Another #sensitivity / #specificity thread.
People are mixing up analytical sens/spec with clinical sens/spec. And translate "clinical" sens/spec in a lab-setting to real-world.
This has implications for the discussions we have about the accuracy of a test.
A #dta 🧵
People are mixing up analytical sens/spec with clinical sens/spec. And translate "clinical" sens/spec in a lab-setting to real-world.
This has implications for the discussions we have about the accuracy of a test.
A #dta 🧵
Analytical sensitivity is how much (or how little) of a certain compound can be measured.
E.g. in blood, what is the lowest glucose concentration or lowest concentration of antibodies to be detected?
Analytical specificity is about the absence of cross-reactivity.
E.g. Does my/
E.g. in blood, what is the lowest glucose concentration or lowest concentration of antibodies to be detected?
Analytical specificity is about the absence of cross-reactivity.
E.g. Does my/
Influenza - antigen test remain negative in samples with other viruses?
This does translate, but is not similar to diagnostic accuracy, as assessed in a lab.
If I have 100 samples of which I am certain that they contain virus particles, how many of those test positive?
And/
This does translate, but is not similar to diagnostic accuracy, as assessed in a lab.
If I have 100 samples of which I am certain that they contain virus particles, how many of those test positive?
And/
If have 100 samples with no or other viruses than the one I am interested in, how many test (correctly) negative?
For many pple, this sens/spec from the lab, is THE sensitivity and specificity of the test: "this is a fair representation of the test's true performance" .
But...
For many pple, this sens/spec from the lab, is THE sensitivity and specificity of the test: "this is a fair representation of the test's true performance" .
But...
From a patient's or policy maker's view, this is not helpful. If you roll out a test in a (symptomatic or asymptomatic) population, the persons tested do not have the same characteristics as the lab samples. And: the test may not measure what you are interested in.
Example: an antibody-test perfectly measures the levels of antibodies in blood. But not all people with the infection you want to test for develop antibodies and not all pple without infection have low titers, then the test will have more false pos and false negs than in the lab
Also, a test may in itself perfectly detect viruses. But if pple use the test in a wrong way, then the effective sens/spec in the population may be lower.
I don't think there is a wrong or right here (although directly translating lab-sens/spec to real world is not a good idea), but if you discuss test performance it may be helpful to be explicit about what performance you are talking about.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
