Are My EHRs Private Enough? Event-Level Privacy Protection

Area of Science:

• Medical Informatics
• Data Privacy
• Health Data Security

Background:

• Sharing human subject data for research is vital but raises significant privacy concerns.
• Current binary consent models (opt-in/opt-out) limit data availability due to patient concerns about specific sensitive conditions.
• Protecting sensitive medical information while enabling secondary data analysis is a key challenge.

Purpose of the Study:

• To propose and evaluate an event-level privacy protection method for electronic medical records (EMRs).
• To develop a feature ablation technique to safeguard sensitive diagnoses within EMRs.
• To assess the feasibility and efficacy of feature ablation for protecting sensitive medical conditions.

Main Methods:

• Developed a feature ablation method for event-level privacy protection in EMRs.
• Utilized a dataset comprising 13 sensitive diagnoses for evaluation.
• Assessed the decrease in identifiability of sensitive medical conditions as features were progressively excluded.

Main Results:

• Identifiability of sensitive diagnoses decreased at different rates with feature ablation.
• Diseases were categorized based on identifiability decline: 5 with fast decline (<400 features), 7 with progressive decline (200-700 features), and 1 with slow decline (>1000 features).
• The majority of sensitive diseases (12/13) showed significant identifiability reduction, indicating method effectiveness.

Conclusions:

• The proposed feature ablation method shows promise for event-level privacy protection in EMRs.
• This approach can potentially increase the amount of sharable patient data for secondary analyses.
• Further research can refine this technique for broader application in health data privacy.