Regulator Loss Functions and Hierarchical Modeling for Safety Decision Making

Area of Science:

• Medical device safety
• Health informatics
• Decision theory

Background:

• Regulatory action is crucial for public protection when medical device safety issues arise.
• Conventional surveillance methods struggle to balance risks and benefits effectively.

Purpose of the Study:

• To enhance medical device decision-making by integrating regulator loss functions with statistical safety signal evidence.
• Develop a framework for optimizing regulatory responses to device safety concerns.

Main Methods:

• Utilized the National Inpatient Sample to identify adverse medical device events (AMDEs) in pediatric admissions.
• Developed hierarchical Bayesian models to estimate hospital-level AMDE rates and compute safety signals.
• Integrated custom loss functions with posterior risk calculations to determine optimal actions, comparing this to the conventional Z-score method.

Main Results:

• The proposed minimum-risk decision method yielded different actions for 45% of hospitals compared to the Z-score approach.
• Simulations demonstrated that Bayes risk minimization decisions are superior to Z-score based decisions, even with model misspecification.

Conclusions:

• A decision-theoretic framework offers a promising approach for acting on medical device safety signals.
• Effective implementation requires careful, expert-informed specification of loss functions.