Citation

Hussain A, Choukairi F, Dunn K. Burns 2013; 39(5): 835-850.

Affiliation

International Burn Injury Database Research Fellow, University Hospital South Manchester, Southmoor Road, Manchester M23 9LT, United Kingdom. Electronic address: amh@doctors.net.uk.

Copyright

(Copyright © 2013, Elsevier Publishing)

DOI

10.1016/j.burns.2012.12.010

PMID

23384617

Abstract

BACKGROUND: The widespread use of mathematical models to predict mortality as an outcome in burn injury is limited by concerns regarding the accuracy of the predictions. This discrepancy in reported and actual model accuracy can be the result of lack of adherence to appropriate methodological standards for the construction of prediction models. AIM: We undertook a systematic review of the methodology of published mortality prediction models against methodological standards. The aim was to identify methodologically superior models for further evaluation and research into outcome prediction. METHODS: Electronic searches were performed on MEDLINE, CINHAL, EMBASE, Web of Science(®), the Cochrane collection and a general web search was performed using Google(®). The searches were complemented by a manual search of the contents of leading burns journals. Methodology of the studies included in the review was evaluated against published standards for composite prediction models. RESULTS: 45 studies reporting composite models specifically for predicting mortality in patients sustaining thermal injury between 1949 and 2010 were included in the review. Only 8 models fulfilled the published methodological standards for composite model construction and validation. These include Modified Baux Score, Abbreviated Burn Severity Index, Total Burn Surface Index and prediction models described by Coste et al., Ryan et al., McGwin et al., Galeiras et al. and Belgian Outcome of Burn Injury (BOBI) study group. CONCLUSION: This review has demonstrated that although a variety of complex models for predicting mortality in thermal injury have been devised, only 8 models have been constructed using appropriate methodological standards. These models warrant further evaluation in independent patient populations and data sets to identify the ones best suited for outcome prediction and performance monitoring.

Language: en