Cohen's kappa statistics as a convenient means to identify accurate SARS-CoV-2 rapid antibody tests

There are many available rapid antibody tests, but the performance of such tests remains unclear. Moreover, it is difficult to compare among the various devices regarding their sensitivity & specificity. In order to compare the performance of such devices, we used Cohen9s kappa statistics to assess the level of agreement between RT-PCR and rapid antibody tests. In doing this study, we considered the term of validity after symptom-onset to compare two tests. It takes more than a week to produce antibodies in the body, and RT-PCR thus gives negative result in the convalescent period. On ELISA data from the literature kappa statistics was calculated as 1.0 beyond 10 days after symptom-onset. By taking these factors into consideration, we evaluated agreement with samples collected beyond 10 days of symptom-onset during the active period. We calculated the data from 9 devices, and the kappa statistics for English data were calculated as 0.64 on average. The same finding was 0.75 for Chinese data. These results corresponded with the values from sensitivity & specificity of their reports. Both reports had no details about the collection procedures. Kappa statistics might become even more accurate, if samples could be restricted to ones collected beyond 10 days. Regarding the data from our hospital9, the kappa statistics was 0.97 when restricted to samples collected beyond 10 days, which thus showed excellent agreement. By using kappa statistics, the performances of rapid antibody tests can be shown as one figure, so that their comparison becomes easy to carry out.