Read on Twitter
THREAD:
A quick lesson in medical statistics for those of you who believe the current "second wave" scare story out of #Victoria.
It starts with the concepts of sensitivity, specificity and positive predictive value.
Even doctors struggle with this..
A quick lesson in medical statistics for those of you who believe the current "second wave" scare story out of #Victoria.
It starts with the concepts of sensitivity, specificity and positive predictive value.
Even doctors struggle with this..
Each medical test that you have has a built-in level of accuracy (and therefore inaccuracy). Most tests will identify most people with a disease correctly and also without a disease correctly. But no tests are perfect.
The #Theranos story for example (but I digress)...
The #Theranos story for example (but I digress)...
Sensitivity is the probability that the test will be positive in people with a disease.
The "pick up rate" if you will. Usually this is high. Expect 100% but we are seeing reports of 50-70% in #covid19
So, of 100 people with Covid, 30 could test negative.
The "pick up rate" if you will. Usually this is high. Expect 100% but we are seeing reports of 50-70% in #covid19
So, of 100 people with Covid, 30 could test negative.
"OMG" you cry. "There are people running around infecting us who have tested negative and are allowed out!!"
Yep. That's one of the (many) reasons why testing people to be allowed out of 2 weeks' #quarantine is pretty much a waste of time. Are you listening @NSWHealth?
Yep. That's one of the (many) reasons why testing people to be allowed out of 2 weeks' #quarantine is pretty much a waste of time. Are you listening @NSWHealth?
OK now the flip side. This one is much more important but most people don't grasp it.
Specificity - is the proportion of people WITHOUT the disease who test negative.
Again, it should be 100% BUT it never is - especially not in the real world
https://www.aruplab.com/news/4-21-2020/How-Accurate-Are-COVID-19-Tests
Specificity - is the proportion of people WITHOUT the disease who test negative.
Again, it should be 100% BUT it never is - especially not in the real world
https://www.aruplab.com/news/4-21-2020/How-Accurate-Are-COVID-19-Tests
Most doctors don't think about specificity and, to be honest, most of the time it's not that important.
But here is a GREAT article in the @bmj_latest (which, BTW, I'm no fan of) that goes through this really well.
(h/t @drjessicawatson) https://www.bmj.com/content/369/bmj.m1808
But here is a GREAT article in the @bmj_latest (which, BTW, I'm no fan of) that goes through this really well.
(h/t @drjessicawatson) https://www.bmj.com/content/369/bmj.m1808
So, although it's difficult to establish the real-world specificity of the #covid19 test, Jessica picks 95% (0.95) as a starting point. I suspect it might be a bit higher but who knows. Almost certainly less than 99%.
Let's split the difference and go for 97%...
Let's split the difference and go for 97%...
OK so now we have a #covid19 test with a 97% specificity. That means that, out of every 100 uninfected people, 97 will test negative.
And being conservative with our test statistic...
3 will test positive.
Now that's important.
And being conservative with our test statistic...
3 will test positive.
Now that's important.
Look at this graph. It's the "official" graph for Australia, from @JohnsHopkins. I'm not going to go into how they have got this data (that's another thread) but let's say it's accurate.
This looks like a few infections at the start, then a peak then some more infections....
This looks like a few infections at the start, then a peak then some more infections....
You could even imagine a "spike" in the last few days (even you have been programmed to imagine that there is one - there isn't)
That one bar sticking up is just a reflection of the lack of tests the day before.
The numbers hover steadily around 30 "cases"/day
That one bar sticking up is just a reflection of the lack of tests the day before.
The numbers hover steadily around 30 "cases"/day
Now we need to remember that Australia is a bit of an anomaly in the Western world, because of its very low infection rates despite being close to China.
And there was a LOT of testing.
And there was a LOT of testing.
Interesting. The overall positive rate is really low. 0.3%. Wow. This is with a test that has a false positive rate which is unknown but expected to be 3%. Maybe it's a deluxe test with a 0.5% false positive rate.
In any case we are into the realm of statistical "noise"
In any case we are into the realm of statistical "noise"
But this is time and prevalence dependent.
You can see this from the wonderful graphing put together by the people at http://covid19data.com.au
You can see this from the wonderful graphing put together by the people at http://covid19data.com.au
I am going to show you the national chart, the NSW chart (which represents the pattern in all the states except one) and the #Victoria chart.
The bars are the daily cases, the line with dots is the testing rate (different scales).
NSW - National - Victoria
The bars are the daily cases, the line with dots is the testing rate (different scales).
NSW - National - Victoria
Now, what is the state of #Victoria up to?
They've gone from a baseline daily testing of less 1000/day to 5000/day
And the case numbers have risen in unison.
No other state has done this.
The graph shows about 16000 tests and 25 cases/day - the "spike"
They've gone from a baseline daily testing of less 1000/day to 5000/day
And the case numbers have risen in unison.
No other state has done this.
The graph shows about 16000 tests and 25 cases/day - the "spike"
Hold on. 25/16000 is 0.15%
So, 0.1% of tests conducted a positive when we don't know the specificity of the test but we know it's not 100%.
We are now at the possibility that the test here is so good the specificity is 99.9% (almost unheard of)
Which means...
So, 0.1% of tests conducted a positive when we don't know the specificity of the test but we know it's not 100%.
We are now at the possibility that the test here is so good the specificity is 99.9% (almost unheard of)
Which means...
If the case rate in Victoria is 0.15% and the specificity of the test is an amazing, unheard of 99.8-99.9%
ALL THE CASES REPORTED IN VICTORIA ARE A CONSEQUENCE OF THE FALSE POSITIVE RATE OF MASS TESTING
It's just noise.
Still confused?
OK it's about prevalence...
ALL THE CASES REPORTED IN VICTORIA ARE A CONSEQUENCE OF THE FALSE POSITIVE RATE OF MASS TESTING
It's just noise.
Still confused?
OK it's about prevalence...
Let's look at the national chart again.
At the peak there were 370 cases/day for a testing rate of 12,000/day. That's 3%. If we're now saying that the specificity of the test is 99.9% (amazing), that is 30 true positives for every false positive.
That's an OK test!
At the peak there were 370 cases/day for a testing rate of 12,000/day. That's 3%. If we're now saying that the specificity of the test is 99.9% (amazing), that is 30 true positives for every false positive.
That's an OK test!
So at the peak of infection, if you tested positive you had a 30/31 chance that it was real.
Not bad for a medical test (honestly).
Now let's turn to this week. Here's the chart again
30 cases/day, 47000 tests (right side of the chart)
Not bad for a medical test (honestly).
Now let's turn to this week. Here's the chart again
30 cases/day, 47000 tests (right side of the chart)
That is less than 0.001%
The false positive rate of this test is almost certainly more than 0.1%
So, the probability that you actually have a true case at this prevalence is tiny.
Almost all the positive tests at this prevalence are FALSE POSITIVES.
The false positive rate of this test is almost certainly more than 0.1%
So, the probability that you actually have a true case at this prevalence is tiny.
Almost all the positive tests at this prevalence are FALSE POSITIVES.
This is a well known phenomenon in medical statistics.
https://www.sciencedirect.com/topics/nursing-and-health-professions/positive-and-negative-predictive-values
https://www.sciencedirect.com/topics/nursing-and-health-professions/positive-and-negative-predictive-values
So, please, @abcnews @SkyNewsAust @australian let's just calm down a little.
There is no spike. There is no second wave.
What there is, is a drive to overtest the general population to generate false positive test results.
There is no spike. There is no second wave.
What there is, is a drive to overtest the general population to generate false positive test results.
I can't imagine why #chairmandan @DanielAndrewsMP would want to do this of course.
I mean, it couldn't possibly be that there is a very big news story that he really really wants to get hidden.
Could it?
#BeltandRoad #corruption
/end
I mean, it couldn't possibly be that there is a very big news story that he really really wants to get hidden.
Could it?
#BeltandRoad #corruption
/end
ADDENDUM:
I have assumed that @VicGovDHHS is using Australian manufactured test kits.
It is possible that
@DanielAndrewsMP's links to China has meant that they have imported test kits, with higher false positive rate.
More fishy "research": https://pubmed.ncbi.nlm.nih.gov/32133832/
I have assumed that @VicGovDHHS is using Australian manufactured test kits.
It is possible that
@DanielAndrewsMP's links to China has meant that they have imported test kits, with higher false positive rate.
More fishy "research": https://pubmed.ncbi.nlm.nih.gov/32133832/
As of yesterday there was 1 person in Australia in ICU with #covid19 - in Victoria no less.
This disease has been eradicated in Australia to all intents and purposes. If there is a new "spike" of significance it's going to be a new disease
This disease has been eradicated in Australia to all intents and purposes. If there is a new "spike" of significance it's going to be a new disease
So, #chairmandan wants us to believe that #victoria is having a "second wave" that is as big as the "first wave" despite this no longer being a novel virus.
Not possible.
Either the kits are tainted or they've imported a new virus.
Not possible.
Either the kits are tainted or they've imported a new virus.
