Long COVID, also known as post-COVID syndrome, causes a very wide range of symptoms. Here, the disease was defined as a diagnosis that cannot be exclusive to another infection: in short, this means that the diagnosis cannot be recorded as that of another unique condition in the patient’s medical record, whether obviously includes COVID symptoms, and that it must have persisted for 2 months or more within a 12-month follow-up window.
The Boston team then leverages artificial intelligence to identify signs of long COVID that are not visible via standard markers, track how different symptoms develop over time, and to rule out “false” alternative explanations. to patients’ symptoms.
23% of the population could have symptoms of long COVID
The new tool results precisely in a prevalence of 22.8%a figure that could correspond to a more realistic long-term toll of the pandemic, write the researchers. The algorithm used was developed by extracting anonymized data from the clinical records of nearly 300,000 patients followed at 14 hospitals and 20 community health centers in the Mass General Brigham system. The team uses a new AI method developed for this tool and named “precision phenotyping” : The technique involves screening individual records to identify conditions and symptoms already known to be related to COVID-19 and to track these symptoms over time to differentiate them from those of other illnesses.
The AI tool is actually able to analyze the history of each patient over time and provide a personalized approach to care and ultimately reduce, at a population level this time, the inequalities observed in current diagnoses and treatments of long COVID. In practice, the tool sifts through electronic medical records in order to more precisely identify cases of long COVID by analyzing the complex litany of persistent symptoms, and often common to other conditions, including fatigue, chronic cough and brain fog.
- For example, the algorithm can detect whether shortness of breath is the result of pre-existing conditions like heart failure or asthma or long COVID. It is only when all other possibilities have been exhausted that the tool concludes on the diagnosis of long COVID.
This analysis protocol not only concludes at a much higher prevalence,
almost a quarter of the population
but will allowidentify more patients who should receive care for this debilitating disease. Indeed, the tool was about 3% more accurate than what the ICD-10 codes capture, with lower bias. Its broader reach ensures that marginalized communities, often left out of clinical studies or with less access to care, can be diagnosed and treated.
“Our AI tool will transform a fuzzy diagnostic process into something precise and focused, giving doctors the ability to make sense of a complex disease,” explains lead author Hossein Estiri, an AI researcher at Center for AI and Biomedical Informatics of the Learning Healthcare System (CAIBILS) at MGB and Harvard professor: “With this tool, we may finally be able to see long COVID for what it really is.”
“Physicians are wading through a tangle of symptoms and medical histories, not knowing which threads to pull, while juggling a high workload. Having an AI-powered tool that can do this methodically could be a game-changer.”
Health
