Assessing the impact of additional inclusion criteria on the representativeness of EHR-based long COVID cohorts identified using U09.9

Electronic health records (EHRs) are widely used to study Long COVID, yet cohort inclusion criteria may systematically affect cohort composition, shaping how findings are interpreted and to whom they may generalize. We analyzed harmonized EHR data from N3C among nondeceased individuals with at least one U09.9 Long COVID diagnosis between October 2021 and October 2023. Within this U09.9-identified cohort, we evaluated three additional inclusion criteria: COVID-19 diagnosis, likely COVID-19–related hospitalization, and availability of ≥ 2 years of prior medical history. Mixed-effects logistic or multinomial regression models estimated associations between selection criteria and sex, race/ethnicity, area-level poverty, and health insurance coverage, while zero-inflated negative binomial models evaluated estimates of healthcare utilization. Data-contributing partner was included as a random effect for all models. Sensitivity analyses examined regional variation and compared findings with COVID-19–only cohort. Cohort size, demographic composition, and healthcare utilization varied substantially by inclusion criteria. Requiring ≥ 2 years of medical history increased representation of women and individuals with higher outpatient utilization but reduced inclusion of individuals from low–insurance coverage areas. COVID-19–related hospitalization was associated with higher inpatient and non-inpatient utilization (acute care clinic visits, emergency room visits) and increased representation of non-Hispanic Black individuals, but lower representation of women. Having a documented COVID-19 diagnosis enriched cohorts with patients from higher-poverty and lower–insurance coverage areas and with greater healthcare utilization. Multinomial models showed that medical history requirements were negatively associated with inclusion of all non-White racial/ethnic groups. Findings were largely consistent across geographic regions and aligned with patterns observed in the COVID-19–only cohort. Routinely used EHR-based inclusion criteria are not analytically neutral. In N3C, cohort composition reflects site participation, data harmonization, common data model implementation, and data quality processes. Although these features are N3C-specific, similar design considerations may affect disparities and generalizability across EHR-based Long COVID research.