Experience With False-Positive Test Results on the TaqPath Real-Time Reverse Transcription-Polymerase Chain Reaction Coronavirus Disease 2019 (COVID-19) Testing Platform

Testing was performed in a 96-well format on nasopharyngeal swab specimens collected from 8174 asymptomatic patients (mostly preoperative testing) and 26 209 symptomatic patients in Utah from August 5, 2020 to August 19, 2020. Viral RNA was extracted using the KingFisher Flex System with MagMAX Viral Nucleic Acid Kits. PCR amplification was performed on Applied Biosystems 7500 PCR systems and a QuantStudio 5 PCR system. Per the manufacturer's instructions for use, the Applied Biosystems COVID-19 Interpretive Software was used to call positive and negative test results. Positive test results required detection of at least 2 of 3 viral genes, and negative test results had to be negative for all 3 viral genes. All test kits and instruments were purchased from ThermoFisher, USA. Positive amplification curves and samples' locations on the plates were reviewed. If 2 adjacent positive samples had a cycle threshold (Ct) difference greater than 5 for any 1 of 3 viral genes, the specimen with the higher Ct was re-extracted and retested twice. If both repeats were negative, the result was classified as a false positive. If a blank was positive or if there was concern about a potential gross splash contamination event, then the entire plate was retested. If any of the laboratory technicians performing the testing reported to the lab supervisor that they were concerned a patient sample had been pipetted into the wrong well on the test plate (eg, if only 7 patient specimens were counted when double checking the number of patient specimens used to fill 1 column of 8 wells on a 96-well plate, then there would be concern that 1 patient specimen was accidentally pipetted out twice), or that the 96-well plate had been improperly sealed during the vortexing or amplification steps (eg, if fluid was noted on the surface of the 96-well plate after removing the cover seal), the entire plate was retested as well. When an entire plate was retested, any specimen that had a discordant result from the original run was tested again, so that there would be 3 test results per specimen. If the original result was positive, and the 2 repeats were negative, then the result was classified as a false positive. If the original result was negative, and the 2 repeats were positive, the result was classified as positive. If 1 of the repeats was positive and the other repeat was negative, then the result was classified as indeterminate with a note to the ordering physician that a new specimen be submitted for testing if clinically indicated. A 2-tailed Student's t-test was used for statistical comparisons.

Of 34 383 tests, 2861 (8.3%) were positive and 336 of those met the criteria for retesting; 40 (0.1%) were identified as false positives. During the study period, there were no discordant viral gene test results; in other words, all positive test results were positive for all 3 viral genes and all negative test results were negative for all 3 viral genes on the original and repeated runs. The most common false-positive pattern was a single, weak positive located directly adjacent to a strong positive (Figure 1). This pattern accounted for 39 of the false positives, which were detected based on testing a total of 336 side-by-side positives with a Ct difference more than 5. The remaining single false-positive event was due to a self-reported technician pipetting error that triggered investigation; 2 adjacent wells on the 96-well test plate were accidentally filled with fluid from the same patient specimen tube. This resulted in a false-positive test result with a Ct value that was only 2 greater than the adjacent true positive well.

Three out of 400 96-well plates run had to be entirely retested owing to positive blanks as follows: 2 were caused by cross-over contamination from a positive patient sample, while the third was due to a splash contamination event caused by improper covering of the plate during the specimen vortexing step. The data obtained from the repeat test of the plate with the gross splash contamination were included in our data set, but not the original run, which had 20 false positives in a radial distribution. The N gene Ct value range for false positives was 21.9 to 37.7 and for the associated true positives was 11.8 to 30.3, with the mean Ct values being significantly different (30.2 versus 19.1, P < .001) (Figure 2). The N gene Ct difference between adjacent false positive and true positive samples ranged from 2 to 22 with an average of 11.1 and a standard deviation of 4.0 (Figure 3). Similar results were observed for the ORF1 and S genes. The ORF1 gene Ct value range for false positives was 20.3 to 40.0 and for the associated true positives was 10.6 to 32.0, with the mean Ct values being significantly different (30.1 versus 19.1, P < .001). The ORF1 gene Ct difference between adjacent false-positive and true-positive samples ranged from 4 to 23. The S gene Ct value range for false positives was 21.6 to 40.0 and for the associated true positives was 11.8 to 35.0, with the mean Ct values being significantly different (30.9 versus 20.2, P < .001). The S gene Ct difference between adjacent false-positive and true-positive samples ranged from 3 to 19.

Forty sided-by-side cross-contamination events led to false-positive test results at a rate of approximately 1 in 1000 specimens tested during the study period. The largest Ct difference observed was 21.8. This indicates that a microscopic cross-over contamination event was sufficient to produce a false-positive result, as the false positive well contained 3.6 million-fold less viral genomic material compared with the adjacent true positive well. We have adopted the following measures to prevent and detect false-positive results: (1) reducing the ethanol wash volume on the KingFisher extractions from 1000 to a 500 μL followed by a 250-uL wash to minimize the risk of splashing, (2) using 8 blanks per plate, 2 randomly placed in each quadrant, (3) more frequent servicing of the testing equipment based on weekly environmental swab results and the rate of false positives per machine, (4) tracking lots of plastics used (ie, tip combs), and (5) re-extracting and retesting positives located adjacent to a stronger positive (Ct difference >5). Manually reviewing all positive amplification curves, analyzing their location on the 96-well plates, and comparing the Ct values of adjacent positive specimens have been critical in our quality assurance process.

Visual inspection of the amplification curves enabled us to identify another important cause of false positives, artifactual rectangular-shaped curves that rise rapidly at low cycle times and then flatline, several months before an FDA letter⁶  that was sent to all health care providers using the TaqPath COVID-19 Combo Kit in regard to this issue. We had already implemented the FDA suggested corrective actions before this study, and no false-positive results in this data set were caused by these artifactual rectangular shaped curves. Manual review of amplification curves has also enabled us to catch false negatives, weak curves arising at high Ct values that do not cross the interpretative software's threshold.

Owing to cross-contamination events usually producing normal appearing amplification curves, we have been able to identify these false-positive results based on their pattern of distribution on the 96-well plates in combination with the Ct value comparisons described in this study. We believe that the recommendations provided above could help to improve the accuracy of SARS-CoV-2 testing for laboratories around the world regardless of the rRT-PCR platform used.