Background

The severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2) virus and resulting COVID‐19 pandemic present important diagnostic challenges. Several diagnostic strategies are available to identify current infection, rule out infection, identify people in need of care escalation, or to test for past infection and immune response. Serology tests to detect the presence of antibodies to SARS‐CoV‐2 aim to identify previous SARS‐CoV‐2 infection, and may help to confirm the presence of current infection.

Objectives

To assess the diagnostic accuracy of antibody tests to determine if a person presenting in the community or in primary or secondary care has SARS‐CoV‐2 infection, or has previously had SARS‐CoV‐2 infection, and the accuracy of antibody tests for use in seroprevalence surveys.

Search methods

We undertook electronic searches in the Cochrane COVID‐19 Study Register and the COVID‐19 Living Evidence Database from the University of Bern, which is updated daily with published articles from PubMed and Embase and with preprints from medRxiv and bioRxiv. In addition, we checked repositories of COVID‐19 publications. We did not apply any language restrictions. We conducted searches for this review iteration up to 27 April 2020.

Selection criteria

We included test accuracy studies of any design that evaluated antibody tests (including enzyme‐linked immunosorbent assays, chemiluminescence immunoassays, and lateral flow assays) in people suspected of current or previous SARS‐CoV‐2 infection, or where tests were used to screen for infection. We also included studies of people either known to have, or not to have SARS‐CoV‐2 infection. We included all reference standards to define the presence or absence of SARS‐CoV‐2 (including reverse transcription polymerase chain reaction tests (RT‐PCR) and clinical diagnostic criteria).

Data collection and analysis

We assessed possible bias and applicability of the studies using the QUADAS‐2 tool. We extracted 2x2 contingency table data and present sensitivity and specificity for each antibody (or combination of antibodies) using paired forest plots. We pooled data using random‐effects logistic regression where appropriate, stratifying by time since post‐symptom onset. We tabulated available data by test manufacturer. We have presented uncertainty in estimates of sensitivity and specificity using 95% confidence intervals (CIs).

Main results

We included 57 publications reporting on a total of 54 study cohorts with 15,976 samples, of which 8526 were from cases of SARS‐CoV‐2 infection. Studies were conducted in Asia (n = 38), Europe (n = 15), and the USA and China (n = 1). We identified data from 25 commercial tests and numerous in‐house assays, a small fraction of the 279 antibody assays listed by the Foundation for Innovative Diagnostics. More than half (n = 28) of the studies included were only available as preprints.

We had concerns about risk of bias and applicability. Common issues were use of multi‐group designs (n = 29), inclusion of only COVID‐19 cases (n = 19), lack of blinding of the index test (n = 49) and reference standard (n = 29), differential verification (n = 22), and the lack of clarity about participant numbers, characteristics and study exclusions (n = 47). Most studies (n = 44) only included people hospitalised due to suspected or confirmed COVID‐19 infection. There were no studies exclusively in asymptomatic participants. Two‐thirds of the studies (n = 33) defined COVID‐19 cases based on RT‐PCR results alone, ignoring the potential for false‐negative RT‐PCR results. We observed evidence of selective publication of study findings through omission of the identity of tests (n = 5).

We observed substantial heterogeneity in sensitivities of IgA, IgM and IgG antibodies, or combinations thereof, for results aggregated across different time periods post‐symptom onset (range 0% to 100% for all target antibodies). We thus based the main results of the review on the 38 studies that stratified results by time since symptom onset. The numbers of individuals contributing data within each study each week are small and are usually not based on tracking the same groups of patients over time.

Pooled results for IgG, IgM, IgA, total antibodies and IgG/IgM all showed low sensitivity during the first week since onset of symptoms (all less than 30.1%), rising in the second week and reaching their highest values in the third week. The combination of IgG/IgM had a sensitivity of 30.1% (95% CI 21.4 to 40.7) for 1 to 7 days, 72.2% (95% CI 63.5 to 79.5) for 8 to 14 days, 91.4% (95% CI 87.0 to 94.4) for 15 to 21 days. Estimates of accuracy beyond three weeks are based on smaller sample sizes and fewer studies. For 21 to 35 days, pooled sensitivities for IgG/IgM were 96.0% (95% CI 90.6 to 98.3). There are insufficient studies to estimate sensitivity of tests beyond 35 days post‐symptom onset. Summary specificities (provided in 35 studies) exceeded 98% for all target antibodies with confidence intervals no more than 2 percentage points wide. False‐positive results were more common where COVID‐19 had been suspected and ruled out, but numbers were small and the difference was within the range expected by chance.

Assuming a prevalence of 50%, a value considered possible in healthcare workers who have suffered respiratory symptoms, we would anticipate that 43 (28 to 65) would be missed and 7 (3 to 14) would be falsely positive in 1000 people undergoing IgG/IgM testing at days 15 to 21 post‐symptom onset. At a prevalence of 20%, a likely value in surveys in high‐risk settings, 17 (11 to 26) would be missed per 1000 people tested and 10 (5 to 22) would be falsely positive. At a lower prevalence of 5%, a likely value in national surveys, 4 (3 to 7) would be missed per 1000 tested, and 12 (6 to 27) would be falsely positive.

Analyses showed small differences in sensitivity between assay type, but methodological concerns and sparse data prevent comparisons between test brands.

Authors' conclusions

The sensitivity of antibody tests is too low in the first week since symptom onset to have a primary role for the diagnosis of COVID‐19, but they may still have a role complementing other testing in individuals presenting later, when RT‐PCR tests are negative, or are not done. Antibody tests are likely to have a useful role for detecting previous SARS‐CoV‐2 infection if used 15 or more days after the onset of symptoms. However, the duration of antibody rises is currently unknown, and we found very little data beyond 35 days post‐symptom onset. We are therefore uncertain about the utility of these tests for seroprevalence surveys for public health management purposes. Concerns about high risk of bias and applicability make it likely that the accuracy of tests when used in clinical care will be lower than reported in the included studies. Sensitivity has mainly been evaluated in hospitalised patients, so it is unclear whether the tests are able to detect lower antibody levels likely seen with milder and asymptomatic COVID‐19 disease.

The design, execution and reporting of studies of the accuracy of COVID‐19 tests requires considerable improvement. Studies must report data on sensitivity disaggregated by time since onset of symptoms. COVID‐19‐positive cases who are RT‐PCR‐negative should be included as well as those confirmed RT‐PCR, in accordance with the World Health Organization (WHO) and China National Health Commission of the People's Republic of China (CDC) case definitions. We were only able to obtain data from a small proportion of available tests, and action is needed to ensure that all results of test evaluations are available in the public domain to prevent selective reporting. This is a fast‐moving field and we plan ongoing updates of this living systematic review.