Setting: Korea.
Objective: Estimation of the prevalence of tuberculous infection from tuberculin skin test surveys can be difficult if cross-reactions resulting from infection with environmental mycobacteria outweigh reactions resulting from tuberculous infection. Mixture analysis was thus chosen as a novel approach for estimating the prevalence of tuberculous infection in Korea.
Design: Seven tuberculin skin test surveys conducted between 1965 and 1995 were analyzed by mixture models, a statistical methodology used either to estimate a prevalence or to classify individuals into predefined homogeneous sub-populations. A Bayesian approach including prior information on component distributions was taken. The final model was selected based on the fit to observed values, and the analysis was therefore stratified by sex and year of survey and included age as a covariate.
Results: The results showed a large decrease in tuberculous infection in the population below 30 years of age: among 10- to 14-year-old boys (girls), infection prevalence decreased from 74.5% (67.9%) in 1965 to 16.5% (16.9%) in 1995. Additionally, the mean induration for individuals with tuberculous infection decreased by 2mm between 1965 and 1995, indicating a changing sensitivity of tuberculin over time.
Conclusions: Mixture analysis is a promising approach for determining the prevalence of tuberculous infection in the presence of substantial interference from infection with environmental mycobacteria and changing tuberculin reaction sizes over time.