351 Predicting the Ability of Wounds to Heal Given Any Burn Size and Fluid Resuscitation Volume: An Analytical Approach

• Published in: Journal of burn care & research 2018-04, Vol.39 (suppl_1), p.S146-S146
• Main Authors: Liu, N T, Rizzo, J A, Shields, B A, Serio-Melvin, M L, Christy, R J, Salinas, J
• Format: Article
• Language: English
• ISSN: 1559-047X; 1559-0488
• DOI: 10.1093/jbcr/iry006.273

Abstract Introduction Various factors affect healing of severe burn injury. The wound burden can be numerically measured by the open wound size (OWS), often expressed as a percentage. The intrinsic relationship between fluid resuscitation and OWS has not been previously examined. Therefore, we conducted this study to investigate whether 1) OWS can be predicted from burn resuscitation volume plus other significant factors and whether 2) machine learning (ML) may perform better in predicting OWS than traditional statistics. Methods This retrospective study involved data from adults admitted to our burn center from April 2011 through March 2015, with at least 20% total body surface area (TBSA) burned. Each patient had initial and final burn mappings. OWS was defined as the sum of the TBSA burned, plus the surface area used as donor sites, minus the surface area healed. Multivariate analysis was used to identify independent predictors of OWS. Various predictive models were then developed, analyzed, and compared using goodness-of-fit statistics (R2, mean absolute error [MAE], root mean squared error [RMSE]). Bland-Altman analysis was also performed to determine bias. Results A total of 121 patients were included in the analysis. Mean age and weight were 43 ± 17 years and 86 ± 22 kg, respectively. Mean TBSA burned was 37 ± 17%, with an average of 5 ± 6 days elapsed until first excision and grafting. Mean final OWS was 11 ± 19%. Average crystalloid resuscitation volumes were 4.0 ± 2.7 mL/kg/TBSA in the first 24 hrs, and 83.1 ± 48.0 L during the hospital stay. There were 24 (20%) patients who died, with 22 (92%) of these not healing their wounds at the time of death. Importantly, multivariate analysis identified seven independent predictors of OWS. Also, ML analysis was able to stratify patients based on the 20th day after admission, ~40% TBSA burn, and fluid resuscitation volumes. Four- and seven-variable models for predicting OWS varied in performance (R2=0.79–0.90, MAE=3.97–7.52, RMSE=7.11–10.69). Notably, a combined ML model using only four features - days since admission, fluid resuscitation volume, TBSA burned, age - performed the best and was sufficient to predict OWS, with >90% goodness of fit and