We found the study by Broussard et al entitled “Maternal treatment with opioid analgesics and risk for birth defects” very interesting; however, we believe there is a serious flaw in the authors’ interpretation of recall bias. Although the authors acknowledge potential recall bias among women who delivered a baby with malformations when compared with control women with healthy children, they state that “even in studies of reproductive outcomes, exposure misclassification is likely to be non-differential and results in estimates biased towards the null.”

We believe this statement is opposite to what it should be, and misquotes. Studies have confirmed that the recall of women who deliver children with malformations would be better than that of the nonaffected controls for various reasons, and estimates would therefore be biased away from the null. Women who have malformed children are much more likely to reflect on anything that they had done and undergone in pregnancy, whereas women with healthy outcome are less likely to remember, eg, codeine taken for a root canal procedure for several days. Use of a second group of controls (so-called affected controls) would probably control for this issue, and is a common means of tackling the issue of recall bias. The fact that the authors found positive signals in different organs that are biologically unconnected further supports the existence of nonspecific recall and reporting bias.