Observational versus randomized controlled trials to inform antibiotic treatment durations: a narrative review.

Background: Studies comparing shorter and longer antibiotic treatment durations are increasingly common. Randomized controlled trials (RCTs) are an ideal methodological approach to study antibiotic treatment durations; however, these trials can be logistically and financially challenging to conduct.

Objectives: In this narrative review, we sought to compare the strengths and limitations of observational study data with those of RCT data in evaluating antibiotic treatment durations. We used uncomplicated Gram-negative bacteraemia as an illustrative case example because several published RCTs and observational studies have been conducted in similar patient populations.

Sources: We searched MEDLINE for articles comparing treatment durations for gram-negative bacteremia from inception to June 9th, 2022. We included studies reporting on all-cause mortality and/or relapse at day 28-30. Data comparing short- versus long-course therapy were pooled by Bayesian random effects meta-analyses to assess the odds ratios (OR) of all-cause mortality and relapse at 30 days, stratified by study design. Parameters were summarized with median and 95% highest-density credible intervals (CrI). Posterior probabilities of OR > 1.0 were estimated. Observational studies were further examined to determine if and how they addressed potential sources of bias.

Content: We identified 1671 unique records and included 10 studies (seven observational and three RCTs). With respect to 30-day mortality, the Bayesian posterior probability that a longer course of therapy was better (i.e. OR >1.0) was 42% in RCTs (OR, 0.94; 95% CrI, 0.51-1.68) and 91% in observational studies (OR, 1.25; 95% CrI, 0.88-1.73). No observational study fully addressed all potential sources of bias.

Implications: On the basis of our findings, we discuss future directions for antibiotic treatment duration trials, including approaches to limit sources of bias in observation data and novel trial designs.