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 multiple cohorts, a three-biomarker tool (procalcitonin, soluble TREM-1, IL-6) accurately quantified immune dysregulation in CAP/sepsis and was independently associated with mortality and secondary infection risk. Post-hoc reanalysis of a hydrocortisone RCT (CAPE COD) suggested survival benefit only in patients with severe dysregulation by this model, supporting biomarker-guided immunomodulation.

Key Findings

• A 3-biomarker ML model (PCT, sTREM-1, IL-6) predicted immune dysregulation (DIP accuracy 91.2%; cDIP RMSE 0.056) derived from 35 biomarkers.
• Greater dysregulation associated with higher mortality (OR 1.26 per 10% cDIP increase) and secondary infections (OR 1.50 per 10% cDIP increase), independent of clinical severity.
• External validation in five cohorts (n=1191) confirmed performance across settings.
• Hydrocortisone reduced 30-day mortality only in severely dysregulated patients (e.g., cDIP ≥0.63; OR 0.21), with faster immune recovery; no effect modification by clinical severity.

Clinical Implications

Enable risk stratification and selection for immunomodulatory therapy using three widely available biomarkers, refining trial design and supporting bedside precision treatment decisions in pneumonia/sepsis.

Limitations

• Post-hoc nature of the hydrocortisone trial reanalysis; not a randomized biomarker-stratified trial
• Biomarker thresholds and operational cut-offs require prospective validation and health-system integration

Future Directions

Prospective biomarker-stratified RCTs to test corticosteroids or other immunomodulators guided by cDIP/DIP, implementation studies of point-of-care panels, and integration with EHR decision support.