REPLY

We appreciate Dr Li’s consideration of our article, and we agree that it is important to consider all research findings in the context of potential limitations. The issues Dr Li identified are minor and should not meaningfully affect the study findings or their interpretation.

Although there is likely some residual confounding in our analysis, as in most observational research, the risk factors that Dr Li listed (parental smoking, alcohol consumption, and recreational drug use) have shown far weaker associations than those we observed for pregestational diabetes. In addition, the factors listed by Dr Li have only been associated with certain specific birth defects, whereas we observed strong associations between pregestational diabetes and most birth defects (46 of 50) considered. Where sample size permitted, we controlled for body mass index, race and ethnicity, and maternal education, factors that are more strongly associated with pregestational diabetes and specific birth defects than the factors listed by Dr Li.

Although recall bias is a theoretical concern for any study in which information on exposure is ascertained after the outcome of interest has occurred, most concerns about recall bias relate to the recall of transient exposures that may be more likely to be remembered by women who had an affected pregnancy. However, diabetes is a chronic medical condition and is unlikely to be forgotten by mothers of controls or cases. Furthermore, approximately half of the women in our analysis who reported pregestational diabetes reported type 1 diabetes, which requires continued use of insulin. As stated in our article, the results we observed for any pregestational diabetes were consistent with those observed for type 1 diabetes alone.

The controls in our study were population based and selected from vital records or hospital birth logs.¹ An assessment of the representativeness of our controls using linked birth certificate data showed that control participants were generally representative of the base populations they were selected to represent.²

The concept of risk is better understood than that of odds. A prevalence of 5%—10% is often used as a cut-point for when an outcome is considered rare enough for the odds ratio (OR) to approximate the risk ratio (RR); the prevalence of the most common birth defect we examined, cleft lip with or without cleft palate, has an estimated US prevalence of 0.1%,³ meaning that the OR should be a close approximation to the RR and the interpretations presented in the article are reasonable.