Table 6 Comparison of our results with related literature
From: Prediction of neonatal deaths in NICUs: development and validation of machine learning models
First Author (Reference) | Numbers of variables | List of variables | Â | Study sample | Models | Tool | External validation | Performance evaluation | Â |
|---|---|---|---|---|---|---|---|---|---|
Jaskari [13] | 10 | Heart rate, blood pressure, GA, BW, medical scores SNAP-II and SNAPPE-II, diagnoses of the patients for BPD, NEC, ROP, information on the survival | Â | 977 VLBW infants | LR, LDA, QDA, KNN, SVM, 3 different Gaussian process, RF | Matlab | No | AUROC (RF): 0.922 F-Score (RF): 0.477 | Â |
Beluzos[42] | 23 | Mother’s age, BW, 1-min Apgar score, 5-min Apgar score, Robson group classification, number of cesarean deliveries, fetal losses, number of previous gestation, GA, number of live births, number of normal deliveries, birth place code, childbirth care, main worker role, child-birth type (delivery), mother race/skin color, marital status, mother’s years of schooling, week of gestation (by ranges), type of pregnancy, prenatal appointments (by range), newborn presentation type, congenital malformation |  | 698 neonates | XGBoost, LR, RF | Python Programming language (3.6) | Yes | Accuracy (RF): 93% AUC (RF): 0.965 |  |
Rezaeian[3] | 15 | Maternal age, GA, number of fetus, premature rapture of membrane, maternal preeclampsia, birth type, gender of neonate, BW, birth height, birth head circumference, after birth crying, delivery room breathing, CPR in delivery room, 1th minute APGAR, 5th minute APGAR | Â | 1618 premature neonate records | NN, LR | MATLAB R2016a | No | AUC: 95.99% Accuracy: 96.789 Sensitivity: 86.20% Specificity: 98.37% | Â |
Cooper[12] | 68 | Not mention | Â | 6499 neonates | Superlearning algorithm (14 regression and machine learning algorithms) | SAS version 9.4, R version 3.3.0 | Yes | Cross-validated MSE Excellent discrimination (AUC development: 0.91 AUC validation: 0.87) Good calibration in model development/ not good in model validation | Â |
Ravelli[43] | 13 | GA, fetal gender, use of antenatal corticosteroids, maternal age, parity, caucasian maternal ethnicity, SES, hypertension/pre-eclampsia, PROM, history of preterm birth, bleeding in the second trimester, non-cephalic presentation and level of hospital for delivery (3rd level versus non 3rd level hospital | Â | 8500 singleton very preterm infants | Multiple logistic regression, Bootstrapping technique | SAS version 9.2, R version 3.01 | No | Discrimination (AUC): 0.83 Accuracy: 65% Calibration: Good calibration | Â |
Our study | 17 | BW, GA, preterm birth, SGA, prenatal care, mother disease, RDS, steroid therapy, surfactant administration, pulmonary hemorrhage, congenital malformation, NEC, sepsis, IVH, asphyxia, intubation, ventilation | Â | Neonates admitted to NICU | ANN, C5.0, SVM, Bayesian network, Ensemble | IBM SPSS Modeler | Yes | Accuracy (SVM): 94% Precision (RF): 98% Specificity (RF): 94% Sensitivity (SVM): 95% F-score (SVM): 0.96 AUC (SVM, ensemble): 0.98 | Â |