(B44) Assessing Risk of Bias in Comparative Effectiveness Studies of Chimeric Antigen Receptor T-Cell Therapies for Relapsed or Refractory Large B-Cell Lymphoma

Publication Number: 654

Background: Chimeric Antigen Receptor T-Cell (CAR-T) therapies are an innovative treatment for relapsed or refractory large B-cell lymphoma (LBCL). Because of lack of randomized clinical trials to compare different CAR-T therapies, real-world comparative effectiveness (CE) studies have emerged to fill the evidence gap.

Objectives: The objective of this study was to identify a scientifically sound method to compare the quality of CE studies on CAR-T therapies and transparently communicate the comparisons to a broad audience.

Methods: We critically assessed the risk of bias (ROB) of four retrospective cohort studies identified via a systematic review that compared tisagenlecleucel (tisa-cel) and axicabtagene ciloleucel (axi-cel) for treating ≥200 patients with LBCL. We leveraged three tools in the assessment: the GRACE Checklist, ROBINS-I, and Newcastle-Ottawa Scale. The ROB assessment findings and the evaluation of the tools’ fit-for-purpose were qualitatively summarized into tables.

Results: The four assessed studies (Bachy 2022, Bethge 2022, Kwon 2022, Riedell 2022) reported data from large multi-center registries in France, Germany, Spain, and the United States, respectively. Overall, the studies showed better effectiveness outcomes (treatment response, survivals) with axi-cel and less neurotoxicity or cytokine release syndrome with tisa-cel.

Details of the ROB assessment will be presented in the poster. In summary, our ROB assessment identified several common methodological issues across the studies. All but the Bachy (2022) study were inadequate in addressing cohort comparability factors, including confounding from patient and disease characteristics, differences in bridging therapies, apheresis-to-infusion time, lymphodepletion regimens, and impact of preceding or subsequent treatments. All studies focused on the subset of patients who actually received CAR T-cell infusion after apheresis or CAR-T therapy order. This could introduce immortal time bias to the comparative effectiveness for the intention-to-treat cohorts.

In general, the three ROB tools were insufficient for comprehensive assessment or consistent comparisons of CE studies on CAR-T therapies. While all tools allow users to assess ROB caused by confounding, patient selection or outcome measurement to certain degree, the depth of the assessment and the fit-for-purpose of the tools were restricted by the limitations in the ROB domain construct and rating mechanisms.

Conclusions: CE studies on CAR-T therapies are subject to various methodological issues that may affect ROB. Existing ROB tools do not meet our need to comprehensively evaluate or transparently compare the rigor of these studies. Future research is warranted to develop better fit-for-purpose tools.