Variations in definitions of mortality have little influence on neonatal intensive care unit performance ratings

Abstract

Objective: To measure the influence of varying mortality time frames on performance rankings among regional neonatal intensive care units (NICUs) in a large state.

Study design: We performed a cross-sectional data analysis of very low birth weight infants receiving care at 24 level 3 NICUs. We tested the effect of 4 definitions of mortality: (1) death between admission and end of birth hospitalization or up to 366 days; (2) death between 12 hours of age and the end of birth hospitalization or up to 366 days; (3) death between admission and 28 days; and (4) death between 12 hours of age and 28 days. NICUs were ranked by quantifying their deviation from risk-adjusted expected mortality and dividing them into 3 tiers: top 6, bottom 6, and in between.

Results: There was wide interinstitutional variation in risk-adjusted mortality for each definition (observed minus expected z-score range, -6.08 to 3.75). However, mortality-based NICU rankings and classification into performance tiers were very similar for all institutions in each of our time frames. Among all 4 definitions, NICU rank correlations were high (>0.91). Few NICUs changed relative to a neighboring tier with changes in definitions, and none changed by more than one tier.

Conclusion: The time frame used to ascertain mortality had little effect on comparative NICU performance.

Figure 1

Patient sample and exclusions.

Figure 2

Four competing definitions of mortality. NICU - neonatal intensive care unit, dc - discharge, d - days, h - hours.

Figure 3

Risk-adjusted observed minus expected mortality z-scores, by hospital, across the four definitions of mortality. The bars do not represent confidence intervals but show the mean and range of z-scores. A lower z-score indicates lower-than-expected mortality. The dotted lines represent two standard deviations below or above expected performance.