Evaluating the statistical utility and information loss in the transformation of a real-world oncology database from...

This project documented the conversion of a French real-world oncology study database from Clinical Data Interchange Standards Consortium (CDISC) Study Data Tabulation Model (SDTM) to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). The goal was to measure syntactic and semantic information loss resulting from this transformation.

Source data originated from a retrospective observational study of HER2-positive early breast cancer patients. Initially in CDISC-SDTM standards, the data included 73 variables detailing patient demographics, disease characteristics, surgery, and follow-up information such as adjuvant treatment. Data conversion to OMOP-CDM utilized extract, load and transform (ELT) procedures with Observational Health Data Sciences and Informatics (OHDSI) tools and the data build tool (dbt), encountering challenges in mapping specific variables and maintaining data granularity. Information loss assessments involved statistical analyses.

The source database was successfully mapped to OMOP-CDM and standardized terminologies. Statistical results from the OMOP-transformed database were consistent with those from the original SDTM database, achieving 100% concordance across all tested equality criteria for univariate, bivariate, logistic, survival, and correlation analyses at a 95% confidence interval or respective P value significance levels. Information loss (<1%) during conversion varied based on the original database's detail and mapping approach.

Statistical utility of the real-world oncology dataset was maintained after transformation to OMOP-CDM, ensuring reproducibility of statistical analyses. Information loss during conversion is significantly dependent on the intrinsic characteristics and level of standardization of the source database, particularly if not originally adhering to standard vocabularies.