Distinguishing viral versus bacterial lower respiratory tract infection (LRTI) is challenging. We previously developed a rapid, host response-based test (Biomeme HR-B/V assay) using peripheral blood samples to identify viral versus bacterial infection. We assessed the performance of this assay when using nasopharyngeal (NP) samples.

Patients with LRTI were enrolled, and a NP swab sample was run using the HR-B/V assay (assessing 24 gene targets) on the FranklinTM platform. The performance of the prior classifier at identifying viral versus bacterial infection was assessed. A novel predictive model was generated for NP samples using the same 24 targets. Results were validated using external datasets with nasal/NP RNA sequence data.

Nineteen patients (median age 62 years, 52.1% male) were included. When using the prior HR-B/V classifier on NP samples of 19 patients with LRTI (12 viral, 7 bacterial), the area under the receiver operator curve (AUC) for viral versus bacterial infection was 0.786 (0.524–1), with accuracy 0.79 (95% CI 0.57–0.91), positive percent agreement (PPA) 0.43 (95% CI 0.16–0.75), and negative percent agreement (NPA) 1.00 (95% CI 0.76–1). The novel model had AUC 0.881 (95% CI 0.726–1), accuracy 0.84 (95% CI 0.62–0.94), PPA 0.86 (95% CI 0.49–0.97), and NPA 0.83 (95% CI 0.55–0.95) for bacterial infection. Validation in two external datasets showed AUC of 0.932 (95% CI 0.90–0.96) and 0.915 (95% CI 0.88–0.95).

We show that host response in the nasopharynx can distinguish viral versus bacterial LRTI. These findings need to be replicated in larger cohorts with diverse LRTI etiologies.