Data source: IMS Health Life Link Health Plan Claims Database

commercial health plan information from managed care plans and other sources
(such as Medicare and Medicaid) throughout the United States
it is generally representative of the national, commercially insured population in
terms of age and sex.
1) Claims files: inpatient and outpatient diagnoses documented as the ICD-9-CM
codes, procedures as the Current Procedural Terminology (CPT-4) codes or the
Healthcare Procedural Coding System (HCPCS), prescriptions as National Drug Code
(NDC), date of services, and days of the prescription supplied.
2) Enrollment files: patients demographic characteristics (including year of birth and
sex) and monthly medical/pharmacy enrollment indicators.
 Study population
244 872 enrollees with a prescription of warfarin, dabigatran, or rivaroxaban
between 1 October 2010 and 31 March 2012.
Each person were entered into the cohort at the date of his or her first prescription for
any of the three study drugs after 1 October 2010.

Inclusion criteria:
Study sample: 46,163 patients
aged 18 years or older
have continuous medical and pharmacy enrollment in the 4907 dabigatran users
six months before the entry date (the baseline period) 1649 rivaroxaban users
have none of the three drugs prescribed in the baseline
period (new user design) 39 607 warfarin users
have the first prescription before 31 March 2012
have known age and sex
not have a previous bleeding event
 Outcome
Outcome of interest: Gastrointestinal bleeding
ICD-9 codes and CPT codes validated in a recent study
 Exposure
Patients observation ending date (the earliest)
The first date: the last date of (continuous) exposure to the same drug.
The last exposure date was the last day of continuous drug at hand plus 14 days (take into
account the clearance of the drug)

if a patient switched to another drug within the 14 day period of the previous exposure, we considered
the day before the start of the second drug to be the last exposure date.

The second date: the date of the loss of medical or pharmacy enrollment.
The third date: the end date of the study, which was 31 March 2012.
The fourth date: the date before the first date of hospital admission not related to
gastrointestinal bleeding (we did not have prescription information during hospital admission)
The fifth date: the first date of gastrointestinal bleeding. If gastrointestinal bleeding occurred
during hospital admission, (the first date of admission as the bleeding date for patients
admitted for gastrointestinal bleeding)
Independent variable

type of drug exposure (warfarin, dabigatran, or rivaroxaban)

Control variables: Using Propensity scores (means of weighting the observations to account
for potential confounding by the variables)
demographics (age groups, sex, and region)
three binary clinical conditions (having any diagnosis of renal failure, trauma, or
Helicobacter pylori infection)
three binary drug indicators (having any prescription of non-steroidal anti-inflammatory
drugs, proton pump inhibitor, or steroid)
six levels of counts of Clinical Classification Software (CCS) (a tool for clustering patients
diagnoses and procedures into a manageable number of clinically meaningful categories;
the higher the CCS score, the higher the comorbidity)
 Propensity score
Model 1: to predict whether a person used dabigatran relative to warfarin among
dabigatran and warfarin users (n=44 514)
Model 2: to predict use of rivaroxaban relative to warfarin among rivaroxaban and
warfarin users (n=41 256)
Propensity score weighting was used because we did not want to lose the observations
of treated patients (compared with matching) and we wanted one interpretable overall
treatment effect (compared with stratification).

Average treatment effect of the treated weighting were applied because this allows us to
estimate the average effect of treatment on patients who received the treatment; that is,
we compared the hazards of GI bleeding among dabigatran or rivaroxaban users with the
hypothesized situation had they taken warfarin instead of the oral anticoagulant.
The balance in baseline covariates before and after weighting were compared using the
standardized difference.
standardized difference less than 0.1 to be a negligible difference between treatment
groups.
 After we applied average
treatment effect of the treated
weighting, standardized
differences of all available
covariates between dabigatran
and warfarin users and between
rivaroxaban and warfarin users
were reduced to 0.05 or smaller
well balanced
Kaplan-Meier survival curves of having gastrointestinal bleeding stratified by the three different drug groups
 Cox-proportional hazard ratio
Two separate Cox proportional hazard models with propensity score average treatment effect
of the treated weighting were created :
To examine the association between anticoagulant exposure (X) (dabigatran vs. warfarin
and rivaroxaban vs. warfarin) and gastrointestinal bleeding (Y)
Control variables could enter the model as either regression covariates or stratification
factors.
if a variable violated the proportional hazard assumption (P=0.1 level), this variable would
enter the model as a stratification factor.
Three variables reached that level: age groups, CCS categories, and having any use of
NSAIDs in the baseline period stratification factors

Post hoc analysis by stratifying for patients above and below the age of 65 years, using the same
model as for the total sample.
 Sensitivity analysis
1) Evaluate two additional models:
including all variables as regression covariates
including all variables as stratification factors
2) Vary the length of washout period from seven to 30 to 45 days
3) Censor all inpatient records owing to the lack of the prescription information during
hospital admission (to examine whether such exclusion would affect our findings)
4) Include the HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding
history or predisposition, labile international normalized ratio, elderly, drugs/alcohol
concomitantly) bleeding risk score in the model (to control for a patients risk of
bleeding). Given that we did not have laboratory data, we excluded the labile international
normalized ratio from construction of this risk score.
Table 1 | Characteristics of study sample. Values are percentages (numbers) unless stated otherwise

Warfarin users: 85.8%

Dabigatran users: 10.6%
Rivaroxaban users: 3.6%
 Non significant

Post hoc analysis

Adjusted for
Appendix 3: Hazard Ratios of all variables included in the Cox Proportional Hazard Models
(The results of the full model for the total sample.)
Dabigatran (ref: Warfarin) Rivaroxaban (ref: Warfarin)
N= 44,514 N= 41,256
Exposure (Dabigatran/ Rivaroxaban) 1.21 (0.961.53) 0.98 (0.36-2.69)
Female 0.64(0.49 0.85) 0.43(0.25 0.71)
Region of Residence
South Reference Reference
East 1.15 (0.78 1.69) 1.50 (0.85 2.62)
Mid-West 1.15 (0.87 1.52) 1.66 (0.86 3.18)
West 0.89 (0.60 1.32) 0.85 (0.47 1.52)
Co-morbidity and co-medications
Use of PPI 0.87 (0.65 1.16) 1.59(0.87 2.90)
Use of Steroids 1.19 (0.85 1.66) 1.09 (0.58 2.05)
Renal Failure 0.82 (0.46 1.47) 4.10 (1.09 15.4)
Trauma 0.77 (0.53 1.11) 0.66 (0.39 1.11)
H Pylori infection 0.01 (0.00-0.04) 0.05 (0.01 0.41)
Age groups, CCS categories, and NSAID use as stratification factors and others as regression covariates
 Sensitivity analysis
Appendix 4: Hazard Ratios Derived from Cox Proportional Hazard Models with Different Covariates-inclusion
Methods
Dabigatran (ref: Warfarin) Rivaroxaban (ref: Warfarin)
N= 44,514 N= 41,256
Crude HR1 1.20 (0.961.52) 0.95 (0.312.94)
Adjusted HR2 1.20 (0.951.52) 0.95 (0.322.86)
Adjusted HR3 1.21 (0.961.53) 0.98 (0.362.69)
Adjusted HR4 1.29 (1.001.65) 1.06 (0.343.30)

1: No control variables
2: All control variables as regression covariates
3: Age groups, CCS categories, and NSAID use as stratification factors and others as regression covariates
4: All control variables as stratification factors
Appendix 5: Hazard Ratios Derived from Cox Proportional Hazard Models with Different Wash-out Periods
Dabigatran (ref: Warfarin) Rivaroxaban (ref: Warfarin)
N= 44,514 N= 41,256
Wash-out Period: 7 Days
Crude HR1 1.21 (0.941.56) 0.83 (0.223.12)
Adjusted HR2 1.25* (0.971.62) 0.76 (0.242.42)
Wash-out Period: 14 Days
Crude HR1 1.20 (0.961.52) 0.95 (0.312.94)
Adjusted HR2 1.21 (0.961.53) 0.98 (0.362.69)
Wash-out Period: 30 Days
Crude HR1 1.22*(0.991.51) 0.89 (0.332.41)
Adjusted HR2 1.13 (0.911.40) 0.99 (0.402.47)
Wash-out Period: 45 Days
Crude HR1 1.23**(1.011.50) 0.85 (0.352.08)
Adjusted HR2 1.11 (0.911.36) 1.10 (0.472.56)

*: p0.10 **: p0.05

1: No control variables
2: Age groups, CCS categories and NSAID use as stratification factors and others as regression covariates
Appendix 6: Hazard Ratios Derived from Cox Proportional Hazard Models Censoring GI Bleeding from
Hospitalization Records
Dabigatran (ref: Warfarin) Rivaroxaban (ref: Warfarin)
All Sample N= 44,514 N= 41,256
Crude HR1 1.18 (0.911.51) 1.34(0.414.36)
Adjusted HR2 1.19 (0.911.54) 1.36(0.493.75)
Non-Elderly N= 34,038 N= 32,099
Crude HR1 1.26 (0.901.75) 1.71 (0.377.83)
Adjusted HR2 1.26 (0.901.77) 1.52 (0.504.60)
Elderly N= 10,476 N= 9,157
Crude HR1 1.08 (0.731.60) 0.80 (0.125.38)
Adjusted HR2 1.08 (0.731.62) 0.67(0.222.04)
1: No control variables
2: Age groups, CCS categories and NSAID use as stratification factors and others as regression covariates
Appendix 7: Hazard Ratios Derived from Cox Proportional Hazard Models Adding the HASBLED Risk Score

Dabigatran (ref: Warfarin) Rivaroxaban (ref: Warfarin)

All Sample N= 44,514 N= 41,256
Adjusted HR1 1.21 (0.961.53) 0.98 (0.362.67)
Non-Elderly N= 34,038 N= 32,099
Adjusted HR1 1.35 (0.991.84) 1.03 (0.343.17)
Elderly N= 10,476 N= 9,157
Adjusted HR1 1.08 (0.761.54) 0.62 (0.182.09)

1: Age groups, CCS categories and NSAID use as stratification factors and others as regression covariates
 Discussion
A meta-analysis of clinical trials: statistically significant increased risk of gastrointestinal bleeding
with novel anticoagulants, with a relative risk of 1.45 (95% CI 1.07 to 1.97) (Holster IL et al., 2013)
Dabigatran: OR 1.58 (95% CI 1.29 to 1.93)
Rivaroxaban: OR 1.48 (95% CI 1.21 to 1.82)
Older US Medicare patients with NVAF: an increase in gastrointestinal bleeding with dabigatran
compared with warfarin (HR 1.28, 1.14 to 1.44) (Graham DJ et al., 2015)
Major gastrointestinal bleeding events were statistically significantly higher with the 150 mg dose
of dabigatran comparedwith warfarin in the RE-LY trial (RR 1.30, 1.08 to 1.56) (Bytzer P et al.,
2013)
Other observational studies have similarly reported either very low rates of gastrointestinal
bleeding or no difference in rates between warfarin and dabigatran in several European cohorts
(Larsen TB et al., 2014; Sorensen R et al., 2013).
A recent observational from the US that reported no statistically significant difference in real
world rates of bleeding between rivaroxaban and warfarin in patients with NAVF (HR 1.06, 0.71 to
1.64) (Laliberte F et al., 2014)
 Discussion
Several of these European cohort studies reporting no difference in bleeding included the lower
110 mg dose of dabigatran, which is widely used in other countries.
USFDA has not allowed the use of 110 mg dabigatran in the United States, so our study did not
include such users.

Limitations of previous studies

Limited efforts in ensuring baseline balance across different exposure groups (propensity score
matching based on up to three simple variables, or none)
Not adopting a new user design so that previous exposure might affect the observed outcomes
Focusing on prevalent instead of incident outcomes
Basing the classification of exposure group on the first eligible claims and not requiring
continuous exposure
 Strength
A new user design was adopted (to eliminate the residual effect of the previous exposure)
Focusing on the incident cases to remove the effect of prevalent GI bleeding
Use of validated algorithm to identify GI bleeding cases
A wide range of potential confounders identified from the literature were included in
constructing propensity scores.
Baseline balance between different exposure groups was achieved through the propensity score
weighting and evaluated through the standardized difference to minimize confounding.
Sensitivity analyses were performed to evaluate the robustness of the results
 Limitations
Information about patients mortality or about laboratory tests such as prothrombin times
were not accessed
Prescription fill data reflect actual patient usage was assumed.
A low number of events resulting in inadequate statistical power inability to detect a
significant difference between dabigatran and warfarin
Outcome of interest, GI bleeding, may not have been fully captured by the algorithm.
This study might underestimate bleeding associated with novel agents, as patients are more
likely to report and seek treatment for bleeding associated with these agents than for warfarin.
Short observation time might not allow us to explore other important long term bleeding
effects. The length of observation was different across the three drug user groups, with the
shortest observation period for rivaroxaban users.
 Limitations
Limited clinical information on patients excluded from the cohort.
The average treatment effect of the treated weighting results in effect estimates standardized
to different populations (rivaroxaban users or dabigatran users). Given the substantial
differences between the two groups, the effect estimates could not be directly compared.
Unmeasured confounding:
it is difficult to capture aspirin, which is obtained OTC and a significant risk factor for GI
bleeding.
Selection bias whereby health professionals may have channeled patients at risk of GI bleeding
to one class of agents is always possible