Quantifying immune dysregulation in pneumonia and sepsis with a parsimonious machine-learning model: a multicohort analysis across care settings and reanalysis of a hydrocortisone randomised controlled trial.

Summary

Across three cohorts, a three-biomarker model (procalcitonin, soluble TREM-1, IL-6) accurately quantified immune dysregulation (DIP stages and cDIP) independent of clinical severity. Validation in five external cohorts showed increasing cDIP associated with higher mortality and secondary infections, and CAPE COD reanalysis indicated hydrocortisone benefited only severely dysregulated patients.

Key Findings

• A three-biomarker ML framework (procalcitonin, soluble TREM-1, IL-6) predicted immune dysregulation stages with 91.2% accuracy; cDIP RMSE 0.056 versus 35 biomarkers.
• Clinical severity was an inadequate proxy for immune dysregulation across CAP cohorts.
• Each 10% increase in cDIP was associated with higher mortality (OR 1.26, 95% CI 1.13–1.40) and secondary infections (OR 1.50, 95% CI 1.22–1.93), independent of severity.
• Hydrocortisone reduced 30-day mortality only in severely dysregulated patients (DIP3 OR 0.25; cDIP ≥0.63 OR 0.21) and accelerated immune recovery.
• The model generalized across five external cohorts (n=1191) spanning infections, severities, and care settings.

Clinical Implications

Incorporating procalcitonin, soluble TREM-1, and IL-6 to compute a dysregulation score could stratify patients for immunomodulatory therapy (e.g., hydrocortisone) and inform trial enrichment, rather than relying solely on clinical severity.

Limitations

• Post-hoc analysis for hydrocortisone effect modification; prospective stratified trials are needed.
• Derivation focused on CAP; applicability to all sepsis phenotypes and assay availability require evaluation.

Future Directions

Prospective, biomarker-stratified RCTs testing immunomodulators; implementation studies integrating cDIP into ED/ICU workflows and multiplex assays.