Wednesday, February 12, 2025 →

I’m delighted to share a new article, Impact of empiric antibiotics on risk of Clostridioides difficile, now published in Antimicrobial Stewardship & Healthcare Epidemiology (ASHE): doi.org/10.1017/a…

Antimicrobials are clearly associated with C. difficile, but that association is highly confounded. (To give one example, we test for C. diff based on loose stools, but loose stools are also a benign side effect of antibiotics.)

And some patients clearly need broad-spectrum antibiotics! It would be nice if we knew how much harm those broad-spectrum antibiotics caused, not what harm they were associated with.

My coauthors and I tried to estimate that by doing a few things. First, we used a cohort of patients admitted with COVID-19 in the first wave of the pandemic (before widespread vaccine availability). All these patients were admitted with a viral infection, so there shouldn’t really be relationships between one antibiotic or another and later outcomes.

Second, we used a causal inference observational design, emulating a hypothetical trial that would have taken patients where it wasn’t clear whether to give an antibiotic or not and randomized them to one of those two conditions.

Patients in shock are very likely to get antibiotics, and completely stable patients are not very likely to get antibiotics. (That’s appropriate!) Our approach should approximate an approach where, when it isn’t so clear, we flip a coin.

To my surprise, this analysis suggests that IV vancomycin is maybe the antibiotic that most increases risk of hospital-onset CDI. I’m not entirely sure why that should be, and hope smarter microbiologists can pick up that thread from here.

But I think we were able to estimate one of the downstream harms of broad-spectrum antibiotics. And I hope this approach–trying to rigorously estimate the harms of therapies for which we don’t have RCTs, rather than taking the associations at face value–can help us be smarter about lots of treatments we use in the hospital.

My thanks to my coauthors, a handful of folks who critiqued the paper and earlier versions of the analysis, and the late David Meyers, who got me started on this work.