Establishment and validation of a 28-day mortality prediction model based on the lactate dehydrogenase/albumin ratio in patients with severe pneumonia.

Background: Severe pneumonia (SP) is a common and often fatal disease. Traditional mortality risk assessments rely on complex scoring systems and lack simple, effective biomarkers. This study aims to explore the potential value of the lactate dehydrogenase to albumin ratio (LAR) in predicting 28-day mortality in patients with severe pneumonia and to develop a predictive model using machine learning techniques. Methods: This retrospective cohort study included patients with severe pneumonia admitted to the Second Xiangya Hospital of Central South University, from January 2020 to May 2025. Clinical data, laboratory indicators, and LAR values were collected. Cox regression analysis, the Boruta feature selection algorithm, and various machine learning models were employed for analysis. The primary outcome was 28-day survival status, and the relationship between LAR and mortality risk was evaluated, leading to the development of a prediction model based on LAR. Results: Among the 491 patients, LAR was significantly associated with 28-day mortality risk and was identified as an independent risk factor for death in severe pneumonia. LAR demonstrated a high area under the curve (AUC) in predicting mortality, exhibiting a significant nonlinear relationship with 28-day mortality risk. Among several machine learning models constructed using key variables selected by the Boruta algorithm, the random forest (RF) model exhibited the best predictive performance. Furthermore, Shapley additive explanations (SHAP) value analysis confirmed the dominant role of LAR in the RF model. Conclusion: LAR is an effective biomarker with significant clinical value in predicting 28-day mortality in patients with severe pneumonia. The LAR-based prediction model enhances the accuracy of mortality risk assessment, especially in non-septic patients. Combined with machine learning techniques, LAR offers a novel tool for early clinical risk evaluation and holds promising potential for clinical application.