Insight into the practical performance of RT-PCR testing for SARS-CoV-2 using serological data: a cohort study

Abstract

Background: Virological detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) through RT-PCR has limitations for surveillance. Serological tests can be an important complementary approach. We aimed to assess the practical performance of RT-PCR-based surveillance protocols and determine the extent of undetected SARS-CoV-2 infection in Shenzhen, China.

Methods: We did a cohort study in Shenzhen, China and attempted to recruit by telephone all RT-PCR-negative close contacts (defined as those who lived in the same residence as, or shared a meal, travelled, or socially interacted with, an index case within 2 days before symptom onset) of all RT-PCR-confirmed cases of SARS-CoV-2 detected since January, 2020, via contact tracing. We measured anti-SARS-CoV-2 antibodies in serum samples from RT-PCR-negative close contacts 2-15 weeks after initial virological testing by RT-PCR, using total antibody, IgG, and IgM ELISAs. In addition, we did a serosurvey of volunteers from neighbourhoods with no reported cases, and from neighbourhoods with reported cases. We assessed rates of infection undetected by RT-PCR, performance of RT-PCR over the course of infection, and characteristics of individuals who were seropositive on total antibody ELISA but RT-PCR negative.

Findings: Between April 12 and May 4, 2020, we enrolled and collected serological samples from 2345 (53·0%) of 4422 RT-PCR-negative close contacts of cases of RT-PCR-confirmed SARS-CoV-2. 1175 (50·1%) of 2345 were close contacts of cases diagnosed in Shenzhen with contact tracing details, and of these, 880 (74·9%) had serum samples collected more than 2 weeks after exposure to an index case and were included in our analysis. 40 (4·5%) of 880 RT-PCR-negative close contacts were positive on total antibody ELISA. The seropositivity rate with total antibody ELISA among RT-PCR-negative close contacts, adjusted for assay performance, was 4·1% (95% CI 2·9-5·7), which was significantly higher than among individuals residing in neighbourhoods with no reported cases (0·0% [95% CI 0·0-1·1]). RT-PCR-positive individuals were 8·0 times (95% CI 5·3-12·7) more likely to report symptoms than those who were RT-PCR-negative but seropositive, but both groups had a similar distribution of sex, age, contact frequency, and mode of contact. RT-PCR did not detect 48 (36% [95% CI 28-44]) of 134 infected close contacts, and false-negative rates appeared to be associated with stage of infection.

Interpretation: Even rigorous RT-PCR testing protocols might miss a substantial proportion of SARS-CoV-2 infections, perhaps in part due to difficulties in determining the timing of testing in asymptomatic individuals for optimal sensitivity. RT-PCR-based surveillance and control protocols that include rapid contact tracing, universal RT-PCR testing, and mandatory 2-week quarantine were, nevertheless, able to contain community spread in Shenzhen, China.

Funding: The Bill & Melinda Gates Foundation, Special Foundation of Science and Technology Innovation Strategy of Guangdong Province, and Key Project of Shenzhen Science and Technology Innovation Commission.

Figure 1

Timing of serological testing and seropositive results relative to last putative exposure Time of serological testing from last putative exposure to an index case, among all PCR-negative close contacts (A) and among those with seropositive results (B). All close contacts had one serological test each. The Shenzhen cohort was defined as individuals who were included in a previous study that characterised the epidemiology and transmission of COVID-19 in Shenzhen, by Bi and colleagues.

Figure 2

Time of RT-PCR test and time of symptom onset from last exposure to an index case (A) Time of RT-PCR tests among seropositive close contacts who were negative on RT-PCR (n=40), for those who had symptoms (n=3) and who did not have symptoms (n=37) before the end of quarantine. (B) Time of RT-PCR tests among infected close contacts who were either positive on RT-PCR (n=75) or who were negative on RT-PCR but later tested seropositive (n=40). (C) Time of symptom onset among symptomatic infected close contacts who were either positive on RT-PCR (n=55) or who were negative on RT-PCR but later tested seropositive (n=3). Two RT-PCR-positive close contacts had missing dates of last contact with an index case and were not shown in panel C. 18 individuals who were included in panel B but excluded in panel C were either asymptomatic or missing symptom onset dates.

Figure 3

False-negative rates of RT-PCR Probability of false-negative RT-PCR test of nasopharyngeal swab, by time since symptom onset (A) and time since last exposure to an index case (B). Point estimates and 95% CIs represent estimates from the Bayesian logistic regression model for test sensitivity with a polynomial spline of third degree. The solid curve represents estimates from the generalised additive model fitted to Shenzhen data only. The dashed curve represents marginal estimates from the generalised additive model with random effects by study fitted to the combined Shenzhen data and pooled data from Kucirka and colleagues. The vertical dashed line corresponds to time of symptom onset.