Presentation (CTI)

OBJECTIVES: With European Health Data Space implementation underway, real-world data (RWD) will become more available for multi-setting observational research. Linking complementary sources can enable follow-up across the care pathway and improve outcome completeness. However, privacy constraints often preclude a unique patient identifier, requiring alternative linkage approaches that can introduce missed and false matches, leading to bias and misinterpretation. The objective is to identify linkage-related challenges in multi-source RWD studies and describe practical approaches used to mitigate their impact on study design and interpretation.
METHODS: We reviewed >1,100 studies and >569 publications (1999-2025) using linked RWD from the PHARMO Data Network in the Netherlands to identify linkage-related challenges and document mitigation approaches.
RESULTS: Three broad challenges emerged: (1) defining patient eligibility in the context of linkage; (2) declining eligible sample size as linkage requirements increase; and (3) ascertaining and comparing outcomes across sources when captured differently. For (1), eligibility was restricted to patients with sufficient match variables to avoid denominator inflation and underestimated incidence. For (2), feasibility constraints were mitigated by using complementary linkage pathways (GP-hospital, outpatient pharmacy-hospital) to maximize capture of exposure and outcomes across care settings. Given differences in (geographical) catchment overlap, population severity heterogeneity, and data provenance, deduplication procedures were implemented and pathway was treated as an analytic covariate . For (3), discordant proxies complicated outcome ascertainment and cross-source comparability, resolved by pre-specifying harmonized, cross-source outcome algorithms with adjudication rules for resolving discrepancies (e.g., source hierarchy), and by reporting sensitivity analyses by data source/definition to assess robustness.
CONCLUSIONS: Linkage can strengthen longitudinal outcomes capture and completeness, but approach can affect eligibility, sample size, and ascertainment consistency, implicating bias and interpretability. Linkage should be treated as an explicit design component and continuously quality-assessed using predefined expectations, standardized checklists (e.g., linkability, deduplication rules), and external benchmarking to detect and mitigate artifacts.