Arterio-occlusive events among patients with chronic myeloid leukemia on tyrosine kinase inhibitors

Tyrosine kinase inhibitors (TKIs) are standard therapy for patients with chronic myeloid leukemia (CML). The first clinical trial for imatinib began in 1998¹ and rapidly changed the CML treatment paradigm. Subsequent generations of TKIs have further improved outcomes and contributed to a near-normal life expectancy.² 

TKIs have generally excellent efficacy and a favorable toxicity profile. However, they each have somewhat unique toxicity profiles. In 2008 ponatinib was introduced, demonstrating significant clinical activity among patients with multi-TKI refractory CML or with T315I. The optimism from the high efficacy was soon tamed by the recognition of the occurrence of arterio-occlusive events (AOEs) in many patients.^3,4 Since then, AOEs have been reported with most TKIs, suggesting that AOEs are possibly a class effect. It is a concern that the recognition of the occurrence of these adverse events (AEs) associated with TKIs was underappreciated for many years. Here, we analyze the history of this delayed recognition and discuss possible actions to prevent this from happening with other agents in the future. The focus of this analysis is AOEs, although other cardiopulmonary events not discussed here are also relevant with TKIs including electrocardiographic abnormalities (arrhythmias and corrected QT prolongation), cardiac failure, and pleural effusion.

In the pivotal phase 2 trial of ponatinib for patients with multiresistant CML (PACE), with a median treatment duration of 12.8 months, serious AOEs were reported in 8.9% of patients. An additional 13 months of exposure revealed a cumulative incidence of 17.1%.³ AOEs had not been reported in pivotal studies of any TKI, although a case series of patients treated with nilotinib suggested a possible association.⁵ Most ponatinib studies were temporarily put on hold or terminated³ and ponatinib was temporarily withdrawn from the market.⁶ Patients enrolled in ongoing trials who were benefiting from ponatinib were required to reduce their dose.

Interestingly, the phase 1 study for ponatinib reported no AOEs.⁴ The only cardiac events reported were corrected QT prolongation in 3 patients (4%; grade ≥3 in 2 patients, 2%), and congestive heart failure (CHF), decreased ejection fraction, and cardiomyopathy in 1 patient each (CHF and decreased ejection fraction were grade ≥3). In contrast, the 5-year PACE analysis described AOEs in 25% (serious 20%) of patients with exposure adjusted rates of 13.8 per 100 patient-years (serious, 10.6 per 100 patient-years). These events were nearly equally distributed among cardiovascular, cerebrovascular, and peripheral vascular. Venous thrombotic events were reported in 6% of patients with chronic phase (CP) disease (2.1 per 100 patient-years).⁷ AOEs, venous thromboembolic events, and heart failure are all in a black-box warning in the US label.

Recognition of the association of AOEs with other TKIs came much later. The initial reports for both second-line and frontline therapy for all other TKIs had little or no mention of AOEs. With dasatinib, nilotinib, and bosutinib, no cardiovascular, cerebrovascular, or peripheral arterial toxicities were reported in patients treated after imatinib failure in the initial analysis with ∼2-year follow-up.^8-10 The pivotal frontline studies of these drugs were very discrete in the description of AOEs. The first-year report of ENESTnd only stated “11 patients in all three study groups combined had an ischemic heart disease event” with no further reports of AOEs.¹¹ Additionally, the first-year report of DASISION only mentioned “One death in each group was attributed to the study treatment; both deaths were the result of myocardial infarction”, with no further discussion of AOEs.¹² The one-year report of BFORE reported “peripheral vascular events” in 1.5% and 1.1% of patients taking bosutinib and imatinib, respectively, 1 case of a “cerebrovascular event” in the imatinib group, and cardiovascular events (CVEs) in 3% and 0.4% of patients taking bosutinib and imatinib, respectively. These events were attributed to a history of cardiac disorders.¹³ In subsequent reports, the 2-year follow-up analysis of DASISION did not mention any AOEs.¹⁴ In contrast, the 2-year analysis of ENESTnd included 2% of patients in each of the nilotinib cohorts with an ischemic heart event (<1% in the imatinib arm). Peripheral arterial occlusive disease was described in 6 patients (3 in each nilotinib dose cohort) but in only 1 was it attributed to nilotinib.¹⁵ These peripheral arterial events were curiously not included in the table of safety findings. Finally, in a phase 1/2 study of bosutinib for patients with imatinib-resistant or imatinib-intolerant CML-CP, and in the frontline BELA trial, no AOEs were reported at the 2-year mark.^10,16 

By the 5-year mark of these trials, AOEs had been described with ponatinib, prompting closer attention to these events (Table 1). The final DASISION analysis described no new AEs but reported ischemic events in 5% of patients treated with dasatinib and 2% of patients treated with imatinib.¹⁷ These included a twofold higher incidence of CVEs with dasatinib than with imatinib (4% vs 2%), and 2 transient ischemic events with dasatinib (none with imatinib). Thus, the incidence of these events with dasatinib was more than double than that with imatinib. In the 5-year analysis of ENESTnd, CVEs (including ischemic heart disease, ischemic cerebrovascular events, and/or peripheral artery disease), occurred in 7.5%, 13.4%, and 2.1% of patients treated with nilotinib 300 mg, 400 mg, and imatinib 400 mg, respectively. These included ischemic heart events in 2.7% with nilotinib 300 mg twice daily and 1.5% with imatinib, and cerebrovascular events in 1.1% and 0.3%, respectively. Overall, the incidence of any event with nilotinib 300 mg twice daily (5.7%) was more than triple than that with imatinib (1.8%). CVE rates increased over time and were higher in patients with preexisting risk factors.²¹ With 10-year follow-up, ENESTnd reports CVEs in 16.5%, 23.5%, and 3.6% with nilotinib 300 mg, 400 mg, and imatinib, respectively.¹⁸ Finally, the 5-year BFORE report describes vascular treatment-emergent events in 7.5% and 3.4% of patients with bosutinib and imatinib, respectively. This included CVEs in 4.9% and 0.4%, cerebrovascular events in 0.7% and 1.1%, and peripheral vascular events in 2.2% and 2.3%, with bosutinib and imatinib, respectively.¹⁹ An earlier randomized trial (BELA) of bosutinib vs imatinib, using bosutinib 500 mg daily, with ∼5-year follow-up, reported AOEs in 4.8% and 3.6% of patients for bosutinib and imatinib, respectively. The rate of CVEs was 2.4% and 2.0%, cerebrovascular events in 0.8% and 0.8%, and peripheral vascular events in 1.6% and 0.8% of patients, respectively.²² Remarkably, the incidence of AOEs for imatinib remained constant across these studies. For example, for CVEs, the incidence was 2%-2.5% in all 5-year reports, except for BFORE in which the incidence was only 0.4%. AOEs were thus occurring in all pivotal TKI trials, more frequently with the second generation TKIs (2G-TKIs) than with imatinib.

AOEs are emerging for asciminib, the most recently approved TKI with a greater selectivity for Abelson murine leukemia viral homolog 1 (ABL1). In a phase 1 study of 141 patients, at 14-month median follow-up, 1 patient taking 150 mg experienced a grade 3 acute coronary syndrome.²³ At the 4-year report of this study, the AOE rate increased to 8.7% .²⁴ In the randomized asciminib vs bosutinib trial (ASCEMBL) at median of 14.9 month follow-up, the frequency of AOEs was 3.2% (n = 5) and 1.3% (n = 1) with asciminib and bosutinib, respectively. These include 2 deaths on asciminib: 1 arterial embolism and 1 ischemic stroke.²⁵ At 2.3-year median follow-up, the frequency increased to 5.1% (n = 8) and 1.3% (n = 1), respectively, with exposure adjusted rates approximately double for asciminib (3.0 vs 1.4 per 100 patient treatment years).²⁰ 

These reports suggest that AOEs could be a class effect. Two meta-analyses have concluded that there is increased risk with dasatinib, nilotinib, and ponatinib compared with imatinib, and a statistically nonsignificant trend with bosutinib.^26,27 This begs the question as to why these events were overlooked for so long as an important consequence of TKI therapy.

How and why the risk for AOEs as a TKI class effect was missed for so many years appear to be multifaceted. Culprits include minimal risk with imatinib, the evaluation criteria, the focus on treatment-related AEs, the association with known risk factors, the primary focus on CML response, the insufficient follow-up, and access to data.

Imatinib likely does not increase the risk of AOEs, although there were conflicting reports of cardiac failure.^28,29 This could have predisposed investigators toward expecting a similar favorable safety profile with the new TKIs. A retrospective analysis even suggested a protective effect of imatinib for AOEs.³⁰ This conclusion, based on pooled retrospective analysis of studies with different patient populations, using an interferon-treated cohort as the no-TKI cohort control, and with various reporting criteria among other confounding variables, may have contributed to a decreased alertness to AOEs. Imatinib likely does not increase the AOE risk beyond baseline risk factors but is unlikely to have a protective effect. In the IRIS trial of imatinib vs interferon and cytarabine, AOEs occurred in 19 (3%) patients with imatinib.³¹ Corresponding data for the patients in the control arm were not reported.

One important element in identifying AOEs is the search criteria. There is great variability on what is described in the 5-year reports of pivotal trials. The Medical Dictionary for Regulatory Activities (MedDRA) provides unifying terms organized in 5 hierarchical tiers to report AEs. The Common Terminology Criteria for Adverse Events is meant mostly for grading of AE severity. These criteria include various types of AOEs, such as myocardial infarction (MI), strokes, and thrombosis (a term that includes deep vein thrombosis, pulmonary embolism, and cardiac thrombosis).³² Although there is some overlap with MedDRA terms, there is no full concordance. The 5-year reports from the 3 main frontline randomized trials, ENESTnd, DASISION, and BFORE, used different strategies and terms and, most important, different depth of search. The number of events identified depends on the specific events being searched. Tables 2-4 detail the search strategy used for AOEs in these trials. Clearly, although AOEs are used generically, what is being reported is very different. This is likely to affect the reported incidence. This would affect both the 2G-TKIs and imatinib in each individual study, but it makes the analysis across the various TKIs difficult, particularly in the absence of no randomized studies of 2G-TKIs.

It is common practice to report only events that reach a certain frequency (eg, 10%-20%) in the overall population. This approach may underestimate reporting events with similar mechanisms or significance. For example, if few patients suffered strokes, few suffered MI, and few suffered deep vein thrombosis, if not reaching the reporting threshold, they could be considered insignificant. However, if they were all considered AOEs (ie, using a higher level MedDRA term), the total number of events could draw attention to the problem. A highly fragmented listing of AEs, failing to group individual events representing AOEs, likely contributed to delay in identification of AOEs.

Most AEs occurring with TKIs are more frequent in the early weeks or months. AOEs in contrast may occur at any time during the course of therapy with the incidence steadily increasing over time. In most pivotal trials, initial reports at the time of the primary end point (ie, ∼12 months), the incidence was low enough for investigators to miss the impact of AOEs. In most trials, very low to no incidence of AOEs was reported at the 2-year mark, with more significant numbers only showing up at the 5-year mark. Even with ponatinib, it was only the sharp increase from the initial analysis to the follow-up analysis ∼12 months later that prompted a more thorough assessment.³ Furthermore, the current design of phase 1 trials aimed at determining the maximally tolerated dose is inadequate to identify AEs that occur late in the course of therapy. Also, the use of early surrogates as primary end points (ie, 6-12 months) contributes to the underappreciation of AEs whose incidence increases over several years.

The possible effect of prior exposure to TKIs that could be contributing to the risk is seldom considered. A modeling of AOEs with ponatinib based on clinical data suggested the risk may remain elevated for ∼6 months after discontinuation (20 days for venous thrombotic events).³⁴ This residual effect should be considered, not only when assessing causality but also in clinical practice when changing therapy. It may also be informative to know the residual risk shortly after treatment discontinuation, requiring persistent alertness if it remains elevated. Future TKI cessation studies would therefore benefit from long-term follow-up to fully characterize AOE risk over time after cessation.

It is troublingly common that AEs in clinical trials are reported by causality, frequently only reporting “treatment related” AEs. When initial findings of AOEs were identified, many may have been written off as unrelated. The 2-year PACE report stated: “Serious arterial thrombotic events were observed in 9% of patients; these events were considered to be treatment-related in 3%.”³ Similarly, the final report from BFORE indicated: “the TEAEs resulting in death within 28 days of last dose occurred in three (1.1%) bosutinib- versus four (1.5%) imatinib-treated patients: acute cardiac failure, myocardial ischemia, and renal failure with bosutinib; and pneumonia, sepsis, cerebrovascular accident, and disease progression with imatinib. Only sepsis (in the imatinib arm) was considered related to study drug by the investigator.”¹⁹ With current understanding we could question whether MI and cerebrovascular accident were truly unrelated to treatment.

Assessing causality was confounded by the fact that AOEs occurred predominantly among patients with additional risks factors. Patients with CML frequently have comorbidities that puts them at an increased risk for AOEs. One analysis suggested that 41% of patients had at least 1 comorbidity.³⁵ Another study found patients with CML to have a significantly higher incidence of AOE risk factors such as hypertension, diabetes, and obesity. There was also a significantly higher prevalence of MI, CHF, atherosclerosis, and stroke in patients with CML than in the general population.³⁶ Thus, when AOEs were identified, it was assumed these were likely because of the patient’s risk factors and not related to the TKI. In the 5-year report of ENESTnd, peripheral artery disease was noted. However, it was argued that Framingham general cardiovascular risk scores correlated with increased chance of a CVE,²¹ seemingly suggesting that these events were likely to happen regardless. Comorbidities do increase the risk of AOEs in these patients. In PACE, patients with no additional risk factors had relative risk of AOEs of 0.4 compared with 0.8 for those with 1, and 2.2 for patients with ≥2 risk factors.⁷ Still, TKIs themselves constitute a risk factor, each with its own weight. Attributing CVEs to preexisting risk factors may have been done excessively, possibly contributing to the delayed identification of the increased risk of AOEs with TKIs.

Overreporting can also be misleading. This is not as common a problem, but when it occurs it creates confusion as to the true frequency of AOEs. Januzzi et al³⁷ suggested that many AEs reported in PACE include nonspecific descriptions. An independent committee of experts reviewed all AOEs to determine whether the events met established criteria for defining AOEs. After adjudication, the AOE rate was 17% vs 25% before adjudication. The most commonly nonadjudicated events were angina pectoris, chest pain, and noncardiac chest pain reported as symptoms or presumptive diagnoses, usually with no accompanying changes in management.³⁷ This approach was used blindly and prospectively in OPTIC.³⁸ This approach might be impractical for all AEs, but for instances such as AOEs that carry special weight and that may determine the willingness to prescribe a potentially useful drug for a patient in need, it might be desirable.

As increasingly more effective TKIs have emerged, treatment goals have become more ambitious. Achieving a major or deep molecular response has become the primary end point even in instances of multiple prior treatment failures.²⁰ In some instances, change of therapy is considered even for patients who have a good response but not a deep molecular response. For example, in the ENESTcmr study, patients receiving imatinib who had a complete cytogenetic response but “persistent minimal residual disease” were randomized to continue imatinib or change to nilotinib.³⁹ Switching resulted in a rate of molecular response with 4.5-log reduction at 48 months of 54% vs 32% with imatinib. AOEs were reported in 13% with nilotinib (including 2 deaths: 1 of MI and 1 of cardiopulmonary failure) and 2% with imatinib. Thus, there was a 1.68-fold higher probability of molecular response with 4.5-log reduction switching to nilotinib in exchange for a 6.5-fold higher probability of having an AOE, some fatal.³⁹ The promising results with ponatinib in the phase 1 trial,⁴ at a time when patients with failure of ≥2 TKIs or with T315I had few treatment options, led to the design of PACE that was very permissive of patients with comorbidities and recent ischemic events (but not within 3 months).³ This is perhaps the most permissive trial ever conducted for TKIs in CML; the consequences were dire. Even with the current understanding of these risks, management of comorbidities is not always optimal. In 1 report, less than a third of patients who developed dyslipidemia while on nilotinib were started on statins.⁴⁰ The all-inclusive management of patients’ medical conditions, particularly those with comorbidities, should not be secondary to the achievement of polymerase chain reaction (PCR) goals.⁴¹ 

The largest and most informative studies have been sponsored by the drug manufacturers. In such studies there might be more limited access to full data sets and more limited ability to do additional review and analysis by investigators. The requirement by many journals to make data available to interested parties is a welcome step to minimize this concern.

It is important to note the disservice to patients in ending trials too early. In the final 10-year report of ENESTnd, rates of CVEs continued to increase, nearly doubling, from 5 to 10 years. Projected 5-year CVE rates with nilotinib 300 mg twice a day were 10.6%, and with imatinib 3.2% (Table 1). Ten-year rates approximately doubled to 24.8% and 6.3%, respectively.¹⁸ This highlights the injudiciousness of terminating studies (eg, DASISION, BFORE, and PACE) after only 5 years. If patients can be treated for >5 years, the follow-up of clinical trials, particularly for late occurring AEs, should match the planned duration of administration to accurately estimate risks of treatment over time. Not doing so is a major disservice to patients, most of whom will be treated for far longer than the reported observation period from clinical trials.

The potential mechanisms underlying AOEs with TKIs are multifaceted, complex, and incompletely understood. A full analysis of these is beyond the scope of this manuscript, but some mechanisms that have been implicated include vascular effects such as vasospasms caused by nilotinib^42,43 and microvascular angiopathy with ponatinib.^44,45 Increased platelet activation promoting thrombus growth has also been suggested.⁴⁶ In contrast, some studies have suggested that dasatinib and ponatinib may have an anti-aggregation effect on platelets.^47,48 This could cast doubt on the potential benefit of using aspirin to decrease the risk. However, in the absence of prospective studies, a possible benefit cannot be ruled out. There is also the potential of off-target effects on VEGFR, FGFR, Tie-2, and the SRC family of kinases,⁴⁹ although not all TKIs associated with AOEs have such effects.

ABL inhibition is a possible consideration if this is a class effect. Although no clear antithrombotic function of ABL is apparent, ABL activation has many tissue-specific and context-dependent effects.⁵⁰ There is some correlation between ABL1 inhibitory potency and risk of AOEs among ATP-competitive inhibitors. Notably, asciminib has an 50% inhibitory concentration of 0.61 nM, the lowest 50% inhibitory concentration of all CML TKIs for ABL, although acting in a different, more selective binding site.⁴⁹ This could explain why imatinib has a lower association with AOEs than the other TKIs, and asciminib a higher risk compared with bosutinib. More research is needed to fully understand the association of TKIs with AOEs.

TKIs have made a major impact in controlling CML in most patients and perhaps even eradicating it in some. However, long-term risks of AOEs (and other AEs) are an important consideration that need attention and a comprehensive management of every patient, including comanagement with cardio-oncology. The historic evolution of our understanding of this risk provides lessons in drug development that should be used for the future of the development of new drugs, and for actions on analyzing existing data to help us better understand the extent of the patient’s risk for AOEs and the best approach to manage them. Some possible actions that could be undertaken now include a comprehensive evaluation of all available data with all TKIs, including a wide and uniform range of MedDRA terms that provide a broader view of all AEs in this class, similar to what has been done with bosutinib.²² Doing this with equal design for all TKIs would provide physicians and patients a more accurate expectation of the incidence and characteristics of AOEs with each drug, a valuable set of information when making treatment decisions for patients. Adjudication of all AOEs and symptoms suggestive of AOEs for all pivotal trials could also be conducted retrospectively to properly judge reported events in a similar way as has been done for ponatinib.³⁷ For future studies, this could be done prospectively as was done with OPTIC.⁵¹ Underreporting should be avoided, but overreporting may be also misleading and may contribute to the deprivation of valuable drugs for patients in need. Continued evaluation of long-term safety of drugs, such as TKIs that are meant for long-term and possibly life-long administration, should be mandated as part of the regulatory process. This applies not only to AOEs but to other AEs possible with long-term administration (eg, pleural effusion, second malignancies, etc). Drugs can only be recommended as stable to the length of the observation when this has been documented (ie, “expiration date”). A similar approach should be used for the allowed recommended duration of exposure to drugs and perhaps tied to loss of patent if violated. There are costs associated with long-term studies, but this should not override patient safety. Additionally, proactive dose reductions in patients meeting certain treatment end points should be considered as done in OPTIC. Dose reductions are common in clinical practice and may contribute to a lower rate of AOEs seen in many real-world reports with TKIs.^52-54 There is evidence of correlation between dose and risk. OPTIC showed a greater risk for patients treated with ponatinib 45 mg daily than for those treated at lower doses.³⁸ In ENESTnd, the risk is also higher with nilotinib 400 mg twice daily than with nilotinib 300 mg twice daily.¹⁸ Reporting of AEs “related to the study drug” should be discouraged. Treatment-emergent AEs should be the norm, and should include a comprehensive list of all events, with groupings by categories, including high-level and high-level group MedDRA terms. Providing this in supplemental information, in a predetermined uniform format, would allow all interested readers to review the whole spectrum of AEs. There should also be a requirement to avoid senseless duplications (eg, thrombocytopenia and low platelets as separate events). It would also be desirable that full, raw data on large, multicenter, pivotal clinical trials are made available more widely, during the course of the study and at final analysis, without filters and interpretations by sponsors and medical writers. This is thankfully becoming more common as a requirement for publication, frequently “upon request.” A common repository, with uniform criteria and clear guidelines, could be created for all drugs at the time of regulatory approval, with regular updates that extend as long as the drug is meant to be used for.

TKIs have saved and improved the lives of many patients. The recognition of AOEs as a consequence of treatment with these drugs was a late development that may have affected patients both by putting patients at unrecognized risks and by avoiding the use of potentially valuable drugs because of a risk that was poorly understood and inadequately reported. TKIs remain generally safe drugs for most patients. More is needed to understand the pathophysiology of these events to better prevent and manage them. More is also needed to better design trials and assess the results to more opportunely and comprehensively understand the risks and benefits of each therapy.

Contribution: J.C. designed and outlined the project; and L.V. and J.C. researched background information, wrote and reviewed the manuscript, and approved the final version.

Conflict-of-interest disclosure: J.C. is a consultant for Novartis, Pfizer, Takeda, Incyte, Sun Pharma, and Terns Pharma; has received research support (to the institution) from Novartis, Sun Pharma, and Ascentage. L.V. declares no competing financial interests.

Correspondence: Jorge Cortes, Georgia Cancer Center, Cecil F. Whitaker Jr, GRA Eminent Scholar Chair in Cancer, 1410 Laney Walker Rd, CN2222, Augusta, GA 30912; email: jorge.cortes@augusta.edu.