Classification under uncertainty: data analysis for diagnostic antibody testing.

Area of Science:

• Biomedical diagnostics
• Decision theory
• Statistical modeling

Background:

• Accurate classification is crucial for diagnostic and antibody tests.
• Ignoring disease prevalence and uncertainty can cause significant classification errors.
• Existing methods may not adequately address prevalence variations.

Purpose of the Study:

• To present a novel classification strategy using optimal decision theory.
• To minimize false positives and false negatives in diagnostic testing.
• To develop a method robust to unknown disease prevalence and measurement uncertainty.

Main Methods:

• Leveraging optimal decision theory to define classification domains.
• Incorporating assumed prevalence and conditional probability models.
• Generalizing the strategy for unknown prevalence using hold-out samples or adaptive algorithms.
• Applying the method to a SARS-CoV-2 serology test case.

Main Results:

• The novel strategy significantly decreases classification error compared to traditional methods.
• Achieved up to a decade improvement in classification accuracy.
• Demonstrated the ability to handle unknown prevalence and measurement uncertainty.
• Provided a theoretical link between optimization and receiver operating characteristic analysis.

Conclusions:

• Optimal decision theory offers a robust framework for diagnostic test classification.
• The proposed method enhances accuracy by accounting for prevalence and uncertainty.
• This approach provides a foundation for improved diagnostic test development and evaluation.