AI-derived cardiotocography age as a digital biomarker for predicting future adverse pregnancy outcomes

Cardiotocography (CTG) is a widely used, cost-effective non-invasive fetal monitor, yet its predictive potential for adverse pregnancy outcomes is understudied. This study constructed an AI model based on CTG signals to estimate fetal biological age (CTGage), and validated CTGage-gap (CTGage minus true gestational age) as a novel digital risk biomarker. Based on 61,140 CTG signals from 11,385 pregnancies, a distribution-aligned augmented 1D residual CNN (Net1d) was trained for CTGage prediction. Subjects were split into five CTGage-gap subgroups, and severe underestimation and overestimation subgroups were pooled as a high-risk cohort for outcome incidence comparison. The average absolute error of the model was 10.86 days. The incidence of preterm birth and gestational diabetes mellitus (GDM) in the overestimation group increased, while in the underestimation group, there were increased risks of low birthweight, hypertensive disorders of pregnancy (HDP), and maternal anemia (all p < 0.05). The AI-driven CTGage can be used as a feasible non-invasive biomarker for predicting adverse pregnancy outcomes.