Living With The Term “Breakthrough Infection”
I started out in this post with a rant about the term “breakthrough infection,” which I still regard as one of the most ineptly-chosen terms in the history of scientific terminology. It’s a term that has done a great deal of damage, both to the clarity of scientific discussion and to the ends of public discourse and policy, because the image that it conveys of a vaccine’s protection—as something hard that can be “broken through”—is so misleading about how vaccine-primed immunity works that it actually causes errors in thinking and language, even among people who should know better. Whoever introduced that term has a great deal to answer for.
Vaccine-primed immunity (or any immunity, for that matter) is “soft”, not hard. It doesn’t work like armor, or anti-aircraft missiles. It works by noticing that an incipient infection is in progress, and reacting to it. Vaccination allows that reaction to occur more swiftly than would be the case otherwise, but there is nonetheless an infection that takes place, and some viral growth occurring at some rate, and reaching some peak viral load, before the immune system’s infection-clearing processes overwhelm it.
Official diagnoses of COVID-19 have largely been made on the basis of an RT-PCR test, which as I discussed in this post is an extremely sensitive assay, miraculously rolled out from biochemistry discovery wet labs to clinical production testing of millions of specimens per day the world over. It is capable of sensing very low viral loads, including levels consistent with infections that are going nowhere, because the immune system in question has a grip on the situation and is about to clear the infection in a routine mopping-up operation that is exactly what the designers of the vaccines intended for it to do under these circumstances. Clearly, a “Positive” RT-PCR test in such a case is a wrong result. It’s not exactly a “false positive”, because there was actually SARS-CoV-2 virus correctly detected, but to call this kind of result a “breakthrough infection” is clearly a travesty. Equally clearly, a number of early reports of “breakthrough infections” were necessarily of this variety.
This is what one might regard as being in the nature of a subtle category error. Early public discussions of “breakthrough infections” were shot through with at least one much coarser category error: every vaccinated COVID-19 case was categorized as a “breakthrough infection”, despite the fact that it was perfectly well-understood from clinical trials that the protective efficacy of the vaccines against infection is not 100%. That is, vaccination reduces the average risk of infection per infective contact, but not to zero. A vaccine efficacy of 95% (Pfizer, or Moderna, say) amounts to a prediction that some vaccinated people will still get infected (more or less, under controlled conditions, that 5% as many vaccinated people will get infected as unvaccinated people). Calling such infections “breakthrough infections” is alarmist idiocy.
Sorting through this kind of nonsense was the reason that I felt impelled to set out the “typology of breakthrough infections” described here. Clearly any effort to purge the term “breakthrough infection” from public discourse is a doomed exercise analogous to removing pee from a swimming pool, given how embedded the term has become. But it might be possible to make some distinctions that allow the term to do some useful work after all. Hence the typology, which is an effort to embed some notions of immunology and RT-PCR diagnostic power in the discussion.
The COVID-19 Vaccines Work, And We Understand Why
The “breakthrough” terminology has been particularly destructive of the effort to give the public the true picture of the tremendous scientific success story attending the development of COVID-19 vaccines. By creating the mistaken impression of a “hard but fragile armor” that is now vitiated as it is being broken through by newer, tougher little variant viruses, it has embedded a landmine in our very language that hampers the very necessary global vaccination effort. Cleaning up the terminology is therefore also connected to aiding vaccination campaigns.
When I first drafted a typology, it had three clean categories: Type 1 (“Vaccine Escape Breakthough”); Type 2 (“Statistical Breakthrough”); and Type 3 (“False Breakthrough”). I broke up Type 3 into sub categories—3a (“False Breakthrough”) and 3b “Infective, or Semi-Breakthrough”)—because it began to dawn on me that the later fast-moving variants (Delta and Omicron) were capable of multiplying themselves so fast that at some point a threshold is likely crossed, where even though the immune system’s infection-clearing processes are still going to wipe out the incipient infection according to program, in the meantime the viral load may in fact reach levels high enough to be consistent with infectivity. In such a case, dismissing the infection as “false” is clearly cavalier, because an infective person is a danger to others, even if that individual is not personally likely to be in any particular danger. Hence Type 3b.
This brings me to a much noted and remarked-upon feature of the vaccines: a vaccinated individual is at much lower risk of severe COVID-19 than an unvaccinated individual, even in case of a “breakthrough” infection. There is not much mystery as to why this is the case: human immunity is a many-layered defensive system, with a number of robust moving parts that operate on different timescales. The now-famous “decline of vaccine immunity” that dominates media discussion is really about the surface defense layer: the antibodies. Antibodies play an important role in immunity, but it’s only a delaying role: they harass and slow down a new, incipient infection while the “cellular” immune system, (which in my simplistic discussions I’ve represented by “the T-cells” because I can only cop to Wikipedia-grade understanding of such matters). In other words, when one reads of the “declining immunity due to vaccines”, one is really reading of the declining ability to slow down the growth of the virus. There is a great deal of evidence that the cellular immunity created by vaccination is preserved long term. It is the T-cells adapted to SARS-CoV-2, and not the antibodies, that are responsible for (a) clearing the infection, and (b) preventing severe disease. The fact that to a very large extent, vaccinated individuals who experience “breakthrough infections” clear their infections and avoid severe disease is, therefore, very straightforward to explain and understand.
Here, again, the typology of breakthroughs creates useful clarity, in my opinion. There is a very clear, useful, and urgent distinction between a Type 3b (“Infective”) breakthrough and a Type 1 (“Vaccine Escape”) breakthrough: In the first case the infection has progressed beyond its incipient phase by virtue of it’s very rapid reproduction rate, but it is doomed in the long term, as the cellular immune system has its number, and is moving to wipe it out; in the second case, there is no cellular response in the process of mobilizing, because the new variant has learned the new, dangerous trick of hiding from SARS-CoV-2-aware (or at least from COVID-19 vaccine-primed) immune systems. So it is actually very important to know what kind of breakthrough infections one is witnessing, and to do that it is necessary to have the right epidemic surveillance tools.
Hiding In Plain View: There Have Been No True Vaccine Escape Variants (Yet)
There are a couple of aspects of the phenomenology of the COVID-19 pandemic that are so “obvious” that they appear to have received essentially no notice beyond minimal comment by epidemiologists and infectious disease specialists, but which appear to me, in light of the present discussion to have enormous significance. I discussed these in this post. Briefly, they are (1) When a new, fast moving variant (such as Alpha, Beta, Delta, or, now, apparently Omicron) appears, it completely annihilates the previous variants from the circulating genomic landscape of SARS-CoV-2; and (2) Co-infections by multiple strains of SARS-CoV-2 are very rare, even at times when comparable prevalences of two major strains are in circulation (such as when Delta was in the process of displacing Alpha). I argued in that post that these facts should not be regarded as obvious and unconnected, but rather are profoundly significant, and explainable by a straightforward hypothesis of cross-immunity, whereby every variant of SARS-CoV-2 that has appeared to date has presented a consistent, recognizable signature to human immune recognition systems, so that when a “fast-transmitting” variant has shown up it has effectively poisoned the landscape of susceptible hosts to the previous slower-moving variant, thus driving that variant out of the genomic landscape.
Another way of phrasing this observation is that thus far, no variant of SARS-CoV-2 has managed the dangerous trick of spoofing human immune systems into believing that it is something other than SARS-CoV-2—if it had, we would probably (and dispiritingly!) refer to the new “variant” as SARS-CoV-3. The implication for the present discussion is that in fact, there has been no observation of a variant capable of creating a Type 1 breakthrough infection. And we can now understand more clearly than before what we mean when we speak of a “true” vaccine-escape variant: such a variant, being impervious to the recognition and infection-clearing processes of the SARS-CoV-2-aware cellular immune system, would circulate independently of SARS-CoV-2, and be capable of co-infection with SARS-CoV-2.
There are some subtleties that I can’t see through at the moment—for example, most vaccines teach immune systems to recognize only regions of the “spike” structure of the virus, whereas natural infections presumably teach more complete recognition patterns; and on the other hand vaccination appears to produce more robust immune response than natural infection. So discussing vaccine-primed immunity as the same thing as “SARS-CoV-2 recognition” is probably a bit naive. Nonetheless it seems incontrovertible to me that should a new variant arise, the development of a new infection pattern, consisting of frequent co-infections (which could be detected right away) followed by co-circulation by two strains (which would probably take weeks or months to establish firmly) would tell us that we’re in new trouble, that SARS-CoV-3 is here, and that a new response is required. In this connection, the recent news of early development of a pan-coronavirus vaccine seems like a very hopeful development, if it pans out.
Omicron Is A Relief From Delta
In a “Good News/Bad News” situation, it is always tempting to focus on the bad news. The fact of the matter is that based on the above discussion, the variant progression suggested that sooner or later a new, faster-moving variant would displace Delta, as Delta had displaced other variants before it, and to do that, the new variant would have to be, well, faster, which is to say, more transmissible. That’s the bad news about Omicron.
The good news about Omicron is, at this point, beyond dispute: it is a much less virulent strain of the virus than Delta. As I argued in this post, that’s actually a prediction of the “Immunosuppressed Patient Hypothesis”, the most popular theory of how Omicron developed its spectacular encrustation of mutations, which causes it to diverge much farther from existing lineages than any previously observed variant. Omicron would have needed some environment in which to develop all those mutations in such a relatively short time, and an immunosuppressed patient (of whom there are unfortunately many in South Africa due to the AIDS epidemic) would have supplied an ideal host environment. But a requirement for success would have been that Omicron “learn” not to kill that host, so that it would necessarily be the case that some of those mutations would code for more benign infection behavior.
So it now appears that the SARS-CoV-2 genome is drifting to lower virulence, although “drifting” seems like too mild a term for a process that is tearing through the whole world in about 2 months. Given the observed pattern of these variants, it seems a foregone conclusion that Delta will be gone from community spread by late January, and with it the mortality rates that we’ve come to associate with COVID-19. What’s to become of “long COVID”, and other strange, poorly-understood clinical aspect of the disease, I have no clue about. It will obviously still be a much better idea to be vaccinated than not, since who would rather have a month-long bronchitis than a 10-day sore throat/cough?
I don’t know whether or not it is reckless to call Omicron the pandemic “exit ramp”. I’d like to think that it is the exit. I am very confident that we’ll at least catch a break for a few months, after the wave crests. In fact, as of this writing, that’s already what is evident in South Africa, where Omicron originally got underway. So I am looking forward to letting my guard down a teensy bit in February.
Mutation Risk
I think that even if we found an endemic Omicron something that we were prepared to live with, it would be reckless not to press on with plans to vaccinate as much of the World as possible against it as quickly as possible, so as to attempt to minimize the chances of a true vaccine-escape mutation—a SARS-CoV-3—evolving, circulating concurrently with Omicron but capable of reinfecting everyone who was infected with Omicron, even recently, even at the same time. The urgency of this effort is not going to go away even if Omicron grants us a break, and relaxing or deferring would be tantamount to adopting a mindset like that of the hapless protagonists of horror films who are certain that they are out of danger because they can no longer see the monster. The cost of jabbing 7-8 billion arms may seem high, but the cost of not doing so is almost certainly higher. Even subtracting vaccine-resisters, we could make a big dent in this risk.
Epidemic Surveillance Tools
I touched on epidemic surveillance at two places in this series. Both times, I noted an interesting aspect of “big data” that data scientists know about and prize, but perhaps needs more advertising in fields that have not been as affected by recent developments in data science and machine learning. The fact of the matter is that feature-rich data, when made available in abundance, can have transformative impacts on the quality of our inferences, predictions, and decisions.
A very simple example: monitoring genome sequences for the emergence of a vaccine-escape variant. As I discussed above, there are two signatures to look for. If all one has is time-series of per-specimen label of most probable variant ID, which is the output of most next-generations sequencing (NGS) methods, then one must wait to see the emergence of the time-series signature of concurrent circulation of two variants. But if one has richer data at one’s disposal, say a 2-dimensional vector of prevalences per specimen, capable of representing co-infections if and when they are observed by whole genome sequencing (WGS), then one need not wait on the development of the time series: as soon as the vectors stop saying almost entirely “(1,0)” or “(0,1)” but instead start frequently saying things like “(0.31,0.69)” and “(0.83,0.17)”, you know it’s time to pick up the phone and call someone you can share a big headache with. This is what I really, really hope is happening, or going to start happening soon, with genomic monitoring, as I described in this post.
The other example is one that I have found alternatively frustrating, infuriating, and depressing, is the situation with respect to viral load, and the so-called Ct measures that could, but don’t, accompany every positive RT-PCR test for COVID-19, as I described in this post. This is a case where we could have a truly vast trove of data, a viral load number for every positive swab administered. As I discussed, histograms of such data are incredibly informative, basically because the data is much richer that a simple “Positive/Negative” value. Histograms of viral load allow us to escape the time domain in the same way that vectors of co-infection permit for genetic monitoring. The reasons given for not mandating that Ct be reported (basically, difficulty of cross-calibration of RT-PCR machines) are relevant to patient diagnosis, not to analysis of the aggregation of millions of data points for the purpose of epidemic surveillance.
Given the urgency of good monitoring, and the danger of bad decisions based on bad reasoning driven by bad information abstracted from bad data, it is simply a scandal, in my view, that because of the CDC’s complete blindness in matters of of data science, it is a bureaucratic impossibility to make “Ct with every positive test” a reality in the US. And yet, for the reasons I discussed, you cannot get there from here. So, I sincerely hope that somewhere else in the world, some health authority will demonstrate the capacity for leadership and mandate this kind of reporting from labs within its purview. The result would be to improve the ability to understand what the virus is doing, and what it will do next.
That’s The Lot
Writing up these essays has been interesting and useful for me, and helped me sort out and clarify a number of things that I can now see, in retrospect, I was a bit muddled about at the outset. I am grateful for the advice and support of friends and colleagues, a number of whom, at Argonne and elsewhere, are actual life scientists, and have helped me keep a few embarrassing errors and miscues from the text. In particular, I’d like to single out my colleagues Jim.Olds (GMU), Ben Blaiszik and Ian Foster (ANL/U Chicago), Gordon Pusch (ANL), Ira Blitz (UC Irvine), and William Parker (U Chicago).
Any remaining dumb stuff I put in myself, of course. I am rather painfully aware of the dangers of delivering myself of scientific editorials in subjects where I have no formal training. I told one of those friends recently that I’d now gotten so accustomed to the feeling of impostor syndrome that if I didn’t wake up every day to that feeling I would probably be plunged immediately into an identity crisis. I’ve written up these pieces despite that feeling, in part because I do believe that I’ve seen a few things that are worth injecting into the scientific discussion. They are nonetheless offered in the spirit of conversation-starters. I feel cheerfully undogmatic about every assertion of substance that I’ve made, although I will also offer vigorous—evidence-based—defenses of any of them. This is science, after all.