Bayesian latent class analysis produced diagnostic accuracy estimates that were more interpretable than composite...

Area of Science:

• Medical Diagnostics
• Statistical Modeling
• Infectious Diseases

Background:

• Accurate evaluation of extrapulmonary tuberculosis (TB) diagnostic tests is hindered by the absence of a gold standard.
• Latent class analysis (LCA) offers a statistical approach to estimate test accuracy, adjusting for imperfect reference tests, unlike composite reference standards (CRSs).

Purpose of the Study:

• To illustrate the application of Bayesian LCA for evaluating extrapulmonary TB tests without a gold standard.
• To compare the performance of Bayesian LCA with CRSs in estimating diagnostic accuracy for extrapulmonary TB.

Main Methods:

• Re-analysis of a dataset from New Delhi, India, involving presumptive extrapulmonary TB cases across three forms: lymphadenitis, meningitis, and pleuritis.
• Application of Bayesian LCA, incorporating data from culture, smear microscopy, Xpert MTB/RIF, and cytopathology/histopathology or adenosine deaminase (ADA) tests.
• Comparison of Bayesian LCA results with estimates derived from sequentially defined CRSs.

Main Results:

• Bayesian LCA provided estimates for the accuracy of all evaluated tests and extrapulmonary TB prevalence.
• Xpert MTB/RIF sensitivity was comparable to culture for TB lymphadenitis and meningitis but lower for TB pleuritis; all microbiological tests showed near 100% specificity.
• Non-microbiological tests had high sensitivity but moderate specificity, limiting their utility for disease rule-in, while CRS estimates varied widely and were not consistently accurate.

Conclusions:

• Bayesian LCA provides more interpretable accuracy estimates for extrapulmonary TB tests by accounting for known test performance characteristics.
• The study advocates for increased consideration of Bayesian LCA in the evaluation of diagnostic tests for extrapulmonary TB, especially when a gold standard is lacking.