In-depth analysis for TKI-driven real-world management of 201 CML patients using TFR.
[BACKGROUND] Tyrosine-kinase inhibitors (TKIs) have reshaped chronic myeloid leukemia (CML) outcomes, but real-world data from Eastern Europe remain scarce.
APA
Micu ML, Cira SF, et al. (2025). In-depth analysis for TKI-driven real-world management of 201 CML patients using TFR.. Frontiers in pharmacology, 16, 1673056. https://doi.org/10.3389/fphar.2025.1673056
MLA
Micu ML, et al.. "In-depth analysis for TKI-driven real-world management of 201 CML patients using TFR.." Frontiers in pharmacology, vol. 16, 2025, pp. 1673056.
PMID
41282616
Abstract
[BACKGROUND] Tyrosine-kinase inhibitors (TKIs) have reshaped chronic myeloid leukemia (CML) outcomes, but real-world data from Eastern Europe remain scarce.
[METHODS] We retrospectively analyzed 201 adult patients with CML managed at the Cluj-Napoca Department of Haematology (January 2001-December 2024). A semi-automated pipeline utilizing a Large Language Model was developed to extract structured data from unstructured medical discharge report text, with all patient identifiers removed to ensure anonymity. We captured demographics, disease phase, line-specific TKI use, adverse events (AEs), treatment-free remission (TFR) eligibility, TFR attempts, continuation, laboratory data and discontinuation. Machine learning models were trained to predict TFR potential.
[RESULTS] Patients <60 years old, 101/201 (50.2%), and ≥60 years old, 100/201 (49.8%) were nearly equal in number. 53.7% were male. At diagnosis, 94.5% were in chronic phase. First-line treatment comprised imatinib in 108/201 (53.7%), dasatinib in 56/201 (27.9%), and nilotinib in 37/201 (18.4%). Second-line therapy ( = 64) was dominated by dasatinib (64.1%) and nilotinib (28.1%). Third- and later-line regimens increasingly incorporated bosutinib, ponatinib, and asciminib. Fourteen patients (7.0%) achieved sustained treatment-free remission (TFR). Among these, 3 had received imatinib, 3 dasatinib, and 8 nilotinib. An additional 31 patients (15.4%) were TFR-eligible but still on therapy-16 of them after imatinib, 9 after dasatinib and 6 after nilotinib. Imatinib achieved MR4+ in 29% of exposures and nilotinib in 43.3% of third-line uses, underscoring its role as the cohort's most effective TFR-enabler. Predictive modelling for TFR potential using a Random Forest classifier achieved high accuracy (85.4%), with top predictors being whether a patient had ever achieved a deep molecular response (achieved_mr4_ever) and the best response after the first year (best_response_after_year1), highlighting the importance of both depth and timing of molecular remission. Adverse events led to discontinuation in 17/105 imatinib (16.1%), 23/105 dasatinib (21.9%), 9/68 nilotinib (13.2%), 6/13 bosutinib (46.2%), 4/11 ponatinib (36.4%), and 2/11 asciminib (18.2%) exposures. The most common imatinib discontinuation cause was loss of therapeutic response (34/105; 32.4%).
[CONCLUSION] In this Romanian center, imatinib was the predominant front-line TKI, reflecting both its earlier availability as the sole treatment option and the durable responses achieved by many long-term patients, but second-generation agents are increasingly used in over 40% of first-line starts. TFR uptake is limited despite a sizeable eligible population. Machine learning models demonstrate that both the depth and kinetics of molecular response are critical for predicting TFR potential. Prospective optimization of molecular monitoring and discontinuation protocols may broaden TFR success.
[METHODS] We retrospectively analyzed 201 adult patients with CML managed at the Cluj-Napoca Department of Haematology (January 2001-December 2024). A semi-automated pipeline utilizing a Large Language Model was developed to extract structured data from unstructured medical discharge report text, with all patient identifiers removed to ensure anonymity. We captured demographics, disease phase, line-specific TKI use, adverse events (AEs), treatment-free remission (TFR) eligibility, TFR attempts, continuation, laboratory data and discontinuation. Machine learning models were trained to predict TFR potential.
[RESULTS] Patients <60 years old, 101/201 (50.2%), and ≥60 years old, 100/201 (49.8%) were nearly equal in number. 53.7% were male. At diagnosis, 94.5% were in chronic phase. First-line treatment comprised imatinib in 108/201 (53.7%), dasatinib in 56/201 (27.9%), and nilotinib in 37/201 (18.4%). Second-line therapy ( = 64) was dominated by dasatinib (64.1%) and nilotinib (28.1%). Third- and later-line regimens increasingly incorporated bosutinib, ponatinib, and asciminib. Fourteen patients (7.0%) achieved sustained treatment-free remission (TFR). Among these, 3 had received imatinib, 3 dasatinib, and 8 nilotinib. An additional 31 patients (15.4%) were TFR-eligible but still on therapy-16 of them after imatinib, 9 after dasatinib and 6 after nilotinib. Imatinib achieved MR4+ in 29% of exposures and nilotinib in 43.3% of third-line uses, underscoring its role as the cohort's most effective TFR-enabler. Predictive modelling for TFR potential using a Random Forest classifier achieved high accuracy (85.4%), with top predictors being whether a patient had ever achieved a deep molecular response (achieved_mr4_ever) and the best response after the first year (best_response_after_year1), highlighting the importance of both depth and timing of molecular remission. Adverse events led to discontinuation in 17/105 imatinib (16.1%), 23/105 dasatinib (21.9%), 9/68 nilotinib (13.2%), 6/13 bosutinib (46.2%), 4/11 ponatinib (36.4%), and 2/11 asciminib (18.2%) exposures. The most common imatinib discontinuation cause was loss of therapeutic response (34/105; 32.4%).
[CONCLUSION] In this Romanian center, imatinib was the predominant front-line TKI, reflecting both its earlier availability as the sole treatment option and the durable responses achieved by many long-term patients, but second-generation agents are increasingly used in over 40% of first-line starts. TFR uptake is limited despite a sizeable eligible population. Machine learning models demonstrate that both the depth and kinetics of molecular response are critical for predicting TFR potential. Prospective optimization of molecular monitoring and discontinuation protocols may broaden TFR success.