Inference about misclassification probabilities from repeated binary responses.

Area of Science:

• Biostatistics
• Epidemiology
• Genetics

Background:

• Assessing diagnostic test reliability is crucial, especially without a gold standard for comparison.
• Misclassification errors (false-positive and false-negative) impact reliability assessments.
• Repeated binary responses offer valuable data for estimating these errors.

Purpose of the Study:

• To develop and illustrate a latent class model for diagnostic test reliability assessment.
• To estimate misclassification error probabilities (false-positive and false-negative) in the absence of a gold standard.
• To test the hypothesis that misclassification error rates are zero.

Main Methods:

• Utilized a latent class model approach.
• Applied the model to repeated binary response data.
• Considered data with inter-individual variation.

Main Results:

• The latent class model effectively estimates misclassification probabilities.
• The model allows for hypothesis testing of zero error rates.
• Demonstrated application using serological data from atomic bomb survivors and their children.

Conclusions:

• The proposed latent class model provides a robust method for evaluating diagnostic test reliability.
• This approach is valuable for understanding measurement error in biological and medical studies.
• The findings contribute to more accurate reliability assessments in situations lacking a definitive reference standard.