Bayesian modelling of tuberculosis clustering from DNA fingerprint data

Area of Science:

• Epidemiology
• Molecular epidemiology
• Biostatistics

Background:

• Classifying subject status in epidemiologic studies often relies on multiple tests lacking a gold standard.
• Tuberculosis (TB) molecular epidemiology uses DNA sequencing to distinguish recent transmission from reactivation.
• Accurate estimation of TB transmission rates is crucial for public health control strategies.

Purpose of the Study:

• To demonstrate the utility of Bayesian latent class models for analyzing multiple TB genotyping data.
• To estimate the proportion of clustered TB cases and the operating characteristics of different typing methods.
• To develop a misclassification-adjusted regression model for estimating risk factors associated with TB clustering.

Main Methods:

• Simultaneous analysis of Mycobacterium tuberculosis DNA data from IS6110 restriction fragment length polymorphism (RFLP), spoligotyping, and mycobacterial interspersed repetitive unit-variable-number tandem repeat.
• Application of Bayesian latent class models to account for the lack of a gold standard.
• Utilized continuous (nearest genetic distance) and dichotomous measures from IS6110 RFLP.

Main Results:

• Estimated the proportion of clustered TB cases and the operating characteristics of each genotyping method.
• Provided misclassification-adjusted estimates of risk factors affecting TB clustering probabilities.
• Demonstrated the feasibility of combining diverse data types for improved epidemiological analysis.

Conclusions:

• Bayesian latent class models offer a robust framework for analyzing complex molecular epidemiology data with multiple imperfect tests.
• This approach enhances the accuracy of estimating TB transmission and identifying associated risk factors.
• Careful interpretation is needed when combining continuous and dichotomous test results in epidemiological studies.