Development of Z-score-based models to predict perinatal mortality in pregnancies complicated by fetal growth restriction managed at a tertiary care hospital

Tipiani O.; Arce E. · European Journal of Obstetrics and Gynecology and Reproductive Biology (2025) · DOI: 10.1016/j.ejogrb.2025.114612

Resumen

Objective: To develop predictive models using fetal weight Z-scores to predict perinatal mortality in pregnancies affected by fetal growth restriction (FGR) in high-risk settings. Study Design: This retrospective cohort included singleton pregnancies without congenital anomalies, delivered after 30 weeks of gestation in a tertiary hospital between 2019 and 2022. FGR was defined as weight below the 3rd percentile according to three fetal growth charts. Optimal Z-score cutoffs were identified to maximize sensitivity and specificity for perinatal mortality. Logistic regression models incorporated dichotomized Z-scores and significant predictors. Model performance was assessed using ROC curves, calibration plots, and decision curve analysis (DCA). Results: Among 17 991 singleton pregnancies, 16 177 met the inclusion criteria. FGR was identified in 369 (2.3 %), 157 (1.0 %), and 335 (2.1 %) foetuses using the Hadlock, INTERGROWTH-21st (IG-21), and local customized charts, respectively. Perinatal mortality occurred in 30/369 (8.1 %), 25/157 (15.9 %), and 31/335 (9.3 %) of these groups. Z-score thresholds (−2.61, −3.17, −2.95) showed strong predictive capacity, with adjusted AUCs of 0.878, 0.884, and 0.917. The final model was logit(p) = β0 + β1Z-score − β2 Gestational age − β3 Preeclampsia (preeclampsia excluded in IG-21). Calibration errors averaged 0.02, 0.04, and 0.02; adjusted odds ratios were 18.7, 37.6, and 23.7. The decision curve analysis suggests that integrating these models into clinical protocols may enhance perinatal risk stratification. Conclusion: Three Z-score-based models accurately predicted perinatal mortality in FGR-complicated pregnancies in a high-risk setting. Their integration into clinical protocols may support individualized risk stratification and decision-making.