Founder & Chief Scientific Officer KREDHERA, LLC Hampton, United States
Background: Early phase, single arm clinical trials may explore trends in biomarkers related to disease activity in the target population for which a product is intended. It is often of interest to contrast estimated trends from the clinical trial to those observed in the target population receiving standard of care in a real world (RW) clinical practice setting. Key challenges in forming a benchmark cohort (BC) are the small sample size in a rare condition, the determination of start of follow up in the benchmark cohort, and the overall ability to balance two independent cohorts.
Objectives: To evaluate different approaches for the construction of a historical BC and assess the comparability of a clinical trial cohort (CT) with the comparator BC among patients with a rare condition.
Methods: We used a retrospective cohort study from a RW registry of patients with a biopsy-confirmed diagnosis of a specific rare condition. Baseline characteristics of participants in CT will be used to assess comparability of the CT cohort with the RW comparator BC, after applying trial eligibility criteria. This allows formation of a BC in which a patient in the RW registry could contribute multiple observations in the study cohort reflecting different study cohort entry (index dates) at which the trial inclusion/exclusion criteria were met.
Results: We considered the advantages and limitations of different approaches to control for confounding. This led us to two causal inference approaches to improve the comparability of the CT and BC: the disease risk score (DRS) and the propensity score (PS). DRS can be derived on BC only and applied to CT baseline data. Differences in baseline characteristics will be assessed using standardized mean differences. To assess the ability of the DRS to account for confounding bias a dry run diagnostic will be implemented in the comparator cohort. The overlap of PS for CT and BC should lend confidence to positivity, as will standardized mean differences of baseline characteristics.
Conclusions: Natural history studies or registries are becoming increasingly used for either benchmarking or as external control arms, particularly for rare disease. Applying clinical trial entry criteria to such studies only identifies patients who might have been trial-eligible but finer matching is typically needed. DRS may be a useful tool for confounder control in rare diseases. The analytical approaches will be described in detail.
