Enabling genomic-phenomic association discovery without sacrificing anonymity

Area of Science:

• Biomedical Informatics
• Data Privacy
• Genomic Research

Background:

• Health information technologies generate vast patient-level data, valuable for biomedical research.
• Sharing de-identified clinical data is encouraged but hindered by re-identification concerns.
• Existing anonymization technologies may rely on unrealistic assumptions about data recipient capabilities.

Purpose of the Study:

• To design robust anonymization algorithms for clinical data dissemination.
• To provide provable guarantees of patient privacy protection.
• To enable large-scale biomedical investigations using de-identified health data.

Main Methods:

• Developed novel anonymization algorithms based on pragmatic assumptions.
• Applied algorithms to a de-identified dataset of over one million medical records.
• Evaluated data utility by assessing the discovery of genotype-phenotype associations.

Main Results:

• Demonstrated that pragmatic anonymization allows detailed clinical profile dissemination with strong privacy guarantees.
• Successfully discovered 192 genotype-phenotype associations from the anonymized dataset.
• Achieved data fidelity for discoveries equivalent to using non-anonymized clinical data.

Conclusions:

• Pragmatic assumptions lead to effective clinical data anonymization strategies.
• Secure sharing of large-scale patient data is feasible for biomedical research.
• The developed methods balance data utility with robust patient privacy protection.