The approval of ibrutinib, a Bruton's Tyrosine Kinase (BTK) inhibitor, has revolutionized the treatment landscape for treatment-naïve and relapsed or refractory chronic lymphocytic leukemia (CLL) patients. Forty-four percent of CLL patients enrolled in the inform CLL registry received ibrutinib—making it the most commonly used CLL treatment in the current era of novel targeted therapies. Ibrutinib demonstrated a 6.5 year progression-free survival (PFS) of 61% in untreated CLL compared to 9% with chlorambucil (RESONATE-2 trial), and also improved PFS in both untreated (E1912 and A041202 trials) and previously treated (HELIOS trial) CLL compared to chemoimmunotherapy. However, bleeding was frequently observed in ibrutinib-treated patients and has become a notable safety concern.

Bleeding of any severity occurs in up to 50% of ibrutinib-treated patients,¹ due to ibrutinib's on-target BTK and potentially off-target (Tec) effects on collagen-induced platelet aggregation. Major bleeding, however, was uncommonly observed in clinical trials, with incidences ranging from 2%–8%.^{1, 2} Although major bleeding was uncommon in these studies, it may occur more frequently in clinical practice due to its use in patients who were excluded from clinical trials, including those with renal dysfunction and/or those taking concomitant warfarin with or without cytochrome P450 3A4/5 (CYP3A4/5) inhibitors.¹ Indeed, higher incidences of major bleeding (8%–18%) have been reported in real-world populations compared to clinical trials.^{3, 4} Few studies have examined this topic, however, and those that have did not directly compare the rate of major bleeding between ibrutinib-treated individuals and individuals treated with therapeutic alternatives. Furthermore, many of these studies were limited to a single institution. Thus, the comparative rate of major bleeding with ibrutinib versus other CLL therapies in a population-based setting remains unknown. Given ibrutinib's current widespread use and the life-threatening potential of major bleeds, a more robust assessment of ibrutinib-related major bleeding may inform its benefit-risk profile, which is particularly important as 2nd-generation BTK inhibitors (e.g., acalabrutinib and zanubrutinib) emerge.

We conducted an incident-user, active comparator retrospective cohort study using 2013–2020 Optum Clinformatics de-identified commercial health insurance data to compare incidence rates of real-world major and clinically-relevant bleeding between ibrutinib- and bendamustine-rituximab (BR)-treated individuals diagnosed with CLL. Additional details on Optum Clinformatics are described in Supplemental Methods 1.1.

We included individuals who were incident users of either ibrutinib or BR between November 1, 2013 and February 29, 2020 and were diagnosed with CLL. We excluded individuals who were co-exposed to an oral anticoagulant on the index date (date of incident exposure to ibrutinib or BR). Additional details on defining the study population are described in Supplemental Methods 1.2–1.5.

Potential confounders were identified during the 6-month baseline period preceding the index date and are listed in Supplementary Table 1. We also used the high-dimensional propensity score (hdPS) approach to empirically identify additional covariates with a high potential for confounding—which are hypothesized to be proxies for unmeasured confounders. This approach ranks and selects potential confounders based on their empiric association with either the exposure or both the exposure and outcome. Details on hdPS are described in Supplemental Methods 1.6.

The primary outcome was major bleeding, defined as bleeding resulting in inpatient hospitalization. We used a previously validated algorithm to identify outcomes, which demonstrated a positive predictive value of 89% in health insurance claims data⁵ (see Supplemental Methods 1.7). The secondary outcome was clinically-relevant bleeding, which was a composite of bleeding events resulting in inpatient hospitalization (i.e., major bleeding) and those resulting in emergency department presentation.

We calculated descriptive statistics for baseline variables, crude incidence rates, and unadjusted association measures, the latter using Cox proportional hazards models. We included pre-specified covariates, along with hdPS-identified covariates, into a logistic regression model to calculate propensity scores. We calculated the stabilized inverse probability of treatment weight (sIPTW) for each individual based on propensity scores. We then estimated hazard ratios (HRs) using Cox proportional hazards regression, weighted based on sIPTW. Additional statistical details on the primary analysis as well as sensitivity and effect modification analyses are described in Supplemental Methods 1.8. The University of Pennsylvania institutional review board deemed research using this dataset to be exempt from review.

We identified 2423 and 1102 incident users of ibrutinib and BR respectively (Supplementary Figure 2). Individuals in the overall cohort were predominantly male (61.6%) and White (70.4%), with a median age of 72 years and a median frailty score of 0.135 (i.e., pre-frail category per frailty index). The majority of ibrutinib-treated individuals (88.1%) received an average daily dose of 420 mg at cohort entry. Differences in unweighted baseline characteristics between groups are highlighted in Supplemental Results 2.1. All weighted baseline characteristics that were included in the propensity score were well-balanced except for geographic region of residence (absolute standardized difference = 0.14) (Supplementary Tables 5 and 6).

Crude incidence rates for major and clinically-relevant bleeding are presented in Table 1. Major bleeding events in ibrutinib-treated individuals were predominantly gastrointestinal (47.6%), followed by cerebral (19.0%), genitourinary (14.3%), other (14.3%), and unspecified (4.8%) (Supplementary Table 7). Ibrutinib (vs. BR) had an elevated, yet statistically compatible with the null, hazard of major bleeding (sIPTW-adjusted HR: 1.61, 95% confidence interval [95% CI]: 0.67–3.84), and an elevated and statistically significant hazard of clinically-relevant bleeding (sIPTW-adjusted HR: 2.66, 95% CI: 1.29–5.49) (Table 1 and Supplementary Figure 4). Results from sensitivity analyses were consistent with our primary findings; additional details on sensitivity analysis and effect modification results can be found in Supplemental Results 2.2.

This population-based cohort study compared the rates of real-world major and clinically-relevant bleeding in ibrutinib-treated individuals versus individuals treated with chemoimmunotherapy. Ibrutinib had an elevated hazard of major bleeding compared to BR, although this finding was compatible with the null. Ibrutinib was associated with a 2.6-fold hazard of clinically relevant bleeding compared to BR, suggesting a potential risk of clinically-relevant bleeding events in patients treated with ibrutinib in routine clinical practice.

The crude incidence rate of major bleeding among ibrutinib-treated individuals in our study (3.1 per 100 person-years [p-y], 95% CI: 2.0–4.7) is comparable to that found in a meta-analysis of ibrutinib clinical trials,¹ which reported a pooled major bleeding incidence rate of 2.8 per 100 p-y. Though prior real-world observational studies did not report incidence rates, their crude incidence proportions (8%–18%)^{3, 4} were notably higher than that of our study (0.87%). Potential reasons for this difference are described in Supplemental Discussion 3.1.

Our adjusted primary outcome analysis found an elevated, yet consistent with the null, rate of major bleeding with ibrutinib compared to BR. It is likely that our study was underpowered due to the low number of outcomes; however, a lack of a difference in the rate of major bleeding between ibrutinib and BR cannot be excluded. Using Austin's method for obtaining the number needed to harm (NNH) from adjusted survival models, we calculated an NNH of 252 for major bleeding. Caron et al. and Wang et al. also found increased pooled relative risks of major bleeding versus comparator therapies in clinical trials ranging from 1.66–2.46.^{1, 2} Our adjusted secondary outcome analysis showed a positive association of ibrutinib (vs. BR) with clinically-relevant bleeding (i.e., bleeding resulting in inpatient admission or emergency department presentation). We calculated an NNH of 58 for clinically-relevant bleeding. Though prior studies did not examine a similarly defined outcome, comparative rates of lower-severity bleeding were reported. Caron et al. and Wang et al. both observed a larger relative risk with all-severity bleeding in clinical trials (pooled relative risks ranging from 2.72–3.08)^{1, 2} than with major bleeding. Abdel-Qadir et al. conducted a population-based cohort study and observed a similar association with hospital-diagnosed bleeding in ibrutinib-exposed patients compared to chemotherapy-exposed patients (HR 2.58, 95% CI 1.76–3.78).⁶ Overall, these findings support current recommendations to consider ibrutinib's potential bleeding risk when assessing it as a treatment option.

This study is the largest to date to investigate major bleeding with ibrutinib in real-world practice. A limitation is that we were unable to obtain CLL characteristics like Rai staging and high-risk genomic features such as the presence of deletion 17p or immunoglobulin heavy-chain gene mutation. However, we included CLL complications that may affect bleeding occurrence (e.g., thrombocytopenia) in the propensity score. Also, though we used a rigorous study design and adjustment method to address confounding, residual differences between ibrutinib- and BR-treated individuals may remain. Additional strengths and limitations are detailed in Supplemental Discussion 3.2–3.3.

Ibrutinib is currently the most commonly used CLL therapy in the United States. Our study found that, though the rate of major bleeding with ibrutinib in a real-world population was comparable to clinical trial populations, an increased rate of clinically-relevant bleeding compared to BR is evident. As the use of 2nd-generation BTK inhibitors increases, clinicians will require information on the real-world risks with individual agents within the therapy class. Furthermore, given the unique antiplatelet mechanism of BTK inhibitors through glycoprotein VI inhibition, research is warranted on their potential protective effect against thrombotic events, especially in patients experiencing BTK inhibitor-related atrial fibrillation. This study will help serve as a benchmark for ibrutinib-related bleeding until such post-market studies among all available BTK inhibitors can be conducted.

布鲁顿酪氨酸激酶 (BTK) 抑制剂依鲁替尼 (ibrutinib) 的批准彻底改变了初治和复发或难治性慢性淋巴细胞白血病 (CLL) 患者的治疗格局。在信息 CLL 登记处登记的 CLL 患者中有 44% 接受了依鲁替尼治疗，使其成为当前新型靶向治疗时代最常用的 CLL 治疗方法。依鲁替尼在未经治疗的 CLL 中表现出 61% 的 6.5 年无进展生存率 (PFS)，而苯丁酸氮芥的无进展生存率 (PFS) 为 9%（RESONATE-2 试验），而且未经治疗（E1912 和 A041202 试验）和既往治疗（HELIOS 试验）的 PFS 也有所改善) CLL 与化学免疫疗法的比较。然而，在接受依鲁替尼治疗的患者中经常观察到出血，这已成为一个值得注意的安全问题。

高达 50% 的接受依鲁替尼治疗的患者会出现任何严重程度的出血¹ ，这是由于依鲁替尼的靶向 BTK 以及对胶原蛋白诱导的血小板聚集的潜在脱靶 (Tec) 影响。然而，在临床试验中很少观察到大出血，发生率在 2%–8% 之间。^{1, 2}虽然大出血在这些研究中并不常见，但由于其用于被排除在临床试验之外的患者，包括肾功能不全患者和/或同时服用华法林（有或没有细胞色素 P450）的患者，因此在临床实践中可能会更频繁地发生这种情况3A4/5 (CYP3A4/5) 抑制剂。¹事实上，与临床试验相比，现实世界人群中大出血的发生率 (8%–18%) 更高。^{3, 4}然而，很少有研究探讨过这一主题，而且那些研究也没有直接比较接受依鲁替尼治疗的个体和接受替代治疗的个体之间的大出血发生率。此外，其中许多研究仅限于单个机构。因此，在基于人群的环境中，依鲁替尼与其他 CLL 疗法的大出血发生率的比较仍不清楚。鉴于依鲁替尼目前的广泛使用以及大出血危及生命的可能性，对依鲁替尼相关的大出血进行更强有力的评估可能会告知其获益风险概况，这作为第二代 BTK 抑制剂（例如 acalabrutinib 和 zanubrutinib）尤为重要出现。

我们使用 2013 年至 2020 年 Optum Clinformatics 去识别的商业健康保险数据进行了一项事件用户、主动比较者回顾性队列研究，以比较依鲁替尼和苯达莫司汀-利妥昔单抗 (BR) 之间真实世界主要出血和临床相关出血的发生率。接受治疗的被诊断患有 CLL 的个体。有关 Optum Clinformatics 的更多详细信息，请参见补充方法 1.1。

我们纳入了 2013 年 11 月 1 日至 2020 年 2 月 29 日期间使用依鲁替尼或 BR 并被诊断患有 CLL 的个人。我们排除了在索引日期（伊布替尼或 BR 事件暴露日期）同时暴露于口服抗凝剂的个体。补充方法 1.2-1.5 中描述了有关定义研究人群的更多详细信息。

在索引日期之前的 6 个月基线期间确定了潜在的混杂因素，并列在补充表 1 中。我们还使用高维倾向评分 (hdPS) 方法凭经验识别具有高混杂潜力的其他协变量，这些协变量是假设是不可测量的混杂因素的代理。这种方法根据潜在混杂因素与暴露或暴露和结果的经验关联对潜在混杂因素进行排名和选择。补充方法 1.6 中描述了有关 hdPS 的详细信息。

主要结局是大出血，定义为导致住院治疗的出血。我们使用之前经过验证的算法来识别结果，该算法在健康保险索赔数据⁵中显示出 89% 的阳性预测值（参见补充方法 1.7）。次要结局是临床相关出血，它是导致住院治疗（即大出血）的出血事件和导致急诊就诊的出血事件的综合结果。

我们计算了基线变量、粗发病率和未经调整的关联措施的描述性统计数据，后者使用 Cox 比例风险模型。我们将预先指定的协变量以及 hdPS 识别的协变量纳入逻辑回归模型中以计算倾向得分。我们根据倾向评分计算了每个人的稳定治疗权重逆概率 (sIPTW)。然后，我们使用 Cox 比例风险回归估计风险比 (HR)，并基于 sIPTW 进行加权。补充方法 1.8 中描述了主要分析以及敏感性和效果修改分析的其他统计细节。宾夕法尼亚大学机构审查委员会认为使用该数据集的研究免于审查。

我们分别确定了 2423 名 ibrutinib 和 1102 名 BR 事件使用者（补充图 2）。整个队列中的个体主要是男性（61.6%）和白人（70.4%），中位年龄为 72 岁，中位虚弱评分为 0.135（即每个虚弱指数的虚弱前类别）。大多数接受依鲁替尼治疗的个体 (88.1%) 在进入队列时接受的平均每日剂量为 420 mg。补充结果 2.1 强调了各组之间未加权基线特征的差异。除居住地理区域外，倾向评分中包含的所有加权基线特征均十分平衡（绝对标准化差 = 0.14）（补充表 5 和 6）。

表1列出了主要出血和临床相关出血的粗略发生率。依鲁替尼治疗个体的主要出血事件主要是胃肠道出血（47.6%），其次是脑出血（19.0%）、泌尿生殖系统出血（14.3%）和其他出血事件（14.3%）。 ）和未指定（4.8%）（补充表7）。依鲁替尼（与 BR 相比）的大出血风险升高，但在统计学上与零值一致（sIPTW 调整后的 HR：1.61，95% 置信区间 [95% CI]：0.67–3.84），并且具有升高的统计学显着性临床相关出血的危险（sIPTW 调整后的 HR：2.66，95% CI：1.29–5.49）（表 1 和补充图 4）。敏感性分析的结果与我们的主要发现一致；有关敏感性分析和效果修正结果的更多详细信息，请参见补充结果 2.2。

这项基于人群的队列研究比较了接受依鲁替尼治疗的个体与接受化学免疫治疗的个体的真实世界主要出血率和临床相关出血率。与 BR 相比，依鲁替尼发生大出血的风险更高，尽管这一发现与无效结果一致。与 BR 相比，依鲁替尼的临床相关出血风险是 BR 的 2.6 倍，这表明在常规临床实践中接受依鲁替尼治疗的患者存在发生临床相关出血事件的潜在风险。

在我们的研究中，接受依鲁替尼治疗的个体大出血的粗发生率（每 100 人年 3.1 例 [py]，95% CI：2.0–4.7）与依鲁替尼临床试验的荟萃分析中发现的结果相当，¹据报道，合并大出血发生率为每 100 py 2.8 例。尽管之前的现实世界观察性研究没有报告发病率，但其粗发病率（8%–18%）^{3, 4}明显高于我们的研究（0.87%）。补充讨论 3.1 中描述了造成这种差异的潜在原因。

我们调整后的主要结果分析发现，与 BR 相比，依鲁替尼的大出血率有所升高，但与零值一致。由于结果数量较少，我们的研究很可能不够有力；然而，不能排除依鲁替尼和 BR 之间大出血发生率没有差异。使用奥斯汀的方法从调整后的生存模型中获得伤害所需的人数 (NNH)，我们计算出大出血的 NNH 为 252。卡伦等人。和王等人。还发现临床试验中与比较疗法相比，大出血的汇总相对风险增加，范围为 1.66-2.46。^{1, 2}我们调整后的次要结果分析显示，依鲁替尼（相对于 BR）与临床相关出血（即导致住院或急诊就诊的出血）呈正相关。我们计算出临床相关出血的 NNH 为 58。尽管之前的研究没有检查类似定义的结果，但报告了较低严重程度出血的比较率。卡伦等人。和王等人。两者都在临床试验中观察到，所有严重程度的出血的相对风险（汇总相对风险范围为 2.72-3.08）^{1, 2}比大出血更大。阿卜杜勒-卡迪尔等人。进行了一项基于人群的队列研究，观察到与接受化疗的患者相比，接受依鲁替尼治疗的患者与医院诊断的出血有类似的关联（HR 2.58，95% CI 1.76-3.78）。⁶总体而言，这些发现支持当前的建议，即在评估依鲁替尼作为治疗选择时考虑其潜在出血风险。

这项研究是迄今为止规模最大的研究，旨在调查依鲁替尼在现实世界中的大出血情况。一个限制是我们无法获得 CLL 特征（如 Rai 分期）和高风险基因组特征（如 17p 缺失或免疫球蛋白重链基因突变的存在）。然而，我们在倾向评分中纳入了可能影响出血发生的 CLL 并发症（例如血小板减少症）。此外，尽管我们使用了严格的研究设计和调整方法来解决混杂问题，但依鲁替尼和 BR 治疗个体之间的残余差异可能仍然存在。其他优点和局限性详见补充讨论 3.2-3.3。

依鲁替尼是目前美国最常用的 CLL 疗法。我们的研究发现，尽管伊布替尼在现实世界人群中的大出血发生率与临床试验人群相当，但与 BR 相比，临床相关出血率明显增加。随着第二代 BTK 抑制剂的使用增加，临床医生将需要有关治疗类别中各个药物的现实风险的信息。此外，鉴于 BTK 抑制剂通过糖蛋白 VI 抑制的独特抗血小板机制，有必要研究它们对血栓事件的潜在保护作用，特别是对于经历 BTK 抑制剂相关心房颤动的患者。这项研究将有助于作为依鲁替尼相关出血的基准，直到可以对所有可用的 BTK 抑制剂进行此类上市后研究。