Test, trace, isolate: evidence for declining SARS-CoV-2 PCR sensitivity in a clinical cohort

Abstract

Real-time reverse transcription-polymerase chain reaction (RT-PCR) on upper respiratory tract (URT) samples is the primary method to diagnose SARS-CoV-2 infections and guide public health measures, with a supportive role for serology. We reinforce previous findings on limited sensitivity of PCR testing, and solidify this fact by statistically utilizing a firm basis of multiple tests per individual. We integrate stratifications with respect to several patient characteristics such as severity of disease and time since onset of symptoms. Bayesian statistical modelling was used to retrospectively determine the sensitivity of RT-PCR using SARS-CoV-2 serology in 644 COVID-19-suspected patients with varying degrees of disease severity and duration. The sensitivity of RT-PCR ranged between 80% - 95%; increasing with disease severity, it decreased rapidly over time in mild COVID-19 cases. Negative URT RT-PCR results should be interpreted in the context of clinical characteristics, especially with regard to containment of viral transmission based on 'test, trace and isolate'. Keywords: SARS-CoV-2, RT-PCR, serology, sensitivity, public health.

Keywords: RT-PCR; SARS-CoV-2; public health; sensitivity; serology.

Fig. 1

Patient cohort flow-diagram

Fig 2

(A) Histogram of Wantai total antibody index values on a logarithmic scale. The red vertical line indicates the manufacturer's cut-off value, the green vertical line our adapted cut-off value. (B) Percentage of PCR-positivity within Wantai total antibody index values. Since each bar represents a number of patients, the percentage of PCR positive test can be calculated within each bar. The histogram is scaled to 100% height per bar, and colored according to percentage of PCR positive (purple) and negative (yellow) status of the patients. When no patients existed in a certain range of Wantai index values, the bar is left blank. The red vertical line indicates the manufacturer's cut-off value, the green vertical line our adapted cut-off value.

Fig. 3

(A) Posterior uncertainty distributions of the sensitivities for the microarray assays, serology and RT-PCR assay. Colors indicate severity category. (B) Estimates of prevalence of infection by severity category. ICU = Intensive Care Unit; OUTP = outpatients; WARD, non-ICU hospitalized patients; EC50_CoV-N, SARS-CoV-2 N protein microarray; EC50_CoV-S1, SARS-CoV-2 S1 protein microarray; serum_our, Wantai serology using cut-off value of 0.25; PCR, polymerase-chain reaction

Fig. 4

Linear increase of Ct-value in relation to days since onset of symptoms across different disease severity categories. The shaded band indicates 95% Bayesian credible interval. The dots are the original data. Dots positioned at a Ct of 50 were right-censored in the inflated model (i.e. count as either above 50 or a negative individual). ICU = Intensive Care Unit; OUTP = outpatients; WARD, non-ICU hospitalized patients; ct, Ct-value

Fig. 5

Modeled sensitivities as a function of days since onset of symptoms. Panels and shading indicate disease severity. The bands indicate 95% Bayesian credible interval, and black dots an estimate of sensitivity directly from the original data, by assuming Wantai serology as the gold standard, and calculating the ratio of Wantai positive to PCR positive for each day since disease onset. Note that since only a limited number of individuals contribute to each point, points are clustered around fractions with low denominators (e.g. 0/2, 1/2 2/2). Concordance of the curves with the data points should be assessed by observing that the curve runs on average between the dots. ICU = Intensive Care Unit; OUTP, outpatients; WARD, non-ICU hospitalized patients.