Integrating Dynamic Red Blood Cell Distribution Width Monitoring and β-Blocker Therapy for Mortality Prediction in Intensive Care Unit Cardiomyopathy Patients: A Bayesian Multivariate Joint Model and Machine Learning Study.

Cardiomyopathy is a key cause of cardiovascular mortality in critically ill patients. Although red blood cell distribution width (RDW) is recognized as a potential prognostic biomarker, its variations during ICU admission and its interaction with treatments such as β-blockers are not well understood across different cardiomyopathy subtypes. To assess the prognostic significance of RDW dynamics and their interaction with β-blocker therapy in predicting 365-day mortality among ICU patients with dilated, hypertrophic, and restrictive cardiomyopathy, utilizing longitudinal data and advanced modeling techniques. A retrospective analysis was conducted on 317 cardiomyopathy patients from the MIMIC-IV database. Their RDW dynamics were assessed over their ICU stay. Cox regression (including time-dependent Cox models) and logistic regression identified independent mortality risk factors. Key predictors were identified using Least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm. Restricted cubic splines (RCSs) were used to examine nonlinear relationships. Machine learning models were used to evaluate predictive performance, with SHapley Additive Explanations (SHAP) and tree-based feature selection identifying influential variables. Repeated-measures ANOVA was used to analyze RDW trends and β-blocker associations. A Bayesian multivariate joint model (BMJM) integrated RDW dynamics and β-blocker therapy, incorporating repeated measures and survival outcomes. RDW was an independent predictor of 365-day mortality (HR = 1.14, 95% CI: 1.01-1.29, = 0.03), alongside the systemic immune-inflammation index (SII) (HR = 1.01, 95% CI: 1.00-1.01, = 0.03), whereas β-blockers significantly reduced mortality risk (HR = 0.2, 95% CI: 0.1-0.39, < 0.001). Comparative analysis demonstrated that RDW exhibited greater predictive value over the aggregate index of systemic inflammation (AISI), systemic inflammation response index (SIRI), and SII. Machine learning identified logistic classification as the best predictive model (AUC = 0.811), with SHAP and tree-based selection confirming RDW and β-blockers as key predictors. A repeated-measures ANOVA revealed a significant interaction between RDW and β-blocker use (F = 6.65, < 0.0001), with β-blockers lowering RDW levels. The BMJM demonstrated strong predictive performance (AUC = 0.80). The patient-specific BMJM indicated that discontinuing β-blockers increased the risk of mortality, while initiating β-blockers reduced it. This study highlights dynamic RDW monitoring and β-blocker therapy as strong predictors of 365-day mortality in ICU-admitted cardiomyopathy patients. The BMJM enables personalized risk assessment by integrating longitudinal biomarker data. These findings support RDW as a dynamic biomarker and advocate for its integration into personalized treatment strategies.