On Anonymizing Medical Microdata with Large-Scale Missing Values - A Case Study with the FAERS Dataset^.

Area of Science:

• Health Informatics
• Data Privacy
• Big Data Analytics

Background:

• Big data analysis drives economic growth, but raises privacy concerns, especially with sensitive medical data.
• Current privacy-preserving data publishing methods assume complete datasets, which is unrealistic for real-world medical data.
• The presence of missing values in medical datasets complicates privacy protection and data utility.

Purpose of the Study:

• To investigate the impact of missing values on medical data privacy.
• To evaluate existing anonymization strategies for datasets with missing values.
• To propose a novel privacy protection strategy for incomplete medical data.

Main Methods:

• Analysis of the US FAERS (Adverse Drug Events) dataset.
• Evaluation of three intuitive anonymization strategies: exclusion, inclusion, and imputation of missing values.
• Development and application of a new 'consolidation' strategy with a privacy protection model and anonymization algorithm.

Main Results:

• Intuitive strategies for handling missing values are inadequate for anonymizing large-scale medical datasets.
• The proposed consolidation strategy effectively prevents privacy disclosure.
• The consolidation method sustains data utility for adverse drug reaction (ADR) signal detection.

Conclusions:

• Missing values significantly challenge privacy in medical big data.
• A novel consolidation strategy offers a robust solution for privacy-preserving anonymization of incomplete medical datasets.
• This approach ensures both data privacy and utility for critical applications like ADR signal detection.