Why Flu Vaccines Work Better in Some Years Than Others

Each February, while winter influenza is still circulating across much of the Northern Hemisphere, experts begin committing to the next season's vaccine. They examine viruses collected around the world, compare their genetic sequences and test whether antibodies can still recognise them. The recommendation must come early enough for manufacturers to prepare millions of doses. By the time those doses reach clinics in autumn, influenza has had months to keep evolving.

This is the central wager of seasonal flu vaccination. The vaccine is not one unchanging product, and influenza is not one virus. Human epidemics are usually caused by influenza A subtypes H1N1 and H3N2, together with influenza B viruses. The vaccine presents selected viral antigens, the molecular features that train an immune response. Its composition is reviewed regularly because the viruses accumulate mutations. For the 2026 to 2027 Northern Hemisphere season, the World Health Organization recommended three updated components after reviewing global surveillance data [1].

When people ask whether the flu vaccine worked, they often expect a single number. Vaccine effectiveness, abbreviated VE, is a relative reduction in a defined outcome among vaccinated people compared with otherwise similar unvaccinated people under real-world conditions. If 2 in every 100 unvaccinated people in a study develop laboratory-confirmed flu and 1 in every 100 vaccinated people does, VE is 50 per cent. The absolute reduction is one case per 100 people in that setting. Both figures can be correct, but they answer different questions.

The outcome changes the answer too. A vaccine might be moderately effective at preventing any medically attended infection while offering different protection against hospitalisation. Studies may count a positive test in primary care, an emergency visit, intensive care admission or death. They enrol different ages and operate in different health systems. A percentage detached from its endpoint, season and confidence interval creates false precision. Flu vaccines work better in some years, but apparent differences can also arise from what researchers chose and were able to measure.

A widely used approach is the test-negative design. Researchers enrol people seeking care for a flu-like illness and test them. Those positive for influenza are cases; those negative are controls. Investigators compare vaccination histories between the groups and adjust for factors such as age, underlying illness and calendar time. Restricting the comparison to people who all sought care helps reduce bias from different health-seeking behaviour. It does not eliminate every difference between vaccinated and unvaccinated patients.

Across many seasons, one pattern is striking: effectiveness varies by viral subtype. A systematic review of 56 test-negative studies found pooled VE against outpatient illness of 61 per cent for the post-2009 H1N1 virus and 54 per cent for influenza B. For H3N2, it was 33 per cent. In older adults, the H3N2 estimate was 24 per cent and its confidence interval included no effect [2]. These historical pooled estimates offer context for understanding why an H3N2-dominated year often feels disappointing; they do not forecast the next season.

The first explanation is antigenic drift. Antigens are the parts of a pathogen recognised by the immune system. In influenza, antibodies commonly target the surface protein haemagglutinin, or HA. As mutations alter exposed portions of HA, antibodies raised against an earlier version may bind less effectively. A drifted virus has changed enough antigenically that existing immunity recognises it less well. H3N2 evolves particularly quickly and contains several co-circulating branches, making selection difficult.

Strain selection is more than reading a genetic family tree. Two viruses with similar sequences can differ in the way antibodies recognise them, while a few strategically placed mutations can cause a large antigenic change. Laboratories test viruses against reference antibodies and increasingly use sequence, structural and computational evidence. Surveillance must also estimate which branch is expanding and likely to dominate months later. A scientifically strong recommendation can still be overtaken by a branch that grows after the decision.

The word mismatch is often used as though it describes a binary failure. In reality, antigenic distance is graded, and a seasonal vaccine contains several components. A drifted H3N2 virus can reduce protection against H3N2 while the H1N1 and influenza B components remain useful. Prior infections and vaccinations may also provide partial cross-protection. In early 2026, a Canadian study estimated about 40 per cent effectiveness against medically attended illness caused by an antigenically distinct H3N2 subclade, despite substantial mismatch [3]. Mismatch lowers expectations; it does not guarantee zero protection.

A second source of variation can enter during manufacturing. Many flu vaccines have historically been produced by growing candidate viruses in fertilised chicken eggs. Influenza viruses adapted to humans do not always grow efficiently there, especially H3N2. As they replicate in eggs, variants with mutations that improve growth can be selected. If those egg-adaptive changes alter an antibody target on HA, the manufactured antigen may become less like the virus circulating among people.

The 2012 to 2013 Canadian season supplied unusually clear evidence. Investigators compared the egg-grown H3N2 vaccine virus with the cell-grown reference strain and circulating viruses. Low effectiveness was associated with a mutation acquired during egg adaptation, while antigenic drift among the circulating viruses did not explain it alone [4]. It was one season, but it established a mechanism that surveillance laboratories now watch closely. The vaccine can diverge from the target on both sides: the wild virus may move, and the production virus may change as it is grown.

Cell-based vaccines grow virus in mammalian cells, avoiding selection in eggs. Recombinant vaccines produce HA protein without growing whole influenza virus. These approaches can reduce egg-adaptation problems, but they do not stop antigenic drift after strain selection. Avoiding eggs does not automatically make every product superior in every season. Observational comparisons can be confounded because recipients differ by age, insurance, location or clinical risk. The manufacturing platform removes one source of mismatch while the others remain.

The recipient contributes another layer. Older immune systems generally respond less vigorously to vaccination, a collection of changes called immunosenescence. Chronic illness and medicines that suppress immunity can also reduce responses. Young children meeting influenza for the first times have a different problem: they may need two doses in their first vaccinated season to establish and boost protection. Pregnancy changes immune physiology, while vaccination also transfers some antibodies to the infant. One formulation and one effectiveness estimate cannot describe all these contexts.

Enhanced vaccines are designed partly for older adults. High-dose products contain more antigen. Adjuvanted vaccines include an ingredient that strengthens or directs the immune response. Recombinant products can also contain more HA than standard-dose vaccines. A 2024 network meta-analysis found that enhanced vaccines as a group offered modestly stronger protection against influenza hospitalisation than standard-dose vaccines in older adults, although direct comparisons among enhanced types did not reveal a clear winner [5]. Product choice can therefore improve protection at the margin without abolishing seasonal variation.

Immune history matters as well. The first influenza infections of childhood can bias later responses, a phenomenon called immune imprinting. Repeated vaccination may reinforce antibodies to familiar features, while also updating protection against the current vaccine strains. Studies of repeat vaccination have produced mixed findings, especially for H3N2. Some report lower relative effectiveness among people vaccinated in consecutive seasons; others find clear added benefit from the current dose. Bias is difficult to exclude because annual vaccine recipients differ systematically from occasional recipients.

A six-season analysis highlighted that difficulty. It found that current-season vaccination generally reduced medically attended flu among people with and without a prior-season dose, and that people vaccinated in the previous season still gained added protection from the current one [6]. This supports annual recommendations. It does not close the research question. The spacing and antigenic similarity of repeated exposures may influence antibody quality, and newer non-egg vaccines could shape memory differently from repeated egg-based products.

Timing creates another trade-off. Protection takes roughly two weeks to develop and can wane over months, particularly in older adults. Vaccinate too early and immunity may be lower during a late peak. Wait too long and the epidemic may arrive first, or the appointment may never happen. Public-health programmes choose a practical window based on expected seasonality, delivery capacity and the groups at risk. An unusually early season can make sound timing look late in retrospect.

The severity of the season also changes how effectiveness feels. During a mild year, a good vaccine may prevent fewer cases because fewer exposures occur. During a severe year, the same VE can prevent many more illnesses, yet clinics may still fill because the starting burden is so large. Coverage matters at population level: a moderately effective vaccine used widely may avert more admissions than a highly effective vaccine received by few. Effectiveness is a property of a vaccine in a particular context; impact combines that performance with uptake and disease pressure.

Uncertainty intervals are essential. An estimate of 40 per cent with a 95 per cent confidence interval from 25 to 52 per cent means the data are compatible with a meaningful range after the model's assumptions. Small subtype samples produce wider intervals. Interim estimates can change as the season grows and circulating viruses shift. Differences of a few percentage points between countries or products may reflect chance or methods. Ranking vaccines from separate observational studies as though they ran one head-to-head trial invites error.

Effectiveness studies also face confounding, a mixing of the vaccine's effect with other differences between groups. People offered vaccination because they are frail may begin with a higher risk of hospitalisation. Conversely, people who accept vaccination may use preventive care more often. Statistical adjustment can reduce these imbalances only for characteristics that were measured well. Randomised trials handle confounding differently but are rarely repeated for every product and outcome each season. The most credible conclusion comes from convergence across designs.

Flu vaccination nevertheless delivers value even when the match is imperfect. A vaccine can protect against another included subtype, reduce the probability of medically attended illness and offer protection against severe outcomes. The relevant decision is not between this season's vaccine and an ideal future vaccine. It is between available vaccination and entering the season without that added layer of immunity. The known limitations explain why research continues; they do not erase the measured benefit.

The longer-term goal is to reduce the annual wager. Researchers are testing vaccines that focus immunity on conserved viral regions, present antigens in new ways or cover a broader range of strains. Faster manufacturing could move strain selection closer to the season. Better global sequencing and antigenic forecasting could improve the choice. Each improvement attacks a different source of variability. No single innovation solves viral evolution, immune ageing and historical exposure at once.

So why does the flu vaccine work better in some years? Sometimes the forecast is closer. Sometimes H1N1 dominates and the more troublesome H3N2 does not. The production process may preserve or distort the chosen antigen, and the people vaccinated may mount stronger or weaker responses. Studies then view those effects through different endpoints and designs. The annual percentage is the visible result of that entire chain, serving as a score and a clue to where the chain can be strengthened next.