Prognostic value of red blood cell distribution width in traumatic brain injury: A mediation and deep learning analysis

Shuting Ding; Zhen Zhang; Qifu Bo; Chenyu Ma; Minghao Wu; Xue Di; Manli Zhao; Kai Luo; Jiani Pan; Xin Zhang; Bingqiang Zhang; Suzhen Wang; Yujia Kong

doi:10.1371/journal.pone.0339879

Abstract

This study aimed to investigate the associations between age, red blood cell distribution width (RDW), and short-term mortality in patients with traumatic brain injury (TBI), with a particular focus on the role of RDW in mediating the impact of age on mortality. We conducted a retrospective cohort analysis of 1,203 ICU-admitted TBI patients from the MIMIC-IV database (v3.1). Cox proportional hazards regression, restricted cubic splines (RCS), and mediation analysis were employed to evaluate the relationships between age, RDW, and mortality outcomes. Both advanced age (adjusted hazard ratio [HR] = 1.022 for 28-day mortality; HR_adj = 1.031 for in-hospital mortality) and RDW (HR_adj = 1.085 for 28-day mortality; HR_adj = 1.094 for in-hospital mortality) were found to predict mortality (all P < 0.05) independently. RDW demonstrated a dose–response relationship with mortality: the highest quartile (Q4) exhibited a 2.061-fold increased risk of 28-day mortality (P = 0.010) and a 2.086-fold increased risk of in-hospital mortality (P = 0.022) compared to the lowest quartile (Q1). RCS analysis revealed significant nonlinear associations between age and 28-day mortality (P < 0.05) and between RDW and in-hospital mortality (P < 0.05). The mediation analysis demonstrated that RDW played a partial mediating role in age-related mortality, accounting for 4.40% of the total effect on 28-day mortality and 4.62% on in-hospital mortality (both P < 0.05). Deep learning survival models (e.g., Deepsurv: C-index: 0.759; IBS: 0.113; AUC (95% CI): 0.824 (0.735–0.900)) that incorporate age, RDW, and other clinical variables achieved robust predictive performance. Age and RDW are independent predictors of short-term mortality in TBI. RDW partially mediates the effect of age on TBI prognosis and shows potential as a practical biomarker for clinical risk stratification.

Citation: Ding S, Zhang Z, Bo Q, Ma C, Wu M, Di X, et al. (2026) Prognostic value of red blood cell distribution width in traumatic brain injury: A mediation and deep learning analysis. PLoS One 21(1): e0339879. https://doi.org/10.1371/journal.pone.0339879

Editor: Aleksandra Klisic, University of Montenegro-Faculty of Medicine, MONTENEGRO

Received: August 23, 2025; Accepted: December 12, 2025; Published: January 2, 2026

Copyright: © 2026 Ding et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data used in this study were obtained from the MIMIC-IV database (v3.1). Due to restrictions in the Data Use Agreement (DUA), redistribution of the data in any form is prohibited. Researchers may access the original data by completing the required certification training and signing the DUA through the PhysioNet platform (https://physionet.org/content/mimiciv/).

Funding: This work was supported by National Key Research and Development Program (Grant No.2024YFA0918404 by BZ), the National Natural Science Foundation of China (Grant No. 82003560 by YK), Key Research and Development Projects of Shandong Province (Grant No.2024TSGC0151 by BZ), the National Natural Science Foundation of Shandong (Grant No. ZR2020MH340 by YK).

Competing interests: The authors have declared that no competing interests exist.

Introduction

Traumatic brain injury (TBI) constitutes an ongoing public health concern worldwide, contributing significantly to mortality and long-term disability, particularly among young adults and the elderly [1,2]. Globally, TBI incidence reached 20.8 million new cases in 2021, corresponding to age-standardized rates of 259 per 100,000 population [3]. TBI typically stems from external mechanical forces, which trigger complex secondary pathophysiological cascades, including intracranial hemorrhage, disruption of the blood–brain barrier, and systemic inflammatory responses [4–6]. Notably, the ongoing demographic shift toward an aging global population exacerbates TBI incidence and severity in older adults [7,8], who tend to exhibit poorer outcomes, likely due to age-related physiological decline, comorbidities, and heightened susceptibility to secondary injury mechanisms [9].

Various prognostic factors have been identified in TBI, including injury severity (e.g., Glasgow Coma Scale [GCS] scores), neuroimaging findings, systemic physiological parameters, and emerging biomarkers [10,11]. Among these, red cell distribution width (RDW), a routinely measured hematologic index reflecting erythrocyte size variability, has gained traction as a prognostic marker in acute and chronic conditions [12–14]. Several investigations have examined the association between RDW and TBI. Elevated RDW has been reported as an independent risk factor for mortality in TBI patients [15,16]. One study found that non-surviving TBI patients had higher RDW levels at admission than survivors [17], whereas another study reported no significant differences [18]. Taken together, these conflicting results suggest that the relationship between RDW and mortality in TBI requires further investigation.

Age is a well-established prognostic determinant in TBI, with mortality and functional decline escalating markedly in older patients [19,20]. Neuroimaging studies further reveal that aging worsens TBI-related brain atrophy, reduces cortical thickness, and increases microstructural damage [21]. Nevertheless, the interplay between age and RDW in determining TBI outcomes remains underexplored. Specifically, it is unclear whether RDW mediates age-related increases in TBI mortality, with few studies addressing this causal pathway.

Given the complex and multifactorial nature of TBI outcomes [22], traditional regression methods may be insufficient to fully capture the nonlinear and high-dimensional relationships underlying these associations. In this context, recent advances in machine learning and deep learning–based survival models have provided practical tools to address these limitations. These approaches have been shown to improve prognostic accuracy and identify complex feature interactions in clinical prediction studies [23–25]. Integrating such methods may help generate additional insights into TBI prognosis beyond those obtained from conventional statistical analyses.

To address these gaps, we leveraged the MIMIC-IV database to investigate the interrelationships among age, RDW, and in-hospital/28-day mortality in TBI patients. Using causal mediation analysis, we examined whether RDW significantly mediates the effect of age on clinical outcomes, thus providing novel insights into prognostic mechanisms and potential avenues for risk stratification.

Materials and methods

Data source

This study is a retrospective analysis of the open-access Medical Information Mart for Intensive Care-IV (MIMIC-IV) database, which includes anonymized clinical data from patients treated at the Beth Israel Deaconess Medical Center between 2008 and 2022. The database was developed with approval from the Institutional Review Boards of both the Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. Access was granted to the first author (Shuting Ding) after successful completion of mandatory online training and acceptance of the data use agreement (Certification Number: 62840007). The data used in this study were obtained from the publicly available MIMIC-IV database, which contains de-identified health-related data for critical care patients. Data extraction for the present analysis occurred from May 17 to June 20, 2025. As the dataset contained no identifiable patient information, further Institutional Review Board oversight was not required, and informed consent was waived. The research process strictly adhered to the Declaration of Helsinki and was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Study population

TBI patients were retrospectively identified from the MIMIC-IV database using ICD codes. The cohort included patients admitted to both hospital wards and the ICU. For individuals with multiple hospitalizations or ICU stays, only data from the first admission were analyzed. Participants were selected based on meeting all of the following criteria: (1) primary diagnosis of TBI at ICU admission, defied by ICD-9 codes (800–801, 803–804, 850–854, 950.1–950.3, 959.01) or ICD-10 codes (S02.0, S02.1, S02.8, S02.91, S04.02, S04.03, S04.04, S06, S07.1); (2) first ICU admission during the study period; and (3) age ≥ 18 years. Exclusion criteria: (1) missing critical data (e.g., RDW, mortality status); or (2) ICU length of stay < 24 hours (to avoid inclusion of rapidly discharged patients with incomplete data). The final cohort included 1203 patients with TBI, as shown in the patient selection process (Fig 1).

Download:

Fig 1. Flow diagram illustrating the selection of ICU patients diagnosed with TBI for inclusion in the analysis cohort.

ICU: intensive care unit; TBI: Traumatic Brain Injury.

https://doi.org/10.1371/journal.pone.0339879.g001

Data extraction

Data were extracted using Structured Query Language (SQL) with PostgreSQL (version 14.8.1) and Navicat Premium (version 15). Variables included: Demographics (Gender, Race, Age, Weight,) Comorbidities (Cerebrovascular Disease, Dementia, Diabetes, Renal Disease), Clinical scores (Logistic Organ Dysfunction System [LODS], Sequential Organ Failure Assessment [SOFA], GCS), Laboratory parameters (RDW, White blood cell count [WBC], Sodium, Potassium, Chloride, Creatinine, Hemoglobin, Calcium, Blood urea nitrogen [BUN], Platelets, Hematocrit, Glucose), Interventions (Invasive mechanical ventilation [IMV], Furosemide, Cefazolin, Ceftriaxone, Vancomycin).

Primary and secondary outcomes

Primary outcome: 28-day all-cause mortality. Secondary outcome: In-hospital all-cause mortality.

Statistical analysis

Continuous variables were first assessed for normality using the Shapiro-Wilk test. Normally distributed variables were summarized as mean ± standard deviation (SD), while skewed variables were reported as median and interquartile range (IQR). Categorical variables were presented as counts and percentages. Comparisons between groups were performed using t-tests or ANOVA for normally distributed data, Mann-Whitney U or Kruskal-Wallis tests for non-normally distributed data, and Chi-square tests for categorical variables.

Cox proportional hazards regression modeled associations between RDW (categorized into quantiles Q1–Q4) and mortality: Model 1 included unadjusted estimates; Model 2 was adjusted for age; and Model 3 was adjusted for age, LODS, invasive ventilation, and WBC. Multicollinearity was evaluated using the variance inflation factor (VIF), with variables exceeding a VIF of 5 being excluded from the analysis. The remaining variables were retained for inclusion in the multivariable analysis (S1 Table). Restricted cubic spline (RCS) analysis with four knots positioned at the 5th, 35th, 65th, and 95th percentiles was applied to examine nonlinear associations of RDW and age with mortality, and to derive the corresponding thresholds for stratification.

Mediation analysis was conducted to assess the potential indirect effect of RDW on the relationship between age and 28-day mortality. We applied a causal steps framework, with age as the independent variable, RDW as the mediator, and mortality as the outcome. To obtain robust estimates of the indirect effect (a × b) and its uncertainty, we implemented non-parametric bootstrapping with 1,000 resamples. The 95% confidence interval for the indirect effect was calculated from the 2.5th and 97.5th percentiles of the bootstrap distribution. Mediation analysis were performed using Python (version 3.7) with the “statsmodels” and “scikit-learn” packages.

For missing data, variables with more than 15% missingness were excluded. Among the remaining variables, calcium (2.41%), admission weight (1.91%), and GCS (0.25%) had missing values. Multiple imputation by chained equations (MICE) was applied to variables with missingness rates of ≤ 15% using the “mice” package in R (version 4.4.1), with five imputations and 20 iterations. Logistic regression was applied to categorical data, and predictive mean matching to continuous data.

All statistical analyses were conducted using Python (version 3.7) and R (version 4.4.1). A two-tailed P < 0.05 was considered statistically significant.

Model development and performance evaluation

To assess the predictive performance of the selected prognostic factors for TBI mortality, we employed Cox regression, random survival forest (RSF), and two deep learning–based survival models, Deepsurv and Coxtime. The dataset was randomly divided into training (70%), validation (15%), and testing (15%) sets. Model hyperparameters were optimized through five-fold cross-validation on the training data, and the corresponding search spaces are summarized in S2 Table. Model performance was evaluated using the time-dependent area under the ROC curve (AUC), concordance index (C-index), integrated Brier score (IBS), and calibration plots to comprehensively assess discrimination and calibration. All models were trained in Python (version 3.7). The Cox and RSF models were implemented using the “lifelines” and “scikit-survival” packages, respectively, while the deep learning models were constructed with “PyCox”.

Results

Population and baseline characteristics

A total of 3404 ICU patients diagnosed with TBI and aged > 18 years were initially screened. After restricting the cohort to patients admitted to the ICU for the first time (n = 1644), we further excluded those with ICU stays <24 hours (n = 526) and cases with missing key variable data (n = 31). Ultimately, 1,203 patients were included. Baseline characteristics stratified by 28-day and in-hospital mortality outcomes are summarized in Tables 1 and 2. The cohort had a median age of 68 years (IQR: 51–81.5), with 40.1% male (n = 483) and 59.9% female (n = 720), the majority (70.5%) identified as White.

Download:

Table 1. Baseline characteristics of participants stratified by 28-day survival.

https://doi.org/10.1371/journal.pone.0339879.t001

Download:

Table 2. Baseline characteristics of participants stratified by in-hospital survival.

https://doi.org/10.1371/journal.pone.0339879.t002

The trends observed for 28-day and in-hospital mortality outcomes were generally consistent. Non-survivors exhibited significantly different characteristics compared to survivors (all P < 0.05): older age, higher invasive ventilation use. Laboratory comparisons revealed significant intergroup differences in multiple biomarkers. Non-survivors exhibited elevated levels of RDW, creatinine, BUN, WBC, and glucose compared to survivors (all P < 0.05). Conversely, hemoglobin, hematocrit, and platelet counts were notably reduced in non-survivors (all P < 0.05). No statistically significant differences were observed for calcium, sodium, potassium, or chloride levels (P > 0.05).

Cox regression analyses

Univariate and multivariate Cox regression analyses were conducted to identify prognostic factors for 28-day and in-hospital mortality. The multivariate model showed that age (HR: 1.022, 95% CI: 1.012–1.033; P < 0.001 for 28-day mortality; HR: 1.031, 95% CI: 1.018–1.045; P < 0.001 for in-hospital mortality) and RDW levels (HR: 1.085, 95% CI: 1.006–1.169; P = 0.034 for 28-day mortality; HR: 1.094, 95% CI: 1.008–1.188; P = 0.032 for in-hospital mortality) were significant independent predictors for mortality. Additionally, LODS (HR: 1.175, 95% CI: 1.074–1.285; P < 0.001 for 28-day mortality; HR: 1.165, 95% CI: 1.051–1.292; P = 0.004 for in-hospital mortality), WBC (HR: 1.015, 95% CI: 1.001–1.029; P = 0.033 for in-hospital mortality), and IMV (HR: 2.551, 95% CI: 1.725–3.772; P < 0.001 for 28-day mortality; HR: 2.762, 95% CI: 1.687–4.522; P < 0.001 for in-hospital mortality) were also demonstrated significantly associated with mortality risk (Table 3).

Download:

Table 3. Univariate and multivariate Cox regression analyses of factors associated with mortality.

https://doi.org/10.1371/journal.pone.0339879.t003

Kaplan-meier survival analysis

Kaplan-Meier (KM) survival analysis was employed to evaluate mortality in TBI patients of different ages and RDW levels. Age stratification analysis revealed that the 28-day mortality rate was significantly higher in the older age group (n = 542, events = 279) compared to the younger age group (n = 663, events = 145; P = 2.06 × 10 ⁻ ¹²; Fig 2A). Similarly, for in-hospital mortality, the older group (n = 381, events = 53) demonstrated a significantly higher rate relative to the younger group (n = 903, events = 71, P = 1.97 × 10 ⁻ ⁵; Fig 2B).

Download:

Fig 2. Kaplan–Meier survival curves for age and RDW levels to 28-day and in-hospital mortality among patients with TBI.

(A) Age and 28-day mortality; (B) Age and in-hospital mortality; (C) RDW levels and 28-day mortality; (D) RDW levels and in-hospital mortality. TBI, traumatic brain injury; RDW, red cell distribution width.

https://doi.org/10.1371/journal.pone.0339879.g002

RDW stratification analysis showed that, for 28-day mortality analysis, 15.30% was determined as the optimal cutoff value for RDW. The high-RDW group (n = 281, events = 164) exhibited a higher mortality than the low-RDW group (n = 923, events = 260, P = 4.35 × 10 ⁻ ⁷; Fig 2C). Regarding the in-hospital mortality analysis, a cutoff value of 15.40% was defined as the optimal RDW value. The high-RDW group (n = 262, events = 53) had a significantly higher mortality compared to the low-RDW group (n = 941, events = 71, P = 3.59× 10 ⁻ ⁶; Fig 2D).

Cox regression analysis of RDW and TBI mortality

As presented in Table 4, RDW levels demonstrated significant associations with both 28-day and in-hospital mortality. Three hierarchical models were constructed. In the fully adjusted Model 3, RDW was independently associated with a higher risk of 28-day mortality (HR = 1.088, 95% CI: 1.011–1.170, P = 0.020) and in-hospital mortality (HR = 1.102, 95% CI: 1.017–1.195, P = 0.017). To evaluate the dose–response relationship, RDW levels were stratified into quartiles (Q1–Q4), with Q1 serving as the reference group. In the unadjusted model, patients in the highest quartile (Q4) had significantly increased risks of both 28-day mortality (HR = 3.453, 95% CI: 2.030–5.873, P < 0.001) and in-hospital mortality (HR = 3.910, 95% CI: 2.133–7.169, P < 0.001). In the fully adjusted model (Model 3), Q4 was associated with higher risks of both 28-day mortality (HR = 2.061, 95% CI: 1.192–3.562, P = 0.008) and in-hospital mortality (HR = 2.086, 95% CI: 1.114–3.906, P = 0.028). Statistically significant linear trends across quartiles were observed (P_trend = 0.014 for 28-day mortality; P_trend = 0.033 for in-hospital mortality).

Download:

Table 4. The association between RDW levels and mortality analyzed using Cox analyses.

https://doi.org/10.1371/journal.pone.0339879.t004

Non-linear analyses

RCS analysis revealed a significant nonlinear association between RDW levels and in-hospital mortality (P_nonlinearity < 0.05; Fig 3A). The risk of mortality exhibited a progressive increase with rising RDW values, with a pronounced acceleration beyond the reflection point; the risk increased significantly faster, with a nonlinear trend that was statistically significant (P_nonlinearity < 0.05). In contrast, although RDW levels demonstrated a positive trend over 28 days, no considerable nonlinearity was observed (Fig 3B).

Download:

Fig 3. RCS analysis of RDW and age in relation to 28-day and in-hospital mortality among patients with TBI.

(A) Association between RDW and 28-day mortality; (B) Association between RDW and in-hospital mortality; (C) Association between age and 28-day mortality; (D) Association between age and in-hospital mortality. The pink dashed line indicates the threshold identified by the RCS analysis.

https://doi.org/10.1371/journal.pone.0339879.g003

Further analysis of the age-related mortality revealed a robust nonlinear relationship between age and 28-day mortality (P_nonlinearity < 0.05; Fig 3C). The curve indicated a steep escalation in mortality risk beyond 72.91 years, indicating a threshold effect of age on short-term prognosis. Regarding in-hospital mortality, the age-related curve displayed a gradual upward slope. Although the overall risk increased with age, the nonlinear trend was relatively feeble (P_nonlinearity > 0.05; Fig 3D).

Mediation role of RDW in the association between age and mortality among TBI patients

Our analysis revealed a significant correlation among age, RDW levels, and both 28-day and in-hospital mortality in TBI patients. To enhance clinical interpretability, age was standardized as a continuous variable scaled in 10-year increments. Mediation analysis was employed to evaluate the potential mediating effect of RDW on the association between age (increased by 10 years) and TBI mortality. For 28-day mortality, RDW partially mediated the effect of age on mortality. The average causal mediation effect (ACME) reached statistical significance (95% bootstrap CI: 0.0032–0.0261), indicating a meaningful indirect pathway via RDW. The average direct effect (ADE) of age remained significant (estimate: 0.2978; 95% CI: 0.1831–0.4125; P = 3.597 × 10⁻⁷), as did the total effect (estimate: 0.3015; 95% CI: 0.1881–0.4149; P = 1.878 × 10⁻⁷). RDW accounted for 4.40% of the total effect size (Fig 4A). For in-hospital mortality, similar mediation patterns were observed. The ACME was statistically significant (95% bootstrap CI: 0.0037–0.0285), indicating an indirect effect of age on mortality via RDW. The ADE (estimate: 0.3205; 95% CI: 0.1894–0.4516; P = 1.649 × 10⁻⁶) and the total effect (estimate: 0.3242; 95% CI: 0.1931–0.4516; P = 1.021 × 10⁻⁶) both remained significant. The proportion of the total effect mediated by RDW was 4.62% (Fig 4B).

Download:

Fig 4. Mediation Analysis of Age, Red Cell Distribution Width (RDW), and Mortality in TBI Patients.

(A) Mediation Path Coefficients. Indirect Effect (a × b): The combined effect of age on mortality mediated through RDW; b (RDW→Death): Direct effect of RDW on mortality; a (Age → RDW): Effect of age on RDW. (B) Decomposition of Total Effect. Direct Effect (c’): The direct effect of age on mortality, accounting for the mediating role of RDW; Indirect Effect (a × b): The proportion of the total effect mediated by RDW; Total Effect (c): The overall effect of age on mortality.

https://doi.org/10.1371/journal.pone.0339879.g004

Prognostic modeling of TBI using traditional and deep learning approaches

The covariates identified through multivariable Cox regression analysis—including age, RDW, LODS, WBC, and IMV—demonstrated robust predictive performance across four survival models: Cox, RSF, DeepSurv, and Cox-Time. The DeepSurv model achieved the highest predictive accuracy (C-index: 0.759; IBS: 0.113; AUC (95% CI): 0.824 (0.735–0.900)), followed by the Cox-Time model (C-index: 0.716; IBS: 0.113; AUC (95% CI): 0.796 (0.705–0.878)), the Cox model (C-index: 0.743; IBS: 0.124; AUC (95% CI): 0.795 (0.696–0.872)), and the RSF model (C-index: 0.613; IBS: 0.122; AUC (95% CI): 0.780 (0.681–0.861)).The ROC curves (Figs 5A–D) confirmed the discriminative capacity of each model, and the calibration plots (Figs 6A–D) indicated good agreement between the predicted and observed probabilities of 28-day mortality.

Download:

Fig 5. ROC curves for the four survival models.

(A) CoxPH; (B) RSF; (C) Deepsurv; (D) Cox-time. The shaded area represents the 95% confidence interval of the AUC. AUC: area under the curve; CI: confidence interval.

https://doi.org/10.1371/journal.pone.0339879.g005

Download:

Fig 6. Calibration curves for predicting 28-day survival probability using four models.

(A) CoxPH; (B) RSF; (C) Deepsurv; (D) Cox-time. The red solid line indicates the calibration performance of each model. The blue error bars represent the 95% confidence intervals of the observed survival probability for each prediction group.

https://doi.org/10.1371/journal.pone.0339879.g006

Discussion

TBI remains a substantial global public health challenge, contributing significantly to mortality and long-term disability [1,2]. In our study of ICU-admitted TBI patients, age and RDW were identified as independent predictors of short-term mortality. Notably, RDW demonstrated a significant dose–response association with mortality risk and partially mediated the impact of age on death. These findings underscore the prognostic importance of RDW in TBI and support its use as a practical biomarker for clinical risk stratification and management.

Consistent with previous observational and registry-based studies, our analysis reaffirms age as an independent and powerful predictor of adverse outcomes following TBI [19,20]. In both 28-day and in-hospital mortality models, older age was associated with a higher risk of death. This association remained significant after adjusting for injury severity, organ dysfunction scores, and physiological derangements. Importantly, our data further reveal that this relationship between age and 28-day mortality is nonlinear, with mortality risk escalating more sharply in advanced age groups. This age-related vulnerability is likely to be caused by several biological and clinical factors. This pattern may reflect age-related decline in neuroplasticity, reduced compensatory capacity, and heightened susceptibility to secondary insults such as inflammation and organ dysfunction [26–28]. Ageing is accompanied by a decline in neuroplasticity and regenerative capacity, an increased prevalence of comorbidities, polypharmacy, and diminished physiological reserve. These age-related changes, which are common in older adults, can compromise the response to acute brain injury [29,30]. Additionally, the ageing brain exhibits a distinct immune-inflammatory response characterized by chronic microglial activation and impaired resolution of inflammation, which can exacerbate secondary brain injury [31]. These considerations emphasize the importance of age-specific management strategies and identifying modifiable factors that can mitigate age-related risks.

RDW, a measure of the heterogeneity of circulating erythrocyte size, has emerged in recent years as a prognostic biomarker in diverse critical illnesses, ranging from cardiovascular disease to sepsis and ARDS [12–14,32]. Our analysis demonstrates that RDW is independently associated with both 28-day and in-hospital mortality in TBI patients. Observational studies have demonstrated that increased RDW is independently associated with adverse outcomes in ICU patients [33]. Moreover, a retrospective study found that in patients with head trauma–induced loss of consciousness, higher RDW was associated with an increased risk of brain death, suggesting it may serve as an early prognostic marker [34]. Biologically, elevated RDW appears to reflect interrelated pathophysiological processes, including systemic inflammation, oxidative stress, and impaired erythropoiesis, all of which contribute to the progression of secondary brain injury following TBI. Inflammatory cytokines can interfere with erythropoiesis, alter red blood cell membrane integrity, and promote anisocytosis, resulting in higher RDW values [35]. RDW may also serve as an indirect marker of underlying chronic disease burden or nutritional deficiencies, both of which themselves are associated with worse outcomes in critical illness [36]. Additionally, red blood cell deformability and rheology affect microcirculatory perfusion and oxygen delivery to injured brain tissue, suggesting that elevated RDW could contribute to tissue hypoxia and further neurological deterioration [15,37].

Our study also showed that RDW slightly mediates the effect of age on mortality in TBI patients. Patel et al. reported that RDW was strongly associated with mortality from multiple causes among middle-aged and older adults, each 1% increase in RDW was associated with approximately a 14% higher risk of death [38]. Additionally, studies in geriatric patients indicated that RDW is a dynamic variable, progressively increasing in the years preceding death, particularly during the last year of life, further supporting its role as a marker of physiological aging and mortality risk [39]. Moreover, hematologic studies have shown that RDW varies with age and reflects immunologic and inflammatory alterations linked to increased mortality risk [40]. These findings suggest that RDW may represent one hematologic pathway through which aging contributes to worse outcomes after TBI, reflecting a biologically plausible mechanism linking aging, hematologic dysregulation, and mortality risk.

Our analysis also reveals that RDW partially mediates impact of age on mortality risk in patients with TBI. Although the mediated proportion is modest, it was statistically significant and biologically meaningful, suggesting that elevated RDW may represent one path through which aging confers increased vulnerability after TBI. This finding is supported by prior studies. One study suggests that the impact of aging on inflammation and oxidative stress may influence patient prognosis by altering erythrocyte morphology, including RDW [38]. Another investigation highlights RDW as a hematologic marker that reflects chronic inflammatory status, considered a crucial mechanism underlying the risk of age-related diseases [41]. Recent evidence suggests that age-related alterations in blood parameters, including RDW, are associated with increased mortality risk. These findings highlight the potential of RDW as a biomarker reflecting systemic aging processes and their detrimental health effects [42].

In recent years, deep learning-based survival models have become increasingly valuable for predicting clinical outcomes, particularly in complex and heterogeneous conditions [43,44]. We applied four survival models to assess the prognostic value of the significant variables identified in our analysis. Deep learning–based survival analysis models, designed to handle time-to-event data and nonlinear relationships, demonstrated strong discriminative performance in predicting 28-day mortality. Notably, RDW emerged as a consistent and robust predictor across all models, reinforcing its potential as a clinically valuable biomarker for risk stratification in TBI patients. Existing prognostic models, such as the IMPACT and CRASH models [45,46], are widely used for TBI outcome prediction. As a readily available and inexpensive hematological parameter, RDW may provide complementary information to existing models, future studies could explore the incremental predictive value of incorporating RDW into the IMPACT and CRASH models to evaluate its potential role in risk stratification and clinical decision-making. Moreover, it may be valuable to assess the incremental prognostic contribution of RDW in combination with widely used TBI indicators, such as GCS, pupillary response, and CT-based scoring systems.

These findings offer practical implications for improving prognostic assessment in TBI. RDW, as a widely available and inexpensive biomarker, could be readily incorporated into clinical risk models to enhance early identification of high-risk patients. The combined evaluation of age and RDW may further refine prognostic accuracy, particularly in older adults. The observed nonlinear association between RDW and mortality suggests that even modest elevations may be clinically meaningful, underscoring the need to establish appropriate thresholds. Finally, the application of deep learning survival models demonstrates the potential of integrating routine clinical variables into advanced predictive frameworks to support individualized TBI management.

However, several limitations should be acknowledged. First, its retrospective design may introduce selection bias and unmeasured confounding. Second, as the analysis was conducted using a single-center dataset, the generalizability of the results to other populations or settings may be limited. Third, although we adjusted for numerous confounders, residual confounding due to unmeasured factors cannot be excluded. Fourth, RDW was measured only at ICU admission, longitudinal changes in RDW and their association with outcomes were not assessed. Finally, while mediation analysis provides insight into potential causal pathways, it cannot establish causality due to the observational nature of the data.

Further research is warranted to validate these findings in larger, multi-center cohorts with diverse populations and to explore the potential mechanisms linking RDW, aging, and TBI outcomes. Prospective studies examining dynamic changes in RDW over time, as well as interventional trials targeting modifiable factors that contribute to elevated RDW, may provide insight into potential therapeutic avenues. In addition, future studies should collect more comprehensive data on factors such as nutritional status, chronic anemia, blood transfusions, injury severity, comorbidities, and interventions, in order to further clarify these associations.

Conclusions

In summary, this study highlights the prognostic relevance of RDW in TBI patients, both as an independent predictor of short-term mortality and as a partial mediator of the age-mortality relationship. Our findings demonstrate that RDW, a routinely available hematologic parameter, provides meaningful prognostic information beyond established clinical indicators. The consistent performance of RDW across traditional and deep learning-based survival models underscores its potential utility in enhancing individualized risk assessment. Incorporating RDW into clinical decision-making frameworks may facilitate early identification of high-risk patients and inform age-sensitive management strategies. These results warrant further validation and mechanistic exploration to better define the role of RDW in the complex pathophysiology of TBI.

Supporting information

S1 Table. VIF of Candidate Variables for Multivariable Cox Regression.

The table presents the Variance Inflation Factor (VIF) for candidate variables considered in the multivariable Cox regression analysis. Variables with VIF > 5 were considered highly collinear and excluded from the final model. VIF: variance inflation factor; SAPSII: Simplified Acute Physiology Score II; LODS: Logistic Organ Dysfunction System; SOFA: Sequential Organ Failure Assessment; BUN: Blood Urea Nitrogen; GCS: Glasgow Coma Scale; IMV: Invasive mechanical ventilation; RDW: red blood cell distribution width; WBC: white blood cell count.

https://doi.org/10.1371/journal.pone.0339879.s001

(DOCX)

S2 Table. The hyper-parameters search space of ML model.

The table summarizes the hyper-parameter search ranges used for three machine learning survival models: Random Survival Forest (RSF), DeepSurv, and Cox-Time. For each model, the hyper-parameters and their corresponding candidate values used in the grid search or optimization procedure are listed. ML: Machine Learning; RSF: Random Survival Forest.

https://doi.org/10.1371/journal.pone.0339879.s002

(DOCX)

Acknowledgments

We thank the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center for their efforts in developing and maintaining the MIMIC project.

References

1. Maas AIR, Stocchetti N, Bullock R. Moderate and severe traumatic brain injury in adults. Lancet Neurol. 2008;7(8):728–41. pmid:18635021
- View Article
- PubMed/NCBI
- Google Scholar
2. Dewan MC, Rattani A, Gupta S, Baticulon RE, Hung Y-C, Punchak M, et al. Estimating the global incidence of traumatic brain injury. J Neurosurg. 2018;130(4):1080–97. pmid:29701556
- View Article
- PubMed/NCBI
- Google Scholar
3. Yan J, Wang C, Sun B. Global, regional, and national burdens of traumatic brain injury from 1990 to 2021. Front Public Health. 2025;13:1556147. pmid:40297033
- View Article
- PubMed/NCBI
- Google Scholar
4. Werner C, Engelhard K. Pathophysiology of traumatic brain injury. Br J Anaesth. 2007;99(1):4–9. pmid:17573392
- View Article
- PubMed/NCBI
- Google Scholar
5. Ziebell JM, Morganti-Kossmann MC. Involvement of pro- and anti-inflammatory cytokines and chemokines in the pathophysiology of traumatic brain injury. Neurotherapeutics. 2010;7(1):22–30. pmid:20129494
- View Article
- PubMed/NCBI
- Google Scholar
6. Shlosberg D, Benifla M, Kaufer D, Friedman A. Blood-brain barrier breakdown as a therapeutic target in traumatic brain injury. Nat Rev Neurol. 2010;6(7):393–403. pmid:20551947
- View Article
- PubMed/NCBI
- Google Scholar
7. Gardner RC, Yaffe K. Epidemiology of mild traumatic brain injury and neurodegenerative disease. Mol Cell Neurosci. 2015;66(Pt B):75–80. pmid:25748121
- View Article
- PubMed/NCBI
- Google Scholar
8. Roozenbeek B, Maas AIR, Menon DK. Changing patterns in the epidemiology of traumatic brain injury. Nat Rev Neurol. 2013;9(4):231–6. pmid:23443846
- View Article
- PubMed/NCBI
- Google Scholar
9. Na. The 2002 Tanner-Vandeput-Boswick Burn Prize Awarded. The Journal of Trauma: Injury, Infection, and Critical Care. 53(2):212.
- View Article
- Google Scholar
10. MRC CRASH Trial Collaborators, Perel P, Arango M, Clayton T, Edwards P, Komolafe E, et al. Predicting outcome after traumatic brain injury: practical prognostic models based on large cohort of international patients. BMJ. 2008;336(7641):425–9. pmid:18270239
- View Article
- PubMed/NCBI
- Google Scholar
11. Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, McHugh GS, et al. Predicting outcome after traumatic brain injury: development and international validation of prognostic scores based on admission characteristics. PLoS Med. 2008;5(8):e165; discussion e165. pmid:18684008
- View Article
- PubMed/NCBI
- Google Scholar
12. Felker GM, Allen LA, Pocock SJ, Shaw LK, McMurray JJV, Pfeffer MA, et al. Red cell distribution width as a novel prognostic marker in heart failure: data from the CHARM Program and the Duke Databank. J Am Coll Cardiol. 2007;50(1):40–7. pmid:17601544
- View Article
- PubMed/NCBI
- Google Scholar
13. Jo YH, Kim K, Lee JH, Kang C, Kim T, Park H-M, et al. Red cell distribution width is a prognostic factor in severe sepsis and septic shock. Am J Emerg Med. 2013;31(3):545–8. pmid:23380094
- View Article
- PubMed/NCBI
- Google Scholar
14. Wang B, Gong Y, Ying B, Cheng B. Relation between Red Cell Distribution Width and Mortality in Critically Ill Patients with Acute Respiratory Distress Syndrome. Biomed Res Int. 2019;2019:1942078. pmid:31016186
- View Article
- PubMed/NCBI
- Google Scholar
15. Bibic A, Sordia T, Henningsson E, Knutsson L, Ståhlberg F, Wirestam R. Effects of red blood cells with reduced deformability on cerebral blood flow and vascular water transport: measurements in rats using time-resolved pulsed arterial spin labelling at 9.4 T. Eur Radiol Exp. 2021;5(1):53. pmid:34935093
- View Article
- PubMed/NCBI
- Google Scholar
16. Lorente L, Martín MM, Ruiz C, Abreu-González P, Pérez-Cejas A, González-Rivero AF, et al. Red blood cell distribution width as mortality biomarker in patients with traumatic brain injury. Acta Neurol Belg. 2021;121(3):715–20. pmid:32572840
- View Article
- PubMed/NCBI
- Google Scholar
17. Zhang B, Zhao J. Red blood cell distribution width as a prognostic biomarker for mortality in traumatic brain injury. Int J Clin Exp Med. 2015;8(10):19172–5. pmid:26770551
- View Article
- PubMed/NCBI
- Google Scholar
18. Sadaka F, Doctors N, Pearson T, Snyders B, O’Brien J. Does Red Cell Distribution Width Predict Outcome in Traumatic Brain Injury: Comparison to Corticosteroid Randomization After Significant Head Injury. J Clin Med Res. 2018;10(1):9–12. pmid:29238428
- View Article
- PubMed/NCBI
- Google Scholar
19. Hukkelhoven CWPM, Steyerberg EW, Rampen AJJ, Farace E, Habbema JDF, Marshall LF, et al. Patient age and outcome following severe traumatic brain injury: an analysis of 5600 patients. J Neurosurg. 2003;99(4):666–73. pmid:14567601
- View Article
- PubMed/NCBI
- Google Scholar
20. Mosenthal AC, Lavery RF, Addis M, Kaul S, Ross S, Marburger R, et al. Isolated traumatic brain injury: age is an independent predictor of mortality and early outcome. J Trauma. 2002;52(5):907–11. pmid:11988658
- View Article
- PubMed/NCBI
- Google Scholar
21. Walter AE, Gugger JJ, Law CA, Brennan DJ, Mosley TH, Reid RI, et al. Neuroimaging Correlates of Traumatic Brain Injury in an Older Community-Dwelling Population: The Atherosclerosis Risk in Communities Study. Neurology. 2025;104(8):e213506. pmid:40184574
- View Article
- PubMed/NCBI
- Google Scholar
22. Capizzi A, Woo J, Verduzco-Gutierrez M. Traumatic Brain Injury: An Overview of Epidemiology, Pathophysiology, and Medical Management. Med Clin North Am. 2020;104(2):213–38. pmid:32035565
- View Article
- PubMed/NCBI
- Google Scholar
23. Satyadev N, Warman PI, Seas A, Kolls BJ, Haglund MM, Fuller AT, et al. Machine Learning for Predicting Discharge Disposition After Traumatic Brain Injury. Neurosurgery. 2022;90(6):768–74. pmid:35319523
- View Article
- PubMed/NCBI
- Google Scholar
24. Teoh L, Ihalage AA, Harp S, F Al-Khateeb Z, Michael-Titus AT, Tremoleda JL, et al. Deep learning for behaviour classification in a preclinical brain injury model. PLoS One. 2022;17(6):e0268962. pmid:35704595
- View Article
- PubMed/NCBI
- Google Scholar
25. Bae C-Y, Kim B-S, Jee S-H, Lee J-H, Nguyen N-D. A Study on Survival Analysis Methods Using Neural Network to Prevent Cancers. Cancers (Basel). 2023;15(19):4757. pmid:37835451
- View Article
- PubMed/NCBI
- Google Scholar
26. Susman M, DiRusso SM, Sullivan T, Risucci D, Nealon P, Cuff S, et al. Traumatic brain injury in the elderly: increased mortality and worse functional outcome at discharge despite lower injury severity. J Trauma. 2002;53(2):219–23; discussion 223-4. pmid:12169925
- View Article
- PubMed/NCBI
- Google Scholar
27. Roozenbeek B, Maas AIR, Menon DK. Changing patterns in the epidemiology of traumatic brain injury. Nat Rev Neurol. 2013;9(4):231–6. pmid:23443846
- View Article
- PubMed/NCBI
- Google Scholar
28. Thompson HJ, McCormick WC, Kagan SH. Traumatic brain injury in older adults: epidemiology, outcomes, and future implications. J Am Geriatr Soc. 2006;54(10):1590–5. pmid:17038079
- View Article
- PubMed/NCBI
- Google Scholar
29. Green TRF, Murphy SM, Ortiz JB, Rowe RK. Age-At-Injury Influences the Glial Response to Traumatic Brain Injury in the Cortex of Male Juvenile Rats. Front Neurol. 2022;12:804139. pmid:35111130
- View Article
- PubMed/NCBI
- Google Scholar
30. Krukowski K, Chou A, Feng X, Tiret B, Paladini M-S, Riparip L-K, et al. Traumatic Brain Injury in Aged Mice Induces Chronic Microglia Activation, Synapse Loss, and Complement-Dependent Memory Deficits. Int J Mol Sci. 2018;19(12):3753. pmid:30486287
- View Article
- PubMed/NCBI
- Google Scholar
31. Moro F, Pischiutta F, Portet A, Needham EJ, Norton EJ, Ferdinand JR, et al. Ageing is associated with maladaptive immune response and worse outcome after traumatic brain injury. Brain Commun. 2022;4(2):fcac036. pmid:35350551
- View Article
- PubMed/NCBI
- Google Scholar
32. Lai H, Wu G, Zhong Y, Chen G, Zhang W, Shi S, et al. Red blood cell distribution width improves the prediction of 28-day mortality for patients with sepsis-induced acute kidney injury: A retrospective analysis from MIMIC-IV database using propensity score matching. J Intensive Med. 2023;3(3):275–82. pmid:37533812
- View Article
- PubMed/NCBI
- Google Scholar
33. Zhang Z, Xu X, Ni H, Deng H. Red cell distribution width is associated with hospital mortality in unselected critically ill patients. J Thorac Dis. 2013;5(6):730–6. pmid:24409348
- View Article
- PubMed/NCBI
- Google Scholar
34. Latifi M, Rahban H, Pourhosein E, Shostak D, Dehghani S. Association between red blood cell distribution width and the prognosis of brain death in patients with a Glasgow Coma Scale < 6. Sci Rep. 2023;13(1):14027. pmid:37640749
- View Article
- PubMed/NCBI
- Google Scholar
35. Mahmood NA, Mathew J, Kang B, DeBari VA, Khan MA. Broadening of the red blood cell distribution width is associated with increased severity of illness in patients with sepsis. Int J Crit Illn Inj Sci. 2014;4(4):278–82. pmid:25625057
- View Article
- PubMed/NCBI
- Google Scholar
36. Patel KV, Ferrucci L, Ershler WB, Longo DL, Guralnik JM. Red blood cell distribution width and the risk of death in middle-aged and older adults. Arch Intern Med. 2009;169(5):515–23. pmid:19273783
- View Article
- PubMed/NCBI
- Google Scholar
37. Arkew M, Gemechu K, Haile K, Asmerom H. Red Blood Cell Distribution Width as Novel Biomarker in Cardiovascular Diseases: A Literature Review. J Blood Med. 2022;13:413–24. pmid:35942475
- View Article
- PubMed/NCBI
- Google Scholar
38. Patel KV, Semba RD, Ferrucci L, Newman AB, Fried LP, Wallace RB, et al. Red cell distribution width and mortality in older adults: a meta-analysis. J Gerontol A Biol Sci Med Sci. 2010;65(3):258–65. pmid:19880817
- View Article
- PubMed/NCBI
- Google Scholar
39. Martínez-Velilla N, Cambra-Contin K, García-Baztán A, Alonso-Renedo J, Herce PA, Ibáñez-Beroiz B. Change in Red Blood Cell Distribution width During the Last Years of Life in Geriatric Patients. J Nutr Health Aging. 2015;19(5):590–4. pmid:25923491
- View Article
- PubMed/NCBI
- Google Scholar
40. Dalton JE, Zidar DA, Krieger NI, Perzynski AT, Gunzler DD. Age-related differences in immuno-hematologic profiles and their association with all-cause mortality. Innovation in Aging. 2019;3(Supplement_1):S103–4.
- View Article
- Google Scholar
41. Hoffmann JJML, Nabbe KCAM, van den Broek NMA. Effect of age and gender on reference intervals of red blood cell distribution width (RDW) and mean red cell volume (MCV). Clin Chem Lab Med. 2015;53(12):2015–9. pmid:26536583
- View Article
- PubMed/NCBI
- Google Scholar
42. Dalton JE, Zidar DA, Krieger NI, Perzynski AT, Gunzler DD. Age-related differences in immuno-hematologic profiles and their association with all-cause mortality. Innovation in Aging. 2019;3(Supplement_1):S103–4.
- View Article
- Google Scholar
43. Guo Y, Yu F, Jiang F-F, Yin S-J, Jiang M-H, Li Y-J, et al. Development and validation of novel interpretable survival prediction models based on drug exposures for severe heart failure during vulnerable period. J Transl Med. 2024;22(1):743. pmid:39107765
- View Article
- PubMed/NCBI
- Google Scholar
44. Teshale AB, Htun HL, Vered M, Owen AJ, Ryan J, Tonkin A, et al. Artificial intelligence improves risk prediction in cardiovascular disease. Geroscience. 2025;47(4):5497–502. pmid:39576563
- View Article
- PubMed/NCBI
- Google Scholar
45. Roberts I, Yates D, Sandercock P, Farrell B, Wasserberg J, Lomas G, et al. Effect of intravenous corticosteroids on death within 14 days in 10008 adults with clinically significant head injury (MRC CRASH trial): randomised placebo-controlled trial. Lancet. 2004;364(9442):1321–8. pmid:15474134
- View Article
- PubMed/NCBI
- Google Scholar
46. Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, McHugh GS, et al. Predicting outcome after traumatic brain injury: development and international validation of prognostic scores based on admission characteristics. PLoS Med. 2008;5(8):e165; discussion e165. pmid:18684008
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Maas AIR, Stocchetti N, Bullock R. Moderate and severe traumatic brain injury in adults. Lancet Neurol. 2008;7(8):728–41. pmid:18635021
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Dewan MC, Rattani A, Gupta S, Baticulon RE, Hung Y-C, Punchak M, et al. Estimating the global incidence of traumatic brain injury. J Neurosurg. 2018;130(4):1080–97. pmid:29701556
View Article
PubMed/NCBI
Google Scholar

[6] View Article

[7] PubMed/NCBI

[8] Google Scholar

[ref3] 3. Yan J, Wang C, Sun B. Global, regional, and national burdens of traumatic brain injury from 1990 to 2021. Front Public Health. 2025;13:1556147. pmid:40297033
View Article
PubMed/NCBI
Google Scholar

[10] View Article

[11] PubMed/NCBI

[12] Google Scholar

[ref4] 4. Werner C, Engelhard K. Pathophysiology of traumatic brain injury. Br J Anaesth. 2007;99(1):4–9. pmid:17573392
View Article
PubMed/NCBI
Google Scholar

[14] View Article

[15] PubMed/NCBI

[16] Google Scholar

[ref5] 5. Ziebell JM, Morganti-Kossmann MC. Involvement of pro- and anti-inflammatory cytokines and chemokines in the pathophysiology of traumatic brain injury. Neurotherapeutics. 2010;7(1):22–30. pmid:20129494
View Article
PubMed/NCBI
Google Scholar

[18] View Article

[19] PubMed/NCBI

[20] Google Scholar

[ref6] 6. Shlosberg D, Benifla M, Kaufer D, Friedman A. Blood-brain barrier breakdown as a therapeutic target in traumatic brain injury. Nat Rev Neurol. 2010;6(7):393–403. pmid:20551947
View Article
PubMed/NCBI
Google Scholar

[22] View Article

[23] PubMed/NCBI

[24] Google Scholar

[ref7] 7. Gardner RC, Yaffe K. Epidemiology of mild traumatic brain injury and neurodegenerative disease. Mol Cell Neurosci. 2015;66(Pt B):75–80. pmid:25748121
View Article
PubMed/NCBI
Google Scholar

[26] View Article

[27] PubMed/NCBI

[28] Google Scholar

[ref8] 8. Roozenbeek B, Maas AIR, Menon DK. Changing patterns in the epidemiology of traumatic brain injury. Nat Rev Neurol. 2013;9(4):231–6. pmid:23443846
View Article
PubMed/NCBI
Google Scholar

[30] View Article

[31] PubMed/NCBI

[32] Google Scholar

[ref9] 9. Na. The 2002 Tanner-Vandeput-Boswick Burn Prize Awarded. The Journal of Trauma: Injury, Infection, and Critical Care. 53(2):212.
View Article
Google Scholar

[34] View Article

[35] Google Scholar

[ref10] 10. MRC CRASH Trial Collaborators, Perel P, Arango M, Clayton T, Edwards P, Komolafe E, et al. Predicting outcome after traumatic brain injury: practical prognostic models based on large cohort of international patients. BMJ. 2008;336(7641):425–9. pmid:18270239
View Article
PubMed/NCBI
Google Scholar

[37] View Article

[38] PubMed/NCBI

[39] Google Scholar

[ref11] 11. Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, McHugh GS, et al. Predicting outcome after traumatic brain injury: development and international validation of prognostic scores based on admission characteristics. PLoS Med. 2008;5(8):e165; discussion e165. pmid:18684008
View Article
PubMed/NCBI
Google Scholar

[41] View Article

[42] PubMed/NCBI

[43] Google Scholar

[ref12] 12. Felker GM, Allen LA, Pocock SJ, Shaw LK, McMurray JJV, Pfeffer MA, et al. Red cell distribution width as a novel prognostic marker in heart failure: data from the CHARM Program and the Duke Databank. J Am Coll Cardiol. 2007;50(1):40–7. pmid:17601544
View Article
PubMed/NCBI
Google Scholar

[45] View Article

[46] PubMed/NCBI

[47] Google Scholar

[ref13] 13. Jo YH, Kim K, Lee JH, Kang C, Kim T, Park H-M, et al. Red cell distribution width is a prognostic factor in severe sepsis and septic shock. Am J Emerg Med. 2013;31(3):545–8. pmid:23380094
View Article
PubMed/NCBI
Google Scholar

[49] View Article

[50] PubMed/NCBI

[51] Google Scholar

[ref14] 14. Wang B, Gong Y, Ying B, Cheng B. Relation between Red Cell Distribution Width and Mortality in Critically Ill Patients with Acute Respiratory Distress Syndrome. Biomed Res Int. 2019;2019:1942078. pmid:31016186
View Article
PubMed/NCBI
Google Scholar

[53] View Article

[54] PubMed/NCBI

[55] Google Scholar

[ref15] 15. Bibic A, Sordia T, Henningsson E, Knutsson L, Ståhlberg F, Wirestam R. Effects of red blood cells with reduced deformability on cerebral blood flow and vascular water transport: measurements in rats using time-resolved pulsed arterial spin labelling at 9.4 T. Eur Radiol Exp. 2021;5(1):53. pmid:34935093
View Article
PubMed/NCBI
Google Scholar

[57] View Article

[58] PubMed/NCBI

[59] Google Scholar

[ref16] 16. Lorente L, Martín MM, Ruiz C, Abreu-González P, Pérez-Cejas A, González-Rivero AF, et al. Red blood cell distribution width as mortality biomarker in patients with traumatic brain injury. Acta Neurol Belg. 2021;121(3):715–20. pmid:32572840
View Article
PubMed/NCBI
Google Scholar

[61] View Article

[62] PubMed/NCBI

[63] Google Scholar

[ref17] 17. Zhang B, Zhao J. Red blood cell distribution width as a prognostic biomarker for mortality in traumatic brain injury. Int J Clin Exp Med. 2015;8(10):19172–5. pmid:26770551
View Article
PubMed/NCBI
Google Scholar

[65] View Article

[66] PubMed/NCBI

[67] Google Scholar

[ref18] 18. Sadaka F, Doctors N, Pearson T, Snyders B, O’Brien J. Does Red Cell Distribution Width Predict Outcome in Traumatic Brain Injury: Comparison to Corticosteroid Randomization After Significant Head Injury. J Clin Med Res. 2018;10(1):9–12. pmid:29238428
View Article
PubMed/NCBI
Google Scholar

[69] View Article

[70] PubMed/NCBI

[71] Google Scholar

[ref19] 19. Hukkelhoven CWPM, Steyerberg EW, Rampen AJJ, Farace E, Habbema JDF, Marshall LF, et al. Patient age and outcome following severe traumatic brain injury: an analysis of 5600 patients. J Neurosurg. 2003;99(4):666–73. pmid:14567601
View Article
PubMed/NCBI
Google Scholar

[73] View Article

[74] PubMed/NCBI

[75] Google Scholar

[ref20] 20. Mosenthal AC, Lavery RF, Addis M, Kaul S, Ross S, Marburger R, et al. Isolated traumatic brain injury: age is an independent predictor of mortality and early outcome. J Trauma. 2002;52(5):907–11. pmid:11988658
View Article
PubMed/NCBI
Google Scholar

[77] View Article

[78] PubMed/NCBI

[79] Google Scholar

[ref21] 21. Walter AE, Gugger JJ, Law CA, Brennan DJ, Mosley TH, Reid RI, et al. Neuroimaging Correlates of Traumatic Brain Injury in an Older Community-Dwelling Population: The Atherosclerosis Risk in Communities Study. Neurology. 2025;104(8):e213506. pmid:40184574
View Article
PubMed/NCBI
Google Scholar

[81] View Article

[82] PubMed/NCBI

[83] Google Scholar

[ref22] 22. Capizzi A, Woo J, Verduzco-Gutierrez M. Traumatic Brain Injury: An Overview of Epidemiology, Pathophysiology, and Medical Management. Med Clin North Am. 2020;104(2):213–38. pmid:32035565
View Article
PubMed/NCBI
Google Scholar

[85] View Article

[86] PubMed/NCBI

[87] Google Scholar

[ref23] 23. Satyadev N, Warman PI, Seas A, Kolls BJ, Haglund MM, Fuller AT, et al. Machine Learning for Predicting Discharge Disposition After Traumatic Brain Injury. Neurosurgery. 2022;90(6):768–74. pmid:35319523
View Article
PubMed/NCBI
Google Scholar

[89] View Article

[90] PubMed/NCBI

[91] Google Scholar

[ref24] 24. Teoh L, Ihalage AA, Harp S, F Al-Khateeb Z, Michael-Titus AT, Tremoleda JL, et al. Deep learning for behaviour classification in a preclinical brain injury model. PLoS One. 2022;17(6):e0268962. pmid:35704595
View Article
PubMed/NCBI
Google Scholar

[93] View Article

[94] PubMed/NCBI

[95] Google Scholar

[ref25] 25. Bae C-Y, Kim B-S, Jee S-H, Lee J-H, Nguyen N-D. A Study on Survival Analysis Methods Using Neural Network to Prevent Cancers. Cancers (Basel). 2023;15(19):4757. pmid:37835451
View Article
PubMed/NCBI
Google Scholar

[97] View Article

[98] PubMed/NCBI

[99] Google Scholar

[ref26] 26. Susman M, DiRusso SM, Sullivan T, Risucci D, Nealon P, Cuff S, et al. Traumatic brain injury in the elderly: increased mortality and worse functional outcome at discharge despite lower injury severity. J Trauma. 2002;53(2):219–23; discussion 223-4. pmid:12169925
View Article
PubMed/NCBI
Google Scholar

[101] View Article

[102] PubMed/NCBI

[103] Google Scholar

[ref27] 27. Roozenbeek B, Maas AIR, Menon DK. Changing patterns in the epidemiology of traumatic brain injury. Nat Rev Neurol. 2013;9(4):231–6. pmid:23443846
View Article
PubMed/NCBI
Google Scholar

[105] View Article

[106] PubMed/NCBI

[107] Google Scholar

[ref28] 28. Thompson HJ, McCormick WC, Kagan SH. Traumatic brain injury in older adults: epidemiology, outcomes, and future implications. J Am Geriatr Soc. 2006;54(10):1590–5. pmid:17038079
View Article
PubMed/NCBI
Google Scholar

[109] View Article

[110] PubMed/NCBI

[111] Google Scholar

[ref29] 29. Green TRF, Murphy SM, Ortiz JB, Rowe RK. Age-At-Injury Influences the Glial Response to Traumatic Brain Injury in the Cortex of Male Juvenile Rats. Front Neurol. 2022;12:804139. pmid:35111130
View Article
PubMed/NCBI
Google Scholar

[113] View Article

[114] PubMed/NCBI

[115] Google Scholar

[ref30] 30. Krukowski K, Chou A, Feng X, Tiret B, Paladini M-S, Riparip L-K, et al. Traumatic Brain Injury in Aged Mice Induces Chronic Microglia Activation, Synapse Loss, and Complement-Dependent Memory Deficits. Int J Mol Sci. 2018;19(12):3753. pmid:30486287
View Article
PubMed/NCBI
Google Scholar

[117] View Article

[118] PubMed/NCBI

[119] Google Scholar

[ref31] 31. Moro F, Pischiutta F, Portet A, Needham EJ, Norton EJ, Ferdinand JR, et al. Ageing is associated with maladaptive immune response and worse outcome after traumatic brain injury. Brain Commun. 2022;4(2):fcac036. pmid:35350551
View Article
PubMed/NCBI
Google Scholar

[121] View Article

[122] PubMed/NCBI

[123] Google Scholar

[ref32] 32. Lai H, Wu G, Zhong Y, Chen G, Zhang W, Shi S, et al. Red blood cell distribution width improves the prediction of 28-day mortality for patients with sepsis-induced acute kidney injury: A retrospective analysis from MIMIC-IV database using propensity score matching. J Intensive Med. 2023;3(3):275–82. pmid:37533812
View Article
PubMed/NCBI
Google Scholar

[125] View Article

[126] PubMed/NCBI

[127] Google Scholar

[ref33] 33. Zhang Z, Xu X, Ni H, Deng H. Red cell distribution width is associated with hospital mortality in unselected critically ill patients. J Thorac Dis. 2013;5(6):730–6. pmid:24409348
View Article
PubMed/NCBI
Google Scholar

[129] View Article

[130] PubMed/NCBI

[131] Google Scholar

[ref34] 34. Latifi M, Rahban H, Pourhosein E, Shostak D, Dehghani S. Association between red blood cell distribution width and the prognosis of brain death in patients with a Glasgow Coma Scale < 6. Sci Rep. 2023;13(1):14027. pmid:37640749
View Article
PubMed/NCBI
Google Scholar

[133] View Article

[134] PubMed/NCBI

[135] Google Scholar

[ref35] 35. Mahmood NA, Mathew J, Kang B, DeBari VA, Khan MA. Broadening of the red blood cell distribution width is associated with increased severity of illness in patients with sepsis. Int J Crit Illn Inj Sci. 2014;4(4):278–82. pmid:25625057
View Article
PubMed/NCBI
Google Scholar

[137] View Article

[138] PubMed/NCBI

[139] Google Scholar

[ref36] 36. Patel KV, Ferrucci L, Ershler WB, Longo DL, Guralnik JM. Red blood cell distribution width and the risk of death in middle-aged and older adults. Arch Intern Med. 2009;169(5):515–23. pmid:19273783
View Article
PubMed/NCBI
Google Scholar

[141] View Article

[142] PubMed/NCBI

[143] Google Scholar

[ref37] 37. Arkew M, Gemechu K, Haile K, Asmerom H. Red Blood Cell Distribution Width as Novel Biomarker in Cardiovascular Diseases: A Literature Review. J Blood Med. 2022;13:413–24. pmid:35942475
View Article
PubMed/NCBI
Google Scholar

[145] View Article

[146] PubMed/NCBI

[147] Google Scholar

[ref38] 38. Patel KV, Semba RD, Ferrucci L, Newman AB, Fried LP, Wallace RB, et al. Red cell distribution width and mortality in older adults: a meta-analysis. J Gerontol A Biol Sci Med Sci. 2010;65(3):258–65. pmid:19880817
View Article
PubMed/NCBI
Google Scholar

[149] View Article

[150] PubMed/NCBI

[151] Google Scholar

[ref39] 39. Martínez-Velilla N, Cambra-Contin K, García-Baztán A, Alonso-Renedo J, Herce PA, Ibáñez-Beroiz B. Change in Red Blood Cell Distribution width During the Last Years of Life in Geriatric Patients. J Nutr Health Aging. 2015;19(5):590–4. pmid:25923491
View Article
PubMed/NCBI
Google Scholar

[153] View Article

[154] PubMed/NCBI

[155] Google Scholar

[ref40] 40. Dalton JE, Zidar DA, Krieger NI, Perzynski AT, Gunzler DD. Age-related differences in immuno-hematologic profiles and their association with all-cause mortality. Innovation in Aging. 2019;3(Supplement_1):S103–4.
View Article
Google Scholar

[157] View Article

[158] Google Scholar

[ref41] 41. Hoffmann JJML, Nabbe KCAM, van den Broek NMA. Effect of age and gender on reference intervals of red blood cell distribution width (RDW) and mean red cell volume (MCV). Clin Chem Lab Med. 2015;53(12):2015–9. pmid:26536583
View Article
PubMed/NCBI
Google Scholar

[160] View Article

[161] PubMed/NCBI

[162] Google Scholar

[ref42] 42. Dalton JE, Zidar DA, Krieger NI, Perzynski AT, Gunzler DD. Age-related differences in immuno-hematologic profiles and their association with all-cause mortality. Innovation in Aging. 2019;3(Supplement_1):S103–4.
View Article
Google Scholar

[164] View Article

[165] Google Scholar

[ref43] 43. Guo Y, Yu F, Jiang F-F, Yin S-J, Jiang M-H, Li Y-J, et al. Development and validation of novel interpretable survival prediction models based on drug exposures for severe heart failure during vulnerable period. J Transl Med. 2024;22(1):743. pmid:39107765
View Article
PubMed/NCBI
Google Scholar

[167] View Article

[168] PubMed/NCBI

[169] Google Scholar

[ref44] 44. Teshale AB, Htun HL, Vered M, Owen AJ, Ryan J, Tonkin A, et al. Artificial intelligence improves risk prediction in cardiovascular disease. Geroscience. 2025;47(4):5497–502. pmid:39576563
View Article
PubMed/NCBI
Google Scholar

[171] View Article

[172] PubMed/NCBI

[173] Google Scholar

[ref45] 45. Roberts I, Yates D, Sandercock P, Farrell B, Wasserberg J, Lomas G, et al. Effect of intravenous corticosteroids on death within 14 days in 10008 adults with clinically significant head injury (MRC CRASH trial): randomised placebo-controlled trial. Lancet. 2004;364(9442):1321–8. pmid:15474134
View Article
PubMed/NCBI
Google Scholar

[175] View Article

[176] PubMed/NCBI

[177] Google Scholar

[ref46] 46. Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, McHugh GS, et al. Predicting outcome after traumatic brain injury: development and international validation of prognostic scores based on admission characteristics. PLoS Med. 2008;5(8):e165; discussion e165. pmid:18684008
View Article
PubMed/NCBI
Google Scholar

[179] View Article

[180] PubMed/NCBI

[181] Google Scholar

Figures

Abstract

Introduction

Materials and methods

Data source

Study population

Data extraction

Primary and secondary outcomes

Statistical analysis

Model development and performance evaluation

Results

Population and baseline characteristics

Cox regression analyses

Kaplan-meier survival analysis

Cox regression analysis of RDW and TBI mortality

Non-linear analyses

Mediation role of RDW in the association between age and mortality among TBI patients

Prognostic modeling of TBI using traditional and deep learning approaches

Discussion

Conclusions

Supporting information

S1 Table. VIF of Candidate Variables for Multivariable Cox Regression.

S2 Table. The hyper-parameters search space of ML model.

Acknowledgments

References