Handling of Allograft As an Intercurrent Event in Randomized Controlled Trials in Acute Leukemias: A Systematic Review

Introduction. Allogeneic hematopoietic stem cell transplantation (allo-HSCT) remains the standard treatment for patients with high-risk acute leukemia (AL), usually associated with a reduced occurrence of relapse at the cost of an increased non-relapse mortality (Burnett et al. 2002). In randomized controlled trials (RCT) evaluating new drugs, allo-HSCT represents an intercurrent event in the estimation of treatment effect. Different statistical approaches have been recommended to handle intercurrent events in the statistical analysis, as stated in the ICH E9(R1) harmonized guideline for clinical trials of the FDA. They differ in terms of treatment effect of interest reflecting distinct clinical questions. Allo-HSCT can be ignored using a treatment policy strategy, where the target is the treatment effect regardless of the occurrence (and impact) of allograft. On the opposite, one may be interested in estimating the treatment effect in a hypothetical population where allo-HSCT does not occur, or in estimating the effect of allo-HSCT in the subsample of allografted patients compared to non-allografted. However, this choice of target obviously impacts the estimand and the statistical analyses to be performed. We wondered whether and how allo-HSCT was handled in the report of RCT in AL.

Methods. We performed a systematic PubMed search of all articles reporting a RCT in patients with acute myeloid leukemia (AML) or acute lymphoblastic leukemia (ALL), published between January 1st, 2014 and April 1st, 2024. We excluded trials evaluating allograft, without two parallel arms, without time-to-event outcomes, or without any patient receiving allo-HSCT after inclusion. Two independent reviewers (RC, LV) evaluated all articles for inclusion and extracted data of selected articles. Method was defined as reported when it was stated in the article text how allo-HSCT was specifically handled in the analysis and not reported otherwise. To evaluate factors associated with reporting, a multivariable logistic regression with random effects was performed.

Results. Among 739 articles retrieved by the PubMed search, 93 RCT were included in the analysis.

There were 71 (76%) trials that included patients with AML and 22 (24%) with ALL; 20 (22%) children and 73 (78%) adults. Median sample size was 330 (IQR: 186-588), with a median proportion of patients receiving allo-HSCT after inclusion of 25% (IQR: 9%-40%).

The primary endpoint was a time-to-event outcome in 71 articles (including two co-primary endpoints in 4 articles). In the main analysis of the 75 primary time-to-event endpoints, method for handling allo-HSCT was not reported in 60 (80%). When the method was reported (n=15), allo-HSCT status was stated to be ignored in 8 and censored in 7.

One sensitivity analysis was reported for 30 (40%) of the 75 primary endpoints, and two sensitivity analyses for 4 (5%). In the whole set of the 38 sensitivity analyses, method of handling allo-HSCT was reported in 36 (95%) with censoring (n=24), a time-dependent variable (n=7), a competing event (n=1), and other methods (n=4).

There was a total of 203 secondary time-to-event endpoints in the 93 articles. Among these secondary endpoints analyses, methods for handling allo-HSCT in the main analysis was not reported in 183 (90%). When the method was reported (n=20), allo-HSCT was censored in 13, ignored in 4, and analyzed as a time-dependent variable in 3. At least one sensitivity analysis was performed for 40 (20%) of 203 secondary endpoints, including 37 with specific report of the method used: mostly censoring (n=20), but also time-dependent variable (n=14) and others (n=3).

In the multivariable logistic regression, the statistical method was more likely to be reported for OS compared to composite endpoints (OR=12.5 [95%CI:2.1-16.7], p=0.0008), for primary compared to secondary endpoints (OR=6.8 [95%CI:2.3-19.7], p=0.0004), and for trials conducted in adults compared to pediatrics (OR=12.5 [95%CI:1.7-100], p=0.01). Type of AL, proportion of allo-HSCT, the promoter (academic or industrial), the number of endpoints, and the phase of the study (II vs III) were not associated with the reporting.

Conclusion. In RCT involving patients treated for AL, the vast majority of trials did not report any statistical method used for handling intercurrent allo-HSCT. Improvement is needed with a systematic reporting of treatment effect of interest and allo-HSCT handling method.

Boissel:Sanofi: Other; Pfizer: Other; Amgen: Other; Adveysa: Other. Dombret:Incyte: Other: Personal Fees, Research Funding; Jazz Pharmaceuticals: Other: Personal Fees, Research Funding; Pfizer: Research Funding; Servier: Research Funding; BMS-Celgene: Research Funding; Amgen: Research Funding; Astellas: Research Funding; Daiichi Sankyo: Other: Personal Fees; Servier: Other: Personal Fees.