aPatients were excluded if they had previous intraocular bleeding, end-stage renal disease, a kidney transplant, mitral valve disease, or a heart valve repair or replacement. ACS indicates acute coronary syndrome; MI, myocardial infarction.
eTable 1. Intraocular Bleeding ICD-9 Codes
eTable 2. Univariate Analysis of Association Between Parameters and Intraocular Bleeding for Warfarin vs Dabigatran/Rivaroxaban Comparison
eTable 3. Univariate Analysis of Association Between Parameters and Intraocular Bleeding for Clopidogrel vs Prasugrel Comparison
Customize your JAMA Network experience by selecting one or more topics from the list below.
Uyhazi KE, Miano T, Pan W, VanderBeek BL. Association of Novel Oral Antithrombotics With the Risk of Intraocular Bleeding. JAMA Ophthalmol. 2018;136(2):122–130. doi:10.1001/jamaophthalmol.2017.5677
Are novel anticoagulant and antiplatelet agents better, the same, or worse than traditional agents with regard to risk of intraocular hemorrhages?
In a cohort study using a large national insurance claims database, the novel anticoagulant agents dabigatran etexilate and rivaroxaban had a 25% decreased hazard of intraocular hemorrhages compared with warfarin sodium. No difference in the hazard of intraocular hemorrhages was identified for the antiplatelet agent prasugrel hydrochloride compared with clopidogrel bisulfate.
These data suggest that novel antithrombotic agents are, at worst, equal in the risk of intraocular hemorrhages and in some instances are safer than their traditional counterparts.
Novel oral anticoagulation and antiplatelet therapies have become a mainstay of treatment for thromboembolic disease. However, the safety profile of these medications has not been completely characterized.
To determine the risk of developing intraocular hemorrhages with novel oral antithrombotic therapy compared with that of traditional antithrombotic agents.
Design, Setting, and Participants
In this retrospective cohort study, a large national insurance claims database was used to generate 2 parallel analyses. All patients with incident use of dabigatran etexilate or rivaroxaban between January 1, 2010, and September 30, 2015, were compared with patients with incident use of warfarin sodium. Similarly, patients with new use of prasugrel hydrochloride were compared with those with new use of clopidogrel bisulfate. Both analyses required the patient to be in the insurance plan for at least 24 months prior to initiation of therapy and excluded patients with any previous diagnosis of intraocular hemorrhages or any prescription for the comparator medications. Furthermore, the antiplatelet analysis required a diagnosis of acute coronary syndrome or a myocardial infarction within 60 days of initiation of pharmacologic therapy. The anticoagulant analysis excluded patients with end-stage renal disease, renal transplants, and those with heart valve disease.
Main Outcomes and Measures
Incident intraocular hemorrhages at 90 and 365 days. Multivariate Cox proportional hazards regression models were used to compare the hazard ratio (HR) of developing an intraocular hemorrhage in individuals taking novel agents compared with those taking traditional medications.
A total of 146 137 patients taking warfarin (76 714 women and 69 423 men; mean [SD] age, 69.8 [11.8] years) were compared with 64 291 patients taking dabigatran or rivaroxaban (31 576 women and 32 715 men; mean [SD] age, 67.6 [11.7] years). Cox proportional hazards regression revealed a decreased hazard for developing an intraocular hemorrhage with dabigatran or rivaroxaban at 365 days (HR, 0.75; 95% CI, 0.58-0.97; P = .03), but not at 90 days (HR, 0.73; 95% CI, 0.22-2.63; P = .13). A total of 103 796 patients taking clopidogrel (37 578 women and 66 218 men; mean [SD] age, 68.0 [11.3] years) were compared with 8386 patients taking prasugrel (1988 women and 6380 men; mean [SD] age, 61.0 [9.6] years) and no increased hazard for developing an intraocular hemorrhage with prasugrel was seen at 90 days (HR, 0.75; 95% CI, 0.29-1.92; P = .55) or 365 days (HR, 1.19; 95% CI, 0.69-2.04; P = .53).
Conclusions and Relevance
These results suggest a decreased risk of intraocular hemorrhage associated with novel direct thrombin inhibitors and direct factor Xa inhibitors, but no difference for P2Y12 inhibitors compared with traditional vitamin K anticoagulation and antiplatelet therapy, respectively.
Millions of patients worldwide are treated with oral anticoagulation and antiplatelet therapy to reduce the risk of thromboembolic events. Although traditional anticoagulation therapy has been with the vitamin K antagonist warfarin sodium, several newer agents have become increasingly popular. The direct thrombin inhibitor dabigatran etexilate and the direct factor Xa inhibitors rivaroxaban and apixaban have been shown to be noninferior to warfarin in the prevention of stroke in patients with atrial fibrillation.1-3 Unlike warfarin, which requires frequent blood tests and dose titration, these novel anticoagulants do not require routine monitoring. Owing to their predictable pharmacokinetics, shorter half-life, reduced drug-drug interactions, and overall ease of use, more patients are being started on these and similar agents in lieu of traditional vitamin K–based therapies.4-6
Similarly, several novel antiplatelet agents are now available for patients who have had acute coronary syndrome or percutaneous coronary intervention. In lieu of traditional dual antiplatelet therapy with aspirin and clopidogrel bisulfate, the newer oral P2Y12 inhibitors prasugrel hydrochloride and ticagrelor are increasingly prescribed owing to improved efficacy, bioavailability, and onset of action compared with that of clopidogrel.7
However, numerous reports of systemic hemorrhagic complications have brought the safety and adverse effect profiles of novel medications into question.7,8 Intraocular hemorrhages are a rare but potentially vision-threatening complication of systemic antithrombotic use. Previous studies have shown an increased risk for all types of intraocular hemorrhages in patients taking aspirin, clopidogrel, or warfarin, raising concerns over the safety of these medications.9-16
An analysis using the World Health Organization’s database of adverse events suggests that the odds of having a retinal or vitreous hemorrhage while taking dabigatran or rivaroxaban far exceeds that of warfarin.17 However, the results of this study are based strictly on voluntary physician report and do not quantify the number of patients using the medication. Contrasting these findings is a meta-analysis that found no increased risk of intraocular bleeding in patients taking novel anticoagulants vs those taking vitamin K antagonists, but this analysis was not geared toward finding eye complications.18
Even less is known about the ocular risks of novel antiplatelet agents, as, to our knowledge, no studies have assessed intraocular hemorrhages in patients taking the newer oral P2Y12 inhibitors. Given the increasing number of patients using novel antithrombotics, the safety profile of these medications with regard to eye disease needs to be better understood. The goal of this study was to assess the risk of intraocular bleeding in patients taking novel oral antithrombotic agents compared with that of traditional antithrombotic agents.
Data were abstracted from the Clinformatics Data Mart Database (OptumInsight, Eden Prairie, MN), which contains the deidentified medical claims of all beneficiaries from a large private insurance network in the United States. The database includes all outpatient medical claims (office visits and associated diagnoses), all outpatient pharmaceutical prescriptions filled, and demographic data for each beneficiary during his or her enrollment in the insurance plan. The subset of data available for this study included all patients in the database from January 1, 2010, to September 30, 2015. The University of Pennsylvania institutional review board deemed this study exempt from review owing to the deidentified nature of the data.
For this study, the risk of intraocular bleeding was tested through comparing multiple cohorts of patients. To best assess the potential incremental risk of the novel medications, we compared them with the older medication that the novel drug intended to replace. Two separate analyses were performed. The index date for all comparison groups was the day the patient first filled a prescription. For all cohorts, patients were required to have been enrolled in the insurance plan for 24 months or more prior to the index date without a previous prescription for any of the study medications. Patients were also excluded if they had any previous diagnosis of intraocular hemorrhages or received a prescription for the comparator study medication prior to the index date.
The first analysis examined patients older than 40 years who had received at least 1 prescription for 1 of 2 P2Y12 receptor antagonists of interest: clopidogrel (older drug) or prasugrel (novel drug) during their time in the insurance plan. In addition, specific to the antiplatelet analysis, patients were required to have a history of acute coronary syndrome or a myocardial infarction within 60 days of initiation of therapy to ensure the medications were used for similar indications. Other novel antiplatelet agents (ie, ticagrelor and cangrelor) were not included owing to their limited use within the database. Patients receiving concomitant anticoagulation were either excluded from the analysis or censored at the initiation of anticoagulation. A sensitivity analysis was also performed with removal of patients with a history of ischemic stroke and transient ischemic attacks (TIAs).
The next analysis compared patients older than 40 years who filled at least 1 prescription for either rivaroxaban or dabigatran with patients who filled at least 1 prescription for warfarin. Other novel anticoagulants (ie, apixaban and edoxaban tosylate) were not tested owing to their limited use within the data set. Patients receiving concomitant P2Y12 receptor antagonists (ie, clopidogrel and prasugrel) were not excluded from the analysis; instead, concomitant exposure to P2Y12 receptor antagonists was measured and controlled for in the analysis. The novel anticoagulants studied (dabigatran and rivaroxaban) are cleared by renal elimination and are contraindicated in patients with end stage renal disease. Accordingly, patients with end stage renal disease or undergoing dialysis were excluded from the anticoagulant cohort. In addition, because the novel anticoagulants are approved only for the treatment of nonvalvular atrial fibrillation, patients with heart valve disease or valve replacement were also excluded. These latter exclusions were not applied to the antiplatelet analysis.
For all study cohorts, follow-up began on the date of filling the first prescription and continued until 1 of the following: occurrence of intraocular bleeding, a prescription was filled for a comparator group (ie, alternate antithrombotic), other bleeding (eg, gastrointestinal bleeding or hemorrhagic stroke), the patient’s exit from the insurance plan, the end of the observation period, or a gap of more than 14 days in prescription coverage.
The primary outcome of interest was evidence of new intraocular hemorrhages, defined as having an incident diagnosis code for a new vitreous hemorrhage, nontraumatic intraocular hemorrhage, hyphema, retinal hemorrhage, or expulsive choroidal hemorrhage made by an eye care professional. Subconjunctival hemorrhages were not included. (See eTable 1 in the Supplement for all International Classification of Diseases, Ninth Revision, codes used in this study.)
Confounding by indication is any association between a drug and an outcome that is owing to reasons underlying why the drug is actually prescribed, rather than a direct effect of the drug itself, and can limit these types of studies. Although using older antithrombotics as a comparator group reduces this confounding, additional covariates were evaluated for potential inclusion in multivariate analysis. These covariates included demographic information (age, sex, and race), year of initiation of treatment, indications for treatment (acute venous thrombosis, pulmonary embolism, atrial fibrillation, or atrial flutter), comorbid conditions (hypertension, types 1 and 2 diabetes, stroke, myocardial infarction, congestive heart failure, chronic liver disease, chronic pulmonary disease, peripheral vascular disease, and any malignant neoplasm) and eye disease states that are associated with bleeding (diabetic retinopathy, age-related macular degeneration, sickle cell anemia, or retinal vein occlusions). In addition, other classes of medications are known to potentiate or inhibit the effectiveness of anticoagulants and antiplatelet drugs, including selective serotonin reuptake inhibitors, statins, prescription nonsteroidal anti-inflammatory drugs (NSAIDs), and amiodarone. Other direct modulators were grouped into cytochrome P450 inhibitors (antifungals, antibiotics, fenofibrate, and calcium channel blockers) and cytochrome P450 inducers (corticosteroids, antiepileptics, rifampin, and leukotriene receptor antagonists). To assess the potential interaction between these medications and the study drugs at the time of a possible outcome, patients were considered to be users of these medications only if they had an active prescription at the time of outcome or censoring.
Baseline demographic characteristics were assessed at the index date and were analyzed using descriptive statistics. Means and ranges were used for continuous variables and percentages were used for categorical variables. Cox proportional hazards regression models were used to analyze the time from prescription index date to either an outcome of interest that occurred or when the patient was censored. Hazard ratios (HRs) were estimated for 2 observation periods: days 1 to 90 after the index date and days 1 to 365 after the index date. All covariates were first assessed in a Cox univariate analysis. Each covariate with an association (defined as P < .20) with intraocular hemorrhages was then included for multivariate analysis. Backward variable selection was used for fitting the final multivariate models for each observation window in each analysis. Each of the multivariate models was assessed for multicolinearity; none was found. Furthermore, the proportional hazard assumption was tested for all variables using log-log plots and only prescription NSAIDs violated this assumption, which was not included in the multivariate analysis. All statistical analysis was performed with SAS, version 9.4 (SAS Institute Inc), software was used for all statistical analysis. P < .05 was considered significant.
A total of 146 137 patients prescribed warfarin and 64 291 patients prescribed dabigatran or rivaroxaban met the inclusion criteria (Figure). Approximately half of the patients were female in each cohort (76 714 [52.5%] in the warfarin cohort and 31 576 [49.1%] in the dabigatran or rivaroxaban cohort) and most patients were white (110 639 [75.7%] in the warfarin cohort and 50 523 [78.6%] in the dabigatran or rivaroxaban cohort) (Table 1). The mean (SD) age was 69.8 (11.8) years in the warfarin cohort and 67.6 (11.7) years in the dabigatran or rivaroxaban cohort. (See Table 1 for complete baseline demographic information.) The mean time to a censoring event (in the 365-day periods) was 124 days in the warfarin cohort and 193 days in the dabigatran or rivaroxaban cohort. The warfarin cohort had 81 incident intraocular hemorrhages at 90 days and 203 incident intraocular hemorrhages at 365 days. The novel therapeutic cohort had 33 incident intraocular hemorrhages at 90 days and 92 incident intraocular hemorrhages at 365 days. The longitudinal prevalence rate of an intraocular hemorrhage was 0.14% (205 of 146 137) among patients treated with traditional anticoagulants and was 0.14% (92 of 64 291) among patients treated with novel anticoagulants.
A total of 103 796 patients prescribed clopidogrel and 8386 patients prescribed prasugrel met the inclusion criteria (Figure). A minority of patients in both cohorts were female (37 578 [36.2%] in the clopidogrel cohort and 1988 [23.7%] in the prasugrel cohort) and most were white (76 842 [74.0%] in the clopidogrel cohort and 6398 [76.3%] in the prasugrel cohort) (Table 1). The mean (SD) age was 68.0 (11.7) years in the clopidogrel cohort and 61.0 (9.6) years in the prasugrel cohort. The mean time to a censoring event was 115 days in the clopidogrel cohort and 180 days in the prasugrel cohort. There were 68 new bleeding events in the traditional clopidogrel cohort at 90 days and 134 new bleeding events at 365 days. The novel antiplatelet cohort had 5 new bleeding events at 90 days and 16 new bleeding events at 365 days. The longitudinal prevalence rate for an intraocular hemorrhage was 0.13% (134 of 103 796) in patients treated with traditional agents and was 0.19% (16 of 8386) in patients treated with novel antiplatelet agents.
Cox univariate analysis revealed that, at 90 days, dabigatran and rivaroxaban were not associated with intraocular hemorrhages compared with warfarin (HR, 0.69; 95% CI, 0.46-1.04; P = .07). However, at 365 days, dabigatran and rivaroxaban were associated with a 32% decreased hazard (HR, 0.68; 95% CI, 0.53-0.87; P = .002). Univariate analysis of prasugrel at both 90 and 365 days showed no association with intraocular hemorrhages compared with clopidogrel (90-day HR, 0.72; 95% CI, 0.29-1.78; P = .47; 365-day HR, 0.98; 95% CI, 0.59-1.65; P = .95). (See eTables 2 and 3 in the Supplement for full univariate analysis results.)
Multivariate analysis again revealed no association for warfarin or dabigatran or rivaroxaban and intraocular hemorrhages in the 90-day period (HR, 0.73; 95% CI, 0.22-2.63; P = .13), but a 25% decreased hazard was seen at 365 days (HR, 0.75; 95% CI, 0.58-0.97; P = .03) (Table 2). No association was also found for developing an intraocular hemorrhage while taking clopidogrel compared with prasugrel in either of the observation windows (90-day HR, 0.75; 95% CI, 0.29-1.92; P = .55; 365-day HR, 1.19; 95% CI, 0.69-2.04; P = .53) (Table 3). A sensitivity analysis that removed all patients with a previous TIA or ischemic stroke from the antiplatelet cohorts found no difference in prasugrel’s hazard for intraocular hemorrhages at either time point (90-day HR, 0.60; 95% CI, 0.19-1.93; P = .39; 365-day HR, 1.13; 95% CI, 0.63-2.02; P = .69). With the exception of retinal vein occlusions in the 90-day antiplatelet comparison, the ocular diseases most associated with bleeding (age-related macular degeneration HRs, 2.48-3.15; P < .001 for all comparisons; retinal vein occlusion HRs, 2.32-5.04; P < .009; and diabetic retinopathy HRs, 1.84-2.96; P < .04) were all found to have increased HRs for intraocular hemorrhages (Tables 2 and 3).
Novel antithrombotic agents are established as safe and effective alternatives to traditional medications for the treatment of atrial fibrillation, thromboembolic disease, and acute coronary syndrome. The direct thrombin inhibitor dabigatran and the direct factor Xa inhibitor rivaroxaban are increasingly being prescribed owing to their ease of use and improved adverse effect profiles.19 Likewise, the oral P2Y12 inhibitors prasugrel and ticagrelor are approved for use in patients with acute coronary syndrome and have demonstrated superior efficacy compared with clopidogrel.20,21
To our knowledge, this study represents the first evaluation of the risk of intraocular bleeding with a novel antiplatelet agent. Our data suggest that prasugrel carries no additional ocular risk compared with its traditional counterpart, clopidogrel. Further inquiry will be needed on ocular safety profiles as more data become available for even newer antiplatelet and anticoagulation agents such as ticagrelor, cangrelor, apixaban, and edoxaban.
This study found a decreased hazard of intraocular hemorrhages at 365 days for dabigatran and rivaroxaban compared with warfarin. Our results, along with the recent meta-analysis of clinical trial data by Sun et al22 contradict the concerns for intraocular bleeding raised by case reports12-16,18 and the World Health Organization adverse events database analysis.17 Information in the World Health Organization adverse events database results is based solely on voluntary reporting. This type of information can be misleading because it is susceptible to reporting bias and lacks the ability to control for confounding factors or to properly quantitate risk since the total number of prescriptions represented is unknown.17
Although our results are similar to those reported in the meta-analysis by Sun et al,22 it is important to differentiate how each study arrived at its conclusions. Our study focuses on real-world use of anticoagulants, whereas the meta-analysis focused on results from clinical trial populations that may differ from our study in terms of baseline bleeding risk and how the drugs are dosed and monitored. We excluded people with a previous history of intraocular hemorrhages (to limit recurrences in those predisposed to intraocular hemorrhages) and controlled for eye diseases most frequently associated with bleeding. Unless they are specifically focused on ocular outcomes, clinical trials (and by extension, the meta-analysis by Sun et al22) are unlikely to exclude patients with previous ocular bleeding or emphasize a history of diabetic retinopathy, retinal vein occlusions, or age-related macular degeneration during randomization. The importance of these risk factors is further highlighted in each of our final models, which show the significant hazard they represent for intraocular bleeding.
Furthermore, the a priori removal of patients predisposed to bleeding from our study also likely explains our lower rate of bleeding (0.14%) in the anticoagulant cohorts compared with that reported by Caldeira et al18 (0.21%-0.33%). Removal of these patients was needed to best equate the cohorts at baseline and allow for a direct assessment of the association of the medications with intraocular hemorrhages, independent of the underlying disease. Although this limits the generalizability of our results, our exceptionally low rate of intraocular hemorrhages throughout the study supports the safety of antithrombotics with regards to the eye in this population.
Other strengths of our study should be noted, including our ability to assess many other nonophthalmic factors that could mitigate or potentiate the effects of anticoagulants that may have been an issue in previous studies. In addition, the cohort design of the study minimizes the potential bias of convenience sampling that frequently occurs in case reports (and, by extension, the World Health Organization report17) when rare outcomes spur publications that do not reflect the true rate seen in a larger population. Last, pharmacy records, not patient reports, determined medication use; thus, recall bias was eliminated.
Owing to the nature of our data and the inability to know if patients temporarily stopped their medication at or around the time of eye surgery, we are unable to comment on the safety profile of these drugs perioperatively. Although recent studies have shown no increased risk of perioperative vitreous hemorrhage with antithrombotic use, balancing the potentially life-threatening risk of stopping these medications and the low risk of hemorrhagic ocular complications is a difficult decision that should be made between the surgeon, the prescribing clinician, and the patient.23-25 Further work should be performed to better inform this decision.
Several limitations need to be considered when reviewing the results of our study. First, the rate of intraocular hemorrhages was exceedingly low, and the power of our study to detect subtle differences between antiplatelet cohorts was limited despite the large number of patients. In another context, however, this is reassuring, given the widespread use of antithrombotic agents. It could even be argued that since the incidence of an intraocular hemorrhage occurring while taking these medications is less than 0.2%, small differences between medication classes may not be clinically significant. Second, not all patients underwent an eye examination while taking the medications studied. Although patients with visually significant bleeding would have presumably been referred for care, it is possible that patients with clinically insignificant intraocular hemorrhages went undiagnosed, suggesting that our incidence rates may be an underestimate. Third, this study used administrative billing data that were unable to be verified with medical record–level data, nor have all the International Classification of Diseases, Ninth Revision codes used been specifically validated. Next, the database consists of claims from 1 insurance network and may not be generalizable to other patient populations.
Last, the database only includes data on prescription NSAIDs. We are unable to account for any over-the-counter use of NSAIDs or other factors that could influence the potency of the antithrombotic medications (ie, the effect of grapefruit juice on warfarin levels). Although this limitation certainly exists, the commonness of use of over-the-counter NSAIDs combined with the extremely low occurrence rate of bleeding found in this study suggest that the overall effect of this issue is likely low.
Owing to the increasing use of novel antithrombotic medications for the treatment of coronary artery disease and atrial fibrillation, the ocular safety profile of these medications must be understood. Our study suggests that there is not an increased risk of intraocular hemorrhage associated with the use of novel antiplatelet therapy, but novel anticoagulants may decrease the hazard of bleeding compared with warfarin.
Corresponding Author: Brian L. VanderBeek, MD, MPH, MSCE, Scheie Eye Institute, Department of Ophthalmology, University of Pennsylvania Perelman School of Medicine, 51 N 39th St, Philadelphia, PA 19104 (firstname.lastname@example.org).
Accepted for Publication: October 26, 2017.
Published Online: December 14, 2017. doi:10.1001/jamaophthalmol.2017.5677
Author Contributions: Dr VanderBeek had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Uyhazi, Miano, VanderBeek.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Uyhazi, VanderBeek.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Uyhazi, Pan, VanderBeek.
Obtained funding: VanderBeek.
Administrative, technical, or material support: VanderBeek.
Study supervision: Miano, VanderBeek.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: This study was supported by the K23 Award 1K23EY025729-01 from the National Institutes of Health and Core Grant for Vision Research 2P30EYEY001583 from the University of Pennsylvania. Additional funding was provided by Research to Prevent Blindness and the Paul and Evanina Mackall Foundation. Funding from each of the above sources was received in the form of block research grants to the Scheie Eye Institute.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Meeting Presentation: This article was presented at the 120th Annual Meeting of the American Academy of Ophthalmology; October 17, 2016; Chicago, Illinois.
Create a personal account or sign in to: