Privacy preserving data anonymization of spontaneous ADE reporting system dataset

Area of Science:

• Pharmacovigilance and Drug Safety
• Data Privacy and Anonymization
• Health Informatics

Background:

• Spontaneous Reporting Systems (SRSs) are crucial for long-term drug safety surveillance.
• Publishing SRS data raises privacy concerns due to sensitive personal health information.
• Existing privacy-preserving data publishing (PPDP) methods are inadequate for SRS data characteristics like rare events and multi-valued attributes.

Purpose of the Study:

• To propose a new privacy model, MS(k, θ(*))-bounding, specifically designed for spontaneous adverse drug event (ADE) reporting data.
• To develop an anonymization algorithm that addresses the unique challenges of SRS datasets.
• To evaluate the effectiveness of the proposed model in balancing privacy protection and data utility.

Main Methods:

• Developed the MS(k, θ(*))-bounding privacy model with flexible privacy thresholds (θ(*)) for varying sensitive values.
• Proposed a greedy-based clustering anonymization algorithm to minimize privacy risk and maintain data utility.
• Conducted empirical studies using the FAERS dataset (2004Q1-2011Q4) and compared with k-anonymity, (X, Y)-anonymity, Multi-sensitive l-diversity, and (α, k)-anonymity.
• Evaluated methods using Danger Ratio (DR) and Information Loss (IL) across uniform, level-wise, and frequency-based threshold settings.

Main Results:

• The MS(k, θ(*))-bounding model effectively prevented sensitive value disclosure (near-zero DRs) across all threshold settings.
• Non-uniform threshold settings (level-wise, frequency-based) demonstrated superior data utility and minimal privacy risk compared to other models.
• Anonymized data showed minimal impact on the strength of discovered adverse drug reaction (ADR) signals (e.g., PRR, ROR).

Conclusions:

• A novel privacy model and anonymization algorithm were developed for protecting SRS data with unique characteristics.
• The proposed method effectively addresses privacy concerns in SRS data without compromising data utility for ADR signal detection.
• Empirical evaluation on the FAERS dataset validates the method's efficacy in real-world scenarios.