Determining the SARS-CoV-2 serological immunoassay test performance indices based on the test results frequency distribution.

Abstract

Introduction: Coronavirus disease 2019 (COVID-19) is known to induce robust antibody response in most of the affected individuals. The objective of the study was to determine if we can harvest the test sensitivity and specificity of a commercial serologic immunoassay merely based on the frequency distribution of the SARS-CoV-2 immunoglobulin (Ig) G concentrations measured in a population-based seroprevalence study.

Materials and methods: The current study was conducted on a subset of a previously published dataset from the canton of Geneva. Data were taken from two non-consecutive weeks (774 samples from May 4-9, and 658 from June 1-6, 2020). Assuming that the frequency distribution of the measured SARS-CoV-2 IgG is binormal (an educated guess), using a non-linear regression, we decomposed the distribution into its two Gaussian components. Based on the obtained regression coefficients, we calculated the prevalence of SARS-CoV-2 infection, the sensitivity and specificity, and the most appropriate cut-off value for the test. The obtained results were compared with those obtained from a validity study and a seroprevalence population-based study.

Results: The model could predict more than 90% of the variance observed in the SARS-CoV-2 IgG distribution. The results derived from our model were in good agreement with the results obtained from the seroprevalence and validity studies. Altogether 138 of 1432 people had SARS-CoV-2 IgG ≥ 0.90, the cut-off value which maximized the Youden's index. This translates into a true prevalence of 7.0% (95% confidence interval 5.4% to 8.6%), which is in keeping with the estimated prevalence of 7.7% derived from our model. Our model can provide the true prevalence.

Conclusions: Having an educated guess about the distribution of test results, the test performance indices can be derived with acceptable accuracy merely based on the test results frequency distribution without the need for conducting a validity study and comparing the test results against a gold-standard test.