Clinical research to improve lymphoma treatment in France relies on randomized clinical trials (CTs). However, in certain situations, it is not feasible to conduct such trials. Patients aged 80 years or older (≥80 y.o) with newly diagnosed diffuse large B-cell lymphomas (DLBCL) are underrepresented in CT. Recruiting patients ≥80 y.o in CT can be more challenging than the younger, explaining partly the lack of focused CT. In this study, we aimed to build an External Control Arm (ECA) from mixed real-world and CT data for ≥80 y.o. DLBCL population (Mixed ECA), and to validate its relevance using the SENIOR trial, the only published randomized CT in this population.

We built two mixed ECA with data from both CT (LNH09-7B) and real-world (REALYSA) selecting patients who received a treatment similar to the control treatment in the SENIOR trial. Patients were also selected to fit SENIOR main inclusion criteria (≥80 y.o. DLBCL, performance status of 3 or 4 and Ann Arbor stage). First, as a preliminary step, we created an ECA (ECA 1) to mimic the control group of the SENIOR trial. Then, we developed another ECA (ECA 2) to mimic the experimental treatment arm of the SENIOR trial. The endpoint for comparison was Overall Survival (OS). All patients were censored at 24 months to ensure comparability between all cohorts. Propensity scores (PS) were estimated for each patient using logistic regression and multiple imputation methods were applied when needed. Patients’ covariates between arms were balanced using the stabilized inverse probability of treatment weighting (sIPTW) statistical approach, allowing to estimate the average treatment effect (ATE) using the marginal Hazard Ratio (HR) as the measure of association.

In total, 73 patients from LNH09-7B trial and 97 patients from REALYSA cohort who met the inclusion criteria were included, as well as 98 and 104 patients from SENIOR control or experimental arm, respectively. For both ECA, standardized Mean Differences for all covariates were <0.1 after weighting regardless of the method used for missing data management, reflecting a good balance between arms. As expected, OS was not significantly different between SENIOR control arm and Mixed ECA 1, with an HR [95%CI] of 0.79 [0.52-1.20]. OS of Mixed ECA 2 vs. SENIOR experimental arm were not statistically different, with a HR [95%CI] of 0.743 [0.494-1.118]. Similar results were obtained with an ECA built with real-world data from REALYSA cohort only.

Our study demonstrates that similar results can be found using an ECA versus an internal control arm, in a difficult to study population. ECA could be used as a bona fide strategy to improve very elderly DLBCL patient's outcomes by allowing more rapid trials with “in silico” control arms and increasing the probability of receiving experimental treatment when included in a trial. Although potential biases should not be overlooked (selection bias, immortality bias, etc.), indirect comparisons represent an interesting complement to randomized CTs for quantifying the effectiveness of therapeutic strategies.