Characteristics of SARS-CoV-2 testing for rapid diagnosis of COVID-19 during the initial stages of a global pandemic

Accurate SARS-CoV-2 diagnosis is essential to guide prevention and control of COVID-19. Here we examine SARS-CoV-2 molecular-based test performance characteristics and summarize case-level data related to COVID-19 diagnosis. From January 11 through April 22, 2020, Public Health Ontario conducted SARS-CoV-2 testing of 86,942 specimens collected from 80,354 individuals, primarily using real-time reverse-transcription polymerase chain reaction (rRT-PCR) methods. We analyzed test results across specimen types and for individuals with multiple same-day and multi-day collected specimens. Nasopharyngeal compared to throat swabs had a higher positivity (8.8% vs. 4.8%) and an adjusted estimate 2.9 Ct lower (SE = 0.5, p<0.001). Same-day specimens showed high concordance (98.8%), and the median Ct of multi-day specimens increased over time. Symptomatic cases had rRT-PCR results with an adjusted estimate 3.0 Ct (SE = 0.5, p<0.001) lower than asymptomatic/pre-symptomatic cases. Overall test sensitivity was 84.6%, with a negative predictive value of 95.5%. Molecular testing is the mainstay of SARS-CoV-2 diagnosis and testing protocols will continue to be dynamic and iteratively modified as more is learned about this emerging pathogen.


RNA extraction and molecular-based assays
Automated processes were used for RNA extraction and included the NucliSENS easyMAG extractor (bioMérieux, QC, Canada), the m2000 platform (Abbott Inc., IL, USA), and the cobas® 8800 (Roche Diagnostics, Germany). All were used according to manufacturer instructions. For the Roche assay, a 400 µl aliquot of specimen was added to 200 µl of cobas® Omni Lysis Reagent (43% guanidine thiocyanate w/v). The instrument was loaded with a 400 µl aliquot of the above 600 µl combination of specimen/lysis reagent for testing.
Next, a laboratory developed test (LDT) rRT-PCR was used for detection of the envelope (E) gene and/or SARS-CoV-2 RNA-dependent RNA polymerase (RdRp) gene using simplex or duplex methods. Duplex rRT-PCR reactions combined the primers and probe for E-gene or RdRp gene with the internal control human RNase P (RNP), S2 At first E-gene and RdRp were tested in simplex reactions on the same plate. After a period, with an increase in testing volumes and percent positivity, specimens were tested first for the E-gene target due to its greater sensitivity and at the end of the day in batches by RdRp for confirmation of E-gene-positive results. This was later revised to single target detection for COVID-19 diagnosis using the E-gene target.
As of April 3, 2020 an additional rRT-PCR assay using the cobas® 8800 (Roche) was brought online and used for detection of the SARS-CoV-2. The cobas® 8800 is an all-in-one system, which includes rRT-PCR detection of the E-gene and ORF1ab. Testing was performed according to manufacturer's instructions. Specimens tested in the cobas® SARS-CoV-2 rRT-PCR assay were reported as detected or not detected; the manufacturer does not include an indeterminate range. Although the Ct value is obtainable from the instrument, the maximum number of cycles of PCR amplification used in the assay is proprietary and not provided in the kit insert and documentation.
An indeterminate result on a real-time PCR assay was defined as a late amplification signal in an rRT-PCR reaction at a predetermined high Ct value range, which is delineated by the testing laboratory at the time of validation of a LDT, or by the manufacturer for commercial assays. This may be due to low viral target quantity in the clinical specimen approaching the limit of detection of the assay or alternatively, in rare cases, may represent nonspecific reactivity (false signal) in the specimen. When clinically relevant, repeat testing is recommended.
An uninterpretable result was classified as invalid, and reflected a failure in one or more of the test's internal controls. This may be due to any number of issues, such as the presence of PCR inhibitors or poor extraction due to a highly viscous specimen. Where it was not possible to repeat the test and obtain a valid result, the invalid result was reported as such. S2

Statistical Analysis
We measured correlations between Ct values of the E-gene and RdRp targets (LDT), and the E-gene and ORF1ab (Roche) using Pearson correlation coefficients. For Ct comparisons of dual targets within the same assay, the Ct for negative results was set to 40 to not bias single target positive results.

A.I.II. Supplementary Test Performance Characteristics
The majority (99.0%, n=86,071) of specimens were tested with a single assay, the largest number of which were tested using PHO's LDT rRT-PCR assay (n=56,928) or the Roche platform (n=29,141), S3 Table. S3

rRT-PCR assays and targets
To explore the performance of the different rRT-PCR targets, we compared test results using one or both targets across different assays. Of the 57,429 specimens tested by LDT, 3,717 (6.5%) were positive for either E-gene or RdRp-gene (S5 Table) -the majority of which was E-gene alone (78.1%). Where both targets were tested for a specimen (n=12,592), 1,077 (8.6%) were positive for both E-and RdRpgenes, 94 (0.7%) were E-gene positive only, and two specimens were indeterminate for the E-gene with one negative and one positive for RdRp. Of the 29,164 specimens tested using the Roche platform, 3,788 (13.0%) were positive for E-gene and/or ORF1ab gene, 3,507 (12.0%) were positive for both targets, 225 (0.8%) were positive for E-gene only, and 56 (0.2%) were positive for ORF1ab only. There were 7 (0.02%) specimens with invalid results which subsequently tested negative using the LDT assay.
S5 We compared Ct values (where available) for each specimen tested using multiple targets on the same platform, which revealed highly correlated Ct values between targets (LDT R 2 =0.91; Roche R 2 =0.98), S2 Fig. However, we noted that when comparing results of the two targets tested for the same specimen the LDT E-gene Ct was lower than the RdRp-gene Ct for 896 (77.4%) of 1,158 specimens (S6 Table), with a median E-gene Ct of 22.4 (IQR 18.3-27.0), and a median RdRp Ct of 23.2 (IQR 19.5-27.5), p<0.001. In contrast, the Roche E-gene Ct was higher than the ORF1ab Ct for 3,339 (88.6%) of 3,768 specimens (S6 Table), with a median E-gene Ct of 26.7 (IQR 21.0-33.7) and a median ORF1ab Ct of 26.2 (IQR 20.6-31.9), p<0.001. Although the largest differences in Ct values for the two the targets on the same platform was up to 16.4 and 17.1 for the LDT and Roche platforms, respectively, it should be noted that the median differences were less than two Ct. While the median Ct for the E-gene on the Roche platform was slightly higher compared to ORF1ab, it is a more sensitive target as more specimens were E-positive/ORF1abnegative (n=225) than ORF1ab-positive/E-negative (n=56), S5

A.I.III. Supplementary Molecular-based Testing Summary
Examining the different targets and assays used over the first few months of the COVID-19 pandemic showed good correlation between targets within an assay; however, we found that percent positivity and median Ct varied across different assays. This can largely be explained by a bias in assays used based on the swab manufacturer (e.g. Roche swabs used with the cobas® 8800 platform are distributed by PHO largely to hospitals), and the timeline of these assays coming online in the context of the prioritized population for testing and case rate at the time.