In their Letter to the Editor, titled “Statistical significance of eating disorders and adverse perinatal outcomes,” Price and Glenn bring out the issue of multiple comparisons and the potential effects of the used statistical methods on the results and on their interpretation.

The authors suggest that the conclusions of the study, “Pregnancy, obstetric, and perinatal health outcomes in eating disorders,” are not supported by the data because of not correcting the standard value of P = .05. The authors suggest the use of the Bonferroni correction or another method to reduce the risk of type 2 error. We believe that the authors actually meant the risk of type 1 error, in which a finding is falsely reported as positive, instead of type 2 error.

We thank the authors for their comments and their suggestions to increase the validity of the study. We acknowledge that considering the high number of outcomes, some of the observed associations may have occurred by chance, as mentioned in the limitations section of the study. However, the use of statistical methods to account for multiple comparisons, as suggested by Price and Glenn, contains certain pitfalls discussed in detail by Perneger (1998) and Morgan (2007). The Bonferroni method decreases type 1 errors but increases type 2 errors, concealing true differences between the groups. Moreover, the question addressed by the method, whether all null hypotheses are true simultaneously, is largely clinically irrelevant. Perneger even states that the Bonferroni method creates more problems than it solves.

We consider that the issue of multiple comparisons is best addressed by defining detailed hypotheses based on previous evidence. Hypotheses presented in the study are based on a number of previous studies (ie, we did not conduct several statistical tests randomly, which reduces demand for Bonferroni correction or similar methods). The findings of this study were largely consistent with the hypotheses.

This study aimed to quantify the magnitude of the risk of pregnancy and obstetric complications and adverse perinatal health outcomes in eating disorders, as opposed to merely exploring whether the risks are elevated among women with an eating disorder. The results for most of the outcomes are thus reported and interpreted based on odds ratios instead of P values. We consider that in this way the reader gets a richer understanding of the phenomena related to pregnancy and childbirth among women with an eating disorder.