Comparison of Neonatal Mortality and Morbidity Prediction Tools for Surgical Nec Patients

Background: Necrotizing enterocolitis (NEC) is a leading contributor of neonatal morbidity and mortality. Few risk-prediction models exist to inform discussions with families before surgery or guide risk-adjustment for between center comparisons of surgical outcomes. Useful prediction models should be parsimonious, validated, well-calibrated, and include inputs using readily accessible pre-operative data. In addition, the performance of currently available prediction tools, including the Score for Neonatal Acute Physiology Perinatal Extension (SNAPPE-II), Vermont Oxford Risk Adjustment Tool (VON) and American College of Surgeons National Surgical Quality Improvement Program Pediatric (NSQIP-P), in predicting outcomes among infants with NEC is uncertain. Purpose: To compare how well existing tools predict death or intestinal failure (IF) in patients with surgical NEC, and to derive and validate a novel hybrid prediction model using elements from existing models. Methods: An observational cohort study was conducted at a quaternary referral NICU from 9/09-5/15. Infants with surgical NEC were identified and variables extracted from medical records per model specifications. Death was evaluated (in hospital) and IF was defined as failure to achieve full feeds by 90 days post-operatively. Logistic regression was used to model the probability of death, alone, and death or IF, as a composite outcome. Model discrimination was assessed using the area under receiver operating curve (AUC). Backward step-wise selection with threshold P-values < 0.1 was used to create a new hybrid model, which was validated using patients with surgical NEC from a separate quaternary referral NICU. Results: Using a sample of 147 patients with surgical NEC, mortality prediction was highest with NSQIP-P (AUC of 0.80; 95% CI 0.72-0.87), outperforming the SNAPPE-II (AUC 0.68; 95% CI 0.53-0.82) and VON (AUC 0.71; 95% CI 0.62-0.80) models. The following 5 variables were selected from these existing models to evaluate in a hybrid model: preoperative inotrope use, mean blood pressure, PO2/FiO2 ratio, hypothermia, and serum pH. The hybrid model predicting mortality demonstrated good discrimination (AUC 0.81; 95% CI 0.74-0.87) and fit (Hosmer-Lemeshow P=0.94). In external validation (n=76), the model retained performance for death alone (AUC 0.92; 95% CI 0.86-0.98) and had slightly lower performance for death or IF (AUC 0.77; 95% CI 0.66-0.86). Conclusions: Prediction of death and IF among infants with surgical NEC is possible using existing tools and to a greater extent using a newly proposed hybrid model. Further external validation is needed before this model can be used clinically.