Citation

McGwin G, George RL, Cross JM, Rue LW. Burns 2007; 34(3): 320-327.

Affiliation

Section of Trauma, Burns, and Surgical Critical Care, Division of General Surgery, Department of Surgery, School of Medicine, University of Alabama at Birmingham, Birmingham, AL, United States.

Copyright

(Copyright © 2007, Elsevier Publishing)

DOI

10.1016/j.burns.2007.06.003

PMID

17869427

Abstract

BACKGROUND: Early efforts to predict death following severe burns focused on age and burn size; more recent work incorporated inhalation injury and pneumonia. Gender, co-morbid illness, and co-existent trauma have been implicated in burn mortality but have rarely been incorporated into predictive models. METHODS: The National Burn Repository (NBR) and the National Trauma Data Bank (NTDB) provided data on 68,661 (54,219 and 14,442, respectively) burn patients that was used to develop and validate, respectively, a predictive model of burn mortality. Logistic regression was used to model the odds of mortality with respect to age, gender, % body surface area burned (BSAB), co-existent trauma, inhalation injury, pneumonia, and co-morbid illness. Performance of the predictive model was assessed using a deviance statistic, receiver operating characteristic (ROC) curves, and the Hosmer-Lemeshow (HL) statistic. RESULTS: The predictive model that demonstrated optimal performance included the variables age, percent total BSAB, inhalation injury, co-existent trauma, and pneumonia. The area under the ROC curve for this model was 0.94 and the HL statistic was 16.0. The inclusion of additional variables, i.e., gender, co-morbid illness, did not improve the performance of the model despite reduction in the model deviance. When the predictive model was applied to the validation data source, the area under the ROC curve was 0.87 and the HL statistic was 10.0, indicating good discrimination and calibration. CONCLUSION: The results of this study suggest that a comprehensive predictive model of burn mortality incorporating certain variables not previously considered in other models provides superior predictive ability.

Language: en