Fig. 3: Comparison of the effect of different positivity thresholds on percent positivity and specificity.

a, The percent positivity of each assay tested on serum from patients who tested positive for SARS-CoV-2 by RT–PCR is plotted by time after patient-reported symptom onset. Biologically independent samples for each test are as follows: n = 126, Biomedomics; n = 126, Bioperfectus; n = 124, DecomBio; n = 128, DeepBlue; n = 114, Innovita; n = 127, Premier; n = 127, Sure; n = 128, UCP; n = 119, VivaChek; n = 124, Wondfo. Squares indicate percent positivity using reader score >0 (‘Weak bands positive’) as the positivity threshold. Triangles indicate percent positivity using reader score >1 (‘Weak bands negative’) as the positivity threshold. ‘IgM or IgG’ signifies detection of either isotype. Wondfo reports a single band for IgM and IgG together, and the results are plotted here as both ‘IgM’ and ‘IgG’ for horizontal comparison across assays. b, Comparison of percent positivity at each timepoint for BioMedomics assay at either the MGH (left, n = 48) or UCSF (right, n = 126) study site using low (square) or high (triangle) positivity thresholds. Note that a weak score at MGH is not directly equivalent to a 1 at UCSF owing to differences in reader training. c, The specificity of all assays on historical pre-COVID-19 serum using low (square) or high (triangle) positivity thresholds. UCSF BioMedomics data are plotted again in the right subpanel column for direct comparison to MGH BioMedomics data. All nodes refer to the calculated percent positivity or specificity (as designated), and all error bars indicate 95% CIs.