Stringent thresholds in SARS-CoV-2 IgG assays lead to under-detection of mild infections

Abstract

Background: Thresholds for SARS-CoV-2 antibody assays have typically been determined using samples from symptomatic, often hospitalised, patients. In this setting the sensitivity and specificity of the best performing assays can both exceed 98%. However, antibody assay performance following mild infection is less clear.

Methods: We assessed quantitative IgG responses in a cohort of healthcare workers in Oxford, UK, with a high pre-test probability of Covid-19, in particular the 991/11,475(8.6%) who reported loss of smell/taste. We use anosmia/ageusia and other risk factors as probes for Covid-19 infection potentially undiagnosed by immunoassays by investigating their relationship with antibody readings either side of assay thresholds.

Results: The proportion of healthcare workers reporting anosmia/ageusia increased at antibody readings below diagnostic thresholds using an in-house ELISA (n = 9324) and the Abbott Architect chemiluminescent microparticle immunoassay (CMIA; n = 11,324): 426/906 (47%) reported anosmia/ageusia with a positive ELISA, 59/449 (13.1%) with high-negative and 326/7969 (4.1%) with low-negative readings. Similarly, by CMIA, 518/1093 (47.4%) with a positive result reported anosmia/ageusia, 106/686 (15.5%) with a high-negative and 358/9563 (3.7%) with a low-negative result. Adjusting for the proportion of staff reporting anosmia/ageusia suggests the sensitivity of both assays in mild infection is lower than previously reported: Oxford ELISA 89.8% (95%CI 86.6-92.8%) and Abbott CMIA 79.3% (75.9-82.7%).

Conclusion: Following mild SARS-CoV-2 infection 10-30% of individuals may have negative immunoassay results. While lowered diagnostic thresholds may result in unacceptable specificity, our findings have implications for epidemiological analyses and result interpretation in individuals with a high pre-test probability. Samples from mild PCR-confirmed infections should be included in SARS-CoV-2 immunoassay evaluations.

Keywords: Ageusia; Anosmia; Antibodies; COVID-19; SARS-CoV-2; Serology.

Fig. 1

SARS-CoV-2 IgG antibody readings in 245 convalescent symptomatic healthcare workers ≥ 14 days following a positive PCR test. Panel a shows readings using the Oxford ELISA assay targeting trimeric spike protein (n = 171) and panel b shows readings using the Abbott CMIA targeting nucleocapsid protein (n = 240). The dashed horizontal lines show the pre-defined threshold for reporting antibody detection (Oxford ELISA 8 million, Abbott CMIA 1.4). The solid line and ribbon shows the fitted mean value and 95% confidence interval using a linear regression model with a 3 knot spline. Points are coloured by whether staff reported previous anosmia (loss of smell) and/or ageusia (loss of taste) since 01 February 2020 when asked prior to serological testing: 159/245 staff (65%) reported anosmia/ageusia

Fig. 2

Proportion of staff with anosmia or loss of taste by antibody reading. Panel a shows the results using a trimeric spike ELISA and panel b the results from the Abbott CMIA targeting nucleocapsid protein, with blue showing results called negative and red showing those called as positive based on pre-defined assay thresholds. The number of individuals with these symptoms is shown in each bar, and the total number of individuals with each antibody reading below the bar. The error bars show 95% confidence intervals. For the Oxford ELISA readings each value is rounded down, such that for example a value of 1.7 million is within the 1 million bar

Fig. 3

The relationship between Abbott CMIA and Oxford ELISA readings and loss of smell or taste in 9191 samples tested by both platforms. Panel a compares the number of individuals with combinations of Abbott CMIA and Oxford ELISA readings, the size of each circle represents the number of individuals and the colour the proportion reporting loss of smell or taste. Panel b groups the data by positive, high-negative (equivocal) and low-negative readings for both assays, the numbers shown beneath each bar are the number of individuals reporting loss of smell/taste and the total number of individuals with the antibody reading

Fig. 4

Ideal (panel a) and actual (panel b) relationship between the probability of Covid-19 and antibody reading. The probability of Covid-19 in panel B was generated from a multivariable model containing risk factors including Covid-19 exposures in the community and at work, ethnicity, healthcare worker role and specialty area worked in [3]