Source: fernandozhiminaicela/Pixabay
New research by Harvard Medical School shows how an artificial intelligence (AI) algorithm created by Mass General Brigham can be used as a precision phenotyping tool to help identify long COVID or PASC (post-acute sequelae of COVID-19), in patients using electronic health record (EHR) data.
“The PASC phenotyping method presented in this study boasts superior precision, accurately gauges the prevalence of PASC without underestimating it, and exhibits less bias in pinpointing long COVID patients,” wrote the study’s senior author Hossein Estiri, Ph.D., an associate professor of medicine at Harvard Medical School and director of the Clinical Augmented Intelligence Group (CLAI) at Massachusetts General Hospital (MGH), along with his research colleagues.
Globally, an estimated 409 million people had PASC in 2023, according to a study published in August 2024 in Nature Medicine by Eric J. Topol, Ziyad Al-Aly, and their co-authors. In the U.S., an estimated 6.9% of adults and 1.3% of children had PASC in 2022, according to a National Health Interview Survey by the U.S. Centers for Disease Control and Prevention (CDC).
PASC is a chronic disease state that affects one or more organs where symptoms last for three months or longer after infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus that causes the coronavirus disease 2019, more widely known as COVID-19.
Patients with long COVID may have one or more symptoms such as cough, persistent fatigue, problems with taste or smell, sleep disturbance, concentration issues, memory changes, recurring headaches, shortness of breath, lightheadedness, bloating constipation, diarrhea, and more, as well as one or more diagnosable conditions such as anxiety, mood disorders, cognitive impairment, migraine, blood clots, stroke, cardiovascular disease, heart arrhythmia, fibromyalgia, diabetes, autoimmune disorders, and many more conditions, according to The New England Journal of Medicine (NEJM).
Personalized medicine, or precision medicine, considers a patient’s individual genetic, environmental, and lifestyle variability for the prevention and treatment of diseases. It is the exact opposite of the one-size-fits-all approach. A person’s observable traits, called phenotypes, are determined by a combination of their genetic makeup (genotype) and environmental factors. For example, weight is an observable trait or phenotype that may be impacted by diet, exercise, and other factors.
One of the aims of studying phenotyping for precision medicine is to understand the phenotypic abnormalities associated with a particular disease in order for clinicians to provide better diagnosis and treatment for the individual patient. Additionally, having knowledge of a disease’s full range of associated phenotypic abnormalities may help with early intervention and treatment for better outcomes.
For this Harvard Medical School study, the researchers used de-identified electronic health record data from more than 295,000 patients of 20 community health centers and 14 hospitals in the Mass General Brigham integrated health care system in Massachusetts.
The AI-based precision phenotyping tool was developed with an attention mechanism to eliminate disease conditions that can be explained by prior factors.
“Compared to relying solely on the U09.9 diagnosis code to identify long COVID patients, our method boasts superior precision, accurately gauges the prevalence of PASC without underestimating long COVID, and exhibits less bias across demographic groups,” the researchers reported.
The researchers reported that their AI tool accuracy rate of 79.9% outperformed the long COVID International Classification of Diseases Tenth Revision diagnosis code for post–COVID-19 condition, unspecified, (ICD-10 code U09.9) of 77.8%. Their precision phenotyping AI tool also discovered a long COVID research cohort of 24,000 patients, which is four times more than the diagnosis code.
This research shows how an AI tool, combined with real-world data, can help find signals and meaningful patterns in massive amounts of complex longitudinal health data. Artificial intelligence deep learning is enabling innovative medial research to accelerate precision medicine and more personalized health care.
Copyright © 2024 Cami Rosso. All rights reserved.
