Re: Do Birthweight‐For‐Gestational Age Centiles Predict Serious Neonatal Morbidity and Neonatal Mortality?

1.

We thank Malik and colleagues for their comments on our paper [1, 2]. Their letter questioned the reliability of our 39‐week diabetes result, the adequacy of the area under the receiver operating characteristic (ROC) curve (AUC) for a rare outcome, and the gap between birthweight and estimated fetal weight. Although their concerns were anticipated and addressed in our paper, we welcome the opportunity to make the issues explicit. Our study was proposed in response to calls questioning the utility of the small‐for‐gestational‐age (SGA) construct as an indicator of fetal growth restriction [3]. Several studies, including a previous study by our group [4], have highlighted the poor performance of SGA as a univariable predictor of perinatal morbidity and mortality. We therefore hypothesised that continuous birthweight‐for‐gestational‐age centiles could improve the prediction of serious neonatal morbidity and neonatal mortality in a multivariable model [1], and showed that they do so in specific populations.

The 39‐week pre‐pregnancy diabetes development model (United States, 2019–2021) showed the greatest optimism on external validation in a different cohort (United States, 2017–2018). External validation is a stringent test, as it corrects for overfitting in the development sample and accounts for temporal and population differences. On adding birthweight centiles, the AUC increased from 0.69 (95% confidence interval [CI] 0.66, 0.72) to 0.77 (95% CI 0.74, 0.79) in development, and from 0.61 (95% CI 0.57, 0.65) to 0.65 (95% CI 0.61, 0.69) in validation. We disagree that the focus should be restricted to the smaller external validation cohort. The increment is unsurprising, since fetal growth is altered in diabetic pregnancy, may reflect poor glycemic control, and is associated with adverse perinatal outcomes.

We agree that the AUC indicates discrimination and not clinical utility. However, threshold‐based decision analysis to determine clinical utility was not the objective we proposed to address. The goal was to assess the predictive ability of birthweight centiles in univariable versus multivariable models, not to build a clinical tool. The calibration plots were truncated for want of high predicted risks for an uncommon outcome, not due to miscalibration.

Finally, we address birthweight versus estimated fetal weight. As acknowledged in the manuscript, the clinical challenge lies in antenatal prediction of adverse perinatal outcomes, not in postnatal prediction, for which excellent predictors of infant health status are readily available. We carried out our analysis as a proof of concept using birthweight centiles. Measurement errors may attenuate prediction, but the degree of attenuation remains an open empirical question. Furthermore, the object of interest is not the estimated fetal weight but the probability of adverse perinatal outcomes, as determined by a multivariable model based on antenatal predictors, including maternal and pregnancy factors, and fetal growth velocity/trajectory, with serial fetal growth measurements likely to be more informative than cross‐sectional assessments.

To reiterate, our study addressed scepticism about SGA as a predictor of adverse perinatal outcomes, particularly because its utility has traditionally been based on univariable analyses. It should be viewed as a case for a deeper examination of fetal weight‐for‐gestational‐age centiles in the antenatal prediction of fetal health status.

Author Contributions

S.J. and K.S.J. discussed the conceptual issues involved, and S.J. drafted the first version of the letter to the editor. All authors critically reviewed the letter, revised its intellectual content, and approved the final version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Linked Articles

This article is linked to the letter to the editor by Malik et al. To view this article, visit https://doi.org/10.1111/ppe.70182.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.