Predicting death in pediatric patients with acute respiratory failure. Pediatric Critical Care Study Group. Extracorporeal Life Support Organization

Abstract

Objective: To estimate mortality risk in pediatric patients with acute hypoxemic respiratory failure (AHRF).

Design: Retrospective chart review.

Setting: Forty-one pediatric ICUs.

Subjects: Four hundred seventy children with AHRF. We defined AHRF as mechanical ventilation with positive end-expiratory pressure > or = 6 cm H2O and fraction of inspired oxygen greater than or equal to 0.5 for 12 or more hours.

Measurements: Physiologic and treatment variables were recorded every 12 h for 14 days. Cases were randomly assigned to score development and score validation subsets. Variables were assessed for their association with mortality in the development subset by logistic regression analysis. The analysis generated a series of logistic equations, which we called the Pediatric Respiratory Failure (PeRF) score, to estimate mortality risk at 12-h intervals over the first 7 days of treatment for AHRF. The predictive ability of the score was assessed in the validation subset by receiver operating characteristic curve area and goodness-of-fit chi 2.

Results: Mortality of the collected cases was 43%. The PeRF score included age, operative status, Pediatric Risk of Mortality score, fraction of inspired oxygen, respiratory rate, peak inspiratory pressure, positive end-expiratory pressure, PaO2, and PaCO2. Area under the receiver operating characteristic curve was 0.769 at entry and increased to greater than 0.8 after 36 h. When the score was applied to the validation subset of patients, goodness-of-fit chi 2 showed no significant difference between estimated and actual mortality between 0 and 96 h.

Conclusions: The PeRF Score accurately estimated mortality risk in this retrospectively sampled group of high-risk pediatric patients with AHRF. This score may be useful in studies of newer therapies for pediatric AHRF, though prospective validation is necessary before it could be used to make clinical decisions.