New Research Suggests Long COVID Affects 1 in 6 Patients, Doubling Current Estimates

A custom AI algorithm identified post-acute sequelae of SARS-CoV-2 infection (PASC) in 16.3% of nearly 458,000 COVID-19 patients across 58 US hospitals, more than double the proportion captured by diagnostic code-based surveillance, according to a retrospective cohort study published in JAMA Network Open.¹

The findings suggest more than 15 million Americans are living with chronic post-COVID conditions invisible to the ICD coding systems clinicians and policymakers use to track disease burden.

“This work demonstrates how longitudinal clinical data in a health system can be structured and analyzed to support more consistent identification of complex post-viral conditions,” Shawn Murphy, MD, PhD, study co-author and Chief Research Information Officer for University of Washington, said in a statement. “There is significant potential for clinical AI when it is designed for public health and integrated across real-world care settings.”

Current reliance on the ICD-10-CM code U09.9 for post-COVID conditions captures fewer than 7% of affected patients.¹ The true toll of long COVID arriving through primary care, cardiology, endocrinology, and neurology offices without the diagnostic label connecting those visits to antecedent infection has remained largely unquantified, until now.

The retrospective cohort study used electronic health record (EHR) data from 4 US regions spanning 2017 to 2025, including New England (n = 12 hospitals), Southeast Texas (n = 1), Southern California (n = 5), and Western Pennsylvania (n = 40).

Researchers from Mass General Brigham deployed the Precision Phenotyping for Research Cohorts (P2RC) algorithm, a custom AI tool that operationalizes the World Health Organization case definition of PASC as a diagnosis of exclusion. The algorithm applies transitive sequential pattern mining to identify temporal symptom patterns emerging 3 or more months after infection and persisting for at least 2 months, while excluding sequelae explained by preexisting conditions. Validation demonstrated 79.9% precision in identifying PASC cases.¹

The P2RC algorithm identified 74,560 PASC cases overall (16.3%), ranging from 13.6% in Western Pennsylvania to 22.7% in Southern California. Among the 74,560 patients identified with PASC, 89.3% (66,587 patients) had ≥ 1 chronic condition requiring ongoing clinical management, representing 14.5% of the entire 457,950-patient COVID-19 cohort. Applying the 14.5% chronic PASC prevalence to approximately 103 million documented US COVID-19 cases suggests roughly 15 million individuals are living with chronic post-COVID conditions.¹

Of the 883 ICD-10-CM codes associated with PASC manifestations, 67.3% represented chronic or potentially chronic conditions; only 4.1% represented acute, self-limited conditions.¹ This profile challenges the framing of long COVID as a transient post-viral syndrome and situates it within the chronic disease management landscape already confronting primary care.

The study identified substantial regional variation in organ system involvement (chi-square = 2504.73; P <.001). Systemic symptoms were most common across all sites (22.6% to 25.1% of manifestations), followed by respiratory and gastrointestinal presentations. Endocrine manifestations differed markedly: New England showed thyroid-predominant patterns (predominantly hypothyroidism), while Southeast Texas, Southern California, and Western Pennsylvania showed metabolic-predominant profiles, with prediabetes emerging as the leading pancreatic diagnosis outside New England. This heterogeneity suggests PASC may comprise distinct endotypes with region- or population-specific pathobiology rather than a single uniform syndrome.

Temporal analysis revealed cumulative PASC prevalence continued to rise through mid-2024, with statistically significant quarterly increases in New England (incidence rate ratio [IRR], 1.01; 95% CI, 1.00-1.01; P <.001), Southern California (IRR, 1.00; 95% CI, 1.00-1.01; P <.001), and Western Pennsylvania (IRR, 1.02; 95% CI, 1.01-1.02; P <.001).¹ Compounded over 40 quarters, these quarterly increments correspond to a 13% to 81% relative increase in cumulative PASC prevalence, contradicting assumptions this condition represents a legacy of early pandemic waves alone.

The investigators noted their cohort requires longitudinal EHR documentation and excludes patients with fragmented or absent medical records, meaning these estimates likely represent a floor rather than a ceiling. The researchers call for expanding federated AI phenotyping infrastructure nationally, stratifying future PASC clinical trials by predominant organ system involvement, and developing biomarker signatures for phenotype-specific interventions.¹

"This study demonstrates how hospitals can leverage AI to help fill surveillance gaps that public health agencies are no longer tracking. What excites me most is what can come next with this new surveillance data,” study corresponding author Hossein Estiri, PhD, a faculty member in the Mass General Brigham Department of Medicine, said in a statement. "Once we can distinguish different clinical and organ-specific manifestations of long COVID, we gain the ability to launch new trials and test targeted treatments for the right patients.”

References

1. Tian J, Azhir A, Decaro M, et al. Long COVID persistence and surveillance gaps across 58 US hospitals. JAMA Network Open. 2026;9(5):e2614909. doi:10.1001/jamanetworkopen.2026.14909

2. Mass General Brigham. Long Covid burden continues to grow, doubling current surveillance estimates, multi-hospital study shows. EurekAlert! May 27, 2026. Accessed May 27, 2026. https://www.eurekalert.org/news-releases/1129452