Predicting mortality risk for infants weighing 501 to 1500 grams at birth: A National Institutes of Health Neonatal Research Network report

OBJECTIVESTo develop and evaluate a model that predicts mortality risk based on admission data for infants weighing 501 to 1500 grams at birth, and to use the model to identify neonatal ICUs where the observed mortality rate differs significantly from the predicted rate. DESIGNValidation cohort study. SETTINGUniversity-based, tertiary care neonatal ICUs. PATIENTSSample of 3,603 infants with birth weights of 501 to 1500 grams who were born at seven National Institute of Child Health and Human Development (NICHHD) Neonatal Research Network Centers, over a 2-yr period of time. INTERVENTIONSNone. MEASUREMENTS AND MAIN RESULTSBased on logistic regression analysis, admission factrs associated with mortality risk for inborn infan weredecreasing birth weight, appropriate size for gestational age, male gender, non-black race, and 1-min Apgar score of ≤3. The mortality prediction model based on these factors had a sensitivity of 0.50, a specificity of 0.92, a correct classification rate of 0.82, and an ar under the receiver operating characteristis curve of 0.82 when applied to a validation sam ple. Goodness-of-fit testing showed that then was a marginal degree of fit between the obses vations and model predictions (X = 15.4, p = .05 The observed mortality rate for 3,603 infa at the seven centers was 24.7%, ranging from 21.8% to 27.7% at individual centers. These were no statistically significant difference between observed and predicted mortality rats at any of the centers. One center had an of served mortality rate that was 2.8% lower the predicted by the model (95% confidence inte −6.0% to 0.5%), and another center had as observed rate that was 3% higher than expects (95% confidence interval −0.3% to 6.2%). CONCLUSIONSMortality risk for infants weiing 501 to 1500 grams can be predicted based admission factors. However, until more ace rate predictive models are developed and dated and the relationships between care p tices and outcomes are better understood, su models should not be relied on for evaluat the quality of care provided in different neo tal ICUs. (Crit Care Med 1993; 21:12–18)