Derivation and Validation of a Clinical and Endothelial Biomarker Risk Model to Predict Persistent Pediatric Sepsis-Associated Acute Respiratory Dysfunction

Abstract

Background

Sepsis-associated ARDS results in high morbidity and mortality in children. However, heterogeneity among patients makes identifying those at risk of persistent acute respiratory dysfunction challenging. Endothelial dysfunction is a key feature of ARDS pathophysiologic characteristics, contributing to lung injury in sepsis. Incorporating endothelial biomarkers into risk models may enhance prediction of those with persistent acute respiratory dysfunction.

Research Question

Can clinical variables and endothelial biomarkers measured early in the course of sepsis predict risk of persistent acute respiratory dysfunction among critically ill children?

Study Design And Methods

This was a multicenter derivation and single center test cohort study of prospectively enrolled children with sepsis. The derivation cohort was split into training and holdout validation sets. We trained TreeNet (Minitab, LLC) and classification and regression tree (CART) models using clinical and endothelial biomarkers measured on day 1 of septic shock to predict risk of sepsis-associated acute respiratory dysfunction (SA ARD) on day 3. The performance of the CART model was tested in the holdout validation data set and in the independent test cohort.

Results

In the derivation (n = 625) and test (n = 162) cohorts, children with day 3 SA ARD showed increased mortality, length of mechanical ventilation, and PICU length of stay compared with those without. The TreeNet and CART models yielded comparable results. The variables included in the final CART model were presence of SA ARD on day 1, Pao₂ to Fio₂ ratio of < 250, soluble thrombomodulin, and vascular cell adhesion molecule 1 concentrations. This model showed an area under the receiver operating characteristic curve (AUC) of 0.88 in the training data set, sensitivity of 0.91 (95% CI, 0.86-0.94), specificity of 0.76 (95% CI, 0.68-0.82), and demonstrated reproducibility in validation data set and test cohort (AUC range, 0.78-0.83).

Interpretation

We derived and validated predictive models incorporating clinical and endothelial biomarkers to identify pediatric patients with septic shock at high risk of persistent acute respiratory dysfunction. Pending prospective validation, such models may facilitate enrichment and targeted intervention in future clinical trials.