Table 3 Performance characteristics for Global Fever classifier models for acute bacterial and viral infection.

From: Host-response transcriptional biomarkers accurately discriminate bacterial and viral infections of global relevance

Cohort

Comparison

Sensitivity, % (95% CI)

Specificity, % (95% CI)

Model accuracy, % (95% CI)

Positive likelihood ratio (95% CI)

Negative likelihood ratio (95% CI)

Global fever bacterial/viral model (GF-B/V)

 Discovery

Bacterial versus viral

84.2 (75.6–90.7)

94.7 (88.6–97.7)

89.7 (85.0–93.4)

14.7 (7.2–30.5)

0.2 (0.1–0.3)

 Validation

Bacterial versus viral

78.8 (65.3–88.9)

84.3 (71.4–93.0)

81.6 (72.7–88.5)

5.0 (2.6–9.6)

0.3 (0.1–0.4)

Global fever bacterial/viral/noninfectious model (GF-B/V/N)

 Discovery

Bacterial versus nonbacterial

87.7 (79.0–89.8)

84.2 (78.2–89.1)

85.2 (80.6–89.1)

5.5 (3.9–7.7)

0.2 (0.1–0.3)

Viral versus nonviral

83.7 (76.0–89.8)

81.5 (74.8–87.1)

82.5 (77.6–86.7)

4.5 (3.3–6.3)

0.2 (0.1–0.3)

 Validation

Bacterial versus nonbacterial

82.7 (69.7–91.8)

80.4 (66.9–90.2)

81.6 (72.7–88.5)

4.2 (2.4–7.5)

0.2 (0.1–0.4)

Viral versus nonviral

76.5 (62.5–87.2)

80.8 (67.5–90.4)

78.6 (69.5–86.1)

4.0 (2.237.1)

0.3 (0.2–0.5)

  1. The top of the table provides performace characteristics for the GF-B/V model and the bottom of the table shows performance of the GF-B/V/N model. In the discovery cohort, performance characteristics are calculated using nested cross validation on the original RNA sequencing data. In the validation cohort, the model is fixed and applied to NanoString data of an independent bacterial and viral cohort. Positive and negative predictive value requires knowledge of prevalence in the community which is not known for global infections. Thus, these could not be calculated.
Back to article page