Mixture analysis may be able to better overcome the problems of specificity observed in tuberculin surveys.

The top left figure shows the observed (thick line) and hypothesized underlying component distributions that sum up to the mixture distribution.

In this example, a normal distribution (thin line) with a mode at around 18 mm, indicates those with tuberculous infection.   A lognormal distribution (dotted line) with a mode at about 7 mm indicates the distribution of cross-reactions.

This is a theoretical mixture model (based on current knowledge about distributions) that needs to be tested in different settings.  If this approach proves successful, then the issue of the specificity (and sensitivity) of the tuberculin skin test is overcome: the operating characteristics are known for each millimeter of induration.  The probability of being infected with M. tuberculosis can thus be estimated.

To top