Comprehensive Summary
The SEPSIS-SHIELD study was a multicenter, prospective trial from March 2020 to May 2024 across 22 emergency departments in Europe and the United States to evaluate the diagnostic and prognostic performance of the machine learning-based TriVerity assay for differentiating viral and bacterial infections and identifying patients requiring intensive care. The trial enrolled 1,444 adults presenting with suspected acute bacterial or viral infection who met predefined vital sign criteria and had blood cultures obtained on admission. Of these, 1,222 patients produced valid TriVerity results, and infectious status was adjudicated by clinical consensus in 729 cases. Among the 729 adjudicated cases, 448 (61.5%), were bacterial, 165 (22.6%) viral, and 12 (1.6%) mixed bacterial-viral infections; 104 patients (14.3%) were determined to have no infection. TriVerity achieved an AUROC of 0.91 for viral and 0.83 for bacterial infection prediction, indicating high discriminative accuracy. These performances exceeded those of standard biomarkers, including procalcitonin (0.71), C-reactive protein (0.74), and white blood cell count (0.76). The authors concluded that TriVerity may support emergency clinicians in optimizing antimicrobial stewardship by accurately identifying patients who require anti-infective therapy while minimizing unnecessary antibiotic use.
Outcomes and Implications
Current triage tools, such as the quick Sequential Organ Failure Assessment (qSOFA), identify patients with a score >2 as high risk for mortality. However, up to 30% of sepsis-related deaths occur among patients with a qSOFA ≤ 1, underscoring the need for more sensitive early detection tools. Retrospective subgroup analysis demonstrated that TriVerity detected high-risk patients whose illness severity was underestimated by qSOFA alone. When integrated with existing clinical scoring systems, TriVerity may enhance diagnostic precision, expedite clinical decision-making, and improve early management of severe infections and sepsis.