Estimating diagnostic test accuracy using a "fuzzy gold standard"

Area of Science:

• Medical diagnostics
• Biostatistics
• Diagnostic accuracy research

Background:

• Evaluating diagnostic test accuracy is crucial in medicine.
• Criterion standards, often called "gold standards," are used for this evaluation.
• Errors in the gold standard can significantly impact accuracy assessments.

Purpose of the Study:

• To analyze the impact of erroneous gold standards on diagnostic test accuracy using ROC curves.
• To propose methods for correcting accuracy estimates when the gold standard is imperfect.
• To investigate the interplay between diagnostic test errors and gold standard errors.

Main Methods:

• Utilized Monte Carlo simulations to model diagnostic accuracy.
• Introduced the concept of a "fuzzy" gold standard (FGS) with inherent errors.
• Developed and evaluated two methods requiring probabilistic gold-standard statements, including the "two-truth" method.

Main Results:

• Inaccurate gold standards can lead to underestimation or overestimation of test accuracy, depending on error dependency.
• When test and FGS errors are independent, accuracy is underestimated.
• When test and FGS errors are dependent, accuracy can be overestimated.
• The "two-truth" method effectively reduces underestimation bias when errors are independent.
• The "two-truth" method may still lead to overestimation if errors are dependent.

Conclusions:

• The accuracy of diagnostic tests can be biased by errors in the gold standard.
• Probabilistic gold standards and methods like "two-truth" offer potential solutions for bias correction.
• Careful consideration of error dependency is necessary when applying bias correction methods to diagnostic accuracy studies.