Spanish COVID-19 seroprevalence study - Printable Version
+- The CardBoard (https://thecardboard.org/board)
+-- Forum: Emergency (https://thecardboard.org/board/forum-11.html)
+--- Forum: Covid-19 (https://thecardboard.org/board/forum-12.html)
+--- Thread: Spanish COVID-19 seroprevalence study (/thread-19710.html)
+- The CardBoard (https://thecardboard.org/board)
+-- Forum: Emergency (https://thecardboard.org/board/forum-11.html)
+--- Forum: Covid-19 (https://thecardboard.org/board/forum-12.html)
+--- Thread: Spanish COVID-19 seroprevalence study (/thread-19710.html)
Spanish COVID-19 seroprevalence study - BostonCard - 05-14-2020
https://www.lamoncloa.gob.es/serviciosdeprensa/notasprensa/sanidad14/Documents/2020/130520-ENE-COVID_Informe1.pdf
The results are in Spanish, but I will summarize:
Random sample of Spanish 102,000 individuals in 36,000 households of which 67,000 participated (5000 were not eligible, 16,000 couldn't be contacted, and 14,500 refused) and about 60,000 had valid tests (there were about 6,000 where the test wasn't run for some reason). The study was fairly representative of the age and sex distribution of Spain (see the figure on page 5).
The overall prevalence of COVID-19 antibodies (IgG) was 5% (95% CI of 4.7 - 5.4%). It was 6.4% in large cities (>100,000). The prevalence in Madrid was 11
Among the people who had already been diagnosed with COVID-19 by PCR, the prevalence was 87% (not sure if that is poor sensitivity of the serology test, or poor history by the study participants).
The test they used claims a sensitivity for IgG of 97% and a specificity of 100%.
Taking the headline number of 5% prevalence, and a population in Spain of 46,750,000, you get 2.34 million infected. With 27,300 deaths (and counting), you get a crude IFR of 1.16%.
The Diamond Princess, which is the one other closed system we have a good handle on, had 13 deaths out of 712 cases, for a crude IFR of 1.8% (95% CI 0.98% to 3.1%), but that was an older population, so if you adjust for age you would get an adjusted IFR of 0.5%.
As this article elegantly states:
https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2766121
not the flu.
BC
RE: Spanish COVID-19 seroprevalence study - M T - 05-14-2020
It appears the test was "Orient Gene IgM/IgG, de la empresa Zhejiang Orient Gene Biotech " .
This particular test does not seem to appear in the 3 evaluation surveys that I've seen (1, 2, 3)
The manufacturer's claims of 87.9% sensitivity for IgM and 97.2% for IgG and 100% specificity for IgG and IgM.
However, those numbers were based on 87/99 and 35/36 tests for sensitivity; 14/14 and 14/14 tests for specificity. (Hmmm.... Do you think they first tried just 36 tests for IgM sensitivity, didn't like the numbers, and tried more? For a company selling >100K units to Spain, that number of tests seems silly small.)
The article also states "In reliability studies carried out for ENE-Covid19, a sensitivity of 73% and 79% respectively was reported for IgM or IgG, and a sensitivity of 85% considering positivity in any of the isotypes, with a specificity of 98% for IgM and 100% for IgG.")
The article reports (via Google Translate)
Quote:During the development of the field work, the greater difficulty of reading that the IgM band presents has been revealed, which we believe has produced greater variability in the interpretation of said band between the different geographical units. For this reason, all the results presented refer to the reading of the IgG band of the rapid test. The specific results referring to total positivity (IgG + or IgM +) are presented in an annex.
Any one know why one band would be harder to read (visually) than the other? Does this make sense? Or maybe "read" was more general (such as following the test procedure properly or something like that).
BC, as I don't read Spanish and am having to use Google Translate, did you see whether the numbers presented are raw numbers of positive tests? Or are they adjusted according to the sensitivity/specificity of the test? Did they adjust for age and gender differences?
If they are only using IgG, and the 79% sensitivity is a better measure than 97%, then the true number of positive results should be 26.6% greater (adjusting for sensitivity), driving the (currently undercounted) IFR downward. (Deaths in the last 24hrs = 0.8% of total deaths. New cases = 0.22% of total cases. So each week adds 5% if it were to stay steady. BTW, Spain deaths =~ NYC deaths but Spain cases =~ 1.4x NYC cases)
Of course, COVID-19 IFR calculated across a wide range of ages is highly dependent on the distribution of those ages in the population (the distribution of ages in the sample seems less important). It would seem better to calculate an infection rate (which seems to be reasonably constant at least for most adult-aged people), and then say the IFR for 80+ is X%, for 70-79 is Y%, etc. or for diabetics, Z%
RE: Spanish COVID-19 seroprevalence study - BostonCard - 05-14-2020
To answer your questions:
It is reasonable to measure IgG than IgM. Whereas IgG provides lasting immunity, IgM is only indicative of recent infection and then fades. The downside of using IgG is that it turns positive later than IgM.
Quote:Las estimaciones de la tasa de seroprevalencia por COVID-19 se realizan asignando a cada participante del estudio un peso de muestreo inversamente proporcional a su probabilidad de selección, ajustado adicionalmente por la tasa de no respuesta específica según sexo y grupo de edad. Debido al diseño complejo del estudio, todos los análisis tienen en cuenta tanto el efecto de la estratificación por provincia y tamaño municipal, como el efecto de la agrupación por hogares y secciones censales, en el error estándar de la tasa estimada de seroprevalencia y en su correspondiente intervalo de confianza.
The prevalence estimates were made by assigning each participant in the study a weight inversely proportional to their probability of being included in the study (inverse probability weighting), further adjusting for the likelihood of non-response by sex and age. Because of the complex study design, all analyses take into account the effect of stratification by province and city size, and clustering by household and census tract to determine the standard error and corresponding confidence interval.
On that basis, it doesn't sound like they adjusted for sensitivity or specificity. Note that the study itself reran the test characteristics and found:
En estudios de fiabilidad realizados para ENE-Covid19 se comunicó una sensibilidad del 73% y del 79% respectivamente para IgM o IgG, y una sensibilidad del 85% considerando positividad en cualquiera de los isotipos, con una especificidad del 98% para IgM y del 100% para IgG.
IgM: 73% sensitivity, 98% specificity
IgG: 79% sensitivity, 100% specificity
IgG or IgM: 85% sensitivity [and by definition, would have 98% specificity]
One last thing
Quote:Para la medición de anticuerpos anti SARS-Cov-2 decidimos utilizar dos tipos de test: incluye un test rápido de inmunocromatografía, que permite obtener resultados in situ para el conocimiento de los participantes y evita la venopunción y, en los participantes que accediesen a donar una muestra de sangre, una determinación de anticuerpos IgG mediante inmunoensayo, dentro de las técnicasactualmente disponibles esta determinación sería nuestro gold-standard.
In order to determine SARS-CoV-2 antibodies they decided to use two types of tests, a rapid immunochromatography test (the one reported), which allowed them two obtain results immediately to allow the participants to learn of their results and avoids venopuncture, and, in the participants who consented to giving a blood sample, a test for IgG antibodies by means of an immunoassay, which would function as their gold standard. Elsewhere in the report they state that they don't have all the results from the immunoassay. However, in the 16,953 samples they did run, the concordence is 97.3%.
BC
BC