The influence of dynamic trajectories of red blood cell distribution width on prognosis and causal mediation analysis in sepsis patients: a longitudinal study with external validation

Background: Sepsis is a highly heterogeneous syndrome, and static clinical data do not provide a comprehensive characterization of the full spectrum of patient conditions. This study investigated sepsis subtypes based on the dynamic trajectory of red cell distribution width (RDW) aiming at better risk stratification of patients. Methods: This retrospective study included a derived cohort of sepsis patients from the Medical Informatics Marketplace Intensive Care Database (MIMIC IV) and an external validation cohort of sepsis patients from the First Affiliated Hospital of AnHui Medical University department of Critical Care medicine Data Mart (AHCCDM) between January 2019 and June 2025. Latent Class Trajectory Modeling (LCTM) was employed to identify distinct RDW trajectories over the first 5 days following ICU admission. We compared the clinical characteristics and outcomes of patients in these subphenotypes to evaluate the prognostic significance of RDW trajectory patterns in sepsis. Causal moderation analyses were applied to determine whether changes in RDW trajectories mediated survival outcomes in patients with sepsis through other variables. Results: A total of 20,485 sepsis patients from the MIMIC-IV database were included in the study as the derived cohort, and 627 sepsis patients from the AHCCDM dataset served as the external validation cohort. Based on the RDW trajectory analysis, three different subtypes were identified, Class 1: stable, normal RDW; Class 2: rapidly increasing and severely abnormal RDW; Class 3: stable, elevated RDW. Among them, the clinical characteristics of patients in category 2 included higher severity, more severe inflammatory response, and more severe organ damage. Using Kaplan-Meier survival curve analysis to compare the incidence of the primary outcome between groups, patients with class 2 sepsis had a significantly higher risk of death at 30, 60, and 90 days than class 3 and 1. External validation revealed that the RDW trajectories were also classified into three distinct groups, with patients in all three groups exhibiting similar mortality trends.Mediated causality analyses indicated that the dynamic trajectory of RDW mediated the risk of death in sepsis patients through an increase in RDW. Conclusion: Dynamic RDW trajectories identify distinct sepsis subphenotypes associated with disease severity and mortality. Serial RDW monitoring may serve as a simple tool for early risk stratification and individualized sepsis management.