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

Comprehensive Summary

Michels et al. developed and validated a machine-learning (ML) tool to quantify immune dysregulation in community-acquired pneumonia (CAP) and sepsis patients. This multicohort analysis and reanalysis of a randomised controlled trial placed 398 CAP patients on a continuum of three Dysregulated Immune Profile stages (DIP1–3) and a continuous score (cDIP) using 35 plasma biomarkers. Researchers then built a parsimonious three-biomarker machine-learning tool (procalcitonin, soluble TREM-1, and IL-6) that was validated across five external cohorts (total n=1191). Higher dysregulation (DIP/cDIP) was associated with increased mortality (p<0.0001) and secondary infections (p=0.0005), regardless of clinical severity. Additionally, in a post-hoc reanalysis of the CAPE COD randomized trial in severe CAP, hydrocortisone’s survival benefit appeared limited to patients classified as severely dysregulated (DIP3 and above a cDIP threshold, p=0.011) alongside faster resolution of dysregulation. Stratification by clinical severity failed to demonstrate a similar effect.

Outcomes and Implications

This study highlights the potential of biomarker-based ML frameworks to improve clinical decision-making by quantifying immune dysregulation beyond clinical severity scoring. Tools such as the DIP/cDIP staging could meaningfully guide clinicians in identifying which patients are most likely to benefit from immunomodulatory treatments such as hydrocortisone. The study is limited by its reliance on retrospective cohort analyses and post-hoc trial reanalysis, the variability of biomarker availability across settings and the need for prospective validation of the data-driven cDIP threshold. Therefore, further research is warranted to evaluate the model’s clinical utility, scalability, and impact on patient outcomes across diverse populations and healthcare environments.