Adverse reaction profiles of hemorrhagic adverse reactions caused by direct oral anticoagulants analyzed using the Food and Drug Administration Adverse Event Reporting System (FAERS) database and the Japanese Adverse Drug Event Report (JADER) database

Direct oral anticoagulants (DOACs) are used in anticoagulant therapy. The purpose of this study was to evaluate the association of DOAC-induced gastrointestinal (GI) and nervous system hemorrhage using the FDA's Adverse Event Reporting System (FAERS) database and the Japanese Adverse Drug Event Report (JADER) database. We identified and analyzed the reports of hemorrhagic reactions between 2004 and 2016 from the FAERS and JADER databases, and calculated the adjusted reported odds ratio (ROR) using the multiple logistic regression method. Additionally, we used the time-to-onset analysis. In the FAERS database, the adjusted ROR of apixaban, rivaroxaban, and dabigatran for GI hemorrhage was 6.79 (5.84-7.91), 19.58 (18.85-20.34), and 14.51 (13.58-15.51), respectively. In the JADER database, the adjusted ROR of apixaban, rivaroxaban, edoxaban, and dabigatran for GI hemorrhage was 11.80 (9.50-14.64), 11.03 (9.18-13.26), 10.17 (6.95-14.88), and 9.85 (7.23-13.42), respectively. We found that the association of GI hemorrhage with DOACs was affected by sex (female). Additionally, 30% of GI hemorrhage was observed after 30 days. Hemorrhagic reactions of both GI and nervous systems were observed in both the spontaneous reporting system databases. We recommend that female patients who experience symptoms related to GI hemorrhage should be closely monitored and advised to adhere to an appropriate care plan. Additionally, our results show that patients should be closely monitored for hemorrhage even after a month.

Renal function is an important factor in DOAC therapy, because each DOAC has varying degrees of renal elimination. The urinary excretion rate of dabigatran, rivaroxaban, apixaban, and edoxaban is reported to be 80% [10], 36% [11], 27% [12], and 50% [13,14], respectively. Moderate to severe renal impairment could increase the risk for hemorrhage due to the accumulation of drugs in the serum, affecting dabigatran the most and apixaban the least [15,16]. Renal impairment is more often observed among elderly patients compared with that in the general population. Because atrial fibrillation is largely a disease of the elderly population, the risk of stroke and hemorrhage with DOAC therapy increases with age. The possible association of DOAC with gastrointestinal (GI) hemorrhage is of interest in elderly patients. Because ischemic strokes and systemic embolisms have greater clinical significance than nonfatal hemorrhage (e.g., GI hemorrhage), higher doses of DOACs (e.g., dabigatran) are more favorable in elderly patients [1]. However, both acute and chronic GI hemorrhages have a negative effect on the patient's quality of life. A limitation of the standard-dose DOACs is an increase in the risk of GI hemorrhage [17][18][19][20]. Moreover, there have been concerns regarding the safety profile of DOACs relating to age and sex; for example, renal functions are affected by sex [21,22].
To monitor the adverse drug reactions, a spontaneous reporting system (SRS) compiles reports of suspected adverse reactions (ARs) either voluntarily reported by patients, clinicians, pharmacists, and other healthcare professionals or mandatorily reported by various pharmaceutical manufacturers. SRS is a valuable tool in post-marketing surveillance that reflects the realities of clinical practice [23]. The US Food and Drug Administration (FDA) has developed the FDA Adverse Event Reporting System (FAERS). The Pharmaceuticals and Medical Devices Agency (PMDA), a regulatory authority in Japan, has developed the Japanese Adverse Drug Event Report (JADER) database.
Several researchers have evaluated hemorrhage risk associated with dabigatran using the FAERS database [24][25][26]. Southworth et al. [24] demonstrated that dabigatran and warfarin were associated with similar hemorrhage rates, which is consistent with the findings of the RE-LY study. McConeghy et al. [25] reported that the reporting rate of hemorrhage related to dabigatran and warfarin was 26% and 32%, respectively. They also reported that intracranial hemorrhage was low with dabigatran, but the rate of GI hemorrhage was increased compared to that with warfarin, both of which are consistent with the findings of RE-LY. We previously demonstrated that GI hemorrhage was significantly increased in patients over the age of 80 years [26]. Despite the insights that these trials provide, the effects of other DOACs on hemorrhagic ARs in a clinical setting are uncertain.
In this study, we evaluated the relationship between DOACs and hemorrhagic ARs using the reporting odds ratio (ROR) adjusted using the multiple logistic regression analysis [26][27][28][29]. Analysis of time-to-onset data has been proposed as a method to detect signals for ARs in SRS. We also analyzed the time-to-onset of hemorrhagic ARs [30][31][32].

Data source
All data from the SRS database were fully anonymized by the regulatory authorities before we used them. Data from January 2004 to December 2016 in the FAERS database are publicly available and can be downloaded from the FDA website (http://www.fda.gov/). The FAERS database permits contributors to register drugs under any name, including a trade name and an abbreviation. The DrugBank database contains information of drugs used globally, including the FDA-approved small molecule drugs; it was utilized as a resource for batch conversion and compilation of drug names [33]. We followed the FDA's recommendation to adopt the most recent CASE number to identify duplicate reports of the same patient from different reporting sources and excluded them from the analysis [34].
Data from April 2004 to November 2016 in the JADER database were extracted from the PMDA website (www.pmda.go.jp). We built a database that integrated data from the FAERS and JADER databases and the DrugBank using FileMaker Pro 13 software (FileMaker, Santa Clara, CA, U.S.A.) following the international safety reporting guidelines (International Council on Harmonization, E2B).

Definition of hemorrhage reactions
The AR definitions used in this study corresponded to those provided by the Medical Dictionary for Regulatory Activities (MedDRA)/Japanese version 19.0 (MedDRA/J, www.pmrj.jp/jmo/php/indexj.php) [35]. To evaluate the effect of DOACs on hemorrhagic reactions, we used a standardized MedDRA inquiry (SMQ) for hemorrhage reactions (SMQ code: 20000039) and the System Organ Class (SOC) for GI disorder, and extracted only reports that met both criteria. The number of selected preferred terms for hemorrhagic reactions, limited by SOC (GI disorder), was 152. Furthermore, to evaluate the nervous system hemorrhage, we used 88 preferred terms that matched the SMQ for hemorrhage reactions (SMQ code: 20000039) and the SOC for the nervous system disorder.

Analysis
To evaluate the effect of age on "hemorrhagic reactions," the reports were stratified into the following age groups: 0-59 years and more than 60 years. According to the definition of the World Health Organization (WHO) of the United Nations, elderly people are those who are aged 65 years or more.
Using established pharmacovigilance indices [23], we evaluated the ROR to establish the effects of DOACs on "hemorrhagic reactions." "Cases" were defined as patients who reported "hemorrhagic reactions," while "non-cases" consisted of patients associated with all other reports. The ROR is the ratio of the odds of reporting ARs versus all other reactions associated with DOACs compared with the reporting odds for all other drugs present in the database. To compare the "cases" and "non-cases," we calculated the RORs as (a:c) / (b:d). The RORs were expressed as point estimates with a 95% confidence interval (CI). The signal was considered positive if the lower limit of 95% CI was > 1 and the reported number was ≥ 2 [36].
The use of ROR allows adjustment using multiple logistic regression analysis and provides the advantage of controlling covariates [37,38]. In this analysis, the results were refined by dedicated correction to detect confounding factors that may be present in the database. We calculated the adjusted ROR to control the covariates using the multiple logistic regression analysis. The report was stratified according to age as follows: 0-59-and ≥ 60-year-old group. To construct a multiple logistic model that coded report year, sex, stratified age group, and drug, the following multiple logistic model was used for analysis: (Y = reporting year, S = sex, A = stratified age group, and D = drug (apixaban, rivaroxaban, edoxaban, and dabigatran)).
The adjusted ROR was calculated using the 0-59-year-old group as the control group. The effectiveness of explanatory variables was evaluated using a stepwise method with a significance level of 0.05 (forward, and backward) [27,28]. Using the likelihood ratio test, the influence of explanatory variables was evaluated. As the difference of -2log likelihood follows chi-square distribution with one degree of freedom, the results with p ≤ 0.05 were considered statistically significant. Data analysis was performed using JMP software version 12.0 (SAS Institute Inc., Cary, NC, USA).
Time-to-onset duration was calculated from the time of a patient's first prescription to the occurrence of hemorrhagic reactions. The records with completed AR occurrence and prescription start date were used for the time-to-onset analysis. It was necessary to consider right truncation when evaluating the time-to-onset of ARs. We determined an analysis period of 365 days after the start of administration. The median duration, quartiles, and Weibull shape parameters (WSPs) were used to evaluate the time-to-onset data. The scale parameter α of Weibull distribution determines the scale of the distribution function. A larger scale value (α) stretches the distribution, whereas a smaller scale value (α) shrinks data distribution. The WSP β of Weibull distribution determines the shape of distribution function. Larger and smaller shape values produce left-and right-skewed curves, respectively. The shape parameter β of Weibull distribution was used to indicate the level of hazard over time without a reference population. When β is equal to 1, the hazard is estimated to be constant over time. If β is greater than 1 and 95% CI of β excluded the value 1, the hazard was considered to increase with time [30,31,39]. The time-to-onset analysis was performed using JMP software version 12.0 (SAS Institute Inc., Cary, NC, USA).

Results
The FAERS database contained 8,867,135 AR reports from January 2004 to December 2016. After excluding duplicates according to the FDA's recommendation, 7,348,357 reports were analyzed. The JADER database contained 430,587 reports from April 2004 to November 2016.

Discussion
In this study, we evaluated the association between DOAC and hemorrhagic reactions using data from the SRS databases. As the crude RORs of hemorrhagic reactions such as GI hemorrhage and nervous system hemorrhage were higher than one in both the SRS databases, our results suggest that DOACs increase the adverse hemorrhagic reactions.
Renal function is affected by age, sex, body weight, clinical condition, and medication. The effect of age on ARs associated with DOAC therapy has been widely reported. We applied the multiple logistic regression analysis to validate the results. The interaction term of age (≥ 60)*dabigatran for GI hemorrhage was significant in the JADER databases. Dabigatran presented high renal excretion rate [21]. Because the renal excretion may be compromised, GI hemorrhage was increased in the elderly patients [21]. Increased adjusted ROR suggested that dabigatran increases ARs with advanced age; this also supports previous observations on GI hemorrhage reactions related to dabigatran [26].
We provided insights into the association between hemorrhagic reactions and sex in the SRS dataset-the results obtained after adjusting the ROR suggested that sex (female) may influence hemorrhagic reactions. Sex-specific differences in creatinine clearance and renal function exist [22]. Creatinine clearance, and presumably renal clearance of drugs, in women tends to be approximately 85% of that in men of the same age and body weight [40]. As women generally have less renal function, the risk of hemorrhagic reaction might be considered higher in females than in males. The interaction of the majority of DOACs and sex (female) in GI hemorrhage was observed in both the FAERS and JADER databases. Although the doses of DOACs are optimized according to the renal function of each patient based on the established guideline and package insert, the sex difference observed in our study should be evaluated further.
After a single 20 mg apixaban administration, the mean Cmax and AUC ∞ were 18% and 15% higher, respectively, in females than in males [41]. As no clinically meaningful age-or sex-related difference in the pharmacokinetics and pharmacodynamics of apixaban has been reported, apixaban is considered safe and well tolerated in both elderly and young subjects of both sexes [41]. Previously, dose adjustment was not required on the basis of body weight, age, or sex alone [41,42]. However, our results suggest that caution is warranted in the presence of unknown additional factors such as renal impairment that could increase GI hemorrhage risk in females.
For rivaroxaban-associated GI hemorrhage, aging (over 60 years) showed no effects in both the FAERS and JADER databases. It has been previously reported that the influence of age and sex on rivaroxaban therapy was small [43]. Neither age nor sex appeared to significantly influence the Emax, time course of inhibition of Factor Xa activity, or prolongation of prothrombin time [43]. Rivaroxaban has been approved for clinical use without dose-adjustment for age or sex alone [43]. Clinical studies have demonstrated that the half-life of rivaroxaban is 7-11 h and 11-13 h for young and elderly patients [44][45][46]. In the J-ROCKET-AF study, 15 mg rivaroxaban instead of 20 mg was used [47], because in Japanese patients, 15 mg rivaroxaban provided exposures comparable to 20 mg dose in Americans [48]. The trial results demonstrated that 15 mg rivaroxaban was non-inferior as compared with warfarin and it lowered intracranial bleeding, suggesting the use of a reduced dose of rivaroxaban (15 mg) for evaluation in Japanese patients with AF [47]. Our study supports the results of these previous reports.
It was difficult to interpret the data of edoxaban, because the number of case reports was small in the FAERS. Edoxaban exposure is affected by the efflux transporter (P-glycoprotein) inhibitors and inducers (amiodarone, quinidine, ketoconazole) [49][50][51]. The effect of concomitantly administered drugs should be investigated in the future.
We also applied time-to-onset analysis to validate the results, which provided novel insights into the time-to-onset of GI hemorrhage, that is, over 30% of GI hemorrhage was observed after 30 days in the real-world data set. We evaluated nervous system hemorrhage as intracranial hemorrhage. It has been reported that Asians are prone to intracranial hemorrhage [52]. For all DOACs, the reporting rates in JADER were higher than those in FAERS. Sex is likely to be a significant factor in DOAC therapy. The interaction terms with DOACs (apixaban, rivaroxaban, and dabigatran) and sex (female) were significant in the FAERS, but not in the JADER.
From the reports of early post-marketing phase vigilance in Japan, severe hemorrhagic ARs in first one month after the use of apixaban, rivaroxaban, and dabigatran were 56.0%, 85.2%, and 56.5%, respectively [53][54][55]. The time-to-onset profile showed that more than 50% of nervous system hemorrhage was observed after 30 days.
Asian patients tended to have lower body weight, lower proportions of prior myocardial infarction, vitamin K antagonist experiences, and concomitant use of gastric antacid drugs, and higher proportion of impaired renal function, prior stroke, nonparoxysmal atrial fibrillation, and antiplatelet medication use [56][57][58]. The differences in safety and efficacy between Asian and Caucasians should be carefully evaluated in the future.
The analysis using SRS such as the FAERS and JADER databases has several notable limitations. The SRS is subject to over-reporting, under-reporting, missing data, exclusion of healthy individuals, lack of a denominator, and the presence of confounding factors [36]. However, reports in the SRS databases could also reflect real-life scenarios. The duration of surveillance period could have been increased to strengthen the data obtained in this study. This might be a future consideration. Nomura et al. reported that there were differences in the reported number of ARs between the FAERS and JADER; however, the number of shared reports between the FAERS and JADER is unknown [59]. It is improbable to evaluate the "true" risk of ARs without information concerning the total number of patients administered DOACs. In general, ROR cannot be used to infer the comparative strength of causality. However, it offers a rough indication of the signal strength that can be used to generate hypotheses to search for unknown potential ARs [60,61]. Patients using any one of the four DOACs (i.e., dabigatran, rivaroxaban, apixaban, and edoxaban) have different risk factors for hemorrhage, and failing to adjust would bias the results. Careful attention must be paid to the interpretation of results. We partially refined the results with a dedicated correction to detect possible confounders present in the database, using multiple logistic regression technique.

Conclusions
We have reviewed hemorrhagic adverse drug reactions from the SRS databases, real-world registries of patients receiving DOACs. The signals of hemorrhagic reactions such as GI hemorrhage and nervous system hemorrhage were observed in both the SRS databases. Despite the limitations inherent to SRS, we demonstrated that the association of GI hemorrhage induced by DOACs was affected by sex (female). To the best of our knowledge, no time-to-onset analysis of hemorrhagic ARs has been performed using the SRS databases. The aim of the time-to-onset analysis was to obtain new information and compare the risks and onset profiles of hemorrhagic ARs for prescription drugs in the real world. We recommend that female patients who experience symptoms related to GI hemorrhage should be closely monitored and advised to adhere to an appropriate care plan. Additionally, our results show that patients should be closely monitored for hemorrhage even after a month. Results of the present study offer practical considerations for the avoidance and management of GI hemorrhage associated with DOACs. These data will be potentially useful to clinicians for improving the management of ARs associated with DOACs.