What "the flu shot is 40% effective" actually measures

"Vaccine effectiveness" is not the share of people the vaccine protects. It is the percentage by which vaccination reduced the risk of one specific, measured outcome — a doctor's visit, or a hospital admission, with laboratory-confirmed influenza — in vaccinated people compared with unvaccinated people, over one season, in one population. For 2025/26 in the United States, that figure was 22-34% against outpatient visits in adults and 30% against hospitalisation, and 38-41% and 41% respectively in children and adolescents [s2].

The vagueness of "does the flu shot work" mostly dissolves once you know which outcome is being counted.

How the number is produced

Interim US estimates were calculated from three national respiratory virus vaccine effectiveness networks, covering patients with acute respiratory illness who presented for outpatient visits or were hospitalised, using a test-negative case-control design [s2].

That design is the reason these estimates are comparable across countries and seasons. Everyone in the analysis has already done the same thing: felt ill enough with a respiratory illness to seek care, and been tested. Those who test positive for influenza are the cases; those who test negative are the controls. The comparison is then between how many in each group had been vaccinated.

The point of restricting to people who sought care and were tested is that it removes a large source of distortion. People who get vaccinated differ systematically from people who do not — in age, health, occupation, and how readily they see a doctor. Comparing tested patients with each other, rather than vaccinated people with the general population, holds much of that constant.

What it does not do is measure everything. A test-negative estimate answers one question about one endpoint. Effectiveness against a hospital admission and effectiveness against a mild outpatient illness are different quantities, and so are effectiveness against influenza A and against influenza B.

What the 2025/26 season produced

The season was a hard test, because the dominant virus had drifted away from the vaccine strain. Across Europe the 2025/26 season was dominated by influenza A(H3N2), with most sequenced viruses belonging to subclade K, genetically drifted from the vaccine virus [s1].

Despite that, estimates from nine European studies covering 19 countries put all-age influenza A vaccine effectiveness at 25-45% for outpatient and hospital settings combined — described by the authors as similar to other seasons — with the highest estimates among children, at 47-72% [s1]. Their conclusion was that vaccination should be encouraged and complemented by other infection prevention and control measures [s1].

Canada's sentinel practitioner network estimated that vaccination reduced the risk of medically-attended acute respiratory illness due to the predominant A(H3N2) viruses, including the antigenically distinct subclade K, by about 40% relative to unvaccinated people, and by about 30% against A(H1N1)pdm09; there were too few influenza B cases for an interim estimate [s3]. The Canadian authors interpret meaningful protection against subclade K despite substantial vaccine mismatch in light of immuno-epidemiological factors including potential viral glycosylation, imprinting and pre-existing immunity [s3].

The US estimates break down further. Among children and adolescents under 18, effectiveness was 38-41% against outpatient visits and 41% against hospitalisation; among adults 18 and over, 22-34% against outpatient visits and 30% against hospitalisation [s2]. Against influenza A specifically, children ranged from 37% (outpatient) to 42% (hospitalisation), adults from 30% (hospitalisation) to 34% (outpatient) [s2]. Against A(H3N2) in children, effectiveness was 35% against outpatient visits and 38% against hospitalisation [s2]. Against influenza B, the numbers were higher: 45-71% among children and adolescents and 63% among adults [s2].

Other estimates in the US analysis were either not statistically significant or not reportable [s2].

How to read a percentage like this

Three things follow from the design, and all three are routinely lost in summary.

First, effectiveness is a relative reduction, not an individual probability. A 30% estimate does not mean 30% of vaccinated people are protected and 70% are not; it means the vaccinated group experienced roughly 30% fewer of that specific outcome than the unvaccinated group did.

Second, the number moves with the endpoint. The same season, the same population and the same vaccine can yield one figure for outpatient illness and another for hospitalisation, and both are correct [s2]. Comparing a hospitalisation estimate from one country with an outpatient estimate from another is not a comparison of vaccine quality.

Third, these are interim estimates, produced mid-season with partial data. Their confidence intervals are wide enough that some strata could not be reported at all [s2].

The US analysis notes that interim effectiveness was lower in 2025/26 than in recent seasons while still demonstrating protection [s2]. That is a more precise, and less satisfying, answer than either "the flu shot works" or "the flu shot failed this year" — and it is the one the data supports.

Whether and when to be vaccinated is a decision for an individual and their clinician, informed by age, health conditions and local recommendations. This article describes how the effectiveness figures in circulation are generated and what they do and do not measure.