Index date selection for external comparators in assessment of cancer therapies in relapsed/refractory follicular lymphoma

Randomized controlled trials (RCTs) are the gold standard for determining the efficacy of new treatments. However, RCTs may not always be feasible, such as in rare disease settings (e.g., when there are few people to potentially randomize) or when there are significant challenges to recruitment and randomization due to a lack of an effective standard of care. Single arm trials are an alternative design in these contexts, wherein there is only one experimental treatment arm, but no randomization to a parallel control arm. This lack of a randomized control arm substantially limits the evidence that can be generated about the true effects of the experimental treatment. In turn, an approach that is gaining increasing interest in the analysis of single arm trials is the use of an external control (or comparator) arm, which (in certain contexts and under specific assumptions) may be able to stand-in for a randomized control arm and used to make direct comparisons with the new experimental treatment, increasing confidence in the estimated treatment effect. External comparator arm data most often come from prior control arms of RCTs (historical controls) or from real-world data sources like well-curated observational cohorts. When designing and analyzing single arm trials that use external comparator arms, researchers face several potential threats to the valid estimation of treatment effects. The most cited sources of bias in the relevant literature include confounding (due to the lack of randomization) and outcome misclassification bias (as outcomes are assessed differently in trials versus real-world data settings).

One lesser studied source of potential bias relevant for some single arm trials with external comparator arms is related to defining the start follow-up for outcome assessment (e.g., often termed the index date or time zero). While the index date for the experimental treatment arm is clear (i.e., the start date of the experimental treatment), the index date for the external comparator arm can be less clear. For example, a recent single arm trial of mosunetuzumab, a new treatment for relapsed or refractory follicular lymphoma, required that individuals had to have failed at least two prior lines of treatment in order to be eligible for the trial. Thus, there were patients in this trial who had previously failed only two lines of therapy, others that failed three lines of therapy, and even others that failed four lines of therapy. When designing an external comparator arm for this study using real-world data, a given individual could be eligible for inclusion several times and thus has several potential index dates to choose from. For example, say that a person in the external comparator arm failed three previous lines of therapy and started their fourth line of therapy. This person would be eligible for inclusion in the external comparator arm two times – first, when they failed their second line of therapy and started their third line and again when they failed their third line of therapy and started their fourth line of therapy. In these scenarios, it is currently unclear how researchers should select the appropriate index date when constructing an external comparator arm.

This definition of an index date is important as it is central to several major study design elements that influence the estimated risk or time-to-event in the external comparator arm. First, the index date determines who is included in the external comparator arm (e.g., the number of people and the characteristics of the people included can change based on different index date definitions). Second, the index date also determines the values of patient time-varying characteristics (e.g., if a researcher selects the index date as the start of the fourth line of therapy, patients may be older than if a researcher had selected the index date as the start of the third line of therapy). Third, the index date determines which outcomes will count (e.g., if a progression event happens between the third and fourth lines of therapy, then choosing the start of the third line of therapy as the index date will ‘count’ this event, whereas choosing the start of the fourth line of therapy as the index date will exclude this event). Depending on the choice of index date, the outcomes observed among patients receiving the experimental treatment may appear better or worse than those of the external comparator arm, separate from the actual treatment benefit (or lack thereof).

The goal of this project is to generate knowledge about the potential impact of different index date definitions in the design and analysis of single arm trials using external comparator arms. We plan to use the mosunetuzumab single arm trial (GO29781), as a motivating example that allows us to conceptualize a variety of index date definitions. We will implement eight different index date definitions (detailed in the statistical analysis plan) using a real-world data source (the Lymphoma Epidemiology of Outcomes Consortium for Real World Evidence, LEO CReWE) including overall survival and progression-free survival as outcomes; these data will not be brought into the Vivli platform. We will instead export aggregate data on the baseline characteristics of the GO29781 trial to serve as a reference population for our external comparator cohorts derived from the LEO CReWE data.

Results from our analyses will generate new knowledge about how the choice of an index date impacts the observed risk of specific outcomes in an external comparator arm. These findings will help to inform future design choices for single arm trials that incorporate external comparator arms, improving the rigor of these studies, and potentially increasing access to novel experimental treatments to patient populations in high need (e.g., patients with rare diseases, conditions where there is no clear standard of care).