Comparison of methods for identifying the optimal treatment duration in randomized trials for antibiotics

The optimal duration of antibiotic treatment must strike a delicate balance: it must be long enough to achieve desirable efficacy yet short enough to prevent the development of toxicities, adverse events, and mitigate other arduous aspects related to patient burden. Historically, the approach used to determine duration of antibiotic treatment has been inefficient, severely impacting the refinement of therapeutics for tuberculosis (TB) where treatment duration, and its complications, can be extensive. Many of the challenges in duration-ranging have parallels and proposed solutions in the field of dose-ranging where the literature is substantially more established and where the traditions of qualitative, pairwise comparison studies have been replaced with model-based approaches. Such methods are more efficient and allow for interpolation between the doses observed.

This work examines the utility of cutting-edge dose-finding methods (such as MCP-Mod) for duration-ranging of TB treatments. We compare the operating characteristics of the adapted model-based duration-ranging methodologies against standard qualitative methods for the purposes of estimating optimal duration and describing the duration-response relationship, using a simulation study motivated by a Multi-Arm Multi-Stage Response Over Continuous Intervention (MAMS-ROCI) clinical trial design. We explore three specific targets: (1) power to detect a duration-response relationship, (2) ability to accurately reproduce the duration-response curve, and (3) ability to estimate the optimal duration within an acceptable margin of error.

We find that model-based methods outperform standard qualitative comparisons on every target examined, particularly when the sample size is constrained to that of a typical Phase II trial.

We conclude that the success of the next era in TB therapeutics duration evaluation trials, and antibiotics duration-ranging more broadly, will meaningfully rely on the ability to simultaneously pair innovative model-based statistical methods with re-imagined study designs such as MAMS-ROCI.