Article on bayesian methods was rightBMJ 1999; 319 doi: https://doi.org/10.1136/bmj.319.7224.1573 (Published 11 December 1999) Cite this as: BMJ 1999;319:1573
- James E Parker, retired paediatrician ()
EDITOR—Bayes's theorem was invoked in a paper about screening for HIV by Meyer and Pauker in 1987.1 The philosophy imparted had much to do with attitudes to screening for HIV infection and its legacy on the present world pandemic.
These authors assumed a false positive rate of 0.005% for combined Western blot testing and enzyme linked immunosorbent assay (ELISA) and a seroprevalence rate in the general population of 0.01% (this represented the incidence of HIV infection in female blood donors in the United States in 1986). Thus testing a population of 100 000 people would detect 10 genuine cases of HIV infection and yield five false positive results. These odds were considered to be an argument against screening. The fact that female blood donors were not truly representative of the general population and that random false positivity in testing could be appreciably reduced by repeat testing went unnoticed.
This would seem to underscore a statement at the end of Spiegelhalter et al's article: “The perceived problems with the bayesian approach largely concern the source of the prior and the interpretations of the conclusions.”2