Test This
One of the refrains we’ve heard repeatedly during the Panicdemic (which is arguably worse than the pandemic) is: “We Need More Tests! We Need More Tests!”
There is a Chicken Little vibe to these calls for testing. A sense that people are running around like the sky is falling, and not thinking through the right testing strategy.
What are tests for? One is for diagnostic purposes in specific cases. To be frank, the value of such tests is minimal. There is no unique therapy for acute Wuhan Virus sufferers. The protocol is to treat the symptoms of acute respiratory distress the same way as one would treat such distress from other causes. So knowing that someone’s acute distress is caused by agent X as opposed to agent Y is of limited therapeutic value.
Insofar as identifying someone expressing symptoms would help identify others so exposed, a much more efficient strategy is to presume that the symptomatic individual is suffering from WV, track his/her contacts, and monitor and quarantine said individuals accordingly. Yes, there will be Type I (false positive) errors, but the cost of such errors is likely to be relatively small if an individual is suffering from an acute condition, regardless of the exact pathogen that caused it. That pathogen is obviously capable of causing severe problems, so why not isolate those exposed to it, even if you don’t know exactly what it is?
Another purpose of testing is to collect information about the prevalence, virulence, contagiousness, and fatality of the disease. Such information can be used to optimize the policy response.
Testing those who are symptomatic and/or have been exposed is exactly the wrong way to go about that. Such a testing strategy is rife with sample selection bias.
For weeks (mainly on Twitter) I advocated construction of a random panel data set. Select people at random. Test them, and test them at regular intervals–including those who tested negative. This would provide an unbiased sample that would permit more reasoned assessments and judgments about the nature of the pathogen. We could see how many people had contracted the virus, how many people they infected, the mortality rate (and how the mortality rate varied with age, health status, etc.), and the trajectory of the virus.
If that had been done, say, in January when shit started to get real in China, perhaps we could have been able to condition policy on better information. (Not to mention if the CCP had done that in, say, December, when it knew it had a problem on its hands–but decided to suppress information rather than suppress the pathogen.)
Why didn’t our Technocrats figure this out? Yeah. We should put more of our lives in their hands.
But that opportunity to get unbiased data has passed. Now we are forced to respond based on the most sketchy and biased data. Chicken Little proposals about testing will generate . . . more biased data. Which is arguably worse than useless.
To prevent the spread of the virus we don’t need testing but social distance maximization which I believe is happening with the cancellation of every event. We need a quick full court press to stop the virus in its tracks.
Comment by Margaret — March 14, 2020 @ 6:03 pm
I’d think it might be wise to do lots of tests on medical staff. You really don’t want hospitals full of doctors and nurses who infect patients.
But then I’m the sort of virus Nazi who can see merit in tight constraints on visitors in hospitals.
Comment by dearieme — March 15, 2020 @ 6:59 am
“Test them, and test them at regular intervals–including those who tested negative”
Err, surely you’d only re-test those whose tested negative?
TBH I think such a testing regime would now only be of academic interest to epidemiologists etc. The virus is already rife is many populations across the world, the US included, hence the policy shift towards trying to minimise its impact on those vulnerable groups.
Comment by David Mercer — March 15, 2020 @ 2:23 pm
“Why didn’t our Technocrats figure this out?”
Because it’s not a very good idea? Infection is not “nicely” (from a statistics perspective) distributed – infection rates are highly “clumped” geographically. Random sampling of the scale to judge political opinions for example, is going to produce highly inaccurate estimates (i.e. useless) for coronavirus mortality rates. So unless you’re prepared to test millions or 10s of millions of Americans and test them repeatedly, the numbers that will come out of such an exercise will be almost certainly useless.
Not all experts in other fields are idiots, prof.
Comment by derrizanile — March 15, 2020 @ 2:24 pm
Just read your 3rd from last para (speed-read it the first time while multitasking. Excuses…), so ignore my last para.
This crisis is unprecedented, so its to be expected there’ll be no end of policy SNAFUs. Who knows what was being discussed – and discounted for whatever reason – back in January? Also, do you honestly believe Trump would have conceded to this?
Comment by David Mercer — March 15, 2020 @ 2:45 pm
@derrizanile. All your issues are matters of sample design. FFS.
Comment by cpirrong — March 15, 2020 @ 3:32 pm
@cpirrong – that’s the point. Maybe there is an easy way to design a way of sampling the population for something like this but it doesn’t seem obvious to me. I’m not a statistics expert by any means but I spend enough time working with stats to be somewhat skeptical. This random sampling idea came up around the coffee machine with work colleagues a while back also – it seems an obvious thing to do given all we had were dodgy numbers from China at the time. But given the dynamic nature of the situation and the highly clustered nature of the infection, I’m inclined to believe that the required number of test administrations would have to be absolutely huge to provide any sort of useful statistics.
The problem is logistics and resources rather than technical – your other idea – tracking/tracing individuals was initially practiced in most European countries but could only be maintained for a while when numbers were low – it simply could not be sustained even just a few weeks after the first cases were identified.
Comment by derrizanile — March 15, 2020 @ 5:54 pm
The Diamond Princess cruise ship population was fairly thoroughly tested. 696 cases, 7 fatalities (so far). I would love to know the demographics for those on board/infected/fatalities since my guess is that a cruise would have an older population on board. At this point it is probably the only population where we really know the denominator for the fatality rate.
Of course, the Diamond Princess cases were treated in a first world medical system that was not overloaded. The fatality rate under other circumstances could be higher.
Comment by Jack — March 15, 2020 @ 9:07 pm
Medical stats: it reminds me of the days when a friend’s job required her to read bits of the medical research literature. Her favourite sentence was “The rats exhibited a 100% mortality response.”
Comment by dearieme — March 16, 2020 @ 11:32 am
The real question: does testing divert resources from something else that is more helpful?
Comment by Andrew Stanton — March 18, 2020 @ 12:45 pm
@dearieme et al. Here is a very good piece that echoes some of the things I said in this post, and my Wuhan Virus: WTF? post.. Including this, @derrizanile:
If the beneficent CCP had given the world a head’s up, instead of throwing sand in our collective eyes, perhaps we would have had a chance to implement such a strategy. Yes, clustering poses issues–but that is something that a sample involving a few 10s of tests would reveal. That is a good thing, not a bad thing.
Resources for testing–like all resources–are constrained. Which is precisely why thought needs to be applied on how to use these resources. The willy-nilly, biased sampling approach has proved wildly inefficient. Worse than providing little information, it has provided “information” that misleads those who do not understand its limitations. Which, I can guarantee you, is a group that includes all of the major decision makers.
This is why Ionnidis is right when he says we are flying blind, and have to play economic/health Russian roulette: kill the economy, or kill people, based on a spin of the cylinder.
The short version of Ionnidis: we don’t know jack–apologies, @Jack.
This radical uncertainty is precisely why we are seeing such unprecedented volatility (realized and forward-looking) in financial markets, and such huge risk premia (as reflected in massive price declines).
Comment by cpirrong — March 18, 2020 @ 1:48 pm
The best testing regime likely depends on the strategy, containment or mitigation. Singapore appears to have used the information on travel history and tracing effectively early on to guide sampling, not clear that other countries have done this as well. This has conserved scarce tracing resources, which in Singapore’s case includes police investigators. It not just about the testing, it’s a system, you have to get all parts working. Containment can be done – many Western countries have simply not learnt the necessary lessons from advanced Asian countries. And to think we once expected them to learn from us.
CP, you can keep fuming at the CCP, but it is hard to argue they hid the fact there was a problem very well. Trump has just shut down global travel at massive cost and no benefit, after weeks of having no ground game. Too late. In your other post, you note countries get the leader they deserve. Too true. And sad.
I agree on the uncertainty. Good leadership would have a communication strategy with credible information and scenario modelling to start to put bounds around the uncertainty. After all, the pandemic itself will be temporary. Instead, we have desperate scramble to throw squillions at a problem that apparently didn’t exist a few weeks ago, even though we also knew that that cagy Xi was hiding something from us. Doh!
Comment by PJD — March 19, 2020 @ 6:30 am