In this multi-country RWD study, we assessed whether transformation of heterogeneous European RWD sources to the OMOP CDM supports semantically aligned and analytically consistent downstream analyses across countries. Using female BC and ALS as two contrasting use-case populations, we evaluated the feasibility of OMOP-based harmonisation, assessed data quality after conversion, and benchmarked OMOP-derived descriptive analyses against analyses conducted on the original native data. The study therefore addresses an important empirical question in regulatory and health-data science: whether CDM conversion improves the comparability of cross-country RWD analyses, and where residual heterogeneity persists despite harmonisation.

Overall, OMOP transformation was successfully implemented for the three participating countries and enabled the application of standardised descriptive analysis workflows across heterogeneous national data sources. The resulting OMOP databases achieved high data quality metrics, and descriptive estimates derived from OMOP-transformed data were largely concordant with estimates obtained from native-format analyses. This concordance indicates that the ETL processes preserved key epidemiological and clinical characteristics and supports the feasibility of OMOP-based harmonisation in diverse European healthcare environments. At the same time, our findings show that successful CDM conversion should not be interpreted as an automatic guarantee of cross-country comparability. Rather, there is a need for systematic quality assessment, iterative benchmarking against native data, and disease- and context-specific interpretation of harmonised data.

Heterogeneity in data completeness and granularity was observed across the countries and between the two showcased diseases, reflecting differences in healthcare organisation, coding practices, and data collection purposes. For BC, core epidemiological estimates including age, sex, incidence and mortality were broadly consistent across the participating countries and aligned with the evidence, supporting the validity of registry-based cancer data for population description [9, 10]. In particular, the stability of incidence estimates across the Nordic datasets is consistent with the well-documented burden of breast cancer in high-income settings with improved screening programmes and comprehensive cancer registration [8, 9]. However, wide variation in the availability of tumour stage and molecular characteristics mirrors known limitations in routine data capture, where clinically relevant prognostic detail is often not consistently captured in the national data sources used in this study. These findings reinforce that, while national registries reliably capture diagnoses and survival trends, more granular information required for stratified or personalised analyses — such as receptor status or detailed staging — may require linkage to additional sources of information in some settings.

The ALS analyses highlighted challenges specific to rare diseases. Variation in ALS incidence estimates across countries is broadly consistent with prior reports of higher rates in populations of European descent and likely reflect differences in diagnostic coding practices and registry coverage rather than true epidemiological differences alone [11, 12]. Despite differences in data availability and capture across countries, observed age and sex distributions were comparable across the datasets and aligned with the known age and sex distributions [31]. The inconsistent capture of ALS treatments, such as riluzole, particularly in settings where hospital-dispensed medications are not recorded at the individual level, highlights an important limitation for treatment-pattern analyses and underscores the need for disease-specific knowledge when interpreting RWD [32]. Together, these findings illustrate that while RWD are well-suited to describing the broad epidemiology and outcomes of rare diseases such as ALS, their use for more detailed clinical characterisation or treatment evaluation requires careful consideration of disease-specific pathways of care and data capture mechanisms.

Across all participating countries, linkage at the individual level within universal healthcare systems enabled longitudinal follow-up and integration of diagnoses, medication dispensing, and mortality. However, the availability of additional variables, such as socioeconomic indicators or molecular tumour markers, was often limited by country-specific data access and governance rules, which determined whether these data could be linked, accessed, or shared for the analyses. Thus, the observed data heterogeneity in RWD reflects not only differences in healthcare delivery, but also differences in the legal and institutional frameworks governing data availability and use.

A key methodological contribution of this work is the comparison of analyses conducted on native data and data transformed to the OMOP CDM. The high concordance observed between native and OMOP-based descriptive estimates is consistent with prior work demonstrating that OMOP enables population-level characterisation across heterogeneous data sources [15, 33]. However, the observed loss of breast cancer stage information in the Danish OMOP dataset highlights an important caveat: clinically relevant variables present in source registries may require additional transformation steps during the ETL process. In particular, complex variables such as cancer stage, which are often recorded through multiple registry fields or local coding conventions, may not be fully preserved through standard OMOP mappings unless explicit transformation logic is implemented [34]. Achieving concordance between native and OMOP-based results required iterative benchmarking against native analyses and refinement of ETL workflows, highlighting that harmonisation requires validation and adjustment rather than purely automated conversion [35]. For example, discrepancies observed during early comparisons of descriptive outputs were traced to differences in the interpretation of source registry variables and their corresponding OMOP mappings, which were subsequently verified and aligned to ensure consistency between native and OMOP-derived estimates. These findings underscore the importance of validation against native data and transparency around ETL decisions, particularly when OMOP-based analyses are intended to inform regulatory or public health decisions [36]. These findings are further supported by earlier European studies that have explicitly compared analyses conducted on native real-world data with those derived from OMOP-transformed datasets [37–39]. In the United Kingdom, fidelity assessments of the Clinical Practice Research Datalink (CPRD) demonstrated that conversion to OMOP preserved key population characteristics and effect estimates from established pharmaco-epidemiological studies, with discrepancies largely attributable to unmapped over-the-counter medications and local coding conventions [40]. Subsequent work using linked UK primary care, hospital, and mortality data similarly reported high concordance between cohorts defined in native formats and those derived from OMOP, with only modest deviations in selected comorbidities and medication capture [39]. Comparable observations have been reported in Germany, where multi-centre hospital implementations of OMOP showed close agreement between native and OMOP-based descriptive analyses while also documenting domain-specific information loss related to national terminologies and local documentation practices [14, 41]. Collectively, these European comparisons indicate that OMOP-based analyses can reliably reproduce population-level descriptive findings obtained from native data when ETL processes are carefully designed and validated, while also highlighting that clinically relevant detail may be attenuated without disease- and context-specific adaptations. In this regard, our observations for breast cancer staging and ALS treatment capture align closely with previously reported findings and reinforce the need for iterative benchmarking against native data when OMOP is used to support regulatory or public-health decision-making. Expert knowledge remains essential to assess fitness for purpose, particularly when using these data for comparative analyses meant to provide evidence about safety and effectiveness of treatments [11].

This study has several strengths, including its multi-country design with different health systems, inclusion of both rare disease and prevalent malignancy, and systematic comparison of native and harmonised analyses. However, also limitations must be acknowledged. Misclassification and missing data remain inherent to routine healthcare data; and access delays limited the completeness of cross-country comparisons. Importantly, these limitations reflect structural characteristics of current RWD ecosystems rather than shortcomings of individual datasets.

In conclusion, this study shows that OMOP CDM conversion can support semantically aligned, reproducible, and analytically consistent cross-country analyses of European RWD when accompanied by rigorous quality assessment and iterative validation against native data. However, OMOP conversion does not by itself eliminate heterogeneity in source-data provenance, clinical practice, coding systems, completeness, or governance constraints. Cross-country comparability remains a study-specific empirical property that must be demonstrated rather than assumed. Robust use of harmonised RWD for regulatory, HTA, or public-health evidence generation therefore requires transparent ETL documentation, native-data benchmarking, and close collaboration between clinical, epidemiological, and data-science experts.