Background: SNAP is used for describing populations, stratifying risk in epidemiology and clinical trials and projecting resource use. We sought to simplify and revalidate it empirically.

Methods: The derivation cohort consisted 12-hour SNAPs on 3432 first admissions to 7 NICUs in Massachusetts and Rhode Island(10/94-6/96). All NICUs collected <31 week cases; 3 NICUs also collected cases ≥31 weeks (10/94-10/95). Outcomes were tracked until discharge home, including transfers. Moribund cases and delivery-room deaths were excluded. The 34 SNAP items had 84 scorable levels. Those items with no univariate association with death were eliminated. Significant predictors of hospital death were tested by organ system and then included in the final logistic model, along with birth weight (BW), gestational age (GA) and Apgar score<7 at 5 min. Separate and combined models were derived and reconciled to a single SNAP II score (7 physiologic variables, plus BW, GA and Apgar). Wevalidated SNAP II on 4864 admissions of all BW to 6 Kaiser NICUs in northern California (7/95-6/97), and evaluated performance using Receiver Operator Characteristic (ROC) curve analysis and Hosmer-Lemeshow goodness-of-fit testing.

Results: See table.

Table 1 No caption available.
Full size table

Conclusions: SNAP II is a 10 item neonatal mortality risk score with excellent discrimination and calibration shown in two large diverse multi-NICU cohorts. It is backward compatible with SNAP I, but takes only 5 minutes per case to collect.