Inclusion of coexisting morbidity in a TBSA% and age based model for the prediction of mortality after burns does not increase its predictive power.

INTRODUCTION Several models for predicting mortality have been developed for patients with burns, and the most commonly used are based on age and total body surface area (TBSA%). They often show good predictive precision as depicted by high values for area under the receiver operating characteristic curves (AUC). However the effect of coexisting morbidity on such prediction models has not to our knowledge been thoroughly examined. We hypothesised that adding it to a previously published model (based on age, TBSA%, full thickness burns, gender, and need for mechanical ventilation) would further improve its predictive power. METHODS We studied 772 patients admitted during the period 1997-2008 to the Linköping University Hospital, National Burn Centre with any type of burns. We defined coexisting morbidity as any of the medical conditions listed in the Charlson list, as well as psychiatric disorders or drug or alcohol misuse. We added coexisting medical conditions to the model for predicting mortality (age, TBSA%, and need for mechanical ventilation) to determine whether it improved the model as assessed by changes in deviances between the models. RESULTS Mean (SD) age and TBSA% was 35 (26) years and 13 (17) %, respectively. Among 725 patients who survived, 105 (14%) had one or more coexisting condition, compared with 28 (60%) among those 47 who died. The presence of coexisting conditions increased with age (p<0.001) among patients with burns. The AUC of the mortality prediction model in this study, based on the variables age, TBSA%, and need for mechanical ventilation was 0.980 (n=772); after inclusion of coexisting morbidity in the model, the AUC improved only marginally, to 0.986. The model was not significantly better either. CONCLUSION Adding coexisting morbidity to a model for prediction of mortality after a burn based on age, TBSA%, and the need for mechanical ventilation did not significantly improve its predictive value. This is probably because coexisting morbidity is automatically adjusted for by age in the original model.