Large-Scale Lung Function Analyses using an automated Data Stream: preliminary data from the SwissPedLung Data Stream

Introduction: Manually extracting lung function (LF) data for research purposes is time-consuming and error-prone. Automated LF data extraction could overcome these problems. Guidelines on pediatric asthma diagnosis differ in their definition of a positive bronchodilator response (BDR): the ERS/ ATS technical standard requires an increase in FEV1 of≥10% relative to the predicted value (PV)) [ 1 ], whereas GINA guidelines require≥12% relative to the baseline value (BV) [ 2 ]. The impact of these differences is not yet known. Aims: In SwissPedLung (SPL), a nested project in the Swiss Personalized Health Network, we build a data stream that integrates routine clinical data from multiple Swiss pediatric pulmonology divisions into one central data platform. We aim to use this data to analyze the impact of differences in BDR definition. Methods: In a multidisciplinary team with pediatric pulmonology experts and data and biomedical engineers, we developed a pipeline that transfers raw LF data from LF devices to local clinical data warehouses, where it is merged with electronic health record data. Combination of data from different hospitals is done in one central data platform. We included children with at least one LF test in the participating clinics in the years 2017 to 2023 and without denial of general consent (GC). In a first analysis, we compared the BDR results based on the two different guidelines. Results: In a first data extraction of one center, we automatically extracted 14.124 spirometries. We excluded 1.792 spirometries due to denial of GC and 780 due to poor quality. 4.285 tests were followed by a BDR test. Mean FEV1 difference pre to post BDR was 8.4% (SD 6.5) relative to PV, and 10% (SD 8.6) from BV. 93% of BDR tests were positive or negative according to both definitions, whereas 7% (n=304) were positive according to only one of the two guidelines. Conclusion: Automated LF data extraction is feasible but requires a robust data infrastructure supported by a multidisciplinary team. In a preliminary analysis we show, how different guidelines influence real-world diagnostic results: In 7% of BDR tests, the definition used alters the outcome. Further analyses will include the entire SPL data from multiple clinics as well as other forms of LF tests. Publication History Article published online: 26 February 2026 © 2026. Thieme. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany