Analyzing bioassay data using Bayesian methods--a primer

Area of Science:

• Health Physics
• Statistical Analysis
• Nuclear Science

Background:

• Classical statistics in health physics often fail to account for rare events, termed "needle in a haystack" effects.
• This deficiency leads to a high false positive fraction in measurements, where positive results are incorrectly identified as non-zero.
• Traditional methods are inadequate for accurately interpreting measurements of infrequent occurrences.

Purpose of the Study:

• To introduce Bayesian statistics as a superior methodology for minimizing incorrect decisions in health physics measurements.
• To demonstrate the application of Bayesian methods for reducing false positives and false negatives.
• To evaluate the effectiveness of Bayesian approaches in scenarios involving rare events.

Main Methods:

• Application of Bayesian statistical methodology to health physics measurement interpretation.
• Utilizing numerically generated and real bioassay data for tritium analysis.
• Employing various analytical models to fit prior probability distributions and assess model sensitivity.
• Conducting parametric studies to determine optimal decision levels.

Main Results:

• Bayesian statistics effectively minimize false positives and false negatives compared to classical approaches.
• For rare events, the normalized Bayesian decision level (k(alpha)) ranges from 3 to 5.
• A decision level of approximately 4 times the measurement uncertainty (sigma0) is proposed as a better choice than the classical 2 times sigma0.

Conclusions:

• Bayesian statistics provide a robust framework for interpreting health physics measurements, especially for rare events.
• The proposed Bayesian decision level enhances the accuracy of identifying true positives while minimizing false alarms.
• Adopting Bayesian methods can significantly improve the reliability of health physics monitoring and safety assessments.