Associations between age, red cell distribution width and 180-day and 1-year mortality in giant cell arteritis patients: mediation analyses and machine learning in a cohort study

The aim of this study was to investigate the correlation between age, red cell distribution width (RDW) levels, and 180-day and 1-year mortality in giant cell arteritis (GCA) patients hospitalized or admitted to the ICU. Clinical data from GCA patients were extracted from the MIMIC-IV (3.0) database. Logistic and Cox regression analyses, Kaplan–Meier (KM) survival analysis, restricted cubic spline (RCS) analysis, and mediation effect analysis were employed to investigate the association between age, RDW levels, and 180-day and 1-year mortality in GCA patients hospitalized or admitted to the ICU. Predictive models were constructed using machine learning algorithms, and SHapley Additive exPlanations (SHAP) analysis was applied to evaluate the contributions of age and RDW levels to mortality in this patient population. A total of 228 GCA patients were eligible for analysis. Our study identified both age and RDW levels (both with OR > 1, P < 0.05) as significant predictors of 180-day and 1-year mortality in GCA patients hospitalized or admitted to the ICU using multivariate logistic regression analysis. In multivariate Cox regression analysis, both age and RDW (both with HR > 1, P < 0.05) also emerged as prognostic risk factors for 180-day and 1-year mortality in this patient population. KM survival analysis further showed that GCA patients hospitalized or admitted to the ICU with higher age or elevated RDW levels had significantly lower survival rates compared to younger patients or those with lower RDW levels (P < 0.0001). Moreover, RCS analysis indicated a strong nonlinear relationship between RDW levels (threshold: 17.53%) and 1-year mortality in this population. Additionally, RDW levels were found to modestly mediate the relationship between age (per 10-year increase) and 180-day or 1-year mortality in GCA patients hospitalized or admitted to the ICU. The results of the machine learning analysis indicated that the model built using the random forest algorithm performed the best, with an area under the curve of 0.879. Furthermore, SHAP analysis revealed that both age and RDW levels made significant contributions to the prediction of mortality in GCA patients hospitalized or admitted to the ICU. Older age and higher RDW levels were identified as independent risk factors for increased 180-day and 1-year mortality in GCA patients hospitalized or admitted to the ICU. Furthermore, elevated RDW levels modestly mediated the relationship between age and 180-day or 1-year mortality in this patient population.