Antimicrobial Drug Use and Resistance in Europe

If a country prescribes more antibiotics, its bacteria should develop more resistance. That logic seems almost too obvious to test. However, for decades, European public health officials couldn't actually prove it at the national scale because the comparable data simply didn't exist. This most intuitive relationship in infectious disease remained unconfirmed. This paper represents the first systematic attempt to close that gap, using six years of surveillance data from 21 countries. Nienke van de Sande-Bruinsma and colleagues set out to test whether country-level antibiotic consumption actually predicts country-level resistance — not in a single hospital, nor in a single city, but aggregated across entire national populations. To do this, they linked two European surveillance systems that had only recently matured enough to support this kind of comparison. The European Surveillance of Antimicrobial Consumption, or ESAC, tracked how much antibiotic each country prescribed. The European Antimicrobial Resistance Surveillance System, known as EARSS, tracked how resistant the bacteria were getting. Together, across 21 countries and the years 2000 to 2005, they provided the raw material for an answer. The consumption side of that equation was measured in defined daily doses per 1,000 inhabitants per day, which we can abbreviate to DID. You can think of it as a population-adjusted prescription rate: how many people, out of every thousand, are taking a standard course of a given drug on any given day. The resistance side focused on two pathogens chosen because they reflect community, not hospital, pressure: Streptococcus pneumoniae, the bacterium behind many cases of pneumonia and meningitis, and Escherichia coli, which is responsible for most urinary tract infections. For each pathogen, the team pulled blood-culture isolates from EARSS and calculated the proportion of samples that were nonsusceptible — this category combines both intermediate and fully resistant results. They distilled each country's resistance profile into a single score: the sum of its quartile rankings across three compound-pathogen combinations. They also tested 11 different exposure-outcome intervals, from same-year comparisons up to a two-year lag, to see whether time delays between prescribing and resistance emergence affected the results. Before getting to what the regression found, it helps to see the raw geography. Across Europe, outpatient antibiotic prescribing varied by a factor of more than three. In 2004, the Netherlands reported just 9.7 defined daily doses per 1,000 inhabitants per day. Greece reported 33.4. France was at 27.1, Belgium at 22.9, Germany at 11.0, and Sweden at 15.0. Northern European countries clustered at the low end, while southern and some eastern countries clustered at the high end. When you overlaid the resistance maps on the consumption maps, the geographic gradients tracked each other almost point for point. The countries prescribing the most antibiotics were the countries with the highest resistant bacteria. Spain, Hungary, and France had the highest overall resistance scores, while Sweden and the Netherlands had the lowest. The contrasts within specific pathogen-drug pairs were stark. France reported thirty-six percent of its pneumococcal isolates as penicillin-nonsusceptible, while the Netherlands reported just 1.3 percent. That's a twenty-seven fold difference. Erythromycin-nonsusceptible pneumococci ranged from forty-one percent in France down to two percent in the Czech Republic. For fluoroquinolone-resistant Escherichia coli, Portugal was at twenty-nine percent and Iceland at three percent. At this point, the map is already telling a story. The regression was built to ask whether that story holds up statistically. To run a linear regression on resistance data, the authors had to transform the raw proportions. A proportion is bounded between zero and one, which breaks the assumptions of ordinary linear regression. So they converted each country's resistance rate into the natural log of the odds of resistance — mathematically, the natural logarithm of R divided by one minus R — which stretches those bounded proportions onto an unbounded scale. That transformation let them model resistance as a proper continuous outcome. The results for two of the three compound-pathogen combinations were specific and strong. For penicillin use and penicillin-nonsusceptible Streptococcus pneumoniae, country-level penicillin consumption alone explained sixty-one percent of the observed variance in resistance, with a regression gradient of 0.29 and a p-value of 0.0002. The median correlation coefficient across all eleven time intervals was 0.78. That is a tight relationship. For fluoroquinolone consumption and fluoroquinolone-resistant Escherichia coli, the association was meaningful but more modest: fluoroquinolone use explained thirty-six percent of the variance, with a median correlation of 0.60. The paper emphasizes that this effect was specific — it was fluoroquinolone use, not total antibiotic use, that drove the association for Escherichia coli. The third combination, erythromycin-nonsusceptible Streptococcus pneumoniae, told a messier story. In univariate analysis, the strongest single predictor wasn't erythromycin or MLS-class agents at all — it was other beta-lactams, mainly cephalosporins, which explained forty-eight percent of the variance. However, when the team added other drug classes to the model, the apparent effect of beta-lactams shrank by about forty percent and lost statistical significance. That pattern suggests co-selection: bacteria developing resistance to one drug class due to pressure from another. The erythromycin-nonsusceptible finding is real in the data, but not clean enough to interpret as a straightforward causal signal. Critically, when the team tested all eleven exposure-outcome intervals — from same-year up to a two-year lag — they found no statistically significant time dependence. The associations didn't sharpen or blur depending on how long a delay they assumed between prescribing and resistance emergence. That stability is actually reassuring. It means the signal isn't just a fluke of timing, and it's why the authors used median correlations across all intervals rather than cherry-picking the strongest one. Van de Sande-Bruinsma and colleagues are candid about what this kind of ecologic analysis cannot resolve. Country averages mask enormous within-country variation – a national defined daily dose figure papers over the difference between a high-prescribing urban clinic and a more cautious rural practice. The two surveillance systems also cover different sampling populations in ways that can introduce inaccuracies. EARSS data come from blood-culture isolates from more than nine hundred laboratories serving roughly one thousand four hundred hospitals. While this provides breadth, it means the resistance picture is limited to invasive infections. Unmeasured confounders — such as differences in hygiene standards, vaccination rates, diagnostic habits, and infection control practices — vary between countries and weren't controlled for. For the erythromycin combination, co-selection between drug classes muddies the causal picture considerably. There's also a statistical limitation worth naming plainly: with only seventeen to twenty-one country-level data points per regression, the models have limited power. The ecologic fallacy is always a risk — a pattern that holds between countries doesn't necessarily hold between individuals within those countries. These aren't reasons to dismiss the findings; they're reasons to be precise about what the findings actually claim. What they claim is this: at the national population level, higher outpatient antibiotic consumption is associated with higher resistance, the relationship is class-specific, and it is stable over time. This is enough, according to the paper, to support national-level policy action. The mechanisms they point to are practical ones: antibiotic stewardship programs that promote prudent prescribing, tighter restrictions on over-the-counter antibiotic sales, public awareness campaigns, and systematic monitoring of the effects of any intervention. They describe antimicrobial effectiveness as a common good — one that cannot be taken for granted and increasingly functions like a nonrenewable resource. The intuition that more prescribing produces more resistance turned out to be correct, in the specific, quantifiable, cross-national sense that van de Sande-Bruinsma and colleagues tested. Penicillin use explained sixty-one percent of the variance in penicillin-resistant pneumococci across twenty-one European countries. Fluoroquinolone use explained thirty-six percent of the variance in resistant Escherichia coli. Neither number explains everything, but both numbers are large enough to tell policymakers something actionable: if you want to influence the resistance map, start by moving the prescription rate. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.