Development of a LightGBM-based survival prediction model for major burn patients using LDH/LYM ratio as the core predictor.

BACKGROUND: Patients with extensive burns face a high mortality risk. Early identification of prognostic indicators may facilitate timely interventions that substantially improve outcomes and reduce mortality. The lactate dehydrogenase-to-lymphocyte ratio (LDH/LYM), a composite biomarker reflecting systemic inflammation severity, has not been previously evaluated for its predictive utility in burn populations. This retrospective study aimed to validate the prognostic value of LDH/LYM in patients with extensive burns and establish a predictive model to guide clinical decision-making. METHODS: A retrospective analysis was conducted on 365 patients with extensive burns admitted to the Burn ICU of Changhai Hospital during 2023. Independent prognostic factors were identified via univariate and multivariate analyses. The predictive performance of these factors was evaluated using receiver operating characteristic (ROC) curve analysis and Kaplan-Meier survival analysis. Additionally, a Light Gradient Boosting Machine (LightGBM) algorithm was applied to identify key predictors and validate the findings. RESULTS: Multivariate analysis identified age, total body surface area (TBSA) burned, and the LDH/LYM ratio as independent prognostic factors. The area under the curve (AUC) for TBSA and LDH/LYM alone was 0.814 and 0.785, respectively, while their combination achieved a superior AUC of 0.88. Stratification using the optimal LDH/LYM cutoff of 186.361 revealed determined by the Youden index significantly different survival curves between high- and low-ratio groups (HR = 0.074, p < 0.001). The LightGBM model ranked TBSA, blood urea nitrogen (BUN), and LDH/LYM as the top three mortality predictors, with a combined prediction accuracy of 0.849 and an AUC of 0.916. CONCLUSION: This study established that the LDH/LYM ratio is an independent prognostic determinant for patients with extensive burns, with predictive significance comparable to the TBSA and age, and its prognostic value was confirmed through LightGBM-based feature importance ranking.