Our study showed a 12-month overall mean TFR rate of 60% [95CI:0.56–0.63] and a 24-month overall mean TFR rate of 55% [95CI:0.52–0.59], indicating that more than half of the CML-CP patients were able to maintain TFR two years after TKI discontinuation. Our subgroup analysis showed that better TFR was associated with prior interferon therapy (12months: P < 0.001,24months: P = 0.006), and molecular response depth MR5.0 (12months:P = 0.005,24months: P = 0.02), which is in line with the findings of previous trials [23, 33, 35], although this analysis exists heterogeneity. According to the previous study [15], the duration of DMR and TKI therapy before discontinuation were found to be associated with TFR. However, our analysis was not, and the possible explanation is the effect of heterogeneity in the inclusion of trials with other factors, further research may significantly impact the estimated effect sizes and potentially alter them. Although subgroup analyses of several trials have shown that TFR rates are not associated with the duration of TKI treatment and DMR duration before TKI discontinuation [28, 29, 31, 38, 39].
Adverse effects in patients maintained on TFR were mainly characterized by skeletal muscle myalgia, i.e., withdrawal syndrome. Although the EURO-SKI and ENESTop study reported rates of 31% and 74%, the rates of grade 3 to 4 adverse reactions were only 1% and 13%, respectively [33, 34]. The D-STOP trial, on the other hand, reported an incidence of only 5.6% [28]. In addition, progression or CML-related deaths occurred after discontinuation in only 2 cases. The above demonstrates the feasibility and safety of TKI discontinuation.
The following limitations are worth considering in this meta-analysis.
Firstly, there was a high degree of heterogeneity in the clinical trials we included. This arose from differences in trial design, including criteria for the discontinuation population, the definition of molecular relapse, and different molecular monitoring methods such as Droplet digital PCR (dd-PCR) and RQ-PCR. Dd-PCR demonstrated more accurate molecular relapse monitoring than RQ-PCR [42, 43]. Since there is no RCT for clinical trials related to drug discontinuation, it also affects the magnitude of heterogeneity. Despite differences in trial design, the trials we included were of high methodological quality. To explore heterogeneity, our analyses were validated by different subcomponents.
Subsequently, we excluded retrospective clinical trials to avoid high heterogeneity and publication bias. A total of 30 retrospective studies were excluded in the rescreening, and we read these articles, which involved a large number of patients and had mostly high second-year TFR, such as the retrospective cohort study of M. Ansuinelli, with a second-year TFR of 74% [44]. Therefore, it would be of interest to analyze the discontinuation population of retrospective studies separately in the future.
In addition, we have not yet analyzed distant relapse. In clinical trials with sufficiently long follow-up, the TFR rate in the fifth year was 44% for DASFREE, 49% for EURO-SKI in the third year, and 53.57% for ENESTfreedom in the fourth year, et al [24, 45, 46]. We note the presence of distant recurrence, but there is a temporary lack of sufficient trial sample sizes for analysis. Similarly, the limited number of prospective trials included in this meta-analysis resulted in a low number of trials in certain subgroups in the subgroup analysis.
Self-renewing leukemic stem cells are associated with relapse, and some patients are still unable to achieve TFR due to the persistence of leukemic stem cells (LSCs) [47]. Paola Pacelli noted that despite fluctuating residual CD26 + LSCs, maintaining a stable TFR is still possible. [48]. Namely, factors other than the presence of LSCs play a positive role in disease recurrence. M Matsushita's trial found a relapse rate of 63.6% in cxorf48-specific ctl-negative patients and 0% in cxorf48-specific ctl-positive patients [49]. Yazad D Irani's results showed that TIM-3 was consistently up-regulated in CD4 + T-cells, CD8 + T-cells, and T-regs in relapsed patients, compared to patients with maintained TFR. regs cells were consistently upregulated [50]. In addition, Gene expression analyses of publicly available datasets revealed increased expression of TIM-3 on CML stem cells in comparison to normal hematopoietic stem cells. These results from recent studies of TKI discontinuation suggest that CXorf48 and TIM-3 may be promising therapeutic targets for durable treatment-free remission in immunotherapy for CML patients.
Management of patients who discontinued TKIs should also focus on their psychological as well as other domains. Giora Sharf investigated that 56% of patients reported fear or anxiety during treatment discontinuation, and 60% reported withdrawal symptoms at the time of discontinuation [51]. Possible reasons for anxiety in patients who discontinue may be due to the uncertainty of relapse, adverse effects, and regular monitoring after discontinuation. The study [15] has shown that patients who failed TFR mostly resumed DMR after re-treatment with TKI. For molecular monitoring after TKIs discontinuation, the expert consensus of the European Leukemia Net proposes lifelong [52]. Our observation that TFR failure occurs mainly within one year, but a few patients still experience distant relapse, albeit slow onset [24, 45, 46], suggests that there is an urgent need for a cost-effective and user-friendly method for monitoring molecular response after discontinuation, as well as for more accurate predictors. Dennis Dong Hwan Kim's results suggest that BCR-ABL1 transcriptional doubling time can be used as a predictor of TFR failure in CML-CP patients after imatinib discontinuation, potentially avoiding the need for frequent monthly monitoring of molecular assays [53].
The results of our analysis suggest a strategy that seems feasible to improve the success of TFR by attempting to rationalize the application of interferon before or after discontinuation of the drug for those who reach a molecular response depth of MR 5.0 or deeper. A Burchert's trial demonstrated that this strategy resulted in high TFR rates, with up to 84% of patients obtaining MR5 before discontinuation after a median follow-up of 7.9 years [54].
Interestingly, our analysis showed no significant effect of DMR duration and duration of TKI therapy before discontinuation of TKIs on TFR rates, but the D-FREE trial tells us that ignoring DMR duration is not desirable. One possible explanation is that the TFR rate depends mainly on the patient's sensitivity to the TKI rather than the duration of treatment, i.e., the depth of the molecular response obtained after treatment with a TKI, and the same factor is also responsible for the time required to reach that depth after TKIs treatment, which has not been recorded in the majority of the trials, and which would be a predictor worth investigating. This factor was reported in only two trials, the TWISTER trial showed an association with better TFR (P = 0.04), but the KID trial did not correlate with TFR rates (P = 0.152) [36, 41]. In Naranie Shanmuganathan's study, time to BCR-ABL1 halving on TKI treatment was the strongest independent predictor of TFR maintenance [55].
Another strategy that seems feasible is dose-escalation discontinuation. A Iurlo's retrospective analysis in Italy found that low-dose TKIs do not seem to affect the likelihood of achieving DMR, but may even improve TFR rates, which were 74% in the second year [56]. The DESTINY trial demonstrated that discontinuation of TKIs after one year of dose halving resulted in a 72% progression-free survival rate in the second year [57], but more trials are needed to validate whether attempted discontinuation after dose reduction promotes TFR.
The number of RCTs and prospective clinical trials related to this area is limited, and more high-quality clinical trials are needed to further explore this. Meanwhile, trials of other second and third-generation TKIs are anticipated. The latest asciminib trial demonstrated superior efficacy and fewer adverse events [58]. We searched ClinicalTrials.gov and found that there are still multiple large trials with no results output, and we will update our opinion if we find an impact on predictors in the future.