The study population was obtained from the MIMIC-IV database. This database is an electronic health record repository that includes extensive, high-quality, deidentified data on intensive care unit (ICU) patients from the Beth Israel Deaconess Medical Center (BIDMC) in Boston, MA, USA, covering the years between 2008 and 2019. Access to this dataset was granted under certification number 58844105. The Institutional Review Board at BIDMC approved a waiver of informed consent and permitted the use of this research resource. This cohort study followed the guidelines established by the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (19).

Our study encompassed all eligible ICU patients from the MIMIC-IV database. We specifically targeted adults diagnosed with sepsis-3, characterized by an increase of at least 2 points in the Sequential Organ Failure Assessment (SOFA) score as a result of a dysregulated host response to infection (1). According to the Surviving Sepsis Campaign guidelines, the criteria for diagnosing sepsis-associated liver injury (SALI) include a total bilirubin level exceeding 2 mg/dL and an INR greater than 1.5 in patients with sepsis (10). These criteria are widely recognized in the literature. The study was restricted to patients with RDW measurements obtained within the first 24 h of ICU admission. The exclusion criteria included the following: (1) individuals younger than 18 years; (2) multiple hospital admissions, excluding their initial ICU admission; (3) absence of RDW data within 24 h of ICU admission or missing information on other variables.

Data pertinent to this study were obtained from the MIMIC-IV database through Structured Query Language (SQL) queries and were subsequently stored in PostgreSQL. At the time of sepsis diagnosis, initial records were systematically collected, including demographic details such as age, sex, and race; vital signs like heart rate and mean blood pressure (MBP); and laboratory findings that encompassed RDW, white blood cell count (WBC), platelet count, hemoglobin, hematocrit, glucose, blood urea nitrogen (BUN), creatinine, and alanine aminotransferase (ALT). Additionally, comorbidities were noted, including congestive heart failure, chronic respiratory disease, diabetes, renal disease, and malignant cancer, along with disease severity metrics such as SOFA score, Charlson comorbidity index, and simplified acute physiology score (SAPS II). The SOFA score evaluates organ dysfunction across six systems, each scored from 0 to 4, with higher scores indicating worse dysfunction (1) (Supplementary Table 1). The SAPS II score is calculated within 24 h of ICU admission, using 12 physiological variables (e.g., heart rate, blood pressure, oxygenation, serum potassium) and patient factors (age, chronic conditions) (20) (Supplementary Table 2). Each variable is assigned a predefined score, with higher scores indicating a higher mortality risk. The Charlson comorbidity index assesses 17 conditions, including cardiovascular disease, diabetes, chronic kidney disease, liver disease, and cancer (21) (Supplementary Table 3). Higher scores indicate a greater comorbidity burden and higher mortality risk. Interventions, including antibiotic therapy and vasoactive agents administered on day one, were also noted. Comorbidity information was obtained using the International Classification of Diseases (ICD) coding systems.

Since none of the patients had an RDW below the normal range of 10.5 to 15.5%, they were classified into two groups: the normal RDW group (nRDW), comprising individuals with RDW ≤ 15.5%, and the high RDW group (hRDW), which included those with RDW > 15.5%. In addition, RDW was categorized into quartiles (Q1-Q4) for sensitivity analysis.

The primary outcome was 30-day mortality. The secondary outcomes included in-hospital mortality, 90-day mortality, ventilator-free days until 28 days, vasopressor-free days until 28 days, and ICU-free days until 28 days.

Due to the retrospective design of this study, no predefined statistical analysis plan or power calculation was implemented; the sample size relied on available database data. Categorical variables are presented as frequencies and percentages, whereas continuous variables are expressed as means with standard deviations (SD) or medians with interquartile ranges (IQR). Continuous variables were assessed based on their distribution, utilizing either the Student’s t-test or the Wilcoxon rank-sum test. Categorical variables were evaluated using either Pearson’s chi-squared test or Fisher’s exact test.

Variable selection was guided by clinical relevance and existing literature. Variables with p-values < 0.1 in univariate analysis were included in the multivariate analysis (Supplementary Table 4). The effect of RDW on 30-day mortality was assessed using multivariable Cox regression models, yielding hazard ratios (HR) and 95% confidence intervals (CI) after adjusting for relevant covariates. Subgroup analyses were performed to explore interactions with covariates such as age, race, and comorbidities. Kaplan–Meier survival curves evaluated 30-day mortality across RDW categories, with significance determined by the log-rank test. Receiver operating characteristic (ROC) curves were generated to assess RDW’s predictive performance for 30-day mortality, calculated as the area under the curve (AUC). In addition, sensitivity analyses were performed by converting the types of variables and excluding special populations.

All analyses utilized R Statistical Software (Version 4.2.2, The R Foundation)¹ and Free Statistics analysis platform (Version 2.0, Beijing, China). A p-value < 0.05 in a two-sided test was deemed statistically significant.

This analysis included a total of 529 patients diagnosed with SALI. The patient selection and data screening processes are illustrated in Figure 1. The baseline characteristics of patients, stratified by survival status, are presented in Table 1. The overall mean age at ICU admission was 68.7 years, and 61.8% of the patients were male. Patients in the non-survival group demonstrated higher scores on the Charlson comorbidity index, SOFA score, and SAPS II. Specifically, the RDW was significantly higher in the non-survival group compared to the survival group (17.2 vs. 15.4%, P < 0.001). Furthermore, the proportion of patients receiving vasopressin in the non-survival group was significantly greater than that in the survival group (75.5 vs. 54%, P < 0.001).

The overall prevalence of 30-day mortality among the cohort was found to be 35.5%. When RDW was considered as a continuous variable, the univariable Cox regression analysis revealed a significant association with 30-day mortality (crude HR 1.17; 95% CI, 1.13 to 1.23; P < 0.001). Following multivariable adjustment for factors including age, race, heart rate, MBP, creatinine, BUN, ALT, Charlson comorbidity index, SOFA, SAPS II, malignant cancer, and vasoactive agent, this association remained significant (adjusted HR 1.14; 95% CI, 1.09 to 1.19; P < 0.001) (Table 2). To further investigate this relationship, we utilized a restricted cubic spline analysis. The result indicated that, after accounting for the previously mentioned covariates, the relationship between RDW and 30-day mortality was linear (P for non-linearity = 0.340) (Supplementary Figure 1).

When RDW was treated as a categorical variable with a cutoff of 15.5%, the hRDW group exhibited a heightened risk of 30-day mortality compared with the nRDW group (adjusted HR 1.62; 95% CI, 1.17 to 2.23; P = 0.004) (Table 2). Kaplan–Meier survival curve estimates further indicated a significantly elevated risk of 30-day mortality in patients with high RDW (log-rank: P < 0.001; Figure 2). The association persisted when RDW was categorized into quartiles. Compared to individuals in Q1, those in Q4 faced a 2.71-fold increased risk of 30-day mortality (adjusted HR, 3.71; 95% CI, 2.19 to 6.28; P < 0.001; Table 2, Model 3).

Subgroup analyses were conducted to assess the influence of age, sex, race, congestive heart failure, diabetes, chronic pulmonary disease, renal disease, and malignant cancer on the relationship between RDW and 30-day mortality. No statistically significant interactions were identified within any subgroup (p > 0.05, Figure 3). These results indicate that the effect of RDW on 30-day mortality remains consistent across different age, race, sex, and comorbidity groups.

ROC analysis revealed that RDW possesses moderate clinical predictive value for 30-day mortality in patients with SALI, achieving an AUC of 0.70 (95% CI, 0.66 to 0.75). We subsequently compared the discriminatory ability for 30-day in-hospital mortality among RDW, SAPS II, and SOFA using ROC curves (Figure 4). RDW exhibited stronger predictive discrimination for 30-day mortality in SALI patients compared to SOFA (AUC, 0.57; 95% CI, 0.52 to 0.62; P < 0.001), but was less effective than SAPS II (AUC, 0.80; 95% CI, 0.76 to 0.84; P = 0.002) (Supplementary Figures 2–4 and Supplementary Table 5).

Increased RDW levels were associated with a higher risk of in-hospital mortality (adjusted HR, 1.13; 95% CI, 1.07 to 1.18; P < 0.001). Similar patterns were observed for 90-day mortality. Each 1% rise in RDW was linked to a reduction of 0.81 days in ICU-free days (95% CI, −1.12 to −0.51), 0.96 days in ventilator-free days (95% CI, −1.30 to −0.62), and 0.98 days in vasopressor-free days (95% CI, −1.32 to −0.65) over a 28-day period (Table 3).

We converted non-normally distributed continuous variables, including BUN, creatinine, ALT, SOFA score, and Charlson comorbidity index, into categorical variables for sensitivity analysis. BUN, creatinine, and ALT were classified according to clinically relevant cut-offs. Normal values range from 6 to 20 mg/dL for BUN and 0.5 to 1.2 mg/dL for creatinine, and 0–40 IU/L for ALT. The SOFA score and Charlson comorbidity index were classified according to the median, which were 4 and 6, respectively. The multivariable Cox regression analysis showed that higher RDW was significantly associated with an increased risk of 30-day mortality (Supplementary Table 6).

In addition, we performed a multivariate COX regression analysis after excluding patients with liver malignancies, anemia, and myelodysplastic syndromes, and the results remained robust (Supplementary Tables 7, 8).