On the measurement and interpretation of birth defect associations in epidemiologic studies.

The extent of clustering of 2 or more defects in the same infant can be expressed as the ratio of the observed number of infants with the defects (O) over the expected number of such infants (E). The expected is usually derived from the product of population rates of individual defects. Because large O/E ratios are obtained for many defect combinations, it has been suggested that clustering of defects is generalized and nonspecific. To control for the tendency of nonspecific clustering of defects, an alternative method is to perform the same calculations among multimalformed infants only. A main limitation of this method is that it adjusts for the clustering tendency of all defects rather than the ones of interest, often resulting in spuriously low O/E ratios. We present a new method to adjust for the tendency for nonspecific clustering between defects that overcomes this limitation. With this method, adjusted O/E ratios are inversely related to the proportion of infants who are multimalformed and have one or more of the defects being examined. Using data from the Metropolitan Atlanta Congenital Defects Program, we apply this method to the previously described associations among VACTERL defects and midline or "schisis" defects. We show that adjusted O/E ratios obtained are greater than those obtained by using multimalformed infants. For midline defects, many of the adjusted ratios were close to one, indicating nonspecific clustering of these defects. Finally, using the example of isotretinoin embryopathy, we show that O/E ratios depend highly on the frequency of exposure in the population, and thus, they should be interpreted with caution.