Predicting death from initial disease severity in very low birthweight infants: a method for comparing the performance of neonatal units.

OBJECTIVES: To investigate (a) which clinical variables and physiological measures of disease severity best predict death in very low birthweight infants and (b) their use in comparing mortality between two neonatal units. DESIGN: Retrospective study of two cohorts of very low birthweight infants from overlapping time periods who received mechanical ventilation. SETTING: Two neonatal intensive care units (hospitals A and B). SUBJECTS: 262 Very low birthweight infants, 130 in hospital A, 132 in hospital B. MAIN OUTCOME MEASURE: Death in hospital. RESULTS: In hospital A the mean level of oxygenation in the first 12 hours of life, whether measured as inspired oxygen requirement (FIO2), arterial/alveolar oxygen (a/AO2) ratio, or alveolar-arterial oxygen difference (A-aDO2), was more closely associated with death than any of four "traditional" risk factors: low birth weight, short gestation, the diagnosis of respiratory distress syndrome, and male sex. Mean pH in the first 12 hours was as strongly associated with death as birth weight. Multiple logistic regression models were derived in infants from hospital A using the four traditional risk factors with measures of oxygenation and pH. The validity of each model was then tested in infants from hospital B. The model based on the four traditional risk factors alone predicted death in hospital B with only 31% sensitivity. Adding mean a/AO2 ratio and mean pH increased its sensitivity to 75%, and when mean a/AO2 ratio was replaced by mean FIO2 its sensitivity increased further to 81%. Based on crude mortality rates alone, the odds of death in hospital A versus hospital B were 0.67 (95% confidence interval 0.37 to 1.23). After correcting for traditional risk factors and mean FIO2 and mean pH, however, the odds of death in hospital A increased to 3.27 (1.35 to 7.92; p less than 0.01). This increased risk persisted after adjusting for the time difference between each cohort. CONCLUSIONS: Crude comparisons of hospital mortality can be highly misleading. Reliable assessment of neonatal outcome is impossible without correcting for major risk factors, particularly initial disease severity. International agreement on a minimum core dataset of clinical and physiological information could improve neonatal audit and help to identify effective treatments and policies.