Introduction

Chronic myeloid leukemia (CML) is a hematological malignancy derived from the fusion gene BCR-ABL1, which is translated into a continuously activated tyrosine kinase, leading to abnormal proliferation and differentiation of hematopoietic cells [1]. In recent decades, tyrosine kinase inhibitors (TKIs) have induced a paradigm shift in the treatment of CML. The first TKI, imatinib, showed significant improvement in clinical outcomes, including a better cytogenetic response and lower progression rate, compared with the traditional combination of interferon and chemotherapy in patients with CML [2]. Second-generation TKIs (dasatinib, nilotinib, and bosutinib) further demonstrated superior molecular responses and even lower progression rates than imatinib [3-9]. The third-generation TKI ponatinib, although restricted to patients with CML with more advanced disease or T315I mutation [10, 11], has also expanded the treatment options for patients with CML. Despite the different clinical profiles of these TKIs, prolonged use of TKIs is usually required for patients with CML. Concerns regarding their “long-term” adverse events, particularly thromboembolic events associated with newer generations of TKIs, have emerged [12-16].

Many observational studies have suggested that newer generations of TKIs may be significantly more associated with vascular adverse events (VAEs) than imatinib is [16-28]. A retrospective cohort study conducted in Sweden indicated that the incidences of VAEs in nilotinib and dasatinib users were higher than that in imatinib users [22]. Another study including pooled patients with CML from different prospective trials also reported that newer-generation TKIs were associated with an increased incidence of arteriothrombotic adverse events, especially ponatinib [21]. The phase II pivotal trial (EPIC trial) of ponatinib was terminated early because of the high incidence of serious VAEs in ponatinib users [10]. A meta-analysis of 29 studies including 15,706 patients found that nilotinib and ponatinib users had a significantly higher risk of encountering major arterial events than users of other TKIs [23]. However, there have been inconsistent results in dasatinib users, with some studies suggesting an increased risk of VAEs and others suggesting the risk is not increased [8, 21-27, 29].

These inconclusive findings regarding the association between TKIs and VAEs may be due to the small sample sizes and/or the inclusion of patients with a history of VAEs in the abovementioned studies. Furthermore, studies regarding the association between TKIs and VAEs in Asian populations are scarce. It thus remains unclear whether the aforementioned findings, based primarily on clinical trials or studies conducted in Western countries, can be translated into clinical settings in Asia. By using the Taiwan Cancer Registry Database (TCRD) and National Health Insurance Research Database (NHIRD), we aimed to examine the association between the use of TKIs and the risk of VAEs. Potential risk factors for this association were also evaluated to provide more clinical insights.

Materials and Methods Data Sources

We incorporated data from the TCRD and NHIRD for analysis. The TCRD is funded by the Ministry of Health and Welfare in Taiwan and was established in 1979. It is considered to be an informative nationwide database for cancer surveillance, with 98.4% completeness and 91.5% morphological verification regarding a cancer diagnosis according to the data in the database in 2012. Comprehensive data, including patient demographics, cancer diagnosis with International Classification of Diseases for Oncology – 3rd edition (ICD-O-3) codes, treatments, and other cancer-related prognostic factors, are recorded in TCRD [30]. Taiwan's NHIRD contains claims data from all beneficiaries enrolled in the National Health Insurance (NHI) program, which was launched in 1995 and covers more than 99% of the population (23.58 million in 2018) in Taiwan. The NHIRD comprises detailed health care information on demographics and health care utilization, including outpatient visits, hospital admissions, and prescription medications [31]. Data linkage was performed by assigning encrypted and unique identification numbers to insured individuals to generate data from the two databases for further analysis.

Ethical Statement

The identification numbers of the beneficiaries were encrypted to ensure their confidentiality. However, unique identification numbers allowed for interconnections among all database subsets of the NHI program. The protocol of this study was approved by the Research Ethics Committee of National Taiwan University Hospital (registration number 201812119RIND).

Study Population

We conducted a retrospective cohort study to assess the incidence of VAEs between different TKI users. Only patients receiving imatinib, nilotinib, or dasatinib were included in this study, as bosutinib and ponatinib were not reimbursed by Taiwan's NHI during our study period.

Patients aged 20 years old and older with a diagnosis of CML (ICD-O-3 codes: 98753) from January 1, 2008, to December 31, 2016, were identified in the TCRD. Their treatments were retrieved from the NHIRD. We excluded patients not treated with any of the three studied TKIs (imatinib, nilotinib, and dasatinib). Patients with a prior history of VAEs, which was defined as having any diagnosis of VAEs, were also excluded. All patients with CML were categorized into three groups according to their first-line TKI treatment (imatinib, nilotinib, or dasatinib). The cohort entry date was defined as the date of diagnosis of CML, whereas the index date was defined as the date of initiation of TKIs.

Outcomes of Interest

The outcomes of interest were VAEs, which included myocardial infarction, other ischemic heart disease, ischemic stroke, other ischemic cerebrovascular disease, arterial occlusive disease, and venous thromboembolism. The corresponding ICD-9-CM codes and ICD-10-CM codes are listed in supplemental online Table 1. Outcome occurrence was defined as a primary diagnosis of VAEs for at least one hospitalization or at least three outpatient visits. All patients with CML were followed from the index date until the date of any of the following censoring criteria: occurrence of VAEs, switch or discontinuation of TKI treatment, stem cell transplantation, death, or receipt of TKIs for 5 years since initiation.

Table 1. Baseline characteristics of study cohort before propensity score matching Variables Nilotinib (n = 306) Dasatinib (n = 240) Imatinib (n = 565) p value Female (%) 143 (46.7) 90 (37.5) 230 (40.7) .0759 Age at diagnosis, yr Mean (SD) 48.3 (14.4) 46.6 (14.6) 49.0 (16.4) .1411 Age at initiation of TKI, yr Mean (SD) 49.1 (14.4) 47.2 (14.5) 49.6 (16.4) .1431 Interval from CML diagnosis until first prescription of TKI, days Mean (SD) 91.2 (277.6) 57.6 (150.0) 35.8 (130.9) .0002 Modified Charlson comorbidity index Mean (SD) 0.6 (1.1) 0.5 (1.0) 0.6 (1.0) .7356 Comorbidities (%) Dyslipidemia 41 (13.4) 21 (8.8) 62 (11.0) .2262 Diabetes mellitus 40 (13.1) 29 (12.1) 76 (13.5) .8703 Hypertension 61 (19.9) 44 (18.3) 136 (24.1) .1332 Congestive heart failure 5 (1.6) ≤3 12 (2.1) .2044 Peripheral vascular disease 5 (1.6) ≤3 5 (0.9) .3748 Cerebrovascular disease 8 (2.6) 4 (1.7) 18 (3.2) .4744 Dementia ≤3 ≤3 4 (0.7) 1.0000 Chronic pulmonary disease 12 (3.9) 6 (2.5) 32 (5.7) .1193 Rheumatic disease ≤3 ≤3 9 (1.6) .1713 Peptic ulcer disease 30 (9.8) 29 (12.1) 61 (10.8) .6958 Mild liver disease 26 (8.5) 26 (10.8) 44 (7.8) .3696 Diabetes without chronic complication 35 (11.4) 28 (11.7) 76 (13.5) .6267 Diabetes with chronic complication 16 (5.2) 5 (2.1) 13 (2.3) .0347 Renal disease 7 (2.3) 4 (1.7) 15 (2.7) .6960 Comedication (%) Aspirin 23 (7.5) 9 (3.8) 30 (5.3) .1510 Circulation enhancersa 7 (2.3) ≤3 14 (2.5) .3343 Antihemorrhagics 6 (2.0) 4 (1.7) 5 (0.9) .5396 Lipid-lowering agents 29 (9.5) 11 (4.6) 45 (8.0) .9440 Anti-hypertensive agents 52 (17.0) 29 (12.1) 106 (18.8) .0681 Antihyperglycemic agents 26 (8.5) 21 (8.8) 64 (11.3) .3178 Hydroxyurea 27 (8.8) 16 (6.7) 40 (7.1) .5600 Hormone replacement therapy 7 (2.3) ≤3 9 (1.6) .1934 Cox II inhibitors ≤3 ≤3 14 (2.5) .0338 Antipsychotics 11 (3.6) 7 (2.9) 11 (1.9) .3274 a Circulation enhancers: pentoxifylline, nicametate citrate, dihydroergotoxine, piracetam and ginkgo biloba extract Abbreviations: CML, chronic myeloid leukemia; Cox II, cyclooxygenase II; TKI, tyrosine. Covariates

Baseline characteristics, including age, sex, comorbidities, and comedications, were collected 1 year prior to the index date. Comorbidities included dyslipidemia, diabetes mellitus, hypertension, and diseases included in the Charlson comorbidity index (CCI) [32, 33]. The items of any malignancy and metastatic solid tumor were omitted from the CCI to avoid overadjustment. Comedications included antiplatelet agents, anticoagulants, other antithrombotic agents, circulation enhancers (pentoxifylline, nicametate citrate, dihydroergotoxine, piracetam, and ginkgo biloba extract), antihemorrhagics, lipid-lowering agents, antihypertensive agents, antihyperglycemic agents, other traditional CML medications, and medications that may also be associated with VAEs (supplemental online Table 2).

Table 2. Multivariable logistic regression model for VAEs since initiation of TKIs Variables OR (95% CI) p value Age at diagnosis 1.04 (1.01–1.07) <.0001 TKI Imatinib Reference Nilotinib 3.43 (1.53–7.69) .0260 Dasatinib 2.39 (0.92–6.23) .5457 Comorbidities Congestive heart failure 3.59 (0.91–14.14) .0673 Cerebrovascular disease 3.49 (1.19–10.3) .0233 Hypertension 2.01 (0.94–4.32) .0734 Abbreviations: CI, confidence interval; OR, odds ratio; TKI, tyrosine kinase inhibitor. Statistical Analysis

The ANOVA tests were applied to compare continuous variables, and χ2 tests were applied to compare categorical variables across different TKI users. Multivariable logistic regression was conducted to investigate the association of TKI use and other risk factors with the risk of VAEs. Variables with a p value < .05 in the univariable logistic regression analysis were selected and assessed in the multivariable logistic regression analysis. Stepwise selection was used to account for potential linearity in related variables.

To reduce the selection bias inherent to retrospective cohort studies, we conducted propensity score matching to control for potential confounders [34]. Baseline characteristics, including age, sex, comorbidities, and comedications, across different TKI users were put into a logistic regression model to compute a propensity score for each patient. Each nilotinib or dasatinib user was separately matched to n (n ≤ 2) imatinib users for further analyses to create matched cohort 1 (nilotinib vs. imatinib) and matched cohort 2 (dasatinib vs. imatinib). Propensity score matching was performed by using greedy matching with a caliper of 0.25. The positivity assumption was tested by using a Kernel density plot.

Differences in baseline characteristics between nilotinib or dasatinib users and imatinib users were compared using standardized mean differences (SMDs) after propensity score matching. An SMD greater than 0.2 implied an imbalanced distribution between the two compared groups (i.e., nilotinib vs. imatinib or dasatinib vs. imatinib). The crude incidence rate and adjusted incidence rate of VAEs were calculated, with the incidence rate ratio (IRR) obtained by conducting Poisson regression. Survival analyses were conducted with the Cox proportional hazard model, and the cumulative incidence curves were compared to assess the association between TKI use and the risk of VAEs. The proportional hazard ratio assumption was tested visually by using log-log plots.

All analyses were performed using the SAS program (version 9.4, SAS Institute, Cary, NC).

Results Demographics of Patients with CML

There were 1,348 newly diagnosed patients with CML identified between 2008 and 2016, with 67 patients under 20 years of age, 30 patients treated without TKIs, and 140 patients having a history of VAEs. After excluding these patients, 1,111 adult patients with CML were included in our study cohort, with 565, 306, and 240 patients treated with first-line imatinib, nilotinib, and dasatinib, respectively (Fig. 1).

Patient inclusion flow chart.

The mean ages at CML diagnosis were 48.3 (SD, 14.4), 46.6 (14.6), and 49.0 (16.4) years for nilotinib, dasatinib, and imatinib users. The modified Charlson comorbidity index values were comparable across different TKI users. However, the proportions of patients who had some specific comorbid conditions were slightly different. For example, the proportion of dasatinib users (8.8%) that had dyslipidemia was lower than the proportions of nilotinib and imatinib users who had dyslipidemia (Table 1). During the follow-up period, there were 12, 16, and 8 VAE cases in imatinib, nilotinib, and dasatinib users, with crude incidence rates of 7.9, 21.0, and 15.1 per 1,000 person-years, respectively. In the multivariable logistic regression analysis, only nilotinib use, older age, and history of cerebrovascular diseases were significantly associated with an increased risk of VAEs (Table 2).

After propensity score matching, 286 nilotinib users were matched to 500 imatinib users (matched cohort 1), with a mean age at TKI initiation of 49 and 48 years, respectively. In addition, 236 dasatinib users were matched to 455 imatinib users (matched cohort 2), with a mean age at TKI initiation of 47 and 47 years, respectively. The baseline characteristics of the two matched cohorts were comparable, as shown in Table 3. The Kernal density plot indicated the positivity assumption was not violated (supplemental online Figs. 1, 2).

Table 3. Baseline characteristics of study cohort after propensity score matching Variables Nilotinib (n = 286) Imatinib (n = 500) SMD Dasatinib (n = 236) Imatinib (n = 455) SMD Female (%) 131 (45.8) 211 (42.2) −0.07 87 (36.9%) 165 (36.3%) −0.01 Age at diagnosis, yr Mean (SD) 47.7 (14.5) 47.4 (16.0) 0.02 46.5 (14.6) 46.4 (15.7) 0 Age at initiation of TKI, yr Mean (SD) 48.5 (14.5) 48.0 (16.0) 0.03 47.1 (14.5) 47.0 (15.7) 0 Interval from CML diagnosis until first prescription of TKI, days Mean (SD) 92.1 (282.5) 37.4 (138.8) 0.25a 58.1 (151.3) 31.7 (117.6) 0.19 Modified Charlson comorbidity index Mean (SD) 0.5 (1.0) 0.5 (1.0) 0.02 0.5 (0.9) 0.4 (0.8) 0.08 Comorbidities (%) Dyslipidemia 32 (11.2) 50 (10.0) 0.04 20 (8.5) 38 (8.4) 0 Diabetes mellitus 31 (10.8) 50 (10.0) 0.03 28 (11.9) 45 (9.9) 0.06 Hypertension 52 (18.2) 95 (19.0) −0.02 43 (18.2) 87 (19.1) −0.02 Congestive heart failure 4 (1.4) 9 (1.8) −0.03 ≤3 ≤3 0 Peripheral vascular disease 4 (1.4) 5 (1.0) 0.04 ≤3 4 (0.9) 0 Cerebrovascular disease 7 (2.4) 7 (1.4) 0.08 0 ≤3 −0.02 Chronic pulmonary disease 11 (3.8) 24 (4.8) −0.05 6 (2.5) 12 (2.6) −0.01 Rheumatic disease ≤3 ≤3 −0.04 ≤3 4 (0.9) −0.06 Peptic ulcer disease 28 (9.8) 53 (10.6) −0.03 28 (11.9) 47 (10.3) 0.05 Mild liver disease 20 (7.0) 35 (7.0) 0 25 (10.6) 38 (8.4) 0.08 Diabetes without chronic complication 30 (10.5) 50 (10.0) 0.02 28 (11.9) 45 (9.9) 0.01 Diabetes with chronic complication 9 (3.1) 11 (2.2) 0.06 4 (1.7) 7 (1.5) 0.09 Renal disease 7 (2.4) 10 (2.0) 0.03 4 (1.7) 8 (1.8) 0.04 Comedication (%) Aspirin 15 (5.2) 25 (5.0) 0.01 9 (3.8) 18 (4.0) −0.01 Circulation enhancers 5 (1.7) 8 (1.6) 0.01 ≤3 4 (0.9) 0.04 Antihemorrhagics 4 (1.4) 5 (1.0) 0.04 4 (1.7) 5 (1.1) 0.05 Lipid-lowering agents 20 (7.0) 35 (7.0) 0 11 (4.7) 24 (5.3) −0.03 Antihypertensive agents 43 (15.0) 76 (15.2) 0 28 (11.9) 55 (12.1) −0.01 Antihyperglycemic agents 21 (7.3) 38 (7.6) −0.01 21 (8.9) 35 (7.7) 0.04 Hydroxyurea 22 (7.7) 37 (7.4) 0.01 15 (6.4) 26 (5.7) 0.03 Hormone replacement therapy 6 (2.1) 9 (1.8) 0.02 ≤3 ≤3 0 Antipsychotics 8 (2.8) 11 (2.2) 0.04 7 (3.0) 10 (2.2) 0.05