Admission Hyperoxia Is a Risk Factor for Mortality in Pediatric Intensive Care.

ABSTRACT

OBJECTIVES: To determine whether the association between hyperoxia and increased risk-adjusted mortality in adult intensive care patients is also observed in a pediatric intensive care population. DESIGN: Single-center retrospective analysis of admissions to ICU over a 5-year period commencing January 1, 2012, examining the relationship between PaO2 measured within the first hour of admission and risk-adjusted mortality. Standardized mortality rates were calculated using the Pediatric Index of Mortality-3, and patients were grouped into 50 mm Hg (6.67 kPa) PaO2 bands to assess the relationship between initial PaO2 and risk-adjusted mortality. SETTING: Tertiary PICU with 17 beds and 1,100 annual admissions located in metropolitan Sydney, Australia. PATIENTS A total of 1,447 patients 0-18 years old with PaO2 recorded at admission to the ICU. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: There were 5,176 patients admitted to the ICU during the study period and 1,447 (28%) with PaO2 recorded at admission. A U-shaped relationship between raw mortality and admission PaO2 was observed, with lowest mortality (2.3% and 2.6%, respectively) observed in the 101-150 (13.5-20.0 kPa) and 151-200 mm Hg (20.1-26.7 kPa) bands and the highest mortality observed in patients with PaO2 less than 50 mm Hg (6.67 kPa) with mortality of 5.3%, or greater than 350 mm Hg (46.7 kPa) with mortality of 18.2%. Hyperoxia at admission was associated with an increase in risk-adjusted mortality, with polynomial regression indicating a strong correlation between PaO2 band and risk-adjusted outcome (r = 0.845). When included in a multivariate model that included the Pediatric Index of Mortality-3 variables, the odds ratio for hyperoxia (defined as PaO2 > 250 mm Hg [33.3 kPa]) predicting death was 2.66 (p = 0.047). CONCLUSIONS: In this single-center study, hyperoxia at admission to the PICU was highly correlated with increased risk-adjusted mortality. Further investigation of these observations in a large multicenter cohort is warranted.