David Pau | Biostatistics & Data Science
Roche

BioTechX EU Day 3 @ 12:30

Evaluating Data Utility in Anonymization, Federated Approaches, and OMOP-CDM

Background: To estimate remaining data utility, we evaluated three data strategies: Anonymization, Federated Approaches, and OMOP-CDM transformation.

Methods: CDISC-SDTM Data from a retrospective HER2+ breast cancer study (73 variables) were anonymized and mapped to OMOP-CDM. Using DataSHIELD, we tested a federated approach by splitting SDTM and OMOP databases into three samples. Statistical analyses (descriptive statistics, regression methods, survival analyses) for each method were compared against the raw CDISC-SDTM gold standard, focusing on information loss, consistency, and reproducibility.

Results: None of the anonymization methods successfully reproduced all statistical analyses. The federated approach demonstrated good consistency but showed decreased accuracy in multivariate models due to database variability. Conversely, CDISC-SDTM was successfully mapped to OMOP-CDM, showing high statistical concordance.Conclusions: Whilst data was successfully mapped to OMOP, utility was reduced when further privacy preserving methods were applied. A trade-off has to be found between privacy and usefulness of data.

David Pau, Biostatistics & Data Science, Roche

×

David Pau / Evaluating Data Utility in Anonymization, Federated Approaches, and OMOP-CDM

David Pau, Biostatistics & Data Science,
Roche

With over 30 years of combined experience working with various pharmaceutical laboratories in the field of biometrics, I currently holds a biostatistics and data science position at Roche France within the real-world data unit. Since 2019, I am also responsible for data sharing with researchers...

CloseMore