Estimation of the diagnostic accuracy of real-time reverse transcription quantitative polymerase chain reaction for SARS-CoV-2 using re-analysis of published data.

Introduction: As the coronavirus disease 2019 (COVID-19) epidemic evolves and test strategies change, understanding the concepts of testing (gold standard and test performance measures) becomes essential. The challenge of any novel disease is that the gold standard has yet to be defined.

Methods: We reanalysed published data on real-time reverse transcription quantitative polymerase chain reaction (RT-qPCR) of severe acute respiratory syndrome coronavirus-2 to illustrate how predictive values vary with disease prevalence, sensitivity (set to values between 30% and 95%) and specificity (set to 99% or 99.98%). We used published data on chest CT and RT-qPCR to examine the potential of latent class analysis to estimate the sensitivity and specificity of RT-qPCR when no single gold standard exists.

Results: For the various sensitivity values, the negative predictive value of a RT-qPCR test remained above 92% until a COVID-19 prevalence of > 10%. The positive predictive value (PPV) was more variable. For a sensitivity of 95% and a specificity of 99%, the PPV was less-than 10% at a prevalence of 0.1%, increasing to about 90% at a prevalence of 10%. This improved to a PPV of 85% and almost 100%, respectively, when specificity increased to 99.98%. In a restricted latent class analysis, the sensitivity was 97.1% and the specificity was 99.9%, which is similar to figures from the Danish Health Authority. However, derived predictive values depended on model specification.

Conclusions: A high risk of false-positives should be considered when extending the testing strategy, whereas false-negatives may occur during local outbreaks. This may have consequences for, e.g., containment strategies and research. A confirmatory test (e.g., demonstrating seroconversion or repeated RT-qPCR) may be warranted.

Funding: none.

Trial Registration: not relevant.