This mixture analysis of a tuberculin skin test survey conducted in India shows the observed (histogram) distribution, the mixture distribution (thin line) and the component distribution of infection with M. tuberculosis (thick line). The component distribution from non-specific cross-reactions is omitted. It is a selection of a particular age group and males, an example that demonstrates the issues best.

In the original analysis, a cut-off point of 10 mm or more was chosen to indicate tuberculous infection. The mixture analysis of the data indicates that the errors made with this cut-off point cancel each other out. The error with the assumption about sensitivity (excluding infected persons) and the error made in the assumption about specificity (including people without tuberculous infection) are similar in magnitude. This is so by chance and does not need to be the case in another setting (or another age group or among females).

Mixture analysis has only recently been applied to tuberculin skin test surveys and is thus in its infancy. It offers, however, an alternative to the crude and frequently erroneous methods used so far.

To top