Perinatal risk assessment in pregnancies complicated by early-onset fetal growth restriction: development and internal validation of a prediction model for composite adverse perinatal outcome

Objective: Early-onset fetal growth restriction (FGR) frequently requires iatrogenic preterm delivery to prevent stillbirth or the sequelae of hypoxia. A prediction model for adverse perinatal outcome could aid clinical decision-making and parental counseling. However, an adequate externally validated model, including predictors that are readily available for this purpose, is currently lacking. The aim of the present study was to develop a prediction model for composite adverse perinatal outcome (CAPO) to be used at the time of hospital admission for fetal monitoring in early-onset FGR. Methods: A model was developed to predict CAPOs (defined as one or more of the following: perinatal or in-hospital mortality, necrotizing enterocolitis ≥ Stage IIA, moderate or severe bronchopulmonary dysplasia, cystic periventricular leukomalacia, intraventricular hemorrhage ≥ Grade III or venous infarction and/or culture-proven sepsis) in early-onset FGR. The model was developed and internally validated in the OPTICORE study, a retrospective, multicenter cohort study of pregnancies complicated by early-onset FGR, in accordance with the consensus-based definition, in six academic hospitals in The Netherlands. Candidate predictors were selected based on the existing literature and expert opinion and comprised maternal medical history, obstetric history, fetal growth and Doppler assessment. A backward stepwise elimination procedure was performed based on the Aikake Information Criterion. Internal validation was performed by bootstrapping and repeating the predictor selection process to determine the shrinkage factor to adjust for overfitting. Internal–external cross-validation was performed as a sensitivity analysis to assess the impact of clustering of patients within each hospital. Results: In total, 560/1453 (38.5%) pregnancies were complicated by CAPO. The developed model included 14 predicting variables, determined at the time of hospital admission for fetal monitoring: maternal history of chronic kidney disease or chronic hypertension, smoker, previous pregnancy complicated by FGR, gestational age at admission, fetal sex, concomitant pre-eclampsia, the use of magnesium sulfate, gestational diabetes mellitus, estimated fetal weight, umbilical artery and middle cerebral artery pulsatility index percentile, absent or reversed end-diastolic flow in the umbilical artery and gestational age at diagnosis of FGR. After internal validation and shrinkage to adjust for optimism, the model performed well (area under the receiver-operating-characteristics curve, 0.83 (95% CI, 0.79–0.87); calibration slope, 1.05 (95% CI, 0.94–1.17); calibration-in-the-large, 0.07 (95% CI, −0.06 to 0.20)). The internal–external cross-validation sensitivity analysis revealed equivalent model performance measures across the three largest hospitals. Conclusions: The developed model, including 14 readily available predictors, showed good performance for the prediction of CAPO at the time of hospital admission and may serve as a helpful tool for clinical decision-making and parental counseling in the setting of early-onset FGR. External validation and assessment of the model's impact on clinical decision-making and patient outcomes are required before it can be implemented in routine clinical practice.