A systematic review and meta-analysis of the effects of antibiotic consumption on antibiotic resistance

Brian Bell, François Schellevis, Ellen E. Stobberingh, Herman Goossens, Mike Pringle•View original

OverviewBalancedalloy voice

If you've ever wondered why your doctor hesitates before writing an antibiotic prescription for a sore throat, here's the big picture. Every pill we swallow nudges the bacteria around us. It doesn't just help one person today; it also shapes which strains will be circulating next month and next year. For decades, we've tracked this evolution mostly inside hospitals, because that's where the worst infections show up. But most antibiotics are prescribed in the community. Most infections start there, too. So the crucial question is simple and uncomfortable: when communities take more antibiotics, do community bacteria become more resistant? Bell and colleagues set out to answer exactly that, not anecdotally, but by pulling together the full sweep of community-based evidence. Think of it like pointing a wide-angle lens at everyday antibiotic use and asking whether the background hum of prescribing maps onto the background level of resistance. They didn't limit themselves to one country or one design. They swept through English and several European languages, pulled in gray literature, and kept the inclusion gates wide, so long as two conditions were met: antibiotics were measured in the community, and the bacteria in question were acquired in the community. And because resistance can take time to show up, they required at least a one-month gap between use and measurement of resistance, and they only kept studies that actually tested for a statistical link. To compare wildly different studies on a common footing, they used two lenses. First, a yes or no lens: did the study find a positive relationship between consumption and resistance or not? That gave them a simple map of the landscape. Second, when studies reported enough data, they calculated an odds ratio, which is the odds of resistance in those exposed to antibiotics versus those not exposed, so they could do a proper meta-analysis. The odds ratio is a workhorse metric in epidemiology; an odds ratio of 2 means the odds of resistance are doubled in the antibiotic-exposed group. With those tools, they built a dataset that was both broad and harmonized enough to synthesize. The breadth was real. Across two hundred and forty-three studies, about two-thirds, or one hundred sixty-four, reported a positive association between community antibiotic use and resistance, while the remaining seventy-nine were negative or equivocal. That's the topography. Underneath, the terrain varied: samples spanned children, adults, and mixed groups; individual-level and country-level analyses; and bacteria from respiratory streptococci to gut-dwelling E. coli. Measures of antibiotic exposure came from self-report, medical records, and pharmacy sales. Not everything was clean. Many studies didn't specify the exact drug, and a frustrating share didn't report the time interval clearly. But as a first pass, this map already says something: the positive signal wasn't rare or niche; it was common. Bell's team didn't stop there. They began by asking a blunt question of the full dataset: are positive findings more common than chance? A simple binomial test said yes, resoundingly so. Then they split the world by bacterial family and found a sharper story. For enteric bacteria—the ones in your gut—and for Streptococcus, positive links were significantly more frequent. For Staphylococcus and a catch-all group of Haemophilus and others, they weren't. That nudged the analysis toward where the signal was strongest and most consistent. They also tried to predict which kinds of studies were more likely to report a positive link. They ran correlations, then a logistic regression with those correlated features—child versus adult samples, antibiotic class, region, and so on. No single factor stood alone after controlling for the others. However, as a set, they mattered. The model distinguished positive from non-positive results better than chance, but only modestly, explaining a little over ten percent of the variation. That's a hint about the world we're in: complex, messy, with many moving parts. For the harder-edged synthesis, the team zoomed in on the eighty-eight studies that let them compute a common effect size. That subset spanned three familiar designs: cross-sectional snapshots, case-control comparisons, and cohorts that follow people over time. They left out ecological studies, which are those at the level of regions or countries, because their sheer size would swamp the others and tilt the math. The question at hand was: when you line up comparable odds ratios across diverse community studies, what's the pooled signal? It was big and clear. The combined odds of resistance were more than doubled with prior community antibiotic exposure—an odds ratio of 2.33—with a tight confidence interval and a z-statistic so high that it leaves little doubt that this wasn't a fluke. To translate that out of stats-speak: across those eighty-eight studies, people or settings with more antibiotic use had substantially higher odds of resistant bacteria. The pattern held no matter how you cut it by study design. Cross-sectional studies landed around 2.46, cohorts closer to 2.93, and case-control studies about 2.26. And when they checked whether a single influential study was doing the heavy lifting—Schneider-Lindner and colleagues from 2007—they dropped it and saw the pooled estimate climb rather than shrink, to 2.74. So the core message didn't depend on one outlier. Then came the question that always follows a big pooled effect: how much are these studies disagreeing with one another, and can we explain it? The heterogeneity was high. A chi-squared test flagged it, and the I-squared statistic—think of that as the share of variability not due to random sampling—sat at roughly three-quarters. That's a lot. High heterogeneity isn't fatal, but it tells you the world isn't uniform. Different bugs, different drugs, different behaviors, different health systems. So the team went hunting for structure in the noise. They started by asking how study-level odds ratios correlated with features like the bacterial species, the region, and the antibiotic class. That exploratory step lit up some suspects: enteric bacteria tended to show stronger associations, staph studies and those centered on methicillin-resistant Staphylococcus aureus—MRSA—tended to be weaker, and southern Europe leaned stronger than northern Europe. Quinolone-resistant E. coli—the workhorse gut bacterium resistant to a common class of broad-spectrum antibiotics—showed up as a hotspot. But correlations are just signposts. To sort out what stands up when you consider everything together, they ran a meta-regression. If you haven't met meta-regression, here's the gist. It's ordinary regression's bigger cousin, used on the effect sizes from multiple studies, weighted by how precise each study is. The more precise a study, the more influence it gets. Bell and colleagues fed in ten predictors and asked which, independently, helped explain why some studies found stronger links between use and resistance. Five mattered. Studies that included both adults and children leaned stronger. Studies in southern Europe leaned stronger, too. Quinolone-resistant E. coli amplified the association. On the flip side, studies focusing on beta-lactams—the penicillin family—and studies on MRSA pulled the link down. Why would that be? The age mix clue probably reflects real-world transmission dynamics. Kids have high contact rates and share bacteria readily with adults, so households and daycares become engines of both colonization and selection. Southern Europe's stronger signal fits with known regional prescribing patterns; higher community antibiotic use has been documented there historically, and the ecology of resistance follows pressure. Quinolones are potent, broad-spectrum drugs that hit the gut hard. When they're used widely, E. coli in the community feels that pressure quickly. Meanwhile, the weaker link for beta-lactams and MRSA points to a different story. MRSA spreads a lot in hospitals and via skin and soft tissue routes, and many MRSA dynamics are decoupled from simple outpatient prescribing totals. Additionally, beta-lactams cover a vast, heterogeneous class; lumping them together may obscure important differences. A quick pause on how we read these estimates. Odds ratios aren't risks, and they aren't destiny. They're a compact way to describe how much more common resistance is in the presence of prior antibiotics in these studies. And because this is a meta-analysis, each study's effect gets pulled toward the center by the rest. Fixed-effects pooling—the approach used here—assumes that, after accounting for measured features, there's one common effect. Bell's team chose that on purpose so they could ask whether study characteristics explained the variability in a follow-on meta-regression. Residual heterogeneity still loomed, meaning we can't fold reality down to a single number and be done with it. But the center of gravity is hard to ignore. What about bias? Publication bias is the perennial suspicion: maybe only positive studies get written up. Funnel plots of the eighty-eight study subset suggested some bias. However, step back to that two hundred forty-three study map. About a third of studies were negative or equivocal, and a simple test showed the surplus of positives wasn't what you'd expect if the field were balanced on a knife edge. That doesn't eliminate bias, but it dulls the sharpest edge of the worry. We should also be honest about measurement. Many studies relied on self-reported antibiotic use, which is noisy—people forget. In this set, roughly four in ten did that. Others used medical records or prescription sales, which are better, but each brings its own blind spots: a filled prescription isn't always a swallowed pill. The time between use and resistance wasn't always clear, and in a sizable slice of reports, the antibiotic class wasn't even specified. All of that adds blur. And remember, the ecological studies—those at the region or country level—were excluded from the core meta-analysis to keep comparisons fair, even though they're important for policy. So what's the take-home? Put plainly, when communities use more antibiotics, their circulating bacteria are more likely to be resistant. The pooled effect is big—on the order of doubling the odds—and it shows up across study designs, across bacteria that matter in daily life, and across countries. The relationship is not uniform. It's stronger in southern Europe than in northern Europe or the United States. It's pronounced in gut bacteria, especially quinolone-resistant E. coli. It's muddier for MRSA and for the grab-bag of beta-lactams as a whole. But the backbone of the finding doesn't snap when you apply pressure. Why does this matter outside a journal club? Because if selection pressure in the community drives resistance in the community, then outpatient stewardship isn't a side quest; it's central. Reducing unnecessary prescriptions should, over time, relieve that pressure. The evidence base here can't name the magic threshold where resistance bends, and it can't guarantee how quickly curves will fall. It does say, loudly, that the lever exists. There's one more layer worth mentioning, and it's a to-do list for all of us who care about getting this right. Better measurement would help—more precise tracking of which antibiotics are used, when, and for how long; clearer documentation of the lag between use and resistance; and routine linkage of individual exposure to individual isolates where possible. Co-selection—the way one drug can select for resistance to another—was rarely examined and needs daylight. Additionally, surveillance that stitches together community clinics, pharmacies, and labs will make these analyses faster and more actionable. Let's end where we began, with that doctor in the clinic and the patient in front of them. Bell and colleagues don't tell us to stop treating infections. They give us a community-scale mirror. Every unnecessary course of antibiotics is a small nudge to the evolutionary dial. Enough nudges, and the soundtrack of everyday bacteria changes. The good news in their synthesis is that the link is clear enough to act on. The hard news is that acting on it means investing in boring, essential work: prudent prescribing, better data, and patience. Evolution is relentless. Stewardship has to be, too.

For decades, we've tracked this evolution mostly inside hospitals, because that's where the worst infections show up. But most antibiotics are prescribed in the community. Most infections start there, too.

So the crucial question is simple and uncomfortable: when communities take more antibiotics, do community bacteria become more resistant?

Bell and colleagues set out to answer exactly that, not anecdotally, but by pulling together the full sweep of community-based evidence. Think of it like pointing a wide-angle lens at everyday antibiotic use and asking whether the background hum of prescribing maps onto the background level of resistance. They didn't limit themselves to one country or one design.

They swept through English and several European languages, pulled in gray literature, and kept the inclusion gates wide, so long as two conditions were met: antibiotics were measured in the community, and the bacteria in question were acquired in the community. And because resistance can take time to show up, they required at least a one-month gap between use and measurement of resistance, and they only kept studies that actually tested for a statistical link.

To compare wildly different studies on a common footing, they used two lenses. First, a yes or no lens: did the study find a positive relationship between consumption and resistance or not? That gave them a simple map of the landscape.

Second, when studies reported enough data, they calculated an odds ratio, which is the odds of resistance in those exposed to antibiotics versus those not exposed, so they could do a proper meta-analysis. The odds ratio is a workhorse metric in epidemiology; an odds ratio of 2 means the odds of resistance are doubled in the antibiotic-exposed group. With those tools, they built a dataset that was both broad and harmonized enough to synthesize.

The breadth was real. Across two hundred and forty-three studies, about two-thirds, or one hundred sixty-four, reported a positive association between community antibiotic use and resistance, while the remaining seventy-nine were negative or equivocal. That's the topography.

Underneath, the terrain varied: samples spanned children, adults, and mixed groups; individual-level and country-level analyses; and bacteria from respiratory streptococci to gut-dwelling E. coli. Measures of antibiotic exposure came from self-report, medical records, and pharmacy sales. Not everything was clean.

Many studies didn't specify the exact drug, and a frustrating share didn't report the time interval clearly. But as a first pass, this map already says something: the positive signal wasn't rare or niche; it was common.

Bell's team didn't stop there. They began by asking a blunt question of the full dataset: are positive findings more common than chance? A simple binomial test said yes, resoundingly so.

Then they split the world by bacterial family and found a sharper story. For enteric bacteria—the ones in your gut—and for Streptococcus, positive links were significantly more frequent. For Staphylococcus and a catch-all group of Haemophilus and others, they weren't.

That nudged the analysis toward where the signal was strongest and most consistent.

They also tried to predict which kinds of studies were more likely to report a positive link. They ran correlations, then a logistic regression with those correlated features—child versus adult samples, antibiotic class, region, and so on. No single factor stood alone after controlling for the others.

However, as a set, they mattered. The model distinguished positive from non-positive results better than chance, but only modestly, explaining a little over ten percent of the variation. That's a hint about the world we're in: complex, messy, with many moving parts.

For the harder-edged synthesis, the team zoomed in on the eighty-eight studies that let them compute a common effect size. That subset spanned three familiar designs: cross-sectional snapshots, case-control comparisons, and cohorts that follow people over time. They left out ecological studies, which are those at the level of regions or countries, because their sheer size would swamp the others and tilt the math.

The question at hand was: when you line up comparable odds ratios across diverse community studies, what's the pooled signal?

It was big and clear. The combined odds of resistance were more than doubled with prior community antibiotic exposure—an odds ratio of 2.33—with a tight confidence interval and a z-statistic so high that it leaves little doubt that this wasn't a fluke. To translate that out of stats-speak: across those eighty-eight studies, people or settings with more antibiotic use had substantially higher odds of resistant bacteria.

The pattern held no matter how you cut it by study design. Cross-sectional studies landed around 2.46, cohorts closer to 2.93, and case-control studies about 2.26. And when they checked whether a single influential study was doing the heavy lifting—Schneider-Lindner and colleagues from 2007—they dropped it and saw the pooled estimate climb rather than shrink, to 2.74. So the core message didn't depend on one outlier.

Then came the question that always follows a big pooled effect: how much are these studies disagreeing with one another, and can we explain it? The heterogeneity was high. A chi-squared test flagged it, and the I-squared statistic—think of that as the share of variability not due to random sampling—sat at roughly three-quarters.

That's a lot. High heterogeneity isn't fatal, but it tells you the world isn't uniform. Different bugs, different drugs, different behaviors, different health systems.

So the team went hunting for structure in the noise. They started by asking how study-level odds ratios correlated with features like the bacterial species, the region, and the antibiotic class. That exploratory step lit up some suspects: enteric bacteria tended to show stronger associations, staph studies and those centered on methicillin-resistant Staphylococcus aureus—MRSA—tended to be weaker, and southern Europe leaned stronger than northern Europe.

Quinolone-resistant E. coli—the workhorse gut bacterium resistant to a common class of broad-spectrum antibiotics—showed up as a hotspot. But correlations are just signposts. To sort out what stands up when you consider everything together, they ran a meta-regression.

If you haven't met meta-regression, here's the gist. It's ordinary regression's bigger cousin, used on the effect sizes from multiple studies, weighted by how precise each study is. The more precise a study, the more influence it gets.

Bell and colleagues fed in ten predictors and asked which, independently, helped explain why some studies found stronger links between use and resistance. Five mattered. Studies that included both adults and children leaned stronger.

Studies in southern Europe leaned stronger, too. Quinolone-resistant E. coli amplified the association. On the flip side, studies focusing on beta-lactams—the penicillin family—and studies on MRSA pulled the link down.

Why would that be? The age mix clue probably reflects real-world transmission dynamics. Kids have high contact rates and share bacteria readily with adults, so households and daycares become engines of both colonization and selection.

Southern Europe's stronger signal fits with known regional prescribing patterns; higher community antibiotic use has been documented there historically, and the ecology of resistance follows pressure. Quinolones are potent, broad-spectrum drugs that hit the gut hard. When they're used widely, E. coli in the community feels that pressure quickly.

Meanwhile, the weaker link for beta-lactams and MRSA points to a different story. MRSA spreads a lot in hospitals and via skin and soft tissue routes, and many MRSA dynamics are decoupled from simple outpatient prescribing totals. Additionally, beta-lactams cover a vast, heterogeneous class; lumping them together may obscure important differences.

A quick pause on how we read these estimates. Odds ratios aren't risks, and they aren't destiny. They're a compact way to describe how much more common resistance is in the presence of prior antibiotics in these studies.

And because this is a meta-analysis, each study's effect gets pulled toward the center by the rest. Fixed-effects pooling—the approach used here—assumes that, after accounting for measured features, there's one common effect. Bell's team chose that on purpose so they could ask whether study characteristics explained the variability in a follow-on meta-regression.

Residual heterogeneity still loomed, meaning we can't fold reality down to a single number and be done with it. But the center of gravity is hard to ignore.

What about bias? Publication bias is the perennial suspicion: maybe only positive studies get written up. Funnel plots of the eighty-eight study subset suggested some bias.

However, step back to that two hundred forty-three study map. About a third of studies were negative or equivocal, and a simple test showed the surplus of positives wasn't what you'd expect if the field were balanced on a knife edge. That doesn't eliminate bias, but it dulls the sharpest edge of the worry.

We should also be honest about measurement. Many studies relied on self-reported antibiotic use, which is noisy—people forget. In this set, roughly four in ten did that.

Others used medical records or prescription sales, which are better, but each brings its own blind spots: a filled prescription isn't always a swallowed pill. The time between use and resistance wasn't always clear, and in a sizable slice of reports, the antibiotic class wasn't even specified. All of that adds blur.

And remember, the ecological studies—those at the region or country level—were excluded from the core meta-analysis to keep comparisons fair, even though they're important for policy.

So what's the take-home? Put plainly, when communities use more antibiotics, their circulating bacteria are more likely to be resistant. The pooled effect is big—on the order of doubling the odds—and it shows up across study designs, across bacteria that matter in daily life, and across countries.

The relationship is not uniform. It's stronger in southern Europe than in northern Europe or the United States. It's pronounced in gut bacteria, especially quinolone-resistant E. coli.

It's muddier for MRSA and for the grab-bag of beta-lactams as a whole. But the backbone of the finding doesn't snap when you apply pressure.

Why does this matter outside a journal club? Because if selection pressure in the community drives resistance in the community, then outpatient stewardship isn't a side quest; it's central. Reducing unnecessary prescriptions should, over time, relieve that pressure.

The evidence base here can't name the magic threshold where resistance bends, and it can't guarantee how quickly curves will fall. It does say, loudly, that the lever exists.

There's one more layer worth mentioning, and it's a to-do list for all of us who care about getting this right. Better measurement would help—more precise tracking of which antibiotics are used, when, and for how long; clearer documentation of the lag between use and resistance; and routine linkage of individual exposure to individual isolates where possible. Co-selection—the way one drug can select for resistance to another—was rarely examined and needs daylight.

Additionally, surveillance that stitches together community clinics, pharmacies, and labs will make these analyses faster and more actionable.

Let's end where we began, with that doctor in the clinic and the patient in front of them. Bell and colleagues don't tell us to stop treating infections. They give us a community-scale mirror.

Every unnecessary course of antibiotics is a small nudge to the evolutionary dial. Enough nudges, and the soundtrack of everyday bacteria changes. The good news in their synthesis is that the link is clear enough to act on.

The hard news is that acting on it means investing in boring, essential work: prudent prescribing, better data, and patience. Evolution is relentless. Stewardship has to be, too.

Related lectures