Cytokine-based strategies to improve prognostic enrichment of pediatric ARDS.

Summary

In a prospective multicenter pediatric ARDS cohort (N=500), the traditional IL-6/TNFR1/bicarbonate hyper-/hypo-inflammatory model had modest utility for 90-day mortality. Adding CCL7 and IL-18 yielded parsimonious models with improved discrimination (AUROC 0.72–0.73 training; 0.66 validation) and ~40% better reclassification; immunocompromised status independently predicted mortality.

Key Findings

• Traditional IL-6/TNFR1/bicarbonate model showed modest performance for 90-day mortality (AUC 0.62; sensitivity 26%; PPV 24%).
• CCL7 (MCP-3) had the strongest association with 90-day mortality (p<0.0001).
• Two-term (CCL7, IL-18) and three-term (CCL7, IL-18, TNFR1) models achieved AUROC 0.72–0.73 (training) and 0.66 (validation) with ~40% improved net reclassification.
• Immunocompromised status independently predicted 90-day mortality and may require tailored prognostic models.

Clinical Implications

Cytokine panels including CCL7 and IL-18 may stratify pediatric ARDS for monitoring and enrollment in immunomodulatory trials, moving beyond IL-6/TNFR1-based approaches.

Limitations

• Moderate discrimination in external validation (AUROC 0.66) indicates need for external replication and calibration.
• Single early timepoint sampling (within 24 h) may miss dynamic cytokine trajectories.

Future Directions

External validation across centers, dynamic longitudinal sampling, and testing whether cytokine-based enrichment improves response to targeted immunomodulatory therapies.