Antibiotics get used a lot in dogs and cats (and other species) that undergo surgery. A lot of the time, it’s unnecessary. Often, I suspect people know it’s unnecessary but still do it out of habit, fear of complications, fear of complaints from owners, lack of consideration of the potential downsides, and because we are programmed to ‘do something’ even though not doing something may be the best approach.

A big challenge we have with antimicrobial prophylaxis guidelines is the lack of good studies. We have lots of small observational studies for various types of surgery. Those are useful but provide quite low certainty evidence. We’d love to have randomized controlled trials but those are expensive and complicated, and funding for this area is poor. So, we’re not likely to get them.

Does that mean we’re stuck with no evidence? No, it just means we have different types of evidence and less certainty.

A major challenge for any type of surgical site infection study is the required sample size. Since surgical site infections are uncommon after most procedures, we’d need very large sample sizes to detect differences, or to confidently say there’s no difference (non-inferiority trial). Studies of hundreds to thousands of animals are not realistically going to be done often, if ever.

I frequently have discussions with people that want to do studies looking at antibiotics and surgical infections. Once we go over the numbers, people quickly get turned off because sample size calculations show the required size would be way too big for what people can do.

That can be disheartening, but it doesn’t mean study is the study is futile. I want these studies done even if they are small and can’t answer all the questions. It’s the old ‘don’t let perfection be the enemy of progress’ argument. We just have to have realistic expectations.

We recently published a commentary that highlighted the value of small studies. A single small study may not be able to answer our big questions, but we can put data together from small but well designed studies and draw stronger conclusions through meta-analyses. We emphasized the concept that there are no underpowered studies, there are underpowered analyses.

Small studies may not be amenable to much or any statistical analysis on their own. Yet, weak or futile analyses are often done. That may be because the authors feel they must or because reviewers expect it. However, underpowered analyses can be useless or even harmful since they can lead to improper conclusions.

So, while we want large studies that can clearly answer a question, we also want small studies to be published. That may just be data, with no analysis, something that can be hard for people to wrap their heads around.

I wasn’t really intending to give that long intro (but regular readers will realize it’s not uncommon!). For more details, here’s the commentary.

The reason I brought this up is a nice small study about bacterial endocarditis in dogs that underwent balloon dilatation because of congenital pulmonary stenosis (Zeedijk and Szatmári et al 2026).

They evaluated dogs that underwent this procedure and that had adequate post-operative followup. They focused on the 83 dogs that didn’t get peri-operative antimicrobials. None developed an infection. There was a small group (11 dogs) that got antibiotics. They didn’t get an infection either.  We could run futile statistical analysis and conclude that there is no statistically significant difference, but we’d have no confidence in that analysis. The study was not adequately powered for that comparison so it’s great they didn’t try to do it. We have those data for a future meta-analysis.

The 83 dogs that didn’t get antibiotics can provide some additional insight.

With 0 infections in 83 dogs, the 95% confidence interval for infection rate would be 0-3.6%. That means we think the true incidence of infection could be between 0 and 36 infections per 1000 dogs that underwent the procedure.

Just using that, we can consider the low rate, the potential severity of disease, the ability to treat disease and potential complications from prophylaxis, and use that to think about the balanced between costs/risks and benefits. That’s admittedly tough since all those are hard to specifically define, but we have to consider the different areas and not just the infection rate.

We can take it a bit further, though. Not all infections are preventable, even when antibiotics are indicated. As a result, we can’t base calculations on the assumption that all infections will disappear. If antibiotics are effective, they will lower, not eliminate the infection rate.

If we use an infection rate of 3.6% (the upper limit of the confidence interval) and estimate that antibiotics would reduce the infection risk by 25% (remembering we don’t know if they reduce it at all), we can calculate the absolute risk reduction of 0.9%….a reduction of 9 infections per 1000 treated dogs. The number needed to treat (NNT) is a concept that’s underused in veterinary medicine and it’s useful to help put rates into context. With this example, the NNT would be 111…. we would need to treat 111 dogs to prevent a single infection.

At first glance, an NNT of 111 might seem quite reasonable, especially since infective endocarditis can be a severe disease.

However, that’s based on a very conservative assumption that the true infection rate is at the upper end of the confidence interval. We also haven’t considered costs, adverse effects or antimicrobial resistance.

While we’re treating those 111 dogs to prevent one infection, some dogs will experience adverse effects from the antibiotic. Most will be minor, but occasionally they can be significant. So we need to weigh the dogs that might benefit against the dogs that will be harmed.

leaves us to piece together data from different studies.

And, remember, we’re basing that on a very (and likely unrealistically) high infection rate. If the infection rate is 1% and antibiotics reduce that risk by 25%, the absolute risk reduction would be 0.25%, or 2.5 infections prevented per 1000 treated dogs. The NNT would be 400.

Drop that infection rate to 0.5%, and it’s 800.

In humans, the endocarditis rate for this procedure has been reported at 0.12%. The NNT for that? 3333. I think it’s pretty safe to say we’d do a lot of damage treating a few thousand dogs. That doesn’t even consider the antimicrobial resistance aspects.

Yet, antibiotics are still very commonly used for low risk procedure like this in dogs, as was shown recently.

Sorry for the math. It’s important to consider and we too often don’t think about absolute risk reduction, NNT, number needed to harm and broader aspects. We tunnel vision on incidence data and P values and act as though they provide the answer. In reality, they rarely tell the whole story and sometimes tell a misleading one