Thoughtful consideration should be given to sampling R/R DLBCL patients that should reflect the purpose of the study.
Background: Real world data provide rich longitudinal information on the entire cycle of care for patients, however sampling R/R DLBCL patient is challenging due to the presence of multiple eligible index dates per patient. In the epidemiological literature, potential sampling methods include to select: 1) the first index, 2) a random index, or 3) all index dates and statistical ly adjust for within-patient correlation in outcome models.
Objectives: To apply three sampling methods to a retrospective real world cohort of R/R DLBCL patients and to evaluate differences of five patient characteristics between each sample by comparing to a reference sample of R/R DLBLC patients enrolled in the control arm of a historic clinical trial (NCT02257567).
Methods: Clinical notes from The Christie NHS Foundation trust in the UK were retrospectively abstracted using a standardized data col- lection form for 90 patients with R/R DLBCL, aged 18 or older, histo- logically confirmed DLBCL, at least one prior anti-lymphoma treatment, who collectively received 192 lines of therapy between 2011-2018. The index date was the start date of each line of therapy. Characteristics included gender, age at index, no. of prior treatment lines, refractory disease, and history of stem-cell transplant; the latter four an vary within patient per eligible index date. The trial enrolled R/R DLBCL patients, of which 39 participants were randomised to the control arm. The trial eligibility criteria were applied to the Christie data resulting in a cohort of 73 patients with 122 eligible index dates. The three sampling approaches were applied to the eligible cohort, and the standardized mean difference (SMD) of the patient characteristics in comparison to the reference sample calculated.
Results: In each sample the SMDs were similar for gender (≈0.05), index age (≈0.16), and prior transplants (≈0.22); while the all-index sample yielded the lowest SMDs for refractory disease (0.61) and no. of prior lines (0.60) in comparison to the reference sample. Proba- bility sampling of a single index to match the distribution of the no. of prior lines from the trial was additionally performed, further reducing SMDs for refractory disease (0.10) and no. of prior lines (0.46) with similar SMDs to the other samples for the remaining variables. Conclusions: Each method yielded samples with differences in patient characteristics, reflected in varying SMDs in comparison to a refer- ence sample. Thoughtful consideration should be given to sampling R/R DLBCL patients that should reflect the purpose of the study