Predictors of Gastrointestinal Bleeding Among Patients with Atrial Fibrillation After Initiating Dabigatran Therapy

Abstract

Study Objectives

To identify demographic and clinical risk factors associated with gastrointestinal (GI) bleeding among a large cohort of patients with atrial fibrillation (AF) who initiated dabigatran therapy for stroke prevention, and to describe patterns of subsequent anticoagulant use after occurrence of the GI bleeding event.

Design

Retrospective cohort study.

Data Sources

Large, nationwide United States commercial insurance database.

Patients

A total of 21,033 patients with nonvalvular AF who initiated dabigatran between October 19, 2010, and December 31, 2012.

Measurements and Main Results

We used multivariate Cox regression analysis to estimate the effect of baseline demographic and clinical characteristics on the probability of a GI bleeding event. Patterns of anticoagulation use after GI bleeding were also examined descriptively. Of the 21,033 patients receiving dabigatran, 446 (2.1%) experienced a GI bleed during follow-up. GI bleeding rates differed across many baseline characteristics. Male sex was associated with a lower risk (adjusted hazard ratio [aHR] 0.78, 95% confidence interval [CI] 0.64–0.95) of GI bleeding. Compared with patients younger than 55 years, those aged 55–64, 65–74, and 75 years or older yielded aHRs of 1.54 (95% CI 0.89–2.68), 2.72 (95% CI 1.59–4.65), and 4.52 (95% CI 2.68–7.64), respectively. Renal impairment (aHR 1.67, 95% CI 1.24–2.25), heart failure (aHR 1.25, 95% CI 1.01–1.56), alcohol abuse (aHR 2.57, 95%CI 1.52–4.35), previous Helicobacter pylori infection (aHR 4.75, 95% CI 1.93–11.68), antiplatelet therapy (aHR 1.49, 95% CI 1.19–1.88) and digoxin use (aHR 1.49, 95% CI 1.19–1.88) were also associated with an increased GI bleeding risk. Of the 446 patients who experienced a GI bleed, 193 (43.3%) restarted an anticoagulant, with most (65.8%) filling prescriptions for dabigatran; the mean time was 50.4 days until restarting any subsequent anticoagulant.

Conclusion

The risk of GI bleeding in patients receiving dabigatran is highly associated with increased age and cardiovascular, renal, and other comorbidities, even after adjusting for other factors. Less than 50% of patients restarted an anticoagulant after experiencing a GI bleed. Clinicians should continue to monitor for these risk factors or consider whether alternative therapies may be appropriate.

More than 460,000 individuals in the United States are newly diagnosed with atrial fibrillation (AF) annually.¹ Thought to affect 3 million Americans, AF is only increasing in prevalence and incidence as the population ages.^1,2 Moreover, the average rate of ischemic stroke among patients with nonvalvular AF is approximately 5% per year, which is thought to be 2–7 times that of the general United States population.³

Use of anticoagulation in patients with AF is recommended to prevent thromboembolic stroke and systemic embolism.⁴ However, use of anticoagulants can lead to an increased risk of bleeding, and, in particular, gastrointestinal (GI) bleeding.^5,6 GI bleeding alone is significantly associated with morbidity and mortality; mortality from GI bleeding has ranged from 1–13%, depending on the location in the GI tract.^7–9 Until recently, warfarin was the only U.S. Food and Drug Administration (FDA)-approved anticoagulant for the prevention of stroke in patients with AF; in October 2010, approval of dabigatran for this indication provided an additional option.^4,10–12 Despite demonstrated advantages over warfarin regarding risk of stroke, systemic embolism, and intracranial bleeding events, dabigatran has been shown to lead to an increased risk of GI bleeding in the major clinical trial used for FDA approval and recent observational studies in clinical practice, with overall rates of GI bleeding ranging from 1–6%.^13–17

Although knowledge about the increased risk of GI bleeding in patients receiving anticoagulants has expanded, the literature on the risk factors associated with GI bleeding in these patients has been restricted to therapy settings where alternatives to warfarin were not available.⁵ Little is known regarding demographic and clinical characteristics that lead to higher risks of GI bleeding among patients receiving dabigatran, especially among patients in real-world practice. The ability to better understand patients who are at increased risk of GI bleeding could help clinicians to better identify appropriate patients for dabigatran use, especially given its increased risk of GI bleeding above and beyond warfarin. In addition, among patients who do experience a GI bleed, restarting anticoagulation may still be necessary to prevent stroke after recovering from the bleed, particularly among patients at high stroke risk.^5,18 However, little is known regarding anticoagulant treatment patterns after a GI bleed in dabigatran users in clinical practice, and particularly whether these patients persist on any anticoagulation therapy.

Therefore, the goal of this study was to identify demographic and clinical risk factors associated with GI bleeding among a large cohort of patients with AF who initiated dabigatran therapy for stroke prevention, and to describe patterns of subsequent anticoagulant use after the occurrence of the GI bleeding event.

Methods

Study Design, Setting, and Patient Population

This was a retrospective cohort study using the Truven Health MarketScan Commercial Claims and Encounters and Medicare supplement databases (Truven Health Analytics, Ann Arbor, MI) for the years 2009–2012. These data include an encrypted unique patient identifier for patient-level medical inpatient and outpatient claims, physician office visits, outpatient pharmaceutical data, and enrollment data for approximately 40 million individuals from over 100 nationwide insurers annually. Prescription medication use was identified through National Drug Codes in the outpatient prescription files.

A cohort of patients with AF was selected by using the following inclusion criteria: filled at least one prescription for dabigatran after October 19, 2010 (FDA approval date for dabigatran), referred to as the “index prescription”; were at least 18 years of age at index prescription fill date; had documentation of at least one inpatient or two outpatient International Classification of Diseases, Ninth Revision (ICD-9) codes for AF (AF; ICD-9 code 427.31), occurring on separate days within 12 months before the index prescription fill date; and were continuously enrolled in the insurance plan for at least 12 months prior to the index prescription fill date to ensure complete information capture. One of the outpatient ICD-9 codes for AF could occur after the index prescription fill date, but the two ICD-9 codes had to be on separate days to lower the possibility of the codes being used as a rule-out condition. Patients were excluded if they had any anticoagulant prescription filled in the 12 months prior to the index prescription fill, including warfarin, dabigatran, or rivaroxaban. Finally, patients with ICD-9 codes relevant to valvular AF, transient AF, or dabigatran contraindications, including end-stage renal disease, during the baseline period were excluded to ensure that only patients initiating dabigatran as chronic therapy for nonvalvular AF were included (supplemental Appendix Table 1 available online at ??????????).

The University of North Carolina at Chapel Hill Institutional Review Board reviewed this study, and it received exempt approval status.

Outcome Measurements

Clinically significant GI bleeding was assessed in the follow-up period after dabigatran initiation by using ICD-9 diagnosis codes in the inpatient claims. Coding algorithms for GI bleeding (supplemental Appendix Table 2 available online at ??????????) were based on published studies with reported positive predictive values ranging from 86–89% in the literature.^14,19,20 Follow-up for a GI bleeding event began on the day after initiating dabigatran and was continued until a loss of continuous eligibility, occurrence of the first documented GI bleeding event, a prescription was filled for a different anticoagulant, or the study period ended (December 31, 2012), whichever came first.

Among patients who experienced a GI bleeding event after initiating dabigatran, subsequent anticoagulant patterns were examined, including the proportion of patients who restarted any anticoagulant and the time until the subsequent anticoagulant prescription fill. For this exploratory analysis, patients were followed from the time of the first documented GI bleeding event until loss of continuous eligibility, the end of the study period (December 31, 2012), or the patient filled another available anticoagulant, including dabigatran, warfarin, or rivaroxaban, another newly approved anticoagulant.²¹

Baseline Characteristics

The patients’ baseline demographic and clinical characteristics and medication use were evaluated during the 12-month baseline period. Demographic characteristics included age, sex, and region of residence (northeast, north central, south, west), as these have been thought to be associated with either GI bleeding or quality of care received in anticoagulation,^5,22,23 and they were measured at the time of dabigatran initiation. Patient comorbidities and clinical characteristics known to be associated with GI bleeding in patients taking warfarin were also identified by using ICD-9 codes in the outpatient or inpatient medical claims (supplemental Appendix Table 2).^5,24 These comorbidities included ischemic stroke, venous thromboembolism, congestive heart failure, hyperlipidemia, hypertension, coronary artery disease, anemia, peripheral vascular disease, renal impairment, diabetes mellitus, major bleeding, peptic ulcer disease, alcohol abuse, and diagnosed Helicobacter pylori infection. Concomitant prescription medications were also evaluated, including antiplatelets, antiarrhythmics, rate-control therapies (e.g., digoxin, beta-blockers, calcium channel blockers), angiotensin-converting enzyme inhibitors and angiotensin receptor blockers (ACEI/ARBs), statins, and GI protective agents, as well as nonsteroidal anti-inflammatory drugs (NSAIDs), prescription aspirin, and corticosteroids which are thought to be possibly associated with overall or GI bleeding. A full list of the specific medications queried is available in supplemental Appendix Table 3 (available online at ??????????). The dose of dabigatran that was first initiated (75 mg or 150 mg) was also identified and examined as a risk factor, as the higher dose may lead to increased GI bleeding risk. Ischemic stroke risk (using the CHA₂DS₂-VASc [heart failure/ejection fraction, hypertension, advanced age, diabetes, stroke/TIA, vascular disease score, female sex] scoring system) and bleeding risk (using the ATRIA [anticoagulation and risk factors in atrial fibrillation] and HAS-BLED [hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/alcohol] scoring systems) were also measured in the 12 months prior to dabigatran initiation by using ICD-9 codes in the inpatient and outpatient claims files and were presented descriptively to avoid colinearity with other risk factors.^25–27

Statistical Analysis

The distributions of sociodemographic and clinical characteristics were described among patients between those who did and those did not experience a GI bleeding event after initiating dabigatran. Absolute standardized differences, indexes that measure the effect size between groups to assess imbalance, were also provided for each characteristic to illustrate any differences in the baseline characteristics between groups; a significant difference in the proportions of patients with that characteristic is usually characterized by an absolute standardized difference > 10 (approximately equivalent to p<0.05).²⁸ Each factor was then tested separately to estimate the impact of its overall effect on the rate of GI bleeding, without controlling for the other covariates, by using Cox proportional hazards regression analysis. Multivariable Cox proportional hazards regression analysis was also used to investigate the independent association between each factor and GI bleeding in patients receiving dabigatran by adjusting for other factors and covariates. Finally, the frequency of patients who filled any anticoagulant after the index GI bleeding event was examined and reported. In the event that an anticoagulant was restarted, the type of anticoagulant was described as well as the mean (± SD) time until the subsequent anticoagulant fill. All analyses were performed by using SAS statistical software, version 9.3 (SAS Institute, Inc., Cary, NC). Statistical significance was determined by using two-sided tests with an α of 0.05.

Results

A total of 21,033 patients with nonvalvular AF were prescribed dabigatran for stroke prevention and met the inclusion criteria, providing a mean time of 336.0 days of follow-up. Of these patients, 446 (2.1%) experienced a clinically significant GI bleeding event after dabigatran initiation. In this cohort, 3,172 (15.1%), 2,926 (13.9%), and 14,489 (68.9%) were censored due to loss of continuous eligibility, filling a different anticoagulant (e.g., warfarin or rivaroxaban), and administrative censoring on December 31, 2012, respectively.

Baseline characteristics are provided in Table 1, stratified by occurrence of subsequent GI bleeding. The mean ± SD age of the cohort was 67.5 ± 12.4 years. The proportion of patients experiencing a GI bleed increased with age, with 58.7% of those who experienced a GI bleeding event aged 75 years or older. As indicated by the absolute standardized differences, those experiencing a GI bleeding event were more likely to be female and had higher rates of almost all measured baseline clinical comorbidities, except for hyperlipidemia. Baseline medication use was also higher in patients experiencing a GI bleed, with the exception of NSAIDs, whereas the proportion of patients having previously filled an NSAID was similar between groups. The proportion of patients filling dabigatran 75mg was also higher in the group that experienced a GI bleeding event compared with those who did not. In addition, of the patients with renal insufficiency, 38 (65.5%) of the patients who had a GI bleeding event used dabigatran 150 mg compared with 755 (67.7%) of those who did not have a GI bleed.

Table 1.

Baseline Demographic and Clinical Characteristics of the 21,033 Patients with Atrial Fibrillation Receiving Dabigatran by Occurrence of Subsequent Gastrointestinal Bleeding

Characteristic	No GI Bleeding (n=20,587)	GI Bleeding (n=446)	Absolute Standardized
Demographic
Age (yrs)
< 55	2,946 (14.3)	16 (3.6)	42.4
55–64	6,374 (31.0)	64 (14.4)	46.8
65–74	4,717 (22.9)	104 (23.3)	1.2
≥ 75	6,550 (31.8)	262 (58.7)	70.2
Male sex	13,104 (63.7)	238 (53.4)	23.3
Region of residenceb
Northeast	3,439 (16.7)	72 (16.1)	2.1
North Central	5,958 (28.9)	139 (31.2)	6.3
South	7,692 (37.4)	151 (33.9)	8.8
West	3,173 (15.4)	82 (18.4)	10.8
Clinical
Ischemic Stroke	1,439 (7.0)	48 (10.8)	20.0
Congestive Heart Failure	3,694 (17.9)	137 (30.7)	42.1
Venous thromboembolism	510 (2.5)	20 (4.5)	17.7
Hyperlipidemia	10,225 (49.7)	212 (47.5)	5.1
Hypertension	14,206 (69.0)	339 (76.0)	17.0
Coronary artery disease	5,717 (27.8)	200 (44.8)	46.3
Peripheral vascular disease	1,092 (5.3)	51 (11.4)	36.6
Renal impairment	1,115 (5.4)	58 (13.0)	44.9
Diabetes mellitus	5,431 (26.4)	153 (34.3)	22.2
Bleeding	1,906 (9.3)	69 (15.5)	28.2
Anemia	2,132 (10.4)	85 (19.1)	37.2
Peptic Ulcer disease	87 (0.4)	4 (0.9)	11.2
Alcohol abuse	403 (2.0)	15 (3.4)	13.9
Helicobacter pylori infection	38 (0.2)	5 (1.1)	28.3
CHA2DS2-VASc score	2.7 ± 1.7	3.8 ± 1.7
ATRIA score	1.9 ± 1.9	3.1 ± 2.2
HAS-BLED score	1.2 ± 0.9	1.5 ± 1.0
Drug therapy
Antiplatelet agent	2,810 (13.7)	127 (28.5)	54.7
Antiarrhythmic agent	5,216 (25.3)	116 (26.0)	2.0
Digoxin	2,874 (14.0)	93 (20.9)	25.8
β-Blocker	13,785 (67.0)	316 (70.9)	9.2
Calcium channel blocker	8,383 (40.7)	201 (45.1)	10.6
ACEI or ARB	11,566 (56.2)	301 (67.5)	26.3
Statin	10,898 (52.9)	285 (63.9)	25.5
Corticosteroid	4,930 (24.0)	130 (29.2)	15.3
NSAID	4,140 (20.1)	88 (19.7)	1.3
GI protective agent	2,199 (10.7)	61 (13.7)	12.9
Dabigatran 150mg	18,830 (91.5)	385 (86.3)	17.1

Data are no. (%) of patients or mean ± SD values.

GI, Gastrointestinal; CHA₂DS₂-VASc = scoring system for ischemic stroke risk using heart failure/ejection fraction, hypertension, advanced age, diabetes, stroke/TIA, vascular disease score, female sex as risk factors; ATRIA = anticoagulation and risk factors in atrial fibrillation scoring system for bleeding risk; HAS-BLED = scoring system for bleeding risk using hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/alcohol as risk factors; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; NSAID, nonsteroidal antiinflammatory drug.

^aA significant difference between the proportions of patients with each characteristic is usually characterized by an absolute standardized difference > 10.

^b327 patients were classified as “other” region.

The results from the unadjusted and full multivariable Cox regression analyses are shown in Table 2, providing estimates of the effects of the baseline demographic and clinical characteristics on the risk of GI bleeding events in patients taking dabigatran. In these analyses, proportional hazards assumptions were also tested and did not reveal any violations of the assumption for the models.

Table 2.

Unadjusted and Multivariable Associations Between Patient Characteristics and Gastrointestinal Bleeding in Patients Receiving Dabigatran

Baseline Characteristic	Unadjusted HR (95% CI)	Adjusted HRa (95% CI)
Demographic
Age (yrs) (reference: < 55 yrs)
55–64	1.91 (1.11–3.31)*	1.54 (0.89–2.68)
65–74	3.95 (2.34–6.69)**	2.72 (1.59–4.65)**
≥ 75	7.07 (4.27–11.71)**	4.52 (2.68–7.64)**
Male sex (reference: female)	0.65 (0.54–0.78)**	0.78 (0.64–0.95)*
Region (reference: northeast)
North Central	1.20 (0.91–1.59)	1.12 (0.85–1.50)
South	0.96 (0.73–1.27)	0.98 (0.74–1.29)
West	1.30 (0.95–1.78)	1.18 (0.86–1.62)
Clinical (reference: none)
Ischemic Stroke	1.58 (1.17–2.13)*	0.96 (0.71–1.30)
Congestive Heart Failure	2.09 (1.71–2.56)**	1.25 (1.01–1.56)*
Venous thromboembolism	1.85 (1.18–2.89)*	1.14 (0.72–1.80)
Hyperlipidemia	0.96 (0.79–1.15)	0.88 (0.72–1.07)
Hypertension	1.47 (1.18–1.82)**	1.07 (0.85–1.35)
Coronary artery disease	2.09 (1.74–2.52)**	1.37 (1.10–1.69)*
Peripheral vascular disease	2.27 (1.70–3.04)**	1.28 (0.94–1.73)
Renal impairment	2.72 (2.06–3.58)**	1.67 (1.24–2.25)**
Diabetes	1.47 (1.21–1.79)*	1.21 (0.98–1.48)
Bleeding	1.76 (1.36–2.28)**	1.32 (1.01–1.72)*
Anemia	2.11 (1.67–2.67)**	1.25 (0.97–1.62)
Peptic Ulcer disease	2.11 (0.79–5.64)	1.59 (0.59–4.28)
Alcohol abuse	1.77 (1.06–2.95)*	2.57 (1.52–4.35)**
Helicobacter pylori infection	6.32 (2.62–15.26)**	4.75 (1.93–11.68)*
Medication use (reference: none)
Antiplatelet agent (including prescription aspirin)	2.40 (1.95–2.94)**	1.49 (1.19–1.88)**
Antiarrhythmic agent	1.04 (0.84–1.29)	1.10 (0.89–1.37)
Digoxin	1.60 (1.27–2.01)**	1.33 (1.05–1.68)*
β-blocker	1.22 (0.99–1.50)	0.97 (0.78–1.19)
Calcium channel blocker	1.20 (0.99–1.45)	0.97 (0.80–1.18)
ACEI or ARB	1.60 (1.32–1.95)**	1.23 (0.99–1.51)
Statin	1.53 (1.26–1.85)**	1.08 (0.87–1.34)
Corticosteroid	1.35 (1.10–1.66)*	1.17 (0.95–1.45)
NSAID	0.99 (0.79–1.25)	1.04 (0.82–1.31)
GI protective agent	1.34 (1.02–1.76)*	1.02 (0.78–1.35)
Dabigatran 150mg (reference: 75mg)	0.54 (0.41–0.70)**	1.14 (0.86–1.53)

HR, Hazard Ratio; CI = confidence interval; GI, Gastrointestinal; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; NSAID, nonsteroidal antiinflammatory drug

^aAdjusted HRs are the reported estimates from the multivariable Cox proportional hazards regression analysis.

^*p<0.05;

^**p<0.001

In the regression analyses, the unadjusted associations indicated similar associations as seen using the absolute standardized differences. Increased age, female sex, and most measured baseline comorbidities were associated with an increased risk of GI bleeding, with the exceptions being hyperlipidemia and peptic ulcer disease. Baseline use of antiplatelets, digoxin, ACEI/ARBs, statins, GI protective agents, and oral corticosteroids were associated with a higher GI bleeding risk. The dabigatran 150-mg dose was associated with lower GI bleeding risk before multivariable adjustment. After multivariate adjustment, all of the estimates for the risk factors moved closer to the null. However, age 65–74 years and age ≥75 years were still strongly associated with increased risks of GI bleeding with adjusted hazard ratios and 95% confidence intervals (CIs) of 2.72 (1.59–4.65) and 4.52 (2.68–7.64), respectively. Previous congestive heart failure, coronary artery disease, renal impairment, bleeding event, alcohol abuse, H. pylori infection, antiplatelet use, and digoxin use were also still found to be associated with the risk of a GI bleeding event after multivariable adjustment. Possibly due to wide 95% CIs resulting from a relatively small number of outcomes, other estimates still trended toward an association in some groups but were no longer significant in the multivariable analysis.

Finally, among the 446 patients receiving dabigatran who experienced a GI bleeding event, 193 (43.3%) restarted a subsequent oral anticoagulant throughout the follow-up period (Table 3). More patients filled prescriptions for dabigatran (127 [28.5%]) than for either warfarin (48 [10.8%]) or rivaroxaban (18 [94.0%]). Of those who did restart an oral anticoagulant after a GI bleed, the mean time until the subsequent anticoagulant fill date was 50.4 days. Of the 446 patients with a GI bleeding event, 12 patients (2.7%) initiated either heparin or a low-molecular-weight heparin (LMWH) after the GI bleed, and 7 (1.6%) patients used prescription aspirin. Fifteen (7.8%) patients who restarted an oral anticoagulant also used a heparin, a LMWH, or prescription aspirin in the follow-up period.

Table 3.

Subsequent Oral Anticoagulant Use Among Patients Experiencing a Gastrointestinal Bleeding Event While Receiving Dabigatran

Anticoagulant	No. (%) of Patients Who Restarted an Oral Anticoagulant (n=446)	Time until Restarting an Oral Anticoagulant (mean ± SD days)
Any oral anticoagulant	193 (43.3)	50.4 ± 51.6
Warfarin	48 (10.8)	48.8 ± 57.2
Dabigatran	127 (28.4)	48.4 ± 48.4
Rivaroxaban	18 (4.0)	68.5 ± 56.7

Discussion

In this study, we examined factors associated with GI bleeding in a large cohort of 21,033 patients nationwide with nonvalvular AF after initiating dabigatran therapy. Of the 446 patients who experienced a GI bleeding event before switching to another anticoagulant or being censored, we found a consistent increased risk of GI bleeding among older adults, especially in those who were older than 65 years, female, had renal impairment, had congestive heart failure, and were diagnosed with alcohol abuse and a previous H. pylori infection. We also found that previous antiplatelet therapy use, including clopidogrel, prasugrel, ticagrelor, and prescription aspirin agents, and digoxin therapy were also still associated in multivariable analyses. Finally, of those who experienced a GI bleeding event, we found that less than 50% of patients restarted a subsequent anticoagulant; however, of those who did, most restarted therapy with dabigatran.

Much is known about predictors of bleeding among warfarin users, but little is known about factors associated with GI bleeding among patients initiating dabigatran in real-world practice. For warfarin, the risk of GI bleeding has been thought to increase with older age, female sex, history of bleeding, H. pylori infection, diabetes, congestive heart failure, alcohol abuse, and antiplatelet use.^5,24,29,30 Our study corroborates that these factors are also independently associated with GI bleeding among users of dabigatran, with the strongest associations being 65 years of age or older, and the GI bleeding risk while using dabigatran may be elevated on the basis of these factors alone. Previous studies that have examined rates of GI bleeding among patients prescribed dabigatran, both in randomized controlled trials and observational studies, have found overall rates of GI bleeding ranging from 1–6%, similar to our study (2.1%), which have indeed been shown to be higher than those usingwarfarin.^13–17 This higher-than-expected risk suggests that other factors are also playing a role in increasing GI bleeding rates in patients using dabigatran.

In addition to finding a similar risk with factors already thought to increase GI bleeding risk in patients using warfarin, our study also found a 67% increased risk among patients with chronic kidney disease, a 37% increased risk among patients with coronary artery disease, and a 33% increased risk among patients who had used digoxin. Due to its predominantly renal excretion, and the potential issues of using dabigatran in patients with renal impairment have been relatively well publicized, we also note that this risk factor appears to be unique to dabigatran compared with warfarin from previous studies.^31,32 We also did not see any major differences in the proportion of patients with renal insufficiency who used the higher dabigatran dose (150mg) who did or did not experience GI bleeding events, suggesting that disproportionate renal dosage adjustments between the two groups cannot fully explain this risk. The increased risk seen in patients with coronary artery disease could be related to the increased risk we also saw in patients who used antiplatelets, which are not recommended for coadministration with dabigatran.³³ In addition, digoxin and dabigatran administered together have been shown to lead to a small, non–clinically significant increase in dabigatran plasma concentrations.³⁴ This study suggests that more exploration into this potential drug interaction could be warranted because the association between an increased risk of GI bleeding and baseline use of digoxin persisted even after adjusting for other risk factors.

Choosing an appropriate anticoagulant in patients at high risk for GI bleeding can be difficult. Our data indicate that when selecting an anticoagulant, careful consideration should be given to patients who are ≥65 years of age, have renal insufficiency, are female, have congestive heart failure, and are using concomitant antiplatelet therapies. Although the focus of this study was to inform risk factors associated with dabigatran GI bleeding, these findings may also have implications for other more recently available novel oral anticoagulants that either were not available during the time of the study or had too few patients to examine risk factors. Evidence from randomized controlled trials examining rivaroxaban and edoxaban have also indicated a potentially increased risk of GI bleeding compared with warfarin; thus, automatic substitution of alternative novel oral anticoagulants may not completely mitigate concerns of GI bleeding.^21,35,36 However, a randomized trial of apixaban did not suggest an increased risk of GI bleeding.³⁷ Although, to our knowledge, our study is the first to use a large database of nationally representative, commercially insured patients, including some Medicare beneficiaries, to specifically examine factors associated with GI bleeding in users of dabigatran, more research is needed to continue to examine this issue for all novel anticoagulants. However, apixaban may be an alternative in patients with other risk factors for GI bleeding, such as renal insufficiency or older age, in whom dabigatran may not be an ideal choice, but vigilance is recommended and further apixaban research is needed. Nonpharmacologic stroke prophylaxis, such as left atrial appendage occlusion, may also be an option.³⁸

Moreover, we found that less than 50% of patients who did experience a clinically significant GI bleed restarted an anticoagulant. Previous evidence has suggested that continued administration of anticoagulants is associated with a very high risk of continued or subsequent bleeding, but it is often unclear how long a drug should be discontinued or whether one should be restarted, especially in patients using a novel oral anticoagulant.^18,39 The decision to restart anticoagulation is usually based on a balance of risk of ischemic stroke or systemic embolism versus risk of bleeding, but evidence has suggested that restarting warfarin after major GI bleeding is associated with improved health outcomes, even when restarting within 7 days.¹⁸ Among those who restarted anticoagulation in our study, the mean time from GI bleeding event to restarting was approximately 50 days, which indicates that a number of patients could be experiencing extended periods without prophylaxis, especially as less than 3% of patients filled either a heparin or LMWH prescription in the outpatient setting; however, they could also be using over-the-counter aspirin. The mean CHA₂DS₂-VASc score of those who experienced a GI bleeding event was 3.8, indicating that many patients were at high risk of ischemic stroke. More investigation into appropriate anticoagulation use after a clinically significant GI bleeding event is warranted, particularly as alternatives to warfarin are now available.

This study has several limitations. First, this is an observational study, and some residual confounding is possible because of unmeasured or inadequately measured confounders, even though we examined many baseline demographic and clinical characteristics. For instance, information about race/ethnicity was not available within the database, and the degree of renal impairment could not be reliably ascertained.^40–42 In addition, refill records from claims databases may not fully reflect medication use, as patients may not take medications despite filling and also may fill prescriptions outside of their pharmacy benefit.^43,44 However, refill records are a widely accepted way to assess medication exposure and have been shown to have good validity, correlation, and similar sensitivity and specificity with other measures, including self-report, pill counts, and electronic records.^45,46 Because over-the-counter medication use data were not available within the claims database, we could not measure nonprescription aspirin or NSAID drug use either at baseline or at the restart of anticoagulant therapy. However, we measured prescription drug use of these agents as well as a number of other clinical factors thought to be associated with GI bleeding. In addition, despite a large sample of over 20,000 dabigatran users, some estimates had wide 95% CIs due to a small number of outcomes and low prevalence of the conditions; nonsignificance of the results, particularly in multivariable adjustment, should not necessarily be interpreted as the factors being unassociated with GI bleeding risk in dabigatran users.

Conclusion

In this study of patients who initiated dabigatran between 2010 and 2012, we found that older age, female sex, renal impairment, congestive heart failure, diagnosed alcohol abuse, previous H. pylori infection, prescription antiplatelet therapy, and digoxin use were associated with increased risks of GI bleeding. Clinicians should be vigilant about monitoring for GI bleeding among patients initiating dabigatran who have these potentially high-risk conditions, and alternative therapies should possibly be considered. Our evidence also suggests the need for clinicians to be attentive in approaching anticoagulation after a GI bleeding event, as reinitiation rates with anticoagulation appeared to be relatively low.