Conditions for unbiased estimation of sensitivity and specificity when using a probabilistic reference standard could...

Area of Science:

• Diagnostic accuracy studies
• Biostatistics
• Medical research methodology

Background:

• Reference standards in diagnostic accuracy studies can be uncertain.
• Probabilistic reference standards offer an alternative to definite decisions.
• Utilizing probabilistic standards for estimating diagnostic accuracy is complex.

Purpose of the Study:

• Identify conditions for unbiased estimation of sensitivity and specificity.
• Investigate the use of probabilistic reference standards in diagnostic accuracy studies.
• Address challenges in estimating diagnostic test accuracy.

Main Methods:

• Derived conditional distribution of index test given probabilistic reference standard.
• Modeled associations between sensitivity/specificity and the reference standard.
• Investigated bias in model-based and model-free estimates via simulation.

Main Results:

• Model-based estimation requires assuming absence of two of three potential bias components for unbiased results.
• Identifiability issues limit the inclusion of all three bias components in models.
• Model-free estimates demonstrate a high risk of bias.

Conclusions:

• Unbiased sensitivity and specificity estimation with probabilistic reference standards requires ruling out two of three conditions: sensitivity association, specificity association, or miscalibration.
• Adequate statistical methods can yield unbiased estimates when conditions are met.
• Careful consideration of potential biases is crucial when employing probabilistic reference standards.