Testing of the models estimated that area-under-the-curve scores (indicating sensitivity and specificity) were excellent in detecting persistent symptoms among all COVID-19 patients (0.92) and hospitalized patients (0.90) and good among outpatients (0.85). The study also revealed important long-COVID risk factors, such as greater healthcare use, increasing age, and shortness of breath. The three resulting machine-learning models were designed to detect patterns of symptoms, healthcare use, demographics, and prescriptions to identify all COVID-19 patients likely to have lingering symptoms, including both hospitalized and non-hospitalized patients.Īs of October 2021, the team had identified more than 100,000 long-COVID patients, a figure that has doubled as of this month. It can be difficult to diagnose, because many of its symptoms mimic those of other conditions, the researchers said.Ī National COVID Cohort Collaborative (N3C) team analyzed electronic health record (EHR) data from 97,995 adults diagnosed as having COVID-19 at least 90 days earlier and 597 survivors undergoing treatment at a long COVID clinic. Long COVID causes a wide variety of symptoms such as shortness of breath, fatigue, fever, headaches, and "brain fog" for months or years after initial diagnosis. Machine-learning models created by a National Institutes of Health (NIH)-supported research team can identify, with high accuracy, patients likely to have long COVID, according to a study yesterday in The Lancet Digital Health. Machine-learning models may detect patients at risk for long COVID-19