© 2008 American Thoracic Society
Estimating the Prevalence of Tuberculosis Infection in the United States PopulationTo the Editor:With great interest I read the analysis of the tuberculin skin test component of the National Health and Nutrition Examination Survey (NHANES) by Dr. Bennett and collaborators (1). I stumbled over the assertion, conveying the impression of an afterthought, that analytic techniques like mixture analysis cannot be easily applied to this survey. I wondered then why a cutoff technique (be it 10 mm or a "mirror technique") could be applied: the idea is the same, to obtain an estimate of the prevalence of tuberculosis infection. As Bennett and coworkers' Figure 1 makes obvious, the test readers had a considerable digit preference for 10 (and 15 and 20) mm. Yet, this was precisely and rather counterintuitively the cutoff chosen for the analysis. While for daily clinical work such a memorizable cutoff point to denote tuberculosis infection seems perfectly adequate as a putative balance between false-negative and false-positive results in an individual, one would have hoped for a slightly more sophisticated approach given the resources that must have been put into such a large national survey. Rather crudely, just to exemplify the point, the frequency of reactors was estimated here from the graph provided by the authors, and the simplest application of publicly available mixture model scripts (2) was used to obtain the results shown in Figure 1. This mixture analysis with a Bayesian approach (obviously in need of refinement with the actual data) would seem to be statistically more solid than a cutoff point. A cutoff point, particularly given the digit preference, has no biological basis; the emphasized part of the histogram contrasts the modeled distribution, which is more commensurate with what might be expected from historical data (3). In fact, this particular cutoff point is precisely that which one should probably avoid using. It would also appear that Bayesian credibility intervals for the model estimates might better express uncertainty than a mechanically calculated confidence interval around a profoundly arbitrary proportion.
International Union Against Tuberculosis and Lung Disease FOOTNOTES Conflict of Interest Statement: H.L.R. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. REFERENCES
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
