Comparing Survival for Different CAR Ts: Need for Addressing Bias Due to Differences in the Pre-Infusion Period

Introduction

Patients with relapsed/refractory large B-cell lymphoma (RR-LBCL) treated with salvage chemotherapy have a poor prognosis, with a median survival of 6.3 months (Crump 2017). Given the recent approval of chimeric antigen receptor T-cell therapies (CAR T) for patients with RR-LBCL, efforts to demonstrate the comparative effectiveness of different CAR Ts are imminent. In the absence of head-to-head trials, such an assessment will likely be based on indirect or between-trial comparisons of reported survival estimates. Due to differences in the multi-step manufacturing process preceding the infusion of CAR T cells, the duration of the pre-infusion period, which accounts for time from leukapheresis to infusion, may vary across CAR Ts. This variability may be associated with differences in the prognostic characteristics of patients who endure the pre-infusion period. As such, comparisons of reported survival among these selected patients with the different CAR Ts evaluated in different trials may be biased. Estimating the comparative effects of CAR T therapies on survival requires careful attention to these potential differences and the use of thoughtful analytic approaches to address bias. We aimed to review current methodological approaches used to address bias related to time-to-treatment initiation and assess their potential utility in comparative analyses of CAR T based on between-trial comparisons.

Methods

A targeted literature review was performed to identify studies that used methods to adjust for bias related to time-to-treatment initiation when estimating relative treatment effects regarding survival outcomes between interventions in oncology. Relevant studies published in English from 2008 onwards were identified by searching the Medical Literature Analysis and Retrieval System Online (MEDLINE) and Excerpta Medica dataBASE (EMBASE). In addition, manual searches of the reference lists of the identified studies were performed. Studies that provided the most recent application of a proposed method were selected for inclusion.

Results

We identified several manuscripts that utilized methods to address survival or survival treatment selection bias (STSB) in observational studies where the classification of "treated" individuals required them to have survived until treatment initiation, and individuals who died early without the opportunity to get treatment were classified as "untreated". The identified methods to adjust for the over-estimation of the treatment effect include landmark survival and time-dependent exposure assignment. The landmark method restricts the analysis sample to patients who are event-free and alive at a given point (i.e, landmark) during follow-up; patients are assigned to risk groups based on their status at the landmark and survival outcomes are estimated from that point. Time-dependent exposure assignment addresses STSB by accounting for changes in patient exposure or treatment status during the course of follow-up. All patients begin in the same risk group; when exposure changes, patients and person-time are switched to the applicable risk group at that point. However, these methods do not address bias resulting from a comparison of reported survival estimates between different CAR Ts for a follow-up starting at the successful infusion of the CAR T cells. When comparing CAR Ts, the differences in prognostic patient characteristics due to differences in the pre-infusion phase are similar to channeling bias or confounding-by-indication. These concepts are well known, and established methods are available. However, those methods have limited applicability without access to individual patient data for the CAR Ts of interest.

Conclusion

Comparing survival upon successful infusion with different CAR Ts reported in separate studies is likely to be biased due to differences in the duration of the pre-infusion period. Methods previously used to address bias related to differences in time-to-treatment initiation may not be relevant or applicable, particularly in the absence of patient-level data. A between-trial comparison of reported survival estimates with the different CAR Ts might benefit from incorporating external evidence to inform the mortality during the pre-infusion periods.