Most COPD hospital records lack captured pulmonary function reports, audit finds

Patients with chronic obstructive pulmonary disease (COPD) often undergo pulmonary function testing, but the reports generated by those tests are frequently stored in outpatient or laboratory systems that never make it into inpatient electronic health record (EHR) extracts. A new retrospective study from Huashan Hospital, Fudan University, quantifies exactly how much of that report text is captured in one hospital’s inpatient analytic export, and the answer is strikingly small: report rows were present for only 10.8% of patients whose hospitalizations carried COPD diagnosis wording.

The study, conducted at the Shanghai hospital over a three-year period between 2022 and 2025, examined an export containing 19,612 hospitalizations from 5,961 patients. The investigators defined EHR-recorded COPD using prespecified diagnosis terms, a definition that identified 6,398 hospitalizations with COPD wording from 2,981 patients. Within that group, a pulmonary function report row was captured for 322 patients, or 10.8%, while the remaining 2,659 patients, 89.2% of the cohort, had no captured report row at all.

An important nuance runs through the entire analysis: report capture meant only that a pulmonary function report row was present in this particular inpatient analytic export. It did not indicate whether testing had ever actually been performed. A patient with no captured report row may have had pulmonary function tests whose results live in systems outside the export, such as outpatient clinics or laboratory databases that were not linked to the inpatient data.

For 278 patients, a report was linked to the same hospitalization record as the COPD wording, a stricter linkage that ties the report directly to the admission where the diagnosis appeared. The researchers also found that capture was not uniform across the cohort. Capture rates differed across sex, age, smoking documentation, respiratory diagnoses, and lung cancer strata, meaning the patients whose reports did appear in the export represent a selected subset rather than a random sample of the COPD cohort.

Beyond measuring capture, the study built a transparent index of the literal terms appearing in the captured reports. The investigators applied direct phrase rules to the ItemResult text field, indexing the exact strings that appeared. The 561 report rows in the full export reduced to just 216 unique normalized strings, a 61.5% reduction in the number of template-level review units that a human reviewer would need to examine. This kind of reduction matters for audit work, because it turns a scattered collection of free-text report rows into a manageable catalogue of recurring phrasings.

In the same-hospitalization subset, where reports were linked to the admission carrying COPD wording, the most common literal terms were ventilatory impairment, appearing in 55.0% of cases, mixed ventilatory impairment at 50.0%, reduced respiratory muscle strength at 48.2%, small-airway abnormality at 44.6%, and explicit obstructive wording at 31.7%. The authors are careful to stress that these are report-text frequencies, not estimates of how commonly these physiologic patterns actually occur in patients. A term appearing in half of captured reports says something about how reports are written, not about the underlying lung function of the broader COPD population.

The study also documents what the export did not contain. There were no International Classification of Diseases (ICD) codes, no timestamps, no structured pulmonary function values, no historical local laboratory rules, no reader metadata, and no clinician-annotated reference standard against which the automated rules could be validated. These absences shape what the analysis can and cannot claim, and the authors state the limits plainly in their conclusions.

The practical implication is that this inpatient export cannot support full-cohort physiologic refinement without additional data linkage. If researchers hoped to use pulmonary function report text to sharpen a COPD research cohort drawn from inpatient records, they would first need to link the export to the outpatient and laboratory systems where most reports presumably reside. The 89.2% of patients with no captured report row represent a gap that no amount of text mining within the export itself can close.

The rule index the authors built is offered as an organizational tool rather than a finished product. It organizes captured text for audit and review, making the literal phrasings of pulmonary function reports visible and countable, but it is explicitly not a clinically validated classifier. Anyone hoping to use the phrase rules to automatically classify patients by ventilatory pattern would need validation against an annotated reference standard, which this dataset did not include.

For the growing field of EHR-based respiratory research, the study offers a cautionary lesson about data provenance. Before report text is used to refine a COPD research cohort, investigators need to know how much of that text is captured and whether the capture is selective. This audit provides a template for answering those questions: quantify capture, test whether it varies across patient strata, reduce the report text to a reviewable index of literal terms, and state clearly what the export lacks. The findings apply to a single center over a defined period, and the authors note that all claims are those of the study team, but the methodological message travels well beyond one hospital’s data warehouse.

Subject of Research: Audit of pulmonary function report capture and rule-based text indexing in an inpatient EHR cohort with recorded COPD

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

Article References: Yang, K., Li, C., Chen, W., He, J., Yalikun, M., & Dong, J. (2026). Pulmonary function report capture and rule-based text indexing in an inpatient EHR cohort with recorded COPD. Frontiers in Medicine, 13, Article 1924553. https://doi.org/10.3389/fmed.2026.1924553

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1924553

Keywords: COPD, electronic health records, pulmonary function tests, report capture, natural language processing, rule-based indexing, data provenance, Huashan Hospital, retrospective study, respiratory medicine, Pulmonary, function