On the use of affected controls to address recall bias in case-control studies of birth defects

Abstract

Inferences regarding causes of birth defects in humans are often based on results of case-control studies conducted after birth. To address bias in these studies caused by potential differential recall of past exposures between case and control mothers, many investigators have advocated the use of affected controls (babies with birth defects other than the one of interest). To evaluate whether the use of affected controls is warranted for a wide range of scenarios, we analyzed data from a population-based case-control study of birth defects in Atlanta, in which there were 4,918 babies with serious defects ascertained in the first year of life and 3,029 babies without defects. We compared the magnitude of the odds ratios for 10 specific defects--risk factor associations between normal and affected controls. These associations included demographic factors (e.g., advanced maternal age and Down syndrome), chronic maternal illnesses (e.g., diabetes and cardiac defects), chronic exposures (e.g., multivitamins and neural tube defects), and acute exposures (e.g., flu and neural tube defects). In all instances, the use of affected controls did not change etiologic inferences derived from using normal controls and there were only moderate changes in odds ratios. On the basis of theoretical considerations, we show that recall bias can lead to spurious inferences only under extreme conditions. We conclude that concerns about recall bias are overrated in birth defects studies and that the use of normal controls is acceptable unless evidence of substantial recall bias exists.