Europe Cut Farm Antibiotics by 53%. The Proof It Helped People Is Thin.

Between 2011 and 2022, sales of veterinary antibiotics fell by 53% in the 25 European countries that reported every year [1]. Resistant E. coli in Dutch pigs, chickens and calves fell too. Did resistant infections in people fall as well? I read the public data to find out. My answer is: probably a little, in some bacteria, and nobody has shown it well.

I should say what I expected. I held a position that cutting routine antibiotic use in livestock lowers resistance in those animals, with a smaller and slower effect in people. The working thesis for this post said resistance fell "in parallel" in people. After reading, I keep the first half and I weaken the second half. One Dutch case runs the other way.

The question

Two questions sit inside one. First: does less antibiotic use in farm animals lower resistant bacteria in those animals? Second: does that lower resistant infections in people?

The first question has a plain biological reason. Antibiotics select for resistant bacteria in the gut of the animal that receives them. Less drug gives less selection. The second question needs a chain: animal bacteria reach people (through meat, handling or the environment), and resistance genes in them matter for human disease. Each link can break.

Data and where it came from

Four public sources carry the argument. I name the evidence type for each.

• Sales data (surveillance). The EMA's ESVAC reports. For 25 countries with continuous data, sales fell from 161.2 mg per population correction unit (mg/PCU) in 2011 to 75.8 mg/PCU in 2022 [2]. The EMA reports this as a 53% fall [1]. Some classes fell more: polymyxins by 81%, other quinolones by 90%, third and fourth generation cephalosporins by 49%. Fluoroquinolones fell by only 25% [1].
• Newer sales data (surveillance). ESVAC ended after its 2022 report. Mandatory sales and use reporting from all EU states began in 2025 (ESUAvet) [1]. Its first full report shows sales for food-producing animals up 5% in 2024 compared with 2023 [3].
• Country panel (observational, ecological). A Frontiers in Public Health study of 31 European countries, 2008 to 2018, linking animal and human antibiotic use to resistance in Salmonella, Campylobacter and E. coli [4].
• Dutch case studies (surveillance and ecological). A 2004 to 2020 study of Campylobacter in people and broilers [5], a 2026 study of E. coli in Dutch livestock [6], and the Dutch national institute (RIVM) summary of 2019 [7].

I also use Tang and colleagues' 2017 systematic review [8]. I read only abstract-level summaries of it, not the full paper. I treat its numbers as a reported result, not as something I checked.

No randomised trial exists here. Nobody randomises countries to cut farm antibiotics. Every human-side number below is observational.

Method

I did two things. I read each source for what it measures. And I did one calculation myself, without the Lab, so a reader can repeat it with a calculator.

The Frontiers study reports elasticities. An elasticity says: if use rises by 1%, resistance rises by this many percent. The authors give bounds, because two model designs (fixed effects and lagged dependent variable) bracket the causal effect [4]. For animal use, the bounds are 0.22 to 0.41 for resistance in animals and 0.03 to 0.40 for resistance in people [4].

I convert these to a 53% cut in use. The formula is:

$\text{relative resistance} = (1 - 0.53)^{e} = 0.47^{e}$

I assume the elasticity holds over a large change and in both directions. The authors did not test that. It is my extrapolation, so treat it as an illustration.

Result

Livestock: the effect is real, then it flattens

The Frontiers bounds give these results by my calculation:

Outcome	Elasticity	Implied change after a 53% cut
Resistance in animals, low bound	0.22	about 15% lower
Resistance in animals, high bound	0.41	about 27% lower
Resistance in people, low bound	0.03	about 2% lower
Resistance in people, high bound	0.40	about 26% lower

(For example, $0.47^{0.22} = 0.85$, so 15% lower; $0.47^{0.03} = 0.98$, so 2% lower.)

The animal range is 15 to 27%. The human range is 2 to 26%. The ranges overlap at the top. But the lower bound for people is close to zero, and the animal range does not go near zero. That asymmetry is the finding. It agrees with my prior, but the human upper bound is wide enough that I cannot call the effect small with confidence.

Tang's meta-analysis points the same way. As summarised, restricting use cut resistance in animals by 10 to 15%, and the pooled human prevalence was 24% lower in intervention groups (95% CI 6 to 42%) [8]. The human interval is wide. It does not exclude a 6% effect. I do not know how many of the human studies were comparisons of communities with and without restriction, so I rate this evidence moderate for animals and weak to moderate for people.

The Dutch livestock data fit the animal story. Dutch sales fell almost 70% from 2009 to 2019 [7]. Resistance in indicator E. coli from broilers, fattening pigs and veal calves fell between 2010 and 2018 for most drugs [6]. The largest recent fall in ESBL-producing gut bacteria was in broilers and chicken meat [7].

Then the curve flattened. From 2019 to 2023, no significant change in resistance appeared for most drugs, and ampicillin, sulfonamide and trimethoprim resistance stayed high in broilers and pigs, even as use kept falling [6]. This is the first lesson for anyone planning policy: the first big cuts bring the largest drop. Later cuts may bring little, because resistance genes persist when they are not costly to the bacterium, or sit on the same element as other genes that drugs still select for. That reasoning is mine, and the 2026 paper's abstract as I read it reports the pattern, not the mechanism.

People: the Dutch case runs the wrong way

The Dutch Campylobacter study is the best test I found of the human link, because it uses a country with a steep cut and long records. From 2004 to 2020, human C. jejuni resistance rose by 66% for tetracycline and 46% for fluoroquinolones. In C. coli, tetracycline resistance rose by 75%, macrolide by 72% and fluoroquinolone by 45% [5]. Livestock use fell throughout. In broilers, tetracycline and fluoroquinolone resistance stayed "stable at a high level" (42 to 81% and 43 to 92%) [5].

Two points matter. First, the authors found no significant link between tetracycline use in broilers and resistance in broiler isolates. Second, broiler and human resistance were moderately to strongly correlated [5]. So the chain from chicken to person looks present. But the first link, less use giving less resistance in the chicken, did not appear for Campylobacter. The authors conclude that cutting use in livestock alone may not suffice [5].

Note the limits they name. The data cover Dutch-raised animals only, so imported meat is not counted. Most human cases have no travel history. And ecological designs cannot assign cause [5]. Human fluoroquinolone resistance may also follow human drug use, travel or imports. I cannot separate these.

For other bacteria the Dutch national summary is calmer. In 2019, hardly any resistance increase was found in people, and some resistance fell slightly [7]. That is "stable", not "fell in parallel". I do not count stable as evidence of a benefit, because I have no counterfactual: what would human resistance have done without the cut?

Sensitivity: which assumption moves the result most

I tested five assumptions. They differ in how much they move the conclusion.

1. The human elasticity bound. This moves the result most. Using the lower bound (0.03), the human effect of a 53% cut is about 2%. Using the upper bound (0.40), it is about 26% [4]. The same data support both. A 13-fold spread in one number decides whether the policy "worked" for people. Everything else is smaller.
2. Which bacterium. Salmonella and E. coli plausibly follow the food chain. Campylobacter in the Netherlands did not follow [5]. A pooled European average hides this. If a reader asks "did human resistance fall?", the honest answer depends on the bacterium and the drug.
3. Sales are not use. ESVAC counts sales by weight per PCU. A kilogram of one drug is not the same dose as a kilogram of another. The 2024 rise of 5% could reflect stockpiling, outbreaks or changes in animal numbers, and the EMA says so [3]. The EMA also says variability between countries limits EU-level aggregation [3]. Since new ESUAvet data cover 2023 onward, the 2011 to 2022 series and the new series are not simple to join.
4. The 53% base. It covers 25 countries with continuous data, not all of Europe [1][2]. Countries that cut most may differ from countries that cut least in ways that also drive resistance, such as farm structure and human drug use. The panel study cannot remove that fully, and it names missing environmental data and surveillance gaps between human and animal systems [4].
5. Direction of causes. Human drug use also raises resistance in people, with elasticities of 0.03 to 0.16 in the panel [4]. If human use fell during the same years, some human gains come from that, not from farms. I did not have human use series for this post, so I cannot split the two. This is a gap in my work.

My own computation is a back-of-envelope conversion of published elasticities. It is not a new analysis. I did not re-run the models and I did not pull the national tables. That work is pending (see the follow-up).

What I now think

For livestock, I rate the evidence moderate to good. Public surveillance shows sales and resistance falling together, and the Dutch time series show it clearly from 2010 to 2018 [6][7]. The flattening after 2018 shows that the effect has a floor, at least at current policy [6].

For people, I rate it weak. The meta-analysis gives 24% (95% CI 6 to 42%) [8]. The country panel gives a range from near zero to about a quarter [4]. The Dutch Campylobacter series shows rising human resistance during falling livestock use [5]. All of it is ecological. My earlier prior, a smaller and slower effect in people, still stands, but I now put less weight on "slower" and more on "uncertain, and maybe absent for some bacteria".

What would change my mind is a study that follows the same strains from farm to patient. Whole-genome sequencing of resistant isolates in people and animals, in a country before and after a sharp cut, would test the chain directly. Counting chickens and counting patients, then drawing a line between them, cannot.

The link to human health runs both ways here. Farm cuts are cheap insurance if they help people even a little, since they cost no patient a treatment they need. But the Dutch case warns against selling them as the main fix. For some pathogens, stopping transmission may do more than the drug cut. The 5% rise in 2024 [3] is a small, early signal. It is the first test of whether Europe can hold the gains.