Oncotherapeutic Protein Kinase Inhibitors Associated With Pro-Arrhythmic Liability

Background Ibrutinib is a protein kinase inhibitor that has been widely successful in treating multiple common variations of B-cell cancers. However, an unfortunate side effect of ibrutinib is that it predisposes patients to development of atrial fibrillation. Objectives The purpose of this study was to assess other commonly prescribed protein kinase inhibitors for similar pro-arrhythmic liability. Methods This study comprehensively evaluated data from the U.S. Food and Drug Administration adverse events reporting system and determined the reporting of cardiac arrhythmia attributed to kinase inhibitor therapy using a multivariable logistic regression model. We evaluated 3,663,300 case reports containing 23,067 cases of atrial fibrillation and 66,262 cases of cardiac arrhythmia. In total, 32 protein kinase inhibitors were evaluated, almost all of which are oncotherapeutics. Results Seven protein kinase inhibitors were associated with a significant increase in the odds of atrial fibrillation (ibrutinib, ponatinib, nilotinib, ribociclib, trametinib, osimertinib, and idelalisib). Assessment of broader pro-arrhythmic toxicity suggested a ventricular-specific liability for nilotinib and a bradyarrhythmia risk with alectinib and crizotinib. Conclusions Compounds that result in the inhibition of a number of protein kinases are associated with an increased risk of cardiac rhythm disturbances. The mechanisms driving the arrhythmogenic effects remain to be discovered, but this study presents an important step in identifying and prioritizing the study of these protein kinase signaling pathways.

kinase inhibitor has been associated with increased risk of several cardiovascular toxicities, and a metaanalysis of 20 studies found the pooled rate of atrial fibrillation (AF) for patients treated with ibrutinib to be 3.3 per 100 person-years (range, 2.5. to 4.1), compared to 0.84 per 100 person-years (0.32 to 1.6) for non-ibrutinib therapy (3)(4)(5)(6). The therapeutic target of ibrutinib is Bruton's tyrosine kinase (BTK), which is irreversibly inhibited by the drug. Bruton's tyrosine kinase is essential for activation of several pathways necessary for chronic lymphocytic leukemia cell survival, including the Akt, ERK, and NF-kB pathways (7). Ibrutinib also inhibits a number of other tyrosine kinases, albeit with lower affinities. A recent study singled-out the off-target inhibition of the tyrosine-protein kinase CSK as a mechanism by which Ibrutinib can cause AF (8).
To assess the extent of proarrhythmic liability within the pharmaceutical class of PKIs, we investi- to Chi-squared test to quantify the association between drug and adverse event (5). Here, we make use of a standard probabilistic machine learning algorithm, logistic regression, to quantify the propensity to cardiac rhythm disorders as a side effect of protein kinase inhibition, while also accounting for con- and those based on multivariable modeling (11). The disproportionality analysis methods use a univariable approach that focuses on the relative proportions of patients experiencing the adverse event. This analysis is predominantly based on 2-by-2 contingency tables, evaluating the effect of treatment from summary statistics allowing for analysis without using patient-level information. This approach is simple and computationally efficient but focuses solely on differences in proportions and cannot account for possible confounding effects in a data set. In contrast, multivariable models analyze multiple variables in parallel, with logistic regression being the most widely used. Logistic regression is a machine learning algorithm, which is preferable due to its ability to quantify variable effects without any assumptions of the underlying data distribution. The regression first applies a nonlinear transformation followed by a 2class regression. Because the model is flexible and uses a probabilistic approach, it has multiple usages.
In this case, the relative effect of variables (treatment with a specific PKI) to separate patients experiencing the specific adverse event from the other reports was quantified. The primary advantage of this method is the ability to account for confounding variables associated with patient characteristics (e.g., age, sex,  (A to C) Summary statistics for age and sex of U.S. Food and Drug Administration Adverse Event Reporting System (FAERS) population and subpopulations. The age distribution for the FAERS population (A) and the subpopulation that reported atrial fibrillation (AF) (B) both tend to be elderly. n indicates number of cases in each population and m is the population mean. The age distribution for the AF population is statistically significantly different from the background FAERS population (p < 0.001, Mann-Whitney U test; ***p < 0.001, Student's t-test). (C) Summary statistics for sex distributions for subpopulations affected by AF or any etiology of cardiac arrhythmia differ significantly from the background FAERS population (***p < 0.001, Chi-squared test). (D and E) Analysis of comorbidities associated with AF reporting. (D) System organ class (SOC) level terms were analyzed by Chi-squared disproportionality testing to identify terms that co-segregated with elevated AF reporting. Significantly enriched terms, included as confounding variables in the final analysis, are shown in red. (E) AF reporting odds ratio (ROR) for the SOC level term "cardiac disorders" was large compared to the other terms and was analyzed for division at the high level group term (HLGT) stratum. "Cardiac arrhythmias" is significantly enriched (shown in red) over other cardiac disorders within AF reporting. RORs and confidence intervals for all terms are presented with Bonferroni corrections for multiple testing, and marker size is the proportional total number of reports containing the term.
Ye et al.  Logistic regression analysis of AF in FAERS assessed the confounding effect size of comorbidities associated with increased incidence of disease, on the patient's likelihood of developing AF (dark red). After controlling for these confounding variables, 7 protein kinase inhibitors have significantly increased likelihood of being reported for AF (red). These are: ibrutinib, ponatinib, nilotinib, ribociclib, trametinib, osimertinib, and idelalisib. The 95% confidence intervals (CIs) are skewed due to being in exponential space, and are not corrected for multiple hypothesis testing. Abbreviations as in Figure 1.   We hypothesized that PKIs may predispose to arrhythmias beyond AF. We discovered that for the 7 PKIs significantly associated with AF, many other diverse arrhythmia types were also disproportionately reported (Supplemental Appendix and Supplemental Figure 6). To test this comprehensively, we analyzed whether there was an association between PKIs and cardiac arrhythmia reporting in general. As the MedDRA classification reports) that assessed the likelihood of developing arrhythmia. After controlling for confounding variables, 9 protein kinase inhibitors have significantly increased likelihood of being reported for arrhythmia, including ibrutinib, ponatinib, nilotinib, ribociclib, trametinib, osimertinib, and idelalisib for atrial fibrillation specifically. The 95% confidence intervals are skewed due to being in exponential space, and are not corrected for multiple hypothesis testing.  We identified 7 PKIs that increase reporting odds for AF, 2 that increase odds for bradycardia, and 1 associated with increased risk for ventricular arrhythmia.
Three of these PKIs have been previously reported as having an association with increased AF risk, whereas another recent report also finds an additional AF association with nilotinib (2,4,5,19). We present novel associations with AF for ribociclib (previously associated with prolonged QT), trametinib, and idelalisib (20). Additionally, we present associations of alecti- HLT-grouped arrhythmias. We hypothesize that this is related to other disease etiologies being more localized and developing more acutely (e.g., sinus bradycardia or extrasystoles), and thus less agedependent than AF which requires broader tissue remodeling to create a proarrhythmic substrate capable of supporting a re-entrant arrhythmia.
EFFECTS OF COMORBIDITIES. Comorbidity effect sizes in the logistic regression models were largely similar to those in the Chi-squared analysis. Minor differences are likely attributable to reports with multiple comorbidities which are assessed independently in the Chi-squared analysis but simultaneously by logistic regression. In the results presented in