Predictive value of red blood cell distribution width–albumin ratio for mortality risks in critically ill patients with valvular heart disease: A retrospective analysis using machine learning-based predictive modelling

Introduction Valvular heart disease (VHD) is a prevalent cardiovascular disorder, wherein malnutrition and chronic inflammation exert substantial adverse impacts on patient prognosis. The red blood cell distribution width–albumin ratio (RAR) emerges as a composite biomarker that comprehensively integrates metabolic and nutritional status, offering a holistic reflection of these interrelated physiological dimensions. Methods The data from the MIMIC-IV database, including 2,358 critically ill VHD patients with complete serum albumin and RDW measurements. Cox proportional hazards regression, restricted cubic spline (RCS) analysis, and threshold effect analysis were performed to assess the association between RAR and mortality. Feature selection was via Lasso regression and the Boruta algorithm, followed by construction of 8 machine learning models; the optimal model was validated with AUC, calibration curves, and Shapley additive explanations (SHAP) analysis for interpretability. Results Kaplan-Meier analysis showed higher RAR quartiles were significantly associated with poorer survival across all follow-up time points. After full adjustment, each 1-unit RAR increase independently elevated mortality risk. RCS analysis confirmed a non-linear association. Subgroup analysis showed consistent prognostic value of RAR except in patients with stroke. Feature selection identified RAR, BUN, anion gap, and disease severity scores as key predictors. The logistic regression model exhibited optimal performance, with SAPS II score, anion gap, and RAR as top predictive features via SHAP analysis. Conclusion RAR is an independent prognostic biomarker for short-term and long-term all-cause mortality in critically ill VHD patients. The LR-based model incorporating RAR provides reliable risk stratification, supporting personalized management and early intervention in this high-risk population.