549 - Inflammatory and Endothelial Biomarker Profiles Reveal Biological Heterogeneity Beyond Oxygenation Severity in Pediatric ARDS

Publication Number: 1528.549

Christine Zhang, University of California, Los Angeles David Geffen School of Medicine, Los Angeles, CA, United States; Colin Sallee, UCLA Mattel Childrens Hospital, Los Angeles, CA, United States; Daniela Markovic, UCLA, Los Angeles, CA, United States; Ana Carolina Costa Monteiro, University of California, Los Angeles David Geffen School of Medicine, Los Angeles, CA, United States; Clove Taylor, Stanford University School of Medicine, Stanford, CA, United States; Matt Zinter, University of California, San Francisco, School of Medicine, San Francisco, CA, United States; Dylan Yu, University of California, Los Angeles David Geffen School of Medicine, Los Angeles, CA, United States; Andreas Schwingshackl, UCLA Mattel Childrens Hospital, Los Angeles, CA, United States; Michael Agus, Harvard Medical School, Boston, MA, United States; Anil Sapru, UCLA, Los Angeles, CA, United States

Background: Pediatric acute respiratory failure (ARF) is clinically and biologically heterogeneous. Although hypoxemia severity is a conventional risk stratification tool, it may not fully capture the diverse clinical trajectories and treatment responses observed in this population. These limitations highlight the need for biologically-based subgrouping to complement physiological metrics and improve risk stratification and treatment.

Objective: We aimed to identify biomarker-defined subphenotypes in pediatric ARF, grouped by hypoxemia severity, and determine whether they improve prognostic discrimination beyond hypoxemia severity alone.

Design/Methods: Plasma biomarker expression was analyzed across hypoxemia severity, defined by oxygenation index (OI), in pediatric patients receiving invasive mechanical ventilation in the CAF-PINT trial. Fourteen inflammatory and endothelial biomarkers were quantified at baseline using Luminex assays. Patients were stratified into mild (OI < 4) and severe (OI > 16) groups after propensity score matching for age, PRISM score, hypotension, and sex. K-means clustering (k = 2) identified biomarker-defined subgroups, and outcomes (90-day mortality, 28-day ventilator free days (VFDs), and ICU length of stay) were compared using Wilcoxon tests. Prognostic performance of biomarker-only vs. biomarker + OI models for 90-day mortality was assessed using AUROC with DeLong testing.

Results: The cohort included 49 patients per group (OI < 4 vs OI > 16), with similar biomarker levels and outcomes, except for higher ICAM1 and TREM1 in severe ARF (p = 0.0112 and 0.0067, FDR = 0.0784). K-means clustering within the OI > 16 group identified two distinct biological subgroups (N1 = 11, N2 = 38): a hyper-inflammatory, endothelial-activated cluster associated with higher mortality (58.3% vs. 8.1%, p< 0.001), fewer VFDs (p=0.001), and longer ICU length of stay (p=0.009) compared to a hypo-inflammatory, endothelial-quiescent cluster. Clustering of the OI < 4 group yielded similar, but less distinct results. Biomarker-only and OI + biomarker models showed similar discrimination for 90-day mortality (AUC = 0.885 vs 0.889, DeLong p=0.78).

Conclusion(s): In pediatric ARF, hypoxemia only partly reflects underlying biology, with similar oxygenation linked to divergent molecular and clinical profiles. Our findings suggest that inflammatory and endothelial activation markers contribute to this heterogeneity independent of hypoxemia severity. Integrating biomarker-based subphenotyping with conventional physiological indices may enhance patient risk stratification and guide targeted therapies in pediatric ARF.

Plasma Biomarker-based Clustering and Principal Component Contributions in Severe ARF (OI > 16)
(A) Unsupervised K-means clustering (k = 2) of patients with severe ARF using 13 plasma biomarkers (SerpinE1/PAI-1, CXCL4/PF4, CRP, IL-6, CXCL8/IL-8, IL-10, Angiopoietin-2, Thrombomodulin, TFPI, TREM-1, P-selectin, ICAM-1, TNFR1) identified two distinct clusters (N1= 11, N2= 38) with visible separation along principle component 1, accounting for 32.7% of total variance (B) Principal component analysis (PCA) of the same 14 biomarkers shows that inflammatory and endothelial markers contribute most strongly to overall variance. Separation along Dim1 corresponds to inflammatory and endothelial activation. Cluster 1 represents hyper-inflammatory profiles, while Cluster 2 reflects hypo-inflammatory, endothelial-stable profiles.

Mean Standardized Biomarker Profiles Define Distinct Severe ARF Subphenotypes
Cluster centroid heatmap illustrating z-score biomarker expression for two K-means-derived subgroups within the severe ARF cohort (OI >16). One subgroup (Cluster 1) shows concurrent elevation of inflammatory (IL-6, TNFR1, TREM-1) and endothelial injury markers (Angiopoietin-2, Thrombomodulin, SerpinE1/PAI-1), while the other (Cluster 2) demonstrates uniformly lower levels across these pathways, indicating biologically distinct inflammatory states.

Predictive Performance of Biomarker and Combined OI + Biomarker Models for 90-Day Mortality
Receiver operating characteristic (ROC) curves comparing a biomarker-only logistic regression model with a combined OI + biomarker model for 90-day mortality prediction. The two models demonstrated similar discrimination (AUC = 0.885 vs 0.889, p - 0.78, DeLong test), indicating that adding OI did not significantly improve predictive performance beyond plasma biomarkers alone for mortality.