Scolaris Content Display Scolaris Content Display

Pruebas de anticuerpos para identificar infecciones pasadas o presentes por SARS‐CoV‐2

This is not the most recent version

Collapse all Expand all

Antecedentes

El virus SARS‐CoV‐2 (síndrome respiratorio agudo grave coronavirus 2) y la pandemia de covid‐19 resultante presentan importantes retos de diagnóstico. Hay varias estrategias de diagnóstico disponibles para identificar la infección actual, descartar la infección, identificar a las personas que necesitan mayor atención o para detectar infecciones anteriores y la respuesta inmunitaria. Las pruebas de serología para detectar la presencia de anticuerpos contra el SARS‐CoV‐2 tienen por objeto identificar la infección previa por SARS‐CoV‐2, y pueden ayudar a confirmar la presencia de la infección actual.

Objetivos

Evaluar la exactitud diagnóstica de las pruebas de anticuerpos para determinar si una persona que se presenta en la comunidad o en la atención primaria o secundaria tiene o ha tenido la infección por el SARS‐CoV‐2, y la exactitud de las pruebas de anticuerpos para su uso en los estudios de seroprevalencia.

Métodos de búsqueda

Se realizaron búsquedas electrónicas en el registro de estudios Cochrane covid‐19 y en la covid‐19 Living Evidence Database de la Universidad de Berna, que se actualiza diariamente con artículos publicados de PubMed y Embase y con "preprints" de medRxiv y bioRxiv. Además, se revisaron los repositorios de las publicaciones de covid‐19. No se aplicaron restricciones de idioma. Se realizaron búsquedas para esta iteración de la revisión hasta el 27 de abril de 2020.

Criterios de selección

Se incluyeron estudios de exactitud de las pruebas de cualquier diseño que evaluaran las pruebas de anticuerpos (incluyendo la prueba enzimática inmunoabsorbente, inmunoensayos de quimioluminiscencia y pruebas de flujo lateral) en personas sospechosas de infección por SARS‐CoV‐2 actual o anterior, o donde se usaron las pruebas para detectar la infección. También se incluyeron estudios de personas de las que se sabía que tenían o no tenían la infección por SARS‐CoV‐2. Se incluyeron todos los estándares de referencia para definir la presencia o ausencia del SARS‐CoV‐2 (incluidas las pruebas de reacción en cadena de la polimerasa con transcriptasa inversa (RT‐PCR) y los criterios de diagnóstico clínico).

Obtención y análisis de los datos

Se evaluó el posible sesgo y la aplicabilidad de los estudios utilizando la herramienta QUADAS‐2. Se extrajeron datos de la tabla de contingencia 2x2 y se presentan la sensibilidad y especificidad de cada anticuerpo (o combinación de anticuerpos) utilizando diagramas de bosque (forest plots) apareados. Se agruparon los datos utilizando una regresión logística de efectos aleatorios cuando fue apropiado, estratificando por tiempo desde el inicio de los síntomas. Se tabularon los datos disponibles por el fabricante de la prueba. Se ha presentado la incertidumbre en estimaciones de sensibilidad y especificidad utilizando intervalos de confianza (IC) del 95%.

Resultados principales

Se incluyeron 57 publicaciones en las que se informaba un total de 54 cohortes de estudios con 15 976 muestras, de las cuales 8526 correspondían a casos de infección por el SARS‐CoV‐2. Se realizaron estudios en Asia (n = 38), Europa (n = 15), y los Estados Unidos y China (n = 1). Se identificaron datos de 25 pruebas comerciales y numerosas pruebas internas, una pequeña fracción de las 279 pruebas de anticuerpos que figuran en la lista de la Foundation for Innovative Diagnostics. Más de la mitad (n = 28) de los estudios incluidos sólo estaban disponibles como "preprints".

Existía preocupación acerca del riesgo de sesgo y la aplicabilidad. Las cuestiones frecuentes fueron el uso de diseños multigrupo (n = 29), la inclusión sólo de los casos de covid‐19 (n = 19), la falta de cegamiento de la prueba en evaluación (n = 49) y del estándar de referencia (n = 29), la verificación diferencial (n = 22) y la falta de claridad sobre el número de participantes, las características y las exclusiones de los estudios (n = 47). La mayoría de los estudios (n = 44) solo incluyeron a personas hospitalizadas debido a una infección por covid‐19 sospechada o confirmada. No hubo estudios exclusivamente en participantes asintomáticos. Dos tercios de los estudios (n = 33) definieron los casos de covid‐19 basándose solo en los resultados de la RT‐PCR, ignorando la posibilidad de resultados RT‐PCR falsos negativos. Se observó evidencia de publicación selectiva de los hallazgos de los estudios mediante la omisión de la identidad de las pruebas (n = 5).

Se observó una heterogeneidad significativa en las sensibilidades de los anticuerpos IgA, IgM e IgG, o combinaciones de los mismos, para resultados agregados a través de diferentes períodos de tiempo posteriores al inicio de los síntomas (rango de 0% a 100% para todos los anticuerpos objetivo). Por lo tanto, los principales resultados de la revisión se basan en los 38 estudios que estratificaron los resultados por el tiempo transcurrido desde el inicio de los síntomas. El número de personas que aportan datos dentro de cada estudio cada semana es pequeño y no suele basarse en el seguimiento de los mismos grupos de pacientes a lo largo del tiempo.

Los resultados agrupados de IgG, IgM, IgA, anticuerpos totales e IgG/IgM mostraron una baja sensibilidad durante la primera semana desde el comienzo de los síntomas (todos menos del 30,1%), aumentando en la segunda semana y alcanzando sus valores más altos en la tercera semana. La combinación de IgG/IgM tuvo una sensibilidad del 30,1% (IC del 95%: 21,4 a 40,7) durante 1 a 7 días, 72,2% (IC del 95%: 63,5 a 79,5) durante 8 a 14 días, 91,4% (IC del 95%: 87,0 a 94,4) durante 15 a 21 días. Las estimaciones de la exactitud más allá de tres semanas se basan en tamaños de muestra más pequeños y menos estudios. Durante 21 a 35 días, las sensibilidades agrupadas para IgG/IgM fueron del 96,0% (IC del 95%: 90,6 a 98,3). No hay suficientes estudios para estimar la sensibilidad de las pruebas más allá de los 35 días posteriores al inicio de los síntomas. Las especificidades resumidas (proporcionadas en 35 estudios) superaron el 98% para todos los anticuerpos objetivo con intervalos de confianza de no más de 2 puntos porcentuales de amplitud. Los resultados falsos positivos eran más frecuentes cuando se sospechaba y se descartaba la presencia de covid‐19, pero los números eran pequeños y la diferencia estaba dentro del rango esperado por el azar.

Suponiendo una prevalencia del 50%, valor que se considera posible en los trabajadores sanitarios que han sufrido síntomas respiratorios, se anticiparía que 43 (28 a 65) se omitirían y 7 (3 a 14) serían falsamente positivos de cada 1000 personas sometidas a pruebas de IgG/IgM en los días 15 a 21 posteriores al inicio de los síntomas. Con una prevalencia del 20%, un valor probable en los estudios en contextos de alto riesgo, se pasarían por alto 17 (11 a 26) por cada 1000 personas sometidas a prueba y 10 (5 a 22) darían un falso positivo. Con una prevalencia inferior al 5%, valor probable en los estudios nacionales, se pasarían por alto 4 (3 a 7) por cada 1000 personas sometidas a prueba, y 12 (6 a 27) serían falsamente positivas.

Los análisis mostraron pequeñas diferencias de sensibilidad entre los tipos de prueba, pero las dudas metodológicas y la escasez de datos impiden las comparaciones entre las marcas de las pruebas.

Conclusiones de los autores

La sensibilidad de las pruebas de anticuerpos es demasiado baja en la primera semana desde la aparición de los síntomas como para desempeñar un papel principal en el diagnóstico de la covid‐19, pero aún pueden tener un papel complementario a otras pruebas en las personas que se presenten más tarde, cuando las pruebas de RT‐PCR son negativas, o no se realizan. Es probable que las pruebas de anticuerpos tengan una función útil para detectar una infección previa de SARS‐CoV‐2 si se utilizan 15 o más días después de la aparición de los síntomas. Sin embargo, actualmente se desconoce la duración de las elevaciones de los anticuerpos y se han encontrado muy pocos datos más allá de los 35 días posteriores a la aparición de los síntomas. Por lo tanto, no existe certeza acerca de la utilidad de estas pruebas para los estudios de seroprevalencia con fines de gestión de la salud pública. Las preocupaciones sobre el alto riesgo de sesgo y la aplicabilidad hacen que sea probable que la exactitud de las pruebas cuando se usen en la atención clínica sea menor que la informada en los estudios incluidos. La sensibilidad se ha evaluado principalmente en pacientes hospitalizados, por lo que no está claro si las pruebas son capaces de detectar niveles de anticuerpos más bajos que probablemente se observan con la enfermedad covid‐19 más leve y asintomática.

El diseño, la ejecución y el informe de los estudios de exactitud de las pruebas de covid‐19 requieren una mejora considerable. Los estudios deben informar datos sobre la sensibilidad, desglosados por tiempo desde la aparición de los síntomas. Se deben incluir los casos positivos de covid‐19 que son RT‐PCR‐negativos, así como los RT‐PCR confirmados, de acuerdo con las definiciones de casos de la Organización Mundial de la Salud (OMS) y la China National Health Commission of the People's Republic of China (CDC). Sólo se pudieron obtener datos de una pequeña proporción de las pruebas disponibles, y es necesario adoptar medidas para garantizar que todos los resultados de las evaluaciones de las pruebas estén disponibles en el dominio público para evitar la presentación de informes selectivos. Este es un campo que evoluciona muy rápidamente y se planean actualizaciones continuas de esta revisión sistemática activa.

¿Qué exactitud diagnóstica tienen las pruebas de anticuerpos para la detección de la infección por el virus de la covid‐19?

Antecedentes

La covid‐19 es una enfermedad infecciosa causada por el virus SARS‐CoV‐2 que se propaga fácilmente entre personas de manera similar al resfriado común o a la gripe. La mayoría de personas con covid‐19 presenta enfermedad respiratoria leve a moderada y es posible que algunas no presenten síntomas (infección asintomática). Otras experimentan síntomas graves y precisan de un tratamiento especializado y cuidados intensivos.

El sistema inmunitario de las personas con covid‐19 responde a la infección desarrollando en la sangre proteínas que pueden atacar al virus (anticuerpos). Las pruebas para detectar los anticuerpos en la sangre de las personas podrían mostrar si presentan covid‐19 en ese momento o si la han padecido con anterioridad.

¿Por qué es importante la exactitud de las pruebas?

Las pruebas exactas permiten identificar a las personas que podrían necesitar tratamiento o que deben aislarse para prevenir la propagación de la infección. Si no se detectan las personas con covid‐19 cuando está presente (un resultado falso negativo) se puede retrasar el tratamiento y se corre el riesgo de que la infección se siga propagando a otras personas. La identificación incorrecta de la covid‐19 cuando no está presente (un resultado falso positivo) puede dar lugar a pruebas, tratamiento y aislamiento innecesarios de la persona y de los contactos cercanos. La correcta identificación de las personas que han tenido covid‐19 anteriormente es importante para medir la propagación de la enfermedad, evaluar el éxito de las intervenciones de salud pública (como el aislamiento) y, potencialmente, para identificar a las personas con inmunidad (si se demuestra en el futuro que los anticuerpos indican inmunidad).

Para identificar resultados falsos negativos y falsos positivos, se comparan los resultados de las pruebas de anticuerpos en personas de las que se sabe que tienen covid‐19 y de las que se sabe que no tienen covid‐19. Los participantes en el estudio se clasifican en función de si se sabe o no que tienen covid‐19, con base en los criterios conocidos como "estándar de referencia". Muchos estudios utilizan muestras tomadas de la nariz y la garganta para identificar a las personas con covid‐19. Las muestras se someten a una prueba llamada reacción en cadena de la polimerasa con transcriptasa inversa (RT‐PCR). Este proceso de prueba, a veces, puede pasar por alto la infección (resultado falso negativo), pero las pruebas adicionales pueden identificar la infección por covid‐19 en personas con un resultado negativo de RT‐PCR. Esto incluye la medición de síntomas clínicos como la tos o la temperatura alta, o pruebas de diagnóstico por imagen como las radiografías de tórax. Las personas de las que se sabe que no tienen covid‐19 a veces se identifican a partir de muestras de sangre almacenadas, tomadas antes de que existiera la enfermedad, o de pacientes con síntomas respiratorios que se ha descubierto que son causados por otras enfermedades.

¿Qué estudió la revisión?

Los estudios observaron tres tipos de anticuerpos, IgA, IgG e IgM. La mayoría de las pruebas miden los niveles tanto de IgG como IgM, pero algunas miden los niveles de un solo anticuerpo o combinaciones de los tres.

Los niveles de anticuerpos aumentan y disminuyen en diferentes momentos después de la infección. La IgG es la última en elevarse, pero es la que se mantiene durante más tiempo. Los niveles de anticuerpos suelen ser más altos unas semanas después de la infección.

Algunas pruebas de anticuerpos necesitan un equipo de laboratorio especializado. Otras utilizan dispositivos desechables, similares a las pruebas de embarazo. Estas pruebas pueden utilizarse en laboratorios o dondequiera que se encuentre el paciente (punto de atención), en el hospital o en el domicilio.

Se quería averiguar si las pruebas de anticuerpos:

‐ son lo suficientemente exactas para diagnosticar la infección en personas con o sin síntomas de covid‐19, y

‐ se pueden usar para averiguar si alguien ya ha tenido covid‐19.

¿Qué se hizo?

Se buscaron estudios que midieran la exactitud de las pruebas de anticuerpos en comparación con los criterios del estándar de referencia para detectar la infección actual o pasada por covid‐19. Los estudios podían evaluar cualquier prueba de anticuerpos comparada con cualquier estándar de referencia. Las personas podían hacerse la prueba en el hospital o en la comunidad. Los estudios podían hacer la prueba a personas de las que se sabe que tienen o no tienen covid‐19, o de las que se sospecha que la tienen.

Características de los estudios

Se encontraron 54 estudios relevantes. Se realizaron estudios en Asia (38), Europa (15), y en los Estados Unidos y China (1).

Cuarenta y seis estudios incluyeron a personas que estaban en el hospital con sospecha de infección o confirmada de covid‐19 solamente. Veintinueve estudios compararon los resultados de las pruebas en personas con covid‐19 con los resultados de las pruebas en personas sanas o con otras enfermedades.

No todos los estudios proporcionaron detalles sobre la edad y el sexo de los participantes. A menudo, no se podía saber si los estudios evaluaban la infección actual o pasada, ya que pocos informaban de si los participantes se estaban recuperando. No se encontró ningún estudio que sólo hiciera pruebas a personas asintomáticas.

Resultados principales

Los hallazgos provienen principalmente de 38 estudios que proporcionaron resultados basados en el tiempo transcurrido desde que las personas notaron los síntomas por primera vez.

Las pruebas de anticuerpos una semana después de los primeros síntomas sólo detectaron el 30% de las personas que tenían covid‐19. La exactitud aumentó en la semana 2 con un 70% detectado, y fue mayor en la semana 3 (más del 90% detectado). Había poca evidencia disponible después de la tercera semana. Las pruebas dieron resultados falsos positivos en el 2% de los que no tenían covid‐19.

Los resultados de las pruebas de IgG/IgM tres semanas después del comienzo de los síntomas sugerían que si 1000 personas se hacían pruebas de anticuerpos, y 50 (5%) de ellas realmente tenían covid‐19 (como se podría esperar en un estudio nacional de seroprevalencia):

‐ 58 personas darían positivo en la prueba de covid‐19. De estas, 12 personas (21%) no tendrían covid‐19 (resultado falso positivo).

‐ 942 personas darían negativo en la prueba de covid‐19. De estas, cuatro personas (0,4%) tendrían realmente covid‐19 (resultado falso negativo).

Si se hiciera la prueba a 1000 trabajadores sanitarios (en un contexto de alto riesgo) que habían tenido síntomas, y 500 (50%) de ellos realmente tenían covid‐19:

‐ 464 personas darían positivo en la prueba de covid‐19. De estas, siete personas (2%) no tendrían covid‐19 (resultado falso positivo).

‐ 537 personas darían negativo en la prueba de covid‐19. De estas, 43 personas (8%) tendrían realmente covid‐19 (resultado falso negativo).

No se encontraron diferencias convincentes en la exactitud de los diferentes tipos de pruebas de anticuerpos.

¿En qué medida son fiables los resultados de los estudios de esta revisión?

La confianza en la evidencia es limitada por varias razones. En general, los estudios fueron pequeños, no utilizaron los métodos más fiables y no informaron sus resultados completamente. A menudo, no incluyeron a pacientes con covid‐19 que pudieron haber tenido un resultado falso negativo en la PCR, y en el caso de las personas sin covid‐19, tomaron los datos de los registros de las pruebas realizadas antes de que apareciera la enfermedad. Esto puede haber afectado a la exactitud de la prueba, pero es imposible identificar en qué medida.

¿Para quiénes son relevantes los resultados de esta revisión?

La mayoría de los participantes estaban en el hospital con covid‐19, por lo que era probable que tuvieran una enfermedad más grave que las personas con síntomas leves que no fueron hospitalizadas. Esto significa que no se sabe el nivel de exactitud de las pruebas de anticuerpos para las personas con enfermedades más leves o sin síntomas.

Más de la mitad de los estudios evaluaron pruebas que ellos mismos habían desarrollado, la mayoría de las cuales no están comercializadas. Muchos estudios se publicaron rápidamente en línea como "preprint" (versión de un manuscrito antes de la revisión por pares). Los "preprints" no se someten a las rigurosas verificaciones normales de los estudios publicados, por lo que no se está seguro de su fiabilidad.

Como la mayoría de los estudios se realizaron en Asia, no se sabe si los resultados de las pruebas serían similares en otros lugares del mundo.

¿Cuáles son las implicaciones de esta revisión?

La revisión muestra que las pruebas de anticuerpos podrían desempeñar un papel útil para detectar si alguien ha tenido covid‐19, pero el momento en que se utilizan las pruebas es importante. Las pruebas de anticuerpos pueden ayudar a confirmar la infección por covid‐19 en personas que hayan presentado síntomas durante más de dos semanas y no se hayan hecho una prueba RT‐PCR, o hayan resultado negativos en este mismo test. Las pruebas son mejores para detectar covid‐19 dos o más semanas después del comienzo de los síntomas, pero no se sabe en qué medida funcionan bien más de cinco semanas después. No se sabe cómo de bien funcionan las pruebas en el caso de las personas que tienen una enfermedad más leve o ningún síntoma, porque los estudios de la revisión se hicieron principalmente en personas que estaban en el hospital. Con el tiempo, se sabrá si el haber tenido covid‐19 anteriormente proporciona a los individuos inmunidad para futuras infecciones.

Es necesario realizar más investigaciones sobre el uso de pruebas de anticuerpos en las personas que se recuperan de la infección por covid‐19, y en las personas que han experimentado síntomas leves o que nunca han manifestado síntomas.

¿Cuál es el grado de actualización de esta revisión?

Esta revisión incluye evidencia publicada hasta el 27 de abril de 2020. Debido a que se están publicando muchas investigaciones nuevas en este campo, esta revisión se actualizará frecuentemente.

Authors' conclusions

Implications for practice

Diagnosis of acute suspected COVID‐19 in symptomatic patients

Based on this analysis, in patients presenting with symptoms of acute suspected COVID‐19, antibody tests have no role on their own as the primary test to use in the diagnosis of COVID‐19 when patients present during the first week since onset of symptoms, as their sensitivity is too low.

A small number of studies showed that the sensitivity of antibody tests is no different in those who were reverse transcription polymerase chain reaction (RT‐PCR)‐negative rather than RT‐PCR‐positive. Thus in hospitalised patients where molecular tests have failed to detect virus, antibody tests have an increasing likelihood of detecting immune response to the infection as time since onset of symptoms progresses.

There may therefore be a role in using antibody tests in COVID‐19 RT‐PCR‐negative but strongly suspected patients where patients are more than two weeks since the onset of symptoms. This is in line with the most recent version of the China CDC (National Health Commission of the People's Republic of China) COVID‐19 case definition (Appendix 2).

Assessment of previous SARS‐CoV‐2 infection and immune response

The data analysed in the review suggest that antibody tests are likely to have a useful role for detecting previous SARS‐CoV‐2 infection if used at 15 days or more after the onset of symptoms. This conclusion needs to be cautioned by the poor study quality, the small sample sizes and restricted number of tests that have undergone evaluation. In addition, we have scant data to inform the accuracy of the test in non‐hospitalised patients with milder disease, and too little data to comment on accuracy beyond 35 days.

Using, for illustration the overall IgG/IgM data at days 15 to 21 (sensitivity 91.4%, 95% CI 87.0 to 94.4 and specificity 98.7%, 93% CI 97.2 to 99.4), we have computed predictive values, and the numbers of true positives, false positives, false negatives and true negatives in a sample of 1000, at a prevalence of 50% (a value seen in healthcare worker populations who have suffered respiratory symptoms in the past months). In this scenario, the positive predictive value is estimated as 99% (95% CI 97 to 99), the negative predictive value as 92% (95% CI 88 to 95), and of 1000 people undergoing testing we would anticipate 7 (95% CI 3 to 14) false positives and 43 (95% CI 28 to 65) false negatives.

Please note that it is not certain whether a detectable immune response indicates that a patient is immune nor no longer infectious.

Seroprevalence surveys for public health management purposes

The duration of antibody rises is not yet known, and this review contains very little data beyond 35 days post‐onset of symptoms. In the 'Summary of findings' table we present scenarios for the likely numbers of missed cases (false negatives) and false positive cases for prevalences of 2%, 5%, (likely values in national surveys), 10% and 20% (likely values in high‐risk settings such as healthcare workers), presuming that the performance of an IgG/IgM test would continue at the same level as for 14‐21 days. Again this conclusion needs to be cautioned by the poor study quality, the applicability of the study settings, the small sample sizes and restricted number of tests that have undergone evaluation. At a prevalence of 20%, a possible value in surveys in high‐risk settings, 17 (95% CI 11 to 26) would be missed per 1000 people tested and 10 (95% CI 5 to 22) would be falsely positive. At a lower prevalence of 5%, a likely value in national surveys, 4 (95% CI 3 to 7) would be missed per 1000 tested, and 12 (95% CI 6 to 27) would be falsely positive.

Implications for research

Many more high‐quality evaluation studies of COVID‐19 antibody tests are needed in patients more than 21 days post‐symptom onset, and in people in the community, particularly those who experience milder symptoms, or who are asymptomatic (but known to be infected).

Future studies must report data on sensitivity disaggregated by time since onset of symptoms. In future updates of this review we will not include studies for analysis of sensitivity where this has not been done. We would suggest that studies standardise how they define time since symptom onset (not, for example, using time since positive RT‐PCR results since this has no biological basis) and present results using standard time groupings (we suggest initially by week up until 35 days and larger time intervals beyond). Studies that sample from the same patients at several time points over time are needed to fully understand how time since symptom onset directly affects performance – our current estimates are based on collation of multiple cross‐sectional studies, which has limitations.

Primary studies need to be undertaken for the many tests that are on the market but as yet have no independent evaluations. Future studies should evaluate test performance in consecutive individuals who are recruited in clinical care with suspected COVID‐19, to estimate both sensitivity and specificity, as this will estimate the likely performance of the tests in practice.

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 CDC case definitions.

Studies should ensure that the test is used as it is intended to be used in clinical practice (i.e. being undertaken at point‐of‐care rather than in laboratories (where appropriate) on the right specimens, by the intended healthcare worker). However, when validating people with suspected COVID‐19 who do not have a positive identification of COVID‐19 by RT‐PCR, these studies need to take care to confirm or rule out COVID‐19 by obtaining standardised evidence from other sources (e.g. repeat RT‐PCR, CT scans, follow‐up). Future studies need to recruit larger sample sizes and consider recruiting from multiple centres. We did not find any multicentre studies for this review.

We would also encourage investigators to utilise blinding in their study designs, such that index tests are undertaken without knowledge of the reference standard diagnosis, and likewise, reference standards are determined without knowledge of the index test findings.

We need good data upon which to compare tests. The strongest comparisons are made by testing the same participants multiple times with different tests. Whilst it is possible for this to be undertaken in prospective studies, it is easier to undertake in laboratory‐based studies utilising serum banks, which will compromise on the applicability of the absolute estimates of test accuracy, but provide some information about comparability.

From these studies we can only draw limited conclusions about cross‐reactivity of COVID‐19 tests with other coronaviruses as these data are summarised in analytical accuracy studies. It would be of value for these results to be reviewed as well as clinical accuracy studies.

Study reporting requires substantial improvement. The STARD checklist outlines standard requirements for the reporting of a test accuracy study, which study investigators should take note of when planning their study to ensure the relevant information is collected and reported. No study was found that reported data using a STARD participant flow‐diagram (Bossuyt 2015).

Due to the speed of new publications in this field, frequent updates of this review are required. Future updates will not include data on tests that are not (or not likely to become) commercially available (thus we will exclude all in‐house assays).

Summary of findings

Open in table viewer
Summary of findings 1. What is the diagnostic accuracy of antibody tests, for the diagnosis of current or prior SARS‐CoV‐2 infection?

Question

What is the diagnostic accuracy of antibody tests, for the diagnosis of current or prior SARS‐CoV‐2 infection?

Population

Adults or children suspected of

  • current SARS‐CoV‐2 infection

  • prior SARS‐CoV‐2 infection

or populations undergoing screening for SARS‐CoV‐2 infection, including

  • asymptomatic contacts of confirmed COVID‐19 cases

  • community screening

Index test

Any test for detecting antibodies to SARS‐CoV‐2, including:

  • laboratory‐based methods

    • ELISA

    • CLIA

    • other laboratory‐based methods

  • rapid tests; lateral flow assays, including

    • tests that can be used at point‐of‐care, such as CGIA

    • rapid diagnostic tests, such as FIA

Target condition

Detection of

  • current SARS‐CoV‐2 infection

  • prior SARS‐CoV‐2 infection

Reference standard

RT‐PCR alone, clinical diagnosis of COVID‐19 based on established guidelines or combinations of clinical features and for non‐COVID‐19 cases, the use of pre‐pandemic sources of samples for testing

Action

The current evidence‐base for antibody tests is inadequate to be clear about their utility (mainly because of small numbers of small studies for each test, few data available outside of acute hospital settings, and many issues in likely bias and applicability of the studies). The sensitivity of antibody tests is too low early in disease for use as a primary test of diagnosis, but they may have value for late diagnosis, for identifying previous infection, and for sero‐prevalence studies.

Limitations in the evidence

Risk of bias

Participant selection: high risk of bias in 48 studies (89%)

Application of index tests: high risk of bias in 14 studies (26%)

Reference standard: high risk of bias in 17 studies (31%)

Flow and timing: high risk of bias in 29 studies (54%)

Concerns about applicability of the evidence

Participants: high concerns in 44 studies (81%)

Index test: high concerns in 17 studies (31%)

Reference standard: high concerns in 33 studies (61%)

Findings

  • We included 54 studies evaluating 15,976 samples. 8256 samples were from COVID‐19 cases.

  • Data were not available for most antibody tests that have regulatory approval.

  • Most studies reported on detection of IgG, IgM, or IgG/IgM antibodies.

  • Test sensitivity was strongly related to time since onset of symptoms, with low sensitivity between 1 and 14 days, and sensitivity for IgG/IgM tests exceeding 90% between 15 and 35 days. Little evidence was available beyond 35 days.

  • Specificity was high (> 98%) for all types of antibody. There was some variation in sensitivity between test methods, with laboratory‐based methods appearing to outperform (point‐of‐care) tests using disposable devices.

  • Small sample sizes, low numbers of studies and concerns and bias and applicability hinder trustworthy comparisons being made between test brands.

Quantity of evidence

Number of studies

Total participants or samples

Total cases

54

15,976

8526

Sensitivity (95% CI)

Studies (TP/COVID cases)

Specificity (95%CI)

Studies (FP/non‐COVID cases)

Days 8‐14

Days 15‐21

Days 22‐35

All time points

IgG

66.5% (57.9 to 74.2)

88.2% (83.5 to 91.8)

80.3% (72.4 to 86.4)

99.1% (98.3% to 99.6%)

22 (766/1200)

22 (974/1110)

12 (417/502)

44 (159/6136)

IgM

58.4% (45.5 to 70.3)

75.4% (64.3 to 83.8)

68.1% (55.0 to 78.9)

98.7% (97.4% to 99.3%)

21 (724/1171)

21 (800/1074)

11 (378/507)

41 (183/6103)

IgG/IgM*

72.2% (63.5 to 79.5)

91.4% (87.0 to 94.4)

96.0% (90.6 to 98.3)

98.7% (97.2% to 99.4%)

9 (441/608)

9 (636/692)

5 (146/152)

23 (78/5761)

Numbers applied to a hypothetical cohort of 1000 patients, using summary data for IgG/IgM at days 15 to 21 as an exemplar (sensitivity 91.4% (87.0 to 94.4) and specificity 98.7% (97.2 to 99.4))

Prevalence of COVID‐19

TP (95% CI)

FP (95% CI)

FN (95% CI)

TN (95% CI)

2%

18 (17 to 20)

13 (6 to 27)

2 (1 to 3)

967 (953 to 974)

5%

46 (44 to 47)

12 (6 to 27)

4 (3 to 7)

938 (923 to 944)

10%

91 (87 to 94)

12 (5 to 25)

9 (6 to 13)

888 (875 to 895)

20%

183 (174 to 189)

10 (5 to 22)

17 (11 to 26)

790 (778 to 795)

50%

457 (435 to 472)

7 (3 to 14)

43 (28 to 65)

494 (486 to 497)

CGIA: colloidal gold immunoassays; CI: confidence interval; CLIA: chemiluminescence immunoassays; ELISA: enzyme‐linked immunosorbent assays; FIA: fluorescence‐labelled immunochromatographic assays; FN: false negative; FP: false positive; RT‐PCR: reverse transcription polymerase chain reaction; TN: true negative; TP: true positive; * Positive if either IgG or IgM positive.

Background

The severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2) virus and resulting COVID‐19 pandemic present important diagnostic evaluation challenges. These range from understanding the value of signs and symptoms in predicting possible infection, assessing whether existing biochemical and imaging tests can identify infection and people needing critical care, and evaluating whether new diagnostic tests can allow accurate rapid and point‐of‐care testing, either to identify current infection, rule out infection, identify people in need of care escalation, or to test for past infection and immunity.

We are creating and maintaining a suite of living systematic reviews to cover the roles of tests and characteristics in the diagnosis of COVID‐19. This review summarises evidence of the accuracy of COVID‐19 antibody tests; both laboratory‐based tests and point‐of‐care tests.

Target condition being diagnosed

COVID‐19 is the disease caused by infection with the SARS‐CoV‐2 virus. The key target conditions for this suite of reviews are current SARS‐CoV‐2 infection, current COVID‐19 disease, and past SARS‐CoV‐2 infection.

Antibody tests are being considered and evaluated for both:

  • identification of past SARS‐CoV‐2 infection, and

  • current infection.

For current infection the severity of the disease is of importance. SARS‐CoV‐2 infection can be asymptomatic (no symptoms); mild or moderate (symptoms such as fever, cough, aches, lethargy but without difficulty breathing at rest); severe (symptoms with breathlessness and increased respiratory rate indicative of pneumonia); or critical (requiring respiratory support due to severe acute respiratory syndrome (SARS) or acute respiratory distress syndrome (ARDS). People with COVID‐19 pneumonia (severe or critical disease) require different patient management, and it is important to be able to identify them. There is no consideration that antibody tests are able to distinguish severity of disease, thus, in this review, we consider their role for detecting SARS‐CoV‐2 infection of any severity (asymptomatic or symptomatic).

Index test(s)

Antibody tests

This review evaluates serology tests to measure antibodies to the SARS‐CoV‐2 virus. Antibodies are formed by the body's immune system in response to infections, and can be detected in whole blood, plasma or serum. Antibodies are specific to the virus, and therefore can be used to differentiate between different infections. There are three types of antibody created in response to infection: IgA, IgG and IgM; these rise and fall at different times after the onset of infection. IgG is used in most antibody tests as it persists for the longest time and may reflect longer‐term immunity, although it is the last to rise after infection. Many tests assess both IgG and IgM. IgM typically rises quickly with infection and declines soon after an infection is cleared. Alternatively tests may combine IgA with IgG, or measure all antibodies (IgA, IgG and IgM).

Antibody tests are available for laboratory use including enzyme‐linked immunosorbent assay (ELISA) methods, or more advanced chemiluminescence immunoassays (CLIA). There are also laboratory‐independent, point‐of‐care lateral flow assays, which use disposable devices, akin to a pregnancy test, that use a minimal amount of blood on a testing strip. Antibody detection is indicated by visible lines appearing on the test strip, or through fluorescence, which can be detected using a reader device. Many of these tests are known as colloidal gold‐based immunoassays, as they use COVID‐19 antigen conjugated to gold nanoparticles.

Following the emergence of COVID‐19 there has been prolific industry activity to develop accurate antibody tests. The Foundation for Innovative Diagnostics (FIND) and Johns Hopkins Centre for Health Security have maintained online lists of these and other molecular‐based tests for COVID‐19. At the time of writing (21 May 2020), FIND listed 279 antibody tests, 196 of which are produced by commercial companies and are commercially available. Reguatory approval in the European Union (EU; CE‐IVD) had been awarded to 185 on the list, whereas in China only seven had been approved, and eight by the FDA (US Food and Drug Administration). For a period of time the FDA allowed commercialisation of antibody tests in the USA without FDA approval, resulting in around 100 tests being placed on the market. Both the content of the list, and these figures will increase over time.

Clinical pathway

Broadly speaking, there are four considered uses of antibody tests.

  1. In diagnosis of acute suspected COVID‐19 in patients who presented with symptoms, particularly where molecular testing had failed to detect the virus.

  2. In assessment of immune response in patients with severe disease.

  3. For individuals to assess whether they have had a SARS‐CoV‐2 infection and have an immune response.

  4. In seroprevalence surveys for public health management purposes.

For 1, the standard approach to diagnosis of COVID‐19 is through a reverse transcription polymerase chain reaction (RT‐PCR) test, which detects the presence of virus in swab samples taken from nose, throat or fluid from the lungs. However, the test is known to give false negative results, and can only detect COVID‐19 in the acute phase of the illness. Both the World Health Organization (WHO) and the China CDC (National Health Commission of the People's Republic of China), have produced case definitions for COVID‐19 that include RT‐PCR‐negative cases that display other convincing clinical evidence (Appendix 1). The most recent case definition from the China CDC includes positive serology tests. Confirming an acute clinical diagnosis using a serology test requires detectable virus‐specific IgM and IgG in serum, or detectable virus‐specific IgG, or a 4‐fold or greater increase in titration to be observed during convalescence compared with the acute phase.

For 2, this is largely a question of monitoring patients, and we will not cover this in this review. Assessment of the accuracy of a test used for assessment of immune response would involve comparison with a reference standard test of antibody response, rather than evidence of infection.

Use 3 involves testing individuals during periods of convalescence (after symptoms have resolved) whereas 4 will involve testing people at a mixture of time points, including long follow‐up. A key difference between 3 and 4 is the likelihood of disease, which is expected to be much higher for 3 than 4.

An extended version of use case scenarios is available in Appendix 2.

Prior test(s)

Prior testing depends on the purpose of the test. For 1 we would anticipate that patients were symptomatic and had most likely undergone RT‐PCR testing and possible computed tomography (CT) imaging. Uses 3 and 4 will most likely include people who have not been tested, and may include people who are asymptomatic as well as symptomatic.

Alternative test(s)

This review is one of six planned reviews that cover the range of tests and characteristics being considered in the management of COVID‐19 (Deeks 2020; McInnes 2020). Full details of the alternative tests and evidence of their accuracy will be summarised in these reviews.

Laboratory‐based molecular tests

Testing for presence of the SARS‐CoV‐2 virus has been undertaken using quantitative RT‐PCR (qRT‐PCR). RT‐PCR tests for SARS‐CoV‐2 identify viral ribonucleic acid (RNA). Reagents for the assay were rapidly produced once the viral RNA sequence was published. Testing is undertaken in central laboratories and can be very labour‐intensive, with several points along the path of performing a single test where errors may occur, although some automation of parts of the process is possible. Although the actual qRT‐PCR test does not take long, the stages of extraction, sample processing and data management mean that test results are typically available in 24 to 48 hours, although faster processes are being implemented. Other nucleic acid amplification methods such as loop‐mediated isothermal amplification (LAMP), or CRISPR‐based nucleic acid detection methods are also being developed, with the potential to reduce the time to produce test results to minutes, but the time for the whole process may still be significant. RT‐PCR tests use upper and lower respiratory samples. Sputum is currently considered better than oropharynx swabs or nasopharynx swabs but is more difficult (and hazardous) to obtain and will only ever be available in a subset of patients.

Laboratory‐independent point‐of‐care and near‐patient molecular and antigen tests

Laboratory‐independent RT‐PCR devices can also be used for identification of infection near patients and even at the bedside. These are small platforms for testing which use matching test cartridges. Several companies have suitable existing technology systems and are producing the required new cartridges for diagnosis of SARS‐CoV‐2 infection. Test results are based on the same samples as those for qRT‐PCR, with results available within minutes or hours. Antigen tests are based on the direct detection of the virus, indicating active infection (i.e. replication of the virus) similar to the detection of RNA. Antigen tests are mainly in the form of lateral flow assays. They will capture the relevant viral antigen using dedicated antibodies, and visualisation is either manual or using a reader device.

Signs and symptoms

Signs and symptoms are used in the initial diagnosis of suspected COVID‐19, and in identifying people with COVID‐19 pneumonia. Key symptoms that have been associated with mild to moderate COVID‐19 include: troublesome dry cough (for example, coughing more than usual over a one‐hour period, or three or more coughing episodes in 24 hours), fever greater than 37.8°C, diarrhoea, headache, breathlessness on light exertion, muscle pain, fatigue, and loss of sense of smell and taste. Red flags indicating possible pneumonia include: breathlessness at rest, increased respiratory rate (above 20 breaths per minute), increased heart rate (above 100 beats per minute), chest tightness, loss of appetite, confusion, pain or pressure in the chest, blue lips or face, and temperature above 38°C. Hypoxia based on measuring pulse oximetry is often used, with various arbitrary thresholds (for example, 93%).

Routinely available biomarkers

Routinely available biomarkers for infection and inflammation may be considered in the investigation of people with possible COVID‐19. For example, many healthcare facilities have access to standard laboratory tests for infection, such as C‐reactive protein (CRP), procalcitonin, measures of anticoagulation, and white blood cell count with different lymphocyte subsets. Evaluation of these commonly available tests, particularly in low‐resource settings, may be helpful for the triage of people with potential COVID‐19.

Imaging tests

Chest X‐ray, ultrasound, and CT are widely used diagnostic imaging tests to identify COVID‐19 pneumonia. Availability and usage varies between settings.

Rationale

It is essential to understand the clinical accuracy of tests and diagnostic features to identify the best way they can be used in different settings to develop effective diagnostic and management pathways. The suite of Cochrane 'living systematic reviews' summarises evidence on the clinical accuracy of different tests and diagnostic features, grouped according to the research questions and settings that we are aware of. Estimates of accuracy from these reviews will help inform diagnosis, screening, isolation, and patient management decisions.

Particularly for antibody tests, new tests are being developed and evidence is emerging at an unprecedented rate during the COVID‐19 pandemic. Tests are being purchased in bulk for seroprevalence studies, and made available for personal purchase online. This review will be updated as often as is feasible to ensure that it provides current evidence about the accuracy of antibody tests.

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.

Secondary objectives

Where data are available, we will investigate the accuracy (either by stratified analysis or meta‐regression) according to:

  • current infection or past infection;

  • test method and brand;

  • days since onset of symptoms;

  • reference standard;

  • study design;

  • setting.

Methods

Criteria for considering studies for this review

Types of studies

We applied broad eligibility criteria in order to include all patient groups and all variations of a test (that is, if patient population was unclear, we included the study).

We included studies of all designs that produce estimates of test accuracy or provide data from which estimates can be computed, including the following.

  • Studies restricted to participants confirmed to have (or to have had) the target condition (to estimate sensitivity) or confirmed not to have (or have had) the target condition (to estimate specificity). These types of studies may be excluded in later review updates.

  • Single‐group studies, which recruit participants before disease status has been ascertained

  • Multi‐group studies, where people with and without the target condition are recruited separately (often referred to as two‐gate or diagnostic case‐control studies)

  • Studies based on either patients or samples

We excluded studies from which we could not extract data to compute either sensitivity or specificity.

We carefully considered the limitations of different study designs in the quality assessment and analyses.

We included studies reported in published articles and as preprints.

Participants

We included studies recruiting people presenting with suspicion of current or prior SARS‐CoV‐2 infection or those recruiting populations where tests were used to screen for disease (for example, contact tracing or community screening).

We also included studies that recruited people either known to have SARS‐CoV‐2 infection or known not to have SARS‐CoV‐2 infection (multi‐group studies).

We excluded small studies with fewer than 10 samples or participants. Although the size threshold of 10 is arbitrary, such small studies are likely to give unreliable estimates of sensitivity or specificity and may be biased.

Index tests

We included studies evaluating any test for detecting antibodies to SARS‐CoV‐2, including laboratory‐based methods and tests designed to be used at point‐of‐care. Test methods include the following.

Laboratory‐based:

  • enzyme‐linked immunosorbent assays (ELISA)

  • chemiluminescence immunoassays (CLIA)

  • other laboratory‐based methods (e.g. indirect immunofluorescence tests (IIFT), luciferase immunoprecipitation system (LIPS)

Rapid diagnostic tests:

  • lateral flow assays, including both colloidal gold or fluorescence‐labelled immunochromatographic assays (CGIA or FIA).

In this first version of the review we have included both commercially available tests, which have regulatory approval, with in‐house assays and assays in development. Future versions of the review are likely to be restricted to only commercially available assays.

We identified the regulatory status of index tests using two main resources:

  • WHO: COVID‐19 listing in International Medical Device Regulators Forum (IMDRF) jurisdictions (www.who.int/diagnostics_laboratory/EUL/en/), which includes listings of FDA, Health Canada, Japan, Australia (Therapeutic Goods Administration), Singapore (Health Sciences Authority), Brazil (Agência Nacional de Vigilância Sanitária), South Korea (Ministry of Food and Drug Safety), China (National Medical Products Administration), and Russia (Roszdravnadzor);

  • FIND: SARS‐COV‐2 Diagnostic Pipeline (www.finddx.org/covid-19/pipeline/), which overlaps with the WHO list, but in addition includes CE‐IVD and IVD India.

In addition, we checked key national websites, including US FDA (www.fda.gov/medical-devices/emergency-situations-medical-devices/emergency-use-authorizations#coronavirus2019) and China FDA (subsites.chinadaily.com.cn/nmpa/2020 03/27/c_465663.htm?bsh_bid=5496527208).

Target conditions

The target conditions were the identification of:

  • current SARS‐CoV‐2 infection (in symptomatic cases);

  • past SARS‐CoV‐2 infection (in convalescent (post‐symptomatic) or asymptomatic cases).

Reference standards

We anticipated that studies would use a range of reference standards to define both the presence and absence of SARS‐CoV‐2 infection but were unclear at the start of the review exactly what methods would be encountered. For the QUADAS‐2 (Quality Assessment tool for Diagnostic Accuracy Studies; Whiting 2011), assessment we categorised each method of defining COVID‐19 cases according to the risk of bias (the chances that it would misclassify COVID‐19 participants as non‐COVID‐19) and whether it defined COVID‐19 in an appropriate way that reflected cases encountered in practice. Likewise, we considered the risk of bias in definitions of non‐COVID‐19, and whether the definition reflected those who, in practice, would be tested.

Search methods for identification of studies

Electronic searches

We conducted a single literature search to cover our suite of Cochrane COVID‐19 diagnostic test accuracy (DTA) reviews (Deeks 2020; McInnes 2020).

We conducted electronic searches using two primary sources. Both of these searches aimed to identify all published articles and preprints related to COVID‐19, and were not restricted to those evaluating biomarkers or tests. Thus, there are no test terms, diagnosis terms, or methodological terms in the searches. Searches were limited to 2019 and 2020, and for this version of the review have been conducted to 27 April 2020.

Cochrane COVID‐19 Study Register searches

We used the Cochrane COVID‐19 Study Register (covid-19.cochrane.org/), for searches conducted to 28 March 2020. At that time, the register was populated by searches of PubMed, as well as trials registers at ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform (ICTRP).

Search strategies were designed for maximum sensitivity, to retrieve all human studies on COVID‐19 and with no language limits. See Appendix 3.

COVID‐19 Living Evidence Database from the University of Bern

From 28 March 2020, we used the COVID‐19 Living Evidence database from the Institute of Social and Preventive Medicine (ISPM) at the University of Bern (www.ispm.unibe.ch), as the primary source of records for the Cochrane COVID‐19 DTA reviews. This search includes PubMed, Embase, and preprints indexed in bioRxiv and medRxiv databases. The strategies as described on the ISPM website are described here (ispmbern.github.io/covid-19/). See Appendix 4.

The decision to focus primarily on the 'Bern' feed was due to the exceptionally large numbers of COVID‐19 studies available only as preprints. The Cochrane COVID‐19 Study Register has undergone a number of iterations since the end of March and we anticipate moving back to the Register as the primary source of records for subsequent review updates.

Searching other resources

We identified Embase records obtained through Martha Knuth for the Centers for Disease Control and Prevention (CDC), Stephen B Thacker CDC Library, COVID‐19 Research Articles Downloadable Database (www.cdc.gov/library/researchguides/2019novelcoronavirus/researcharticles.html), and de‐duplicated them against the Cochrane COVID‐19 Study Register up to 1 April 2020. See Appendix 5.

We also checked our search results against two additional repositories of COVID‐19 publications including:

Both of these repositories allow their contents to be filtered according to studies potentially relating to diagnosis, and both have agreed to provide us with updates of new diagnosis studies added. For this iteration of the review, we examined all diagnosis studies from either source up to 16 April 2020.

In addition we have used the list of potentially eligible index tests (documented in Criteria for considering studies for this review), to search company and product websites for studies about test accuracy and to contact companies to request further information or studies using their tests. We will include the result of this process in a future iteration of this review.

We have also contacted research groups undertaking test evaluations (for example, UK Public Health England‐funded studies, and FIND studies (www.finddx.org/). We appeal to researchers to supply details of additional published or unpublished studies at the following email address, which we will consider for inclusion in future updates ([email protected]).

We did not apply any language restrictions.

Data collection and analysis

Selection of studies

A team of experienced systematic reviewers from the University of Birmingham screened the titles and abstracts of all records retrieved from the literature searches. Two review authors independently screened studies in Covidence. A third, senior review author resolved any disagreements. We tagged all records selected as potentially eligible according to the Cochrane COVID‐19 DTA review(s) that they might be eligible for and we then exported them to separate Covidence reviews for each review title.

We obtained the full texts for all studies flagged as potentially eligible. Two review authors independently screened the full texts for one of the COVID‐19 molecular or antibody test reviews. We resolved any disagreements on study inclusion through discussion with a third review author.

Data extraction and management

One review author carried out data extraction, which was checked by a second review author. Items that we extracted are listed in Appendix 6. Both review authors independently performed data extraction of 2x2 contingency tables of the number of true positives, false positives, false negatives and true negatives. They resolved disagreements by discussion.

We encourage study authors to contact us regarding missing details on the included studies ([email protected]).

Where possible we extracted 2x2 tables according to time since onset of symptoms. We predefined groups of interest as 1‐7, 8‐14, 15‐21, 22‐35 and over 35 days since onset of symptoms. Where the data presented did not exactly match these categorisations we entered data in the time group that had the greatest overlap with our groupings. Where a study presented data for a group without stating an upper time limit (e.g. more than 21 days) we placed the data in the first category above the stated value (e.g. 22‐35 days).

Where possible, we separately extracted data related to each class of antibody (IgA, IgG and IgM), and combinations of classes (IgA/IgM, IgA/IgG, IgG/IgM, where a positive is defined as either or both classes of antibody being detected). We also extracted data on total antibodies where this was reported.

Assessment of methodological quality

Two review authors independently assessed risk of bias and applicability concerns using the QUADAS‐2 checklist tailored to this review (Appendix 7; Whiting 2011). The two review authors resolved any disagreements by discussion.

Ideally, studies should prospectively recruit a representative sample of participants presenting with signs and symptoms of COVID‐19, either in community or primary care settings or to a hospital setting, and they should clearly record the time of testing after the onset of symptoms. Studies should perform antibody tests in their intended use setting, using appropriate sample types as described in the 'Instructions for use' sheet (e.g. fingerprick blood for tests being evaluated for use as point‐of‐care tests), and tests should be performed by relevant personnel (e.g. healthcare workers), and should be interpreted blinded to the final diagnosis (COVID‐19 or not). Serology samples should be taken at time points that reflect the intended use (either whilst symptomatic for diagnosis of infection, or during a convalescent period (after resolution of symptoms) for diagnosis of previous infection). The reference standard diagnosis should be blinded to the result of the antibody test, and should not incorporate the result of the index test or any other serology test. If the reference standard includes clinical diagnosis of COVID‐19, then established criteria should be used. Studies including samples from participants known not to have COVID‐19 should use pre‐pandemic sources or contemporaneous samples with at least one RT‐PCR‐negative test result. Data should be reported for all study participants, including those where the result of the antibody test was inconclusive, or participants in whom the final diagnosis of COVID‐19 was uncertain. If studies obtained multiple samples for testing over time from the same study participants, then they should disaggregate results by time post‐symptom onset.

Statistical analysis and data synthesis

We grouped data by study and test. Thus studies that evaluated multiple tests in the same participants were included multiple times. We present estimates of sensitivity and specificity for each antibody (or combination of antibodies) using paired forest plots in tables, and also summarise them in tables as appropriate.

For analysis purposes, unlike in most DTA reviews we considered estimates of sensitivity and specificity separately, because many of the included studies presented only estimates of sensitivity. Estimates of specificity were typically exceptionally high, thus the correlation between sensitivity and specificity across studies was unlikely to be high (Macaskill 2010; Takwoingi 2017). We considered the heterogeneity in the study findings through visual inspection of forest plots when deciding to meta‐analyse study estimates, and have not computed summary estimates where they were likely to be regarded as misleading.

Where we pooled results, we fitted random‐effects logistic regression models using the meqrlogit command in Stata v15.1 (Stata). In a small number of instances, the random‐effects logistic regression analyses failed to converge (usually when there were very small numbers of studies), and we have computed estimates and confidence intervals by summing the counts of true positive, false positive, false negative and true negative across 2x2 tables. These analyses are clearly marked in the tables. We present all estimates with 95% confidence intervals.

Investigations of heterogeneity

We investigated sources of heterogeneity in two ways. First, for analysis of sensitivity for time since onset of symptoms, we extracted data by week and extended the random‐effects logistic regression model to include indicator variables for each week. There was a strong relationship between time since onset of symptoms and sensitivity, thus we elected to fit all subsequent models for investigation of heterogeneity in sensitivity stratifying by week. We excluded studies for which stratified data were not available at this stage. For analysis of sensitivity according to the RT‐PCR status of patients (RT‐PCR positive ‘confirmed’ and RT‐PCR negative ‘suspect’), we extracted 2x2 tables stratified by RT‐PCR result (as well as week) and extended the random‐effects logistic regression to include terms for week and RT‐PCR status.

We investigated heterogeneity related to study design, reference standard and test technology by including indicator variables in the random‐effects logistic regression model alongside the variables for week since onset of symptoms. We present estimates from these models by test or reference standard type for the sensitivity of the test in the third week since onset of symptoms (since this is the time point most commonly recommended for post‐infection testing to start to be undertaken).

We did not fit models to compare test brands due to the small number of studies available, but we do report estimates with confidence intervals for each brand.

Sensitivity analyses

We planned to undertake sensitivity analyses by excluding:

  • unpublished studies;

  • studies identified only from industry 'Instructions for use' documentation;

  • studies using sample banks or spiked samples;

  • studies with inadequate reference standards;

  • for previous infection, we also planned to assess increasing lengths of time since symptoms cleared.

In this version of the review we did not undertake any of these analyses because the majority of studies were preprints, we did not include any company documents, and no study used spiked samples. We investigated issues with reference standards and time as part of the investigations of heterogeneity.

Assessment of reporting bias

We made no formal assessment of reporting bias. However we were aware of the manner in which results in studies could be suppressed by test developers or manufacturers, and detail where we believe this may have happened.

Summary of findings

We summarised key findings in a 'Summary of findings' table indicating the strength of evidence for each test and findings, and highlighted important gaps in the evidence.

Updating

We are aware that a substantial number of studies have been published since the search date of 27 April 2020 and plan to update this review imminently. We have already completed searches for the update up until 25 May 2020, and report the number of studies that we anticipate will be added to this review in the first update.

Results

Results of the search

We screened 10,965 unique references (published or preprints) for inclusion in the complete suite of reviews to assist in the diagnosis of COVID‐19 (Deeks 2020; McInnes 2020). Of 1430 records selected for further assessment for inclusion in any of the six reviews, we assessed 267 full‐text reports for inclusion in this review. See Figure 1 for the PRISMA flow diagram of search and eligibility results (McInnes 2018; Moher 2009). We included 54 studies from 57 reports in this review, three studies are awaiting assessment including two foreign language papers and one study of neutralising antibodies (Characteristics of studies awaiting classification), 34 are ongoing studies (Characteristics of ongoing studies), and we excluded 172 publications. Exclusions were mainly due to ineligible study designs (n = 84) or index tests (n = 40), or because we could not extract or reconstruct 2x2 data (n = 21). The reasons for exclusion of all 172 publications are provided in Characteristics of excluded studies.


Study flow diagram

Study flow diagram

The 57 included study reports relate to 54 separate studies, six studies (Gao 2020a; Liu 2020d [A]; Pan 2020a; Okba 2020a; Wang 2020a [A]; Zhao 2020a), having two publications each, and three studies providing data for two separate cohorts of participants (Cassaniti 2020 (A); Cassaniti 2020 (B); Garcia 2020 (A); Garcia 2020 (B); Long 2020 (A); Long 2020 (B)). Of the 57 study reports, 28 studies are available only as preprints and four as preprints with subsequent journal publications. (Please note when naming studies, we use the letters (A), (B), (C) in standard brackets to indicate multiple studies from the same publication, and the letters [A], [B], [C] etc. in square brackets to indicate data on different tests evaluated in the same study).

Description of included studies

The 54 studies include a total of 15,976 samples, with 8526 samples from cases of COVID‐19. Summary study characteristics are presented in Table 1 with further details of study design and index test details in Appendix 8 and Appendix 9. The median sample size across the included studies is 129.5 (interquartile range (IQR) 57 to 347) and median number of COVID‐19 cases included is 62 (IQR 31 to 151). Thirty‐eight studies were conducted in Asia: China (n = 36); Hong Kong (n = 1); or Singapore (n = 1). Fifteen studies were conducted in Europe, and the remaining study included samples from more than one country (Bendavid 2020). Forty‐four studies included only hospital inpatient cases, one included hospital outpatients, two included participants attending emergency departments, two, community screening (including one study of close contacts). Five studies were conducted in mixed or unclear settings.

Open in table viewer
Table 1. Description of studies

Participants

Studies (percentage)

(n=54 studies)

Sample size

Median (IQR) 129.5 (57 to 347)

Min 10, max 3481

Number of COVID‐19 cases

Median (IQR) 62 (31 to 151)

Min 3, max 555

Setting

Hospital inpatient

44 (81%)

Hospital outpatient

1 (2%)

Hospital accident and emergency

2 (4%)

Community

2 (4%)

Mixed or unclear

5 (9%)

Patient group

Asymptomatic

0 (0%)

Asymptomatic and acute

1 (2%)

Acute

23 (43%)

Acute and convalescent

22 (41%)

Convalescent

2 (4%)

Mixed or unclear

6 (11%)

Study design

Recruitment structure

Single group, both COVID‐19 and non‐COVID‐19 cases

6 (11%)

Single group, only COVID‐19 cases

19 (35%)

Two or more groups with COVID‐19 and non‐COVID‐19 cases

29 (54%)

Reference standard for COVID‐19 cases

All RT‐PCR‐positive

32 (59%)

China CDC criteria including RT‐PCR‐negative patients

11 (20%)

WHO criteria including RT‐PCR‐negative patients

1 (2%)

Other criteria including RT‐PCR‐negative patients

3 (6%)

Other

2 (4%)

Mixed or unclear

5 (9%)

Reference standard for non‐COVID19

Pre‐pandemic healthy

4 (7%)

Pre‐pandemic other disease

3 (6%)

Pre‐pandemic healthy + other disease

4 (7%)

Current healthy (untested)

5 (9%)

Current other disease (untested)

1 (2%)

Current healthy + other disease (untested)

2 (4%)

Current healthy + other disease (RT‐PCR‐negative)

2 (4%)

COVID suspects, single RT‐PCR‐negative

8 (15%)

COVID suspects, two or more RT‐PCR–negative results

3 (6%)

Mixed/other

3 (6%)

Tests

Number of tests per study

1

40 (74%)

2

8 (15%)

3‐5

4 (8%)

6‐10

2 (2%)

Test technology (n = 89)

CGIA

23 (26%)

CLIA

20 (22%)

ELISA

28 (31%)

FIA

2 (2%)

IIFT

1 (1%)

LFA (no details)

10 (11%)

LIPS

4 (4%)

S‐flow

1 (1%)

Test brand (n = 89)

Withheld

13 (%)

Acro Biotech ‐ IgG/IgM

1 (1%)

Artron Laboratories IgM/IgG

1 (1%)

Autobio Diagnostics IgM/IgG

1 (1%)

Beijing Beier Bioengineering CGIA

1 (1%)

Beijing Beier Bioengineering CLIA

1 (1%)

Beijing Beier Bioengineering ELISA

1 (1%)

Beijing Diagreat

1 (1%)

Beijing Hotgen CGIA

1 (1%)

Beijing Hotgen ELISA

2 (3%)

Beijing Wantai CGIA

1 (1%)

Beijing Wantai ELISA

3 (3%)

Bioscience Co (Chongqing)

3 (3%)

CTK Biotech OnSite IgG/IgM

1 (1%)

Darui Biotech

1 (1%)

Dynamiker Biotechnology IgG/IgM

1 (1%)

EUROIMMUN

3 (3%)

EUROIMMUN Anti‐SARS‐Cov

1 (1%)

EUROIMMUN Beta

1 (1%)

Hangzhou Alltest ‐ IgG/IgM

3 (3%)

Innovita Biological ‐ Ab test (IgM/IgG)

2 (3%)

Jiangsu Medomics IgG‐IgM

1 (1%)

Shenzhen YHLO

7 (8%)

Snibe Diagnostic ‐ MAGLUMI

2 (3%)

Vivachek ‐ VivaDiag IgM/IgG

3 (3%)

Xiamen InnodDx Biotech

1 (1%)

Zhuhai Livzon CGIA

2 (3%)

Zhuhai Livzon ELISA

5 (6%)

In‐house, S‐based ELISA

1 (1%)

In‐house, S‐based LIPS

1 (1%)

In‐house, rN‐based ELISA

1 (1%)

In‐house, rS‐based ELISA

1 (1%)

In‐house CGIA

2 (2%)

In‐house CLIA

5 (6%)

In‐house ELISA

6 (7%)

In‐house FIA

1 (1%)

In‐house S‐flow

1 (1%)

In‐house ‐ N‐based ELISA

1 (1%)

In‐house ‐ N‐based LIPS

2 (2%)

In‐house ‐ S1‐based LIPS

1 (1%)

In‐house ‐ tri‐S‐based ELISA

1 (1%)

In‐house Anti‐SARS‐Cov ELISA

1 (1%)

Ab: antibody; CDC: Center for Disease Control and Prevention; CGIA: colloidal gold immunoassay; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IQR: interquartile range; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay; LIPS: luciferase immunoprecipitation system; max: maximum; min: minimum; N‐based: nucleocapsid protein; RT‐PCR: reverse transcription polymerase chain reaction; S‐based: spike protein; S‐flow: flow‐cytometry assay; WHO: World Health Organization

Participant characteristics

Twenty‐three studies included cases during the early phase of illness only (< 21 days post‐symptom onset), two only included cases 21 days or more post‐symptom onset, 23 included mixed groups and six did not report days post‐symptom onset. Few studies were clear whether participants were symptomatic or convalescent (i.e. symptoms had resolved) at the time of testing. It is therefore difficult to clearly separate out studies that detected current infection from studies that detected past infection. Thus the two target conditions we defined cannot clearly be distinguished. There were no studies exclusively in asymptomatic participants.

The mean or median age of included COVID‐19 cases ranges from 37 to 76 years (reported in 31 studies), and 26% to 87% of participants were male (reported in 31 studies). Full details are in the Characteristics of included studies table.

Study designs

We identified six studies that recruited suspected COVID‐19 cases before it was ascertained whether the patients did or did not have COVID‐19. These six studies identified people with suspected COVID‐19 based on symptoms or as close contacts of confirmed cases (symptomatic and asymptomatic). Sample sizes of these studies ranged from 50 to 814 with between 3 and 154 COVID‐19 cases. Four of these studies defined the presence or absence of COVID‐19 based on RT‐PCR alone, and two also included clinically confirmed RT‐PCR‐negative cases based on undefined clinical suspicion or CT findings. The absence of SARS‐CoV‐2 infection was confirmed by a single RT‐PCR‐negative result in five of the six and by two or more negative RT‐PCR results in one study.

The other forty‐eight studies retrospectively recruited patients when it was already known whether or not they had COVID‐19.

Twenty‐nine studies used two‐ or multi‐group study designs with separate selection of COVID‐19 cases and healthy participants or non‐COVID‐19 participants with another disease. Sample sizes ranged from 17 to 3481 with between 7 and 276 COVID‐19 cases. Nineteen of these studies defined COVID‐19 cases based on a positive RT‐PCR test, six included clinically defined RT‐PCR‐negative cases in addition to RT‐PCR‐positive cases and the remaining four studies used mixed or unclear criteria to define the presence of COVID‐19. Four of the 29 studies included participants with suspected COVID‐19 but who had subsequently been ruled out on the basis of one (2 studies) or more (2 studies) negative RT‐PCR tests. Ten included contemporaneous non‐COVID‐19 groups, including samples from healthy participants (5 studies), patients with other diseases (one study) or both (4 studies), only two of which used RT‐PCR testing to exclude the presence of SARS‐CoV‐2. Twelve studies included pre‐pandemic non‐COVID 19 groups, using samples from either healthy people (n = 5), participants with other diseases (n = 3), or both (n = 4). The remaining three studies included control samples from mixed sources including pre‐pandemic and contemporaneous samples, with or without RT‐PCR testing.

Nineteen studies included only a single group of only COVID‐19 cases, thus only allowing estimation of sensitivity. They determined COVID‐19 cases based on positive RT‐PCR alone (n = 9), clinically defined criteria including RT‐PCR‐negative cases (n = 8, 7 of which used Chinese government‐issued COVID‐19 guidelines to define cases), one using undefined clinical criteria, and one study that did not report how COVID‐19 cases were defined.

Index tests

Forty‐three studies evaluated only one test, five compared two tests, three compared 3 tests, one 5 tests, one 9 and one 10 tests. In total the 54 studies reported on a total of 89 test evaluations.

There were 52 evaluations of laboratory‐based methods (27 ELISA, 19 CLIA, 6 other methods), including 32 using commercially available laboratory‐based kits produced by 11 different commercial companies (16 ELISAs, 15 CLIAs and 1 IIFT), two where the manufacturer name was withheld, and 20 classified as using in‐house methods (11 ELISA, 4 CLIA and 5 other approaches).

There were 34 evaluations of lateral flow assays, 23 were described as or discovered to be CGIA, two were FIAs and nine were not described. Thirty‐one of the 34 evaluations used commercially available lateral flow assays and three were in‐house (including two CGIA and one FIA). Of the 34 evaluations, only three used whole blood (two using the Vivadiag test), and only two used the assays as point‐of‐care tests rather than in a laboratory setting.

Methodological quality of included studies

We report the overall methodological quality assessed using the QUADAS‐2 tool for all included studies (n = 54) in Figure 2 (Whiting 2011). See Appendix 10 for study‐level ratings by quality.


Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Overall, we judged risk of bias to be high in 48 (89%) studies concerning how participants were selected, 14 (26%) studies related to application of the index test, 17 (31%) through concerns about the reference standard and 29 (54%) for issues related to participant flow and timing. No study had low risk in all domains. We judged that there were high concerns about the applicability of the evidence related to participants in 44 (81%) studies, 17 (31%) related to the index test and 32 (59%) related to the reference standard. Explanations of how we have reached these judgements are given below and in the Characteristics of included studies table.

Participant selection

For participant selection, we judged only one study to be at low risk of bias and five to be of unclear risk. The remaining 48 (89%) we judged to be at high risk of bias (n = 44) either due to the use of a multi‐group design with healthy or other disease controls (n = 26) or recruitment of only COVID‐19 cases (n = 19), inappropriate exclusions (n = 2) or inappropriate inclusions (n = 15). Numbers per group are not mutually exclusive. Eleven studies (20%) reported consecutive or random recruitment of participants.

We had high concerns about the applicability of the selection of participants in 44 studies (81%) meaning that the participants who were recruited were unlikely to be similar to those in whom the test would be used in clinical practice. This was largely because studies only recruited hospitalised, confirmed cases of COVID‐19, often with severe symptoms (18 studies) or recruited healthy or other disease non‐COVID‐19 groups (26 studies). We judged 10 (19%) studies likely to have selected an appropriate patient group, including the six studies that recruited participants suspected of COVID‐19 prior to definitive testing and four multi‐group studies that separately recruited COVID‐19 cases and suspected COVID‐19 control groups.

Index tests

Eight studies explicitly reported that they had undertaken the index test with knowledge of whether individuals did or did not have COVID‐19, and eight studies determined the threshold to define test positivity by analysing the data, rather than it being pre‐determined. In 37 studies, reporting of one or both of these issues was too unclear to be able to rule out the possibility of bias. These issues led to the index test performance in 14 studies being rated as at high risk of bias. We judged only three studies to have implemented the index test in a way that protected against the risk of bias.

In 34 studies (63%) we judged the test to be implemented as it would be in practice. Twenty‐two of these were evaluations of laboratory‐based, commercially available tests, and 12 were evaluations of lateral flow assays associated with commercial test manufacturers, primarily evaluated in an inpatient setting. Two of the 12 evaluated the assays as point‐of‐care tests in an emergency room setting. Sixteen studies raised concerns that the tests could not be purchased (high concerns for applicability). The remaining four studies provided inadequate information to make a judgement due to withholding of the names of the commercial tests (one additional study also withheld the names of the lateral flow assays evaluated but scored high concerns as it also reported results for an in‐house ELISA test).

Reference standards

We judged 13 studies (24%) to have used an appropriate reference standard and implemented it in ways that prevented bias. In six studies there was a risk of misclassification, as they had used a single, negative RT‐PCR result to define the absence of disease in people with suspected COVID‐19; eight studies did not report any RT‐PCR testing to confirm COVID‐19 status for contemporaneous healthy or other disease non‐COVID‐19 groups; and one study used serology results in part to determine the reference standard diagnosis, thus risking incorporation bias. We judged 24 studies as having unclear risk of bias due to lack of information about blinding of the reference standard to the index test (19/24) or unclear descriptions of the reference standards used (6/24).

We judged the reference standard to be equivalent to WHO or China CDC definitions of COVID‐19 in 15 studies (28%). We judged studies that used a definition based only on RT‐PCR‐positive results as high concern (32 (59%) of studies), and seven studies reported inadequate detail to assess the reference standard.

Flow and timing

Twenty‐nine (54%) studies were at high risk of bias due to using different reference standards to verify COVID‐19 and non‐COVID‐19 cases (n = 19), participants being excluded from the analysis (n = 15), or the inclusion of multiple samples per participant (n = 7). In 20 (37%) studies we could not make judgements on one or more of these issues, primarily due to lack of clarity around participant inclusion and exclusion from analyses. Five studies reported adequate detail to rule out these risks of bias. None of the included studies reported a Standards of Reporting Diagnostic Accuracy Studies (STARD)‐style participant flow diagram (Bossuyt 2015), and none mentioned that they aimed to report in line with STARD reporting recommendations for test accuracy studies.

In 39 studies all authors declared no conflicts of interest although four included co‐authors affiliated to test manufacturers. Ten studies did not provide a conflict of interest statement (two of these included co‐authors affiliated to test manufacturers or biotechnology companies); and in the five remaining studies at least one author declared conflicts of interest in relation to test manufacturers (four studies) or vaccine companies (one study).

Nine studies provided no funding statement, six reported no funding sources to declare, and 39 studies reported one or more funding sources. The reported funding sources were primarily public funding sources. Two studies reported receipt of equipment ‘in kind’ from test manufacturers and two studies reported private donors.

Findings

We included 54 different studies, which were reported in 57 publications. Fourteen of the 54 studies evaluated more than one test (Table 1), up to a maximum of 10 tests per study. To incorporate all results from all tests, in these analyses we have treated results from different tests of the same samples within a study as separate data points, such that data are available on 89 test‐study combinations. This leads to individual samples being included in some analyses multiple times where they have been evaluated using different tests. To identify where estimates are based on multiple assessments of the same sample sets, the tables include both the number of test‐study combinations and the number of studies. The numbers of true positives, false positives, COVID‐19 samples and non‐COVID samples are based on test result counts.

Overall analyses

We are unable to distinguish between studies that evaluated the accuracy of antibody tests to identify current infection from past infection. Whilst time since onset of symptoms is strongly related to whether an infection was current or past, few studies reported whether participants' symptoms had resolved (and thus they were in a convalescent state) when serology samples were taken. Whilst 21 days post‐symptom onset is assumed to be a point where COVID‐19 cases are likely to be convalescent, many participants in these studies were hospitalised for prolonged periods and likely to reflect those with more severe and long‐lasting symptoms.

A key aspect of interpreting the sensitivity of the tests is the relationship between accuracy and days since onset of symptoms. Sixteen (30%) studies only presented results aggregated over 0 to more than 35 days since onset, and did not present data (or provide datasets) that disaggregated data by week. The figures in Appendix 11 show forest plots of sensitivity and specificity estimates including these studies for IgG, IgM, and IgG/IgM (either positive), which clearly depict substantial heterogeneity in sensitivity, with estimates ranging from 0% to 100% for all three markers. Forest plots of results for IgA, total antibodies, IgA/IgG, IgA/IgM (Appendix 11), show similar heterogeneity with smaller numbers of studies. Given the heterogeneity and the known strong relationship of sensitivity with time, computation of an average estimate of sensitivity from these studies would be misleading and serves no purpose.

Sensitivity by time since onset of symptoms

Table 2 and Figure 3 present the results disaggregated by week of testing since onset of symptoms for IgG (from 23 studies), IgA (from 4 studies), IgM (from 24 studies), total antibodies (from 5 studies), combination of IgG/IgM (from 21 studies), and IgA/IgG (from 1 study; these results are based on a maximum of 12 participants per time period and we will not comment on them further). We did not find any data disaggregated by week for IgA/IgM. Forest plots of these data are given in Figure 4, Figure 5 and Figure 6. We have undertaken meta‐analyses of data stratified by week as heterogeneity, whilst still present, is substantially less. As indicated in Table 2, the strength of the relationship of time with sensitivity shows exceptionally high levels of statistical significance (P < 0.0005). All further analyses of sensitivity in this report are thus stratified by week since symptom onset.

Open in table viewer
Table 2. Test sensitivity by time since onset of symptoms

Days 1‐7

Days 8‐14

Days 15‐21

Days 22‐35

Days > 35

Comparison

Test groups [studies] (true positives/COVID cases)

Sensitivity (95% CI)

IgG

33 [23] (165/568)

34 [22] (766/1200)

34 [22] (974/1110)

20 [12] (417/502)

11 [4] (213/252)

29.7% (22.1 to 38.6)

66.5% (57.9 to 74.2)

88.2% (83.5 to 91.8)

80.3% (72.4 to 86.4)

86.7% (79.6 to 91.7)

P < 0.00005

IgM

34 [24] (207/608)

32 [21] (724/1171)

32 [21] (800/1074)

19 [11] (378/507)

11 [4] 118/215

23.2% (14.9 to 34.2)

58.4% (45.5 to 70.3)

75.4% (64.3 to 83.8)

68.1% (55.0 to 78.9)

53.9% (38.4 to 68.6)

P < 0.00005

IgA

4 [4] (54/100)

3 [3] (38/53)

3 [3] (66/68)

2 [2] (81/82)

1 [1] (23/23)

28.4% (0.9 to 94.3)

78.1% (9.5 to 99.2)

98.7% (39.0 to 100)

98.7% (91.9 to 99.8)

100% (85.2 to 100)

*

Total antibodies

5 [4] (62/144)

6 [5] (220/247)

6 [5] (174/176)

4 [3] (11/19)

2 [1] (15/28)

24.5% (9.5 to 50.0)

84.0% (64.1 to 93.9)

98.1% (90.1 to 99.6)

69.5% (34.8 to 90.7)

79.0% (49.8 to 93.4)

P < 0.00005

IgG/IgM

17 [9] (81/259)

21 [9] (441/608)

21 [9] (636/692)

16 [5] (146/152)

9 [2] (122/153)

30.1% (21.4 to 40.7)

72.2% (63.5 to 79.5)

91.4% (87.0 to 94.4)

96.0% (90.6 to 98.3)

77.7% (66.0 to 86.2)

P < 0.00005

IgA/IgG

1 [1] (0/12)

1 [1] (5/10)

1 [1] (7/8)

1 [1] (1/1)

0 [0]

0% (0 to 26.5)

50.0% (18.7 to 81.3)

87.5% (47.3 to 99.6)

100% (2.5 to 100)

*

IgA/IgM

0 [0]

0 [0]

0 [0]

0 [0]

0 [0]

CI: confidence interval; * inadequate data to make a formal statistical comparison


Meta‐analytical estimates of sensitivity (with 95% CI) by antibody class and time since onset of symptoms

Meta‐analytical estimates of sensitivity (with 95% CI) by antibody class and time since onset of symptoms


Forest plot of studies evaluating tests for detection of IgG according to week post‐symptom onset and type of test

Forest plot of studies evaluating tests for detection of IgG according to week post‐symptom onset and type of test


Forest plot of studies evaluating tests for detection of IgM according to week post‐symptom onset and type of test

Forest plot of studies evaluating tests for detection of IgM according to week post‐symptom onset and type of test


Forest plot of studies evaluating tests for detection of IgG/IgM according to week post‐symptom onset and type of test

Forest plot of studies evaluating tests for detection of IgG/IgM according to week post‐symptom onset and type of test

The numbers of individuals contributing data within each study within each week are very small, thus by pooling these data across studies these meta‐analyses contribute clarity to the relationship between sensitivity and time, although the important limitations of these studies as described above should be considered when interpreting all findings.

Pooled results for IgG, IgM, IgA, total antibodies and IgG/IgM all show the same general pattern over the first three weeks, with sensitivity being low when tests were used in the first week since onset of symptoms, rising in the second week, and reaching their highest values in the third week. For IgG, sensitivity across the three weeks were 29.7% (95% confidence interval (CI) 22.1 to 38.6), 66.5% (95% CI 57.9 to 74.2) and 88.2% (95% CI 83.5 to 91.8); for IgM they were 23.2% (95% CI 14.9 to 34.2), 58.4% (95% CI 45.5 to 70.3) and 75.4% (95% CI 64.3 to 83.8); and for IgG/IgM they were 30.1% (95% CI 21.4 to 40.7), 72.2% (95% CI 63.5 to 79.5) and 91.4% (95% CI 87.0 to 94.4). Values for total antibodies and IgA are also given in Table 2.

It is important to note that these estimates are based on pooling multiple cross‐sectional studies, and are not based on tracking the same groups of participants over time or even using the same tests. The reasons why individuals are included at some particular time points and not at others is mostly not reported.

Estimates of sensitivity beyond three weeks are based on smaller sample sizes, with a maximum of 12 studies contributing data in weeks 4 and 5, and only four studies providing any follow‐up information beyond week 5. Estimates for IgA and total antibodies are based on fewer than 100 samples/participants and we will not comment upon them further. In weeks 4 and 5, pooled sensitivities of IgG were 80.3% (95% CI 72.4 to 86.4); IgM were 68.1% (95% CI 55.0 to 78.9); and for IgG/IgM were 96.0% (95% CI 90.6 to 98.3).

The data beyond week 5 gave sensitivity estimates of 86.7% (95% CI 79.6 to 91.7; IgG), 53.9% (95% CI 38.4 to 68.6; IgM) and 77.7% (95% CI 66.0 to 86.2; IgG/IgM). The expected decline in the sensitivity of IgM is evident.

Overall specificity

We estimated antibody test specificity from 35 studies. Specificity estimates for all studies are presented in Appendix 11 for IgG, IgM, IgG/IgM, IgA, total antibodies, and IgA/IgG. Results pooled across all studies are in Table 3 and show specificity exceeding 98% for all antibody types, with precise estimates (confidence intervals up to 2 percentage points wide), particularly for IgG, IgM, total antibodies and IgG/IgM, where estimates are based on several thousand non‐COVID samples. Inspection of the figures shows low heterogeneity in study estimates of specificity across studies. Nine studies provided some information on the cross‐reactivity of other infections, including other coronaviruses, with the SARS‐CoV‐2 antigens used in the assays (Table 4).

Open in table viewer
Table 3. Specificity and impact of reference standard for non‐COVID cases

Overall specificitya

COVID suspects deemed negative

Current healthy or other disease

Pre‐pandemic

Comparison of control groups

Test groups [studies] (false positives/non‐COVID cases)

Specificity (95% CI)

IgG

62 [44] (159/6136)

6 [6] (10/396)

14 [10] (60/2614)

19 [10] (88/2633)

99.1% (98.3% to 99.6%)

98.0% (91.0% to 99.6%)

99.2% (97.6% to 99.8%)

99.2% (97.8% to 99.7%)

P = 0.56

IgM

59 [41] (183/6103)

5 [5] (12/384)

14 [10] (89/3069)

17 [9] (38/2075)

98.7% (97.4% to 99.3%)

98.1% (89.9% to 99.7%)

98.6% (96.0% to 99.5%)

99.3% (98.0% to 99.8%)

P = 0.50

IgG/IgM

34 [23] (78/5761)

7 [7] (33/454)

7 [5] (20/506)

18 [6] (22/1104)

No formal comparison possible

98.7% (97.2% to 99.4%)

92.8% (89.7% to 95.0%)

99.9% (65.2% to 100%)

98.7% (96.6% to 99.5%)

Total antibodies

16 [10] (41/3585)

99.2% (98.3% to 99.6%)

IgA

4 [4] (10/663)

98.5% (97.2% to 99.2%)

IgA/IgGb

2 [2] (1/528)

99.8% (98.9% to 100%)

IgA/IgMb

1 [1] (1/483)

99.8% (99.2% to 100%)

CI: confidence interval

aIncludes studies that are categorised as mixed/other not included in the subgroups.
bConfidence intervals computed using binomial exact on totals.

Open in table viewer
Table 4. Reported cross‐reactivity with SARS‐CoV‐2 antigens

Study

Test(s) evaluated

What the study says about cross‐reactivity

Cai 2020

In‐house CLIA

Reported no cross‐reactivity in 167 sera from patients with infection with other pathogens (influenza A virus (25), respiratory syncytial virus (7), parainfluenza virus (8), influenza B virus (5), adenovirus (6), Klebsiella pneumoniae (8), Streptococcus pneumoniae (3), mycoplasma (5), Acinetobacter baumannii (10), Candida albicans (2), Staphylococcus aureus (3), Mycobacterium tuberculosis (4), hepatitis B virus (33), hepatitis C virus (22), syphilis (23) and saccharomycopsis (3)).

Freeman 2020

In‐house ELISA

Reported cross‐reactivity to SARS‐CoV‐2 spike protein in sera from patients with SARS‐1 and MERS‐CoV, and no cross‐reactivity with NL63, OC43, HKU1, 229E

Guo 2020a

In‐house ELISA

Reported Western Blot cross‐reactivity analysis in plasma samples positive for human CoV‐229E, ‐NL63, ‐OC43, ‐HKU1, and SARS‐CoV. Strong cross‐reactivity was observed only for SARS‐CoV.

Infantino 2020

Shenzhen YHLO CLIA

Observed no cross‐reactivity in sample from blood donors from the COVID‐19 era (winter 2019) but positive results in two samples from people with CMV infections and 2 with rheumatic disease.

Lassauniere 2020 [A]

[A] Beijing Wantai ELISA

[B] EUROIMMUN IgG ELISA

[C] EUROIMMUN IgA ELISA

[D] Dynamiker Biotechnology LFA

[E] CTK Biotech ‐ OnSite LFA

[F] Autobio Diagnostics LFA

[G] Artron Laboratories LFA

[H] Acro Biotech LFA

[I] Hangzhou Alltest LFA

Included sera from patients with acute viral respiratory tract infections caused by other coronaviruses (n = 5) or non‐coronaviruses (n = 45), and sera from patients positive for dengue virus (n = 9), CMV (n = 2) and Epstein Barr virus (n = 10). Cross reaction was observed for the EUROMIMMUN IgA ELISA (> 1 respiratory virus present, adenovirus, dengue virus) and for the EUROMIMMUN IgG ELISA (coronavirus HKU1 and adenovirus). Some cross‐reactivity also observed for CGIA tests. Study authors suggest related to antigen target and ELISA format.

Ma 2020a

In‐house CLIA

Limited detail but suggests limited cross‐reaction

Wang 2020a [A]

A. Beijing Hotgen IgM CGIA

B. Beijing Hotgen IgM ELISA

Demonstrated considerable cross‐reaction with rheumatoid factor IgM (22/36 false positive results). Other pathogens included influenza A virus (n = 5), influenza B virus (n = 5), Mycoplasma pneumoniae (n = 5), Legionella pneumophila (n = 5), HIV infection (n = 6), hypertension (n = 5) and diabetes mellitus (n = 5)

Zhang 2020b

Shenzhen YHLO CLIA

Observed false positive results in influenza A and B (2 each), adenovirus (n = 4) and Mycoplasma pneumoniae (n = 17).

Zhang 2020d

In‐house CGIA (co‐author Beijing Hotgen)

Appears to report a separate cross‐reactivity study for influenza A, influenza B, respiratory syncytial virus, Mycoplasma pneumoniae and Chlamydia pneumoniae. No cross reactions were observed.

CGIA: colloidal gold immunoassay; CLIA: Chemiluminescence immunoassay; CMV: cytomegalovirus; ELISA: enzyme‐linked immunosorbent assay; LFA: lateral flow assay; MERS: Middle East respiratory syndrome; SARS: severe acute respiratory syndrome

Impact of reference standard for COVID‐19 cases on sensitivity

The majority of studies only included participants who were diagnosed with COVID‐19 based upon observing a positive RT‐PCR test. However, in clinical practice it is common to encounter patients from whom positive RT‐PCR results are never obtained, but who demonstrate clinical and imaging features of COVID‐19. Diagnostic criteria for COVID‐19 produced by WHO and the China CDC include definitions for suspected COVID‐19 in RT‐PCR‐negative patients. Twelve studies defined the presence of COVID‐19 using these criteria, thus including RT‐PCR‐negative patients in the COVID‐19 group as well as RT‐PCR‐positive patients. We compared estimates of sensitivity between studies using a RT‐PCR‐positive reference standard definition with a criteria‐based reference standard (including both RT‐PCR‐positives and RT‐PCR‐negatives; Table 5). We stratified the analysis for weeks since onset of symptoms. All the observed differences were within magnitudes expected by chance.

Open in table viewer
Table 5. Investigation of impact of reference standard on sensitivity

RT‐PCR‐positive COVID‐19 cases

RT‐PCR‐negative COVID‐19 cases

Comparison

Test groups [studies] (true positives/COVID cases)

Sensitivity (95% CI)a

IgG

26 [15] (1555/2280)

8 [8] (925/1300)

87.9% (82.7 to 91.7)

91.2% (83.9 to 95.4)

P = 0.36

IgM

23 [13] (1368/2166)

10 [9] (792/1292)

70.8% (56.3 to 82.0)

87.5% (73.7 to 94.6)

P = 0.06

IgG/IgM

17 [6] (966/1278)

4 [4] (400/499)

90.6% (86.6 to 93.5)

93.6% (88.9 to 96.4)

P = 0.22

CI: confidence interval; RT‐PCR: reverse transcription polymerase chain reaction

aWe obtained sensitivity estimates from a model of all data stratified by week, estimating the average difference in sensitivity across follow‐up. The figures quoted correspond to the week 3 strata (15‐21 days) in the model.

In a further analysis, we separated COVID‐19 participants who were RT‐PCR‐positive from those who were RT‐PCR‐negative, where studies allowed, and subgrouped the results to investigate whether there is a difference in accuracy according to RT‐PCR status. Data from only three studies could be included in this analysis (Figure 7; Figure 8; Figure 9). Differences in estimates of sensitivity (pooled stratifying for weeks since onset of symptoms), varied in direction for IgG and IgM, and were very similar for IgG/IgM (Table 6). All differences were within magnitudes expected by chance. There was no consistent evidence that the accuracy of serology tests was lower in RT‐PCR‐positive patients, although there is high uncertainty in these findings.


Sensitivity of IgG in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Sensitivity of IgG in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.


Sensitivity of IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Sensitivity of IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.


Sensitivity of IgG/IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Sensitivity of IgG/IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Open in table viewer
Table 6. Studies reporting sensitivity in both RT‐PCR‐positive and RT‐PCR‐negative subgroups

RT‐PCR‐positive COVID‐19 cases

RT‐PCR‐negative COVID‐19 cases

Test groups [studies]

(True positives/COVID‐19 cases)

Sensitivity (95% CI)

Test groups [studies]

(True positives/COVID‐19 cases)

Sensitivity (95% CI)

IgG

Days 1‐7b

2 [2] (1/28)

2 [2] (8/13)

Days 8‐14b

2 [2] (21/33)

3 [3] (25/30)

Days 15‐21b

2 [2] (39/40)

3 [3] (64/72)

Pooleda (stratified by time)

72.6% (46.2% to 89.1%)

84.0% (64.4% to 93.9%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative groups: P = 0.18

IgM

Days 1‐7b

2 [2] (3/28)

2 [2] (4/13)

Days 8‐14b

2 [2] (25/33)

3 [3] (11/30)

Days 15‐21b

2 [2] (8/16)

3 [3] (31/72)

Pooleda (stratified by time)

64.6% (49.7% to 77.1%)

49.0% (34.2% to 63.9%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative group: P = 0.07

IgG/IgM

Days 1‐7b

2 [2] (8/36)

2 [2] (4/17)

Days 8‐14b

2 [2] (37/53)

3 [3] (29/40)

Days 15‐21b

2 [2] (141/150)

3 [3] (104/113)

Pooleda (stratified by time)

71.9% (58.7% to 82.2%)

71.1% (57.0% to 82.0%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative group: P = 0.90

CI: confidence interval; RT‐PCR: reverse transcription polymerase chain reaction

aThe sensitivity estimates are produced from a model that combines all data from both subgroups and time‐groups, stratifying by time‐group. The estimate corresponds to sensitivity in Days 15‐21.
bRT‐PCR‐positive data have only been included here when the study includes a RT‐PCR‐negative subgroup as well.

Impact of reference standard for non‐COVID‐19 cases on specificity

We classified the reference standard used to verify non‐COVID cases into three main groups: pre‐pandemic controls (both healthy and with other diseases) who underwent no RT‐PCR testing, current controls from healthy or other disease groups (typically who also did not undergo RT‐PCR testing), and individuals who were investigated for COVID‐19 but deemed non‐COVID cases. Whilst results were similar for IgG and IgM, we noted more false positives for the IgG/IgM outcome in the studies using a COVID suspect group than in other studies (Table 3).

Sensitivity and specificity by assay type

We further investigated the heterogeneity in sensitivity estimates at any time point according to test technology type. We considered differences between CGIA, CLIAs, ELISAs and tests we can only describe as lateral flow assays due to lack of any names or detail (this group originate from the UK National COVID Testing Scientific Advisory Panel, which withheld names of the tests evaluated due to confidentiality clauses in the legal contracts with the manufacturers Adams 2020 [A]). There were inadequate numbers of studies evaluating FIAs and indirect immunofluorescence tests, luciferase immunoprecipitation assays and 'S‐flow' assays to analyse, and we were only able to assess IgG, IgM and IgG/IgM targets. In a sensitivity analysis we restricted the included studies to those that used commercial (rather than in‐house) tests.

We obtained estimates from a model that included all data stratified by weeks since onset of symptoms. The results presented in Table 7 and below correspond to estimates from the model of performance in week 3 post‐symptom onset.

Open in table viewer
Table 7. Sensitivity and specificity by test technology

Test method

Test method

CGIA

CLIA

ELISA

LFA

Comparison

IgG

Test groups [studies]

(True positives/COVID cases)

6 [5] (268/397)

10 [10] (1112/1432)

12 [11] (1014/1552)

7 [1] (133/238)

Sensitivity (95% CI)a

87.3% (77.0 to 93.4)

94.6% (90.7 to 97.0)

85.8% (78.0 to 91.1)

76.0% (61.0 to 86.5)

P = 0.004

Test groups [studies]

(True negatives/non‐COVID cases)

11 [11] (409/415)

12 [12] (318/322)

18 [16] (2003/2102)

6 [1] (354/360)

Specificity (95% CI)a

99.5% (96.5 to 99.9)

99.0% (91.6 to 99.9)

98.8% (96.5 to 99.6)

99.0% (95.3 to 99.8)

P = 0.85

IgM

Test groups [studies]

(True positives/COVID cases)

7 [6] (109/411)

10 [10] (884/1355)

12 [11] (1083/1568)

7 [1] (78/228)

Sensitivity (95% CI)a

69.5% (44.3 to 86.7)

80.9% (63.8 to 91.0)

84.5% (70.7 to 92.5)

51.4% (26.5 to 75.6)

P = 0.11

Test groups [studies]

(True negatives/non‐COVID cases)

12 [11] (455/487)

13 [13] (609/621)

14 [12] (1674/1710)

6 [1] (357/360)

Specificity (95% CI)a

97.3 (90.0 to 99.3)

98.5 (92.3 to 99.7)

99.1 (97.2 to 99.7)

99.6 (97.3 to 99.9)

P = 0.40

IgG/IgM

Test groups [studies]

(True positives/COVID cases)

4 [3] (232/316)

3 [3] (344/420)

5 [4] (595/770)

11 [2] (255/358)

Sensitivity (95% CI)a

90.7% (82.7 to 95.2)

97.5% (94.0 to 99.0)

90.7% (83.3 to 95.0)

88.6% (82.0 to 93.0)

P = 0.02

Test groups [studies]

(True negatives/non‐COVID cases)

11 [11] (330/353)

5 [4] (230/244)

5 [4] (387/391)

13 [3] (3797/3827)

Specificity (95% CI)a

96.0 (90.1 to 98.5)

94.1 (82.7 to 98.2)

99.4 (97.4 to 99.9)

98.2 (96.3 to 99.1)

P = 0.05

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; LFA: lateral flow assay (no further detail)

aWe obtained sensitivity estimates from a model of all data stratified by week, estimating the average difference in sensitivity across follow‐up. The figures quoted correspond to the Week 3 (15‐21 days) strata in the model.

For IgG, there were clear differences in the sensitivity of assays, with CLIA (94.6%), CGIA (87.3%) and ELISA (85.8%) all outperforming the unknown lateral flow assay tests (76.0%). The differences between the groups was beyond that expected by chance (P = 0.004), but largely driven by the low value for lateral flow tests (all of the data coming from 40 COVID‐19 patients in the UK National COVID Testing Scientific Advisory Panel study tested multiple times).

For IgM, although laboratory‐based ELISA (84.5%) and CLIA (80.9%) outranked lateral flow CGIA (69.5%) and the unknown lateral flow assays (51.4%), the differences observed were in the realms of those expected by chance (P = 0.11).

In the smaller subset of studies that evaluated tests combining IgM/IgG, the performance of laboratory CLIA tests (97.3%) ranked above those of CGIA (91.4%), ELISA (90.5%) and unknown lateral flow tests (85.8%). These differences were beyond those expected by chance (P = 0.01)

Excluding the in‐house tests, and thus restricting the analysis to only commercial tests, made little difference to estimates of sensitivity.

Analyses of specificity presented by assay type are also given in Table 7. Differences in specificity of IgG and IgM between assay types were small, CLIA and CGIA tests showed lower specificity for IgG/IgM tests than ELISA and LFIA, but confidence intervals on all estimates are wide.

Sensitivity and specificity by brand

We have tabulated the results by brand for the 27 commercial tests: 15 tests for IgG Table 8; 14 tests for IgM Table 9; and nine tests for IgG/IgM Table 10. The study data for these estimates are provided in Figure 4, Figure 5 and Figure 6. Appendix 12 tabulates the information that we have been able to derive regarding the current availability of these commercially produced tests. Data for sensitivity are stratified by week of onset of symptoms and we present the numbers of studies and samples from which data are available for each time interval. Caution is required in the interpretation of these data as many are based only on single studies with small sample sizes. We present confidence intervals to quantify the uncertainty in the estimates. We would advise focusing on estimates based on at least 100 samples/participants per week further. Three tests have estimates of sensitivity based on more than 100 samples (Beijing Wantai ELISA, Bioscience Co. (Chongqing) CLIA, Zuhai Livzon ELISA). We evaluated the studies that we pooled to create these estimates as having multiple domains at risk of bias and having concerns about the applicability of the findings (all studies having at most 2 of the 7 ratings in the QUADAS‐2 assessment described as low risk or low concern).

Open in table viewer
Table 8. Sensitivity and specificity by test brand (IgG)

Test namea

Test

method

IgG sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitivity (95% CI)

IgG specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Beijing Beier Bioengineering

CGIA

1 (2/10)

1 (6/13)

1 (11/14)

20.0% (2.5 to 55.6)

46.2% (19.2 to 74.9)

78.6% (49.2 to 95.3)

Beijing Beier Bioengineering

CLIA

1 (4/10)

1 (6/13)

1 (9/14)

40.0% (12.2 to 73.8)

46.2% (19.2 to 74.9)

64.3% (35.1 to 87.2)

Beijing Beier Bioengineering

ELISA

1 (4/10)

1 (8/13)

1 (12/14)

40.0% (12.2 to 73.8)

61.5% (31.6 to 86.1)

85.7% (57.2 to 98.2)

Beijing Hotgen

ELISA

1 (9/22)

1 (60/92)

1 (51/55)

1 (39/45)

2 (22/172)

40.9% (20.7 to 63.6)

65.2% (54.6 to 74.9)

92.7% (82.4 to 98.0)

86.7% (73.2 to 94.9)

87.2% (81.3 to 91.8)

Beijing Wantai

ELISA

2 (31/133)

2 (130/210)

2 (127/149)

2 (2/297)

23.3% (16.4 to 31.4)

61.9% (55.0 to 68.5)

85.2% (78.5 to 90.5)

99.3% (97.6 to 99.9)

Beijing Wantai

CGIA

1 (1/209)

99.5% (97.4 to 100)

Bioscience Co (Chongqing)

CLIA

2 (43/92)

2 (129/212)

2 (208/244)

2 (98/164)

1 (75/76)

46.7% (36.3 to 57.4)

60.8% (53.9 to 67.5)

85.2% ( 80.2 to 89.4)

59.8% (51.8 to 67.3)

98.6% (92.9 to 100)

Darui Biotech

ELISA

1 (0/64)

100% (94.4 to 100)

EUROIMMUN

ELISA

1 (2/13)

2 (13/25)

2 (14/15)

2 (98/164)

2 (3/82)

15.4% (1.9 to 45.4)

52.0% (31.3 to 72.2)

93.3% (68.1 to 99.8)

59.8% (51.8 to 67.3)

96.3% (89.7 to 99.2)

EUROIMMUN Anti‐SARS‐Cov

IIFT

1 (1/4)

1 (3/5)

1 (3/3)

1 (1/1)

1 (0/10)

25.0% (0.6 to 80.6)

60.0% (14.7 to 94.7)

100% (29.2 to 100)

100% (2.5 to 100)

100% (69.2 to 100)

EUROIMMUN Beta

ELISA

1 (0/12)

1 (3/10)

1 (7/8)

1 (1/1)

1 (0/45)

0% (0 to 26.5)

30%.0% (14.7 to 94.7)

87.5% (47.3 to 99.7)

100% (2.5 to 100)

100% (92.1 to 100)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (21/42)

2 (57/68)

2 (0/45)

12.5% (0.3 to 52.7)

50.0% (34.2 to 65.8)

83.8% (72.9 to 91.6)

100% (92.1 to 100)

Innovita Biological ‐ Ab test (IgM/IgG)

CGIA

1 (7/13)

1 (7/8)

1 (21/23)

53.8% (25.1 to 80.8)

87.5% (47.3 to 99.7)

91.3% (72.0 to 98.9)

Shenzhen YHLO

CLIA

2 (2/8)

2 (28/29)

2 (25/26)

2 (64/64)

1 (7/7)

7 (4/322)

25.0% (3.2 to 65.1)

96.6% (82.2 to 99.9)

96.2% (80.4 to 99.9)

100% (94.4 to 100)

100% (59.0 to 100)

98.8% (96.9 to 99.7)

Snibe Diagnostic ‐ MAGLUMI

CLIA

2 (11/40)

2 (35/48)

25/25

27.5% (14.6 to 43.9)

72.9% (58.2 to 84.7)

100.0% (86.3 to 100)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

2 (0/42)

100% (91.6 to 100)

Zhuhai Livzon

CGIA

1 (5/36)

1 (20/34)

1 (35/38)

2 (0/35)

13.9% (4.7 to 29.5)

58.8% (40.7 to 75.4)

92.1% (78.6 to 98.3)

100% ( 90.0 to 100)

Zhuhai Livzon

ELISA

4 (17/80)

3 (163/288)

3 (197/223)

2 (91/104)

5 (5/351)

21.3% (12.9 to 31.8)

56.6% (50.7 to 62.4)

88.3% (83.4 to 92.2)

87.5% (79.6 to 93.2)

98.6% (96.7 to 99.5)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Open in table viewer
Table 9. Sensitivity and specificity by test brand (IgM)

Test namea

Test method

IgM sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitiivity (95% CI)

IgM specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Artron Laboratories IgM/IgG

CGIA

1 (5/7)

1 (12/15)

1 (8/8)

71.4% (29.0 to 96.3)

80.0% (51.9 to 95.7)

100% (63.1 to 100)

Autobio Diagnostics IgM/IgG

CGIA

1 (6/7)

1 (14/15)

1(8/8)

85.7% (42.1 to 99.6)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

Beijing Hotgen

ELISA

1 (10/22)

1 (72/92)

1 (72/92)

1 (41/45)

1 (0/100)

45.5% (24.4 to 67.8)

78.3% (68.4 to 86.2)

78.3% (68.4 to 86.2)

91.1% (78.8 to 97.5)

100% (96.4 to 100)

Beijing Hotgen

CGIA

1 (22/72)

69.4% (57.5 to 79.8)

Beijing Wantai

ELISA

1 (3/513)

99.4% (98.3 to 99.9)

Beijing Wantai

CGIA

1 (4/209)

98.1% (95.2 to 99.5)

Bioscience Co (Chongqing)

CLIA

1 (34/67)

1 (34/67)

1 (131/134)

1 (13/13)

50.7% (38.2 to 63.2)

50.7% (38.2 to 63.2)

97.8% (93.6 to 99.5)

100% (75.3 to 100)

CTK Biotech OnSite IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

Darui Biotech

ELISA

1 (14/64)

78.1% (66.0 to 87.5)

Dynamiker Biotechnology IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

EUROIMMUN

ELISA

1 (76/82)

92.7% (84.8 to 97.3)

EUROIMMUN Anti‐SARS‐Cov

IIFT

1 (1/10)

90.0% (55.5 to 99.7)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (23/42)

2 (58/68)

2 (0/45)

12.5% (0.3 to 52.7)

54.8% (38.7 to 70.2)

85.3% (74.6 to 92.7)

100% (92.1 to 100)

Shenzhen YHLO

CLIA

7 (10/321)

96.9% (94.3 to 98.5)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

2 (1/42)

97.6% (87.4 to 99.9)

Xiamen InnodDx Biotech

CLIA

1 (2/300)

99.3% (97.6 to 99.9)

Zhuhai Livzon

CGIA

1 (7/36)

1 (31/34)

1 (35/38)

2 (0/35)

19.4% (8.2 to 36.0)

91.2% (76.3 to 98.1)

92.1% (78.6 to 98.3)

100% ( 90.0 to 100)

Zhuhai Livzon

ELISA

3 (14/66)

2 (150/202)

2 (159/166)

1 (43/45)

5 (3/351)

21.2% (12.1 to 33.0)

74.3% (67.7 to 80.1)

95.8% (91.5 to 98.3)

95.6% (84.9 to 99.5)

99.1% (97.5 to 99.8)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Open in table viewer
Table 10. Sensitivity and specificity by test brand (IgG/IgM)

Test namea

Test method

IgG/IgM sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitiivity (95% CI)

IgG/IgM specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Acro Biotech ‐ IgG/IgM

CGIA

1 (3/15)

80.0% (51.9 to 95.7)

Artron Laboratories IgM/IgG

CGIA

1 (5/7)

1 (12/15)

1 (8/8)

1 (0/17)

71.4% (29.0 to 96.3)

80.0% (51.9 to 95.7)

100% (63.1 to 100)

100% (80.5% to 100)

Autobio Diagnostics IgM/IgG

CGIA

1 (6/7)

1 (14/15)

1(8/8)

1 (0/32)

85.7% (42.1 to 99.6)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Beijing Hotgen

ELISA

1 (10/22)

1 (72/92)

1 (72/92)

1 (41/45)

1 (0/100)

45.5% (24.4 to 67.8)

78.3% (68.4 to 86.2)

78.3% (68.4 to 86.2)

91.1% (78.8 to 97.5)

100% (96.4 to 100)

Bioscience Co (Chongqing)

CLIA

1 (34/67)

1 (34/67)

1 (131/134)

1 (13/13)

2 (7/148)

50.7% (38.2 to 63.2)

50.7% (38.2 to 63.2)

97.8% (93.6 to 99.5)

100% (75.3 to 100)

95.3% (90.5 to 98.1)

CTK Biotech OnSite IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

1 (0/32)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Dynamiker Biotechnology IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

1 (0/32)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (23/42)

2 (58/68)

3 (2/60)

12.5% (0.3 to 52.7)

54.8% (38.7 to 70.2)

85.3% (74.6 to 92.7)

96.7% (88.5 to 99.6)

Shenzhen YHLO

CLIA

2 (7/96)

92.7% (85.6 to 97.0)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

3 (14/162)

91.4% (85.9 to 95.2)

Zhuhai Livzon

CGIA

1 (7/36)

1 (31/34)

1 (35/38)

2 (0/35)

19.4% (8.2 to 36.0)

91.2% (76.3 to 98.1)

92.1% (78.6 to 98.3)

100% (90.0 to 100)

Zhuhai Livzon

ELISA

3 (14/66)

2 (150/202)

2 (159/166)

1 (43/45)

4 (4/291)

21.2% (12.1 to 33.0)

74.3% (67.7 to 80.1)

95.8% (91.5 to 98.3)

95.6% (84.9 to 99.5)

98.6% (96.5 to 99.6)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Eight tests have estimates of specificity based on more than 100 samples, with estimates over 98% for five tests (Bejing Hotgen ELISA, Beijing Wantai ELISA, Beijing Wantai CGIA, Xiamen InnodDx Biotech ELISA, Zhuhai Livzon ELISA). Again please note the concerns in the risk of bias and applicability of these findings.

Other sources of heterogeneity

Our protocol included additional planned analyses by:

  • current infection or past infection;

  • study design; and

  • setting.

We could not investigate these sources because of lack of variability across the studies in these features. Only two studies explicitly stated that they recruited only convalescent patients, and 48 (85%) studies recruited hospital inpatients. For study design only five out of 54 (11%) studies recruited a single group of suspected COVID‐19 patients, and did not use a 'COVID‐19 cases only' study, or a 'two‐group' study design.

Investigation of publication bias

We observed direct evidence of selective reporting through the withholding of names of the nine lateral flow assay testing brands from the UK National COVID Testing Scientific Advisory Panel study (Adams 2020 [A]). The paper states, "Individual manufacturers did not approve release of device‐level data, so device names are anonymised" (Adams 2020 [A]). The sensitivity estimates for the lateral flow assays in this study (which are most likely to be CGIA) were noted to be lower than estimates for CGIA tests from other studies. Four other studies also did not identify the test that they were evaluating.

Discussion

This is the first version of a Cochrane living review summarising the accuracy of antibody tests for detecting current or previous SARS‐CoV‐2infection. This version of the review is based on published studies or studies available as preprints up until the 27 April 2020. The speed of development and publication of studies for COVID‐19 antibody tests is unprecedented, and the content of this review will always be out of date. We are continuously identifying new published studies, and plan to update this review several times during the next few months.

The studies included in this version are largely from China, evaluating tests from Chinese universities and manufacturers. Many of the studies are the first that have been published for each test, and thus are early‐phase studies. Whilst there is no recognised stage classification of diagnostic studies, there are several common features of those undertaken during test development. These include multiple tests being described as 'in‐house', that thresholds for tests are determined from the data collected during the study, that all tests are undertaken by technical experts in laboratories, that the samples used are from collections easily available to the research team, and that multiple samples are used from the same participants. These limitations explain much of the rating for high risk of bias and concerns about applicability in this review. Many of these issues make it likely that the accuracy of tests when used in clinical care will be lower than that observed here. We did locate six evaluations recruiting patients identified in clinical pathways before it was established whether they had COVID‐19. This is more likely to produce results that reflect clinical practice, and we encourage future evaluations to consider this study design.

A concern with this review, and with its updates, is the high likelihood of selective reporting of results, particularly by manufacturers. We have already noted manufacturers being unwilling to be identified in the UK National COVID Testing Scientific Advisory Panel study (Adams 2020 [A]). Unlike randomised controlled trials of interventions, there are no requirements for test accuracy studies to be prospectively registered on study registers, nor to publish their findings. Many industry studies are only briefly described on 'Information for use' documents included with the tests, and study reports submitted to regulators are regarded as confidential. We are also aware that there are independent studies undertaken by National Public Health bodies, some of which have been submitted to FIND's data tracking tool for speedy data sharing. We plead for greater transparency and full publication in this field and continue to encourage laboratories to submit data and reports via FIND's portal. We request sharing of any unpublished reports for inclusion in future updates (please send to [email protected]). We have contacted test manufacturers to request full study reports which we will include in a future update of this review.

Summary of main results

We summarise 10 key findings from this review.

  1. Evaluations of most antibody tests on the market are not available as publications or even as preprints. This review has evaluated data from 25 commercial tests and numerous in‐house assays. These represent a small fraction of the antibody assays currently available. We have identified 66 additional studies of antibody tests published or available as preprints up until 25 May 2020, which we will appraise for inclusion in the review update, but there still remain no published data for the majority of tests on the current FIND list.

  2. The design and execution of the current studies limits the strength of conclusions that we are currently able to draw. Nearly all studies sampled COVID‐19 cases and non‐COVID cases separately, and methods for selecting participants were not described. Only four studies reported blinding reference standard and index tests, and some reference standards may misclassify individuals.

  3. Many studies only applied tests in laboratory settings on plasma or serum, whilst they are also approved for use as point‐of‐care tests using whole blood. From these data it is not possible to ascertain the clinical accuracy of these tests in lower resource and more accessible settings.

  4. Sensitivity varies with the time since of onset of symptoms. Figures from the studies showed the ability of antibody tests to detect SARS‐CoV‐2infection is very low in the first week (average sensitivity 30.1%, 95% CI 21.4 to 40.7) and only moderate (average sensitivity 72.2%, 95% CI 63.5 to 79.5) in the second week post‐symptom onset. These estimates are based on patients who have been hospitalised with COVID‐19, and remain in hospital at the time of sampling, and thus are likely to represent the more severe end of the disease spectrum and are potentially individuals with higher antibody responses.

  5. Tests have higher sensitivity when done later in the course of the disease. The average sensitivity across all the included studies for IgG/IgM tests was estimated from the included studies as 91.4% (95% CI 87.0 to 94.4) for 15 to 21 days, and 96.0% (95% CI 90.6 to 98.3) for 22 to 35 days. Too few studies had evaluated tests beyond 35 days to estimate accuracy. These findings are expected given the delayed rise of IgG antibodies.

  6. Studies estimate the specificity of tests precisely, and it appears to be high. The average from the studies for IgG/IgM is 98.7% (95% CI 97.2% to 99.4%). However, estimates of specificity are mainly based on testing pre‐pandemic, healthy people, or people known to have other disorders, and not those being investigated for possible COVID‐19.

  7. From the limited evaluations studied, some differences were noted by test technology, CLIA methods appearing more sensitive (97.5%, 95% CI 94.0 to 99.0) than ELISA (90.7%, 95% CI 83.3 to 95.0) or CGIA‐based lateral flow assays (90.7%, 95% CI 82.7 to 95.2) for IgG/IgM, (there are also differences for IgG but no differences for IgM). There was little clear evidence of differences in specificity between technology types.

  8. There is currently too little data on individual tests to be able to consider comparisons of their performance.

  9. Study reports did not include many of the key items listed on the STARD reporting guideline for test accuracy studies (Bossuyt 2015), which has hindered assessment and data extraction. No study utilised a STARD participant flow diagram to enable identification of missing, indeterminate or unavailable test results.

  10. We observed partial reporting (suppression of the identify of tests) in five studies, indicating the likelihood of publication bias.

Strengths and weaknesses of the review

Our review used a broad search screening all articles concerning COVID‐19. We undertook all screening and eligibility assessments, QUADAS‐2 assessments (Whiting 2011), and data extraction of study findings independently and in duplicate. Whilst we thus have reasonable confidence in the completeness and accuracy of the findings up until the search date, should errors be noted please inform us at [email protected] so that we can check and correct in our next update.

Weaknesses of the review primarily reflect the weaknesses in the primary studies and their reporting. Many studies omitted descriptions of sample recruitment, and key aspects of study design and execution. Some studies omit information that allows the tests to be identified. We have had to treat studies that describe their data as being based on 'samples' as if the samples were individual patients. We have been explicit about these issues where they arose.

More than half (28/54) of the studies we have included are currently only available as preprints, and as yet, have not undergone peer review. As published versions of these studies are identified in the future, we will double‐check study descriptions, methods and findings, and update the review as required.

We also did not make within‐study comparisons between tests. Two studies (Adams 2020 [A]; Lassauniere 2020 [A]), evaluated panels of nine or 10 tests, nine other studies evaluated two, three, or five tests. As we could not identify tests in Adams 2020 [A], and the sample of Lassauniere 2020 [A] was very small, it is not possible from the studies available at this time to make direct comparisons between alternative tests.

We identified only one study that included comparison of test results with a reference standard of a neutralisation assay in studies identified for inclusion in this first version of the review (Thompson 2020), but we did not include these data in this version of the review. We are aware of several more studies of these assays in more recent publications and will include this as a new target condition in the next update of the review.

In such a current and fast moving field searches will always be out of date. However we are committed to ongoing updates of this living review

Applicability of findings to the review question

In the background we outlined four main roles for antibody testing that would be addressed in this review.

  1. In diagnosis of infection in patients presenting with symptoms of suspected COVID‐19, particularly where molecular testing had failed to detect the virus. Most studies included in the review collected data from patients in the acute phase of disease in hospital settings and thus provide evidence to address this question amongst hospitalised patients. The review showed that antibody tests had very low sensitivity in the first week following onset of symptoms, but sensitivity rose in the second week, and only exceeded 90% in the third week. In addition we saw no difference in sensitivity of tests according to RT‐PCR status. We had no data to inform the accuracy of the test in primary care and community settings for the purpose of diagnosis, where patients are likely to have milder symptoms.

  2. In assessment of immune response in patients with severe disease. We stated in the Background that we would not cover this in this review. In any case, we found no studies that directly addressed this question. Assessment of the accuracy of a test used for assessment of immune response would involve comparison with a reference standard test of antibody response, rather than evidence of infection.

  3. To assess whether individuals have had a SARS‐CoV‐2 infection. As above, we found no studies that directly addressed this question, and very few studies were undertaken in community settings in patients who had not undergone RT‐PCR testing during their symptomatic period. Conclusions about the likely value of tests for this purpose rely on the sensitivity of the tests being no different in mild disease than in severe disease that requires hospital admission.

  4. In seroprevalence surveys for public health management purposes. We also found no studies that directly addressed this question (although Bendavid 2020 is a seroprevalence study, it did not evaluate the accuracy of the test in the seroprevalence sample). High specificity of tests is essential in seroprevalence testing, which appears likely for many of the tests included in this review. However, the suitability of pre‐pandemic samples to establish specificity requires further discussion. We found no difference in specificity between pre‐pandemic and current non‐COVID‐19 samples, but lower specificity in those where COVID‐19 was ruled out after initially being suspected. This either reflects misclassification, or a true lower specificity in those presenting with symptoms. As sensitivity of the tests was mainly evaluated in hospitalised patients it is also unclear whether the tests have the ability to detect lower antibody levels likely in non‐hospitalised COVID‐19 patients.

Study flow diagram

Figures and Tables -
Figure 1

Study flow diagram

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Figures and Tables -
Figure 2

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Meta‐analytical estimates of sensitivity (with 95% CI) by antibody class and time since onset of symptoms

Figures and Tables -
Figure 3

Meta‐analytical estimates of sensitivity (with 95% CI) by antibody class and time since onset of symptoms

Forest plot of studies evaluating tests for detection of IgG according to week post‐symptom onset and type of test

Figures and Tables -
Figure 4

Forest plot of studies evaluating tests for detection of IgG according to week post‐symptom onset and type of test

Forest plot of studies evaluating tests for detection of IgM according to week post‐symptom onset and type of test

Figures and Tables -
Figure 5

Forest plot of studies evaluating tests for detection of IgM according to week post‐symptom onset and type of test

Forest plot of studies evaluating tests for detection of IgG/IgM according to week post‐symptom onset and type of test

Figures and Tables -
Figure 6

Forest plot of studies evaluating tests for detection of IgG/IgM according to week post‐symptom onset and type of test

Sensitivity of IgG in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Figures and Tables -
Figure 7

Sensitivity of IgG in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Sensitivity of IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Figures and Tables -
Figure 8

Sensitivity of IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Sensitivity of IgG/IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Figures and Tables -
Figure 9

Sensitivity of IgG/IgM in PCR+ve and PCR‐ve COVID‐19 cases by week since onset of symptoms.

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study

Figures and Tables -
Figure 10

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study

Forest plot of studies evaluating tests for detection of IgG at all time post‐symptom onset

Figures and Tables -
Figure 11

Forest plot of studies evaluating tests for detection of IgG at all time post‐symptom onset

Forest plot of studies evaluating tests for detection of IgM at all time post‐symptom onset

Figures and Tables -
Figure 12

Forest plot of studies evaluating tests for detection of IgM at all time post‐symptom onset

Forest plot of studies evaluating tests for detection of IgG/IgM at all time post‐symptom onset.

Figures and Tables -
Figure 13

Forest plot of studies evaluating tests for detection of IgG/IgM at all time post‐symptom onset.

Forest plot of tests: 19 IgA (all time points), 25 Total antibodies (Ab) (all time points), 38 IgA/IgM (all time points).

Figures and Tables -
Figure 14

Forest plot of tests: 19 IgA (all time points), 25 Total antibodies (Ab) (all time points), 38 IgA/IgM (all time points).

IgG (all time points)

Figures and Tables -
Test 1

IgG (all time points)

IgG (1 to 7 days)

Figures and Tables -
Test 2

IgG (1 to 7 days)

IgG (8 to 14 days)

Figures and Tables -
Test 3

IgG (8 to 14 days)

IgG (15 to 21 days)

Figures and Tables -
Test 4

IgG (15 to 21 days)

IgG (22 to 35 days)

Figures and Tables -
Test 5

IgG (22 to 35 days)

IgG (over 35 days)

Figures and Tables -
Test 6

IgG (over 35 days)

IgM (all time points)

Figures and Tables -
Test 7

IgM (all time points)

IgM (8 to 14 days)

Figures and Tables -
Test 8

IgM (8 to 14 days)

IgM (1 to 7 days)

Figures and Tables -
Test 9

IgM (1 to 7 days)

IgM (15 to 21 days)

Figures and Tables -
Test 10

IgM (15 to 21 days)

IgM (22 to 35 days)

Figures and Tables -
Test 11

IgM (22 to 35 days)

IgM (over 35 days)

Figures and Tables -
Test 12

IgM (over 35 days)

IgG/IgM (all time points)

Figures and Tables -
Test 13

IgG/IgM (all time points)

IgG/IgM (1 to 7 days)

Figures and Tables -
Test 14

IgG/IgM (1 to 7 days)

IgG/IgM (8 to 14 days)

Figures and Tables -
Test 15

IgG/IgM (8 to 14 days)

IgG/IgM (15 to 21 days)

Figures and Tables -
Test 16

IgG/IgM (15 to 21 days)

IgG/IgM (22 to 35 days)

Figures and Tables -
Test 17

IgG/IgM (22 to 35 days)

IgG/IgM (over 35 days)

Figures and Tables -
Test 18

IgG/IgM (over 35 days)

IgA (all time points)

Figures and Tables -
Test 19

IgA (all time points)

IgA (1 to 7 days)

Figures and Tables -
Test 20

IgA (1 to 7 days)

IgA (8 to 14 days)

Figures and Tables -
Test 21

IgA (8 to 14 days)

IgA (15 to 21 days)

Figures and Tables -
Test 22

IgA (15 to 21 days)

IgA (22 to 35 days)

Figures and Tables -
Test 23

IgA (22 to 35 days)

IgA (over 35 days)

Figures and Tables -
Test 24

IgA (over 35 days)

Total antibodies (Ab) (all time points)

Figures and Tables -
Test 25

Total antibodies (Ab) (all time points)

Total antibodies (Ab) (1 to 7 days)

Figures and Tables -
Test 27

Total antibodies (Ab) (1 to 7 days)

Total antibodies (Ab) (8 to 14 days)

Figures and Tables -
Test 29

Total antibodies (Ab) (8 to 14 days)

Total antibodies (Ab) (15 to 21 days)

Figures and Tables -
Test 30

Total antibodies (Ab) (15 to 21 days)

Total antibodies (Ab) (21 to 35 days)

Figures and Tables -
Test 31

Total antibodies (Ab) (21 to 35 days)

Total antibodies (Ab) (over 35 days)

Figures and Tables -
Test 32

Total antibodies (Ab) (over 35 days)

IgA/IgG (all time points)

Figures and Tables -
Test 33

IgA/IgG (all time points)

IgA/IgG (1 to 7 days)

Figures and Tables -
Test 34

IgA/IgG (1 to 7 days)

IgA/IgG (8 to 14 days)

Figures and Tables -
Test 35

IgA/IgG (8 to 14 days)

IgA/IgG (15 to 21 days)

Figures and Tables -
Test 36

IgA/IgG (15 to 21 days)

IgA/IgG (22 to 35 days)

Figures and Tables -
Test 37

IgA/IgG (22 to 35 days)

IgA/IgM (all time points)

Figures and Tables -
Test 38

IgA/IgM (all time points)

IgG in PCR+ve (all time points)

Figures and Tables -
Test 39

IgG in PCR+ve (all time points)

IgG in PCR +ve (1 to 7 days)

Figures and Tables -
Test 40

IgG in PCR +ve (1 to 7 days)

IgG in PCR+ve (8 to 14 days)

Figures and Tables -
Test 41

IgG in PCR+ve (8 to 14 days)

IgG in PCR+ve (15 to 21 days)

Figures and Tables -
Test 42

IgG in PCR+ve (15 to 21 days)

IgG in PCR‐ve (all time points)

Figures and Tables -
Test 43

IgG in PCR‐ve (all time points)

IgG in PCR‐ve (1 to 7 days)

Figures and Tables -
Test 44

IgG in PCR‐ve (1 to 7 days)

IgG in PCR‐ve (8 to 14 days)

Figures and Tables -
Test 45

IgG in PCR‐ve (8 to 14 days)

IgG in PCR‐ve (15 to 21 days)

Figures and Tables -
Test 46

IgG in PCR‐ve (15 to 21 days)

IgM in PCR+ve (all time points)

Figures and Tables -
Test 47

IgM in PCR+ve (all time points)

IgM in PCR+ve (1 to 7 days)

Figures and Tables -
Test 48

IgM in PCR+ve (1 to 7 days)

IgM in PCR+ve (8 to 14 days)

Figures and Tables -
Test 49

IgM in PCR+ve (8 to 14 days)

IgM in PCR+ve (15 to 21 days)

Figures and Tables -
Test 50

IgM in PCR+ve (15 to 21 days)

IgM in PCR‐ve (all time points)

Figures and Tables -
Test 51

IgM in PCR‐ve (all time points)

IgM in PCR‐ve (1 to 7 days)

Figures and Tables -
Test 52

IgM in PCR‐ve (1 to 7 days)

IgM in PCR‐ve (8 to 14 days)

Figures and Tables -
Test 53

IgM in PCR‐ve (8 to 14 days)

IgM in PCR‐ve (15 to 21 days)

Figures and Tables -
Test 54

IgM in PCR‐ve (15 to 21 days)

IgG/IgM in PCR+ve (all time points)

Figures and Tables -
Test 55

IgG/IgM in PCR+ve (all time points)

IgG/IgM in PCR+ve (1 to 7 days)

Figures and Tables -
Test 56

IgG/IgM in PCR+ve (1 to 7 days)

IgG/IgM in PCR+ve (8 to 14 days)

Figures and Tables -
Test 57

IgG/IgM in PCR+ve (8 to 14 days)

IgG/IgM in PCR+ve (15 to 21 days)

Figures and Tables -
Test 58

IgG/IgM in PCR+ve (15 to 21 days)

IgG/IgM in PCR‐ve (all time points)

Figures and Tables -
Test 59

IgG/IgM in PCR‐ve (all time points)

IgG/IgM in PCR‐ve (1 to 7 days)

Figures and Tables -
Test 60

IgG/IgM in PCR‐ve (1 to 7 days)

IgG/IgM in PCR‐ve (8 to 14 days)

Figures and Tables -
Test 61

IgG/IgM in PCR‐ve (8 to 14 days)

IgG/IgM in PCR‐ve (15 to 21 days)

Figures and Tables -
Test 62

IgG/IgM in PCR‐ve (15 to 21 days)

IgG (moderate)

Figures and Tables -
Test 63

IgG (moderate)

IgG (severe)

Figures and Tables -
Test 64

IgG (severe)

IgG (critical)

Figures and Tables -
Test 65

IgG (critical)

IgM (moderate)

Figures and Tables -
Test 66

IgM (moderate)

IgM (severe)

Figures and Tables -
Test 67

IgM (severe)

IgM (critical)

Figures and Tables -
Test 68

IgM (critical)

RT‐PCR (all time points ‐ throat)

Figures and Tables -
Test 69

RT‐PCR (all time points ‐ throat)

RT‐PCR (1 to 7 days throat)

Figures and Tables -
Test 70

RT‐PCR (1 to 7 days throat)

RT‐PCR (8 to 14 days ‐ throat)

Figures and Tables -
Test 71

RT‐PCR (8 to 14 days ‐ throat)

RT‐PCR (15 to 21 days ‐ throat)

Figures and Tables -
Test 72

RT‐PCR (15 to 21 days ‐ throat)

RT‐PCR (all time points ‐ sputum)

Figures and Tables -
Test 73

RT‐PCR (all time points ‐ sputum)

RT‐PCR (1 to 7 days ‐ sputum)

Figures and Tables -
Test 74

RT‐PCR (1 to 7 days ‐ sputum)

RT‐PCR (8 to 14 days ‐ sputum)

Figures and Tables -
Test 75

RT‐PCR (8 to 14 days ‐ sputum)

RT‐PCR (15 to 21 days ‐ sputum)

Figures and Tables -
Test 76

RT‐PCR (15 to 21 days ‐ sputum)

Summary of findings 1. What is the diagnostic accuracy of antibody tests, for the diagnosis of current or prior SARS‐CoV‐2 infection?

Question

What is the diagnostic accuracy of antibody tests, for the diagnosis of current or prior SARS‐CoV‐2 infection?

Population

Adults or children suspected of

  • current SARS‐CoV‐2 infection

  • prior SARS‐CoV‐2 infection

or populations undergoing screening for SARS‐CoV‐2 infection, including

  • asymptomatic contacts of confirmed COVID‐19 cases

  • community screening

Index test

Any test for detecting antibodies to SARS‐CoV‐2, including:

  • laboratory‐based methods

    • ELISA

    • CLIA

    • other laboratory‐based methods

  • rapid tests; lateral flow assays, including

    • tests that can be used at point‐of‐care, such as CGIA

    • rapid diagnostic tests, such as FIA

Target condition

Detection of

  • current SARS‐CoV‐2 infection

  • prior SARS‐CoV‐2 infection

Reference standard

RT‐PCR alone, clinical diagnosis of COVID‐19 based on established guidelines or combinations of clinical features and for non‐COVID‐19 cases, the use of pre‐pandemic sources of samples for testing

Action

The current evidence‐base for antibody tests is inadequate to be clear about their utility (mainly because of small numbers of small studies for each test, few data available outside of acute hospital settings, and many issues in likely bias and applicability of the studies). The sensitivity of antibody tests is too low early in disease for use as a primary test of diagnosis, but they may have value for late diagnosis, for identifying previous infection, and for sero‐prevalence studies.

Limitations in the evidence

Risk of bias

Participant selection: high risk of bias in 48 studies (89%)

Application of index tests: high risk of bias in 14 studies (26%)

Reference standard: high risk of bias in 17 studies (31%)

Flow and timing: high risk of bias in 29 studies (54%)

Concerns about applicability of the evidence

Participants: high concerns in 44 studies (81%)

Index test: high concerns in 17 studies (31%)

Reference standard: high concerns in 33 studies (61%)

Findings

  • We included 54 studies evaluating 15,976 samples. 8256 samples were from COVID‐19 cases.

  • Data were not available for most antibody tests that have regulatory approval.

  • Most studies reported on detection of IgG, IgM, or IgG/IgM antibodies.

  • Test sensitivity was strongly related to time since onset of symptoms, with low sensitivity between 1 and 14 days, and sensitivity for IgG/IgM tests exceeding 90% between 15 and 35 days. Little evidence was available beyond 35 days.

  • Specificity was high (> 98%) for all types of antibody. There was some variation in sensitivity between test methods, with laboratory‐based methods appearing to outperform (point‐of‐care) tests using disposable devices.

  • Small sample sizes, low numbers of studies and concerns and bias and applicability hinder trustworthy comparisons being made between test brands.

Quantity of evidence

Number of studies

Total participants or samples

Total cases

54

15,976

8526

Sensitivity (95% CI)

Studies (TP/COVID cases)

Specificity (95%CI)

Studies (FP/non‐COVID cases)

Days 8‐14

Days 15‐21

Days 22‐35

All time points

IgG

66.5% (57.9 to 74.2)

88.2% (83.5 to 91.8)

80.3% (72.4 to 86.4)

99.1% (98.3% to 99.6%)

22 (766/1200)

22 (974/1110)

12 (417/502)

44 (159/6136)

IgM

58.4% (45.5 to 70.3)

75.4% (64.3 to 83.8)

68.1% (55.0 to 78.9)

98.7% (97.4% to 99.3%)

21 (724/1171)

21 (800/1074)

11 (378/507)

41 (183/6103)

IgG/IgM*

72.2% (63.5 to 79.5)

91.4% (87.0 to 94.4)

96.0% (90.6 to 98.3)

98.7% (97.2% to 99.4%)

9 (441/608)

9 (636/692)

5 (146/152)

23 (78/5761)

Numbers applied to a hypothetical cohort of 1000 patients, using summary data for IgG/IgM at days 15 to 21 as an exemplar (sensitivity 91.4% (87.0 to 94.4) and specificity 98.7% (97.2 to 99.4))

Prevalence of COVID‐19

TP (95% CI)

FP (95% CI)

FN (95% CI)

TN (95% CI)

2%

18 (17 to 20)

13 (6 to 27)

2 (1 to 3)

967 (953 to 974)

5%

46 (44 to 47)

12 (6 to 27)

4 (3 to 7)

938 (923 to 944)

10%

91 (87 to 94)

12 (5 to 25)

9 (6 to 13)

888 (875 to 895)

20%

183 (174 to 189)

10 (5 to 22)

17 (11 to 26)

790 (778 to 795)

50%

457 (435 to 472)

7 (3 to 14)

43 (28 to 65)

494 (486 to 497)

CGIA: colloidal gold immunoassays; CI: confidence interval; CLIA: chemiluminescence immunoassays; ELISA: enzyme‐linked immunosorbent assays; FIA: fluorescence‐labelled immunochromatographic assays; FN: false negative; FP: false positive; RT‐PCR: reverse transcription polymerase chain reaction; TN: true negative; TP: true positive; * Positive if either IgG or IgM positive.

Figures and Tables -
Summary of findings 1. What is the diagnostic accuracy of antibody tests, for the diagnosis of current or prior SARS‐CoV‐2 infection?
Table 1. Description of studies

Participants

Studies (percentage)

(n=54 studies)

Sample size

Median (IQR) 129.5 (57 to 347)

Min 10, max 3481

Number of COVID‐19 cases

Median (IQR) 62 (31 to 151)

Min 3, max 555

Setting

Hospital inpatient

44 (81%)

Hospital outpatient

1 (2%)

Hospital accident and emergency

2 (4%)

Community

2 (4%)

Mixed or unclear

5 (9%)

Patient group

Asymptomatic

0 (0%)

Asymptomatic and acute

1 (2%)

Acute

23 (43%)

Acute and convalescent

22 (41%)

Convalescent

2 (4%)

Mixed or unclear

6 (11%)

Study design

Recruitment structure

Single group, both COVID‐19 and non‐COVID‐19 cases

6 (11%)

Single group, only COVID‐19 cases

19 (35%)

Two or more groups with COVID‐19 and non‐COVID‐19 cases

29 (54%)

Reference standard for COVID‐19 cases

All RT‐PCR‐positive

32 (59%)

China CDC criteria including RT‐PCR‐negative patients

11 (20%)

WHO criteria including RT‐PCR‐negative patients

1 (2%)

Other criteria including RT‐PCR‐negative patients

3 (6%)

Other

2 (4%)

Mixed or unclear

5 (9%)

Reference standard for non‐COVID19

Pre‐pandemic healthy

4 (7%)

Pre‐pandemic other disease

3 (6%)

Pre‐pandemic healthy + other disease

4 (7%)

Current healthy (untested)

5 (9%)

Current other disease (untested)

1 (2%)

Current healthy + other disease (untested)

2 (4%)

Current healthy + other disease (RT‐PCR‐negative)

2 (4%)

COVID suspects, single RT‐PCR‐negative

8 (15%)

COVID suspects, two or more RT‐PCR–negative results

3 (6%)

Mixed/other

3 (6%)

Tests

Number of tests per study

1

40 (74%)

2

8 (15%)

3‐5

4 (8%)

6‐10

2 (2%)

Test technology (n = 89)

CGIA

23 (26%)

CLIA

20 (22%)

ELISA

28 (31%)

FIA

2 (2%)

IIFT

1 (1%)

LFA (no details)

10 (11%)

LIPS

4 (4%)

S‐flow

1 (1%)

Test brand (n = 89)

Withheld

13 (%)

Acro Biotech ‐ IgG/IgM

1 (1%)

Artron Laboratories IgM/IgG

1 (1%)

Autobio Diagnostics IgM/IgG

1 (1%)

Beijing Beier Bioengineering CGIA

1 (1%)

Beijing Beier Bioengineering CLIA

1 (1%)

Beijing Beier Bioengineering ELISA

1 (1%)

Beijing Diagreat

1 (1%)

Beijing Hotgen CGIA

1 (1%)

Beijing Hotgen ELISA

2 (3%)

Beijing Wantai CGIA

1 (1%)

Beijing Wantai ELISA

3 (3%)

Bioscience Co (Chongqing)

3 (3%)

CTK Biotech OnSite IgG/IgM

1 (1%)

Darui Biotech

1 (1%)

Dynamiker Biotechnology IgG/IgM

1 (1%)

EUROIMMUN

3 (3%)

EUROIMMUN Anti‐SARS‐Cov

1 (1%)

EUROIMMUN Beta

1 (1%)

Hangzhou Alltest ‐ IgG/IgM

3 (3%)

Innovita Biological ‐ Ab test (IgM/IgG)

2 (3%)

Jiangsu Medomics IgG‐IgM

1 (1%)

Shenzhen YHLO

7 (8%)

Snibe Diagnostic ‐ MAGLUMI

2 (3%)

Vivachek ‐ VivaDiag IgM/IgG

3 (3%)

Xiamen InnodDx Biotech

1 (1%)

Zhuhai Livzon CGIA

2 (3%)

Zhuhai Livzon ELISA

5 (6%)

In‐house, S‐based ELISA

1 (1%)

In‐house, S‐based LIPS

1 (1%)

In‐house, rN‐based ELISA

1 (1%)

In‐house, rS‐based ELISA

1 (1%)

In‐house CGIA

2 (2%)

In‐house CLIA

5 (6%)

In‐house ELISA

6 (7%)

In‐house FIA

1 (1%)

In‐house S‐flow

1 (1%)

In‐house ‐ N‐based ELISA

1 (1%)

In‐house ‐ N‐based LIPS

2 (2%)

In‐house ‐ S1‐based LIPS

1 (1%)

In‐house ‐ tri‐S‐based ELISA

1 (1%)

In‐house Anti‐SARS‐Cov ELISA

1 (1%)

Ab: antibody; CDC: Center for Disease Control and Prevention; CGIA: colloidal gold immunoassay; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IQR: interquartile range; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay; LIPS: luciferase immunoprecipitation system; max: maximum; min: minimum; N‐based: nucleocapsid protein; RT‐PCR: reverse transcription polymerase chain reaction; S‐based: spike protein; S‐flow: flow‐cytometry assay; WHO: World Health Organization

Figures and Tables -
Table 1. Description of studies
Table 2. Test sensitivity by time since onset of symptoms

Days 1‐7

Days 8‐14

Days 15‐21

Days 22‐35

Days > 35

Comparison

Test groups [studies] (true positives/COVID cases)

Sensitivity (95% CI)

IgG

33 [23] (165/568)

34 [22] (766/1200)

34 [22] (974/1110)

20 [12] (417/502)

11 [4] (213/252)

29.7% (22.1 to 38.6)

66.5% (57.9 to 74.2)

88.2% (83.5 to 91.8)

80.3% (72.4 to 86.4)

86.7% (79.6 to 91.7)

P < 0.00005

IgM

34 [24] (207/608)

32 [21] (724/1171)

32 [21] (800/1074)

19 [11] (378/507)

11 [4] 118/215

23.2% (14.9 to 34.2)

58.4% (45.5 to 70.3)

75.4% (64.3 to 83.8)

68.1% (55.0 to 78.9)

53.9% (38.4 to 68.6)

P < 0.00005

IgA

4 [4] (54/100)

3 [3] (38/53)

3 [3] (66/68)

2 [2] (81/82)

1 [1] (23/23)

28.4% (0.9 to 94.3)

78.1% (9.5 to 99.2)

98.7% (39.0 to 100)

98.7% (91.9 to 99.8)

100% (85.2 to 100)

*

Total antibodies

5 [4] (62/144)

6 [5] (220/247)

6 [5] (174/176)

4 [3] (11/19)

2 [1] (15/28)

24.5% (9.5 to 50.0)

84.0% (64.1 to 93.9)

98.1% (90.1 to 99.6)

69.5% (34.8 to 90.7)

79.0% (49.8 to 93.4)

P < 0.00005

IgG/IgM

17 [9] (81/259)

21 [9] (441/608)

21 [9] (636/692)

16 [5] (146/152)

9 [2] (122/153)

30.1% (21.4 to 40.7)

72.2% (63.5 to 79.5)

91.4% (87.0 to 94.4)

96.0% (90.6 to 98.3)

77.7% (66.0 to 86.2)

P < 0.00005

IgA/IgG

1 [1] (0/12)

1 [1] (5/10)

1 [1] (7/8)

1 [1] (1/1)

0 [0]

0% (0 to 26.5)

50.0% (18.7 to 81.3)

87.5% (47.3 to 99.6)

100% (2.5 to 100)

*

IgA/IgM

0 [0]

0 [0]

0 [0]

0 [0]

0 [0]

CI: confidence interval; * inadequate data to make a formal statistical comparison

Figures and Tables -
Table 2. Test sensitivity by time since onset of symptoms
Table 3. Specificity and impact of reference standard for non‐COVID cases

Overall specificitya

COVID suspects deemed negative

Current healthy or other disease

Pre‐pandemic

Comparison of control groups

Test groups [studies] (false positives/non‐COVID cases)

Specificity (95% CI)

IgG

62 [44] (159/6136)

6 [6] (10/396)

14 [10] (60/2614)

19 [10] (88/2633)

99.1% (98.3% to 99.6%)

98.0% (91.0% to 99.6%)

99.2% (97.6% to 99.8%)

99.2% (97.8% to 99.7%)

P = 0.56

IgM

59 [41] (183/6103)

5 [5] (12/384)

14 [10] (89/3069)

17 [9] (38/2075)

98.7% (97.4% to 99.3%)

98.1% (89.9% to 99.7%)

98.6% (96.0% to 99.5%)

99.3% (98.0% to 99.8%)

P = 0.50

IgG/IgM

34 [23] (78/5761)

7 [7] (33/454)

7 [5] (20/506)

18 [6] (22/1104)

No formal comparison possible

98.7% (97.2% to 99.4%)

92.8% (89.7% to 95.0%)

99.9% (65.2% to 100%)

98.7% (96.6% to 99.5%)

Total antibodies

16 [10] (41/3585)

99.2% (98.3% to 99.6%)

IgA

4 [4] (10/663)

98.5% (97.2% to 99.2%)

IgA/IgGb

2 [2] (1/528)

99.8% (98.9% to 100%)

IgA/IgMb

1 [1] (1/483)

99.8% (99.2% to 100%)

CI: confidence interval

aIncludes studies that are categorised as mixed/other not included in the subgroups.
bConfidence intervals computed using binomial exact on totals.

Figures and Tables -
Table 3. Specificity and impact of reference standard for non‐COVID cases
Table 4. Reported cross‐reactivity with SARS‐CoV‐2 antigens

Study

Test(s) evaluated

What the study says about cross‐reactivity

Cai 2020

In‐house CLIA

Reported no cross‐reactivity in 167 sera from patients with infection with other pathogens (influenza A virus (25), respiratory syncytial virus (7), parainfluenza virus (8), influenza B virus (5), adenovirus (6), Klebsiella pneumoniae (8), Streptococcus pneumoniae (3), mycoplasma (5), Acinetobacter baumannii (10), Candida albicans (2), Staphylococcus aureus (3), Mycobacterium tuberculosis (4), hepatitis B virus (33), hepatitis C virus (22), syphilis (23) and saccharomycopsis (3)).

Freeman 2020

In‐house ELISA

Reported cross‐reactivity to SARS‐CoV‐2 spike protein in sera from patients with SARS‐1 and MERS‐CoV, and no cross‐reactivity with NL63, OC43, HKU1, 229E

Guo 2020a

In‐house ELISA

Reported Western Blot cross‐reactivity analysis in plasma samples positive for human CoV‐229E, ‐NL63, ‐OC43, ‐HKU1, and SARS‐CoV. Strong cross‐reactivity was observed only for SARS‐CoV.

Infantino 2020

Shenzhen YHLO CLIA

Observed no cross‐reactivity in sample from blood donors from the COVID‐19 era (winter 2019) but positive results in two samples from people with CMV infections and 2 with rheumatic disease.

Lassauniere 2020 [A]

[A] Beijing Wantai ELISA

[B] EUROIMMUN IgG ELISA

[C] EUROIMMUN IgA ELISA

[D] Dynamiker Biotechnology LFA

[E] CTK Biotech ‐ OnSite LFA

[F] Autobio Diagnostics LFA

[G] Artron Laboratories LFA

[H] Acro Biotech LFA

[I] Hangzhou Alltest LFA

Included sera from patients with acute viral respiratory tract infections caused by other coronaviruses (n = 5) or non‐coronaviruses (n = 45), and sera from patients positive for dengue virus (n = 9), CMV (n = 2) and Epstein Barr virus (n = 10). Cross reaction was observed for the EUROMIMMUN IgA ELISA (> 1 respiratory virus present, adenovirus, dengue virus) and for the EUROMIMMUN IgG ELISA (coronavirus HKU1 and adenovirus). Some cross‐reactivity also observed for CGIA tests. Study authors suggest related to antigen target and ELISA format.

Ma 2020a

In‐house CLIA

Limited detail but suggests limited cross‐reaction

Wang 2020a [A]

A. Beijing Hotgen IgM CGIA

B. Beijing Hotgen IgM ELISA

Demonstrated considerable cross‐reaction with rheumatoid factor IgM (22/36 false positive results). Other pathogens included influenza A virus (n = 5), influenza B virus (n = 5), Mycoplasma pneumoniae (n = 5), Legionella pneumophila (n = 5), HIV infection (n = 6), hypertension (n = 5) and diabetes mellitus (n = 5)

Zhang 2020b

Shenzhen YHLO CLIA

Observed false positive results in influenza A and B (2 each), adenovirus (n = 4) and Mycoplasma pneumoniae (n = 17).

Zhang 2020d

In‐house CGIA (co‐author Beijing Hotgen)

Appears to report a separate cross‐reactivity study for influenza A, influenza B, respiratory syncytial virus, Mycoplasma pneumoniae and Chlamydia pneumoniae. No cross reactions were observed.

CGIA: colloidal gold immunoassay; CLIA: Chemiluminescence immunoassay; CMV: cytomegalovirus; ELISA: enzyme‐linked immunosorbent assay; LFA: lateral flow assay; MERS: Middle East respiratory syndrome; SARS: severe acute respiratory syndrome

Figures and Tables -
Table 4. Reported cross‐reactivity with SARS‐CoV‐2 antigens
Table 5. Investigation of impact of reference standard on sensitivity

RT‐PCR‐positive COVID‐19 cases

RT‐PCR‐negative COVID‐19 cases

Comparison

Test groups [studies] (true positives/COVID cases)

Sensitivity (95% CI)a

IgG

26 [15] (1555/2280)

8 [8] (925/1300)

87.9% (82.7 to 91.7)

91.2% (83.9 to 95.4)

P = 0.36

IgM

23 [13] (1368/2166)

10 [9] (792/1292)

70.8% (56.3 to 82.0)

87.5% (73.7 to 94.6)

P = 0.06

IgG/IgM

17 [6] (966/1278)

4 [4] (400/499)

90.6% (86.6 to 93.5)

93.6% (88.9 to 96.4)

P = 0.22

CI: confidence interval; RT‐PCR: reverse transcription polymerase chain reaction

aWe obtained sensitivity estimates from a model of all data stratified by week, estimating the average difference in sensitivity across follow‐up. The figures quoted correspond to the week 3 strata (15‐21 days) in the model.

Figures and Tables -
Table 5. Investigation of impact of reference standard on sensitivity
Table 6. Studies reporting sensitivity in both RT‐PCR‐positive and RT‐PCR‐negative subgroups

RT‐PCR‐positive COVID‐19 cases

RT‐PCR‐negative COVID‐19 cases

Test groups [studies]

(True positives/COVID‐19 cases)

Sensitivity (95% CI)

Test groups [studies]

(True positives/COVID‐19 cases)

Sensitivity (95% CI)

IgG

Days 1‐7b

2 [2] (1/28)

2 [2] (8/13)

Days 8‐14b

2 [2] (21/33)

3 [3] (25/30)

Days 15‐21b

2 [2] (39/40)

3 [3] (64/72)

Pooleda (stratified by time)

72.6% (46.2% to 89.1%)

84.0% (64.4% to 93.9%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative groups: P = 0.18

IgM

Days 1‐7b

2 [2] (3/28)

2 [2] (4/13)

Days 8‐14b

2 [2] (25/33)

3 [3] (11/30)

Days 15‐21b

2 [2] (8/16)

3 [3] (31/72)

Pooleda (stratified by time)

64.6% (49.7% to 77.1%)

49.0% (34.2% to 63.9%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative group: P = 0.07

IgG/IgM

Days 1‐7b

2 [2] (8/36)

2 [2] (4/17)

Days 8‐14b

2 [2] (37/53)

3 [3] (29/40)

Days 15‐21b

2 [2] (141/150)

3 [3] (104/113)

Pooleda (stratified by time)

71.9% (58.7% to 82.2%)

71.1% (57.0% to 82.0%)

Test for difference in sensitivity between RT‐PCR‐positive and RT‐PCR‐negative group: P = 0.90

CI: confidence interval; RT‐PCR: reverse transcription polymerase chain reaction

aThe sensitivity estimates are produced from a model that combines all data from both subgroups and time‐groups, stratifying by time‐group. The estimate corresponds to sensitivity in Days 15‐21.
bRT‐PCR‐positive data have only been included here when the study includes a RT‐PCR‐negative subgroup as well.

Figures and Tables -
Table 6. Studies reporting sensitivity in both RT‐PCR‐positive and RT‐PCR‐negative subgroups
Table 7. Sensitivity and specificity by test technology

Test method

Test method

CGIA

CLIA

ELISA

LFA

Comparison

IgG

Test groups [studies]

(True positives/COVID cases)

6 [5] (268/397)

10 [10] (1112/1432)

12 [11] (1014/1552)

7 [1] (133/238)

Sensitivity (95% CI)a

87.3% (77.0 to 93.4)

94.6% (90.7 to 97.0)

85.8% (78.0 to 91.1)

76.0% (61.0 to 86.5)

P = 0.004

Test groups [studies]

(True negatives/non‐COVID cases)

11 [11] (409/415)

12 [12] (318/322)

18 [16] (2003/2102)

6 [1] (354/360)

Specificity (95% CI)a

99.5% (96.5 to 99.9)

99.0% (91.6 to 99.9)

98.8% (96.5 to 99.6)

99.0% (95.3 to 99.8)

P = 0.85

IgM

Test groups [studies]

(True positives/COVID cases)

7 [6] (109/411)

10 [10] (884/1355)

12 [11] (1083/1568)

7 [1] (78/228)

Sensitivity (95% CI)a

69.5% (44.3 to 86.7)

80.9% (63.8 to 91.0)

84.5% (70.7 to 92.5)

51.4% (26.5 to 75.6)

P = 0.11

Test groups [studies]

(True negatives/non‐COVID cases)

12 [11] (455/487)

13 [13] (609/621)

14 [12] (1674/1710)

6 [1] (357/360)

Specificity (95% CI)a

97.3 (90.0 to 99.3)

98.5 (92.3 to 99.7)

99.1 (97.2 to 99.7)

99.6 (97.3 to 99.9)

P = 0.40

IgG/IgM

Test groups [studies]

(True positives/COVID cases)

4 [3] (232/316)

3 [3] (344/420)

5 [4] (595/770)

11 [2] (255/358)

Sensitivity (95% CI)a

90.7% (82.7 to 95.2)

97.5% (94.0 to 99.0)

90.7% (83.3 to 95.0)

88.6% (82.0 to 93.0)

P = 0.02

Test groups [studies]

(True negatives/non‐COVID cases)

11 [11] (330/353)

5 [4] (230/244)

5 [4] (387/391)

13 [3] (3797/3827)

Specificity (95% CI)a

96.0 (90.1 to 98.5)

94.1 (82.7 to 98.2)

99.4 (97.4 to 99.9)

98.2 (96.3 to 99.1)

P = 0.05

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; LFA: lateral flow assay (no further detail)

aWe obtained sensitivity estimates from a model of all data stratified by week, estimating the average difference in sensitivity across follow‐up. The figures quoted correspond to the Week 3 (15‐21 days) strata in the model.

Figures and Tables -
Table 7. Sensitivity and specificity by test technology
Table 8. Sensitivity and specificity by test brand (IgG)

Test namea

Test

method

IgG sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitivity (95% CI)

IgG specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Beijing Beier Bioengineering

CGIA

1 (2/10)

1 (6/13)

1 (11/14)

20.0% (2.5 to 55.6)

46.2% (19.2 to 74.9)

78.6% (49.2 to 95.3)

Beijing Beier Bioengineering

CLIA

1 (4/10)

1 (6/13)

1 (9/14)

40.0% (12.2 to 73.8)

46.2% (19.2 to 74.9)

64.3% (35.1 to 87.2)

Beijing Beier Bioengineering

ELISA

1 (4/10)

1 (8/13)

1 (12/14)

40.0% (12.2 to 73.8)

61.5% (31.6 to 86.1)

85.7% (57.2 to 98.2)

Beijing Hotgen

ELISA

1 (9/22)

1 (60/92)

1 (51/55)

1 (39/45)

2 (22/172)

40.9% (20.7 to 63.6)

65.2% (54.6 to 74.9)

92.7% (82.4 to 98.0)

86.7% (73.2 to 94.9)

87.2% (81.3 to 91.8)

Beijing Wantai

ELISA

2 (31/133)

2 (130/210)

2 (127/149)

2 (2/297)

23.3% (16.4 to 31.4)

61.9% (55.0 to 68.5)

85.2% (78.5 to 90.5)

99.3% (97.6 to 99.9)

Beijing Wantai

CGIA

1 (1/209)

99.5% (97.4 to 100)

Bioscience Co (Chongqing)

CLIA

2 (43/92)

2 (129/212)

2 (208/244)

2 (98/164)

1 (75/76)

46.7% (36.3 to 57.4)

60.8% (53.9 to 67.5)

85.2% ( 80.2 to 89.4)

59.8% (51.8 to 67.3)

98.6% (92.9 to 100)

Darui Biotech

ELISA

1 (0/64)

100% (94.4 to 100)

EUROIMMUN

ELISA

1 (2/13)

2 (13/25)

2 (14/15)

2 (98/164)

2 (3/82)

15.4% (1.9 to 45.4)

52.0% (31.3 to 72.2)

93.3% (68.1 to 99.8)

59.8% (51.8 to 67.3)

96.3% (89.7 to 99.2)

EUROIMMUN Anti‐SARS‐Cov

IIFT

1 (1/4)

1 (3/5)

1 (3/3)

1 (1/1)

1 (0/10)

25.0% (0.6 to 80.6)

60.0% (14.7 to 94.7)

100% (29.2 to 100)

100% (2.5 to 100)

100% (69.2 to 100)

EUROIMMUN Beta

ELISA

1 (0/12)

1 (3/10)

1 (7/8)

1 (1/1)

1 (0/45)

0% (0 to 26.5)

30%.0% (14.7 to 94.7)

87.5% (47.3 to 99.7)

100% (2.5 to 100)

100% (92.1 to 100)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (21/42)

2 (57/68)

2 (0/45)

12.5% (0.3 to 52.7)

50.0% (34.2 to 65.8)

83.8% (72.9 to 91.6)

100% (92.1 to 100)

Innovita Biological ‐ Ab test (IgM/IgG)

CGIA

1 (7/13)

1 (7/8)

1 (21/23)

53.8% (25.1 to 80.8)

87.5% (47.3 to 99.7)

91.3% (72.0 to 98.9)

Shenzhen YHLO

CLIA

2 (2/8)

2 (28/29)

2 (25/26)

2 (64/64)

1 (7/7)

7 (4/322)

25.0% (3.2 to 65.1)

96.6% (82.2 to 99.9)

96.2% (80.4 to 99.9)

100% (94.4 to 100)

100% (59.0 to 100)

98.8% (96.9 to 99.7)

Snibe Diagnostic ‐ MAGLUMI

CLIA

2 (11/40)

2 (35/48)

25/25

27.5% (14.6 to 43.9)

72.9% (58.2 to 84.7)

100.0% (86.3 to 100)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

2 (0/42)

100% (91.6 to 100)

Zhuhai Livzon

CGIA

1 (5/36)

1 (20/34)

1 (35/38)

2 (0/35)

13.9% (4.7 to 29.5)

58.8% (40.7 to 75.4)

92.1% (78.6 to 98.3)

100% ( 90.0 to 100)

Zhuhai Livzon

ELISA

4 (17/80)

3 (163/288)

3 (197/223)

2 (91/104)

5 (5/351)

21.3% (12.9 to 31.8)

56.6% (50.7 to 62.4)

88.3% (83.4 to 92.2)

87.5% (79.6 to 93.2)

98.6% (96.7 to 99.5)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Figures and Tables -
Table 8. Sensitivity and specificity by test brand (IgG)
Table 9. Sensitivity and specificity by test brand (IgM)

Test namea

Test method

IgM sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitiivity (95% CI)

IgM specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Artron Laboratories IgM/IgG

CGIA

1 (5/7)

1 (12/15)

1 (8/8)

71.4% (29.0 to 96.3)

80.0% (51.9 to 95.7)

100% (63.1 to 100)

Autobio Diagnostics IgM/IgG

CGIA

1 (6/7)

1 (14/15)

1(8/8)

85.7% (42.1 to 99.6)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

Beijing Hotgen

ELISA

1 (10/22)

1 (72/92)

1 (72/92)

1 (41/45)

1 (0/100)

45.5% (24.4 to 67.8)

78.3% (68.4 to 86.2)

78.3% (68.4 to 86.2)

91.1% (78.8 to 97.5)

100% (96.4 to 100)

Beijing Hotgen

CGIA

1 (22/72)

69.4% (57.5 to 79.8)

Beijing Wantai

ELISA

1 (3/513)

99.4% (98.3 to 99.9)

Beijing Wantai

CGIA

1 (4/209)

98.1% (95.2 to 99.5)

Bioscience Co (Chongqing)

CLIA

1 (34/67)

1 (34/67)

1 (131/134)

1 (13/13)

50.7% (38.2 to 63.2)

50.7% (38.2 to 63.2)

97.8% (93.6 to 99.5)

100% (75.3 to 100)

CTK Biotech OnSite IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

Darui Biotech

ELISA

1 (14/64)

78.1% (66.0 to 87.5)

Dynamiker Biotechnology IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

EUROIMMUN

ELISA

1 (76/82)

92.7% (84.8 to 97.3)

EUROIMMUN Anti‐SARS‐Cov

IIFT

1 (1/10)

90.0% (55.5 to 99.7)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (23/42)

2 (58/68)

2 (0/45)

12.5% (0.3 to 52.7)

54.8% (38.7 to 70.2)

85.3% (74.6 to 92.7)

100% (92.1 to 100)

Shenzhen YHLO

CLIA

7 (10/321)

96.9% (94.3 to 98.5)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

2 (1/42)

97.6% (87.4 to 99.9)

Xiamen InnodDx Biotech

CLIA

1 (2/300)

99.3% (97.6 to 99.9)

Zhuhai Livzon

CGIA

1 (7/36)

1 (31/34)

1 (35/38)

2 (0/35)

19.4% (8.2 to 36.0)

91.2% (76.3 to 98.1)

92.1% (78.6 to 98.3)

100% ( 90.0 to 100)

Zhuhai Livzon

ELISA

3 (14/66)

2 (150/202)

2 (159/166)

1 (43/45)

5 (3/351)

21.2% (12.1 to 33.0)

74.3% (67.7 to 80.1)

95.8% (91.5 to 98.3)

95.6% (84.9 to 99.5)

99.1% (97.5 to 99.8)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Figures and Tables -
Table 9. Sensitivity and specificity by test brand (IgM)
Table 10. Sensitivity and specificity by test brand (IgG/IgM)

Test namea

Test method

IgG/IgM sensitivity by time since onset of symptoms

Studies (true positives/COVID‐19 cases)
Sensitiivity (95% CI)

IgG/IgM specificity

Studies (false positives/COVID‐19 cases)
Specificity (95% CI)

1‐7 days

8‐14 days

15‐21 days

22‐35 days

> 35 days

Acro Biotech ‐ IgG/IgM

CGIA

1 (3/15)

80.0% (51.9 to 95.7)

Artron Laboratories IgM/IgG

CGIA

1 (5/7)

1 (12/15)

1 (8/8)

1 (0/17)

71.4% (29.0 to 96.3)

80.0% (51.9 to 95.7)

100% (63.1 to 100)

100% (80.5% to 100)

Autobio Diagnostics IgM/IgG

CGIA

1 (6/7)

1 (14/15)

1(8/8)

1 (0/32)

85.7% (42.1 to 99.6)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Beijing Hotgen

ELISA

1 (10/22)

1 (72/92)

1 (72/92)

1 (41/45)

1 (0/100)

45.5% (24.4 to 67.8)

78.3% (68.4 to 86.2)

78.3% (68.4 to 86.2)

91.1% (78.8 to 97.5)

100% (96.4 to 100)

Bioscience Co (Chongqing)

CLIA

1 (34/67)

1 (34/67)

1 (131/134)

1 (13/13)

2 (7/148)

50.7% (38.2 to 63.2)

50.7% (38.2 to 63.2)

97.8% (93.6 to 99.5)

100% (75.3 to 100)

95.3% (90.5 to 98.1)

CTK Biotech OnSite IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

1 (0/32)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Dynamiker Biotechnology IgG/IgM

CGIA

1 (5/7)

1 (14/15)

1 (8/8)

1 (0/32)

71.4% (29.0 to 96.3)

93.3% (68.1 to 99.8)

100% (63.1 to 100)

100% (89.1 to 100)

Hangzhou Alltest ‐ IgG/IgM

CGIA

1 (1/8)

2 (23/42)

2 (58/68)

3 (2/60)

12.5% (0.3 to 52.7)

54.8% (38.7 to 70.2)

85.3% (74.6 to 92.7)

96.7% (88.5 to 99.6)

Shenzhen YHLO

CLIA

2 (7/96)

92.7% (85.6 to 97.0)

Vivachek ‐ VivaDiag IgM/IgG

CGIA

3 (14/162)

91.4% (85.9 to 95.2)

Zhuhai Livzon

CGIA

1 (7/36)

1 (31/34)

1 (35/38)

2 (0/35)

19.4% (8.2 to 36.0)

91.2% (76.3 to 98.1)

92.1% (78.6 to 98.3)

100% (90.0 to 100)

Zhuhai Livzon

ELISA

3 (14/66)

2 (150/202)

2 (159/166)

1 (43/45)

4 (4/291)

21.2% (12.1 to 33.0)

74.3% (67.7 to 80.1)

95.8% (91.5 to 98.3)

95.6% (84.9 to 99.5)

98.6% (96.5 to 99.6)

CGIA: colloidal gold immunoassay; CI: confidence interval; CLIA: chemiluminescence immunoassay; ELISA: enzyme‐linked immunosorbent assay; FIA: fluorescence immunoassay; IIFT: indirect immunofluorescence assay; LFA: lateral flow assay

aSee Appendix 12 for details of manufacturer product codes, where available.

Figures and Tables -
Table 10. Sensitivity and specificity by test brand (IgG/IgM)
Table Tests. Data tables by test

Test

No. of studies

No. of participants

1 IgG (all time points) Show forest plot

62

11748

2 IgG (1 to 7 days) Show forest plot

33

585

3 IgG (8 to 14 days) Show forest plot

34

1220

4 IgG (15 to 21 days) Show forest plot

32

1108

5 IgG (22 to 35 days) Show forest plot

20

495

6 IgG (over 35 days) Show forest plot

12

259

7 IgM (all time points) Show forest plot

58

11436

8 IgM (8 to 14 days) Show forest plot

31

1166

9 IgM (1 to 7 days) Show forest plot

34

658

10 IgM (15 to 21 days) Show forest plot

30

1057

11 IgM (22 to 35 days) Show forest plot

18

492

12 IgM (over 35 days) Show forest plot

12

222

13 IgG/IgM (all time points) Show forest plot

44

9496

14 IgG/IgM (1 to 7 days) Show forest plot

17

259

15 IgG/IgM (8 to 14 days) Show forest plot

21

608

16 IgG/IgM (15 to 21 days) Show forest plot

21

692

17 IgG/IgM (22 to 35 days) Show forest plot

16

152

18 IgG/IgM (over 35 days) Show forest plot

9

153

19 IgA (all time points) Show forest plot

5

1278

20 IgA (1 to 7 days) Show forest plot

4

100

21 IgA (8 to 14 days) Show forest plot

4

65

22 IgA (15 to 21 days) Show forest plot

3

78

23 IgA (22 to 35 days) Show forest plot

3

90

24 IgA (over 35 days) Show forest plot

1

23

25 Total antibodies (Ab) (all time points) Show forest plot

17

5339

27 Total antibodies (Ab) (1 to 7 days) Show forest plot

5

144

29 Total antibodies (Ab) (8 to 14 days) Show forest plot

6

247

30 Total antibodies (Ab) (15 to 21 days) Show forest plot

6

176

31 Total antibodies (Ab) (21 to 35 days) Show forest plot

4

19

32 Total antibodies (Ab) (over 35 days) Show forest plot

2

28

33 IgA/IgG (all time points) Show forest plot

2

775

34 IgA/IgG (1 to 7 days) Show forest plot

1

12

35 IgA/IgG (8 to 14 days) Show forest plot

1

10

36 IgA/IgG (15 to 21 days) Show forest plot

1

8

37 IgA/IgG (22 to 35 days) Show forest plot

1

1

38 IgA/IgM (all time points) Show forest plot

1

699

39 IgG in PCR+ve (all time points) Show forest plot

4

558

40 IgG in PCR +ve (1 to 7 days) Show forest plot

2

28

41 IgG in PCR+ve (8 to 14 days) Show forest plot

2

33

42 IgG in PCR+ve (15 to 21 days) Show forest plot

2

40

43 IgG in PCR‐ve (all time points) Show forest plot

6

252

44 IgG in PCR‐ve (1 to 7 days) Show forest plot

2

13

45 IgG in PCR‐ve (8 to 14 days) Show forest plot

3

30

46 IgG in PCR‐ve (15 to 21 days) Show forest plot

3

72

47 IgM in PCR+ve (all time points) Show forest plot

6

740

48 IgM in PCR+ve (1 to 7 days) Show forest plot

2

28

49 IgM in PCR+ve (8 to 14 days) Show forest plot

2

33

50 IgM in PCR+ve (15 to 21 days) Show forest plot

2

16

51 IgM in PCR‐ve (all time points) Show forest plot

8

352

52 IgM in PCR‐ve (1 to 7 days) Show forest plot

2

13

53 IgM in PCR‐ve (8 to 14 days) Show forest plot

3

30

54 IgM in PCR‐ve (15 to 21 days) Show forest plot

3

72

55 IgG/IgM in PCR+ve (all time points) Show forest plot

2

177

56 IgG/IgM in PCR+ve (1 to 7 days) Show forest plot

2

36

57 IgG/IgM in PCR+ve (8 to 14 days) Show forest plot

2

53

58 IgG/IgM in PCR+ve (15 to 21 days) Show forest plot

2

150

59 IgG/IgM in PCR‐ve (all time points) Show forest plot

4

215

60 IgG/IgM in PCR‐ve (1 to 7 days) Show forest plot

2

17

61 IgG/IgM in PCR‐ve (8 to 14 days) Show forest plot

3

40

62 IgG/IgM in PCR‐ve (15 to 21 days) Show forest plot

3

113

63 IgG (moderate) Show forest plot

1

44

64 IgG (severe) Show forest plot

1

52

65 IgG (critical) Show forest plot

1

37

66 IgM (moderate) Show forest plot

1

44

67 IgM (severe) Show forest plot

1

52

68 IgM (critical) Show forest plot

1

37

69 RT‐PCR (all time points ‐ throat) Show forest plot

2

276

70 RT‐PCR (1 to 7 days throat) Show forest plot

2

67

71 RT‐PCR (8 to 14 days ‐ throat) Show forest plot

2

142

72 RT‐PCR (15 to 21 days ‐ throat) Show forest plot

2

73

73 RT‐PCR (all time points ‐ sputum) Show forest plot

1

53

74 RT‐PCR (1 to 7 days ‐ sputum) Show forest plot

1

13

75 RT‐PCR (8 to 14 days ‐ sputum) Show forest plot

1

8

76 RT‐PCR (15 to 21 days ‐ sputum) Show forest plot

1

23

Figures and Tables -
Table Tests. Data tables by test