TY - JOUR
T1 - Antenatal prediction of neonatal mortality in very premature infants
AU - Ravelli, Anita C. J.
AU - Schaaf, Jelle M.
AU - Mol, Ben Willem J.
AU - Tamminga, Pieter
AU - Eskes, Martine
AU - van der Post, Joris A. M.
AU - Abu-Hanna, Ameen
PY - 2014
Y1 - 2014
N2 - To develop a prognostic model for antenatal prediction of neonatal mortality in infants threatening to be born very preterm ( <32 weeks). Nationwide cohort study in The Netherlands between 1999 and 2007. We studied 8500 singletons born between 25(+0) and 31(+6) weeks of gestation where fetus was alive at birth without congenital anomalies. We developed a multiple logistic regression model to estimate the risk of neonatal mortality within 28 days after birth, based on characteristics that are known before birth. We used bootstrapping techniques for internal validation. Discrimination (AUC), accuracy (Brier score) and calibration (graph, c-statistics) were used to assess the model's predictive performance. Neonatal mortality occurred in 766 (90 per 1000) live births. The final model consisted of seven variables. Predictors were low gestational age, no antental corticosteroids, male gender, maternal age ≥35 years, Caucasian ethnicity, non-cephalic presentation and non-3rd level of hospital. The predicted probabilities ranged from 0.003 to 0.697 (IQR 0.02-0.11). The model had an AUC of 0.83, the Brier score was 0.065. The calibration graph showed good calibration, and the test for the Hosmer Lemeshow c-statistic showed no lack of fit (p=0.43). Neonatal mortality can be predicted for very preterm births based on the antenatal factors gestational age, antental corticosteroids, fetal gender, maternal age, ethnicity, presentation and level of hospital. This model can be helpful in antenatal counseling
AB - To develop a prognostic model for antenatal prediction of neonatal mortality in infants threatening to be born very preterm ( <32 weeks). Nationwide cohort study in The Netherlands between 1999 and 2007. We studied 8500 singletons born between 25(+0) and 31(+6) weeks of gestation where fetus was alive at birth without congenital anomalies. We developed a multiple logistic regression model to estimate the risk of neonatal mortality within 28 days after birth, based on characteristics that are known before birth. We used bootstrapping techniques for internal validation. Discrimination (AUC), accuracy (Brier score) and calibration (graph, c-statistics) were used to assess the model's predictive performance. Neonatal mortality occurred in 766 (90 per 1000) live births. The final model consisted of seven variables. Predictors were low gestational age, no antental corticosteroids, male gender, maternal age ≥35 years, Caucasian ethnicity, non-cephalic presentation and non-3rd level of hospital. The predicted probabilities ranged from 0.003 to 0.697 (IQR 0.02-0.11). The model had an AUC of 0.83, the Brier score was 0.065. The calibration graph showed good calibration, and the test for the Hosmer Lemeshow c-statistic showed no lack of fit (p=0.43). Neonatal mortality can be predicted for very preterm births based on the antenatal factors gestational age, antental corticosteroids, fetal gender, maternal age, ethnicity, presentation and level of hospital. This model can be helpful in antenatal counseling
U2 - https://doi.org/10.1016/j.ejogrb.2014.02.030
DO - https://doi.org/10.1016/j.ejogrb.2014.02.030
M3 - Article
C2 - 24666798
SN - 0301-2115
VL - 176
SP - 126
EP - 131
JO - European journal of obstetrics, gynecology, and reproductive biology
JF - European journal of obstetrics, gynecology, and reproductive biology
ER -