Current scoring systems, which adjust prediction for severity of illness, do not account for higher observed mortality in neonatal intensive care units (NICUs) of children's hospitals than that of perinatal centers. We hypothesized that three potential predictors, (a) admission from another NICU, (b) presence of congenital anomalies and (c) need for surgery, would modify expected mortality and/or length of stay for infants admitted to NICUs in children's hospitals. We reviewed consecutive admissions to two NICUs in children's hospitals in Canada. We performed regression analyses to evaluate these potential predictors and severity-of-illness indices for the outcomes of mortality and length of stay. Of 625 neonatal admissions, transfer from another NICU, congenital anomalies requiring admission and surgery were identified in 371 (59%). Using logistic regression, mortality was predicted based on admission from another NICU (odds ratio (OR) 1.92; 95% confidence interval (CI) 1.04, 3.57), congenital anomalies (OR 7.28; 95% CI 3.69, 14.36) and a validated severity-of-illness score, the Score for Neonatal Acute Physiology Perinatal Extension Version II (SNAPPE-II; OR 1.07; 95% CI 1.05, 1.09 per point). By contrast, surgical intervention was predictive of survival (OR 0.35; 95% CI 0.18, 0.67). Length of stay >or=21 days was predicted by SNAPPE-II (OR 1.02; 95% CI 1.01, 1.03 per point), congenital anomalies (OR 2.47; 95% CI 1.60, 3.79) and surgery (OR 2.73; 95% CI 1.77, 4.21). Fair performance comparisons of NICUs with different case-mixes, such as children's hospital and perinatal NICUs, in addition to severity-of-illness indices, should account for admissions from another NICU, congenital anomalies and surgery.