Can a host-response bacterial-viral classifier safely guide antibiotic avoidance in COVID-19? A diagnostic accuracy study in hospitalized adults

Dagfinn Lunde Markussen; Christina Skår Saghaug; Siri Tandberg Knoop; Christian Ritz; Elling Ulvestad; Synne Jenum; Harleen M S Grewal

doi:10.1093/jacamr/dlag021

Can a host-response bacterial-viral classifier safely guide antibiotic avoidance in COVID-19? A diagnostic accuracy study in hospitalized adults

JAC Antimicrob Resist. 2026 Feb 20;8(1):dlag021. doi: 10.1093/jacamr/dlag021. eCollection 2026 Feb.

Authors

Dagfinn Lunde Markussen^{1

2}, Christina Skår Saghaug¹, Siri Tandberg Knoop³, Christian Ritz^{1

4}, Elling Ulvestad^{1

3}, Synne Jenum^{1

5}, Harleen M S Grewal^{1

3}

Affiliations

¹ Department of Clinical Science, Bergen Integrated Diagnostic Stewardship Cluster, Faculty of Medicine, University of Bergen, Bergen 5020, Norway.
² Department of Emergency Medicine, Haukeland University Hospital, Bergen 5021, Norway.
³ Department of Microbiology, Haukeland University Hospital, Bergen 5021, Norway.
⁴ National Institute of Public Health, University of Southern Denmark, Copenhagen, Denmark.
⁵ Department of Infectious Diseases, Oslo University Hospital, Oslo, Norway.

Abstract

Background and objectives: The MeMed BV^® test distinguishes bacterial from viral infections by integrating circulating levels of TRAIL, IP-10, and CRP into a likelihood score. Pre-COVID studies reported high diagnostic accuracy in respiratory infections, but evidence in adults hospitalized with SARS-CoV-2 is sparse. To evaluate the ability of MeMed BV to identify hospitalized COVID-19 patients who could be safely managed without antibiotics.

Methods: We included adults with PCR-confirmed SARS-CoV-2 infection from two prospective Norwegian cohorts (October 2020-April 2023). The reference standard for true viral infection was a composite safety outcome: 30-day survival, no readmission within 30 days, and no antibiotic exposure before, during, or after hospitalization. MeMed BV performance was assessed using sensitivity, specificity, predictive values, and area under the ROC curve (AUC).

Results: Of 90 patients, 28 (31%) met the reference standard for safe antibiotic avoidance. Median MeMed BV scores were 38 (IQR 1-66) in the safely managed group versus 96 (IQR 78-99) in others (P < 0.001). Sensitivity for identifying safe antibiotic avoidance was 50.0% (95% CI 32.6-67.4), specificity 88.7% (95% CI 78.5-94.4), PPV 66.7%, NPV 79.7%, and AUC 0.69 (95% CI 0.59-0.80). Patients safely managed without antibiotics misclassified as bacterial/equivocal had markedly lower TRAIL (median 38 versus 177 pg/mL, P < 0.001) and higher CRP.

Conclusions: In hospitalized adults with COVID-19, MeMed BV showed reasonable specificity but limited sensitivity for identifying patients who could safely avoid antibiotics, with performance considerably lower than that observed in pre-COVID studies. These findings highlight the need for context-specific validation of host-response diagnostics.