Agreement Between ICD-10 Code–Only and Composite Definitions of Asthma Exacerbations in Electronic Health Records: A Multicenter Cohort Study in Taiwan, 2017-2023

Abstract

Background Accurate identification of asthma exacerbations in electronic health records (EHRs) is essential for research and clinical decision-making. While composite definitions incorporating healthcare encounters and systemic corticosteroid use are commonly used, ICD-10 code-only definitions are often applied when detailed data are unavailable. However, their agreement remains unclear. This study aimed to (1) compare an ICD-10 code–only definition of asthma exacerbation with a composite definition, and (2) identify factors associated with concordance between both definitions in EHR data. Methods We conducted a multicenter retrospective cohort study using the Taipei Medical University Clinical Research Database (2017–2023). Asthma exacerbations were identified using (1) an ICD-10 code–only definition and (2) a composite definition incorporating asthma-related emergency department visits, hospitalization, or systemic corticosteroid use. Episodes were defined using a seven-day gap rule. Agreement was assessed using overall percent agreement (OPA), positive predictive value (PPV), negative predictive value (NPV), positive percent agreement (PPA), and negative percent agreement (NPA), with generalized estimating equations accounting for within-patient clustering. Results Among 1,486,449 episodes, 40,180 (2.7%) met the code-only definition and 35,875 (2.4%) met the composite definition. After covariate adjustment, OPA was 0.930 (95% CI, 0.926–0.933). PPV and PPA were 0.313 (95% CI, 0.293–0.334) and 0.309 (95% CI: 0.291–0.328), respectively, while NPV and NPA were high (0.968 and 0.962). Comorbidity burden, hospital, asthma severity, and index year were associated with concordance. Conclusion ICD-10 code–only definition demonstrated high overall agreement with the composite definition, driven primarily by agreement on non-events, but limited agreement for identified exacerbations. These findings suggest that the two definitions may capture different levels of exacerbations. Several factors were associated with concordance between both definitions, highlighting the importance of accounting for these factors when analyzing EHR data. Code-only definitions may be used cautiously in EHR-based studies, given potential heterogeneity and misclassification.