Self-care practices for puerperal sepsis prevention and their determinants among postpartum women: a community-based cross-sectional study in Halaba Town, Central Ethiopia.

Electronic health record (EHR) data comprise large volumes of medical information accumulated during routine clinical practice and are widely utilized in clinical epidemiology, healthcare quality assessment, and machine learning applications. In recent years, distributed analytical approaches such as federated learning, which enable the integration and analysis of EHR data across multiple healthcare institutions, have gained increasing attention, highlighting the growing importance of multicenter data research [, ]. In such research contexts, ensuring the reproducibility of findings requires a clear understanding of the temporal stability of the underlying EHR data [, ]. Previous studies investigating temporal variability in healthcare data have demonstrated that the distribution of EHR data can change over time due to factors such as system updates and revisions to clinical guidelines. For example, a study using laboratory data from a clinical data warehouse in France reported both gradual trends and abrupt changes in many test items, primarily attributed to updates in measurement instruments and software []. Similarly, research based on primary care and hospital data in the United Kingdom showed that revisions to clinical coding guidelines led to substantial changes in the frequency of cardiovascular disease codes []. However, these studies have focused on distributional changes between data recorded at different time periods within the same database, and it remains unclear whether data recorded for the same target period may vary depending on the timing of data extraction. Studies specifically examining variability arising from differences in extraction timing itself—rather than from changes in clinical practice or coding conventions—remain limited, despite its potential relevance to the reproducibility of multicenter EHR research. A prior study using a Canadian primary care database reported data integrity issues associated with transitions between EHR systems, including difficulties in linking records across old and new systems due to the absence of common patient identifiers, incorrect overwriting of historical procedure dates, and anomalous reductions in prescription records []. These findings suggest that EHR data may be modified even after initial recording due to system updates and data migration processes, raising the possibility that data corresponding to the same target period may differ depending on the extraction time point. Consequently, in study designs that rely on extracting EHR data at different time points for analysis, variations in data composition across extraction times may affect the reproducibility of research findings.