- The chance of H5N1 or one of its descendants causing a pandemic is slim, but the impact would probably be more severe than COVID-19
- A reasonable forecast that H5N1 causes a pandemic as bad as or worse than COVID-19 beginning in the next year is ~4%, which means the expected cost in terms of potential harms to the U.S. is at least $640 billion
- The federal government can reduce that risk by:
- Scaling up testing & sequencing
- Banning mink farms
- Not pursuing H5N1 gain-of-function research
- Laying the groundwork for mass production of an H5N1 vaccine
- Investing in general pandemic prevention
Though COVID-19 is often referred to as a “once-in-a-century event,” this expression should not be taken literally. Another COVID-like pandemic at some point is almost inevitable, and the chances of a devastating pandemic materializing in any given year are low but constant. Recently, media figures have raised concerns that the highly pathogenic avian influenza virus A (H5N1) may have the potential to cause a pandemic. These concerns largely stem from probable transmission between minks late last year, which would constitute the first observed mammal-to-mammal transmission of H5N1.
An opinion piece by Zeynep Tufekci appropriately raises the alarm about H5N1 as a potential pandemic threat. Her piece may imply a higher probability of an H5N1 pandemic than is actually the case. But even with conservative assumptions, our forecasts indicate a roughly 4% chance of an H5N1 pandemic as bad as or worse than COVID-19 over the next year. If outbreaks in avian populations and spillover events return to baseline levels, then risk in subsequent years will be lower but still uncomfortably high. There are concrete steps the federal government can take to guard against that risk.
H5N1 is a subtype of influenza virus A, the kind of flu that causes pandemics. The modern strain of H5N1 was first detected in infected birds in 1996, and human cases and deaths were first confirmed in 1997. The virus spreads most readily between birds, though it has resulted in a total 887 confirmed human cases and 463 confirmed deaths, with most occurring 2004-2015. To date, known human infections have only occurred as a result of direct contact with infected birds, and there has been no observed human-to-human transmission.
The current clade of H5N11, known as 220.127.116.11b, has infected hundreds of millions of birds worldwide since 2021, including at least 58 million birds in the U.S. alone and a similar number in Europe. This is the main driver of the recent spike in egg prices, as H5N1 has killed or caused the culling of millions of egg-laying hens. Beyond this horrible reality, the spread of H5N1 creates many opportunities for spillover into mammals (like humans!). The more avian-to-mammal cases occur, the more opportunities there are for genetic changes in the virus to optimize it for mammal-to-mammal transmission — which is probably what occurred in mink on a farm in Spain in October 2022, and possibly in sea lions as well. Human and bird viruses can combine in mink. This makes mink-to-human transmission a genuine pandemic risk.
H5N1’s likely spread between mammals poses an urgent question: will it cause a full-blown pandemic? Many experts hesitate to assign probabilities to this sort of question — reasonably, since forecasting the future is hard. But we need at least rough guesses of the relative likelihoods of different scenarios to inform how concerned we should be about particular threats and how best to prepare.
A good starting point for forecasting is to understand the relevant reference class—in this case, the frequency and characteristics of past influenza pandemics since 1900. The past is more often than not a useful guide to what might happen in the future. There have been five flu pandemics in this time frame, which implies an annualized probability of 5/122 ≈ 4% of a flu pandemic beginning in a given year. Only one of these pandemics, the Spanish Flu, was COVID-like in its impact on global mortality.2 COVID-19, according to excess mortality figures, has killed an estimated 20.8 million people. By this accounting, the probability of a flu pandemic that is COVID-like or worse starting in a given year is ~0.8% (1/122).
None of these recent flu pandemics were caused by H5N1, since so far H5N1 has never been observed to transmit between humans. Indeed, in its current form H5N1 cannot transmit between humans; thus far it has not acquired the mutations that facilitate binding to human airway receptors. One way to estimate the probability of something occurring that has never occurred is Laplace's rule of succession,3 which in this case implies that, given there have been 0 H5N1 pandemics since the discovery of the modern form of H5N1 in 1996, there is a 1/29 ≈ 3.4% probability of it occurring in this next year.
These two methods are very rough and not in agreement (chance of a flu pandemic of any kind ≠ chance of an H5N1 pandemic). But they do guide our thinking: we can say that the percent chance of a flu pandemic starting in a given year is in the low single-digits, and that the chance of an H5N1 pandemic is probably not less than 1%.
Of course, the most relevant factors for our purposes are not related to the past, but are instead about the current situation. We have to account for the fact that the current H5N1 outbreak in birds is the largest ever outbreak of the H5N1 subtype of the virus, and that it is the first time probable mammal-to-mammal transmission has occurred. Since we do not have sufficiently reliable data to construct reference classes for an H5N1 pandemic, these are the author’s best guesses given his forecasting experience. Unfortunately, we do not have comprehensive historical data for any type of virus, given that large-scale genomic surveillance has only started in the last few years.
We can decompose the question of whether there will be an H5N1 pandemic into parts, keeping in mind the circumstances about our current situation. Each step is conditional on the previous step, and our objective here is to reason through the process that might lead to a pandemic from the current situation.
There is a 95% chance that non-human, mammal-to-mammal transmission has actually occurred
- Although we have strong evidence transmission has occurred in farmed mink, it is not definitive. There is perhaps a 90% chance transmission actually occurred in farmed mink in Spain in October 2022 and a 5% chance it did not occur in farmed mink in Spain but has instead occurred unobserved between other mammals.
There is a 90% chance that sustained transmission of any kind among non-human mammals has already occurred or will soon occur.
- Sustained transmission is different from any transmission whatsoever. Roughly, sustained transmission would mean multiple consecutive transmission events (e.g., mink A to mink B to mink C to mink D) instead of sporadic events (e.g., just mink A to mink B). The transmission mode can be of any kind (fecal-oral, airborne, etc).
There is an 85% chance notable genetic changes that facilitate airborne transmission in particular have occurred or will soon occur.
- We estimate a 80% chance that there has been or will soon be reassortment of H5N1 with another flu virus in mammals, and that this will result in notable genetic changes facilitating eventual transmission (especially airborne) between humans, and an additional 5% chance that this reassortment will occur but not in non-human mammals. This would likely involve an altered tropism, or ability to infect, where H5N1 more readily infects the upper airway.
There is a 40% chance that the H5N1 virus from the above step is at least somewhat transmissible between humans.
There is a 20% chance that the H5N1 virus that meets the above minimum transmissibility is actually substantially transmissible between humans.
- In particular, where R0>1, which would allow for a pandemic to occur. It is very difficult for R0 of a virus that is currently poorly adapted for human-to-human transmission to have a R0 that exceeds 1.0.
There is a 90% chance that a new H5N1 virus with a R0>1 would not be contained and would spread globally.
- We operationalize this outcome as global spread that is fast and extensive enough that it leads the World Health Organization (WHO) to declare a Public Health Emergency of International Concern.
There is a 95% chance that the new H5N1 virus would result in at least 10k confirmed deaths.
- WHO estimates a 56% case fatality rate of H5N1, but this is likely a large overestimate given ascertainment bias — that is, non-fatal cases are far less likely to be counted in current estimates. It is therefore surprising that well-informed people like Zeynep Tufekci cite this 56% figure without comment. Still, the real fatality rate would probably exceed 10%. It is possible that if, in the process of adapting to humans, H5N1 gained alpha-2,6 binding and lost alpha-2,3 binding (leading to ability to infect the upper airway instead of lower airway), this would result in fewer severe cases. SARS was simultaneously less transmissible and more severe than SARS-CoV-2 for this reason.
There is a 80% chance that this new H5N1 virus would become a COVID-like pandemic or worse.
- It’s likely that an H5N1 virus with R0>1 and which has already caused >10k deaths would go on to be at least COVID-like in its impact. There is a chance that appropriate actions such as massive vaccine production efforts could avert this outcome.
The above decomposition process leads to an overall ~4% chance of an H5N1 pandemic starting in the next year that is at least COVID-like in its impact. Since the total harms of COVID-19 to the U.S. are about $16 trillion, the expected cost of H5N1 is at least $640 billion — though it is probably far greater, since the mortality rate would probably be much higher than that of COVID-19. Yet we are hardly preparing for this costly possibility, though adequate preparation would cost far less than $640 billion. Our analysis does not take into account the annualized risk for all future years, so we are being very conservative in estimating the value of preventing an H5N1 pandemic.
COVID-19 was a clear reminder that low-probability, high-consequence events are worth taking seriously. The ~4% chance arrived at in this piece is just for H5N1, so it doesn’t account for the possibility of a different flu pandemic starting this year. There will also be a continued (albeit probably lower) risk of an H5N1 pandemic starting in subsequent years.
There is not much the public can do at this time about H5N1 — with the exception of not touching dead or dying birds — but public health authorities and policymakers have several options:
1. Scale up testing and genomic sequencing of H5N1
- So we can keep comprehensive track of human infections and keep an eye out for any notable genetic changes. We especially need to scale this up in Asia and Africa, where we have disproportionately little testing and surveillance.
2. Ban or buy out mink farms
- We almost did this in the U.S. last year. According to the fur industry itself, the value of mink pelts produced in the U.S. is only $80 million.
- If this is not possible, we should at least ensure mink are not fed with poultry products, reduce crowding in mink cages, and take steps to prevent all contact between mink and wild birds.
3. Do not do H5N1 gain-of-function research
- Some may suggest that we should research how to increase the airborne mammal-to-mammal transmissibility of H5N1 (researchers have done such experiments before) in order to better characterize it. There is little actionable knowledge we gain from this, and that enhanced H5N1 virus might leak out of the lab, which would not be unprecedented.
4. Lay the groundwork for mass production of an H5N1 vaccine.
- Support work by the Biomedical Advanced Research and Development Authority to develop candidate vaccines against the currently-circulating strain of H5N1 and to ramp general vaccine manufacturing capacity
- If we get to a point where we observe sustained human-to-human transmission of H5N1, then we should begin mass-producing and stockpiling H5N1 vaccines specific to the that strain using an Operation Warp Speed-like process
- Support efforts to vaccinate as many birds as possible against H5N1, both to protect the birds and to reduce the number of opportunities for H5N1 spillover from birds to mammals
5. Invest more in overall pandemic prevention to protect us against not only H5N1 but also all other pandemic threats
- There is a lot we can do to prevent future pandemics, like funding the Chemical Biological Defense Program. President Biden asked for $81.7 billion over 5 years for pandemic preparedness, which is about a tenth of the expected cost of an H5N1 pandemic potentially that begins in the next year, let alone the risk of H5N1 over the next decade or even the combined risk of all other pathogens with pandemic potential.
- One area we at IFP think is particularly promising is making our indoor spaces pathogen-resistant, and we’ll be releasing more on this soon.
Finally, forecasting sites like Metaculus and GoodJudgment.io can help us keep track of how the probability of H5N1 being declared a public health emergency, and how it will evolve over the coming weeks and months.