Federated Analysis With Differential Privacy in Oncology Research: Longitudinal Observational Study Across Hospital...

Area of Science:

• Health Informatics
• Medical Oncology
• Data Privacy

Background:

• Federated analytics allows statistical queries on remote health datasets without accessing raw data, addressing privacy and governance concerns.
• This approach is crucial for multi-center studies but remains underutilized in real-world settings.
• Combining federated analytics with differential privacy (DP) has not been previously explored in real-world studies.

Purpose of the Study:

• To deploy a federated architecture in a real-world clinical setting for a multicenter oncology study.
• To evaluate the practicality and utility of differential privacy (DP) as a privacy-enhancing technology within this federated framework.

Main Methods:

• Established a federated architecture across three healthcare centers (Toulouse, Reims, Foch).
• Harmonized data within each center and performed statistical analyses using DataSHIELD.
• Implemented a novel open-source differential privacy package (dsPrivacy) for DataSHIELD.

Main Results:

• Successfully conducted a multicenter study on metastatic non-small cell lung cancer patient data without centralizing raw information.
• Demonstrated the practicality of DataSHIELD for efficient, secure, multi-center data analysis.
• Showcased the feasibility of implementing differential privacy with acceptable privacy-utility trade-offs.

Conclusions:

• The federated architecture successfully supported a real-world, multicenter oncology study with robust privacy protections.
• Differential privacy integration proved practical, offering a viable solution for enhancing data security in federated health research.