Temporal Bias in Predicting Adverse Perinatal Outcomes in Dichorionic Twins

Giorgione et al. [1] report findings from their cohort study investigating the prediction of adverse perinatal outcomes using estimated fetal weight (EFW) and Doppler indices measured within 2 weeks of delivery or an adverse event. The primary outcome was defined as either stillbirth (of one or both twins) or medically indicated early preterm birth (ePTB) before 34 weeks' gestation due to fetal concerns. The combination of EFW discordance and umbilical artery pulsatility index (UA PI) discordance showed strong predictive performance, with an area under the ROC curve (AUC) of 0.90. However, the median gestational age at delivery was reported as 36.71 weeks. If EFW and UA PI measurements were taken within 2 weeks of delivery, as described, the median gestational age at measurement would have been 34.71 weeks or later. This suggests that in most cases, the predictive assessments were conducted after 34 weeks' gestation—after the window for the outcome of ePTB < 34 weeks had closed. Consequently, the model's predictive performance for this outcome would be artificially inflated, as the outcome could no longer occur in many cases. Additionally, scan timing is generally linked to clinical management. If an iatrogenic preterm birth is planned, a scan is likely to occur shortly beforehand. This introduces a form of reverse causality, where the outcome influences the timing of measurement, potentially exaggerating the apparent predictive performance of the model. These temporal mismatches raise important concerns about the validity and clinical applicability of the findings. To properly evaluate the prediction of iatrogenic preterm birth before 34 weeks, analyses should be restricted to routine assessments performed prior to that gestational age. All authors contributed equally. The authors have nothing to report. The authors declare no conflicts of interest.