Selection of Comparator Group in Observational Drug Safety Studies: Alternatives to the Active Comparator New User Design

A valid study design is essential when assessing the safety of drugs based on observational data. The comparator group is a key element of the design and can greatly influence the results. The active comparator new user design is a go-to design in observational drug safety research where a target trial of initiation of a study drug versus usual care is emulated. A comparison with another treatment that targets similar patients as the study drug and has no effect on the outcome has great potential to reduce bias. However, the active comparator new user design can be difficult to implement because no suitable comparator drug is available or because it requires extensive exclusion of study drug initiators. In this analysis, we evaluated alternative study designs that can be used in drug safety assessments when the active comparator new user design is not optimal. Using target trial emulation as a common framework, we defined and evaluated the following designs: traditional no use, no-use episodes, active comparator new user, prevalent new user, generalized prevalent new user, and hierarchical prevalent new user. We showed that all designs can be implemented by using sequential cohorts and simply altering the patient selection criteria, i.e., identifying increasingly restrictive cohorts. In this way, all designs are nested in each other and the differences between them can be demonstrated clearly. We concluded that many study-specific factors need to be considered when choosing a design, including indication, available comparator drugs, treatment patterns, potential effect modification, and sample size.

Automated summary

A valid study design is crucial for assessing drug safety from observational data. The active comparator new user design is commonly used but can be challenging to implement. This analysis evaluated alternative designs for drug safety assessments when the active comparator new user design is not optimal. By using sequential cohorts and altering patient selection criteria, all designs can be nested in each other and differences between them can be clearly demonstrated. Many study-specific factors need to be considered when choosing a design.