No Differences in Molecular Relapse-Free Survival after Stopping Imatinib Treatment of Chronic Myeloid Leukemia Between Patients with Prior 4.5 Log Reduction (MR4.5) but Detectable and Patients with Undetectable Disease in the EURO-SKI Trial

Introduction: With imatinib (IM), 60% of patients with chronic myeloid leukemia (CML) achieved deep molecular responses (MR⁴: at least a 4-log reduction in BCR-ABL1 transcript level according to the International Scale) within 4 years (Hehlmann 2014). In the EURO-SKI trial, 62% of the patients were in molecular relapse-free survival (RFS) i.e. alive and in major molecular remission (3-log reduction in BCR-ABL1) 6 months after stopping tyrosine kinase inhibitor treatment. Duration of MR⁴ was a prognostic factor associated with RFS at this time (Richter EHA 2016).

Aims: Apart from MR⁴ duration, depth of molecular response is suspected to influence RFS. To investigate this, 3 different status of deep molecular response at time of treatment stop were compared: Status A, "MR⁴ only": Patients with MR⁴ but not MR^4.5 (MR^4.5: at least a 4.5-log reduction in BCR-ABL1); status B, "MR^4.5 detectable disease": Patients with MR^4.5 but still detectable BCR-ABL1 transcripts; status C, "MR^4.5 undetectable disease": Patients with MR^4.5 but undetectable BCR-ABL1 transcripts.

Methods: For MR^4.5 assessment, the number of control gene transcripts had to be at least 32 000, if the control gene was ABL1 and 77 000 in case of GUS. To an unknown extent, detectability of BCR-ABL1 in patients with MR^4.5 depended on the number of control gene transcripts. To reduce bias when comparing status B and C, propensity score (PS) matching (Rosenbaum, Rubin 1983) was applied in order to receive two samples with a similar sensitivity of identifying BCR-ABL1. Additional to type (ABL1 or GUS) and number of control gene transcripts, matching variables were interferon alpha pre-treatment, duration of MR⁴_,and the IM treatment time before observation of MR⁴. Logistic regression was used to compare RFS at 6 months between response status. Significance level was 0.05.

Results: In the EURO-SKI trial, a total of 357 patients fulfilled all in- and exclusion criteria, had eligible and complete data regarding variables part of any prognostic score, and sufficient molecular data prior to and within the first 6 months after stopping IM treatment. Thirty-three patients (9%) had status A "MR⁴ only", 125 (35%) had detectable disease (status B) and 199 patients undetectable BCR-ABL1 (56%, status C). Between the groups, there was hardly a difference in the type of control gene: ABL1 in 79%, 85%, and 83% of patients with status A, B, and C. Prior to PS matching, median numbers of evaluated ABL1 transcripts were 81 490 (A), 96 040 (B), and 77 250 (C), respectively. Numbers were significantly higher with status B as compared with C (U test: P=0.0283). In patients with MR⁴ only, RFS at 6 months was 61% (95% confidence interval (CI): 42-77%), in patients with MR^4.5 (status B) 51% (CI: 42-60%), and with undetectable disease 62% (CI: 56-69%). For 119 patients with status B, 119 suitable matching partners with undetectable disease were identified. Now, after PS matching, median numbers of evaluated ABL1 transcripts were 103 759 (B) and 89 250 (C) (not significant). In patients with status B, RFS at 6 months was 52% (CI: 43-61%), and with undetectable MR^4.5 57% (CI: 48-66%). In a logistic regression model stratified for the matched pairs and adjusted for interferon alpha pre-treatment, duration of MR⁴ and IM treatment time before its observation, regarding RFS at 6 months, the odds ratio for MR^4.5 with undetectable to detectable disease was 1.301(CI: 0.750-2.258).

Conclusions: Detectability of BCR-ABL1 transcripts in patients with MR^4.5 depends on the number of control gene transcripts. The higher the number, the higher is the sensitivity i.e. the chance to identify BCR-ABL1. This hampers the assessment of whether a patient really is without BCR-ABL1 or whether the observation of undetectable disease was due to insufficient sensitivity. The sensitivity problem introduces a bias when, at MR^4.5 level, the analysis of outcome in dependence on detectable vs. undetectable BCR-ABL1 is in focus. Also in the EURO-SKI trial, numbers of ABL1 transcripts were significantly higher in patients with MR^4.5 but detectable BCR-ABL1. After matching, differences and potential bias were reduced. Between patients with and without BCR-ABL1, the difference in RFS at 6 months was not significant - neither before nor after PS matching. Refined statistical analyses combined with adherence to laboratory recommendations support unbiased analyses when differentiating detectable and undetectable disease.