14 April 2020

Response to an objection to the point that even imperfect testing/tracing/isolation is likely our optimal strategy

I looked at the Romer argument and find an important flaw/omission in it and your argument: a certain percentage of those false negatives will go out into the community with impunity due to their clean diagnosis, and be in contact with exponentially more people each day thereafter.  As far as I see from your link, Romer's model is mired in math and doesn't account for the reality of human behaviour.  A very serious flaw.

Please link me to a place where he does address this, and I will review his argument and my opinion, as we all should be willing to do in the face of reasoned argument. 

I got the above comment as a response to my posting or Paul Romer's (https://paulromer.net/ ) analysis purporting to show that even imperfect massive scale testing, contact tracing, and isolation of positives is a better course of action than alternatives. Below is my attempt at a response. 

First, I make no claim to be a statistical economist, demographer or epidemiologist, although all of us are delving into these areas more than ever before because they've suddenly become important to the continuity of our everyday lives. I did a quick look and couldn't find exactly what was asked for, which if I understand it, is a piece by Romer in which he specifically addresses what is described as a flaw or omission in his argument. Folks might want to look at this (1), which gives some more detail to the thought process behind the somewhat more general argument that testing, even less than perfect testing, combined with contact tracing and quarantine of positives, can achieve the same or better results than more general (but necessarily, because of the same "human behavior" cited, less rigorous) "social distancing," and, indeed, when applied to the majority of the population, is much more effective. 

I have to say dismissing an argument because it is "mired in math" is not particularly constructive. Economics, finance, demographics, and epidemiology (as well as climate science and other areas where direct measurement and analysis is effectively impossible and modeling is the only option)... all depend on mathematical modeling of aggregate human behavior. It's almost a truism that on an individual level human behavior is chaotic and unpredictable, but on a population level it can often be quite accurately described using statistical models and data

But the more fundamental objection, if I understand it correctly, is one that needs to be addressed or at least clarified. We're all in this together, as has become the watchword of this time, so we do need to work on a common understanding of the problem and how best to deal with it. I think part of the immediate problem is that we're starting in the middle, by asserting a response to an earlier objection to the basic argument for massive scale testing/tracing/isolation. And then the response is a response to the response. Let's try starting at the beginning, and I think I see where what is being objected objecting to drops out as something that actually is included in the original argument. If anyone still doesn't think so, I'm happy to try to understand better what they're objecting to and work towards a better understanding of the whole problem. 

The original premise is that it is not good enough to accept the status quo of generalized "social distancing," which is largely the result of inadequate built capacity for scaling up testing for this thing. We just didn't anticipate this need, and prepare for it, and we have been caught flat footed. The advocacy is for major mobilization of public resources to quickly create new and greatly upscaled testing capacity so that a large percentage of the population can be tested (both serum antibody tests, to find people exposed and either recovered or otherwise no longer positive and possibly immune; and for active viral infection). This would be enormously expensive and require "command economy" action on the part of government, which the Federal government under Trump is clearly unwilling to do. Included in the regime would be massively upscaled capacity to do contact tracing and provide for effective quarantine of people who are positive and incubating the virus or mildly ill. (The seriously ill, of course, have to be treated in hospitals). 

The rationale for doing this has been explained in a number of places, which I've cited to on Facebook several times. See this (2) for a relatively mainstream version of this idea. Even Joe Biden just yesterday set forth a version of it; Tony Fauci has talked about it; indeed it's become a consensus view, which has, to some extent, already been effectively put in place in other countries. Most notably Iceland, which, with only 300,000 people, can accomplish this more easily, but can still serve as a test of concept. By having much more information about who is infected, and who is potentially exposed but immune (serum testing), and doing the usual epidemiologic response of contact tracing and isolation, it should be possible, from the inexorable math of herd immunity, to drive the presence of the virus in the community to very low levels. Subsequent outbreaks could then be dealt with using tried and true response techniques (which were blown through in the case of this virus early on). It should be emphasized that this kind of process is envisioned for AFTER the current social distancing regime has already succeeded in bringing down the rate of new infections to very low levels. But we need to prepare for that next phase NOW, because a great deal of work and development will be needed for it to work. And time is being wasted and we are NOT doing this. 

The core argument is that this kind of approach is likely the only way to achieve a sufficient reduction in the risk of infection at some time before effective antivirals and/or vaccines become available for anything like a return to normal social and economic activity. Probably not including large crowds or close contact such as lines in amusement parks, but people will, the thought goes, be able for the most part to go to work, and with some sensible physical distancing, go to restaurants, shop, meet with friends, etc. The current regime will likely lead to a short term reduction to low levels of new cases, but the belief is that if the physical distancing is then relaxed, in the absence of a strict testing/contact tracing/isolation regime, it is the nature of epidemics that it will likely break out again, and there will be little public health infrastructure in place to prevent it from spreading almost as fast and wide as it did the first time. But WITH adequate massive scale testing/tracing/isolation, outbreaks can be quickly quelled and the epidemic can be brought to some kind of control, even before there is effective treatment or a vaccine. The modeling in these scenarios shows that the overall cumulative infection numbers, deaths, and economic impact, will all be lower, and that, even taking the huge costs and effort involved in mobilizing this kind of response, which greatly exceeds the ability of current laboratory and manufacturing facilities to scale up and would therefore require major public health infrastructure mobilization, the medium term cost both in terms of public health outcomes and economic impact would be more than worth it. Unfortunately, this understanding has not penetrated the minds of the people in charge of Federal policy to date. 

So far so good, but then there is the objection that Romer was addressing, which is that what if the scaled up testing is unreliable to some degree? The short analysis I linked to was his attempt to show, through statistical analysis, that even with some (not huge, but some) degree of inaccuracy in the testing, the course of action (testing/tracing/isolation) is STILL the best available option, because all the other options result in WORSE outcomes, again, both in terms of public health and in terms of economic impact overall. Of course, there are limits. If the testing is so inaccurate that the average rate of spread R0 ("R-naught") is not brought below 1, which is the threshold for herd immunity, then it will not work. In case this isn't clear, if on average each infected person infects 3 people, which is about what this virus is without any measures to prevent spread, then it quickly envelopes the entire population and spreads uncontrollably until a large percentage of the entire world population is infected. Epidemics in those cases only die down when natural immunity of recovered people reaches a certain level, and the R0 drops below one. Once the rate of infection drops to less than one additional person for each newly infected person, whether through public health intervention or just through natural immunity, the epidemic will quite rapidly disappear. This is not intuitively obvious to everyone, but it is mathematically demonstrable and backed up by historical experience of literally hundreds of past epidemics. (Sorry if this is all pretty basic, but I'm trying to be as comprehensive as I can in case anyone reading this is unfamiliar with this concept). 

We could achieve this R0<1 by social distancing alone, with no or insufficient information from testing, but there is every reason to believe that if we tried to ease up on this social distancing without effective treatments or vaccines, the epidemic would break out again and we'd have not much better ability to control it than we did in the first place. And grossly incorrect testing obviously would have the same result, since as is noted, the people who test false negative will then be the same as untested positives without the testing regime. But testing which is above a certain threshold of accuracy will achieve the desired effect, just not quite as efficiently as a "perfect test." Obviously, perfect testing is better, better accuracy is better than worse accuracy, but what is possible and optimal will fall somewhere short of perfection. This is also true of vaccination and treatment options: no treatment or vaccine is 100% effective, nor do they reach 100% of the population, but by getting most of the way there, you can get the R0 under one and the epidemic will fairly rapidly disappear. In some cases, even without anywhere near 100% penetration, pathogens actually become extinct, but not usually in cases where there are nonhuman hosts, which is unfortunately the case with this virus, in all probability. 

So, to the point. Let me restate it verbatim: "a certain percentage of those false negatives will go out into the community with impunity due to their clean diagnosis, and be in contact with exponentially more people each day thereafter." I believe that the reason this is not a serious flaw in the argument is that it just isn't true; the keys being "certain percentage" and "exponentially." If, say, there is a 15% false negative rate, those people will encounter a population which is not "virgin," but is, to whatever extent of penetration, already itself subject to testing and isolation of positives. So, in other words, the false negatives are a drag on the process of limiting the overall rate of infection, but they do not, beyond a certain threshold, completely derail that process. They do NOT spread infection exponentially, because the people they infect are themselves getting tested, and if positive, being isolated or treated. The testing does not occur all at once and one time only, but is a regime, in place over time, and continuing until the disease is completely controlled, either because of natural decline or due to treatment or vaccines. Just as herd immunity does not require 100% penetration to be effective, neither does the accuracy of the testing have to be 100% to achieve the reduction of the spreading rate R0 to less than one. R0, in other words, doesn't have to be 0, it just has to be less than one, for the epidemic to begin to disappear. And it does disappear. If one person infects two people, and they each infect two (=4), it isn't long before you have millions infected. But by the same token, if ten people infect 7, who infect 3.5, who infect 1.75, etc., it is surprisingly rapid for the epidemic to simply wink out (because people eventually recover or die, so the number of active infections begins to decline, slowly, then rapidly). We are nowhere near that point yet, but it is possible, and this kind of regime appears to be the fastest, most reliable, and most cost effective way to achieve this goal, short of treatments and vaccines. It would be VERY costly, but the alternatives, versions of what we're doing now or doing nothing, would be far MORE costly. 

I hope this makes sense. The factor of reinfection by false negatives actually IS included in the kind of analysis Prof. Romer was referring to, but that is not perhaps as clear as it could be. 

No comments:

Post a Comment

Gyromantic Informicon. Comments are not moderated. If you encounter a problem, please go to home page and follow directions to send me an e-mail.