I'll get back to you in maybe a week, by which time I'll have worked my
way through that little lot.
(I've got a couple of bits of homework to do before I can start on the
'implications for my modelling' part of the job).
Makes yer think about the talent that's sitting around just waiting to
be baited into delivering the goods (:-)}
Seriously, many thanks for this Trent, not to mention for your ongoing
hard work.
_________________
On 20/4/20 7:16 pm, Trent Yarwood wrote:
Hi Roger,
I haven't got the time to give this the attention it probably deserves,
because you've obviously put quite a lot of thought into it. I'm obviously
just a little bit busy at the moment.
In defence of the SEIR model, it's a pretty well-established epidemiologic
tool that has the advantage that it's a lot easier to model because it
works on far fewer parameters than your model does. You're correct that it
does make some assumptions to simply but (although I'm not an
epidemiologist) it works well enough for most of the models that I use as
an infection clinician.
Your third document (CVCT) does have some holes that jump out to even a
cursory glance by a busy clinician.
The Doherty group released another model a week later which models (but
doesn't measure) community transmission:
https://www.doherty.edu.au/uploads/content_doc/Estimating_changes_in_the_transmission_of_COVID-19_April14-public-release.pdf
and strongly suggests that it's not that much of a thing. I wrote an
explainer here:
https://theconversation.com/latest-coronavirus-modelling-suggests-australia-on-track-detecting-most-cases-but-we-must-keep-going-136518
The problems with your suggestions lie with the random sampling.
(Apologies if this is vastly over-simplifying it and readers already know
some of it)
We currently test for viruses in two main ways.
PCR (essentially: direct detection of the viral DNA/RNA. Generally
indicates active infection, but remains positive for a while after
resolution and suggests but doesn't absolutely imply infectivity)
Serology (antibody testing. Generally, but not always indicates past
infection, although sometimes gives an indication that the infection is
recent).
Prevalence surveys (which is what we call your idea of random-sample
testing) usually rely on serology, because the testing material is
available (eg: by doing antibody screening on blood donors, or people who
have blood tests for other reasons). This way we can measure how many
people *have had* the disease as opposed to the number of people who
*currently do have* it - which is what we need to know in terms of
loosening lock-down.
Prevalence surveys of active infection are difficult because:
1) there's a shortage of testing reagents (world wide) and doing lots of
screens on asymptomatic people will restrict our ability to do PCR testing
not only on people with suspect coronavirus infection, but all other
diagnostics done by PCR (because they have a common reagent)
2) the testing is more invasive as it involves having a swab stuck up your
nose until your eyes water and isn't likely to be highly popular (
https://youtu.be/DVJNWefmHjE)
3) The performance of the PCR test isn't well established in asymptomatic
people
Three is the one worth expanding on the most. PCR based-tests are
generally very sensitive (unlikely to miss an infection if the DNA is
there) and specific (unlikely to have false positives if it isn't). But
when you're testing for a disease which is uncommon (which the epidemiology
and the second Doherty modelling study strongly suggests it is), even very
sensitive and specific tests can have low positive predictive values (
https://en.wikipedia.org/wiki/Positive_and_negative_predictive_values) - ie
the likelihood that a positive test is the real deal, or that the negative
test is actually legit.
In addition to the technical performance of the test, there is also the
issue of biological negative tests providing false reassurance. For
example, if someone is in the incubation period of the illness (ie have
come into contact the disease but have not yet developed symptoms) there
are three phases.
The incubation period proper in which they don't have symptoms and tests
are negative because the virus hasn't manifested yet. (correctly negative
test, but will develop symptoms and infectivity soon)
The pre-symptomatic phase of the infection (this is what people worry about
when they talk about "asymptomatic transmission" - which by the way is very
unlikely at a population level to lead to large transmission clusters; most
infections come from symptomatic patients) - this individual if tested
would have a positive test, but no symptoms (this is what you're looking
for when you're talking about screening)
Then there are symptomatic patients (positive test, based on the
sensitivity, and symptoms).
There's also talk about screening with CT scans which are a bad idea
because the findings are very non-specific (ie: quite a lot of things look
like COVID changes; in a low-COVID prevalence environment like ours, it
would have a very poor negative predictive value) and because it has
limited availability and involves radiating people which is probably not
justified by how good a screening tool it is. The maths for CT screening
would probably be quite different in say, New York where there is lots and
lots and lots of COVID, so the positive predictive value is probably pretty
good, even though the specificity of the test isn't that high.
Now I have to get back to work, so I'll leave it there.
Trent.
--
Roger Clarke mailto:[email protected]
T: +61 2 6288 6916 http://www.xamax.com.au http://www.rogerclarke.com
Xamax Consultancy Pty Ltd 78 Sidaway St, Chapman ACT 2611 AUSTRALIA
Visiting Professor in the Faculty of Law University of N.S.W.
Visiting Professor in Computer Science Australian National University
_______________________________________________
Link mailing list
[email protected]
http://mailman.anu.edu.au/mailman/listinfo/link
