Pentraxin-3, MyD88, GLP-1, and PD-L1: Performance assessment and composite algorithmic analysis for sepsis identification.

Objectives: Accurate diagnosis of sepsis is needed to initiate life-saving treatment decisions. Biomarkers capable of identifying both acute infection and sepsis are required to assist clinicians.

Methods: A real-life heterogeneous cohort of 388 patients with suspected acute infections was recruited at presentation to the ED. Nine emerging host-response biomarkers (MyD88, MMP-8, leptin, ENA-78, fractalkine, PD- L1, pentraxin-3, TRAIL, and GLP-1) were quantified using a multiparameter assay. We performed AUROC analysis for the endpoints bacterial infection, sepsis, and 30-day mortality. We further assessed diagnostic performance when combining these biomarkers using a machine learning algorithm.

Results: Particularly, MyD88, PD-L1, and pentraxin-3 presented high AUROCs for the endpoints bacterial infection (≥0.87), sepsis (≥0.81), and 30-day mortality (≥0.71). Seven out of the nine investigated biomarkers showed statistically significant discrimination for all three endpoints. A combined algorithm via the XGBoost model using pentraxin-3, MyD88, and GLP-1 was used for sepsis prediction, with an AUROC of 0.89, higher than clinical assessment via NEWS-2 (0.83) or procalcitonin (0.81).

Conclusion: Pentraxin-3, MyD88, GLP-1, and PD-L1 are a promising complementary set of biomarkers for risk assessment and stratification. When a trained multiparameter classifier is used, the combination of biomarkers results in a valid tool for sepsis diagnosis.

Trial Registration: DRKS00020521, DRKS00017395.