一代酪氨酸激酶抑制剂（tyrosine kinase inhibitor，TKI）伊马替尼的上市，使慢 性髓性白血病（chronic myeloid leukemia，CML）患者实现了与普通人群相似的生 存期。然而，由于分子异质性，部分患者仍然会经历治疗失败，少数高危患者甚至 面临疾病进展，预后极差。二代TKI能快速诱导深度缓解，是高危患者和追求无治 疗缓解患者的更优一线选择，但失败后治疗存在挑战。基于此，我们结合日常临床 工作中的相关经验，针对以下问题展开研究：1）不典型转录本对患者结局和治疗反 应的潜在影响；2）低比例异常B淋巴母细胞的进展意义；3）预测二代TKI一线治 疗的失败因素。本研究旨在基线识别治疗失败和疾病进展的高危患者，为早期个体 化干预提供理论依据，以进一步优化CML精准治疗，改善患者结局。

第一章 不典型转录本对患者治疗反应的潜在影响

背景：大约 0.3%~4.6%的 CML 患者携带不典型 BCR::ABL1 转录本，与典型 e13a2/e14a2 转录本患者有着差异化临床表现，该类患者的生存结局及对 TKI 的治 疗反应存在争议。

方法：系统回顾就诊CML患者的转录本信息，筛选出携带不典型转录本的患者。 按1:4比例（不典型:典型）进行倾向性评分匹配，校正年龄和性别差异后，比较典 型转录本患者和不典型转录本患者的结局和疗效反应。并进一步按不典型转录本类 型，分别表征患者在一代TKI和二代TKI治疗下的完全细胞遗传学反应（complete cytogenetic response，CCyR）获得率。

结果：2724例CML患者中52例携带不典型转录本，占CML患者的1.9%，涵盖8 种亚型，最常见的3种亚型为e19a2（n=20，38.5%）， e1a2（n=10，19.2%），以及 e13a3/e14a3（n=12，23.1%）。其中 e19a2 多见于女性（p=0.038），患者年龄更大 （p=0.037），血小板水平显著升高（p=0.012）；e1a2也以女性患者为主（p=0.010）。 与典型转录本患者相比，e19a2 患者（85.3% vs. 41.2%，p＜0.0001）和 e1a2 患者 （87.5% vs. 16.7%，p=0.003）的 12 个月 CCyR 显著更低，e19a2患者（p=0.025） 和e1a2患者（p＜0.0001）无失败生存期（failure-free survival，FFS）显著下降，e19a2 患者无进展生存期（progression-free survival，PFS）降低（p=0.048）。 典型转录本患 者与不典型转录本患者的总生存期（overall survival，OS）无统计学差异（p>0.05）。 e19a2 患者（p＜0.0001）和 e1a2 患者（p=0.008）的 CCyR 获得率在一代TKI和二 代 TKI 治疗下具有显著差异，提示 e19a2 和 e1a2 对伊马替尼的潜在耐药性。 e13a3/e14a3 患者与典型转录本患者在临床表现、治疗反应及生存结局方面均无差异。携带罕见类型不典型转录本的患者临床表现各异，其中e8a2患者具有相对惰性 的病程，而e6a2患者具有更侵袭性的病程。

结论：与典型转录本患者相比，不典型转录本患者具有差异化的临床表现和对TKI 的疗效反应，其中e19a2和e1a2患者对伊马替尼具有潜在耐药性，但在二代TKI的 治疗下，能获得与典型转录本患者相似的临床结局。

第二章 低比例异常B淋巴母细胞提示疾病进展高风险

背景：CML进展期患者治疗反应受限，预后不良。不同指南对于CML急淋变定义 不一致，提示异常B淋巴母细胞（aberrant B-lymphoblasts，ABLB）的最佳临界值 及临床意义不明。

方法：系统回顾初诊时形态为慢性期且进行流式细胞术检测的CML患者资料，将 ABLB比例＜5%定义为低比例，纳入携带低比例ABLB的患者，系统分析其临床表 现，异常细胞免疫表型，治疗反应及临床结局。进一步将单中心队列作为训练组确 定ABLB最佳截断值，既往报道患者作为验证组明确该截断值对进展风险的区分度。

结果：802 例患者在初诊时进行流式细胞术检测，其中 16 例（2.0%）携带低比例 ABLB，10例经历疾病进展。单中心数据明确与进展相关的ABLB比例最佳截断值 为0.2%，按截断值将训练组和验证组患者分为进展高风险和和进展低风险，两组患 者的PFS均具有显著差异（p=0.011；p=0.044）。进一步分析单中心患者的生存和治 疗信息，发现5例患者携带ABLB≤0.2%，没有患者发生进展，但其中2例患者经 历治疗失败，1例患者始终未达到主要分子学缓解（major molecular response，MMR）； 11 例患者携带ABLB＞0.2%，10例患者发生进展，中位无进展时间3个月（范围： 1~41 个月），其中携带1.69% ABLB的患者可能获益于强化治疗而未发生进展。

结论：ABLB比例与CML进展风险相关， ABLB＞0.2%的患者具有高进展风险， ABLB≤0.2%的患者虽然进展风险较低，但难以达到满意疗效，容易经历治疗失败。 考虑将0.2%作为风险参考值，结合 ABLB 定量监测进行风险分层，优化治疗强度 选择，可能改善患者生存结局。

第三章 基于基线特征识别一线二代TKI治疗失败的高危患者

背景：二代TKI一线使用逐步增多，但其失败后治疗存在挑战。早期识别失败的高 危患者，给予恰当干预措施，避免失败事件，可有效改善患者预后。

方法：系统回顾CML患者治疗信息，纳入接受一线二代TKI治疗的患者，并按8：2 随机划分为训练集和验证集。通过单因素Cox和限制性立方样条（restricted cubic spline，RCS）进行特征变量筛选，最小绝对值收缩和选择算子（least absolute shrinkage and selection operator，LASSO）压缩变量，结合 C 指数和赤池信息准则（Akaike information criterion，AIC）进行模型优化，构建多因素 Cox 模型，列线图量化风险 评分，进一步利用时间依赖性ROC，校准曲线以及临床决策曲线评估模型的预测性 能及临床应用价值。

结果：共521例患者接受一线二代TKI治疗，124例（23.8%）经历治疗失败，其中 29 例（ 23.4%）因不耐受药物副作用导致失败。多因素Cox模型纳入6个关键变量： 性别，骨髓纤维化（myelofibrosis，MF）分级，脾脏肋下厘米数，血红蛋白水平， 骨髓原始细胞比例以及骨髓嗜碱性粒细胞比例，其中骨髓嗜碱性粒细胞比例与治疗 失败呈非线性相关性。最终模型C指数为0.762。模型在早期（6个月）和晚期（24 个月和60个月）失败事件的预测中，AUC值均高于ELTS评分和Sokal评分，其 中6个月、24个月和60个月的AUC值分别为0.812 vs. 0.713 vs. 0.692、0.777 vs. 0.712 vs. 0.690，以及 0.762 vs. 0.674 vs. 0.665（模型 vs. ELTS 评分 vs. Sokal 评分）。 校准曲线显示预测风险与实际风险高度吻合，决策曲线提示模型可带来临床收益。 根据风险评分最佳截断值将训练集和验证集患者分为失败高危组和低危组，两组患 者的FFS具有显著差异。

结论：本研究开发的预测模型整合基线时关键临床变量，量化评估一线二代TKI治 疗失败风险，其效能显著优于传统预后评分，可针对性识别一线二代TKI失败的高 危患者，辅助临床治疗决策。  

ABSTRACT

The first-generation tyrosine kinase inhibitor (TKI), imatinib, has enabled patients with chronic myeloid leukemia (CML) to achieve comparable survival to those of the general population. Nevertheless, molecular heterogeneity leads to diverse treatment responses, with some experiencing treatment failures and even disease progression, resulting in an inferior prognosis. Second-generation TKIs can rapidly induce deep remission, making them a promising first-line option for high-risk patients and those aiming for treatment free remission; however, managing post-failure treatment remains challenging. In light of this, we combined our clinical experiences with a dedicated study to address the following issues: 1) the potential impact of atypical transcripts on patient outcomes and treatment response; 2) the progression significance of low levels of abnormal B-lymphoblasts in patients with chronic phase disease; and 3) the factors predicting the failure of upfront second-generation TKIs. The aim of this study is to early identify patients at high risk of treatment failure and disease progression, thereby optimizing precision therapy for CML and improving patient outcomes.

Part I The Impact of Atypical Transcripts on Treatment Response to Tyrosine Kinase Inhibitors in Chronic Myeloid Leukemia

Background: While the majority of CML patients present with typical BCR::ABL1 transcripts, a small percentage (0.3%-4.6%) display atypical variants with poorly defined clinical implications.

Methods: A systematic retrospective review was conducted on CML patients with atypical transcripts. Propensity score matching was performed to minimize age and gender differences between patients with typical and atypical transcripts. The clinical features, treatment response, and outcomes were estimated based on transcript types. Results: Out of 2724 CML patients, 52 (1.9%) had atypical transcripts, including eight different types. The three most common types were e19a2 (n=20, 38.5%)，e1a2 (n=10, 19.2%), and e13a3/e14a3 (n=12, 23.1%). Compared to patients with typical transcripts, those with e19a2 (n=17) were older (p=0.037), more frequently female (p=0.038), and had elevated platelet levels (p=0.012). Similarly, patients with e1a2 (n=6) were predominantly female (p=0.010). Regarding treatment response to TKIs, the lower cumulative rates of complete cytogenetic response (CCyR) by 1 year were seen in patients with e19a2 (41.2% vs 85.3%, p<0.0001) and e1a2 (16.7% vs 87.5%, p=0.003) compared to those with typical  transcripts. The probabilities of failure-free survival (FFS) were lower in patients with e19a2 (p=0.025) and e1a2 (p<0.0001). Additionally, the probabilities of progression-free survival (PFS) were lower in patients with e19a2 (p=0.048). However, overall survival (OS) was comparable between atypical and typical cohorts (p>0.05), likely attributable to salvage therapy with second-generation TKIs. No differences were found in clinical features, treatment response and outcomes among patients with e13a3/e14a3 (n=11). Notably, rare subtypes exhibited diverse disease manifestations, with e8a2 linked to indolent course and e6a2 associated with rapid progression.

Conclusions: Atypical transcript subtypes define a clinically heterogeneous CML subgroup with differential TKI sensitivity. Patients with e19a2 and e1a2 transcripts exhibit potential resistance to imatinib but achieve similar clinical outcomes to patients with typical transcripts under second-generation TKI therapy.

Part II The Progression Significance of Low-level Aberrant B- lymphoblasts in Patients with Chronic Myeloid Leukemia

Background: CML patients with progression events typically exhibit limited treatment responses and outcomes. The definition of blast crisis varies across treatment criteria, making the optimal cutoff for lymphoblasts and the significance of low-level (<5%) aberrant B-lymphoblasts (ABLB) unclear.

Methods: A systematic review was conducted to analyze flow cytometry results in CML patients. A low level of ABLB was defined as less than 5%. The review further examined the clinical characteristics, immunophenotype of abnormal cells, treatment responses, and clinical outcomes of patients with low ABLB levels. The single-center cohort was utilized as a training group to establish the cutoff value for ABLB, while patients from previously reported studies were included as a validation group to confirm the effectiveness of this cutoff value.

Results: Flow cytometry analysis was conducted on 802 patients at diagnosis. Among these patients, 16 (2.0%) had a low level of ABLB, and 10 of them experienced disease progression. The optimal cutoff value of ABLB associated with progression was determined to be 0.2%. Patients were classified into high-risk and low-risk groups based on this cutoff, showing a significant difference in PFS between the two groups in both training and validation cohorts (p=0.011; p=0.044). Further analysis indicated that five patients with an ABLB≤0.2% had a low risk of progression but struggled to achieve satisfactory treatment responses, with two of them experiencing treatment failures. In contrast, 11 patients with an ABLB＞0.2% showed a high risk of progression, with 10 of them experiencing progression. The median time to FFS was 3 months (range: 1-41 months). Notably, the patient with an ABLB of 1.69% who did not experience progression may benefit from intensive therapy.

Conclusion: Patients with ABLB＞0.2% are at high risks of progression. Monitoring ABLB levels for risk stratification and optimizing treatment options can improve patient outcomes.

Part III Predicting High-risk Patients for Upfront Second-generation Tyrosine Kinase Inhibitors Failure Using Baseline Characteristics

Background: The use of upfront second-generation TKIs has gradually increased. However, the treatment of post-failure is challenging. Early identification of high-risk patients and timely interventions to prevent failure events can significantly enhance patient prognosis.

Methods: The treatment information of CML patients was systematically reviewed. Patients treated with upfront second-generation TKIs were included and randomly divided into training and validation sets in an 80:20 ratio. Characteristic variables were selected using one-way Cox and restricted cubic spline (RCS) regression. These variables were then compressed using LASSO regression. Model optimization was performed by combining the C-index and Akaike Information Criterion (AIC). A multifactorial Cox model was constructed, and risk scores were quantified using column-line plots. The predictive performance of the model and its clinical application value were further evaluated using time-dependent receiver operating characteristic (ROC) curves, calibration curves, and clinical decision curves.

Results: Treatment failure occurred in 124 out of 521 patients (23.8%), with 29 cases (23.4%) attributed to intolerance of side effects. A multifactorial Cox model was developed, incorporating six key variables: gender, myelofibrosis, spleen size, hemoglobin level, percentage of blasts and basophils in bone marrow. This model achieved a C-index of 0.762. The model demonstrated higher AUC values compared to both the ELTS score and the Sokal score when predicting early (6 months) and late (24 and 60 months) failure events. Specifically, the AUC values of 6 months, 24 months, and 60 months were 0.812 vs. 0.713 vs. 0.692, 0.777 vs. 0.712 vs. 0.690, and 0.762 vs. 0.674 vs. 0.665, respectively (model vs. ELTS score vs. Sokal score). The calibration curve indicated that the predicted risk aligned closely with the actual risk observed. Patients in the training and validation sets were classified into high-risk and low-risk failure groups based on the optimal cutoff value, with a significant difference in FFS between these two groups.

Conclusion: The prediction model incorporates clinical variables at baseline, providing a quantitative assessment of failure risk associated with upfront second-generation TKI therapy in CML patients. The effectiveness of this model is better than traditional prognostic scores, enabling early identification of high-risk patients and facilitating informed clinical treatment decisions.