Early predictors of MR4.5 attainment and eligibility for TKI discontinuation in CML treated with second-generation TKIs

Key Points

• •Six-month MR with 2G-TKI therapy in patients with CML enables early stratification for MR4.5 attainment and TKI discontinuation eligibility.
• •BCR::ABL1 IS ≤0.01% at 6 months during 2G-TKI therapy predicts a high likelihood of achieving TKI discontinuation eligibility by 36 months.

Visual Abstract

Abstract

Among patients with chronic myeloid leukemia (CML) aiming for tyrosine kinase inhibitor (TKI) discontinuation, second-generation TKIs (2G-TKIs) are used as one of the first-line therapy options because they induce faster and deeper molecular responses (MR) than imatinib. However, predictors of attaining and sustaining MR4.5 (BCR::ABL1 International Scale [IS] ≤0.0032%) during 2G-TKI therapy remain undefined. We analyzed 431 patients enrolled in the phase 3 Japan Adult Leukemia Study Group CML212 trial comparing nilotinib at 300 mg twice daily (n = 218) and dasatinib at 100 mg once daily (n = 213) as first-line therapy. The objective was to identify predictors of MR4.5 attainment and TKI discontinuation eligibility (TDE). Candidate variables assessed within 6 months included the European Treatment and Outcome Study (EUTOS) Long-Term Survival (ELTS) score; BCR::ABL1 IS at baseline, 3 months, and 6 months; and IS-derived halving times (HT), HT (0-3) and HT (3-6). TDE was defined as ≥3 years of TKI therapy with ≥2 consecutive years of sustained MR4.5. With a median follow-up of 3.0 years, the cumulative incidence of MR4.5 was 46.6%, and 36.3% of patients achieving MR4.5 met TDE criteria. In multivariable models, HT (0-3) (subdistribution hazard ratio [HR], 2.23; 95% confidence interval [CI], 1.09-4.55) and the 6-month IS value (HR, 0.38; 95% CI, 0.31-0.47) were independent predictors of MR4.5 attainment, whereas only the 6-month IS value predicted TDE (odds ratio, 0.37; 95% CI, 0.24-0.56). This study demonstrates that MR at 6 months after TKI initiation has important clinical value in early stratification of MR4.5 attainment and TDE in patients receiving 2G-TKI therapy.

Introduction

The development of BCR::ABL1 tyrosine kinase inhibitors (TKIs) has significantly improved clinical outcomes in patients with chronic-phase chronic myeloid leukemia (CP-CML),1, 2, 3, 4, 5 resulting in a life expectancy for these patients that is now comparable to that of the general population.^6,7 Furthermore, treatment-free remission (TFR) has emerged as a realistic therapeutic goal in clinical practice, as recognized by several guidelines.8, 9, 10

Early molecular monitoring has become increasingly crucial for CML management in the TFR era. According to both the National Comprehensive Cancer Network⁸ and the European LeukemiaNet 2025 guidelines,¹⁰ BCR::ABL1 transcript levels on the International Scale (IS) at 3, 6, and 12 months are defined as critical milestones for evaluating TKI efficacy and guiding therapeutic decisions, including continuation, dose modification, or switching of treatment. Among these, the 6-month assessment represents a particularly important clinical time point that is widely recognized as one of the key decision points for assessing treatment efficacy in routine practice. Therefore, identifying patients who are likely to achieve TKI discontinuation eligibility (TDE) based on molecular and clinical parameters available within the first 6 months of therapy is of considerable clinical relevance.

In this context, several studies have attempted to identify early predictors of TDE during the course of TKI therapy. Sasaki et al¹¹ developed a regression-based predictive model showing that early molecular response (MR) thresholds were associated with subsequent sustained MR4.5, suggesting that deeper responses than conventional optimal criteria¹⁰ may be required. Shanmuganathan et al reported that early decline kinetics of BCR::ABL1 transcripts, specifically the halving time (HT) after TKI initiation as defined by Branford et al,¹² was a stronger predictor of subsequent TDE than the transcript level at 3 months.¹³ In contrast, Saugues et al demonstrated that the BCR::ABL1 transcript level at 6 months more accurately predicted achievement of deep MR (DMR) than either HT or the 3-month transcript level, although it was not predictive of a sustained DMR sufficient to meet TDE criteria.¹⁴ Thus, previously reported predictors of MR4.5 attainment or TDE have varied across studies, and consistent early predictors have not been clearly defined.

Second-generation TKIs (2G-TKIs), which achieve DMR more rapidly and more frequently than imatinib,2, 3, 4 are selected as one of the first-line options in newly diagnosed CP-CML when achieving TFR is a major therapeutic goal.8, 9, 10 Nevertheless, prolonged use of 2G-TKIs has been associated with a higher incidence of adverse events, including cardiovascular complications, pleural effusions, and pulmonary hypertension, compared with imatinib.^2,3 Therefore, identifying patients who are likely to achieve TDE early and who can safely discontinue TKI therapy within the shortest appropriate timeframe is of particular clinical importance in patients treated with 2G-TKIs. However, previous studies investigating early predictors of TDE have included only a limited number of patients treated with 2G-TKIs, thereby limiting the generalizability of their findings to the broader population of patients receiving 2G-TKIs in current clinical practice.^11,13,14

To address this gap, we conducted a comprehensive exploratory subanalysis, the Japan Adult Leukemia Study Group (JALSG) CML212 Road to MR4.5 study, using data from the prospective phase 3 JALSG CML212 trial, which directly compared nilotinib and dasatinib as first-line therapy for the treatment of CP-CML.¹⁵ Leveraging a large cohort of ∼400 patients treated with 2G-TKIs, we aimed to identify independent predictors of achieving MR4.5 and TDE based on baseline characteristics and early treatment response indicators, including the EUTOS Long-Term Survival (ELTS) score¹⁶ at diagnosis, as well as MR and HT metrics during the first 6 months of therapy. By focusing on parameters measurable within the first 6 months of treatment, our goal was to establish a clinically practical model for early stratification of patients most likely to achieve MR4.5 and TDE.

Patients and methods

The JALSG CML212 study

The JALSG CML212 study was a multicenter, open-label, prospective, randomized phase 3 trial involving patients with newly diagnosed CP-CML (clinical trial registry number: UMIN000007909). The detailed treatment protocol and primary results have been reported previously.¹⁵ Briefly, a total of 461 patients were enrolled from 82 institutes from May 2012 to January 2016. Of these 461 patients, 454 were randomly assigned in a 1:1 ratio to receive either nilotinib (300 mg twice daily) or dasatinib (100 mg once daily), with 227 patients allocated to each arm. Among the intention-to-treat population of 454 patients, 2 were excluded because of withdrawal of consent before drug administration, resulting in a safety population of 452. An additional 11 patients were excluded because of the presence of variant or minor BCR::ABL1 transcripts (n = 8), concurrent lung cancer (n = 1), recent history of malignancy (n = 1), or misdiagnosis (not CML; n = 1). Consequently, the per-protocol population comprised 441 patients (nilotinib arm, n = 224; dasatinib arm, n = 217; Figure 1).

Figure 1.

Flow diagram of study progress. Flow of patient selection in the JALSG CML212 study. The per-protocol population included patients who received at least 1 dose of study treatment and excluded those determined to be ineligible. bid, twice daily; qd, once daily.

Peripheral-blood BCR::ABL1 transcript levels as reported on the IS (BCR::ABL1 IS) were assessed every 3 months through month 18 and subsequently every 3 to 6 months until month 36. Quantification was performed using a highly sensitive reverse-transcription quantitative polymerase chain reaction assay capable of detecting MR4.5 or deeper. The ratio of BCR::ABL1 to the control gene ABL1 was converted to BCR::ABL1 IS using a laboratory-specific conversion factor. MRs were defined according to standard criteria based on BCR::ABL1 IS values as major MR (MMR; BCR::ABL1 IS ≤0.1%) and MR4.5 (BCR::ABL1 IS ≤0.0032%). The primary end point of the JALSG CML212 study was the cumulative rate of MR4.5 achievement within 18 months.

Patient selection for the JALSG CML212 Road to MR4.5 study

In this study, we retrospectively analyzed clinical and molecular data from patients enrolled in the CML212 trial to identify independent predictors of MR4.5 achievement and subsequent TDE focusing on molecular and clinical parameters obtained within the first 6 months after TKI initiation, as defined in the "Definitions and outcome evaluation" section. Among the 441 patients in the per-protocol population, 10 were excluded because of missing BCR::ABL1 IS data, resulting in a final study cohort of 431 patients (Figure 1).

Definitions and outcome evaluation

Patients who experienced loss of MR4.5 at any point during the 2-year period after its initial achievement were considered as not having sustained MR4.5.

TDE in this study was defined as receipt of at least 3 years of TKI therapy with sustained MR4.5 for at least 2 consecutive years, reflecting the maximum evaluable follow-up duration of the CML212 trial.¹⁵ For cases in which BCR::ABL1 IS was assessed shortly before the mark at 36 months (eg, at 35 months), patients were considered to have completed at least 3 years of treatment for the purpose of determining TDE. The HT from time point t₁ to t₂ was calculated using the following formula based on BCR::ABL1 IS values: HT = (ln2 × [t₂ − t₁])/(ln[IS₁] − ln[IS₂]). BCR::ABL1 IS values measured at diagnosis (baseline), 3 months, and 6 months after TKI initiation were used to calculate HT for 2 distinct intervals, including baseline to 3 months (HT [0-3]) and 3 to 6 months (HT [3-6]). These calculated values were used in subsequent analyses.

Statistical analysis

All continuous variables, except for sex and treatment arm, were log-transformed before statistical modeling. Because patients who died or discontinued therapy were no longer at risk of achieving MR4.5, these events were treated as competing risks. Cumulative incidences of MMR and MR4.5 were estimated using the Aalen-Johansen estimator, accounting for these competing events.

Associations between clinical/molecular variables and MR4.5 achievement were evaluated using the Fine-Gray subdistribution hazard model for competing risks. This approach was chosen to appropriately account for competing risks such as death or treatment discontinuation before MR4.5 achievement. For multivariable modeling, variables with a P value <.05 in univariable analysis were selected for further evaluation, with additional filtering to avoid multicollinearity. Model performance was evaluated by the time-dependent concordance index (C-index), which is defined as the probability that, at a given time point, patients who achieve the event earlier are assigned higher predicted risks, while accounting for censoring.

Predictors of TDE were assessed using a logistic regression model. Logistic regression was chosen over the Fine-Gray model because of the limited temporal resolution of BCR::ABL1 IS assessments around the 36-month mark, which was the key time point for determining TDE. Univariate and multivariate analyses were conducted as per the Fine-Gray approach. To enhance clinical applicability and interpretability, we also developed a simpler bivariate model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). All statistical analyses were performed using the scikit-learn library in Python and the cmprsk and riskRegression packages in R.

Ethical approval for the study was obtained from the institutional review board of Hamamatsu University School of Medicine (approval number 24-162). The study was conducted in accordance with the principles of the Declaration of Helsinki. Study information was publicly disclosed on the hospital website, which also provided an opt-out option for participants.

Results

Patient characteristics

A total of 431 patients from the JALSG CML212 trial were included in this exploratory analysis. Baseline characteristics are summarized in Table 1. The median age was 53 years (range, 17-90), and 63.8% of patients were male. First-line 2G-TKI therapy consisted of dasatinib in 213 patients (49.4%) and nilotinib in 218 patients (50.6%). Based on the ELTS score, 335 patients (77.7%) were classified as low risk, 73 (16.9%) as intermediate risk, and 23 patients (5.4%) were classified as high risk. Additional chromosomal abnormalities were observed in 13% of patients. Baseline features were similar between treatment arms.

Table 1.

Patient characteristics at baseline

Factor	N = 431
Age, median (range), y	53 (17-90)
Sex, n (%)
Female	156 (36.2)
Male	275 (63.8)
Treatment, n (%)
Dasatinib	213 (49.4)
Nilotinib	218 (50.6)
ECOG performance status, n (%)
0	374 (86.8)
1	45 (10.4)
2	3 (0.7)
Unknown	9 (2.1)
Sokal score, n (%)
Low	191 (44.3)
Intermediate	161 (37.4)
High	79 (18.3)
ELTS score, n (%)
Low risk	335 (77.7)
Intermediate risk	73 (16.9)
High risk	23 (5.4)
ACA, n (%)
0	371 (86.1)
1	56 (13.0)
Unknown	4 (0.9)

ACA, additional chromosomal abnormality; ECOG, Eastern Cooperative Oncology Group.

Clinical outcomes

The median follow-up duration was 3.0 years (range, 0.01-6.50). The cumulative incidence rates of achieving MMR and MR4.5 during the observation period were 81.2% and 46.6%, respectively (supplemental Figure 1). The median times to achieve MMR and MR4.5 were 6 months (range, 2.5-38.7) and 12.2 months (range, 3.0-39.8), respectively.

The median HT (0-3) and HT (3-6) were 12.0 days (range, 6.8-183.1) and 38.0 days (range, 7.8-2292.4), respectively. Detailed clinical courses of all patients during the observation period, including the status after MR4.5 achievement of the 201 patients who reached MR4.5, are shown in supplemental Figure 2. Among these, 73 patients who sustained MR4.5 for at least 2 years and received TKI therapy for a minimum of 3 years were considered to have achieved TDE. In contrast, 53 patients lost MR4.5 after initially achieving it.

Predictors for achievement of MR4.5

To refine variable selection and minimize redundancy among correlated early-phase parameters, we first performed an exploratory principal component analysis using baseline and early molecular variables. The details of this analysis are provided in supplemental Table 1 and supplemental Figure 3. In addition, we analyzed early MR trajectories, which illustrate BCR::ABL1 IS kinetics over 0 to 6 months (supplemental Figure 4). These results suggest that molecular and clinical parameters available within the first 6 months of therapy may contain sufficient predictive information for this analysis.

To identify early predictors of achieving MR4.5, we first performed univariable analyses. Significant predictors included BCR::ABL1 IS at baseline, 3 months, and 6 months, HT (0-3), HT (3-6), and the ELTS score (Table 2). Time-dependent C-indices further demonstrated strong discriminative ability for BCR::ABL1 IS at 6 months (C-index = 0.95 at 6 months, 0.79 at 36 months) and HT (0-3) (C-index = 0.79 at 6 months, 0.69 at 36 months), suggesting their strong clinical relevance.

Table 2.

Univariate Fine-Gray subdistribution hazard model for MR4.5 achievement

Covariate	Coef	HR (95% CI)	P value	C-index (6 months)	C-index (36 months)
Age	0.03	1.03 (0.71-1.5)	.864	0.49	0.5
Female (ref male)	−0.13	0.88 (0.66-1.17)	.374	0.48	0.52
Sokal score	0.07	1.08 (0.9-1.28)	.41	0.62	0.51
ELTS score	−1.04	0.35 (0.24-0.53)	<.001	0.65	0.62
Dasatinib (ref nilotinib)	0	1.0 (0.76-1.31)	.983	0.55	0.5
BCR::ABL1 IS at baseline	−0.39	0.68 (0.49-0.95)	.023	0.59	0.55
BCR::ABL1 IS at 3 months	−0.57	0.57 (0.48-0.67)	<.001	0.85	0.73
BCR::ABL1 IS at 6 months	−0.72	0.49 (0.44-0.54)	<.001	0.95	0.79
HT (0-3)	−2.06	0.13 (0.07-0.24)	<.001	0.79	0.69
HT (3-6)	−0.47	0.62 (0.48-0.82)	<.001	0.89	0.66

Coef, coefficient; ref, reference variable.

These significant variables were incorporated into a Fine-Gray model after filtering for multicollinearity to construct a multivariable model. Based on this preselection, the final multivariable model included BCR::ABL1 IS at 6 months, HT (0-3), HT (3-6), and the ELTS score. Among these, BCR::ABL1 IS at 6 months (hazard ratio [HR], 0.38; 95% confidence interval [CI], 0.31-0.47; P < .001) and HT (0-3; HR, 2.23; 95% CI, 1.09-4.55; P = .03) remained independent predictors of MR4.5 attainment (Table 3).

Table 3.

Multivariate Fine-Gray subdistribution hazard model for MR4.5 achievement

Covariate	Coef	HR (95% CI)	P value
BCR::ABL1 IS at 6 months	−0.96	0.38 (0.31-0.47)	<.001
HT (0-3)	0.80	2.23 (1.09-4.55)	.03
HT (3-6)	0.25	1.28 (0.94-1.75)	.12
ELTS score	−0.20	0.82 (0.49-1.36)	.44

Coef, coefficient.

To further validate these findings, cumulative incidences of MR4.5 attainment were estimated after stratification by quartiles of HT (0-3) and by BCR::ABL1 IS categories at 6 months (Figure 2A-B). Stratification by HT (0-3) showed a decreasing trend, with 36-month MR4.5 probabilities ranging from 78.6% (Q1) to 23.5% (Q4). Similarly, BCR::ABL1 IS at 6 months demonstrated a clear discriminative trend, in which 96.8% of patients with BCR::ABL1 IS ≤0.01% (G1) achieved MR4.5 by 36 months, whereas no patient with BCR::ABL1 IS >1.0% (G4) did. These findings confirm the superior predictive value of 6-month BCR::ABL1 IS and the complementary prognostic contribution of early transcript kinetics.

Figure 2.

Prediction of MR4.5 attainment and TKI discontinuation eligibility. (A) Cumulative incidence of MR4.5 by quartiles of HT (0M-3M; Aalen-Johansen method with competing risks). The 36-month probabilities ranged from 78.6% in Q1 (HT <10.29 days) to 23.5% in Q4 (HT >15.8 days). MR4.5 was defined as a BCR::ABL1 IS value of ≤0.0032%. (B) Cumulative incidence of MR4.5 by 6-month BCR::ABL1 IS categories, including ≤0.01%, 0.01% to 1%, and >1%. No patient with IS >1% achieved MR4.5 by 36 months. MR4.5 was defined as a BCR::ABL1 IS value of ≤0.0032%. (C) Cumulative incidence of achieving TDE (Aalen-Johansen method) according to 6-month IS categories, including <0.01% (63.5% at 36 months), 0.01% to 0.1% (39.6%), and >0.1% (0%). No patient with IS >0.1% achieved TDE by 36 months. (D) Bivariate logistic model for evaluating TDE. Predicted probability of TDE within 36 months based on the BCR::ABL1 IS at 6 months and the ELTS score (AUC = 0.855). The model demonstrates that the probability of TDE by 36 months declines once IS exceeds 0.01%, regardless of ELTS risk category. The vertical color scale represents the predicted probability of achieving TDE within 36 months, with warmer colors indicating higher probabilities and cooler colors indicating lower probabilities. G1, group 1; HT (0M-3M), halving time during the first 3 months of therapy; Q1, quartile 1.

Predictors for achievement of TDE

Building on the same principal component analysis–guided variable selection and early molecular framework described previously (supplemental Table 1; supplemental Figures 3 and 4), we next explored predictors of TDE, defined as sustained MR4.5 or DMR for at least 2 consecutive years and a total TKI treatment duration of at least 3 years.

In univariable logistic regression, BCR::ABL1 IS at 3 and 6 months, HT (0-3), HT (3-6), and the ELTS score were identified as significant predictors of achieving TDE (Table 4). In the multivariable model, only BCR::ABL1 IS at 6 months remained an independent predictor of achieving TDE (odds ratio [OR], 0.37; 95% CI, 0.24-0.56; P < .001; Table 5). The overall model demonstrated excellent discrimination (AUC = 0.855).

Table 4.

Univariate logistic regression model for achievement of TKI discontinuation eligibility

Covariate	Coef	OR (95% CI)	P value	AUC
Age	−0.241	0.79 (0.34-1.81)	.571	0.53
Female (ref male)	−0.405	0.67 (0.32-1.40)	.283	0.542
Sokal score	0.207	1.23 (0.75-2.01)	.412	0.528
ELTS score	−1.231	0.29 (0.12-0.68)	.005	0.684
Dasatinib (ref nilotinib)	−0.566	0.57 (0.29-1.10)	.095	0.57
BCR::ABL1 IS at baseline	−0.376	0.69 (0.37-1.28)	.239	0.543
BCR::ABL1 IS at 3 months	−0.673	0.51 (0.37-0.70)	<.001	0.73
BCR::ABL1 IS at 6 months	−0.81	0.44 (0.35-0.56)	<.001	0.841
HT (0-3)	−2.533	0.08 (0.02-0.32)	<.001	0.711
HT (3-6)	−0.903	0.41 (0.25-0.67)	<.001	0.717

Coef, coefficient; ref, reference variable.

Table 5.

Multivariate logistic regression model for achievement of TKI discontinuation eligibility

Covariate	Coef	OR (95% CI)	P value
BCR::ABL1 IS at 6 months	−0.995	0.37 (0.24-0.56)	<.001
ELTS score	−0.521	0.59 (0.23-1.51)	.275
HT (0-3)	0.772	2.16 (0.35-13.22)	.403
HT (3-6)	0.404	1.50 (0.80-2.81)	.209

The overall discriminative performance of the model was high (AUC = 0.855).

Coef, coefficient.

Cumulative incidence curves stratified by 6-month BCR::ABL1 IS further supported this association (Figure 2C). Patients with BCR::ABL1 IS levels of 0.01% to 0.1% and <0.01% at 6 months had 39.6% and 63.5% probabilities of achieving TDE by 36 months, respectively, whereas none of the patients with BCR::ABL1 IS >0.1% achieved TDE. Collectively, these findings suggest that an early MMR, particularly BCR::ABL1 IS ≤0.01% at 6 months, serves as a robust indicator for identifying patients who may safely discontinue TKI therapy within an optimal timeframe. In further analysis focusing on outcomes beyond 6 months, among patients with BCR::ABL1 IS 0.01% to 0.1% at 6 months, 69.6% of those who achieved MR4.5 by 12 months subsequently met the criteria for TDE by 36 months (supplemental Figure 5).

As a sensitivity analysis, we performed additional analyses defining TDE based on sustained MR4.0 (BCR::ABL1 IS ≤0.01). In univariable logistic regression, dasatinib treatment arm, the ELTS score, BCR::ABL1 IS at 3 and 6 months, HT (0-3), and HT (3-6) were identified as significant predictors of MR4.0-based TDE (supplemental Table 2). In multivariable analysis, 6-month BCR::ABL1 IS remained the only independent predictor of MR4.0-based TDE (OR, 0.38; 95% CI, 0.26-0.55; P < .001, supplemental Table 3). When stratified by BCR::ABL1 IS at 6 months, a stepwise difference in TDE probability was observed across G1 to G4 (69.4%, 59.3%, 1.9%, and 0.0%, respectively; supplemental Figure 6), with patients achieving DMRs at 6 months showing higher TDE rates. Overall, these findings indicate that the predictive value of the 6-month BCR::ABL1 IS remains consistent even when using an MR4.0-based definition of TDE.

Modeling for achievement of TDE

We developed a logistic regression model to predict achievement of TDE by 36 months in patients with CP-CML treated with 2G-TKIs. This model incorporated BCR::ABL1 IS at 6 months, which showed the strongest association with TDE in multivariable analysis, and the ELTS score, the most widely used baseline risk stratification tool.

This bivariate model demonstrated discriminative performance comparable to that of the 4-variable model (AUC, 0.855), underscoring its clinical practicality. The model emphasized the predominant predictive value of BCR::ABL1 IS at 6 months, with a modest contribution from the ELTS score. As illustrated in Figure 2D, patients with BCR::ABL1 IS ≤0.01% at 6 months were more likely to achieve TDE, whereas those with BCR::ABL1 IS >0.1% rarely met the criteria, irrespective of ELTS risk category. This finding indicates that, even under 2G-TKI therapy, patients who achieve MR4.0 within the first 6 months are likely to meet the criteria for TDE within 36 months of TKI initiation, underscoring the importance of molecular monitoring during this period of treatment.

Discussion

Our findings demonstrate that the BCR::ABL1 IS value at 6 months is the most robust early predictor of both the attainment of MR4.5 and TDE in patients with CP-CML treated with frontline 2G-TKIs, based on analyses of ∼400 patients enrolled in the multicenter phase 3 JALSG CML212 trial¹⁵ who received either nilotinib or dasatinib. Randomized studies have consistently shown that 2G-TKIs achieve faster and deeper MRs than imatinib, including higher rates of MMR and MR4.5,8, 9, 10 supporting their use in patients for whom attaining DMR and TFR represent major therapeutic priorities. Our results highlight that, in patients receiving 2G-TKIs, integrating early molecular monitoring into treatment planning is crucial when aiming toward DMR and TFR as key therapeutic goals.

Although several previous studies have proposed early molecular predictors of achieving TDE, their findings have been inconsistent across cohorts, underscoring the need to clarify contextual factors contributing to these discrepancies. Although Shanmuganathan et al identified HT (0-3) as an independent predictor of subsequent TDE,¹³ and Saugues et al reported that the BCR::ABL1 IS level at 6 months predicted MR4.5 but not TDE,¹⁴ our analysis demonstrated that BCR::ABL1 IS value at 6 months was the most robust early predictor of both end points. These partially divergent findings are likely attributable to differences in the definition of TDE and the analytical timeframe. In our study, TDE required at least 3 years of TKI therapy and 2 consecutive years of sustained MR4.5, with the BCR::ABL1 IS value at 36 months serving as the final assessment point. In contrast, Shanmuganathan et al evaluated cumulative TDE attainment within 5 years of therapy,¹³ whereas Saugues et al applied stricter criteria (≥5 years of TKI and ≥2 years of MR4.5) with a median follow-up of 62 months.¹⁴ Variation in TKI composition may have further influenced the identified predictors, as MRs generally occur more slowly in imatinib-treated patients compared with those receiving 2G-TKIs. All patients in our cohort received 2G-TKIs, whereas only 14.8% and 28.9% of patients in the studies by Shanmuganathan et al¹³ and Saugues et al,¹⁴ respectively, were treated with 2G-TKIs. Consistent with these treatment differences, the median HT (0-3) in our cohort was shorter (12 days) than in the previous studies (13.95 and 17 days, respectively). Collectively, these observations highlight the importance of evaluating early molecular markers within a TKI-specific treatment framework to establish accurate predictive benchmarks for clinical decision-making.

The optimal duration of TKI exposure and sustained DMR required to achieve durable TFR remains unclear and may vary depending on treatment response. The European LeukemiaNet 2025 guidelines recommend a minimum of 4 years of 2G-TKI therapy and at least 2 years of sustained MR4.5 before attempting TFR, whereas the National Comprehensive Cancer Network guidelines suggest at least 3 years of therapy.^8,10 In the EURO-SKI trial,^17,18 which primarily included imatinib-treated patients, longer TKI duration was associated with higher TFR success rates, and each additional year reduced the risk of molecular relapse by ∼3%. In contrast, TFR studies using 2G-TKIs as first-line therapy have demonstrated outcomes comparable to those achieved with imatinib despite shorter treatment durations.^19,20 In the first-line DADI trial, the median duration of dasatinib therapy was 40.4 months, and treatment duration (≥40.4 vs <40.4 months) did not significantly affect the probability of maintaining TFR at 6 months after discontinuation.²⁰ Moreover, a meta-analysis of 10 discontinuation trials including 1689 patients found minimal additional benefit beyond 30 months of DMR.²¹ Together, these findings indicate that a subset of patients receiving 2G-TKIs may achieve durable TFR after a relatively shorter course of therapy, underscoring the importance of identifying predictive biomarkers that enable early selection of suitable candidates and help avoid unnecessary long-term exposure and treatment-related toxicity.

From the perspective of MR kinetics, the BCR::ABL1 IS decline pattern observed in this study closely followed the established biphasic exponential model of leukemic burden reduction during TKI therapy.22, 23, 24 The initial α-phase (≤3 months) reflects the clearance of differentiated leukemic cells, whereas the subsequent β-phase (6-12 months) corresponds to reduction of leukemic stem and progenitor cells. Tang et al reported daily decay rates of 0.0424 ± 0.0154% for the α-phase and 0.0014 ± 0.0017% for the β-phase under nilotinib therapy,²⁴ values that closely aligned with our observations (supplemental Figure 4). Patients who failed to reach MR4.5 exhibited significantly slower decay rates in both phases than those who achieved MR4.5, and the α-to-β transition point occurred later in patients who attained TDE (4.1 vs 3.7 months). The most pronounced between-group difference in BCR::ABL1 IS was observed at 6 months, supporting the hypothesis that patients who ultimately achieve TDE may have a higher proportion of differentiated and TKI-sensitive leukemic cells at diagnosis, enabling more efficient early clearance and facilitating DMRs. Consistent with this interpretation, a recent study employing artificial intelligence-based image analysis of diagnostic bone marrow demonstrated that greater granulocytic abundance and higher levels of maturation predicted improved TFR outcomes after TKI discontinuation.²⁵ Although persistent or fluctuating polymerase chain reaction positivity during DMR and TFR has been attributed in part to lineage-specific BCR::ABL1 expression in lymphoid compartments,²⁶ the cellular composition underlying the observed BCR::ABL1 kinetics in the present study remains incompletely clarified. Elucidation of these biological mechanisms was beyond the scope of the current analysis and warrants further investigation.

This study has some limitations. First, the relatively short median follow-up duration of 36 months limits evaluation of predictors beyond this time point, and longer observation may allow additional patients to meet TDE criteria. In addition, the relatively limited follow-up period may also constrain the evaluation of later molecular milestones, such as the 12-month BCR::ABL1 IS, in relation to long-term TDE outcomes. However, no patients with BCR::ABL1 IS >0.1% at 6 months achieved TDE within the 36-month follow-up period, indicating that poor early responders rarely met the threshold within the evaluated timeframe, supporting previously established cutoffs.¹¹ Second, we did not directly assess whether early molecular predictors, particularly the 6-month BCR::ABL1 IS value, correlate with successful TFR after treatment discontinuation. Currently, patients from the JALSG CML212 trial who have met TDE criteria are being prospectively enrolled into the dasatinib and nilotinib discontinuation studies (JALSG D-STOP and N-STOP trials),^27,28 and future analyses integrating these data sets are expected to validate our predictive model and clarify whether early predictors are associated with sustained TFR after treatment cessation. In addition, this study was conducted in a predominantly Japanese population; therefore, potential ethnic or regional differences in toxicity profiles and response kinetics may limit the generalizability of these findings to more diverse patient populations.

In conclusion, within 36 months of follow-up, the BCR::ABL1 IS value at 6 months represents a robust and practical early biomarker for predicting both achievement of MR4.5 and TDE in patients with CP-CML treated with 2G-TKIs. Although transcript kinetics such as HT (0-3) provide complementary prognostic information, their incremental value becomes limited once the 6-month IS was available. Early identification of patients likely to achieve TDE may enable optimization of treatment duration, reduce unnecessary long-term exposure, and support advancement of individualized, cost-effective management strategies in the modern TFR era.

Conflict-of-interest disclosure: T. Ono and S.F. received honoraria from Otsuka Pharmaceutical, Pfizer Japan, and Novartis Pharma. N.T. received honoraria from Novartis Pharma, Otsuka Pharmaceutical, and Pfizer Japan; a grant from Asahi Kasei Pharma and Otsuka Pharmaceutical; and research funding from Novartis Pharma, Otsuka Pharmaceutical, Pfizer Japan, and Astellas Pharma. Y. Minami received consulting fees and honoraria from Astellas Pharma, Novartis Pharma, Pfizer Japan, Takeda Pharmaceutical, and Daiichi Sankyo; and research funding from Novartis Pharma, Ono Pharmaceutical, and CMIC Co, Ltd. C.N. received honoraria from Pfizer Japan, Janssen Pharmaceutical, and Fujimoto Pharmaceutical. N.I. received honoraria and speaker fees from Novartis Pharma, Pfizer Japan, Ono Pharmaceutical, Otsuka Pharmaceutical, AbbVie, Nippon Shinyaku, Chugai Pharmaceutical, Janssen Pharmaceutical, Meiji Seika Pharma, and Asahi Kasei Pharma. K.F. received honoraria from Bristol Myers Squibb, Otsuka Pharmaceutical, Pfizer Japan, and Novartis Pharma. Y.A. received honoraria from Otsuka Pharmaceutical Co, Ltd and Chugai Pharmaceutical; and consulting fees from JCR Pharmaceuticals Co, Ltd. T.M. received honoraria from Genmab, Eisai, Chugai Pharmaceutical, Kyowa Kirin, and Gilead Sciences; and research funding from LSI Medience Corporation. E.S. received honoraria from Novartis Pharma, Janssen Pharmaceutical, Pfizer Japan, and Sanofi. H.K. received honoraria from Astellas Pharma, AbbVie, and Chugai Pharmaceutical; and research funding from Fujifilm, Kyowa Kirin, Bristol Myers Squibb, Otsuka Pharmaceutical, Perseus Proteomics, Daiichi Sankyo, AbbVie, CURED Inc, Astellas Pharma, Chugai Pharmaceutical, Zenyaku Kogyo, Nippon Shinyaku, Eisai, Takeda Pharmaceutical, and Sumitomo Pharma. Y. Miyazaki received honoraria from Incyte Biosciences Japan, Novartis Pharma, Chugai Pharmaceutical, Bristol Myers Squibb, Daiichi Sankyo, and AbbVie. I.M. received grants from Chugai Pharmaceutical, Asahi Kasei Pharma, Kyowa Kirin, Sumitomo Pharma, Eisai, Taiho Pharmaceutical, AbbVie, and Otsuka Pharmaceutical; and honoraria from Chugai Pharmaceutical, Novartis Pharma, Bristol Myers Squibb, Janssen Pharmaceutical, Pfizer Japan, Otsuka Pharmaceutical, AbbVie, Takeda Pharmaceutical, Ono Pharmaceutical, SymBio Pharmaceuticals, Sanofi, Astellas Pharma, and AstraZeneca. The remaining authors declare no competing financial interests.