Predictors of hospital mortality in a population-based cohort of patients with acute lung injury

Abstract

Objective: Studies describing predictors of mortality in patients with acute lung injury were primarily derived from selected academic centers. We sought to determine the predictors of mortality in a population-based cohort of patients with acute lung injury and to characterize the performance of current severity of illness scores in this population.

Design: Secondary analysis of a prospective, multicenter, population-based cohort.

Setting: Twenty-one hospitals in Washington State.

Patients: The cohort included 1,113 patients with acute lung injury identified during the year 1999–2000.

Interventions: None.

Measurements and Main Results: We evaluated physiology, comorbidities, risk factors for acute lung injury, and other variables for their association with death at hospital discharge. Bivariate predictors of death were entered into a multiple logistic regression model. We compared Acute Physiology and Chronic Health Evaluation (APACHE) II, APACHE III, and Simplified Acute Physiology Score II to the multivariable model using area under the receiver operating characteristic curve. The model was validated in an independent cohort of 886 patients with acute lung injury. Modified acute physiology score, age, comorbidities, arterial pH, minute ventilation, Paco2, Pao2/Fio2 ratio, intensive care unit admission source, and intensive care unit days before onset of acute lung injury were independently predictive of in-hospital death (p < .05). The area under the receiver operating characteristic curve for the multivariable model was superior to that of APACHE III (.81 vs. .77, p < .001) but was no different after external validation (.71 vs. .70, p = .64).

Conclusions: The predictors of mortality in patients with acute lung injury are similar to those predictive of mortality in the general intensive care unit population, indicating disease heterogeneity within this cohort. Accordingly, APACHE III predicts mortality in acute lung injury as well as a model using variables selected specifically for patients with acute lung injury.