Trajectory-Defined Thrombo-Inflammatory Phenotypes Predict 30-Day ICU Mortality in Post-cardiac Arrest Syndrome: A Multicenter Retrospective Longitudinal Cohort Study

Abstract

Current risk stratification for Post-cardiac Arrest Syndrome (PCAS) relies mainly on static admission variables and may fail to capture the dynamic systemic evolution. This study aimed to identify trajectory-defined thrombo-inflammatory phenotypes in PCAS using longitudinal trajectories of platelets, white blood cells (WBC), hemoglobin, and body temperature, and to evaluate their association with 30-day ICU mortality. We conducted a multicenter retrospective cohort study using the MIMIC-IV, MIMIC-III, and eICU-CRD databases, including adult patients with ICU stays of 2-90 days after cardiac arrest. A Multivariate Process Joint Latent Class Mixed Model (mJLCMM) identified latent classes from 30-day biomarker trajectories. The primary outcome was 30-day ICU mortality. Associations were evaluated using Inverse Probability Weighting (IPW) and Doubly Robust Estimation (DRE). Prognostic accuracy was compared against SOFA and OASIS scores using time-dependent Receiver Operating Characteristic (ROC) analysis. A total of 5,099 patients were included. Two phenotypes were identified: Class 1 (“Rapid Decline and Recovery”) and Class 2 (“Mild Decline and Recovery”). Class 1 was associated with higher 30-day ICU mortality (eICU: 46.7%; MIMIC: 24.6%). In doubly robust analyses, the class 2 remained associated with lower ICU mortality in both cohorts, with odds ratios (ORs) of 0.82 (95% CI, 0.72-0.96) in eICU and 0.74 (95% CI, 0.55-0.95) in MIMIC. By Day 30, the trajectory model outperformed SOFA and OASIS, with an AUC of 0.74 versus 0.54 and 0.59, respectively. This trajectory-based classification showed superior prognostic performance for 30-day ICU mortality and highlights the potential value of dynamic monitoring in post-cardiac arrest management.

Keywords

cardiac arrest post-cardiac arrest syndrome intensive care units thrombo-inflammation trajectory analysis longitudinal studies prognosis platelets count

Introduction

Sudden cardiac arrest (SCA) remains a significant global public health crisis and a leading cause of death worldwide. Despite advancements in resuscitation, survival rates for out-of-hospital cardiac arrest (OHCA) remain low, plateauing near 10%.¹ Approximately 350,000 individuals in the United States alone suffer from OHCA annually.² Although only a minority of patients achieve return of spontaneous circulation (ROSC), successful resuscitation at hospital admission represents a critical determinant of survival.³ However, ROSC does not guarantee favorable neurological outcomes. Many patients experience post-resuscitation complications, including re-arrest, multi-organ dysfunction, and neurological deterioration.⁴ Consequently, mortality rates among resuscitated patients admitted to the intensive care unit (ICU) remain substantially high.⁵

Among the multifactorial mechanisms driving multiple organ dysfunction in Post-Cardiac Arrest Syndrome (PCAS), a key driver is a “sepsis-like” systemic inflammatory response. This phenomenon is increasingly recognized as “thrombo-inflammation”—a pathological collision between the immune system and the hemostatic machinery.⁶ Following global ischemia, reperfusion-induced endothelial activation and glycocalyx shedding expose the subendothelial matrix, catalyzing platelet aggregation and leukocyte recruitment that culminate in microvascular thrombosis and disseminated intravascular coagulation (DIC).⁷ As a conceptual background for our analysis, platelets and white blood cells (WBCs) are understood to serve as active effectors within this cascade. Although these specific cellular interactions were not directly measured in the current study, biologically plausible models suggest that activated platelets recruit neutrophils to the injured endothelium, promoting Neutrophil Extracellular Trap (NET) formation—a process that, in theory, propagates thrombosis and tissue injury.⁸ While modern prognostication for PCAS typically employs a multimodal approach—integrating neurological examination, electrophysiology, and neuroimaging⁹ —the laboratory component often relies on static biomarkers obtained at admission¹⁰ (e.g., initial lactate, pH) or snapshot severity scores (e.g., SOFA, APACHE).¹¹ These static measures fail to capture the decline in physiological deterioration or the trajectory of recovery, often leading to inaccurate risk stratification.^12,13 While traditional statistical methods, such as logistic regression or linear mixed models, are robust for many clinical applications, analyzing longitudinal ICU data presents a specific challenge known as ‘informative dropout'—the bias introduced when the most severely ill patients die early and cease to contribute data to the longitudinal record.^14,15 The Multivariate Process Joint Latent Class Mixed Model (mJLCMM) specifically addresses this limitation by jointly modeling longitudinal biomarkers and the survival outcome, enabling more accurate estimation of dynamic physiological trajectories.¹⁶

In this study, we leveraged longitudinal data from multiple multicenter critical care databases, including two versions of the Medical Information Mart for Intensive Care (MIMIC-III and MIMIC-IV version 3.1) and the eICU Collaborative Research Database (eICU-CRD version 2.0). We hypothesized that specific joint trajectories of thrombo-inflammatory markers (platelets, WBCs, hemoglobin) and body temperature during the early post-resuscitation phase would identify distinct trajectory-defined subgroups associated with divergent survival outcomes. By applying mJLCMM, we aimed to identify these model-derived latent classes to better understand disease heterogeneity, thereby providing a conceptual foundation for future prospective studies exploring early prognostication and personalized management strategies for patients with PCAS.

Method

Data Source

This multicenter retrospective cohort study, utilizing data from the MIMIC-III, MIMIC-IV (version 3.1), and eICU-CRD (v2.0) databases, was conducted and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines.¹⁷ MIMIC-III contains records from over 23,000 patients admitted to the Intensive Care Unit (ICU) at Beth Israel Deaconess Medical Center (BIDMC) between 2001 and 2008.¹⁸ MIMIC-IV v3.1, the subsequent version, includes comprehensive medical records for all patients admitted to the ICU or emergency department at BIDMC from 2008 to 2022.¹⁹ Additionally, the eICU-CRD v2.0 comprises records from over 200,000 ICU patients across multiple centers during 2014 and 2015.²⁰

The use of the MIMIC database was approved by the Institutional Review Board (IRB) of Beth Israel Deaconess Medical Center (BIDMC), which granted a waiver of informed consent. Data collection and release were additionally approved by the Massachusetts Institute of Technology (MIT) Committee on the Use of Humans as Experimental Subjects (Protocol No. 0403000206). All data were deidentified in compliance with the Health Insurance Portability and Accountability Act (HIPAA). As this study involved secondary analysis of publicly available, anonymized data, additional ethical approval was not required.

Participants

The study population comprised adults (≥18 years) extracted from the MIMIC-IV v3.1, MIMIC-III, and eICU-CRD v2.0 databases. Eligibility was determined based on the following criteria. 1) Patients who experienced cardiac arrest. 2) ICU length of stay between 2 and 90 days.

The following exclusion criteria were applied: (1) patients with multiple ICU admissions (only the first admission was included); (2) age <18 years; (3) ICU length of stay <24 hours or >90 days; (4) missing any baseline measurement within the first 24 hours of ICU admission (vital signs: heart rate, respiration, systolic/diastolic blood pressure, mean arterial pressure, temperature; laboratory values: white blood cell count, hemoglobin, platelet count; (5) having fewer than two measurements of vital signs, white blood cell count, or platelet count during the 30-day post-admission period.

Data Extraction, Handling Missing Data

Clinical data were extracted from the databases using Structured Query Language (SQL), encompassing five categories of information: (1) demographics, laboratory parameters, and severity scores obtained within the first 24 hours of intensive care unit (ICU) admission, including age, sex, serum calcium, alanine aminotransferase (ALT), aspartate aminotransferase (AST), pH, fraction of inspired oxygen (FiO₂), arterial partial pressure of oxygen (PaO₂), arterial partial pressure of carbon dioxide (PaCO₂), bicarbonate (HCO₃^-), peripheral capillary oxygen saturation (SpO₂), anion gap, albumin, creatinine, urea nitrogen (BUN), glucose, hematocrit, hemoglobin, potassium, international normalized ratio (INR), prothrombin time (PT), Sequential Organ Failure Assessment (SOFA) score, and Oxford Acute Severity of Illness Score (OASIS); (2) longitudinal measurements collected over a 30-day period following ICU admission, including WBC, hemoglobin, body temperature, and platelet count; (3) ICU clinical details, including duration of ICU stay, ICU and hospital survival status, mechanical ventilation and renal replacement therapy within the first 24 hours, blood and platelet transfusion, and vasoactive agent administration; (4) comorbidities, including chronic renal disease, diabetes mellitus, history of myocardial infarction, congestive heart failure (CHF), peripheral vascular disease, chronic obstructive pulmonary disease (COPD), connective tissue disease, peptic ulcer disease, cerebrovascular disease, and severe liver disease; and (5) life support interventions administered within the first 24 hours of ICU admission, specifically mechanical ventilation and renal replacement therapy. For variables with multiple measurements within the same day, the value reflecting the most severe condition was selected for analysis.

Variables with more than 30% missing data were excluded from further analysis (Figure S1). For variables with ≤30% missing values, multiple imputation was performed using predictive mean matching (PMM) via the ‘mice’ package (version 3.16.0) in R.²¹

Outcome

The primary outcome was 30-day ICU mortality after ICU admission; the secondary outcomes comprised 30-day survival status and time to death within 30 days. Survival time was measured from ICU admission to death, with right-censoring at ICU discharge or 30 days for survivors.

Statistical Analysis

Step 1: Identification of Subpopulations

To identify subpopulations with distinct trajectories of platelets, WBC, hemoglobin, and body temperature, we employed a Multivariate Process Joint Latent Class Mixed Model (MPJLCMM). This model jointly analyzes longitudinal measurements of the four biomarkers while identifying latent classes with heterogeneous trajectory patterns. Nonlinear temporal trends were modeled using natural cubic splines with three internal knots placed at percentiles of the time distribution. The model structure was specified as follows: Time Scale and Fixed Effects: Time was defined continuously as days from ICU admission up to a maximum of 30 days. To capture nonlinear physiological dynamics, the temporal trends for the biomarkers were modeled using natural cubic splines with 3 degrees of freedom. Random Effects Structure: To account for intra-individual correlation and baseline heterogeneity, the models incorporated patient-specific random intercepts and random slopes for the spline terms. Class Membership Sub-model: The probability of latent class assignment was estimated without baseline covariates to ensure that the derived trajectory patterns were purely data-driven. The optimal number of latent classes was determined by evaluating multiple fit indices: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-adjusted BIC (SABIC), and entropy.²² Analyses were performed using the lcmm package (version 2.0.2) in R.

Step 2: Trajectory Visualization

We compared clinical characteristics and laboratory values between the identified latent classes using appropriate statistical tests (t-test/Mann-Whitney for continuous variables; chi-square/Fisher’s exact for categorical variables) (Table S2 and S3). Biomarker trajectories were plotted for each class, and 30-day survival was compared using Kaplan-Meier analysis with the log-rank test.

Step 3: Independent Association With Outcomes (Double Robust Estimation)

Variable selection employed a two-step approach. First, baseline variables with pairwise correlation coefficients ≥ 0.5 were excluded to address high intercorrelation (Figure S2 and S3). Second, from the remaining variables, those demonstrating clinical relevance and a univariate logistic association with the outcome (p < 0.05), while maintaining acceptable collinearity (variance inflation factor [VIF] < 5), were retained as potential confounders (Tables S6 and S7). Subsequently, propensity scores for latent class membership were estimated using gradient boosting machine models implemented in the R ‘twang’ package, with models iteratively optimized to achieve covariate balance.

Step 4: Sensitivity Analysis

To evaluate the robustness of our findings across different statistical methodologies, sensitivity analyses were performed using three distinct analytical approaches: (1) a fully adjusted multivariate logistic regression model was fitted, in which trajectory phenotype was entered as the main exposure and all selected baseline covariates were included directly in the outcome model (2) an inverse probability weighting (IPW) analysis was conducted using propensity scores for latent class membership estimated by gradient boosting machine models; these weights were used to generate a pseudo-population with improved balance of measured baseline covariates between the two phenotype groups, after which a weighted logistic regression model was fitted (3) a doubly robust estimation (DRE) approach that combined inverse probability weighting based on the propensity score with an additional adjusted outcome regression model, thereby addressing potential confounding from imbalanced covariates.

Step 5: Subgroup Analysis

A subgroup analysis was conducted on the cardiac arrest cohort derived from the MIMIC and eICU-CRD databases to evaluate the consistency of the association between two distinct trajectory patterns and cardiac arrest outcomes across various patient subgroups. Subgroups were stratified according to demographic and clinical characteristics, including age, comorbidities, platelet transfusion, red blood cell (RBC) transfusion, mechanical ventilation, renal replacement therapy, anticoagulation therapy, and antiplatelet therapy. Interaction effects between these factors and trajectory patterns were assessed by comparing two logistic regression models: a baseline model and an extended model incorporating interaction terms.

To assess the prognostic accuracy of the derived clinical trajectory phenotypes relative to established severity scores (SOFA and OASIS), we performed time-dependent Receiver Operating Characteristic (ROC) curve analyses. Time-dependent Area Under the Curve (AUC) values were estimated at 3-, 10-, and 30 days post-admission to capture dynamic changes in predictive utility over the clinical course. All time-dependent ROC analyses were conducted using the ‘timeROC’ package in R.

All statistical analyses were conducted using R software (version 4.0.5). Continuous variables are presented as mean ± standard deviation (SD) for normally distributed data or as median with interquartile range (IQR) for non-normally distributed data. Between-group comparisons for continuous variables were performed using Student's t-test for normally distributed data and the Mann-Whitney U test for non-normally distributed data. Categorical variables are expressed as frequencies and percentages and were analyzed using the chi-square test or Fisher’s exact test, as appropriate.

Results

Baseline Characteristics

A total of 3,776 patients from the eICU-CRD database were included in the analysis, of whom 1,477 (39.1%) died before hospital discharge; the MIMIC cohort included 1,323 patients (Figure 1). As illustrated in Table 1, Non-survivors were slightly older than survivors (median age: 66.0 vs. 64.0 years; P = 0.001) and had a significantly higher proportion of females (44.6% vs. 37.7%; P < 0.001). Significant metabolic and organ dysfunction were observed in the non-survival group. Liver enzymes were markedly elevated in non-survivors compared to survivors, including ALT (85.0 vs. 54.0 U/L; P < 0.001) and AST (135 vs. 75.0 U/L; P < 0.001). Metabolic disorder was also more pronounced in non-survivors, characterized by hyperglycemia (239 vs. 192 mg/dL; P < 0.001) and an increased anion gap (17.0 vs. 14.0 mmol/L; P < 0.001). Additionally, non-survivors exhibited worse renal function (creatinine: 1.75 vs. 1.36 mg/dL; P < 0.001; BUN: 30.0 vs. 25.0 mg/dL; P < 0.001) and coagulopathy, indicated by prolonged PT (16.1s vs. 15.1s, P < 0.001) and elevated INR (1.40 vs. 1.23) (P < 0.001 for both). Hemodynamic instability was apparent in non-survivors, who presented with higher heart rates (113 vs. 106 bpm; P < 0.001) and elevated blood pressures. Regarding interventions, non-survivors had higher rates of vasopressor use (59.5% vs. 42.8%; P < 0.001) and mechanical ventilation within the first 24 hours (97.9% vs. 89.6%; P < 0.001), but received anticoagulation (42.9% vs. 53.5%; P < 0.001) and platelet infusion less frequently (1.30% vs. 2.91%; P = 0.001). Among comorbidities, only congestive heart failure and COPD showed significant associations with ICU mortality. Consequently, disease severity scores were markedly higher in non-survivors, including OASIS (41.0 vs. 36.0; P < 0.001) and SOFA scores (10.0 vs. 8.00; P < 0.001). This distribution was similarly observed in the MIMIC database (Table S1).

Figure 1.

Flowchart of patient selection and study cohort derivation.

Table 1.

Clinical Baseline Characteristics of Patients in eICU-CRD

	Survival	Non-survival	P value
	N=2299	N=1477	P value
Class			<0.001
Rapid decline and recovery	701 (30.5%)	615 (41.6%)
Mild decline and recovery	1598 (69.5%)	862 (58.4%)
Age (years)	64.0 [53.0-74.0]	66.0 [55.0-76.0]	0.001
Gender			<0.001
Female	866 (37.7%)	659 (44.6%)
Male	1433 (62.3%)	818 (55.4%)
PT (s)	15.1 [13.2-18.8]	16.4 [13.8-21.9]	<0.001
INR (s)	1.23 [1.10-1.60]	1.40 [1.20-1.94]	<0.001
ALT (IU/L)	54.0 [28.0-134]	85.0 [37.0-230]	<0.001
AST (IU/L)	75.0 [37.0-191]	135 [57.0-386]	<0.001
FiO2 (%)	100 [60-100]	100 [80-100]	<0.001
Albumin (g/dl)	3.10 [2.70-3.60]	3.00 [2.50-3.40]	<0.001
Anion gap (mmol/L)	14.0 [10.0-18.0]	17.0 [13.0-21.0]	<0.001
pH	7.41 [7.35-7.47]	7.40 [7.34-7.47]	<0.001
Paco2 (mmHg)	46.0 [39.0-56.5]	49.8 [40.1-63.6]	<0.001
SpO2 (%)	92 [87-96]	91 [83-95]	0.032
PaO2 (mmHg)	197 [120-310]	218 [130-334]	<0.001
Respiratory rate (bpm)	27.0 [23.0-33.0]	29.0 [24.0-35.0]	<0.001
Systolic Blood Pressure (mmHg)	151 [134-172]	155 [135-178]	0.001
Diastolic Blood Pressure (mmHg)	92.0 [78.0-108]	94.0 [80.0-111]	0.001
Mean Arterial Pressure (mmHg)	111 [97.7-128]	115 [101-132]	<0.001
Heart rate (bpm)	106 [90.0-124]	113 [95.0-131]	<0.001
HCO3 (mmol/L)	25.0 [22.0-28.0]	24.0 [21.0-27.0]	<0.001
Temperature (°C)	37.3 [36.8-37.8]	37.1 [36.3-37.9]	<0.001
Calcium (mg/dl)	8.60 [8.10-9.10]	8.50 [8.10-9.10]	0.201
Glucose (mg/dl)	192 [144-257]	239 [172-319]	<0.001
Creatinine (mg/dl)	1.36 [0.98-2.28]	1.75 [1.20-2.70]	<0.001
Sodium (mmol/L)	140 [138-143]	142 [138-145]	<0.001
Blood urea nitrogen (mg/dl)	25.0 [17.0-40.0]	30.0 [21.0-46.0]	<0.001
Potassium (mmol/L)	4.40 [4.10-5.00]	4.70 [4.10-5.30]	<0.001
Hematocrit (%)	36.9 [30.7-42.2]	36.9 [31.8-42.8]	0.022
Hemoglobin (g/dl)	12.0 [10.0-14.0]	12.0 [10.3-14.0]	0.193
Platelet (10⁹/L)	222 [172-281]	220 [165-281]	0.486
WBC (10⁹/L)	14.4 [10.9-19.8]	17.1 [12.3-23.1]	<0.001
Vasopressors			<0.001
No	1314 (57.2%)	598 (40.5%)
Yes	985 (42.8%)	879 (59.5%)
Anti-coagulation treatment			<0.001
No	1069 (46.5%)	843 (57.1%)
Yes	1230 (53.5%)	634 (42.9%)
Anti-platelet treatment			0.499
No	2197 (95.6%)	1419 (96.1%)
Yes	102 (4.44%)	58 (3.93%)
Renal disease			0.313
No	1870 (81.3%)	1181 (80.0%)
Yes	429 (18.7%)	296 (20.0%)
Diabetes disease			0.803
No	1500 (65.2%)	957 (64.8%)
Yes	799 (34.8%)	520 (35.2%)
Myocardial Infarction			0.596
No	2012 (87.5%)	1302 (88.2%)
Yes	287 (12.5%)	175 (11.8%)
Congestive Heart Failure			0.005
No	1878 (81.7%)	1151 (77.9%)
Yes	421 (18.3%)	326 (22.1%)
Peripheral Vascular Disease			0.142
No	2170 (94.4%)	1376 (93.2%)
Yes	129 (5.61%)	101 (6.84%)
COPD			<0.001
No	1999 (87.0%)	1182 (80.0%)
Yes	300 (13.0%)	295 (20.0%)
Connective Tissue Disease			0.523
No	2250 (97.9%)	1440 (97.5%)
Yes	49 (2.13%)	37 (2.51%)
Peptic ulcer disease			0.870
No	2266 (98.6%)	1454 (98.4%)
Yes	33 (1.44%)	23 (1.56%)
Cerebrovascular disease			0.398
No	2206 (96.0%)	1408 (95.3%)
Yes	93 (4.05%)	69 (4.67%)
Liver cirrhosis			0.107
No	2279 (99.1%)	1455 (98.5%)
Yes	20 (0.87%)	22 (1.49%)
Renal replace treatment within the first 24 h			0.979
No	2087 (90.8%)	1342 (90.9%)
Yes	212 (9.22%)	135 (9.14%)
Mechanical ventilation within the first 24 h			<0.001
No	240 (10.4%)	31 (2.10%)
Yes	2059 (89.6%)	1446 (97.9%)
Platelet infusion			0.001
No	2269 (98.7%)	1434 (97.1%)
Yes	30 (1.30%)	43 (2.91%)
RBC infusion			0.075
No	2126 (92.5%)	1341 (90.8%)
Yes	173 (7.53%)	136 (9.21%)
Oasis	36.0 [28.0-43.0]	41.0 [35.0-48.0]	<0.001
Sofa	8.00 [5.00-11.0]	10.0 [7.00-13.0]	<0.001
ICU duration (days)	4.62 [2.58-8.15]	3.88 [2.50-6.83]	<0.001
Unit discharged status			<0.001
Alive	2277 (99.0%)	379 (25.7%)
Expired	22 (0.96%)	1098 (74.3%)

Identification of Subpopulations Using the MJLCMM

MJLCMM was fitted to identify subpopulations using multivariate longitudinal markers (platelets, temperature, hemoglobin, and WBC). The model fit indices are summarized in Table 2. We observed a consistent decrease in maximum log-likelihood, AIC, and SABIC as the number of classes increased. Across all univariate models (platelets, WBC, temperature, and hemoglobin), the two-class models consistently exhibited higher entropy values than the three-class models, indicating greater classification certainty with a two-group structure. Specifically, the entropy for the two-class models was 0.79 for platelets, 0.53 for temperature, 0.9 for hemoglobin, and 0.75 for WBC, whereas the values dropped to 0.68, 0.45, 0.75, and 0.64, respectively, in the three-class solutions. This trend of superior classification quality in the two-class solution was preserved in the multivariate integrated model. Although AIC and SABIC continued to decrease with increasing numbers of classes, model selection was not based solely on entropy. The two-class multivariate integrated model was retained because it provided high classification quality (entropy = 0.95), balanced class proportions, and clinically interpretable trajectory patterns, whereas the three-class solution produced a very small class (0.32%), suggesting reduced stability and limited practical usefulness. Consequently, based on the high entropy and distinct clinical interpretability, the two-class multivariate integrated model was selected for further analysis. The trajectories of platelets, hemoglobin, and WBC from the combined model are presented in Figures 2-4. Class 1 (Rapid decline and recovery) exhibited a dynamic fluctuation pattern. This group started with an initial sharp decline, then showed a robust, rapid recovery. Class 2 (Mild decline and recovery) experienced a milder initial decrease, followed by a modest, slow recovery. Similar trajectory patterns were observed in the MIMIC dataset, as depicted in Figures 2-4. No notable change in the temperature trajectory was observed between the two classifications in either database (Figure 5). The high average posterior probabilities and class separation, detailed in Tables S4 and S5, demonstrate the model’s robust classification certainty.

Table 2.

Metric for Determining Optimal Class Quantity

No. of classes	Source	Log likelihood	AIC	BIC	SABIC	Entropy	Class proportion (%)
No. of classes	Source	Log likelihood	AIC	BIC	SABIC	Entropy	Class1	Class2	Class3
One	Platelets	-139,489.85	279,019.70	279,144.43	279,080.88	1.00	100
Two	Platelets	-139,509.30	279,068.60	279,224.51	279,145.07	0.79	44.68	55.32
Three	Platelets	-139,394.18	278,848.37	279,035.46	278,940.14	0.68	94.07	3.39	2.54
One	WBC	-70,600.06	141,240.12	141,364.85	141,301.30	1.00	100
Two	WBC	-70,611.23	141,272.47	141,428.38	141,348.94	0.75	46.48	53.52
Three	WBC	-70,443.03	140,946.06	141,133.15	141,037.83	0.64	2.17	96.05	1.77
One	Hemoglobin	-213,378.26	426,876.52	427,250.71	427,060.06	1.00	100
Two	Hemoglobin	-213,127.92	426,401.83	426,857.09	426,625.13	0.9	29	71
Three	Hemoglobin	-213,038.21	426,248.43	426,784.76	426,511.49	0.75	21.45	59.69	18.86
One	Temperature	-21,695.00	43,430.00	43,554.72	43,491.17	1.00	100
Two	Temperature	-21,707.08	43,464.16	43,620.07	43,540.64	0.53	48.12	51.88
Three	Temperature	-21,112.47	42,284.95	42,472.04	42,376.72	0.45	0.32	71.58	28.1
One	Multivariate integrated model	-70600.06	141240.1	141364.8	141301.3	1.00
Two	Multivariate integrated model	-70495.04	141240.1	141196.0	141116.6	0.95	48.12	51.88
Three	Multivariate integrated model	-70443.03	141240.1	141196.2	141037.8	0.75	0.32	71.58	28.1

AIC: akaike information criterion, BIC: bayesian information criteria, SABIC: sample-adjusted information criteria.

Figure 2.

Longitudinal trajectories of platelet count in the derived latent classes.

Figure 3.

Longitudinal trajectories of hemoglobin levels across latent classes.

Figure 4.

Longitudinal trajectories of white blood cell (WBC) count across latent classes.

Figure 5.

Longitudinal trajectories of body temperature across latent classes.

Clinical Characteristics of Differential Trajectory Patterns and Their Association With 30-day ICU Mortality: A Propensity Score-weighted Analysis

Comparisons of clinical baseline characteristics between the two trajectory patterns are detailed in Table S2 and Table S3. Across both databases, Class 1 exhibited a more severe pro-inflammatory and metabolically deranged profile. Specifically, in the eICU-CRD dataset, Class 1 patients presented with markedly higher white blood cell counts (20.9 vs. 13.0 × 10⁹/L; P < 0.001), elevated ALT levels (93.0 vs. 54.0 IU/L; P < 0.001), and severe hyperglycemia (247 vs. 191 mg/dL; P < 0.001) compared to Class 2. This pattern of higher severity was mirrored in the MIMIC cohort; Class 1 necessitated more aggressive life support interventions; vasopressor use (65.5% vs. 42.4%; P < 0.001) and mechanical ventilation (88.1% vs. 80.3%; P < 0.001) were significantly more frequent in Class 1 than in Class 2. Class 1 exhibited higher ICU mortality rates in both eICU-CRD (46.7% vs. 35.0%; P < 0.001) and MIMIC (24.6% vs. 16.0%; P < 0.001).

As shown in Table 3, multivariable logistic regression showed that membership in the Mild Decline and Recovery group was associated with lower odds of 30-day ICU mortality in both cohorts., with adjusted odds ratios (ORs) of 0.78 (95% CI: 0.64–0.94) in the eICU-CRD cohort and 0.51 (95% CI: 0.37–0.68) in the MIMIC cohort. To adjust for baseline covariate imbalances between the two trajectory patterns, Inverse Probability Weighting (IPW) was applied, which substantially reduced confounding and maintained the robustness of the observed associations (Figure 6). Furthermore, DRE yielded consistent findings, with the Class 2 pattern remaining associated with lower 30-day mortality in both datasets, with ORs of 0.82 (95% CI: 0.72–0.96) in the eICU-CRD cohort and 0.74 (95% CI: 0.55–0.95) in the MIMIC cohort.

Table 3.

Primary Outcome Analysis With Three Different Models in the Cardiac Arrest Cohort of eICU-CRD and MIMIC-III and IV

Method	Comparison	eICU-CRD			MIMIC-III and IV
Method	Comparison	OR	95% CI	P Value	OR	95% CI	P Value
Multivariate logistic regression	Class two vs Class one	0.78	0.64-0.94	0.011	0.51	0.37-0.68	<0.001
IPW	Class two vs Class one	0.86	0.73-0.93	0.006	0.72	0.54-0.96	0.025
Doubly Robust Estimation	Class two vs Class one	0.82	0.72-0.96	0.007	0.74	0.55-0.95	0.031

IPW: Inverse Probability Weighting. ASMD: Absolute Standardized Mean Difference.In the doubly robust estimation model, covariates that remained unbalanced after propensity score weighting (ASMD > 0.1) were included in the outcome regression model for further adjustment to account for residual confounding. Specifically, these adjusted variables included body temperature and white blood cell (WBC) count for the eICU cohort, and vasoactive agent administration for the MIMIC cohort.

Figure 6.

Information on balance statistics.

Subgroup Analysis of Clinical Trajectory Phenotypes and Their Association With 30-Day ICU Mortality

As illustrated in Figure 7 and Figure S4, the Class 1 pattern was associated with significantly higher odds of 30-day ICU mortality than the Class 2 pattern in the majority of subgroups in both MIMIC and eICU-CRD, except for a small subset of subgroups with insufficient sample sizes. However, significant heterogeneity was observed regarding the use of vasoactive agents (P for interaction < 0.01) and RBC infusion (P for interaction = 0.04). Specifically, the association was more pronounced in patients not receiving vasoactive agents (OR 2.11, 95% CI 1.68–2.65) compared to those who did, while the risk relationship was attenuated and lost statistical significance in patients receiving RBC infusion (OR 0.97, 95% CI 0.57–1.68), indicating a potential interaction between these treatments and trajectory patterns regarding ICU mortality risk. Despite these specific interactions, the discriminative direction of the main results remained stable across all other prespecified subgroups and in the MIMIC database (Figure S4). To further validate the prognostic value of the classification, we performed a Kaplan-Meier survival analysis over a 30-day observation period in both datasets (Figure 8). The survival curves revealed a significant divergence between the two groups shortly after admission. Patients with Class 1 exhibited a significantly steeper decline in cumulative survival probability than those with Class 2 (log-rank test, P < 0.001), indicating that Class 1 is strongly associated with worse clinical outcomes and higher short-term ICU mortality.

Figure 7.

Forest plot of subgroup analysis for the association between trajectory patterns and 30-day ICU mortality in the eICU-CRD cohort

Figure 8.

Kaplan-Meier survival estimates stratified by trajectory phenotype.

To evaluate the prognostic accuracy of the derived trajectory phenotypes relative to established clinical severity scores, we performed a time-dependent Receiver Operating Characteristic (ROC) analysis within the eICU-CRD cohort. We compared the discriminative ability of the trajectory model with that of the OASIS and SOFA scores at 3-, 10-, and 30-day post-admission. In the acute phase (Day 3), the trajectory model yielded an Area Under the Curve (AUC) of 0.65, performing comparably to the SOFA (AUC=0.64) and OASIS (AUC=0.63) scores (Figure 9A). However, the predictive superiority of the Trajectory model became increasingly pronounced at later time points as the predictive power of static admission scores decayed. By Day 10, the Trajectory model maintained an AUC of 0.67, surpassing both OASIS (AUC=0.55) and SOFA (AUC=0.58) (Figure 9B). Most notably, at Day 30, the trajectory model demonstrated robust predictive capability with an AUC of 0.74, whereas the performance of conventional scoring systems deteriorated significantly, with OASIS and SOFA dropping to 0.59 and 0.54, respectively (Figure 9C). This divergence highlights that while static admission scores are sufficient for immediate risk assessment, the dynamic trajectory-based classification offers significantly superior long-term risk stratification for PCAS patients.

Figure 9.

Time-dependent Receiver Operating Characteristic (ROC) analysis comparing prognostic accuracy.

Discussion

Current clinical frameworks for risk stratification in PCAS, including the APACHE and SOFA scores, predominantly rely on static biomarkers obtained at admission or at fixed time points.²³ While valuable for risk assessment, these snapshots fail to capture the dynamic progression of the “sepsis-like” syndrome that defines PCAS.²⁴ The physiological state after ROSC is dynamic, marked by the complex interaction between ischemia-reperfusion injury (IRI) and compensatory physiological mechanisms.^25,26 PCAS represents a complex clinical entity characterized by significant disturbances in hemostatic balance following resuscitation.²⁷ A key manifestation of this dysregulation involves measurable alterations in platelet dynamics, reflecting widespread systemic injury and potential activation of pro-inflammatory and pro-coagulant pathways. These platelet-related changes have emerged as valuable prognostic biomarkers for predicting in-hospital mortality after ROSC.^28,29 Several studies have identified specific hematological markers associated with PCAS outcomes. Mehmet K. Ero et al demonstrated that during the initial 48 hours post-cardiac arrest, low albumin, hemoglobin, hematocrit, and mean hemoglobin concentration, along with elevated red blood cell distribution width and urea levels, correlate with increased mortality.³⁰ Furthermore, the white blood cell-to-platelet ratio (WPR) exhibits a significant U-shaped relationship with mortality risk and serves as an independent prognostic biomarker for 28-day mortality in cardiac arrest patients.³¹ Despite these advances, conventional static biomarker measurements remain limited in their ability to reflect the dynamic pathophysiology of PCAS. Single-time-point platelet counts cannot distinguish between stable levels and ongoing consumption during PCAS progression. Previous studies suggest that static measures may inadequately capture trajectories of deterioration or recovery, thereby constraining risk stratification.^32,33 These limitations underscore the need to augment conventional “snapshot” with trajectory-based methodologies,³⁴ integrating them as complementary tools. In this setting, MPJLCMM provides a statistical framework for modeling complex longitudinal relationships among multiple variables and survival outcomes in heterogeneous populations.³⁵ This framework enables the identification of patient subphenotypes and provides real-time, dynamic prognostic predictions as new longitudinal data become available, complementing initial baseline assessments.³⁶

The pathological basis of PCAS involves ischemia-reperfusion injury that triggers a sepsis-like inflammatory response^37,38. In this theoretical framework, this process leads to endothelial cell denudation, exposing the prothrombotic subendothelial matrix and catalyzing systemic platelet adhesion and activation. Activated platelets may form platelet-leukocyte aggregates (PLAs), which could become trapped in capillary beds and disappear from peripheral circulation.³⁹ Leveraging MPJLCMM with data from the MIMIC and eICU-CRD databases, our study identified two distinct clinical phenotypes based on joint longitudinal trajectories of four key biomarkers: platelets, white blood cells (WBC), temperature, and hemoglobin. Class 1 (“Rapid Decline and Recovery”) exhibits a precipitous initial decline in platelets, WBC, and hemoglobin followed by a rapid rebound. This high-risk phenotype demonstrates a strong association with ICU mortality. In contrast, Class 2 (“Mild Decline and Recovery”) shows an attenuated physiological response and significantly higher survival rates. The sharp decline in platelet and WBC counts observed in Class 1 may serve as a macroscopic surrogate for large-scale systemic activation and PLA formation. However, without direct molecular assays in our dataset, these pathophysiological interpretations remain hypothesis-generating. The MPJLCMM method enables inference of potential trajectories, even for patients who died early, by jointly modeling longitudinal biomarkers. The high entropy value (0.95) indicates a high degree of class separation and classification certainty, suggesting that combining platelets, white blood cells, temperature, and hemoglobin into a single dynamic structure provides robust and improved phenotypic discrimination compared to individual biomarkers alone (Table 2). Notably, no significant difference in temperature trajectory was observed between Class 1 and Class 2 patients. A possible explanation for this phenomenon is that the widespread application of targeted temperature management (TTM) in this population may have homogenized temperature curves, masking the endogenous “febrile” response that could otherwise differentiate inflammatory phenotypes. On the other hand, it also suggests that even with strict control of core temperature, the macroscopic biomarker trajectories indicate that thrombo-inflammatory cascades (such as proposed NETosis or platelet consumption) might still progress aggressively.^40-42 Crucially, our time-dependent ROC analysis (Figure 9) provides evidence for the limitations of static “snapshot” assessments in the post-cardiac arrest population. While traditional scoring systems like SOFA and OASIS demonstrated comparable predictive accuracy to our trajectory model in the acute phase (Day 3), their discriminative ability deteriorated rapidly over time, dropping to minimal discriminatory capacity (AUCs of 0.54 and 0.59, respectively) by Day 30. This “predictive decay” likely occurs because admission scores predominantly reflect the initial severity of insult but fail to account for the subsequent efficacy of compensatory mechanisms or the progression of secondary injury (e.g., thrombo-inflammation). In contrast, the trajectory-based model maintained robust prognostic utility (AUC = 0.74) at Day 30. This finding underscores that the directionality of physiological change—the slope of recovery or deterioration—is a more reliable indicator of long-term survival than the absolute magnitude of organ dysfunction at a single time point. The clinical applicability of these dynamic phenotypes lies in both prognostic triage and therapeutic targeting. While the mathematical complexity of trajectory modeling precludes simple manual calculation, its true translational potential resides in integration into modern Electronic Health Record (EHR) systems as an automated clinical decision support tool. Such systems could continuously monitor these trajectories to identify patients exhibiting the high-risk ‘Rapid Decline’ pattern in real time, prompting heightened vigilance and timely interventions for secondary microvascular complications.

However, our study is subject to several limitations that are worth noting. Firstly, as a retrospective cohort study, this research design inherently limits our ability to infer causality and cannot establish causal relationships as definitive as those in randomized controlled trials. While this approach efficiently leverages existing data, it remains susceptible to several methodological limitations, including confounding factors, selection bias, and information bias. Although we employed comprehensive statistical methodologies—such as propensity score matching and doubly robust estimation—to adjust for measured confounders, it is important to note that these findings represent robust observational associations rather than confirmed causal effects, as the potential for unmeasured confounding remains. Moreover, despite consistent findings across the eICU-CRD and MIMIC cohorts, prospective validation in geographically and clinically distinct populations remains necessary, particularly for subgroups with limited sample size. Secondly, as a retrospective analysis of EHR databases, the study infers immunothrombosis and Neutrophil Extracellular Trap (NET) formation from trajectory patterns of standard markers (platelets, WBC). It does not directly measure specific biomarkers of NETs (e.g., H3Cit, MPO-DNA complexes)^43,44 or platelet activation (e.g., P-selectin, soluble CD40L).^45,46 Consequently, the mechanistic links discussed herein are inherently inferential and serve primarily as hypothesis-generating concepts. Future prospective translational studies are required to measure these specific mediators alongside clinical trajectories to validate these mechanistic links. Thirdly, although the overall sample size was large, some subgroup analyses may have been underpowered because of limited sample sizes within specific strata, potentially precluding the detection of significant associations in certain populations. In addition, because the subgroup analyses involved multiple interaction tests, the possibility of Type I error inflation cannot be excluded. Therefore, these subgroup findings should be interpreted cautiously as exploratory and hypothesis-generating, and their generalizability warrants further validation in future, adequately powered prospective studies.

Fourth, comparisons of clinical characteristics and outcomes between the derived trajectory phenotypes were based on assigning each patient to the latent class with the highest posterior probability. While this hard-classification approach facilitates conventional between-group statistical testing, it does not fully account for the probabilistic nature of latent class membership and the resulting classification uncertainty. However, the high entropy of the final multivariate integrated model (0.95), together with the favorable posterior classification metrics shown in Tables S4 and S5, suggests strong class separation and likely reduces, though does not eliminate, the impact of potential misclassification on these comparisons. Fifth, our study has inherent limitations stemming from the retrospective nature of the MIMIC and eICU-CRD databases. Specifically, the lack of granular pre-hospital emergency medical services (EMS) data prevented reliable differentiation between out-of-hospital and in-hospital cardiac arrest. At the same time, critical peri-arrest variables (e.g., exact time to ROSC, initial shockable rhythm) were largely unavailable, necessitating reliance on admission severity scores (SOFA and OASIS) as surrogates for initial insult severity.

Furthermore, the exclusion of patients with ICU length of stay <24 hours—implemented to ensure sufficient data points for robust trajectory modeling—introduces survivorship bias by omitting early deaths. Sixth, our primary outcome was limited to 30-day ICU mortality due to inherent constraints of the eICU and MIMIC databases, which precluded reliable tracking of post-discharge out-of-hospital survival. Additionally, these administrative datasets lack standardized, structured neurological assessments, such as the Cerebral Performance Category (CPC) scale, thereby preventing the evaluation of long-term neurological outcomes, which are critical endpoints in post-cardiac arrest care. Future prospective studies are essential to validate these dynamic thrombo-inflammatory phenotypes in relation to both long-term survival and neurological recovery.

Conclusion

This multicenter observational study identifies distinct clinical phenotypes in patients with PCAS based on specific dynamic trajectories of thrombo-inflammatory biomarkers. The “Rapid Decline and Recovery” was associated with higher 30-day ICU mortality and may represent a high-risk clinical subgroup. Trajectory-based classification showed promising long-term prognostic performance compared with static admission scores (e.g., SOFA, OASIS) in this cohort. Future prospective studies are warranted to validate these phenotypes and investigate whether precision interventions targeting the thrombo-inflammatory cascade can meaningfully inform therapeutic strategies and improve survival in this vulnerable population.

Supplemental Material

Supplemental Material -Trajectory-Defined Thrombo-Inflammatory Phenotypes Predict 30-Day ICU Mortality in Post-cardiac Arrest Syndrome: A Multicenter Retrospective Longitudinal Cohort Study

Supplemental Material for Trajectory-Defined Thrombo-Inflammatory Phenotypes Predict 30-Day ICU Mortality in Post-cardiac Arrest Syndrome: A Multicenter Retrospective Longitudinal Cohort Study by Guyu Zhang, Le An, ChenChen Hang, XingSheng Wang, Rui Shao, ZhenYu Shan, Ziren Tang in INQUIRY: The Journal of Health Care Organization, Provision, and Financing

Footnotes

Acknowledgments

We thank all Emergency Medicine Clinical Research Center participants, Beijing Chaoyang Hospital, Capital Medical University.

ORCID iD

Ziren Tang

Ethical Considerations

The MIMIC and eICU databases are publicly available, and the creation of the research resource was reviewed by the Institutional Review Board at the Beth Israel Deaconess Medical Center. Permission to use the data was available in the Supplemental Material. The patient consent was waived as the data are wholly deidentified and retrospective from public databases.Access to this database was granted after one of the authors (GuyuZhang, certification ID: 55849941) completed the requisite NIH-mandated online training.

Consent to Participate

The database is publicly available and all patient data has been de-identified. The Institutional Review Boards (IRB) of the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC) approved the database establishment and waived the need for informed consent.

Author Contributions

GZ collected the data, analyzed the data, and drafted the manuscript. LA, CCH, RS, and XSW extracted the data and participated in its design. ZYS participated in the literature research. ZT acted as the guarantor for the entire project, with responsibilities including manuscript review, study design, and overall supervision. All authors contributed to the article and approved the submitted version.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Beijing Municipal Science & Technology Commission, Administrative Commission of Zhongguancun Science Park No. Z251100006025022 and the Beijing Hospitals Authority’s Ascent Plan (DFL20240302).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The raw data supporting the conclusions of this article were available at Medical Information Mart for Intensive Care (PhysioNet), and eICU Collaborative Research Database (PhysioNet).*

Clinical Trial Registration

This study is a retrospective analysis of a publicly available database and does not require clinical trial registration.

Supplemental Material

Supplemental material for this article is available online.

References

Holmström

Reinier

Toft

, et al. Out-of-hospital cardiac arrest with onset witnessed by emergency medical services: Implications for improvement in overall survival. Resuscitation. 2022;175:19-27. doi:10.1016/j.resuscitation.2022.04.003

Bowman

Tulsky

Ouchi

. Mortality and healthcare resource utilization after cardiac arrest in the United States: A decade of unclear progress and stark disparities. Resuscitation. 2023;193:109985. doi:10.1016/j.resuscitation.2023.109985

de Graaf

Donders

Beesems

Henriques

JPS

Koster

. Time to return of spontaneous circulation and survival: when to transport in out-of-hospital cardiac arrest? Prehosp Emerg Care. 2021;25:171-181. doi:10.1080/10903127.2020.1752868

El-Menyar

Wahlen

. Cardiac arrest, stony heart, and cardiopulmonary resuscitation: An updated revisit. World J Cardiol. 2024;16:126-136. doi:10.4330/wjc.v16.i3.126

Katircioglu

Ayvat

Gunturkun

. Cardiac arrest patients admitted to intensive care unit after cardiopulmonary resuscitation: a retrospective cohort study to find predictors for mortality. Braz J Anesthesiol. 2023;73:401-408. doi:10.1016/j.bjane.2021.03.013

Gianforcaro

Kurz

Guyette

, et al. Association of antiplatelet therapy with patient outcomes after out-of-hospital cardiac arrest. Resuscitation. 2017;121:98-103. doi:10.1016/j.resuscitation.2017.10.007. 20171012.

Seppä

AMJ

Skrifvars

Pekkarinen

. Inflammatory response after out-of-hospital cardiac arrest-Impact on outcome and organ failure development. Acta Anaesthesiol Scand. 2023;67:1273-1287. doi:10.1111/aas.14291

Liang

Wang

Guan

Wang

Gong

. Predictive Value of Neutrophil Extracellular Trap Components for 28-Day All-Cause Mortality in Patients with Cardiac Arrest: A Pilot Observational Study. Shock. 2023;60:664-670. doi:10.1097/SHK.0000000000002225

Vanat

Lee

Elkhider

, et al. Multimodal prediction of favorable outcome after cardiac arrest: a cohort study. Crit Care Med. 2023;51:706-716. doi:10.1097/CCM.0000000000005841

10.

Mueller

Grafeneder

Schoergenhofer

, et al. Initial Blood pH, Lactate and Base Deficit Add No Value to Peri-Arrest Factors in Prognostication of Neurological Outcome After Out-of-Hospital Cardiac Arrest. Front Med (Lausanne). 2021;8:697906. doi:10.3389/fmed.2021.697906

11.

Laimoud

Alanazi

. The Validity of SOFA Score to Predict Mortality in Adult Patients with Cardiogenic Shock on Venoarterial Extracorporeal Membrane Oxygenation. Crit Care Res Pract. 2020;2020:3129864. doi:10.1155/2020/3129864

12.

Elmer

Jones

Zadorozhny

, et al. A novel methodological framework for multimodality, trajectory model-based prognostication. Resuscitation. 2019;137:197-204. doi:10.1016/j.resuscitation.2019.02.030

13.

Abudurexiti

Sun

Jiang

Gong

. Glucose metabolic reprogramming-related parameters for the prediction of 28-day neurological prognosis and all-cause mortality in patients after cardiac arrest: a prospective single-center observational study. World J Emerg Med. 2024;15:197-205. doi:10.5847/wjem.j.1920-8642.2024.047

14.

Alnabelsi

Annabathula

Shelton

, et al. Predicting in-hospital mortality after an in-hospital cardiac arrest: A multivariate analysis. Resusc Plus. 2020;4:100039. doi:10.1016/j.resplu.2020.100039

15.

Soloperto

Magni

Farinella

, et al. A Comparison of Prognostic Factors in a Large Cohort of In-Hospital and Out-of-Hospital Cardiac Arrest Patients. Life (Basel). 2024;14:403. doi:10.3390/life14030403

16.

Proust-Lima

Saulnier

Philipps

, et al. Describing complex disease progression using joint latent class models for multivariate longitudinal markers and clinical endpoints. Stat Med. 2023;42:3996-4014. doi:10.1002/sim.9844

17.

von

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Clin Epidemiol. 2008;61:344-349. doi:10.1016/j.jclinepi.2007.11.008

18.

Johnson

Pollard

Shen

, et al. MIMIC-III, a freely accessible critical care database. Sci data. 2016;3:1-9. doi:10.1038/sdata.2016.35

19.

Johnson

Bulgarelli

Shen

, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci data. 2023;10:1-1. doi:10.1038/s41597-022-01899-x

20.

Pollard

Johnson

AEW

Raffa

Celi

Mark

Badawi

. The eICU Collaborative Research Database, a freely available multicenter database for critical care research. Sci Data. 2018;5:180178. doi:10.1038/sdata.2018.178

21.

van Buuren

Groothuis-Oudshoorn

. mice: Multivariate Imputation by Chained Equations in R. J Stat Softw. 2011;45:1-67. doi:10.18637/jss.v045.i03

22.

Kim

. Determining the Number of Latent Classes in Single- and Multi-Phase Growth Mixture Models. Struct Equ Modeling. 2014;21:263-279. doi:10.1080/10705511.2014.882690

23.

Pekkarinen

Ristagno

Wilkman

, et al. Procalcitonin and presepsin as prognostic markers after out-of-hospital cardiac arrest. Shock. 2018;50:395-400. doi:10.1097/SHK.0000000000001081

24.

Mai

Prifti

Rininger

Bazarian

Halterman

. Endotoxemia induces lung-brain coupling and multi-organ injury following cerebral ischemia-reperfusion. Exp Neurol. 2017;297:82-91. doi:10.1016/j.expneurol.2017.07.012

25.

Gando Sand Wada

Wada

. Disseminated intravascular coagulation in cardiac arrest and resuscitation. JThromb Haemost. 2019;17:1205-1216. doi:10.1111/jth.14489

26.

Kim

Michael

Wee

, et al. Coagulation measures after cardiac arrest (CMACA). PLoS One. 2023;18:e0279653. doi:10.1371/journal.pone.0279653

27.

Rojas-Salvador

Kalra

Steiner

, et al. Consumptive Coagulopathy is Associated With Poor Neurological Outcome in Patients With Refractory VT/VF Cardiac Arrest Requiring Extracorporeal Cardiopulmonary Resuscitation. Circulation. 2023;148:A246-A246. doi:10.1161/circ.148.suppl_1.246

28.

Zhang

, et al. Development and validation of machine learning-based prediction model for outcome of cardiac arrest in intensive care units. Sci Rep. 2025;15:8691. doi:10.1038/s41598-025-93144-w

29.

Huang

Peng

Wang

. High platelet-lymphocyte ratio is a risk factor for 30-day mortality in in-hospital cardiac arrest patients: a case-control study. Expert Rev Clin Immunol. 2021;17:1231-1239. doi:10.1080/1744666X.2021.1979393

30.

Erol

Kankılıc

Kaya

Atlas

Pehlivan

. The relationship between hematological parameters and mortality in cardiovascular patients with postcardiac arrest syndrome. Cureus. 2019;11:e5795. doi:10.7759/cureus.5795

31.

Huang

Ren

Sun

, et al. Correlation between the white blood cell/platelet ratio and 28-day all-cause mortality in cardiac arrest patients: a retrospective cohort study based on machine learning. Front Pharmacol. 2025;15:1527664. doi:10.3389/fphar.2024.1527664

32.

De Rosa

Lassola

Visconti

, et al. Acute Kidney Injury in Patients After Cardiac Arrest: Effects of Targeted Temperature Management. Life. 2025;15:265. doi:10.3390/life15020265

33.

Misumi

Hagiwara

Kimura

, et al. External Validation of the CAST and rCAST Score in Patients With Out‐of‐Hospital Cardiac Arrest Who Underwent Extracorporeal Cardiopulmonary Resuscitation: A Secondary Analysis of the SAVE‐J II Study. J Am Heart Assoc. 2024;13:e031035. doi:10.1161/JAHA.123.031035

34.

Chen

Gao

Shen

, et al. Association of longitudinal platelet count trajectory with ICU mortality: A multi-cohort study. Front Immunol. 2022;13:936662. doi:10.3389/fimmu.2022.936662

35.

Andrew

Pieper

Cherry

, et al. Identification of trajectory-based acute kidney injury phenotypes among cardiac surgery patients. The Annals of thoracic surgery. 2022;114:2235-2243. doi:10.1016/j.athoracsur.2021.10.021

36.

Nam

J-W

Lee

W-R

Kim

G-H

, et al. Association between healthcare utilization patterns and long-term all-cause mortality after acute myocardial infarction discharge in South Korea. Front Public Health. 2025;12:1530274. doi:10.3389/fpubh.2024.1530274

37.

Sugita

Kinoshita

Sakurai

, et al. Systemic impact on secondary brain aggravation due to ischemia/reperfusion injury in post-cardiac arrest syndrome: a prospective observational study using high-mobility group box 1 protein. Crit Care. 2017;21:247. doi:10.1186/s13054-017-1834-3

38.

Wada

. Coagulofibrinolytic changes in patients with post-cardiac arrest syndrome. Front Med. 2017;4:156. doi:10.3389/fmed.2017.00156

39.

Zahn

Schanze

Staudacher

, et al. The Effect of Cytokine Adsorption on Leukocyte and Platelet Activation after Extracorporeal Cardiopulmonary Resuscitation. Thromb Haemost.Thromb Haemost. 2024;124:533-545. doi:10.1055/s-0043-1777259

40.

Wienkamp

Erpenbeck

Rossaint

. Platelets in the NETworks interweaving inflammation and thrombosis. Front Immunol. 2022;13:953129. doi:10.3389/fimmu.2022.953129

41.

Zdanyte

Borst

Munzer

. NET-(works) in arterial and venous thrombo-occlusive diseases. Front Cardiovasc Med 2023; 10: 1155512. doi:10.3389/fcvm.2023.1155512

42.

Colicchia

Perrella

Gant

Rayes

. Novel mechanisms of thrombo-inflammation during infection: spotlight on neutrophil extracellular trap-mediated platelet activation. Res Pract Thromb Haemost. 2023;7:100116. doi:10.1016/j.rpth.2023.100116

43.

Melamud

Tolmacheva

Sizikov

, et al. NETosis-Related Biomarkers in Systemic Lupus Erythematosus, Rheumatoid Arthritis, Psoriatic Arthritis and Ankylosing Spondylitis: A Comparative Study. Int J Mol Sci. 2025;26:1217. doi:10.3390/ijms262412127

44.

, et al. Positive correlations between plasma BPI level and MPO-DNA and S100A8/A9 in myocardial infarction. Platelets. 2022;33:603-611. doi:10.1080/09537104.2021.1961427

45.

Shen

Yang

Xia

, et al. P-selectin (CD62P) and soluble TREM-like transcript-1 (sTLT-1) are associated with coronary artery disease: a case control study. BMC Cardiovasc Disord. 2020;20:387. doi:10.1186/s12872-020-01663-2

46.

Rupp

Bergmann

Jung

, et al. The posttraumatic response of CD4+ regulatory T cells is modulated by direct cell-cell contact via CD40L- and P-selectin-dependent pathways. Cent Eur J Immunol. 2021;46:283-294. doi:10.5114/ceji.2021.109171

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.12 MB