Researchers have created a new advanced AI tool designed to detect hidden cases of long COVID by analyzing patient medical records. This innovative method proved to be more effective than existing tools that rely solely on diagnostic codes. According to the study’s authors, approximately 22.8% of individuals experience long COVID symptoms, which is a higher percentage than previously estimated, potentially offering a more accurate representation of national trends.
At Mass General Brigham, the research team has developed an AI tool that analyzes electronic health records to assist healthcare providers in identifying long COVID cases. This often perplexing condition can involve lasting symptoms such as fatigue, chronic cough, and cognitive impairments following COVID-19 infection. The findings, published in the journal Med, indicate that more individuals may be in need of care for this possibly debilitating condition, suggesting that the actual prevalence of long COVID may be significantly underestimated.
“Our AI tool has the potential to transform an unclear diagnostic process into a clear and precise one, empowering clinicians to better understand this complex condition,” stated senior author Hossein Estiri, PhD, who leads AI Research at the Center for AI and Biomedical Informatics of the Learning Healthcare System (CAIBILS) at Mass General Brigham and also serves as an associate professor at Harvard Medical School. “With this advancement, we may finally begin to recognize long COVID in its true form and, importantly, determine effective treatments.”
Long COVID, also called Post-Acute Sequelae of SARS-CoV-2 infection (PASC), encompasses a broad array of symptoms. For their research, Estiri and his team defined it as a diagnosis that is both associated with COVID-19 infection and cannot be explained by any other existing condition in the patient’s medical history. Additionally, the symptoms must have lasted for at least two months within a year-long follow-up period.
The AI tool’s algorithm was developed using anonymized patient data from the clinical records of nearly 300,000 patients spread across 14 hospitals and 20 community health centers within the Mass General Brigham system. Instead of relying on a singular diagnostic code, the AI employs a new technique called “precision phenotyping” to examine individual records for symptoms and conditions related to COVID-19 and monitor those symptoms over time to distinguish them from other illnesses. For instance, the algorithm can ascertain if shortness of breath is due to existing conditions like heart failure or asthma rather than being a result of long COVID, only flagging the patient as having long COVID after all alternative explanations have been thoroughly evaluated.
“Physicians often face the daunting task of navigating a complex web of symptoms and medical histories while managing hefty caseloads. An AI-powered tool that can systematically analyze this data for them could revolutionize the approach to diagnosis,” remarked Alaleh Azhir, MD, co-lead author and internal medicine resident at Brigham Women’s Hospital, a core component of the Mass General Brigham healthcare system.
By providing patient-centered diagnoses, this new method may help reduce biases ingrained in current long COVID diagnostic practices, as noted by the researchers. They point out that patients who are diagnosed using the official ICD-10 code for long COVID often have better access to healthcare services. While previous studies estimated that around 7% of the population suffers from long COVID, this new strategy indicates a significantly higher estimate of 22.8%. The authors believe this figure aligns more closely with actual national trends and offers a more accurate portrayal of the pandemic’s long-term impact.
The research team found that their tool was about 3% more accurate than the figures captured by ICD-10 codes and less biased overall. Their study showed that the individuals identified as having long COVID reflect the broader demographic landscape of Massachusetts, setting it apart from long COVID algorithms that depend on a single diagnostic code or isolated clinical interactions, which often skew results towards populations with better access to medical care. “By adopting this wider lens, we can ensure that marginalized communities, which are often overlooked in clinical research, are no longer rendered invisible,” said Estiri.
Some limitations of the study and AI tool include that the health record data utilized in the algorithm might not be as comprehensive as information documented by physicians in their post-visit clinical notes. Additionally, the algorithm may overlook possible worsening of pre-existing conditions that could signify long COVID symptoms. For instance, if a patient suffers from COPD and previously experienced worsening episodes before contracting COVID-19, the algorithm might mistakenly dismiss them even if their long-standing symptoms are indicators of long COVID. The decline in COVID-19 testing rates in recent years also complicates the identification of when a patient first contracted the virus. Furthermore, the research was conducted only on patients within Massachusetts.
Future studies may assess the algorithm’s effectiveness in specific patient cohorts, such as those with COPD or diabetes. The researchers also intend to make this algorithm publicly available for use by healthcare providers and systems around the globe.
By paving the way for improved clinical care, this research may also establish a foundation for future investigations into the genetic and biochemical factors underlying the various subtypes of long COVID. “We can now explore questions concerning the true impact of long COVID, which have remained challenging to address until now,” noted Estiri.
Authorship: Alongside Estiri, the team from Mass General Brigham includes Alaleh Azhir, Jonas Hügel, Jiazi Tian, Jingya Cheng, Ingrid V. Bassett, Emily S. Lau, Yevgeniy R. Semenov, Virginia A. Triant, Zachary H. Strasser, Jeffrey G. Klann, and Shawn N. Murphy. Additional contributors comprise Douglas S. Bell, Elmer V. Bernstam, Maha R. Farhat, Darren W. Henderson, Michele Morris, and Shyam Visweswaran.
Funding: This work was supported by the National Institutes of Health, National Institute of Allergy and Infectious Diseases (NIAID) R01AI165535, National Heart, Lung, and Blood Institute (NHLBI) OT2HL161847, and National Center for Advancing Translational Sciences (NCATS) UL1 TR003167, UL1 TR001881, and U24TR004111. J. Hügel’s research was also partially funded through a fellowship within the IFI program of the German Academic Exchange Service (DAAD) and the Federal Ministry of Education and Research (BMBF), as well as the German Research Foundation (426671079).
