How likely is a bird flu pandemic right now?

(Yes, there will be math.)

Getty Images/KTSDESIGN (wheel) Kateryna Kon (particle). Mashup: Inside Medicine.

Everyone has the same question: Are we about to have another pandemic? With all the H5N1 circulating in animals in the US, the chances of one are higher than usual. Some models estimate the risk as anywhere from 5%-40% this year.

But the odds of events that could reasonably trigger a pandemic are, in my view, temporarily much higher than that.

Whether they will is a separate matter.

These are distinct questions. Let’s consider them both. Let’s go Inside Medicine…

How do flu pandemics start?

There are different ways for a flu pandemic to occur. Broadly speaking, though, we can think of this happening via two mechanisms.

Mechanism 1:

Novel mutations. Example: an animal carries a flu virus that can’t spread to and among people. But a mutation occurs that allows this. The virus then jumps to a person during a close encounter. Whether this novel virus causes severe illness depends on the nature of the mutation and any prior human exposures to similar viruses.

Mechanism 2:

Co-infection. Co-infection is when two versions of the same virus simultaneously infect the same person. Because influenza has eight strands of genetic material, a novel combination of these strands can form a new variant. This is called reassortment.

The chances of Mechanism 1 increase when the virus is prevalent in animals (more chances for mutations to occur) and when there is more animal-human contact (e.g. dense farms). That’s why controlling outbreaks in animals with potential human contact is advisable and why keeping high-risk humans safe (e.g., giving farmworkers adequate PPE and training) is essential.

The first mechanism applies to a vast array of pathogens including SARS-CoV-2 and influenza. The second mechanism basically only applies to viruses like influenza that have segmented genomes (albeit, “mosaics” can occur in coronaviruses, the process is a different one and much less likely).

Which is more likely overall?

It’s hard to know which mechanism more likely to cause a novel pandemic, and in some ways, the mechanisms are intertwined; the chances of Mechanism 2 depend both on Mechanism 1 conditions and the prevalence of seasonal influenza in humans.

So, it’s also difficult to say which mechanism is responsible for past flu pandemics, partly because the implications of co-infections can be changed by Mechanism 1 features (e.g., high prevalence in animals, new mutations).

Now (January, though February and March in some places) is a crucial moment.

January in many places (February and March in others) is the most dangerous time for a pandemic to erupt by way of Mechanism 2. By far. The probabilities of Mechanism 2 co-infection occurring during the peak of seasonal influenza season in humans are somewhat knowable, I believe, and worth interrogating. If we can get through the next few weeks (in the Northeast) and the next couple of months in places like Michigan (where flu often peaks later), we’ll be in a better place with respect to Mechanism 2. That’s because the prevalence of seasonal flu from December-March is orders of magnitude higher than the fall and spring (let alone summer, when it’s essentially zero).

I was told there would be math.

You came for the math. Let’s do it. (Or you can skip to the next section other than the bold text.) We’ll use California as the laboratory for this, as that is where most H5N1 in animals is circulating right now.

The question is: How likely is a co-infection of seasonal flu and H5N1 to occur in California during the peak two weeks of flu season?

Assumptions:

• 10,000 cattle farmers in California with elevated H5N1 exposure risks.

• During a moderate flu season, around 20% of the US population gets flu, but the peak two weeks accounts for around 25% of that. That means around 5% of the US population gets flu during the peak two weeks (2.5% per week).

• California farmworker H5N1 infection rates mirror the evidence that 7% of farmworkers in Colorado and Michigan were infected the 13 weeks during June-August of 2024 (i.e., 0.54% per week).

Given this, we would expect that 250 California farmworkers get seasonal flu in a peak week. Of those, 99.46% will not get H5N1, but 0.54% (0.0054) will. Therefore, the probability that one of the 250 farmworkers with seasonal flu will also get H5N1 in a one-week period is equal to: 1-(1-0.0054)^250, or 74.2%.

So, for the peak two weeks of seasonal flu in California, the math suggests a 93.3% chance that at least one co-infection of seasonal flu and H5N1 will occur in an at-risk farmworker. We get this by simply replacing 250 cases of flu in a week with 500 over two weeks, or by calculating the probability that no H5N1 infection occurs among two sets of 250 people for two distinct weeks (the math there is: 1-(0.2583)^2≈(1-0.0667)≈93.3%). On top of that, given these assumptions we actually expect there to be 2 or 3 such cases in a two-week period.

By the way, I’m concerned that influenza modelers at the CDC and elsewhere haven’t taken the seroprevalence data (from the CDC itself) into account in their threat assessments. If you believe there have only been the 67 documented cases of H5N1 in the US so far, then, yes, the estimates above are way too high. But if you think the CDC’s antibody data in farmworkers is correct (or even ballpark), then these figures are not unreasonable.

There are also some other factors to consider, including whether co-infections are rendered less likely because getting one (on day 0) might make it less likely to get the other (say, after day 3, if symptoms cause someone to isolate). But just as staying home from work could decrease the odds of an H5N1 infection in a farmworker, it could increase the odds of a seasonal flu infection by adding extra hours of exposure to a sick child in the home.

What does this all mean?

Does this math imply that a pandemic is 93.3% likely to occur in California this month? No—and I implore you not to misquote this figure; these are the odds of a co-infection in the high-risk California farmworker population, not the risk that such an event is guaranteed to cause a serious outbreak of public health concern. The reason the odds of that are lower than 93.3% is that we have no idea how often co-infections cause a virus possessing two key ingredients for a serious outbreak or pandemic: severity of illness and transmissibility. It’s a black box.

So, while we can estimate how often co-infections occur, we can’t know how often these events create scary-bad pathogens that actually take off. If the models that stake the rates of a pandemic this year as between 5%-40% incorporate this (and I don’t think they do), the implication is that 53%-95% of events like co-infections never cause sustained chains of transmission leading to epidemics/pandemics. But, again, I don’t think the models take that into account. (Someone can correct me if I’m wrong.)

Indeed, we simply don’t know how often these events occur but flame out. For example, how many people with a co-infection are appropriately isolated during the transmission window? Is it most? Maybe. It depends on the overlap of symptoms and transmissibility, as well as social factors.

Consider two co-infection scenarios:

• Scenario 1: A co-infected person who lives alone returns to work as soon as he feels well enough—but he is still contagious. His work puts him in close contact with 50 people each day. If 10% of them catch the virus during his contagious window, his “reproductive number” is 5. Whether they spread it, depends on their own behaviors. But when the average person spreads a virus to more than 1 person, an epidemic takes off.

• Scenario 2: the same patient with the same virus happens to have no work lined up for the week after he gets sick. Instead of going to work, he stays home, plays video games, and orders delivery. He spreads it to nobody (i.e., his reproductive number is 0). The “epidemic” starts and ends with him.

When Covid-19 began, there was a lot of attention around its reproduction number (“R”). What most people don’t realize is how much the R of a pathogen depends on our own behaviors. Staying home or wearing a good mask can lower the R substantially. These things can make small or large differences. Imagine a virus with an R of 1.2. What if masking reduces the R by 25%? Now the R is less than 1. It will die out. But for a bug with an R of 1.2, after 25 generations we’d have >2 million infections. While 25 generations sounds like a lot, each generation might be 7 days. That means you can go from a single case to 2 million in just 175 days. Of course, the R might decrease once we become aware of the illness and take measures to slow it down. That’s why the R of Covid-19 in the earliest days of the pandemic might have actually been double digits, but quickly dropped.

A game of chance, where we can alter the odds.

When we think about co-infections and novel pandemics, there are key unknowns: How often do co-infections occur? How many such events create variants with pandemic potential? How many of these infections happen to flame out during the transmission window of the initial case?

I think our situation regarding the frequency and importance of co-infections really boils down to a 2x2 table like the one below. The lower left cell (red stars) is where we are now. My concern is that we are in a moment of increased probability.

Co-infection implication matrix:

The key to success, I believe, is knowing when the risk is high (i.e., now), and acting to improve our odds. That means bolstering PPE and testing for farmworkers. That means detecting animal outbreaks earlier (by pooled testing of milk and other products) so that there are fewer infected vectors. That means asking high-risk infected people to stay home longer than they might think is strictly necessary.

On this last point, we need the government to act. Remember when the federal government sent out checks during the Covid-19 “shutdown”? Hopefully, we are not headed towards anything like that. But if I were the federal government, I’d be ensuring that high-risk farmworkers have well-paid sick leave over the next few months.

We don’t know what our overall odds of a pandemic are right now. But we roughly know the odds of a co-infection with pandemic potential, and they will be uncomfortably high in the coming weeks. We also know how to alter the odds in our favor. Are we doing all we can?