Clinical validation of an AI-based blood testing device for diagnosis and prognosis of acute infection and sepsis
Recommended Citation
Liesenfeld O, Arora S, Aufderheide TP, Clements CM, DeVos E, Fischer M, Giamarellos-Bourboulis EJ, House S, Humphries RL, Gill JK, Liu E, Mace SE, May L, Michelson E, Osborn TM, Panacek E, Rothman RE, Self WH, Smithline HA, Steingrub J, Van Heukelom P, Weissman A, Wilson M, Wolk DM, Wright DW, Buturovic L, Hasin-Brumshtein Y, Damaraju N, Lu C, Shak JR, Whitfield NN, Khatri P, Sweeney TE, and Shapiro NI. Clinical validation of an AI-based blood testing device for diagnosis and prognosis of acute infection and sepsis. Nat Med 2025;31(12):4044-4045.
Document Type
Article
Publication Date
12-1-2025
Publication Title
Nature medicine
Keywords
Humans, Sepsis, Female, Prognosis, Male, Middle Aged, Aged, Procalcitonin, Bacterial Infections, ROC Curve, Machine Learning, Artificial Intelligence, C-Reactive Protein, Severity of Illness Index, Virus Diseases, Algorithms
Abstract
Lack of reliable diagnostics for the presence, type and severity of infection in patients presenting to emergency departments with non-specific symptoms poses considerable challenges. We developed TriVerity, which uses isothermal amplification of 29 mRNAs and machine learning algorithms on the Myrna instrument to determine likelihoods of bacterial infection, viral infection and need for critical care interventions within 7 days. To validate TriVerity, the SEPSIS-SHIELD study enrolled 1,222 patients with clinically adjudicated infection status and need for critical care intervention within 7 days as endpoints. The TriVerity Bacterial and Viral scores had higher accuracy than C-reactive protein, procalcitonin or white blood cell count for the diagnosis of bacterial infection with area under the receiver operating characteristic (AUROC) of 0.83, and viral infection (AUROC = 0.91). The TriVerity Severity score had an AUROC of 0.78 for predicting illness severity and allowed reclassification of risk for critical care interventions compared to clinical assessment (quick Sequential Organ Failure Assessment) alone. Each of the three scores had rule-in specificity >92% and rule-out sensitivity >95%. Comparison of antibiotics administration at presentation with post-follow-up adjudication found that TriVerity could potentially reduce false positives and false negatives for inappropriate antibiotics use by 60-70%. Further clinical testing in an interventional setting is needed to prove actionability and clinical benefit of TriVerity.
Medical Subject Headings
Humans; Sepsis; Female; Prognosis; Male; Middle Aged; Aged; Procalcitonin; Bacterial Infections; ROC Curve; Machine Learning; Artificial Intelligence; C-Reactive Protein; Severity of Illness Index; Virus Diseases; Algorithms
PubMed ID
41028541
ePublication
ePub ahead of print
Volume
31
Issue
12
First Page
4044
Last Page
4054
