Estimating the false positive rate of highly automated SARS-CoV-2 nucleic acid amplification testing

Abstract

Molecular testing for infectious diseases is generally both very sensitive and specific. Well-designed PCR primers rarely cross-react with other analytes, and specificities seen during test validation are often 100%. However, analytical specificities measured during validation may not reflect real-world performance across the entire testing process. Here, we use the unique environment of SARS-CoV-2 screening among otherwise well individuals to examine the false positivity rate of high throughput so-called sample-to-answer nucleic acid amplification testing (NAAT) on three commercial assays: the Hologic Panther Fusion(R), Hologic Aptima(R) transcription mediated amplification (TMA), and Roche cobas(R) 6800. We used repetitive sampling of the same person as the gold standard to determine test specificity rather than retesting of the same sample. We examined 451 people repetitively sampled over 7 months via nasal swab, comprising 7,242 results. During the study period there were twelve positive tests (0.17%) from 9 people. Eight positive tests (0.11%, five individuals) were considered bona fide true positives based on repeat positives or outside testing and epidemiological data. One positive test had no follow-up testing or metadata and could not be adjudicated. Three positive tests (three individuals) did not repeat as positive on a subsequent collection, nor did the original positive specimen test positive on an orthogonal platform. We consider these three tests false positives and estimate the overall false positive rate of high-throughput automated, sample-to-answer NAAT testing to be approximately 0.041% (3/7242). These data help laboratorians, epidemiologists, and regulators understand specificity and positive predictive value associated with high-throughput NAAT testing.