This is a guest post by Mike Hearn, a software engineer who between 2006-2014 worked at Google in roles involving data analysis.
The Daily Sceptic has for some time been reporting on the apparent negative vaccine effectiveness visible in raw U.K. health data. Despite some age ranges now showing that the vaccinated are more than twice as likely to get Covid as the unvaccinated, this is routinely adjusted out, leading UKHSA to un-intuitively claim that the vaccines are still highly effective even against symptomatic disease. A recent post by new contributor Amaneunsis explains the Test Negative Case Control approach (TNCC) used by authorities and researchers to adjust the data, and demonstrates that while a theoretically powerful way to remove some possible confounders, it rests on an initially reasonable-sounding assumption that vaccines don’t make your susceptibility to infection worse:
A situation where this assumption may be violated is the presence of viral interference, where vaccinated individuals may be more likely to be infected by alternative pathogens.Chua et al, Epidemiology, 2020
Amanuensis then compares results between the two different statistical approaches in a Qatari study to explore whether violation of this assumption is a realistic possibility and concludes that the multi-variate logistic regression found in their appendix supports the idea that viral interference can start happening a few months after initial vaccination.
What other angles can we explore this idea through? One way is to read the literature on prior epidemics.
Between 2009-2010 there was a pandemic of H1N1 influenza, better known as Swine Flu. In April 2009 a small outbreak was detected in northern British Columbia. Researchers from Canada’s public health agencies researched the outbreak by doing interviews, testing and sero-surveys of the affected population. They were especially interested in the question of how effectively the routine trivalent influenza vaccine (TIV) was protecting people against H1N1.
The effect they saw was unexpected and previously unknown: people who had taken the flu vaccine had a more than doubled chance of getting sick with flu during the H1N1 outbreak:
We present the first observation of an unexpected association between prior seasonal influenza vaccination and pH1N1 illness … participants reporting pH1N1-related ILI during the period 1 April through 5 June 2009 were more than twice as likely to report having previously received seasonal influenza vaccine.Janjua et al, Clinical Infectious Diseases, 2010
This result was shocking to the researchers. They were well aware of the impact these results could have on public support for the influenza vaccine programme and thus they didn’t merely double check their results, or request another team replicate their findings. They waited a year and a half, until six different investigations were all saying the same thing:
Canadian investigators thus embarked on a series of confirmatory studies… these showed 1.4–2.5- fold increased risk of medically attended, laboratory-confirmed pH1N1 illness among prior 2008–2009 TIV recipients… 6 observational studies based on different methods and settings, including the current outbreak investigation, consistently showed increased risk of pH1N1 illness during the spring and summer of 2009 associated with prior receipt of the 2008–2009 TIV
After the sixth study they seem to have accepted that the effect they were seeing was real.
One reason for their hesitation was that studies reported in other countries were inconclusive. Some suggested protective effects; nearly as many suggested no effect at all, and one other report showed increased risk. However, there was a very real risk of the so-called ‘file drawer’ problem, where inconvenient research simply doesn’t get published at all, and the Canadians had by this point made an enormous effort to make the conclusions go away via further research. The follow-up investigations left them with a high degree of confidence in what they were seeing, thus they explained contradictory foreign studies as being likely a result of either Canada-specific factors or flawed studies:
Findings of pH1N1 risk associated with TIV – consistent in Canada but conflicting elsewhere – may have been due to methodological differences and/or unrecognised flaws, differences in immunisation programs or population immunity, or a specific mechanistic effect of Canadian TIV. High rates of immunisation and the use of a single domestic manufacturer to supply >75% of the TIV in Canada may have enhanced the power within Canada to detect a vaccine-specific effect.
How robust is this research? This is an epidemiological study and by now it’s worth being extremely sceptical of such papers, even if they run counter-narrative. Surprisingly, this paper seems quite good. It’s not written by epidemiologists and bears little resemblence to the sort of modelling papers that now dominate policy making. In particular, it:
- Makes no predictions, only studies past events to learn from them.
- Puts actual boots on the ground to gather the data they need.
- Correlates self-reported symptoms with a sero-survey.
- Makes restrained use of statistical methods (the primary results are a standard logistic regression).
- Controls for age, chronic conditions, Aboriginal status and household density, a selection which looks reasonable (the epidemic affected an Aboriginal reserve and they differ from the normal Canadian population health wise in several aspects).
- Stratifies by age. Note that Swine Flu was the opposite of COVID: it affected the young worse than the elderly.
- Honestly discusses the weaknesses of their study, which are primarily due to the small size of the epidemic rather than anything they could have addressed.
If there are errors in this work they are of a type that aren’t easily spotted by outsiders. Although we should give a tip of the hat to this team, after reading so many absurd public health papers over the past two years it’s nonetheless hard to escape the feeling that when researchers are about to violate some tenet of vaccine dogma they suddenly become model scientists, presumably in the hope that by applying higher standards they’ll find a reason why their results are wrong.
In 2018 Rikin et al published a study in the journal Vaccine designed to solve “the misperception that inactivated vaccine can cause influenza” which was acting as “a barrier to influenza vaccination“. They concluded that the folk intuition they were fighting wasn’t actually wrong in any meaningful way, due to the presence of viral interference:
Among children there was an increase in the hazard of [acute respiratory illness] caused by non-influenza respiratory pathogens post-influenza vaccination compared to unvaccinated children during the same period. Potential mechanisms for this association warrant further investigation. Future research could investigate whether medical decision-making surrounding influenza vaccination may be improved by acknowledging patient experiences, counseling regarding different types of ARI, and correcting the misperception that all ARI occurring after vaccination are caused by influenza.Rikin et al, Vaccine, 2018
Although the paper claims that the mechanisms warrant further investigation, in reality at least one mechanism had been hypothesised as far back as 1960. In a seminal paper Thomas Francis Jr. coined the term “original antigen sin” to describe the way the immune system appears to prefer re-manufacturing antibodies for antigens similar to those it’s seen before, versus developing new antibodies customised for a slightly different invader. The odd name may be due to Francis Jr. having a Presbyterian priest as a father, thus OAS is sometimes summarised as “the first flu is forever”. This imprinting process can cause the immune system to misfire when challenged with a similar but different virus.
Some evidence for this comes from a 2017 review paper in the Journal of Infectious Diseases titled “The Doctrine of Original Antigenic Sin”, which stated:
Approximately 40 years ago, it was observed that sequential influenza vaccination might lead to reduced vaccine effectiveness (VE). This conclusion was largely dismissed after an experimental study involving sequential administration of then-standard influenza vaccines. Recent observations have provided convincing evidence that reduced VE after sequential influenza vaccination is a real phenomenon.Monto et al, Journal of Infectious Diseases, 2017
Amusingly, the paper also states that, “Hoskins et al concluded at that time that prior infection is more effective than vaccination in preventing subsequent infection, an observation that remains undisputed.” How times change.
Speculating for a moment, viral interference might explain why despite influenza vaccines being advertised as having positive efficacy multiple studies have failed to find any impact on mortality at the population level (effectiveness). For example, in 2004 a U.S. government study concluded that they “could not correlate increasing vaccination coverage after 1980 with declining mortality rates in any age group” and “observational studies substantially overestimate vaccination benefit”. This is difficult to reconcile with trials and studies showing efficacy at sizes smaller than overall population, but could be explained if vaccines merely redirect immune resources towards one pathogen away from equally dangerous variants. The same phenomenon was found in Italy.
There are also counter-studies. By 2018 awareness was growing of the problem of viral interference and the impact it can have on TNCC effectiveness metrics. In 2020 Wolff published a study of flu outbreaks in the U.S. military. It opens by confirming the problem highlighted by Amanuensis:
The virus interference phenomenon goes against the basic assumption of the test-negative vaccine effectiveness study that vaccination does not change the risk of infection with other respiratory illness, thus potentially biasing vaccine effectiveness results in the positive direction.Wolff, Vaccine, 2020
This time “receipt of influenza vaccination was not associated with virus interference among our population”. However the results of this study are rather contradictory and confusing, e.g. it also says “Examining non-influenza viruses specifically, the odds of both coronavirus and human metapneumovirus in vaccinated individuals were significantly higher when compared to unvaccinated individuals (OR = 1.36 and 1.51, respectively)”. Overall, Wolff seems to have found a mixed bag of effects in which the vaccines worked against influenza, but made some other viruses easier to catch and still others harder.
Despite the institutional pedigree of the Canadian public health researchers reporting the problem, other researchers have struggled to accept it. They are subject to the same systematic social conditioning as everyone else, which is why the HSA’s explanation of why they use the TNCC methodology starts by simply saying “vaccines work”, even though their raw data actually shows the exact opposite – for the original definition of “work”, at least.
As a consequence researchers sometimes hide this problem when it arises by deleting negative effectiveness from data sets or models. Recently UCL modellers responded to the changing UK data by simply imposing a zero lower bound. No justification was given for this, and as the above papers show, presumably no literature survey was done to sanity-check this “fix”. The Qatari study initially also did this, and thus their key results (see table 2) vary wildly between initial and final versions. Fortunately, they realised that this was not scientific and changed their approach before publication.
The problem seems to go like this: everyone knows vaccines work, thus data showing they don’t must be in error and in need of fixing. Different adjustments are tried for confounders (sometimes real, sometimes hypothetical) until the data comes good, at which point the results are published and the idea that vaccines work is reinforced, leading to a greater propensity to view opposing data as flawed and in need of correction… ad infinitum.
The raw data now departs so seriously from the conclusions drawn from it that it would require a staggeringly huge behavioural change between the two camps to explain, one which stretches credulity past breaking point. The argument that the data requires adjustment/replacement due to speculated behavioural differences has another problem: that’s a sword that cuts in both directions. UKHSA is keen to stress that its raw data shows some effectiveness against hospitalisation. But that data is hopelessly confounded at this point by the fact that vaccine recipients are being told, in no uncertain terms, that while they might well get sick with Covid after taking it, the vaccine means their case won’t be “severe” and they definitely won’t need to go to hospital. “Severe” is a vague standard. Because Covid has a wide range of severities there will be many borderline cases where going to hospital is effectively a choice that could go either way.
Opinion polling shows consistently that governments and media have catastrophically failed to educate the population about Covid correctly: people routinely estimate that the unvaccinated infection:fatality ratio is orders of magnitude higher than it really is. In a recent French survey the population estimated the IFR at an astounding 16% (the true level is closer to 0.1%-0.3%) and their understanding of severity has got worse over time. If you previously believed that you had a 16% chance of dying if you got Covid, you were very likely to rush to hospital immediately on presentation of more or less any Covid-like symptoms. If you now believe that the vaccine reduces this risk to negligible levels then you’re very unlikely to bother unless you become quite seriously sick indeed, because to do so would effectively be a repudiation of the advice of government, scientific and medical authority. And if there’s one behavioural difference between the vaccinated and unvaccinated that is more plausible than any other, it’s that the vaccinated are self-selecting for strong faith in scientific claims by authority figures. I’ve not yet seen any recognition by public health that this confounder exists – they are literally telling people what to do, and then declaring victory when people do it. If hospitalisation was 100% a force of nature that involved no element free will this wouldn’t matter, but the 50% drop in A&E admissions at the start of lockdown showed quite clearly that it’s not.
Negative effectiveness is important because if a vaccine halves your risk of getting one virus but doubles your risk of getting a closely related virus, you can end up back at square one. In fact, you’d end up in a worse position than when you started because vaccination programmes aren’t free: they consume enormous resources, both financially and in terms of public health staffing, and cause collateral damage via vaccine injuries (hence why vaccine manufacturers refuse to accept liability for harm caused by their products). It’s therefore of critical importance to understand the gestalt effect of vaccination on the immune system, and not merely on the specific variant of a virus that was originally targeted.
The fact that papers published as recently as 2018 are talking about negative vaccine effectiveness as a new, not really understood effect should give governments serious pause for thought. Most people in public health are clearly unfamiliar with this phenomenon – as indeed we all are – and are thus tempted to either ignore it, delete it from their data, or try to convince the public that it must be a statistical artefact and anyone talking about it is guilty of spreading “misinformation”. The reports in these papers provide recent evidence that vaccines making epidemics worse is in fact a real phenomenon and that it has been previously detected by serious researchers who took every effort to avoid that conclusion.
Nonetheless, despite my harsh words about IFR education above, we must acknowledge that the UKHSA is so far standing by the basic moral and foundational principles of public statistics. Their answer to the confounders and denominators debate is clearly written, straightforward, reasonable and ends by saying:
We believe that transparency – coupled with explanation – remains the best way to deal with misinformation.
That’s absolutely true. The deep exploration of obscure but important topics by independent parties is possible in the U.K. largely because the HSA is not only publishing statistics in both raw and processed forms, but has continued to do so even in the face of pressure tactics from organisations like Full Fact and the so-called Office for Statistical Regulation (whose contribution to these matters has so far been quite worthless). England is one of the very few countries in the world in which this level of conversation is possible, as most public health agencies have long ago decided not to trust the population with raw data in useful form. While the outcomes may or may not be “increasing vaccine confidence in this country and worldwide”, as the HSA goes on to say, there are actually things more important than vaccines that people need confidence in – like government and society itself. Trustworthy and rigorously debated government statistics are a fundamental pillar on which democratic legitimacy and thus social stability rests. Other parts of the world should learn from the British government’s example.
Many questions now lie open:
- To what extent does negative effectiveness require viruses to be different? For example, is the difference between H1N1 and the flu strains targeted by the Canadian TIV bigger, smaller or the same as the gap between COVID Alpha and COVID Delta, as perceived by the immune system?
- Although highly suggestive, is this genuinely happening with COVID vaccines, or is raw negative effectiveness due to something else, e.g. a temporal artefact caused by splitting waves into two overlapping waves as effectiveness wears off, or indeed, due to lack of adjustments for factors that TNCC fixes even though it may introduce other problems?
- Should this cause health authorities to abandon TNCC as a methodology, despite its speed and cost advantages?
The fact that TNCC can artificially make vaccines appear more effective than they really are, and that this would actually have happened during the Swine Flu pandemic, should really be addressed at the highest levels before anyone uses terms like “misinformation” again.
Thanks to Amanuensis and Will Jones for their review.