Analysis of latent tuberculosis and mycobacterium avium infection data using mixture models

Area of Science:

• Epidemiology
• Infectious Diseases
• Public Health

Background:

• Estimating latent tuberculosis infection (LTBI) prevalence is difficult in regions with low tuberculosis rates but high exposure to non-tuberculous mycobacteria (NTM) and BCG vaccination.
• Assessing LTBI and M avium infection requires accurate methods, especially in infant populations with complex exposure histories.

Purpose of the Study:

• To assess LTBI and M avium infection in infants.
• To estimate the probability of these infections based on skin test responses in a population with high NTM and BCG exposure.

Main Methods:

• A population-based survey using tuberculin skin tests (TST) and M avium sensitin on 7-year-old infants in Biscay, Spain.
• Comparison of standard estimation methods with a mixture model approach using the Expectation Maximization algorithm.
• Development of functions to estimate LTBI and M avium infection probabilities based on skin test results for vaccinated and unvaccinated children.

Main Results:

• LTBI prevalence estimates varied significantly by method; the mixture model yielded higher estimates than some standard approaches.
• Exposure to BCG vaccination caused an average 3 mm upward shift in TST induration size for a given infection probability.
• The study confirmed widespread NTM exposure impacting tuberculin survey interpretations.

Conclusions:

• NTM exposure significantly affects tuberculin survey results and must be considered in assessments.
• Mixture analysis within an empirical Bayes framework enhances LTBI probability estimation in BCG-vaccinated populations with NTM exposure.
• The developed models, combined with new blood tests, can improve the specificity and cost-effectiveness of tuberculosis surveys.