N. Rothman et al., MISCLASSIFICATION OF GENETIC SUSCEPTIBILITY BIOMARKERS - IMPLICATIONSFOR CASE-CONTROL STUDIES AND CROSS-POPULATION COMPARISONS, Cancer epidemiology, biomarkers & prevention, 2(4), 1993, pp. 299-303
Phenotype and genotype markers of genetic susceptibility are of increa
sing interest in case-control studies of cancer. It is well establishe
d that bias to the odds ratio is caused by less-than-perfect assay sen
sitivity and specificity and varies with risk factor prevalence. As su
ch, the observed variation in odds ratio between studies of genetic ma
rkers and cancer risk may be real, or may be attributed, in part, to v
ariation in assay accuracy or in risk factor prevalence (e.g., prevale
nce differences between racial groups). The latter can be a particular
concern when the prevalence of the ''at risk'' polymorphism in one or
more populations is either very high (e.g., >85%) or very low (e.g.,
<15%). For example, even very high sensitivity (e.g., 98%) can produce
substantial bias to the odds ratio when the risk factor prevalence is
high. Under some prevalence conditions, however, assays with only mod
erate accuracy are sufficient and result in minimal bias to the odds r
atio. Understanding misclassification in the context of marker prevale
nce may help to explain disparate findings in the literature and shoul
d assist investigators in selecting markers that are appropriate for f
uture studies.