Great point
Sensitivity = proportion of actual cases (ppl with a disease) that a test picks up
99% sensitive means 99 out of 100 truly affected ppl will be detected.
1/n
Sensitivity = proportion of actual cases (ppl with a disease) that a test picks up
99% sensitive means 99 out of 100 truly affected ppl will be detected.
1/n
https://twitter.com/thebluelady3/status/1340241806836060162
What’s the “gold standard” to determine what is a true case?
It depends on the condition in question, but for a viral RT-PCR a range of concentrations of RNA or virus can be added to a negative sample to help determine the limit of detection
2/n
It depends on the condition in question, but for a viral RT-PCR a range of concentrations of RNA or virus can be added to a negative sample to help determine the limit of detection
2/n
Specificity = proportion of unaffected ppl that a test correctly identifies as not having the disease
99% specific means 99 out of 100 truly unaffected ppl will be correctly labelled as being negative.
Many known negative samples will be examined to determine this value.
3/n
99% specific means 99 out of 100 truly unaffected ppl will be correctly labelled as being negative.
Many known negative samples will be examined to determine this value.
3/n
But it’s extremely important to consider the test in the context of what’s called the “pre-test probability”
4/n
4/n
This takes a second to think through, but the more common a disease is in a population, the more likely you are to actually have the condition if given a positive result.
5/n
5/n
Bayes's theorem
the probability you test positive AND are sick is the product of the likelihood that you test positive GIVEN that you are sick and the "prior" probability that you are sick (the prevalence in the population).
6/n
the probability you test positive AND are sick is the product of the likelihood that you test positive GIVEN that you are sick and the "prior" probability that you are sick (the prevalence in the population).
6/n
This is a really interesting point. Is a less sensitive test (or less specific) better is it’s more available?
It very much depends, in my opinion, on the consequences of an error. A readily available but insensitive test is inadequate if it puts vulnerable ppl at risk.
7/n
It very much depends, in my opinion, on the consequences of an error. A readily available but insensitive test is inadequate if it puts vulnerable ppl at risk.
7/n
https://twitter.com/sciencescanner/status/1340305672043376640
If it’s used to assess changes in the population at large, well then it’s less of a problem.this could still be very useful to guide policy decisions.
8/n
8/n
The trade-offs between sensitivity / specificity are therefore complex & often subtle. Maybe Twitter isn’t the ideal medium for this discussion, but it’s better than no discussion at all.
Testing is, however, key to controlling this pandemic.
9/9
Testing is, however, key to controlling this pandemic.
9/9
*by “this” I mean David’s comment, not Bayes’ theorem, which is also very interesting but not related to the question of test availability*
• • •
Missing some Tweet in this thread? You can try to
force a refresh
