A Bayesian finite mixture model approach to evaluate dichotomization method for correlated ELISA tests

Area of Science:

• Veterinary diagnostics
• Biostatistics
• Biomarker analysis

Background:

• Diagnostic accuracy studies often dichotomize continuous biomarker data.
• Bayesian latent class models (BLCM) are frequently used with dichotomized data, employing binomial or multinomial distributions.
• This dichotomization process can result in significant information loss and reduced outcome reliability.

Purpose of the Study:

• To evaluate the limitations and disadvantages of dichotomizing continuous biomarkers.
• To compare model estimates with true values when continuous data are dichotomized.
• To emphasize the importance of data preservation in diagnostic test evaluation.

Main Methods:

• Comprehensive simulation studies were conducted.
• Continuous biomarker data from two correlated tests were analyzed.
• The impact of dichotomization on Bayesian latent class models was assessed.

Main Results:

• Dichotomization of continuous biomarkers led to notable disparities between true values and model estimates.
• Information loss during dichotomization significantly impacted the reliability of diagnostic accuracy and prevalence estimates.
• The accuracy of the reference test is critical for obtaining dependable estimates.

Conclusions:

• Veterinary researchers should exercise caution when dichotomizing continuous biomarker data.
• Preserving the continuous nature of biomarker data is essential for accurate diagnostic test evaluation.
• Reliable estimation of test accuracy and prevalence requires avoiding data dichotomization.