Antibiotic Safety in Early Pregnancy: When “Caution” Becomes Confusion
Guidelines should translate data into action, not adjectives. When every recommendation hides behind flexible language, clinical responsibility is omitted.
When a scientific conclusion blurs more than it clarifies, clinicians are left guessing—and patients are left uninformed. The new JAMA Network Open study by Osmundson et al. on antibiotic safety in early pregnancy ends with a puzzling statement: “Our results support the current ACOG recommendation for caution in using TMP-SMX during the first trimester but do not support current recommendations to limit nitrofurantoin use.”
This large U.S. cohort of more than 71,000 first-trimester antibiotic exposures offers some of the clearest data yet on congenital malformation risks. Trimethoprim-sulfamethoxazole (TMP-SMX) exposure was linked to a 35% higher relative risk of any malformation and a two-fold risk of severe cardiac defects and cleft lip or palate compared with β-lactams. Nitrofurantoin and fluoroquinolones showed no statistically significant excess risk.
The design was strong: an active comparator, propensity-weighted modeling, and sensitivity analyses that largely confirmed the signal for TMP-SMX. Still, the authors’ conclusion language deserves scrutiny. What does it mean to “support caution” for one drug but to “not support limits” for another? Both “caution” and “limit” are undefined, non-quantitative, and subject to interpretation. ACOG itself avoids clarity by using both verbs without operational thresholds. In medicine, such rhetorical ambiguity becomes policy camouflage: enough to imply safety without accepting liability.
If TMP-SMX doubles the risk of certain malformations, then “caution” is not an adequate descriptor—it implies a watchful eye, not a warning.
Conversely, stating that the data “do not support limiting nitrofurantoin” risks overinterpretation. A relative risk of 1.12 (95% CI 1.00–1.26) hardly justifies a safety endorsement; it merely fails to reach conventional statistical significance. The absence of proof of harm is not proof of safety. When the upper confidence interval includes a 26% increase, clinical humility—not reassurance—is warranted.
The underlying issue is not just semantics but ethics. When guideline committees adopt deliberately soft verbs, they shift risk communication from the institution to the individual clinician. A provider who “used with caution” can still be accused of negligence if harm occurs, while an organization that “recommended caution” remains shielded. This linguistic fog is the opposite of evidence translation—it is evidence dilution.
ACOG’s own 2017 opinion (“avoid when possible”) was already an exercise in hedging. Osmundson’s data could have clarified practice by proposing concrete thresholds: for example, to avoid TMP-SMX entirely before 10 weeks or to favor β-lactams whenever culture sensitivity allows. Instead, we are left with the same verbs that started the problem—caution and limit—two words that sound protective but are scientifically indistinguishable.
Guidelines should translate data into action, not adjectives. When every recommendation hides behind flexible language, clinical responsibility migrates from the collective to the individual. The ethical question is simple: If “caution” cannot be defined, is it really guidance—or just a disclaimer?