Development and validation of a simple prediction model for length of hospital stay in a nationwide Japanese burn registry.

Abstract

Length of hospital stay (LOS) during burn care affects discharge planning and resource allocation. A simplified model (Taylor's model) was developed in the US; however, its applicability in other countries remains unclear. This study validated this model using a Japanese national registry and developed a simple prediction model applicable to Japan. This retrospective study, conducted using the Japan Society for Burn Injury Burn Registry 2011, included adult patients who survived hospitalization. Using a derivation cohort, we constructed a model based on generalized estimating equations (GEE) using age, % total body surface area burned (TBSA), inhalation injury, and their interactions. A simplified model was then established using GEE. Model performance was assessed in a validation cohort using the mean residual (observed-predicted), coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE). The performance of our Simplified Japanese Model [LOS = 0.3 × Age + 3 × %TBSA - 8 (with inhalation injury); or 0.3 × Age + 2 × %TBSA - 1 (without inhalation injury)] [R², 0.175 (95% CI, 0.124-0.230); RMSE, 49.0; MAE, 22.0; mean residual, -0.77] was comparable to that of the GEE model [R², 0.175 (95% CI, 0.125-0.229); RMSE, 49.0; MAE, 22.3; mean residual, -1.4] and superior to that of the Taylor model [R², 0.012 (95% CI, -0.025-0.047); RMSE, 53.6; MAE, 23.6; mean residual, 20.0]. We established a Simplified Japanese Model to predict LOS that outperformed the US-based model based on the Japanese context. The model supports early resource planning and benchmarks for burn care patients.