Association between the endothelial activation and stress index and 30-day in-hospital mortality rate in patients with traumatic lung injury: a retrospective analysis based on the MIMIC-IV database

Abstract

Background

The endothelial activation and stress index (EASIX) quantifies endothelial injury and serves as a validated surrogate marker for endothelial damage. Individuals with traumatic lung injury (TLI) in the intensive care unit (ICU) often exhibit endothelial impairment. However, the prognostic implication of EASIX in this population remains unclear. Therefore, this study aims to investigate the relationship between EASIX and prognosis in patients with TLI.

Methods

Clinical data of individuals with TLI in the ICU were extracted from the Medical Information Mart for Intensive Care (MIMIC)-IV-3.1 database. The EASIX was calculated as lactate dehydrogenase (LDH) (U/L) × creatinine (mg/dL) / platelet count (PLT) (10⁹/L). Subjects were stratified into tertiles based on log₂-transformed EASIX. The Boruta algorithm was applied for feature selection. Cox proportional hazards models were constructed to examine the association between EASIX and 30-day in-hospital mortality (IHM). Restricted cubic spline (RCS) curves were generated to evaluate nonlinear relationships. Subgroup analyses were then conducted. The predictive efficacy of EASIX was validated by analyzing the area under the receiver operating characteristic curve (AUC). Random forest (RF) models incorporating EASIX were also developed.

Results

The analysis encompassed 231 participants. After full adjustment for confounding variables, Cox regression revealed a significant correlation between elevated levels of EASIX and an increased risk of IHM [hazard ratio (HR) =2.895, 95% confidence interval (CI): 1.245–6.735]. RCS analysis demonstrated an approximately linear relationship between EASIX and mortality (P>0.05 for nonlinearity). The AUC of EASIX alone for predicting IHM was 0.709 (95% CI: 0.621–0.797) in the overall cohort. A RF model incorporating EASIX achieved AUC values of 0.973 and 0.753 in the training and validation sets, respectively, indicating robust predictive performance.

Conclusions

EASIX is a reliable prognostic indicator for individuals with TLI. Larger prospective investigations are required to confirm causality.

Highlight box.

Key findings

• Endothelial activation and stress index (EASIX) functions effectively as a prognostic marker for patients with traumatic lung injury (TLI). However, larger prospective investigations are necessary to establish causality.

What is known and what is new?

• The EASIX quantifies endothelial injury and serves as a validated surrogate marker for endothelial damage. Individuals with TLI in the intensive care unit often exhibit endothelial impairment.

• Elevated EASIX is significantly associated with an increased risk of 30-day in-hospital mortality in individuals with TLI. EASIX may serve as a practical clinical tool for assessing prognosis in this population.

What is the implication, and what should change now?

• EASIX offers a simple yet accurate prognostic tool for critically ill patients, including those with TLI. As a retrospective cohort study, despite efforts to minimize reverse causality in the study design, the findings should be validated through prospective longitudinal cohort studies.

Introduction

Traumatic lung injury (TLI) denotes a pathological state of structural and functional impairment of the lung caused by direct or indirect thoracic or systemic external forces. It is commonly observed in incidents such as motor vehicle accidents, high falls, crush injuries, or penetrating trauma. TLI poses risks across multiple organ systems and can be life-threatening. The post-traumatic mortality rates can reach up to 30% in severely injured chest patients (1). The pathogenesis of TLI involves direct disruption of alveolar structures and pulmonary capillary endothelial cells by high-energy impact (2). This is followed by the massive release of inflammatory cytokines—such as tumor necrosis factor-α and interleukin-6 (IL-6)—which trigger inflammatory responses that damage the alveolar–capillary barrier (3). These processes lead to increased pulmonary vascular permeability, result in diffuse pulmonary edema (4), and ultimately cause severe hypoxemia and respiratory failure (5,6). Current literature indicates that acute respiratory distress syndrome (ARDS) is associated with a mortality rate of 30–40% (7,8). Therefore, early prevention or intervention to halt the progression of TLI and reduce the mortality rate is crucial. However, endothelial function assessment may have clinical limitations. Thus, a convenient and highly specific biomarker is urgently needed to predict the severity and prognosis of TLI.

Recently, a novel index, endothelial activation and stress index (EASIX), derived from lactate dehydrogenase (LDH), creatinine, and platelet count (PLT), has been introduced as a promising biomarker for endothelial injury. The formula of EASIX is LDH (U/L) × creatinine (mg/dL)/PLT (10⁹/L). EASIX has been employed to evaluate the status and function of endothelial cells. This index shows associations with mortality risk in critically ill and cancer patients (9-11). Current research demonstrates significant correlations between EASIX and endothelial activation markers, including IL-18, C-X-C motif chemokine 8, insulin-like growth factor-1, and pro-tumorigenic factor-2 (12,13). This supports the use of EASIX in assessing endothelial damage (14). Nonetheless, the relationship between EASIX and in-hospital mortality (IHM) in individuals with TLI remains under-investigated. This retrospective study explores the association between EASIX and the prognosis of TLI. The aim is to provide a basis for early identification of high-risk individuals, timely intervention, and improved outcomes. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0030/rc) (15).

Methods

Data source

Clinical data on individuals with TLI were evaluated. The data were acquired from the Medical Information Mart for Intensive Care (MIMIC)-IV-3.1 database, a publicly accessible and free resource (16). One author (H.W.) completed the requisite training and obtained access (record ID: 73445735). The MIMIC project received approval from the Institutional Review Boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. Due to the anonymized nature of the data, ethical exemption was granted, and there was no need for further approval. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Eligibility criteria

Inclusion criteria: Individuals diagnosed with TLI per International Classification of Diseases (ICD)-9 or ICD-10 codes, admitted to the intensive care unit (ICU) for the first time. Exclusion criteria: (I) individuals aged under 18 years; (II) individuals not admitted to the ICU; (III) cases with missing follow-up data; (IV) individuals with absent creatinine, LDH, or PLT.

Clinical data collection

Relevant medical information was collected from the MIMIC-IV database using structured query language (SQL) to obtain the following data: (I) demographic information: age, gender, body mass index (BMI), ethnicity; (II) vital signs: systolic blood pressure (SBP), diastolic blood pressure (DBP); (III) comorbidities, identified based on ICD-9 or ICD-10 codes, including acute kidney injury (AKI), ARDS, chronic kidney disease (CKD), hypovolemic shock (HS), hypertension (HP), septic shock (SS), lung surgery; (IV) laboratory parameters: white blood cells (WBCs), hemoglobin (HGB), hematocrit (HCT), PLT, prothrombin time (PT), partial thromboplastin time (PTT), lactate, albumin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), calcium, creatinine, creatine kinase-x (CK-x), creatine kinase-y (CK-y), total bilirubin (TB), blood glucose, urea nitrogen, LDH, fibrinogen; (V) ICU scoring systems: Simplified Acute Physiology Score II (SAPS II), Sepsis-related Organ Failure Assessment (SOFA), systemic inflammatory response syndrome (SIRS). For repeatedly measured data, the first measurement within 72 hours of initial admission was analyzed.

Handling of missing data: To avoid bias stemming from the exclusion of samples, multiple imputation using random forests (RFs) was implemented for variables with <20% missing values. For those with >20% missing values, the missing values were categorized and incorporated into the analysis as dummy variables. Albumin, BMI, CK, DBP, fibrinogen, and SBP had >20% missing values.

Exposure and outcome

The EASIX was computed as LDH (U/L) × creatinine (mg/dL)/PLT (10⁹/L) (9). Since EASIX displayed a skewed distribution, the log₂ transformation was applied for standardization. Log₂-transformed EASIX values were then employed to arrange individuals into tertiles for analysis. The primary outcome was the risk of 30-day IHM in individuals with TLI, with a follow-up duration starting from admission until death after discharge.

Confounder selection

The study cohort was partitioned into training (70%) and validation (30%) sets. The training cohort was employed for feature selection and model training, while the validation cohort was used to validate model performance. The Boruta algorithm was applied for selecting variables. Boruta is an RF-based algorithm for feature selection. It systematically evaluates the importance of each original feature by constructing shadow features. This algorithm categorizes each feature as ‘Important’, ‘Rejected’, or ‘Tentative’ by comparing the original features to randomly permuted shadow features. Thus, outcome-associated variables are identified.

Statistical analysis

Categorical variables were summarized as numbers (percentages) and assessed via Pearson’s chi-squared or Fisher’s exact test. Continuous variables with normal distribution were reported as mean [standard deviation (SD)] and compared using Student’s t-test or analysis of variance (ANOVA); non-normally distributed variables were expressed as median [interquartile range (IQR)] and analyzed using the Mann-Whitney U or Kruskal-Wallis tests.

The Cox proportional hazards model was employed to investigate the association between EASIX and the risk of 30-day IHM in individuals with TLI, with stepwise adjustment for confounders. Model 1 was unadjusted. Model 2 included adjustments for gender, race, and age. Model 3 incorporated adjustments for age, race, gender, and variables selected by the Boruta algorithm. Variance inflation factors (VIFs) were then calculated to quantify multicollinearity; variables with VIFs >5 were removed. Restricted cubic splines (RCS) were used to examine possible nonlinear relationships between EASIX and outcomes. Subgroup analyses were implemented to assess consistency across diverse patient groups. An RF model was constructed using variables from Model 3. Optimal RF hyperparameters were identified via random search with 10-fold cross-validation. The final model was trained using these parameters, and Shapley Additive exPlanations (SHAP) analysis was conducted to enhance interpretability. Receiver operating characteristic (ROC) curves were generated to validate the performance of the model. All analyses were performed in R (v4.4.1) with a significance level of α=0.05.

Results

Study population

Following the eligibility criteria, 231 individuals were enrolled. The selection of subjects is illustrated in Figure 1.

Figure 1.

Subject selection flowchart. ICD, International Classification of Diseases; LDH, lactate dehydrogenase.

The overall cohort comprised 166 males and 65 females, with a median age of 53 years. Non-white individuals accounted for 118 cases. Table 1 presents a comparison based on IHM. Among patients who died during hospitalization, we observed shorter hospital stays, a higher proportion of males, a greater representation of white individuals, as well as elevated levels of markers, including DBP, SBP, CK-y, glucose, lactate, PTT, creatinine, and LDH. SS occurred more frequently, and SAPS II and SOFA scores were higher (Table 1).

Table 1. Baseline characteristics stratified by IHM.

Characteristic	Overall (N=231)	Alive (N=183)	Dead (N=48)	P value†
Hospital stays (days)	11 [5–22]	12 [7–24]	4 [2–12]	<0.001
Age (years)	53 [29–69]	50 [28–65]	61 [38–81]	0.005
Gender				0.007
Female	65 [28]	44 [24]	21 [44]
Male	166 [72]	139 [76]	27 [56]
Race				0.15
Non-White	118 [51]	89 [49]	29 [60]
White	113 [49]	94 [51]	19 [40]
SBP, mmHg				<0.001
<90	1 [0.4]	1 [0.5]	0 [0]
>120	59 [26]	54 [30]	5 [10]
90–120	54 [23]	53 [29]	1 [2.1]
Missing	117 [51]	75 [41]	42 [88]
DBP, mmHg				<0.001
<60	12 [5.2]	12 [6.6]	0 [0]
>90	10 [4.3]	10 [5.5]	0 [0]
60–90	92 [40]	86 [47]	6 [13]
Missing	117 [51]	75 [41]	42 [88]
BMI, kg/m2				0.003
<30	53 [23]	49 [27]	4 [8.3]
≥30	22 [9.5]	20 [11]	2 [4.2]
Missing	156 [68]	114 [62]	42 [88]
AKI	46 [20]	35 [19]	11 [23]	0.56
ARDS	6 [2.6]	4 [2.2]	2 [4.2]	0.61
CKD	11 [4.8]	7 [3.8]	4 [8.3]	0.25
HP	58 [25]	47 [26]	11 [23]	0.69
Lung surgery	27 [12]	21 [11]	6 [13]	0.84
Hypovolemic shock	5 [2.2]	2 [1.1]	3 [6.3]	0.06
Septic shock	16 [6.9]	5 [2.7]	11 [23]	<0.001
Albumin, g/dL				0.04
<3.5	112 [48]	82 [45]	30 [63]
3.5–5.2	44 [19]	40 [22]	4 [8.3]
Missing	75 [32]	61 [33]	14 [29]
ALT, U/L	60 [28–161]	57 [25–156]	70 [34–183]	0.32
AST, U/L	94 [45–258]	87 [42–251]	128 [63–427]	0.07
Calcium, mmol/L	8.10 [7.50–8.60]	8.10 [7.60–8.60]	7.75 [7.20–8.70]	0.14
CK-x, IU/L				0.76
<40	2 [0.9]	2 [1.1]	0 [0]
>200	109 [47]	83 [45]	26 [54]
40–200	16 [6.9]	13 [7.1]	3 [6.3]
Missing	104 [45]	85 [46]	19 [40]
CK-y, IU/L				<0.001
<40	76 [33]	61 [33]	15 [31]
>200	2 [0.9]	0	2 [4.2]
40–200	7 [3.0]	2 [1.1]	5 [10]
Missing	146 [63]	120 [66]	26 [54]
Fibrinogen, g/L				0.052
<200	60 [26]	40 [22]	20 [42]
>400	11 [4.8]	10 [5.5]	1 [2.1]
200–400	26 [11]	21 [11]	5 [10]
Missing	134 [58]	112 [61]	22 [46]
Glucose, mg/dL	138 [116–176]	133 [115–161]	171 [138–228]	<0.001
Hematocrit, %	35 [30–39]	35 [30–39]	33 [27–39]	0.09
Hemoglobin, g/dL	11.40 [9.70–12.90]	11.60 [10.00–13.00]	10.70 [8.75–12.35]	0.03
Lactate, mmol/L	2.70 [1.60–4.50]	2.50 [1.50–3.90]	4.75 [2.80–7.35]	<0.001
PT, s	13.8 [12.3–16.5]	13.7 [12.0–15.8]	15.3 [13.3–19.0]	0.002
PTT, %	30 [26–36]	29 [26–34]	35 [29–49]	<0.001
Total bilirubin, mg/L	0.60 [0.40–1.00]	0.60 [0.40–1.00]	0.55 [0.40–1.10]	>0.99
Urea, mg/dL	15 [11–20]	15 [11–18]	17 [13–24]	0.03
WBC, 109/L	13 [9–17]	13 [10–17]	13 [9–18]	0.56
SAPS II	32 [23–43]	30 [22–39]	45 [35–58]	<0.001
SIRS				0.26
0	2 [0.9]	2 [1.1]	0 [0]
1	13 [5.6]	12 [6.6]	1 [2.1]
2	54 [23]	43 [23]	11 [23]
3	81 [35]	68 [37]	13 [27]
4	81 [35]	58 [32]	23 [48]
SOFA	5.0 [2.0–8.0]	4.0 [2.0–7.0]	8.0 [6.0–12.0]	<0.001
Creatinine, mg/dL	0.90 [0.70–1.20]	0.90 [0.70–1.10]	1.15 [0.85–1.45]	<0.001
LDH, IU/L	392 [277–659]	373 [267–589]	541 [338–1,155]	<0.001
Platelets, 109/L	180 [128–237]	183 [137–239]	158 [115–214]	0.03

Data are presented as median [interquartile range] or n [%]. ^†, Wilcoxon rank sum test, Pearson’s chi-squared test, or Fisher’s exact test. AKI, acute kidney injury; ALT, alanine aminotransferase; ARDS, acute respiratory distress syndrome; AST, aspartate aminotransferase; BMI, body mass index; CK, creatine kinase; CKD, chronic kidney disease; DBP, diastolic blood pressure; HP, hypertension; IHM, in-hospital mortality; LDH, lactate dehydrogenase; PT, prothrombin time; PTT, partial thromboplastin time; SAPS, Simplified Acute Physiology Score; SBP, systolic blood pressure; SIRS, systemic inflammatory response syndrome; SOFA, Sepsis-related Organ Failure Assessment; WBC, white blood cell.

Table 2 presents baseline characteristics stratified by tertiles of EASIX. Higher EASIX groups (T3) showed increased IHM rates, a greater proportion of males, a higher proportion of non-white individuals, and levels of platelet and fibrinogen lower than the normal range. Elevated levels of ALT, AST, PT, PTT, creatinine, lactate, and LDH were noted, along with higher SAPS II and SOFA scores.

Table 2. Baseline characteristics stratified by EASIX tertiles.

Characteristic	Overall (N=231)	T1 (N=77)	T2 (N=77)	T3 (N=77)	P value†
Dead	48 [21]	8 [10]	12 [16]	28 [36]	<0.001
Hospital stays (days)	11 [5–22]	10 [6–22]	10 [5–17]	12 [4–24]	0.576
Age (years)	53 [29–69]	57 [35–75]	47 [26–66]	50 [29–65]	0.087
Gender					0.757
Female	65 [28]	21 [27]	20 [26]	24 [31]
Male	166 [72]	56 [73]	57 [74]	53 [69]
Race					0.151
Non-White	118 [51]	33 [43]	40 [52]	45 [58]
White	113 [49]	44 [57]	37 [48]	32 [42]
SBP, mmHg					0.394
<90	1 [0.4]	0	1 [1.3]	0
>120	59 [26]	19 [25]	22 [29]	18 [23]
90–120	54 [23]	23 [30]	17 [22]	14 [18]
Missing	117 [51]	35 [45]	37 [48]	45 [58]
DBP, mmHg					0.316
<60	12 [5.2]	7 [9.1]	2 [2.6]	3 [3.9]
>90	10 [4.3]	2 [2.6]	4 [5.2]	4 [5.2]
60–90	92 [40]	33 [43]	34 [44]	25 [32]
Missing	117 [51]	35 [45]	37 [48]	45 [58]
BMI, kg/m2					0.946
<30	53 [23]	19 [25]	18 [23]	16 [21]
≥30	22 [9.5]	8 [10]	6 [7.8]	8 [10]
Missing	156 [68]	50 [65]	53 [69]	53 [69]
AKI	46 [20]	11 [14]	12 [16]	23 [30]	0.027
ARDS	6 [2.6]	0	0	6 [7.8]	0.004
CKD	11 [4.8]	1 [1.3]	6 [7.8]	4 [5.2]	0.195
HP	58 [25]	23 [30]	18 [23]	17 [22]	0.490
Lung surgery	27 [12]	6 [7.8]	12 [16]	9 [12]	0.322
Hypovolemic shock	5 [2.2]	2 [2.6]	0	3 [3.9]	0.375
Septic shock	16 [6.9]	4 [5.2]	4 [5.2]	8 [10]	0.341
Albumin, g/dL					0.019
<3.5	112 [48]	30 [39]	33 [43]	49 [64]
3.5–5.2	44 [19]	18 [23]	18 [23]	8 [10]
Missing	75 [32]	29 [38]	26 [34]	20 [26]
ALT, U/L	60 [28–161]	29 [16–57]	60 [33–139]	133 [66–326]	<0.001
AST, U/L	94 [45–258]	46 [28–84]	83 [53–207]	261 [130–611]	<0.001
Calcium, mmol/L	8.10 [7.50–8.60]	8.20 [7.70–8.70]	8.10 [7.50–8.50]	7.90 [7.30–8.50]	0.242
CK-x, IU/L					0.005
<40	2 [0.9]	0	2 [2.6]	0
>200	109 [47]	25 [32]	39 [51]	45 [58]
40–200	16 [6.9]	10 [13]	3 [3.9]	3 [3.9]
Missing	104 [45]	42 [55]	33 [43]	29 [38]
CK-y, IU/L					0.293
<40	76 [33]	27 [35]	26 [34]	23 [30]
>200	2 [0.9]	0	0	2 [2.6]
40–200	7 [3.0]	1 [1.3]	1 [1.3]	5 [6.5]
Missing	146 [63]	49 [64]	50 [65]	47 [61]
Fibrinogen, g/L					<0.001
<200	60 [26]	8 [10]	19 [25]	33 [43]
>400	11 [4.8]	4 [5.2]	3 [3.9]	4 [5.2]
200–400	26 [11]	6 [7.8]	7 [9.1]	13 [17]
Missing	134 [58]	59 [77]	48 [62]	27 [35]
Glucose, mg/dL	138 [116–176]	134 [115–153]	139 [115–174]	151 [121–216]	0.028
Hematocrit, %	35 [30–39]	34 [31–39]	36 [30–38]	34 [27–40]	0.605
Hemoglobin, g/dL	11.40 [9.70–12.90]	11.50 [10.20–12.80]	11.60 [9.90–12.60]	11.20 [8.90–13.20]	0.401
Lactate, mmol/L	2.70 [1.60–4.50]	1.90 [1.30–2.80]	2.70 [1.80–4.00]	4.20 [2.60–6.20]	<0.001
PT, s	13.8 [12.3–16.5]	13.3 [12.2–14.3]	13.7 [11.9–15.8]	15.8 [13.7–19.1]	<0.001
PTT, %	30 [26–36]	28 [26–31]	28 [26–33]	35 [31–59]	<0.001
Total bilirubin, mg/L	0.60 [0.40–1.00]	0.60 [0.30–0.80]	0.60 [0.40–0.90]	0.70 [0.40–1.20]	0.123
Urea, mg/dL	15 [11–20]	14 [10–17]	16 [12–21]	17 [13–22]	0.001
WBC, 109/L	13 [9–17]	14 [10–18]	13 [10–16]	12 [9–17]	0.286
SAPS II	32 [23–43]	31 [22–40]	29 [20–39]	37 [29–49]	<0.001
SIRS					0.009
0	2 [0.9]	0	1 [1.3]	1 [1.3]
1	13 [5.6]	7 [9.1]	5 [6.5]	1 [1.3]
2	54 [23]	24 [31]	16 [21]	14 [18]
3	81 [35]	26 [34]	33 [43]	22 [29]
4	81 [35]	20 [26]	22 [29]	39 [51]
SOFA	5.0 [2.0–8.0]	3.0 [1.0–5.0]	4.0 [2.0–6.0]	8.0 [6.0–11.0]	<0.001
Creatinine, mg/dL	0.90 [0.70–1.20]	0.80 [0.60–0.90]	1.00 [0.80–1.10]	1.20 [0.90–1.50]	<0.001
LDH, IU/L	392 [277–659]	263 [217–315]	392 [316–532]	779 [561–1,205]	<0.001
Platelets, 109/L	180 [128–237]	231 [192–266]	169 [125–219]	140 [95–182]	<0.001

Data are presented as median [interquartile range] or n [%]. ^†, Wilcoxon rank sum test, Pearson’s chi-squared test, or Fisher’s exact test. AKI, acute kidney injury; ALT, alanine aminotransferase; ARDS, acute respiratory distress syndrome; AST, aspartate aminotransferase; BMI, body mass index; CK, creatine kinase; CKD, chronic kidney disease; DBP, diastolic blood pressure; EASIX, endothelial activation and stress index; HP, hypertension; LDH, lactate dehydrogenase; PT, prothrombin time; PTT, partial thromboplastin time; SAPS, Simplified Acute Physiology Score; SBP, systolic blood pressure; SIRS, systemic inflammatory response syndrome; SOFA, Sepsis-related Organ Failure Assessment; WBC, white blood cell.

Feature selection results

Based on the Boruta analysis, SS, calcium, glucose, PTT, and log₂-EASIX were identified as important variables, PT and BMI were identified as tentative, and the remaining variables were identified as unimportant. Those categorized as ‘Important’ and ‘Tentative’ were included in the current analysis (Figure 2).

Figure 2.

Boruta feature selection results. The Boruta algorithm categorizes variables as ‘Important’, ‘Rejected’, or ‘Tentative’. The y-axis indicates feature importance. Red denotes ‘Rejected’ features, green signifies ‘Important’ features, and blue denotes tentative features. AKI, acute kidney injury; ARDS, acute respiratory distress syndrome; BMI, body mass index; CK, creatine kinase; CKD, chronic kidney disease; EASIX, endothelial activation and stress index; HP, hypertension; PT, prothrombin time; PTT, partial thromboplastin time; SIRS, systemic inflammatory response syndrome; WBC, white blood cell.

Association between EASIX and risk of 30-day IHM in individuals with TLI

When EASIX was analyzed as a continuous variable, increasing levels of EASIX were associated with a significantly elevated risk of IHM in individuals with TLI. This significance persisted after adjusting for all confounders (Table 3).

Table 3. Cox proportional HR.

Categories	Model 1			Model 2			Model 3
	HR (95% CI)	P value		HR (95% CI)	P value		HR (95% CI)	P value
EASIX (continuous)	1.615 (1.311–1.988)	<0.001		1.646 (1.327–2.042)	<0.001		1.519 (1.177–1.962)	0.001
EASIX† (categorical)
T1 (N=77)	Ref.			Ref.			Ref.
T2 (N=77)	1.635 (0.668–4.001)	0.282		1.897 (0.772–4.664)	0.163		1.628 (0.653–4.060)	0.296
T3 (N=77)	3.707 (1.688–8.141)	<0.001		4.131 (1.866–9.141)	<0.001		2.895 (1.245–6.735)	0.014
P of trend	2.526 (1.448–4.405)	0.001		2.726 (1.555–4.781)	<0.001		2.121 (1.167–3.852)	0.014

Model 1 had no adjustments. Model 2 incorporated adjustments for sex, age, and race. Model 3 comprised adjustments for the variables in Model 2 and further adjusted for BMI, SS, calcium, glucose, PT, and PTT. ^†, log₂-transformed EASIX values: T1, 0.356–1.307; T2, 1.307–2.274; T3, 2.274–5.89. BMI, body mass index; CI, confidence interval; EASIX, endothelial activation and stress index; HR, hazard ratio; PT, prothrombin time; PTT, partial thromboplastin time; SS, septic shock.

Compared to individuals with lower levels of EASIX (T1), those with higher levels (T3) faced a significantly increased risk of IHM. This trend was consistent across Model 1 [hazard ratio (HR) =3.707, 95% confidence interval (CI): 1.688–8.141, P<0.001], Model 2 (HR =4.131, 95% CI: 1.866–9.141, P<0.001), and Model 3 (HR =2.895, 95% CI: 1.245–6.735, P=0.01), as detailed in Table 3.

RCS analysis revealed an approximately linear increase in the risk of 30-day IHM with rising levels of EASIX in individuals with TLI (Model 1: P=0.97 for linearity; Model 2: P=0.91 for linearity; Model 3: P=0.93 for linearity). This association demonstrated consistency across the three models (Figure 3).

Figure 3.

RCS plots of log2EASIX. (A) Model 1; (B) Model 2; (C) Model 3. Model 1 had no adjustments. Model 2 incorporated adjustments for sex, age, and race. Model 3 comprised adjustments for the variables in Model 2 and further adjusted for BMI, SS, calcium, glucose, PT, and PTT. BMI, body mass index; CI, confidence interval; EASIX, endothelial activation and stress index; HR, hazard ratio; PT, prothrombin time; PTT, partial thromboplastin time; RCS, restricted cubic spline; SS, septic shock.

Subgroup analyses based on age, gender, race, HP, CKD, and SS indicated a significant positive association between EASIX and the risk of 30-day IHM in individuals aged ≤60 years (HR =1.678, 95% CI: 1.253–2.246), males (HR =1.622, 95% CI: 1.220–2.157), females (HR =1.533, 95% CI: 1.129–2.080), non-white individuals (HR =1.493, 95% CI: 1.149–1.940), those without HP (HR =1.637, 95% CI: 1.300–2.062), those without CKD (HR =1.545, 95% CI: 1.240–1.924), and those without SS (HR =1.589, 95% CI: 1.234–2.047). Furthermore, no interactions between EASIX and other factors were observed (Figure 4).

Figure 4.

Subgroup analysis. CI, confidence interval; CKD, chronic kidney disease; HP, hypertension; HR, hazard ratio.

Validation of predictive performance for EASIX and the RF model

ROC analysis was conducted to evaluate the predictive performance of EASIX alone. The results demonstrated that the AUC values of EASIX for predicting 30-day IHM in patients with TLI were 0.709 (95% CI: 0.621–0.797) in the overall population, 0.751 (95% CI: 0.649–0.853) in the training cohort, and 0.617 (95% CI: 0.452–0.783) in the validation cohort.

Using 10-fold cross-validation with 500 iterations, optimal parameters were selected via a random search method. The final parameter set (ntree =399; mtry =3; maxnodes =9; nodesize =5) was incorporated into the model for construction. Subsequent ROC analysis of the RF model revealed an AUC of 0.973 (95% CI: 0.950–0.995) and 0.753 (95% CI: 0.593–0.913) in the training and validation sets, respectively (Figure 5).

Figure 5.

ROC curves: (A) ROC curve for log2EASIX; (B) ROC curve for the RF model. AUC, area under the curve; CI, confidence interval; EASIX, endothelial activation and stress index; RF, random forest; ROC, receiver operating characteristic.

Interpretability of the RF model was enhanced using the SHAP method to evaluate the contribution of each feature to the prediction outcome. The results indicated that BMI, age, calcium, glucose, PTT, and EASIX exhibited positive effects on the outcome, and higher EASIX values were associated with an increased risk of mortality as predicted by the model. Feature importance analysis identified EASIX as the most influential variable, followed by PTT and glucose (Figure 6).

Figure 6.

SHAP analysis. (A) SHAP value plot; (B) SHAP feature importance plot. BMI, body mass index; EASIX, endothelial activation and stress index; PTT, partial thromboplastin time; SHAP, Shapley Additive exPlanations.

Discussion

This is the first retrospective investigation utilizing the MIMIC-IV database to examine the association between EASIX and outcomes in patients with TLI. The analysis demonstrated a significant independent correlation between elevated EASIX and an increased risk of 30-day IHM among patients with TLI, and the findings were consistent across all subgroups. The study further revealed a near-linear positive relationship between EASIX and the risk of 30-day IHM, which was robust among all subgroups. Unlike traditional scoring systems that mainly capture acute physiological disturbances, EASIX offers a more targeted assessment by integrating LDH, creatinine, and PLT to quantify endothelial injury and coagulation abnormalities. As a result, it captures microcirculatory dysfunction more accurately throughout the disease course of TLI. The RF model incorporating EASIX exhibited strong predictive performance, and SHAP analysis further identified EASIX as the most important predictive feature. These findings suggest potential clinical utility: EASIX, as a low-cost tool, may assist clinicians in the early identification of high-risk TLI patients upon admission, facilitating more intensive monitoring or timely intervention.

Endothelial cells form the inner lining of blood vessels and are particularly abundant in the lungs, accounting for approximately half of all pulmonary cells (17-19). This high density of endothelial cells provides an extensive surface area for gas exchange, facilitating efficient oxygen delivery to tissues (20,21). Beyond gas exchange, these cells play critical roles in pathogen defense (19,20,22) and inflammatory responses, thereby maintaining pulmonary homeostasis (23,24). In individuals with TLI, trauma-induced stress disrupts endothelial integrity, increasing vascular permeability and leading to fluid leakage into tissues. This process can cause pulmonary edema and severe respiratory distress (25-28), further worsening clinical outcomes. Currently, no routine clinical markers are available to assess the function of endothelial cells, and practical indicators are urgently needed. Recently, Luft et al. proposed the EASIX as a novel marker for systemic endothelial activation in allogeneic stem cell transplantation (14). EASIX incorporates routinely measured laboratory values—LDH, creatinine, and PLT (29)—all of which are components of thrombotic microangiopathy diagnostic criteria and reflect endothelial pathophysiology (14,30,31). EASIX offers a simple yet accurate prognostic tool for critically ill patients, including those with TLI. Although few studies have focused on predicting outcomes in TLI, an Italian team recently combined levels of plasma cytokines with chest computed tomography (CT) to assess lung injury in patients with COVID-19 (32). However, imaging-based approaches may lag behind clinical changes and lack the practicality of EASIX.

Machine learning models can capture nonlinear relationships and interactions, offering strong predictive performance. Nonetheless, their ‘black-box’ nature often limits interpretability. SHAP analysis partially addresses this issue by explaining feature contributions. In the current study, SHAP results aligned with analyses of associations and identified EASIX as the most important predictor.

Subgroup analysis revealed that EASIX was significantly positively associated with the risk of 30-day IHM in TLI patients without HP, CKD, or SS. This may be explained by the fact that post-traumatic stress response can lead to sympathetic activation, increased release of catecholamine, and insulin resistance (2), generating substantial reactive oxygen species and triggering oxidative stress and inflammatory responses. These processes rapidly induce endothelial dysfunction, ultimately resulting in elevated EASIX and increased 30-day IHM. In contrast, among patients with HP or CKD, endothelial injury tends to be chronic and prolonged due to persistently elevated blood pressure, rather than manifesting as acute dysfunction. Consequently, these patients may not exhibit markedly elevated EASIX in the short term.

Our study defines the role of EASIX in ICU-admitted individuals with TLI, but several limitations should be noted. First, this is a retrospective analysis. Although reverse causality was minimized through study design, the causality cannot be definitively established. Prospective longitudinal cohorts are needed for validation. Second, data were sourced solely from the MIMIC-IV database, which lacks details on TLI-specific factors such as hemorrhage volume, pulmonary function, and intraoperative conditions. These unmeasured confounders were not adjusted for. Finally, the study population was predominantly White, and the generalizability to other ethnic groups requires further investigation.

Conclusions

Elevated EASIX is significantly associated with an increased risk of 30-day IHM in individuals with TLI. EASIX may serve as a practical clinical tool for assessing prognosis in this population.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.