Feasibility of using automatically extracted routine clinical data in a respiratory cohort study: The SPHN-SPAC demonstrator project.

Objectives To assess the feasibility of using clinical data automatically extracted via the Swiss Personalized Health Network (SPHN) to complement or replace manually abstracted clinical data in the Swiss Paediatric Airway Cohort (SPAC). Materials and Methods We studied 1,075 SPAC participants enrolled between 2017-2023 at two Swiss children's hospitals. Clinical data were extracted from electronic health records via SPHN in Resource Description Framework format, transformed into visit-centered datasets, and compared with manually abstracted SPAC clinical data and parent-reported emergency department (ED) visits and hospitalizations from follow-up questionnaires. We assessed feasibility by identifying challenges in acquiring data and evaluated data quantity, completeness, and agreement between datasets. Results We obtained analysis-ready SPHN-derived datasets from two hospitals after 24 months. SPHN-derived data captured more pneumology outpatient visits than manual abstraction (Hospital A: 1,963 vs 1,049; Hospital B: 2,343 vs 1,010) and identified clinical events among children without follow-up questionnaires. Completeness of variables varied across hospitals and encounters, reflecting differences in local clinical documentation practices. SPHN-derived and manually abstracted data showed high agreement for structured clinical variables, including spirometry measurements (concordance correlation coefficient >0.99). Self-reported and SPHN-derived ED visits and hospitalizations showed high absolute agreement but moderate concordance. Discussion and Conclusion Automated extraction of routine clinical data increased the completeness of longitudinal information compared with manual abstraction, suggesting that SPHN-derived data can complement manual data collection in cohort studies. Broader use remains limited by heterogeneous clinical documentation practices and the substantial effort required to harmonize and transform extracted data into analysis-ready research datasets.