LAUSR

With great interest, we read the study by Wilcox et. al. “The limits of small‐for‐gestational‐age as a high‐risk category” [1], which showed that small‐for‐gestational age (SGA) was a poor marker for predicting neonatal mortality and cerebral palsy, whilst gestational age performed well. Following the same selection criteria, we used birth certificate data of Aus‐ trian newborns between 1984 and 2019 (n = ~ 2,920,000) (provided by Statistics Austria) and found an AUC value of 85.3 (95% CI: 84.7–85.9) for gestational age as a predictor of neonatal mortality. This estimate perfectly mirrors the ROC values reported in Wilcox et al. analyzing data of the US and Norwegian births. In addition to confirming the findings reported in Wilcox et al. [1] we would like to draw attention to two additional points as follows: In the discussion outlined in Wilcox et al. [1], authors used the term “birthweight” several times; a few examples are “The discriminating power of birthweight is even fur‐ ther reduced in preterm weeks”, “This suggests that ROC analyses of birthweight for virtually any other birth out‐ comes are likely to show even worse prediction,” and “It may seem incongruous that birthweight so poorly discrimi‐ nates neonatal mortality and cerebral palsy”. We believe the “birthweight” mentioned in the discussion stands for SGA because their method sections explicitly noted that birthweight percentiles (i.e., SGA) were analyzed. Although birthweight was not analyzed in Wilcox et al., we were able to conduct this analysis in our data based on Austrian births. We examined the effect of birthweight (in grams) on neo‐ natal mortality and calculated the corresponding AUC and maximum of the Youden index. We found an AUC value of 87.2 (95% CI: 86.7,87.7) for birthweight as a predictor of neonatal mortality (see Online Resource 1), which was even numerically higher than that for gestational age. In the “History of SGA” section, Wilcox et al. described the difference between birthweight (specifically low birth‐ weight that defines babies < 2500 g) and SGA, and that SGA was preferred to using low birthweight as a binary variable. Interestingly, in our analysis using birthweight as a continu‐ ous variable in lieu of gestational age the maximum of the Youden index turned out to be at a cutoff of 2240 g, which is close to the cutoff of 2500 g to define low birthweight. Dif‐ ferent criteria can be used to identify optimal cut‐offs to cre‐ ate subgroups using the Youden index. Wilcox et al. use the maximum of the Youden index as a criterion to group births into pre‐and full‐term deliveries [2]. Using this criterion in our analysis, we observed a cutoff of < 36 weeks which is similar as in Wilcox et al. [1]. Another well‐known criterion is to use the Youden index with minimal distance to the top‐ left corner of the ROC curve. Following this criterion, our data showed a shift of the optimal cutoff from < 36 weeks to < 37 weeks and from 2240 to 2435 g for defining pre‐term delivery. The cutoff value of < 37 weeks for preterm delivery with the highest Youden index found in the study by Wil‐ cox et al. as well our cutoff surprisingly coincides well with the often‐used “accumulated clinical judgment”. Without a doubt, grouping newborns into pre‐and full‐term deliveries by gestational age, SGA, or exact birthweight makes sense in a clinical setting in practical terms. Nevertheless, using a cutoff to create a binary variable for gestational age as the outcome (dependent variable) or exposure (independent variable) in an analysis may lead to arbitrary effect estimates * Thomas Waldhoer [email protected]