorts of patients with specific conditions, like chronic obstructive pulmonary disease (COPD) or diabetes. In the meantime, the team plans to make its algorithm open-access so it can be deployed by other healthcare systems.
Image by Gerd Altmann from Pixabay