Read on Twitter
I’m commonly asked “Why don’t we just test everyone for #COVID2019 ?”
I’ll explain below in this #epitwitter #medtwitter #tweetorial #thread –
Please share/RT to help spread real understanding in the midst of all the panic and guesswork!
1/10
I’ll explain below in this #epitwitter #medtwitter #tweetorial #thread –
Please share/RT to help spread real understanding in the midst of all the panic and guesswork!
1/10
The accuracy of medical tests is typically described using two terms:
Sensitivity (what percentage of ppl WITH the disease will test POSITIVE?)
Specificity (what percentage of ppl WITHOUT the disease will test NEGATIVE?)
2/10
Sensitivity (what percentage of ppl WITH the disease will test POSITIVE?)
Specificity (what percentage of ppl WITHOUT the disease will test NEGATIVE?)
2/10
… but what the patient wants to know is measured by two DIFFERENT terms:
Positive Predictive Value (PPV- if you test POSITIVE, do you actually HAVE the disease?)
Negative Predictive Value (NPV- if you test NEGATIVE, do you actually NOT HAVE the disease?)
3/10
Positive Predictive Value (PPV- if you test POSITIVE, do you actually HAVE the disease?)
Negative Predictive Value (NPV- if you test NEGATIVE, do you actually NOT HAVE the disease?)
3/10
PPV and NPV are calculated using sensitivity and specificity, but ALSO using the disease’s PREVALENCE.
Prevalence: how many people currently have the disease
In a case like #COVID2019 , prevalence is low, making PPV and NPV very unstable.
Let's explore.
4/10
Prevalence: how many people currently have the disease
In a case like #COVID2019 , prevalence is low, making PPV and NPV very unstable.
Let's explore.
4/10
Let’s imagine the current prevalence of #COVID2019 in the US is 1 in 1,000 (it’s probably much, much lower than this).
5/10
5/10
Let’s also assume we have a test with 99% sensitivity and 99% specificity. A near-perfect test!
6/10
6/10
So when 1 in 1,000 people HAVE a disease (prevalence), only 99/1,098 will have a POSITIVE TEST – that’s only 9% (PPV)!
… though almost everyone who TESTS negative DOESN’T HAVE the disease (99.9% NPV)
7/10
… though almost everyone who TESTS negative DOESN’T HAVE the disease (99.9% NPV)
7/10
So if we test everyone, VERY FEW of the people who HAVE the disease will TEST positive.
If we limit testing to people WITH SYMPTOMS who have had KNOWN CONTACT with disease carriers, PREVALENCE goes up (you’re more likely to actually have the disease).
8/10
If we limit testing to people WITH SYMPTOMS who have had KNOWN CONTACT with disease carriers, PREVALENCE goes up (you’re more likely to actually have the disease).
8/10
If we assume a HIGHER prevalence, PPV goes up. Let’s assume prevalence of 50%. Using the same 99% sensitivity and specificity:
In this case, 99% of people who TEST POSITIVE actually HAVE the disease (PPV), and 99% of people who TEST NEGATIVE actually DON’T HAVE it (NPV).
9/10
In this case, 99% of people who TEST POSITIVE actually HAVE the disease (PPV), and 99% of people who TEST NEGATIVE actually DON’T HAVE it (NPV).
9/10
SO, for now, with low disease prevalence, the “test everyone” approach wouldn’t be very helpful. If #COVID2019 infection rates (prevalence) rise, it may be more appropriate to test more people.
10/10
10/10
