Evaluation of the clinical performance of seven serological assays for SARS-CoV-2 for use in clinical laboratories.

Background Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) serological assays have emerged as a response to the global pandemic, warranting studies evaluating their clinical performance. This study investigated seven commercially available SARS-CoV-2 serological assays in samples from non-infected individuals and hospitalized patients. Methods SARS-CoV-2 qualitative serological assays by Abbott (IgG), Beckman (IgG), DiaSorin (IgG), EUROIMMUN (IgG and IgA), Roche and Bio-Rad (Total) were evaluated using specimens collected pre-December 2019 (n=393), from nucleic acid amplification testing (NAAT) negative patients (n=40), and from 53 patients with COVID-19 by NAAT collected 3-21 days post-onset of symptoms (POS) (N=83). Negative agreement (NA), positive agreement (PA), and positive and negative predictive values (PPV and NPV) at prevalences of 5% and 10% were calculated. Results The overall %NA;95% CI in the negative samples were: Roche 99.8%; 99.3-100.2, Beckman 99.8%; 98.7-100.0, Abbott and Bio-Rad 99.3%; 98.0-99.9, DiaSorin 98.4; 97.2-99.6, EUROIMMUN IgG 97.5%; 95.5-98.7, and EUROIMMUN IgA 79.7%; 75.9-83.5), accounting for positive/equivocal results as false positives. The %PA; 95% CI in samples collected 14+days POS (n=24) were: Bio-Rad 83.3%; 68.4-98.2, Abbott and Roche 79.2%; 62.9-95.4, EUROIMMUN IgA 70.8%; 52.6-89.0, Beckman 58.3%; 38.6-78.1, DiaSorin 54.2; 34.2-74.1, and EUROIMMUN IgG 50.0%; 30.0-70.0, accounting for negative/equivocal results as false negatives. NPVs ranged from 97.4-98.9% and 94.7-97.7% for prevalences 5% and 10%, respectively. PPVs ranged from 15.5-94.8% and 27.9-97.4% for prevalences 5% and 10%, respectively. Conclusions The Roche and Beckman assays resulted in fewer false positives followed by the Bio-Rad and Abbott assays. While the Bio-Rad assay demonstrated higher antibody detection in COVID-19-positive patients, PA claims cannot be established with a high level of confidence in our sample population.