Multimorbidity patterns and phenotype transitions in patients with clinician-coded long COVID: a multicenter US...

Long COVID is clinically heterogeneous, but longitudinal changes in documented chronic disease burden and transitions in multimorbidity phenotypes after infection are not well characterized in routine care.

We considered 425,614 patients with clinician-coded long COVID (ICD-10-CM U09.9) and a definable COVID-19 index date between October 1, 2021 and September 16, 2024 using deidentified electronic health record data from Epic Cosmos, a multicenter US network. Pre-index and post-index windows were defined as days - 365 to - 1 and days 91 to 455 relative to infection, respectively; follow-up was available through December 15, 2025. Chronic condition groups derived from ICD-10-CM codes were compared across windows using adjusted generalized estimating equation models. K-modes clustering was used to identify multimorbidity phenotypes, and multinomial regression was used to estimate adjusted transition probabilities.

Two pre-index phenotypes were identified: low burden (87.9%) and multimorbid (12.1%). Four post-index phenotypes emerged: low burden (63.3%), multimorbid/systemic (21.0%), respiratory-dominant (4.3%), and high-utilization/low-coded multimorbidity (11.3%). Higher baseline multimorbidity was associated with greater probability of transition to the multimorbid/systemic phenotype, whereas respiratory-dominant and high-utilization phenotypes arose from both baseline groups. Increases were concentrated in neurologic/autonomic, respiratory, hypercoagulable, endocrine/metabolic, sleep-related, and symptom-based domains. The high-utilization/low-coded phenotype was younger, predominantly female, and had greater emergency department and outpatient use. Fewer changes reached statistical significance in children than in adults.

Among patients with clinician-coded long COVID, chronic disease burden increased after infection and diversified into interpretable post-index phenotypes with distinct utilization profiles, supporting phenotype-informed follow-up and health system planning.