Pulmonary function report capture and rule-based text indexing in an inpatient EHR cohort with recorded COPD

Background Inpatient electronic health record (EHR) extracts may not contain pulmonary function reports stored in outpatient or laboratory systems. Before report text is used to refine a chronic obstructive pulmonary disease (COPD) research cohort, investigators need to know how much of that text is captured and whether this capture is selective. Herein, we audited pulmonary function report capture and built a transparent index of literal report terms. Methods We conducted a single-center retrospective EHR study at Huashan Hospital, Fudan University, during a 3-year period between 2022 and 2025. The export contained 19,612 hospitalizations from 5,961 patients. EHR-recorded COPD was defined by prespecified diagnosis terms. Report capture meant that a pulmonary function report row was present in this inpatient analytic export; it did not indicate whether testing had ever been performed. Direct phrase rules indexed literal ItemResult text. The export did not contain International Classification of Diseases (ICD) codes, timestamps, structured pulmonary function values, historical local laboratory rules, reader metadata, or a clinician-annotated reference standard. Results The main definition identified 6,398 hospitalizations with COPD wording from 2,981 patients. A pulmonary function report row was captured for 322 patients (10.8%), whereas 2,659 (89.2%) had no captured report row. For 278 patients, a report was linked to the same hospitalization record as COPD wording. Capture differed across sex, age, smoking documentation, respiratory diagnoses, and lung cancer strata. The 561 report rows in the full export were reduced to 216 unique normalized strings, a 61.5% reduction in template-level review units. In the same-hospitalization subset, common literal terms included ventilatory impairment (55.0%), mixed ventilatory impairment (50.0%), reduced respiratory muscle strength (48.2%), small-airway abnormality (44.6%), and explicit obstructive wording (31.7%). These are report-text frequencies and not physiologic prevalence estimates. Conclusion Report rows were captured for only a small, selected subset, and therefore, this inpatient export cannot support full-cohort physiologic refinement without additional data linkage. The rule index organizes captured text for audit and review but is not a clinically validated classifier.