The Prognostic Value of an RT-PCR Test for Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Is Contingent on Timing across Disease Time Course in addition to Assay Sensitivity.

To the Editor-in-Chief:

We read with great interest the article by Tian et al in a recent issue of the Journal of Molecular Diagnostics that showed a significant overlap in RT-PCR cycle threshold (Ct) value among spreader and non-spreader individuals. The study also discussed the limited potential to identify individuals as spreaders based on the viral load as detected from a nasal swab, showing that single Ct values obtained from serial surveillance testing at the individual level provides little diagnostic value for differential case management. The study also comments that “instead, a sensitive method to detect the presence of virus is needed to identify asymptomatic individuals who may carry a low viral load but can still be infectious.” , pp.1078 However, RT-PCR Ct value is a sensitive method to detect and quantify the presence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus.2, 3, 4 The reason that single Ct values obtained from serial surveillance testing provide little diagnostic value for differential case management is that the Ct value obtained depends critically on when during the disease time course the measurement is taken, , much more than the spreader or non-spreader status of the subject. Therefore, a more sensitive test than RT-PCR may not address the issue of overlap in Ct among spreaders and nonspreaders and among symptomatic patients and nonsymptomatic patients.

A single measurement of viral load at a single time point in an asymptomatic individual provides very little insight into prospective transmission because of the dynamic changes in viral load that are characteristic of the SARS-CoV-2 disease time course , , and consequent RT-PCR diagnostic specificity (Figure 1 ). , , , 7, 8, 9 During the latent period immediately following exposure, viral load is very low and likely localized, and nasal swabs are unlikely to recover any virus: an RT-PCR test may yield a negative result. Viral load increases steeply later in incubation, when both symptomatic and asymptomatic cases become infectious and the RT-PCR test becomes sensitive enough to detect virus. , , , ,

Figure 1

Sensitivity of RT-PCR for detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) across the COVID-19 disease time course from infection. The approximately 3.5-day period during which students surveilled in the study by Tian et al are likely to first test positive spans a rapid rise in viral load early in disease. Sensitivity when sampling only symptomatic cases (green) is based on Miller et al, He et al,, and Qin et al using the approach of Wells et al. Sensitivity when sampling currently asymptomatic cases (blue) is based on data from Hellewell et al, He et al,, and Qin et al using the approach of Wells et al.

In the college population studied by Tian et al, currently asymptomatic students were tested twice a week (approximately once per 3.5 days). Because of the distinct temporal sensitivity of RT-PCR (or indeed, any conceivable sample-based test), a consequence of testing every approximately 3.5 days is that positive tests will typically have a uniform probability of identifying infected students across a biweekly window starting at the point when swabs begin to sample virus and RT-PCR becomes sensitive. Because RT-PCR is a sensitive test for quantitative detection of specific RNA,2, 3, 4 this 3.5-day window spans an extremely dynamic range (many orders of magnitude) of viral load (Figure 1). , , Most of the variance in these measurements corresponds to the uniformly distributed variate (happenstance) of when during their infection the student was sampled, and has much less to do with peak viral load. Peak viral load is very poorly measured by randomly sampling one time point across 3.5 days that includes the viral exponential growth phase, and it is the eventual peak viral load that is most likely to correspond with future symptomatic versus asymptomatic status.

The high variance in viral load during late incubation due to rapid viral growth across this period explains why Ct measurements by Tian et al spanned such a wide range of Ct values (Figure 3E in Tian et al), with such a slight difference in cumulative frequency toward lower Ct values (higher viral load) in symptomatic individuals compared to asymptomatic cases. Sampling the RT-PCR cycle threshold of many infected individuals in the population enables detection of that shift because of the law of large numbers: a very high sample size addresses the very high inter-individual variance, enabling Ct values to be useful for estimating epidemiological dynamics from cross-sectional viral load distributions.

The interesting and useful results published by Tian et al suggest that a sensitive method to detect the presence of virus is needed. An alternative conclusion could be that consideration of the timing of measurements of viral load across the disease time course is vital to their prognostic value. A high prognostic value test would not come from increased sensitivity, but instead from many fine-scale or exquisitely timed measurements that capture dynamic changes in viral load that may be statistically associated with the peak levels experienced by subjects when they become most infectious.