In 1997, a 3-year-old boy in Hong Kong died of a respiratory illness that turned out to be H5N1 avian influenza. Seventeen other people were infected. Six died. Authorities slaughtered 1.5 million chickens in three days and stopped the outbreak. For the next 25 years, H5N1 circulated in poultry and wild birds, occasionally jumping to humans who had direct contact with infected animals. Global cumulative count through early 2024: 888 confirmed human cases, 463 deaths. A case fatality rate above 50%.

That number has always come with a caveat. Mild cases are likely undercounted. The true fatality rate is almost certainly lower. But even if it's 10%, or 5%, or 1%, an influenza virus with pandemic potential and a fatality rate that high demands attention. COVID-19 had an infection fatality rate around 0.5-1% and killed over 7 million people worldwide. H5N1 is a different animal.

What changed with US dairy cattle?

In March 2024, H5N1 was confirmed in dairy cattle herds across multiple US states, marking the first known instance of this virus establishing sustained mammal-to-mammal transmission in a livestock species. By December 2024, the USDA had confirmed infections in herds across more than a dozen states, with evidence of spread through contaminated milking equipment and possibly respiratory droplets between cows.

Nothing like it had happened before. H5N1 had always been a bird virus that occasionally spilled over to mammals. Cats, dogs, foxes, bears, seals, and mink had all been infected in isolated events, usually after eating infected birds. But dairy cattle represented something qualitatively different: a mammalian population of millions, in close physical contact, being handled daily by human workers.

Several dairy farm workers tested positive for H5N1 during 2024. Most experienced only conjunctivitis (eye infections) and mild respiratory symptoms. One patient in Louisiana was hospitalized with severe pneumonia. No confirmed human-to-human transmission occurred. But each human infection is a lottery ticket the virus didn't ask for. Every time H5N1 replicates inside a human cell, it has a chance to acquire the mutations that would make it spread efficiently between people.

What mutations are scientists watching?

One mutation gets the most attention. PB2 E627K is the most closely tracked genetic change because it improves the virus's ability to replicate in mammalian respiratory tracts. Avian influenza viruses replicate best at 40-41 degrees Celsius, the temperature of a bird's respiratory tract. Human airways run at 33-37 degrees. PB2 E627K helps the virus adapt to that cooler environment.

H5N1 isolates from some mammalian infections have already shown PB2 E627K. It appeared in the Louisiana patient's virus. It's been found in cat and seal isolates. Each occurrence is an independent adaptation event, suggesting strong selective pressure toward this mutation whenever H5N1 enters a mammal.

But E627K alone isn't enough. Efficient airborne transmission between humans likely requires multiple coordinated changes. Research published by Ron Fouchier's lab in 2012 (the controversial gain-of-function studies) identified that as few as five mutations could make H5N1 transmissible through respiratory droplets between ferrets, the standard mammalian model for influenza transmission. Some of those five mutations have already been observed in circulating strains.

Other changes scientists track include mutations in the hemagglutinin (HA) protein that shift receptor binding preference from avian-type (alpha-2,3 sialic acid) to human-type (alpha-2,6 sialic acid) receptors. Human upper airways are rich in alpha-2,6 receptors. A virus that binds preferentially to those receptors can infect and replicate in the nose and throat, which is where airborne transmission starts.

How does this compare to 1918?

Between 50 and 100 million people died in the 1918 influenza pandemic. The virus was H1N1, and genomic analysis confirmed it was avian in origin. It adapted to humans through a series of mutations, possibly via an intermediate mammalian host, and acquired efficient airborne transmissibility. The basic playbook for an avian influenza pandemic has been written before.

Key differences between 1918 and a potential H5N1 pandemic:

Medical capacity. In 1918, there were no antiviral drugs, no ventilators, no ICUs, and no understanding of viral respiratory pathogens. We now have oseltamivir (Tamiflu) and baloxavir (Xofluza), both active against H5N1. We have ventilators, ECMO, and intensive care protocols refined through COVID. But we also learned from COVID that ICU capacity runs out fast. The US has approximately 100,000 ICU beds for a population of 340 million.

Vaccine timeline. Influenza vaccines are well-understood technology. The US has pre-pandemic H5N1 vaccine stockpiled (though the current stockpile may not match circulating strains well). From the moment a pandemic strain is identified, manufacturing at scale takes 4-6 months using egg-based production. mRNA vaccine platforms could potentially cut that to 3-4 months. During that gap, the virus spreads unopposed among the immunologically naive global population.

Global connectivity. In 1918, a virus moved by steamship and rail. An H5N1 pandemic strain would reach every continent within weeks via commercial air travel. Modeling studies estimate that a novel influenza pandemic would infect 30-50% of the global population within 6-9 months.

Population vulnerability. 1918 disproportionately killed healthy young adults through cytokine storm responses. H5N1's pattern in humans so far skews toward severe disease across age groups, with immunocompromised and elderly patients at highest risk. But the sample size of human H5N1 cases is too small to predict a pandemic age distribution with confidence.

What would an H5N1 pandemic actually look like?

If H5N1 acquired efficient human-to-human airborne transmission while retaining significant pathogenicity, the initial weeks would look deceptively like COVID's early phase: clusters of severe pneumonia in a specific geography, followed by rapid international spread through air travel.

Differences would become apparent fast. COVID's initial case fatality rate in Wuhan was estimated at 2-3%. H5N1's dairy-strain cases have been mild so far, but historical H5N1 infections with the classic avian strain have a much higher hospitalization and death rate. Even a pandemic strain with a 1-2% fatality rate, attenuated from the 50% seen in sporadic cases, would overwhelm hospital systems globally.

Antiviral supply is a bottleneck. The US Strategic National Stockpile holds approximately 65 million courses of oseltamivir. For a pandemic infecting 100+ million Americans, that's not enough. Baloxavir production capacity is even more limited. Resistance to neuraminidase inhibitors (oseltamivir's drug class) can emerge during treatment, as it has in seasonal H1N1.

Non-pharmaceutical interventions, the same toolbox used for COVID (masking, distancing, ventilation, isolation), would be the first line of defense during the vaccine gap. Countries that maintained pandemic preparedness infrastructure after COVID would respond faster. Countries that dismantled it would struggle.

Why does PandemicAlarm rate H5N1 at severity 4/5?

PandemicAlarm's severity scoring weighs five factors: pathogen lethality, transmission potential, geographic spread, healthcare system capacity to respond, and availability of medical countermeasures. H5N1 scores high on the first two and middling on the rest.

Lethality: historically extreme, though the dairy cattle-associated strain has produced milder human cases so far. We weight historical data because viral evolution is not predictable, and a reassortment event could restore high pathogenicity overnight.

Transmission potential: currently rated as limited (no sustained human-to-human spread). But the virus is accumulating mammalian adaptations in an enormous cattle reservoir. Each season it circulates in dairy herds and wild birds increases the probability of the right mutation combination appearing. We rate transmission potential based on proximity to efficient human spread, not just current status.

Geographic spread: H5N1 clade 2.3.4.4b is present on every continent except Antarctica, in wild bird populations. It's in US dairy cattle across multiple states. The geographic preconditions for a pandemic are already met.

Medical countermeasures: vaccines exist but aren't deployed. Antivirals exist but aren't stockpiled in sufficient quantity. Hospital capacity is finite. These factors don't reduce our severity rating.

What should you watch for?

Watch for one signal above all others: confirmed human-to-human transmission. Every H5N1 human case to date has been traced to animal contact. The day epidemiologists confirm a chain of transmission between people who had no animal exposure, the risk calculus changes fundamentally.

Other signals PandemicAlarm monitors:

You don't need to live in fear of H5N1. You need to watch it. The virus is telling us what it's doing through its genetics and its host range expansion. PandemicAlarm translates that signal into severity scores and alerts so you can calibrate your preparedness. Right now, that means maintaining your baseline supplies, knowing where to find N95 masks quickly, and checking whether your household has had a seasonal flu vaccine (which won't protect against H5N1 but keeps you out of the hospital for something preventable while a different threat is building).

Stay informed. That's the whole point.