In the early 20th century, Kermack and McKendrick (1927) developed a useful mathematical model for describing the progression of an epidemic, called the “Susceptible/Infected/Recovered (SIR)” model [74]. This model is a set of mathematical equations (differential equations) that compartmentalises the population into susceptible (S), infected (I), and then recovered (R) or dead. It offered an explanation as to how an epidemic of a highly infectious disease could end before everybody in the community had been infected [74].

Although the original SIR model was proposed in 1927 in the pre-computing era, it is also very amenable to being solved by more modern analytical methods [45, 61, 62]. It also could easily be adapted to include additional compartments such as an exposed (E) population that was not yet infectious, leading to the related SEIR model. Hence, the SIR framework, or its derivatives, still dominate the field of epidemic modelling. Indeed, most of the COVID-19 models published during the pandemic used some implementation of this SIR/SEIR framework [75] including those used to advise governments [39–46].

Some COVID-19 models also used computationally expensive agent-based models based on the SIR/SEIR framework, e.g., UK and USA [6, 43]; or Austria [44], but most studies used the simpler population-averaged models, e.g., USA [76]; Canada [77]; Ireland [45]; Spain and Italy [78].

In Figure 1, we explain the main features of the basic SEIR framework and demonstrate how it led to alarming initial projections [39–46, 76–78] when applied to the COVID-19 pandemic:

a. The default SIR/SEIR framework assumes that 100% of the population starts in the susceptible compartment but as the epidemic progresses everybody exposed to the virus will pass through each of the compartments until they have recovered (or die): S→E→I→R.

Even now, it is still unclear what is the average R ₀ of COVID-19 [79, 80]. However, many of the early estimates suggested values in the range 2–6. Therefore, as can be seen from Figure 1F, the implications of the model projections were highly alarming. They projected that–in the absence of NPIs or a population-wide vaccination programme–there would be a major pandemic wave that would only take a few months to infect more than 70%–80% of the population. And at its peak, up to 10%–25% of the population would be sick at the same time–overwhelming health services, as well as devastating society.

When we are familiar with the implications of this model, we can understand why so many modellers were alarmed [39–46, 76–78] and why policymakers advised by these modellers could easily have shared this alarm. However, while the original SIR model remains a powerful useful epidemiological tool [61, 62], especially for highly infectious diseases that are predominantly spread through close contact, e.g., measles, there are multiple reasons why these simplistic projections were unrealistic for modelling the COVID-19 pandemic [39–42, 46, 51, 61, 63–66, 81].

Separate problems with how the models were calibrated and evaluated arose due to the inconsistencies in how the incidence of COVID-19 in the community was measured over time. Most models were fitted using either COVID-19 cases or COVID-19 deaths. However, the methods by which both of these metrics were measured changed throughout the pandemic [110–117].

In terms of “COVID-19 cases,” the proportion of the population tested, the priority of testing and the case definitions for COVID-19 changed dramatically over the course of the pandemic [110–116]. In the beginning of the pandemic, RT-PCR tests to reliably identify SARS-CoV-2 were still being developed [118] and testing capacity was limited. Hence, testing was often prioritized for patients with the most severe symptoms or healthcare workers (to reduce nosocomial spread in hospitals) [111, 119]. The case definition was also initially stricter and based more on symptoms than test results [113]. Therefore, the identified “cases” were initially skewed by the “clinical iceberg” effect [119, 120], i.e., many of those who were infected may have gone undetected because they did not present to a doctor for diagnosis and treatment.

This led to two major biases in the modelling of the early waves. First, the true number of infections during the first wave was substantially underestimated [111, 113–116]. It also meant that the early estimates of the infection fatality rate (IFR) and hospitalization risks were too high [1, 63, 111, 112, 121], since most of the infections with mild or even no symptoms would have remained unidentified [67].

As testing capacity increased (and also demand temporarily lessened) after the first wave, the case definitions were loosened [113] and testing priorities expanded. Hence, since mid-2020, the more number of cases identified was partially a function of testing capacity. The more tests undertaken, the more cases could be identified. This does not mean that the number of cases was merely a function of testing capacity. The number of tests carried out was a function of both the supply of tests and the demand for tests [122–124]. However, it meant that case numbers were an unreliable metric for studying the course of the pandemic.

Additional problems in the use of “case numbers” arose because COVID-19 testing laboratories often used very high cycle thresholds (Ct) of 40 cycles or higher in the RT-PCR test to reduce the possibility of giving a false negative to a sample from a person at the early stages of infection, i.e., when the viral load was still very low. This was useful in terms of containment strategies because the infectious period for COVID-19 seems to begin before or at the time of symptom onset [125], i.e., before the infected person realises they are sick.

The problem was that “case numbers” conflated both high-Ct and low-Ct positive specimens as identical, meaning that many of the identified “cases” were neither infectious nor symptomatic [126–130]. Hence, since mid-2020, many of the cases might have been completely asymptomatic and non-infectious, yet still treated identically to symptomatic and infectious cases [127, 128]. Some solutions might have been to either (a) include the Ct for a positive result [130]; (b) revert to a case definition that also required symptoms [113]; (c) provisionally treat a high Ct “positive” as possibly being pre-symptomatic (and advising quarantining), but following up with the person before confirming them as a “case”.

An alternative metric sometimes used for evaluating the pandemic progression was the number of COVID-19 deaths [117]. However, this metric also was surprisingly inconsistent, because some patients who had tested positive for COVID-19 but died of other causes were still counted as a “COVID-19 death”. For example, in the UK, deaths were recorded for people who died up to 28 days of having a positive COVID test or if COVID-19 was entered on their death certificate [117]. Therefore, “COVID-19 death” statistics combined together “deaths from COVID-19,” “deaths where COVID-19 was a contributing factor” and “deaths with COVID-19,” i.e., patients whose death had nothing to do with COVID-19 but who had coincidentally tested positive for COVID-19. This conflation of different causes of deaths into “COVID-19 deaths” has made it challenging to use these mortality statistics as a reliable metric for tracking the pandemic.

Meanwhile, testing of all hospital patients rapidly became routine to control nosocomial infections. This was understandable from a healthcare perspective, but it often led to confusion over the pandemic progression, because hospital patients who tested positive for COVID-19 while in hospital were typically counted among the “hospitalized with COVID” statistics even if the reason for hospitalisation had nothing to do with COVID-19 infection [131–133].

As well as leading to uncertainties in (a) the severity of COVID-19 morbidity and mortality risks, and (b) the relative magnitudes of the different waves of the pandemic over time, the above inconsistencies also would have introduced real-time errors into the data inputted into the models used for advising governments on how the pandemic was progressing in each country. Therefore, there are multiple reasons for strengthening the surveillance of infectious diseases.

A major concern with the over-reliance of policymakers on model scenarios is the lack of mechanisms implemented to try and assess the accuracy (or otherwise) of the models as the pandemic progressed [39–42, 46, 51]. This was especially necessary given that some models provided a range of plausible scenarios and/or provided less alarming scenarios [51, 63–68]. Hence, the most alarming scenarios were not necessarily the most plausible, yet the less alarming scenarios were apparently given less weight [69–72] than the more alarming ones [43, 47, 48, 134, 135].

Policies were determined based on what models projected would happen in the absence of those policies, i.e., essentially the dramatic scenarios outlined in Figure 1. However, typically, modellers did not attempt to model what would occur with those policies. Hence, once the policies were implemented, the model scenarios could never be tested by either the modellers or the policymakers, i.e., they were “non-falsifiable” [136].

This put policymakers in an unfortunate position. They were warned by modellers that if they did not implement major unprecedented policies, the consequences would be dire. Yet, once they implemented the policies, they were unable to assess how realistic those warnings had been. Nor had the modellers any feedback to establish whether they needed to modify their models going forward.

Meanwhile, most of the studies that retrospectively evaluated COVID-19 policies as relatively successful [7, 10, 12, 55–59] in reducing the spread of COVID-19 did so using so-called “counterfactual scenarios”. That is, researchers would compare the observed COVID-19 statistics following the introduction of the policies to the modelled scenarios of what might have happened in the absence of the policies. This comparison of an observed reality to a hypothetical “counterfactual scenario” was used to claim both NPIs [7, 10, 12, 55–58] and vaccination programmes [59, 60] were effective.

However, as we will discuss in the next sections, assessments that were not solely based on counterfactual scenarios often found that the progression of the pandemic was largely independent of government measures [64, 81, 96, 103, 105, 137–145].

Indeed, in a few rare cases where the policies recommended by the model scenarios were not implemented, thereby allowing a comparison of the model scenarios to the observed reality, the model scenarios were widely wrong [134, 135].

For instance, Sweden was somewhat unique in choosing to not adopt the wide range of NPIs implemented by neighbouring countries [31–33]. While many in Sweden may still have voluntarily modified their behaviour [32, 141, 142], these voluntary measures were widely regarded as less restrictive than mandated NPIs [31–33]. It therefore offers a rare counter-example for evaluating the effects of the NPIs implemented by neighbouring countries. Using Imperial College London’s own model, researchers estimated 34,895 first-wave deaths in Sweden under “social distancing of the whole population” — the most stringent measure short of full lockdown, and arguably the closest to the measures actually implemented in Sweden [134, 142, 146]. By the end of July 2020, the actual number of deaths reported in Sweden was 5,741 [31].

More broadly, the countries that implemented the least stringent NPIs did not generally experience more COVID-19 deaths than the countries with the most stringent NPIs [145, 147]. This is the opposite of what the models had predicted.

Lastly, after repeatedly following the model-based recommendations to implement strict NPIs for most of the pandemic, in December 2021, the UK government finally decided to overrule the model-based recommendations that were calling for a (fourth) lockdown in December 2021. On that occasion, the model predictions of deaths exceeded the actual numbers by a factor of 20 [70].

Before 2020, pandemic preparedness plans had been carefully prepared and they advised against most of the NPIs [49] that ended up being implemented. In particular, a systematic WHO review in autumn 2019 concluded that the evidence for the effectiveness of most NPIs was limited [49]. Additionally, the limited data in their favour was mostly based on either observational or modelling studies [49]. It was also predicted that NPIs might also have considerable harms [49]. These predictions were later confirmed at the end of the COVID-19 pandemic by a UKHSA evidence review of studies conducted in the UK [160]. Yet these pandemic preparedness plans that had been developed in advance were apparently abruptly abandoned with little discussion [161, 162].

Even during the early months of the pandemic, the WHO Strategic and Technical Advisory Group for Infectious Hazards (STAG-IH) recommended a range of measures for countries to prepare and respond to the pandemic that did not involve most of the NPIs that were ultimately used [163]. So, while the implementation of an unprecedented array of NPIs was frequently promoted as “following the science” [11, 48, 164–166], the reality was that most of the science that had been available at the time was hastily replaced on the basis of the alarming model projections [39–46, 76–78] described earlier.

During the pandemic, while the NPIs were in play, many researchers called on the community to reconsider this strategy and proposed alternative approaches to managing the pandemic going forward [54, 137, 149, 150, 157–159, 167, 168], yet these calls were either ignored or dismissed without adequate consideration [102, 168–173]. For instance, when an eminent epidemiologist wrote a generally favourable assessment in May 2020 of how Sweden had fared without the more stringent NPIs of its neighbouring countries [167] it led to six separate critiques disputing this assessment [102, 168–173]. Although Sweden did experience a relatively severe first wave due to a high death toll in elderly care services [174, 175] ultimately, it did not experience a noticeably more severe pandemic than its neighbours over the long-term [31–33, 96, 142].

In late 2020, three prominent epidemiologists wrote the Great Barrington Declaration (https://gbdeclaration.org/), advocating for a more “focused protection” strategy that prioritised the most vulnerable to COVID-19, particularly the elderly [176], rather than the diffuse strategies of population-wide NPIs that they were “gravely concerned” were leading to “damaging physical and mental health impacts”. At the time of writing, the declaration has had 941,261 signatories including 16,176 medical and public-health scientists and 47,839 medical practitioners. Yet, rather than leading to a public discussion over whether alternative strategies could be adopted, the declaration was immediately dismissed as allegedly promoting “potentially dangerous fallacies” [102]. Behind the scenes, the then-director of the National Institutes of Health apparently tried to organise “a quick and devastating published take down of its premises” because it “seems to be getting a lot of attention” [177].

We appreciate that governments were trying to respond quickly to the modelling advice provided to them. However, we believe that if governments had taken seriously existing pandemic strategies [49, 163] and/or sought feedback from researchers with multiple perspectives [54, 137, 147, 149, 150, 157–159, 167, 168, 178], they could have formed more well-rounded, evidence-based strategies than they did.

Given the speed of the adoption and implementation of NPIs, it is conceivable that governments neglected the proper critical appraisal of their effectiveness. This is perhaps reflected in early studies which suggested benefits of NPIs throughout 2020 and much of 2021, based on counterfactual scenarios [7, 12, 57, 179]. However, as discussed earlier, once the NPIs began being implemented, there seemed to be insufficient interest from policymakers in encouraging research to critically re-evaluate the effectiveness of the NPIs as more data accumulated. Additionally, governments likely felt a pressure to “do something” and to “keep up with” the more stringent NPIs implemented by their neighbours [112, 161, 162].

Governments appeared content to rely on model-based studies that concluded the NPIs must collectively be working whenever a pandemic wave was in decline [7, 10, 12, 55–58]. However, studies that looked closely at how the timing and magnitude of the NPIs in different countries compared to the dynamics of the pandemic in those countries failed to identify a clear or consistent influence of the NPIs [64, 81, 96, 103, 105, 137–145].

For instance, several studies found that COVID-19 waves had often peaked before the NPIs had been implemented [68, 103–105, 139]. This suggested that the rises and falls in viral incidence were largely independent of the stringency of NPIs. Indeed, retrospective analysis of sewage samples suggested that, in some countries at least, the disease may already have been present months before NPIs were even considered [180, 181]. Instead, several studies have suggested that the natural seasonality of human coronaviruses, reaching peaks in winter months and lows in summer months [82, 85–91], may have played a much greater role in the pandemic dynamics than the NPIs [81, 88, 91, 96–98].

Another challenge in assessing the effectiveness of NPIs was that, for much of the pandemic, governments typically simultaneously implemented a diverse array of completely different NPIs. This often made it very difficult to empirically isolate the relative effectiveness of any individual NPI. We note that many of the modellers, who argued that collectively the NPIs were effective, also shared this frustration–see the perspective review by Lison et al. [182]. We agree with Lison et al. that the simultaneous implementation of multiple different NPIs by multiple neighbouring countries severely hindered the scientific community’s ability to assess the relative effectiveness of individual measures [182].

We appreciate that many of the NPIs might intuitively feel like they should have been having a significant influence on the pandemic progression. However, scientific analysis often contradicts our intuitions.

Let us consider the wearing of masks, as a case study, since NPIs from mid-2020 often involved mask requirements [9, 145]. Mechanistic studies suggested that masks could potentially reduce the transmission of viral particles [180, 181]. However, meta-analyses of studies before the pandemic had failed to identify a statistically significant reduction in the spread of viral transmission for influenza [183]– even if some had identified a non-statistically significant possible effect [184]. Indeed, a randomized control trial in Denmark during early 2020 found that 2.1% (53/2,994) of the control group and 1.8% (42/3,030) in the mask-wearing group caught COVID-19 before the trial had to be discontinued as the government had introduced mask-wearing regulations [185]. Again, this difference was not statistically significant, and suggested that any effect was, at best, modest.

Including the Danish study, three randomized control trials (RCTs) have now assessed whether facemasks and respirators were effective in preventing COVID-19 transmission. In two of these studies, surgical or cloth masks were investigated [185, 186] and in the third study, N95 respirators were compared with medical masks in a multicentre, randomised study of health workers who had direct contact to patients with suspected or confirmed COVID-19 [187]. A Cochrane meta-analysis showed that, in conjunction with previous studies, these trials failed to demonstrate that masks significantly reduced viral transmission in the community or among healthcare workers [188]. In fact, high-quality data from randomized trials consistently failed to demonstrate a significant effect of masks on viral transmission, while evidence supporting the beneficial effect of masks was derived almost exclusively from lower-quality observational studies [189]. Additionally, a comparison of 35 European countries during the 2020–2021 winter failed to identify a statistically significant relationship between mask usage and COVID-19 outcomes [145].

Public health involves much more than dealing with a pandemic [61, 147, 153, 190]. Yet, throughout the pandemic, many long-standing public-health policies and strategies from the pre-pandemic period [147, 153, 154, 156, 190] were abandoned or severely deprioritised to focus almost exclusively on just one public-health issue, i.e., minimising the spread of the SARS-CoV-2 virus [42, 147, 149–153, 155, 156]. Therefore, although NPIs were designed with a public-health goal of reducing the health burden of the pandemic [7, 102], public-health policies should have explicitly weighted the potential benefits of the proposed NPIs against their many unintended harmful consequences, not only on public health, but also on the wider society, economy, and natural environment [42, 147, 149–156].

While the NPIs often led to a mixture of beneficial as well as adverse consequences for different people and groups, the net consequences were often adverse, e.g., see ÓhAiseadha et al. [153] for an extensive review.

The harmful consequences of the NPIs included stress [191], adverse changes in diet, nutrition, body weight and obesity [192, 193], substance abuse [194, 195], tobacco smoking [196], emotional and mental health impacts [197, 198], impaired healthcare delivery [199], adverse economic, social and environmental impacts [200], interruptions of education [198, 199], reduced hospital attendance [155, 156, 201], reduced vaccine uptake [202, 203] and impaired care and health of the elderly [149, 204, 205]. The adverse impacts on lifestyle and population health were exacerbated by greater health inequalities according to age, gender, socioeconomic status, pre-existing health and location [153].

The harmful impacts of NPIs for body weight and obesity are particularly noteworthy in that there is some evidence for an association between overweight/obesity and an increased risk of long COVID [206, 207]. As mentioned in the Background, the possibility of long COVID was one of the concerns associated with the pandemic (after the risks of hospitalization or death) [3].

Therefore, we believe that most of the NPIs that were implemented during COVID-19 should be avoided, if possible, in future pandemics. If any of these measures are to be considered again, governments should ensure that this is in conjunction with rigorous and holistic cost-benefit analyses. Because of the far-reaching impacts of the NPIs, a proper impact analysis should also involve a more multi-disciplinary range of scholars from the social sciences and the humanities as well as public-health practitioners [147, 178].

From the start of the pandemic, physicians were told that no anti-coronaviral therapy existed and that healthcare workers were recommended to provide “supportive care only” and that other therapies should be avoided outside of randomized controlled trials [211–213]. The WHO explicitly prohibited the use of corticosteroids outside of clinical trials, until September 2, 2020 [221], when the WHO switched to recommending corticosteroids for severe and critical patients. Within hospital, supplemental oxygen therapy and potentially mechanical ventilation could be considered if necessary. However, patients that did not require hospitalization were typically told to rest at home without treatment, but to “return to hospital if they develop any worsening of illness” [222].

Given the explicit absence of any treatment options, it is not surprising that some physicians and researchers began looking at the possibility of developing potential protocols for treatment and/or prevention by repurposing promising drugs [223, 224] – preferably well-studied and affordable candidates with known safety profiles [209–213, 223–230]. What was surprising was the strongly hostile over-reaction from the medical community and health authorities whenever a potential protocol was identified as promising [210, 231, 232]. This was often accompanied by media campaigns to create the public impression that these recently-proposed protocols were not simply ineffective, but potentially dangerous (even if the protocol had only recently been proposed and it was based on widely used drugs), and that physicians that were considering these protocols were therefore behaving dangerously and recklessly [233]. A recurring theme used in media campaigns to discredit these drugs, that are so widely used which they have applications for both humans and animals, was to emphasize the veterinary usage, hence creating the false impression that the drugs might not be suitable for human usage [234, 235].

Probably the most high-profile potential protocols were those that included either hydroxychloroquine (HCQ), ivermectin or anti-inflammatory medications such as corticosteroids (dexamethasone) [209, 210, 213, 223, 224, 227–230]. These drugs had been widely used by the medical community before the pandemic and therefore their safety profiles were well understood, making them promising candidates for repurposing [223, 224]. Several promising protocols involving their use were suggested as a “short-term option for the early treatment of most symptomatic high-risk outpatients” [209, 226]. Different studies of HCQ have given conflicting results, with some studies finding no mortality benefits to the use of HCQ [236, 237] and others finding a statistically significant mortality benefit [238]. However, on the basis of the former, in June 2020, the US FDA rescinded its temporary emergency use authorisation for HCQ [239]. Physicians who attempted to provide patients with access to protocols involving HCQ–even as a “right to try” – were either stonewalled or faced professional censure [210–212, 231–233].

In contrast, even though the Solidarity Therapeutics Trial coordinated by the WHO found in October 2020 similarly negative results for both HCQ and remdesivir (a relatively expensive drug still under patent by Gilead Sciences) [237], the WHO still allows the use of remdesivir in certain circumstances [221].

Another proposed repurposed drug, ivermectin, had been in use for several decades as a safe, inexpensive antiparasitic drug [225]. Ivermectin already demonstrated antiviral activity against other RNA viruses including HIV, influenza A and SV40 (DNA virus) in lab tests by a host-directed nuclear import protein inhibitor (IMP) [229]. Again, different studies have given conflicting results on its efficacy [225, 228, 240–243]. However, the tested protocols have not shown any toxicity issues [225, 228, 240, 241]. Moreover, one double-blind, randomized clinical trial of ivermectin reported it to be a potentially safe and effective medication for COVID-19 patients with moderate disease [242] and another review and meta-analysis states that “ivermectin could reduce the risk of mechanical ventilation requirement and adverse events in patients with COVID-19, without increasing other risks. In the absence of a better alternative, clinicians could use it with caution” [243].

A detailed review on the relative effectiveness of the various protocols involving HCQ, ivermectin or any other cheap repurposed drug is beyond the scope of this article–indeed several of our co-authors have different views on these ongoing debates. So, in this current article, we are not necessarily drawing any definitive conclusions as to their effectiveness. Nor are we saying that treatments needed to be based on cheap drugs. Indeed, monoclonal antibodies were granted EUA in November 2020 by the FDA in the USA and are used to treat and also detect COVID-19 [244].

However, all of the co-authors are alarmed at the manner in which research into the potential use of protocols involving the use of widely-used cheap repurposed drug was not just discouraged but vilified. Even if none of these protocols had been effective, given the major worldwide usage and well-understood safety profiles of many of these drugs before the pandemic, open-minded exploration into their potential should have been welcomed [209, 210, 225], rather than attacked [210, 226, 231, 232]. Especially since most patients were not provided with any alternative treatment–other than, from 2021, hoping that the mRNA/DNA vaccines that we will discuss below would help.

For decades before the pandemic, major public campaigns to encourage various population-wide and individual vaccine programmes, coupled with the labelling of opposition to any vaccine as “unscientific” and “anti-vax”, had led to a common public perception that any vaccine that was offered to the public by health authorities should automatically be trusted as being “safe” and relatively “effective” and based on well-established science [245–247].

This meant that when the various “mRNA vaccines” and “DNA vaccines” introduced in late 2020 were described as “vaccines”, many people–including high-profile doctors and scientists that took and promoted the mRNA/DNA vaccines–did not realise until much later that these new vaccine technologies were not the same as traditional forms of vaccination and had not previously been used for any public vaccination programmes before 2020 [218].

From a marketing perspective, this remarkable public trust in the term “vaccine” was very useful for developers of these new technologies (e.g., Moderna, AstraZeneca, Pfizer), who had been trying to develop an mRNA or DNA “vaccine” that would be commercially relevant for more than a decade [248, 249] It also seems to have contributed to many people (including researchers and health authorities) mistakenly conflating concerns over these specific vaccines with general “vaccine hesitancy” [250].

However, mRNA and DNA vaccines were very different from the conventional vaccines the public had been used to up to 2021. Traditional vaccination strategies use the whole or part of an inactivated or weakened infectious agent to stimulate the immune system into providing protection against further attack. In contrast, the theory behind these new mRNA/DNA vaccine technologies is to use DNA or mRNA to instruct human cells to produce part(s) of the viral protein which are subsequently displayed on the outside of the cell membrane. The idea is to mimic a part of the infective process of the virus. In theory, the human immune system should then recognise viral protein as non-self or foreign, try to destroy the cells that display it, and then create antibodies that can recognise it in future encounters [251].

Although the technology that these genetic vaccines were derived from promised much for the biotechnological industry back in the early 2000s, its regulatory approval was delayed for more than a decade due to significant adverse events. To overcome this hurdle, the genetic code of parts of the virus (that would act as a vaccine) was extensively modified so that it could bypass the human innate immune system [251] paving the way for repurposing this technology for the new vaccines. However, there are still concerns that this new type of vaccine technology was only approved by emergency use authorisation (EUA) via suspension of normal testing processes and review, rather than the more fitting regulatory approval as gene therapy [252].

However, COVID-19 vaccines were also developed and received WHO approval using more conventional processes especially in China and India, e.g., Sinopharm, Sinovac and Covaxin [34, 35, 253] and these were used in many countries, especially in the developing world–see Figure 2. In many countries, only the mRNA or DNA vaccines were accepted as valid in terms of COVID-19 vaccine regulations [219].

We believe that if the “mRNA/DNA vaccines” had been labelled using a terminology to better indicate (a) their novel (and relatively untested) genetic nature [252, 255–261] and (b) that they were not the traditional “vaccines” that people were familiar with, this would have allowed health authorities to better assess their suitability for public-health purposes and help the public make their decisions with informed consent.

Reports of the phase 3 trial results for the various COVID-19 vaccines claimed that each vaccine was safe (as well as efficacious) [13–15, 34–37]. This was typically achieved by noting that the incidence of severe adverse events and/or fatalities were (a) rare and (b) similar in both the vaccinated group and the control group. On this basis, governments began population-wide vaccination programmes in early-2021.

The trials for evaluating safety of each vaccine were relatively short (often <2–3 months) and at the end of the trials, participants in the control groups were unblinded and offered the vaccine–thus preventing continued evaluation of the groups over longer periods. This is despite the fact that vaccines require a very high safety standard because they are administered to healthy individuals [249, 291] and because adverse events might not be confirmed until years after the roll-out [291]. The brevity of these safety trials was particularly concerning in light of the novel nature of the mRNA and DNA technology [257–261].

As the vaccination programmes progressed–and towards the end of 2021 as they switched to booster programmes involving third or subsequent doses–the narrative of what was “safe” seems to have progressively evolved, as the estimated incidence of concerning adverse events increased by orders of magnitude over time as we will discuss below.

Typically, the frequency of side effects associated with a medication are commonly annotated as follows:

• “Very rare” denotes a side effect that occurs in less than 1 in 10,000 people, i.e., <0.01%

In early 2021, concerns initially focused on “rare” cases of serious blood clotting events associated with two of the DNA vaccines–AstraZeneca and Johnson & Johnson–especially among women [292–294]. More than 30 of the first 222 suspected cases in Europe subsequently died [292].

In parallel, concerns over the mRNA vaccines initially focused on cardiovascular issues chiefly, myocarditis/pericarditis. Although early assessments based on passive adverse event reporting systems initially suggested these events were “very rare” [295], it soon became apparent that the rates were higher among younger men [296, 297] and increased substantially from the first to second dose [297–299].

In the United States in June 2021, the Advisory Committee on Immunization Practices (ACIP) debated the benefits/risks of continuing the programmes for young adults. They explicitly assumed that the vaccines were 95% effective “in preventing COVID-19 cases and hospitalization” [300]. Hence, they calculated that the benefits outweighed the risks [300]. As we saw above, such optimistic assumptions of vaccine effectiveness were soon abandoned.

Some studies conceded that the mRNA vaccines led to an increased risk of myocarditis and pericarditis, but argued that the risks were higher from COVID-19 infection [270]. It was later realised that such estimates were flawed, because the numbers of COVID-19 infections during the first waves had been severely underestimated and the risks from infection had been overestimated [301]. Studies that avoided this problem confirmed that the risks of myocarditis and pericarditis were much higher from vaccination than from infection. For example, between May and October 2021, 32 million people aged 12 to 50 were vaccinated with the mRNA vaccines in France and 3,225 members of that population developed myocarditis or pericarditis over the same period (i.e., 0.01%), but 97% of these cases were due to the vaccines, not infections [299].

By 2022, studies of vaccine-induced myocarditis/pericarditis that were stratified by sex, age, dose number and vaccine brand found much higher estimates of the incidence of this condition, at greater than 1.5 in 10,000 (“rare”) among young males after the second dose [301].

More recently, several high-sensitivity observational studies of vaccination programmes within individual institutions have indicated that cardiovascular manifestations after the second dose are “very common”, but usually “mild and temporary” [302, 303]. Participants with either abnormal ECGs or elevated cardiac biomarkers were “common” [302–304], but all patients in these studies had fully recovered by the end of the study [302–304]. These studies suggest that the incidence of mild and transient myocarditis and/or pericarditis is not “rare”, but rather “uncommon” [302–304].

As time has elapsed, new information has raised concerns about how accurately the safety data during the original clinical trials had been reported and interpreted. A reanalysis of the Pfizer and Moderna clinical trial data suggested that there were 0.10%–0.15% serious adverse reactions relative to the placebo arm [305], i.e., “uncommon” rather than “very rare” as originally claimed.

Meanwhile, an analysis of suspected adverse events following Pfizer vaccination in Denmark suggested that some vaccine batches had much higher rates than others [306]. That said, this study was controversial and disputed on multiple fronts [307–310]. So, while many of these criticisms have been responded to [306], the results should be treated cautiously. Nonetheless, subsequent studies have found similar findings for Sweden [311] and the Czech Republic [312], suggesting that this controversial finding may require further investigation. If the finding turns out to have some validity, the potential inconsistencies between batches could increase the difficulties in accurately estimating the risks of adverse reactions from these vaccines.

In November 2021, a former regional director of one of the Pfizer clinical trial sites provided whistle-blower testimony alleging serious concerns over data integrity and regulatory oversight in the sites she was working at [268, 269]. This included concerns that the adverse reactions of some trial participants were not being adequately followed up. Indeed, Brianne Dressen (AstraZeneca trial participant) and Maddie de Garay (Pfizer 12–15 years old trial participant) have both publicly spoken about how:

(a) Their severe adverse reactions were omitted from the papers describing the trial results.

Only 1,131 participants had been in the vaccine arm of Maddie de Garay’s clinical trial [313], so her case accounted for 0.089% of vaccinated participants. Yet, neither of these two trial participants were mentioned in the studies describing the safety and efficacy of the relevant vaccines [14, 313].

As the estimates of the rates of adverse reactions increase over time, the range of possible adverse reactions confirmed to be associated with the mRNA and DNA vaccines also continue to increase: e.g., Guillain-Barré syndrome [314]; cerebral venous sinus thrombosis [314]; encephalomyelitis [314]; psychiatric adverse reactions [315] menstrual irregularities [316, 317] tinnitus [318] and shingles [257]. In some cases, at least, the vaccines have been linked to fatal outcomes [319, 320].

We also note that some studies have found instances of the genetic material of the vaccine or spike protein fragments remaining in circulation for months [257, 321–323]. For example, one study detected recombinant spike protein fragments in the blood of patients 187 days after vaccination [321]. Another noted the presence of mRNA concentrated in extracellular vesicles of human breast milk [322]. This is concerning as it was initially assumed that this genetic material and its protein product would be short-lived [249].

Rasmussen Reports have carried out a series of polls of the general public in the United States that have included questions on the COVID-19 vaccines. Below, we list a few recent examples, but see https://www.rasmussenreports.com/search?SearchText=covid for the most recent results:

• May 20–22, 2024 (N = 1,250): “Do you know someone personally who died from side effects of the COVID-19 vaccine?” 19% yes; 74% no; 7% not sure. [For comparison, the results for “Do you know someone personally who died from the COVID-19 virus?” were 42% yes; 53% no; 5% not sure.]

Obviously, such polls only capture personal non-expert opinion, but they indicate that (rightly or wrongly) a sizeable minority of the public personally believe that they or people they know suffered a severe adverse reaction from the COVID-19 vaccines.

Finally, there are still concerns over significant increases in excess mortality during the period associated with the vaccine roll-out compared to either the pre-vaccination period of the COVID-19 pandemic (2020) or pre-pandemic years [261, 324–326]. Indeed, some studies comparing all-cause mortalities between vaccinated and unvaccinated groups found slightly increased risks among vaccinated groups compared to the unvaccinated [327, 328] i.e., the opposite of what might be expected if the vaccines were reducing mortality risks. Although there are many problems in the attribution of excess deaths in any particular country, mainly due to the number of possible causes involved, and in identifying a reasonable and consistent measurement system, the relationships of the timings of increased mortality to vaccine roll-outs is an important “safety signal” [325], and the fact that “excess mortality has remained high in the Western World for three consecutive years, despite the implementation of containment measures and COVID-19 vaccines” [326] raises serious concerns. It also suggests that the vaccine adverse events reporting system (VAERS) is not operating as intended and that safety signals are being missed [329].

It is known that when people perceive certain behaviours as being associated with the risk of illness or death, their assessment of that behaviour (subconsciously) becomes moralised [348, 349], perhaps even contradicting a sacred value [348, 350, 351]. When we are evaluating an issue from a moral or even sacred perspective, it can be difficult to evaluate it dispassionately, based on purely evidentiary or logical grounds. Moreover, our moral views are often aligned with our political ideology [348, 349], meaning that pre-existing political divisions can aggravate societal divisions on these issues [351, 352].

From the beginning of the pandemic, media coverage, social media and government efforts effectively propagated an escalating sense of fear and panic among many people about the severity and dangers of COVID-19 [161, 344, 353, 354]. Hence, many people found it difficult to listen to different scientific perspectives on COVID-19 policies neutrally and dispassionately. Surveys in mid-2020 found that people with a greater personal fear of contracting COVID-19 were more likely to evaluate COVID-19 policies from a moral or even “sacred values” perspective [351], and that this effect was more pronounced among left-wing participants [351].

Concerns had already been expressed before the pandemic about the tendency for people to self-select into “echo chambers” where people seek information that agrees with their views and avoid dissenting perspectives [355]. Using value-laden terms such as “conspiracy theory” and “the science” to uncritically dismiss or promote particular perspectives seems to have increased this polarisation during the pandemic [356], and appears to have encouraged some people to “scapegoat” others [352]. Meanwhile, those who felt that expressing their opinion might lead to vilification, mocking or harassment, might have increasingly self-censored their views–leading to a “spiral of silence” [357].

Article 19 of the Universal Declaration of Human Rights states that, “Everyone has the right to freedom of opinion and expression; this right includes freedom to hold opinions without interference and to seek, receive and impart information and ideas through any media and regardless of frontiers” [358]. The Global Charter of Ethics for Journalists emphasizes that a journalist’s responsibility to the public in aiding this specific human right “takes precedence over any other responsibility, in particular towards their employers and the public authorities” [359].

On the other hand, in recent years, many in the journalistic profession have revisited these ideals–especially with regard to scientific reporting–because when the public is presented with multiple perspectives they might not reach the same conclusions as the journalists [360, 361]. Hence, many media outlets have begun prioritizing “reliable reporting” (where journalists and editors pre-decide what information the public should be provided with) over more conventional “balanced reporting” (where the journalists try to provide the public with all relevant perspectives) [361]. A consequence of this shift in approaches to scientific reporting by the media is that many editors and journalists have effectively decided to act as gatekeepers of what scientific information to convey to the public. During COVID-19, this was often accompanied with “a tendency to overuse linguistic items implying certainty” in “the science” even as the science in question was repeatedly re-evaluated over time [362].

In recent years, social media platforms and internet search engines have allowed the public the opportunity to research information for themselves and to share with each other information that they think is important. However, this has allowed the public access to a wider range of information than that provided by media outlets, thereby partially undermining the “reliable reporting” project described above. Hence, in recent years, these internet companies have found themselves under increasing pressure to artificially reduce the spread of some information and increase the spread of other information [363, 364].

At the start of COVID-19, platforms such as Facebook and Twitter expressed a willingness to suppress alleged “misinformation” on COVID-19 and promote alleged “accurate information” [28]. However, it is now being increasingly recognised that often the “misinformation” that was being suppressed transpired to be “accurate information” and vice versa. In a June 2023 podcast interview, the CEO of Meta (Facebook, Instagram and WhatsApp) admitted that his platforms had been “asked for a bunch of things to be censored that, in retrospect, ended up being more debatable or true” [365]. Decisions over what information to promote versus downrank were often determined by pressure from US government agencies, and often mistakenly involved downranking or censoring genuine scientific opinions [30, 366–368]. At one point, Facebook began penalising users for attempting to share an article published in the British Medical Journal, one of the world’s oldest general medical journals [269].

The narratives of popular discourse have increasingly been controlled by fact-checking organizations. These organisations are often partially funded by corporations such as Google and Meta (i.e., Facebook) and purport to identify certain articles as “false,” “misleading” or “missing context”. These fact-checks are then used as justification by social media platforms and media outlets for suppressing the information [269, 369].

These fact-checks can be based on flimsy grounds, misrepresent an article, or even make verifiably false claims, yet the organisations are set up in such a way that successfully appealing the claim is largely impossible [269, 369]. Hence, rather than improving the quality of information available to the public or providing relevant context, fact-checks are effectively no more than a narrative-check, censoring or discrediting scientific opinions that differ from theirs. Not only does this limit the public from being exposed to different scientific opinions, but the reputational damage of sharing information that ends up being “fact-checked” can also lead to self-censorship.

The academic basis for the use of fact-checking for narrative control appears to be based on “inoculation theory” [370, 371]. This is a strategy to minimise the availability of multiple perspectives on certain topics by “pre-bunking” the public with deliberate misrepresentations of compelling arguments into less compelling versions and then countering these strawman versions of the arguments. Analogous to vaccination with a weakened virus, when the public later encounters the original arguments, they will then be “inoculated” into dismissing the arguments without due consideration [370, 371].

Ultimately, governments and other policymakers decided national COVID-19 policies, even if they often insisted that they were “following the science” [48, 164–166]. However, there have been conflicting scientific opinions on most aspects of COVID-19 policies, throughout the pandemic. Therefore, they could only have been following some of the science at best. Indeed, it seems that governments were often simply following their neighbours [8, 112, 162, 372].

The scientific advisory groups used by governments for deciding COVID-19 policies generally included expertise from only a few relevant aspects of public health and were particularly influenced by modelling groups [48, 70, 165, 166]. Indeed, it has been argued that many of the policies implemented “would be considered unacceptable according to pre-pandemic norms of public health ethics” [373]. Meanwhile, there is concern that government policies were often heavily influenced by scientists and other appointees with potential conflicts of interest [165, 374, 375].

Others have noted that many policies were “authoritarian” and led to “violations of democratic standards” [159, 178], yet did “not correlate with better public health outcomes” [159, 376]. Apparently, many governments used psychological and sociological “nudge” manipulation techniques [377] to promote particular policies and behaviours [69, 220, 354].

Surprisingly, however–despite proclaiming that policy decisions were based on solid and incontrovertible evidence–behind-the-scenes information has revealed that often decisions were made without a concrete rationale [69–72]. The former UK Prime Minister, Rishi Sunak, who was Chancellor of the Exchequer from 2020 to 2022, admitted that UK’s “lockdown” policies were a direct consequence of preliminary model predictions [70]. It was only after the government rejected the modelling-based advice of SAGE, their scientific advisory group, calling for a fourth lockdown in December 2021 that the government realised SAGE’s model predictions were off by a factor of 20. Meanwhile, the chair of SAGE’s modelling committee admitted that they had intentionally excluded the possibility that a new variant might have been less virulent [71].

Legg et al. developed a framework for evaluating how and why corporations influence science and the use of science in policy and practice, suggesting that the pharmaceutical industry is potentially using 5 of their identified macro-strategies and 16 of their 19 proposed meso-strategies [378]. However, directly identifying biases specifically caused by the influence of the pharmaceutical industry on the media, social media platforms, government policies and the scientific community is difficult because the industry is so influential that interconnections between all of these different sectors are widespread [231, 379].

Although the direct advertising budget of many pharmaceutical companies on media and social media is only a fraction of their total expenditure budget, this can still represent a large source of advertising revenue for media or social media platforms. For instance, it has been estimated that the pharmaceutical industry spent nearly US$4 billion on TV advertisements in 2021 [380] – potentially biasing the neutrality of those platforms. Recently, it was revealed that Royal Medical Colleges in UK receive large payments from drug and medical devices companies [381]. An analysis of the payments received by French medical doctors from Gilead Science (developers of remdesivir) and their public statements on the potential use of HCQ (a cheap repurposed drug, proposed as an alternative treatment to remdesivir) for COVID-19 treatment during 2020 revealed that those who were most critical of HCQ had received the most funding from Gilead Science [382].

Working relationships between pharmaceutical corporations (including Pfizer, AstraZeneca, Johnson & Johnson, and Moderna), mainstream media (including Thomson Reuters) and tech companies (including Facebook and Google), as outlined by Deruelle, call the impartiality of the social media platforms’ fact-checking initiatives into question [231].

One of the main mechanisms by which scientists communicate their findings and analysis to other scientists is by publishing their work in peer-reviewed academic journals. However, over the years, the number of published papers has accelerated–especially in subjects where considerable funding is available.

Within just the first year of the pandemic, more than 100,000 COVID-19-related scientific articles were published [383]. Hence, there has clearly been a lot of scientific information, analysis and opinions expressed during the pandemic. This emphasises the erroneous nature of the idea that the science on COVID-19 was ever clear, simple or unambiguous. However, as we will discuss, it appears that many journals ensured research that supported certain scientific narratives was prioritised over research that contradicted those narratives.

Some papers seemed to speed through publication. For example, the paper describing the original RT-PCR test for SARS-CoV-2 was published on 23 January 2020 within less than 48 h of submission [118]. However, other papers have faced extended peer-review processes involving multiple rounds of revisions (often involving resubmitting revised manuscripts to different journals) that have delayed publication for months or even years. As an example the original version of Simandan et al. was first submitted to a journal on 12 December 2020 and had to go through multiple rounds of peer review in multiple journals before being accepted for publication on 11 January 2023 [178]. Given the rapid developments during the pandemic, these delays meant that COVID-19 policy decisions were typically heavily influenced by the papers that were published first. When different perspectives were eventually published months or years later, it was generally too late to make much difference.

For example, both Flaxman et al. and Wood analysed a similar scientific question: How much of an influence did NPIs have on the progression of the first wave of the pandemic (February to May 2020) in Europe? Flaxman et al. (2020) compared the COVID-19 death statistics to a SEIR-modelled counterfactual scenario of what would have occurred in the absence of NPIs and concluded that the NPIs were very effective and that 2.8–3.5 million deaths across 11 countries had been averted [7]. The paper was submitted on 30 March 2020 and accepted for publication in the journal Nature on 22 May 2020. Wood also studied this first wave but found that the wave had already been in decline in the UK before the full implementation of NPIs and in Sweden without NPIs [103]. This study received the first of several rejections in May 2020 and did not appear in print until September 2022.

Given the urgency of developing COVID-19 policies, many researchers took advantage of “pre-print servers” to try and speed up the process of delivering relevant scientific information into the public domain before peer review [383–386]. Some of these preprints influenced COVID-19 policies–especially during the early stages of the pandemic [6, 43, 383, 384]. Some also received media coverage [383, 386]. So, for some researchers, preprints offered a way of conveying their findings to the public and/or policymakers in time for them to be relevant. However, given the large number of COVID-19-related preprints published, unless the researchers had considerable influence on policymakers and/or good access to the media, many of these insights could have been overlooked. Moreover, some researchers have cautioned that many preprint servers have begun using opaque and/or poorly justified reasons for refusing to publish preprints, seemingly based on the conclusions of the manuscripts rather than their scientific merit [385].

An alarming trend we have noticed (and some of us have personally experienced) during the pandemic has been the misuse by some journals of the retraction/withdrawal process. When misapplied, this process can have the effect of both silencing and discrediting researchers for publishing peer-reviewed scientific articles that raise concerns over certain COVID-19 policies. Until recently, journal articles were very rarely retracted or withdrawn following publication in a peer-reviewed journal [387, 388]. Moreover, in a 2012 survey of retracted articles, it was found that 67.4% were accused of misconduct (including 43.4% fraud or suspected fraud) and 21.3% due to errors [387]. It was also found that it took an average of 2.5 years from publication until retraction [387].

These historic factors have erroneously convinced most of the scientific community that, if a paper is “retracted” or “withdrawn,” it should be treated as a major warning flag that the researchers involved were either involved in scientific fraud or else made serious scientific errors in their analysis that took several years to identify. However, during COVID-19, several journals began using this previously very rare and cautious tool of retraction/withdrawal as a new form of censorship [24, 388, 389].

Below, we give examples of several high-profile COVID-19 studies that were retracted or withdrawn by a journal without providing an adequate scientific justification or offering the authors a chance to issue a response:

• Two papers by Walach et al. were abruptly retracted with inadequate explanation [390, 391]. One raised concerns about the COVID-19 vaccination programmes, given the low observed efficacy of the COVID-19 vaccines in reducing death or severe illness (absolute risk reduction versus relative risk reduction, see table 2), coupled with the relatively high incidence of severe adverse events (as of June 2021). The other raised concerns over COVID-19 policies promoting the prolonged use of face masks by children in schools. Revised versions of both articles were subsequently published in different journals [389, 392, 393], and other researchers have reached similar findings [342, 394]. None of the journals gave sufficiently robust reasons why the studies should be retracted, and they refused to publish the authors’ responses.