Error Rates in SARS-CoV-2 Testing Examined with Bayesian Inference

Abstract

AbstractA literature review on SARS-CoV-2 reverse-transcription polymerase chain reaction (RT-PCR) is used to construct a clinical test confusion matrix. A simple correction method for bulk test results is then demonstrated with examples. The required sensitivity and specificity of a test are explored for societal needs and use cases, before a sequential analysis of common example scenarios is explored. The analysis suggests that many of the people with mild symptoms and positive test results are unlikely to be infected with SARS-CoV-2 in some regions. It is concluded that current and foreseen alternative tests can not be used to “clear” people as being non-infected. Recommendations are given that regional authorities must establish a programme to monitor operational test characteristics before launching large scale testing; and that large scale testing for tracing infection networks in some regions is not viable, but may be possible in a focused way that does not exceed the working capacity of the laboratories staffed by competent experts. RT-PCR tests can not be solely relied upon as the gold standard for SARS-CoV-2 diagnosis at scale, instead clinical assessment supported by a range of expert diagnostic tests should be used.