Bird Flu: Human Pandemic Unlikely, but Will Poultry and Cattle Industries Take Action?
Illustration by James Daw
The forecasters were asked to predict whether the H5N1 bird flu virus, currently circulating in birds and in some non-human mammal species, will lead to a major human health concern or to economic damage to the livestock industry.
Key takeaways
The forecasters thought that it was unlikely, but not vanishingly so (4.5%), that significant human-to-human spread of the virus (minimum 10,000 confirmed cases) would happen this year
They thought it more plausible (but still unlikely – 14.5%) that the World Health Organisation would declare H5N1 a “Public Health Emergency of Imminent Concern”, because sometimes the WHO declares PHEICs over diseases that have relatively few human cases
They foresaw essentially two main futures when it comes to the number of deaths: either the disease will not start spreading significantly among humans, in which case there will be very few, or it will. Conditional on 10,000 human cases, they thought there could well be millions of deaths
Most forecasters thought that whether or not there are 10,000 human cases, mass poultry vaccination programmes or dairy cattle culling in the US and UK were unlikely, both because of the economic costs of doing so and because they would likely be ineffective at preventing the spread of the virus
Chance of many human to human cases remains low but slight increase due to spreading in mammals
(hover/tap to see comments and rationales from forecasters within each range)
The average forecast was 4.5%, and the forecasters clustered quite closely around that figure, although their reasons for the decision differed.
Some forecasters felt that the risk of a serious human outbreak of the disease was more likely than in a normal year, thanks to the virus’ ongoing spread in several mammal species: notably cats, cows, seals and pigs.
One representative forecaster (estimate 4.5%) said that the base rate for an H5N1 pandemic should be “somewhere in the range of 0.7% to 1% per year,” given previous research, suggesting a roughly once-a-century rate. That forecaster felt the risk was increasing, thanks to the aforementioned mammal infections, but that they weren’t higher because “I'd expect that if it was going to happen, it would have already happened” – the virus has been spreading for some years now.
They added that they would update their probability if there were significant infections in pigs, which are quite human-like in their immune profile, and if “multiple cases of untraceable H5N1 show up in random hospital screenings (this one would massively increase the probability).”
Another one (estimate 3.5%) reached a similar conclusion but in a somewhat opposite direction. They assumed a base rate of 4% for an influenza pandemic – which would imply one every 25 years or so – but adjusted that down, because “an avian flu in the past has never achieved the sort of sustained human-to-human transmission” that a pandemic would require, and “I come in pretty sceptical that this time is particularly different.”
Another (estimate 4%) said “I am fairly optimistic that this won't happen before the end of 2025 so I am anchoring to the base rate.” They said they expected more human infections, but mostly among farm workers from animals and not from human-to-human transmission. “I am not confident this holds out in further years,” they said, “but we aren't forecasting on that.” One important point they added was that if the virus does start to spread rapidly among humans, “I think the political resistance to the measures taken to protect against Covid 19 will be far stronger,” which may mean the chance of a few cases turning into many might be higher.
Chance of a ‘Public Health Emergency of Imminent Concern’ declaration by WHO
The forecasters thought this was more likely than the 10,000 confirmed cases – average forecast of 14%.
Several thought the question was essentially the 10,000 cases one in disguise. “This is roughly equivalent to my forecast for 10,000 cases resulting from human-to-human transmission,” one (estimate 8%) said. “The WHO declared a PHEIC around the day when 10,000 COVID-19 cases were reached.”
Others noted that the WHO has declared PHEICs at much lower levels in the past. One (estimate 8.5%) noted that the H1N1 virus was declared a PHEIC at around 1,000 cases, polio at around 68, ebola 1,779, Zika 4,000, and Mpox 16,000. Nonetheless, “all the PHEICs seem to have been in cases where human-to-human transmission was pretty rapid and worrying, so this seems fairly similar to the main question. Arbitrarily, maybe twice the chance.”
Others were much higher, in some cases because they thought a COVID hangover would create institutional overreaction: “I anticipate any H5N1 PHEIC declaration to come before it might have been warranted on more objective measures,” said one (estimate 27.5%).
Predicted relatively low deaths globally by the end of 2027
(hover/tap to see comments and rationales from forecasters within each range. You can also hover over the out of bound %s to see further comments)
The forecasters thought that the most likely outcome was a relatively small number of deaths (median estimate ~670). But they had a long tail: they thought there was a 20% chance of over 120,000 deaths, and a 10% chance of over a million.
They were also asked what the number of human deaths would be, conditional on there being 10,000 confirmed cases of human-to-human transmission, and they thought that was far higher: an average of 910,000.
The forecasters essentially envisioned two scenarios: in one, the most likely, the virus never really takes hold in humans at all; in the other, it spreads rapidly and becomes a pandemic. One (whose forecast conditional on 10,000 cases was 20 million) said: “More than 10,000 human-human cases I think would imply more than a 95% chance of a pandemic.” In that case, it depends on the lethality of the virus, and that is hugely unknown. “I think it could plausibly be anywhere from 0.1% to 20%.” If it’s substantially above 1%, then “we should expect that healthcare facilities rapidly become overwhelmed, with most patients receiving no medical treatment, pushing the IFR up even further.”
Others were more sanguine, suggesting that even given significant human-to-human transmission, the best comparator was a bad flu season, not a devastating COVID-19-style pandemic. “Flu is much less transmissible than Covid,” one (conditional forecast 1 million) said, “and the new virus about equally may end up being milder or more severe than Covid. Aiming for the high end of seasonal flu.”
Predictions on case fatality rate of H5N1 by the end of 2025
(hover/tap to see comments and rationales from forecasters within each range)
The impact of a disease is not just mediated by how rapidly it spreads but by how deadly it is for each person infected. The usual measure of that is the “case fatality rate”, or CFR – the percentage of known cases who go on to die.
It is an imperfect measure since not all cases are known: many people might get the virus but never be tested, so they don’t show up as a non-fatal case. Ideally we would have the “infection fatality rate”, IFR, which is the percentage of all people infected who go on to die. CFRs are usually significantly higher than IFRs because the denominator is smaller.
The forecasters had widely differing estimates on this question, but mostly that was because of disagreements on how much the testing regime would reveal the true underlying numbers, rather than because some of them genuinely thought a quarter of all infected people would die.
One forecaster (estimate 1.4%) noted that there have so far been 66 cases and one death, for a CFR of 1.5%. Whether that figure will change, they said, depends on the path the disease takes: “In an epidemic scenario, I expect the CFR to be around 0.4%,” they said, because there would be widespread testing which would pick up more weakly symptomatic cases, and serological testing has suggested that the true number of infections is higher than estimated. But “if the status quo (sporadic infections due to contact with infected wild birds and dairy cows) continues, I think the CFR will hover around 1.5%,” because only severe or high-risk cases will be tested, artificially boosting the apparent CFR.
Another broadly agreed with that prognosis – they noted that H5N1’s CFR in the past has been around 50%, but that that was due to only very severe cases being tested. “Even without an emergency, they will do more testing than was available in past outbreaks, so the number will be lower,” they said: Initially they forecast 25% but adjusted down to 5%. Even that was an overstating of what they thought was the “true” rate: “If there is an emergency and a high number of cases and mass testing is conducted, the fatality rate will drop dramatically (possibly below 1%).” The vast majority of the 2024 cases in North America were very mild, consisting primarily of conjunctivitis. The notable exceptions were a severe cases in a teen in Canada that required intubation, and the death of a man in Louisiana that was over 65. Both had comorbidities and the man in Louisiana was infected by poultry, rather than the more dominant vector of cows. The virus circulating in infected poultry is more dangerous.
Chance of mass poultry culling remains relatively low
(hover/tap to see comments and rationales from forecasters within each range)
One obvious way of reducing the risk of infections from birds would be to vaccinate the stock on poultry farms. The forecasters, though, thought this was unlikely (estimate 4.8%), although somewhat more likely conditional on there being 10,000 human cases (11%).
“It doesn’t seem like a particularly good measure to prevent an H5N1 pandemic,” one (forecast 2%) said. “There are better ones, and this one could face significant opposition.” They also noted that US public health bodies have been slow to react to previous health crises, notably COVID-19 but also HIV/AIDS, opioids, and others. “I expect the will to take this kind of measure wouldn't even be present until it was too late and made even less sense.”
They did say that it was more likely (conditional forecast 4%) in the event of 10,000 human cases, not because the 10,000 cases would spark action – “there is no point then, it’s far too late” – but because in futures in which we reach 10,000 cases, we’d have seen concerning signs before then which might spark a mass poultry vaccination programme.
Another had a higher estimate (18%), but still felt it was the unlikely outcome, partly because of public “hysteria” about vaccines but also for simple economic reasons: “Chicken are relatively cheap so culling is a potential option.” Another (estimate 3%) noted that the US chicken industry is wary of vaccinations because it has implications for export: many countries restrict the import of vaccinated chickens in case it hides the presence of the disease itself.
The highest conditional estimate was 35%, from a forecaster who noted that a similar programme had been used in China in 2017, apparently successfully, “squashing an H7N9 outbreak with mass vaccination of poultry with a bivalent H5/H7 vaccine.” If there was a full-blown epidemic, they noted, and if the human-adapted virus could still infect birds, then “the bird vaccination might very well be viewed as a critical tool in combating human epidemic,” and “if China could to it in 2017, we certainly should be able to do it in 2025.”
Cattle culling predictions in 2025
(hover/tap to see comments and rationales from forecasters within each range. You can also hover over the out of bound %s to see further comments)
H5N1 is spreading rapidly through US cattle herds, and at the moment that is probably the likeliest vector for human transmission to start, since they are mammals in regular contact with humans. The forecasters were asked to predict the numbers that might be culled, both given the status quo and conditional on 10,000 human cases. Their central estimate was 46,000 in a business-as-usual scenario and 425,000 in the conditional scenario, but the latter was very uncertain, with the probability distribution very widely spread.
There are around 10 million dairy cattle in the USA, one (estimate 534) noted. So if “somewhere between hardly any and 10%” are infected, and 2% of those are culled, the number would end up in the hundreds or low thousands.
Another (conditional estimate 1.7 million) noted that if there are 10,000 human cases, and since cows are the most likely vector, culling is likely to occur, but it would likely be limited because cows are generally not getting very sick and because they are valuable: “As long as the severity of symptoms continues to primarily be non-lethal conjunctivitis, the push for mass culling will likely remain low. The financial pressure to not implement mass culling will be very high.”
Another (status quo estimate 76,000) said that if mass culling was going to happen, it would probably have already happened: “This correlates strongly with futures where there is no pandemic in humans AND there are lots of problems with cattle. So... sort of like where we are now. Except it's not happening, and doesn't seem to be on the horizon.” They thought that in the 10,000 human cases scenario (conditional estimate 592,000) it was more likely, because it “correlates with bad futures,” as with chicken vaccination.
Will UK dairy herd stock be culled due to H5N1 concerns?
(hover/tap to see comments and rationales from forecasters within each range)
The forecasters were similarly asked to estimate the chance of major culls in the UK, both in a business-as-usual scenario and conditional on 10,000 human cases. The difference was that they were asked to give a probability of a certain number of cullings, rather than a probability distribution. Their forecast was 2.% given the status quo and 6.9% in the widespread transmission scenario.
Some thought culling more likely in the UK than the US, perhaps because of the memory of bovine spongiform encephalitis (mad cow disease). “The UK seems more prone to culling in such situations,” one (estimate 35%) said, “and 50k is not too high a threshold,” given that there are about 2.65 million dairy cows in the UK.
Another (status-quo estimate 1%), though, thought that it was still unrealistically high. “This is about 2-3% of all dairy cows in the UK,” they said. “Only around 7% of cattle were culled during the foot-and-mouth outbreak in 2001, and culling seems much less appropriate or necessary in this case because most dairy cows recover.” They note that about 3% of dairy cows in the US have been infected but only about 2% of them culled, and so far UK dairy herds have not been affected. Even in the conditional scenario, they thought (estimate 2%) it was still unlikely. “Even if there is sustained human-to-human transmission,” they said, “British dairy cows won't necessarily get the blame.”
Conclusions
The forecasters, on the whole, thought that H5N1 is unlikely to cause a human pandemic this year. But they thought that if significant human spread did begin – 10,000 confirmed cases or more – it would be very likely to cause large numbers of deaths, at least as much as a bad flu season and potentially many millions more.
They argued that potential indicators of seriousness, such as whether the WHO declares a public health emergency, or whether the case fatality rate is high, are likely to be of limited use in determining how bad the situation is: the WHO often declares PHEICs on relatively few human cases, and the CFR will be more influenced by the level of testing than the inherent lethality of the virus.
They also suggested that steps to protect humans from infection via livestock, vaccinating birds or culling cattle, would be ineffective, poorly implemented and economically devastating to the industries, so were unlikely even in the event that the virus was spreading – although if it does spread, those steps would be somewhat more likely.
