Mixture analysis of tuberculin survey data from northern Malawi and critique of the method

Various methods have been used to estimate the prevalence of Mycobacterium tuberculosis infection from tuberculin survey data. All are complicated by prior sensitisation to environmental mycobacteria and bacille Calmette-Guérin (BCG) vaccination. Mixture analysis has recently been proposed as a means of overcoming misclassification and improving infection prevalence estimates.

To compare conventional and mixture model estimates of M. tuberculosis infection prevalence.

Mixture models with two or three univariate normal components were fitted to the results of 53 909 tuberculin tests conducted in northern Malawi during 1980-1984. Data were stratified by BCG status, sex and age and corrected for digit preference. Prevalence estimates derived from mixture models were compared with those of conventional methods.

The optimal model was age-dependent, with three- and one-component solutions preferred in younger and older age groups, respectively. In contrast with findings from elsewhere, a component corresponding to BCG vaccination was indistinguishable from that attributable to environmental mycobacterial exposure, and infection prevalence estimates in younger individuals with a BCG scar were inflated, irrespective of the method used.

The validity of infection prevalence and incidence estimates based on mixture modelling is probably locale-dependent, and the assumptions underlying mixture models may not realistically reflect underlying immunological processes.