Objective: Prediction of subsequent mortality among very low birth weight infants (<1500 grams) on days of life 3 and 14 using the Score for Neonatal Acute Physiology (SNAP) and traditional risk factors.

Methods: We prospectively abstracted clinical and demographic data on a cohort of 1670 infants (<1500 grams) at seven regional NICUs from October 1994 to July 1996 and identified all NICU deaths. We measured severity of illness using the Score for Neonatal Acute Physiology (SNAP) at days 3 and 14. The risk of subsequent mortality at days 3 and 14 was determined using sequential logistic models of the traditional risk factors - male sex, white race, SGA status, birth weight and low 5-minute Apgar - as well as SNAP at days 1, 3 and 14 (SNAP1, SNAP3, SNAP14). We constructed Receiver-operator curves (ROC) for the competing models and compared the differences in area under the ROC-curves.

Results: There were 198 deaths in the cohort. Ninety-three occurred after day of life 3 and 43 after day of life 14. On both days only birth weight and SNAP improved mortality prediction. Sequential addition of increasingly proximate SNAP improved the predictive power of the models at both days 3 and 14. ROC-areas at day 3 were significantly less for the traditional model (.82 ±.02) than for the traditional model plus SNAP1 and SNAP3 (.88 ±.02) with p-value <.02. ROC-areas at day 14 were.80±.02 for the traditional model and.84 ±.03 for the traditional model plus SNAP1, SNAP3 and SNAP14 (p-value=NS).

Conclusion: SNAP does improve the prediction of subsequent mortality at day 3 over traditional risk factors. Serial measurement of severity of illness yields additional information about evolving mortality risk among infants <1500grams