Study design and data source
We conducted a retrospective cohort study using data obtained from electronic medical records of patients admitted to En Chu Kong Hospital, which a >500-bed regional teaching hospital and the main referral hospital in the Sanxia district in Taiwan [13], from January 1, 2014 to December 31, 2019. Unlike administrative data, the electronic medical records contain extensive laboratory data, which can provide a less biased estimate of the association between exposures and outcomes [14].
Study cohort identification
The study cohort included patients aged ≥20 years, who were newly treated with NOACs (dabigatran or rivaroxaban) or warfarin between January 2015 and December 2018 for AF, diagnosed according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes of 427.31 and ICD-10-CM codes of I48.0, I48.2, and I48.91. We defined the cohort entry date as the first prescribed date of the anticoagulant drug. To improve the impact of prevalent user bias, patients with any anticoagulant prescription in the year before the cohort entry date were excluded [15]. To ensure data quality and avoid loss to follow up, we excluded patients with <1 year of records in the year before the cohort entry date and those without any following record after the cohort entry date.
Patients were classified as either NOAC initiators or warfarin initiators, with the two groups further categorized into the exposure and reference groups, respectively. NOACs and warfarin are similar in treatment indications, which helps mitigate the risk of confounding factors such as differences in indication, disease severity, and unmeasured confounding factors in the evaluation of NOACs [16]. We followed each individual of the cohort from the cohort entry date until the first occurrence of the study outcome, treatment discontinuation, switching to a different anticoagulant group, an add-on of other anticoagulants, or the study end date (December 31, 2019) for as-treated analysis, whichever came first. Termination of treatment was defined as a period between two successive prescriptions exceeding 30 days .
Outcome measures
The primary safety outcome was any major bleeding, which was the composite of intracranial, gastrointestinal tract, or other bleeding events, from inpatients or emergency room visits. The primary effectiveness outcome was major thromboembolism, which was defined as arterial thromboembolism (for example, ischemic stroke, transient ischemic attack, and peripheral embolism), myocardial infarction, and venous thromboembolism from inpatients or emergency room visits. Secondary outcomes separately analyzed a number of individual outcomes among the composite outcomes. Definitions of outcomes were described previously [17] and are reported in Online Resource 1.
Covariate measurement
We extracted five classes of confounders according to a literature review, including demographics, laboratory data, occurrence of the study outcomes, comorbidities, and comedications in the year before cohort entry. Specifically, we identified INR, activated partial thromboplastin time (APTT), serum creatinine, aspartate aminotransferase, and alanine aminotransferase to measure parameters of blood coagulation, renal function, and liver function. Despite the therapeutic INR range being between 2.0 and 3.0 for American or European subjects, we set the normal range of INR between 1.8 and 2.4 because the cut-off value is more sensitive to warfarin for Asians [18]. We estimated the CHA2DS2-VASc and HAS-BLED scores as risk predictors of bleeding and thrombotic events. The HAS-BLED score could not be reliably calculated because of the unavailability of blood pressure measurements. We evaluated the presence of comorbidities using frequently observed diagnoses in patients with AF, including cardiovascular disease, diabetes, and dyslipidemia. In addition, we identified comedications for these conditions, such as antiarrhythmic drugs, antiplatelet monotherapy, dual antiplatelet therapy, and antidiabetic drugs.
Statistical analysis
To control for imbalances of multiple covariates in patient characteristics between the two groups, we calculated a propensity score (PS) for each subject as the predicted probability of receiving NOACs using a multiple logistic regression model with predictors of all potential confounders listed in Table 1. We used 1:1 matching of the cohorts on their estimated PS using a caliper width equal to 0.05 of the PS scale without replacement. Categorical variables were analyzed using the chi-square test and Fisher’s exact test, and continuous variables were analyzed using the independent t-test or Wilcoxon test. Before and after PS matching, the covariate balance between the NOAC and warfarin groups was assessed, and a two-sided p-value < 0.05 was considered statistically significant. Cox regression models were employed for the hazard ratios (HRs) with 95% confidence intervals (CIs) for outcomes of bleeding or thromboembolism. The variables were adjusted in the regression model if the variables had significant associations with the outcome in the bivariate analysis at each time period. We assessed the proportional hazards assumption by testing means of Schoenfeld residuals and the graphical methods and confirmed that the assumption was not violated. The analyses were performed using SAS software version 9.4 (College Station, TX, USA) for data management and statistical analyses were performed using STATA software version 15 (Cary, NC, USA).
Table 1
Clinical characteristics of NOAC users and matched warfarin users
Characteristics*
|
Before matching, no. (%)
|
After matching, no. (%)
|
NOACs
(n = 687)
|
Warfarin
(n = 388)
|
p-value‡
|
NOACs
(n = 246)
|
Warfarin
(n = 246)
|
p-value‡
|
Age, mean ± SD
|
75.5 (10.8)
|
72.5 (11.9)
|
0.000
|
73.5 (11.6)
|
74.4 (11.1)
|
0.375
|
Sex, male no. (%)
|
310 (45.1)
|
183 (47.2)
|
0.519
|
112 (45.5)
|
113 (45.9)
|
0.928
|
Entry year (%)
|
|
|
|
|
|
|
2015
|
276 (40.2)
|
261 (67.3)
|
0.000
|
145 (58.9)
|
146 (59.4)
|
0.984
|
2016
|
136 (19.8)
|
46 (11.9)
|
|
38 (15.5)
|
35 (14.2)
|
|
2017
|
141 (20.5)
|
50 (12.9)
|
|
38 (15.5)
|
39 (15.9)
|
|
2018
|
134 (19.5)
|
31 (8.0)
|
|
25 (10.2)
|
26 (10.6)
|
|
Serum creatinine (mg/dL)
|
|
|
0.000
|
|
|
0.683
|
≤ 1.4
|
494 (71.9)
|
233 (60.1)
|
|
158 (64.2)
|
152 (61.8)
|
|
> 1.4
|
88 (12.8)
|
72 (18.6)
|
|
35 (14.2)
|
42 (17.1)
|
|
Unknown
|
105 (15.3)
|
83 (21.4)
|
|
53 (21.5)
|
52 (21.1)
|
|
GOT (IU/L)
|
|
|
0.009
|
|
|
0.681
|
≤ 38
|
299 (43.5)
|
135 (34.8)
|
|
98 (39.8)
|
100 (40.7)
|
|
> 38
|
56 (8.2)
|
28 (7.2)
|
|
20 (8.1)
|
15 (6.1)
|
|
Unknown
|
332 (48.3)
|
225 (58.0)
|
|
128 (52.0)
|
131 (53.3)
|
|
GPT (IU/L)
|
|
|
0.043
|
|
|
0.775
|
≤ 41
|
458 (66.7)
|
253 (65.2)
|
|
152 (61.8)
|
154 (62.6)
|
|
> 41
|
59 (8.6)
|
20 (5.2)
|
|
19 (7.7)
|
15 (6.1)
|
|
Unknown
|
170 (24.8)
|
115 (29.6)
|
|
75 (30.5)
|
77 (31.3)
|
|
APTT (s)
|
|
|
0.000
|
|
|
0.911
|
< 22.5
|
12 (1.75)
|
4 (1.03)
|
|
5 (2.0)
|
3 (1.2)
|
|
22.5–32.5
|
244 (35.5)
|
69 (17.8)
|
|
53 (21.5)
|
55 (22.4)
|
|
> 32.5
|
58 (8.4)
|
37 (9.5)
|
|
27 (11.0)
|
27 (11.0)
|
|
Unknown
|
373 (54.3)
|
278 (71.7)
|
|
161 (65.5)
|
161 (65.5)
|
|
INR
|
|
|
0.000
|
|
|
0.000
|
< 1.8
|
232 (33.8)
|
134 (34.5)
|
|
59 (24.0)
|
87 (35.4)
|
|
1.8–2.4
|
13 (1.9)
|
81 (20.9)
|
|
4 (1.6)
|
47 (19.1)
|
|
> 2.4
|
8 (1.2)
|
41 (10.6)
|
|
4 (1.6)
|
23 (9.4)
|
|
Unknown
|
434 (63.2)
|
132 (34.0)
|
|
179 (72.8)
|
89 (36.2)
|
|
CHA2DS2–VASc score
|
|
|
0.000
|
|
|
0.672
|
0–2
|
113 (16.5)
|
101 (26.0)
|
|
55 (22.4)
|
47 (19.1)
|
|
3–4
|
233 (33.9)
|
146 (37.6)
|
|
88 (35.8)
|
91 (37.0)
|
|
≥ 5
|
341 (49.6)
|
141 (36.3)
|
|
103 (41.9)
|
108 (43.9)
|
|
HAS-BLED score
|
|
|
0.000
|
|
|
0.786
|
0–2
|
306 (44.5)
|
237 (61.1)
|
|
129 (52.4)
|
132 (53.7)
|
|
≥ 3
|
381 (55.5)
|
151 (38.9)
|
|
117 (47.6)
|
114 (46.3)
|
|
Previous bleeding
|
|
|
|
|
|
|
Intracranial
|
123 (17.9)
|
16 (4.1)
|
0.000
|
15 (6.1)
|
16 (6.5)
|
0.853
|
Gastrointestinal bleedings
|
66 (9.6)
|
34 (8.8)
|
0.647
|
22 (8.9)
|
21 (8.5)
|
0.873
|
Other bleedings
|
40 (5.8)
|
21 (5.4)
|
0.780
|
15 (6.1)
|
14 (5.7)
|
0.848
|
Previous thromboembolism
|
|
|
|
|
|
|
Myocardial infarction
|
19 (2.8)
|
5 (1.3)
|
0.115
|
2 (0.8)
|
3 (1.2)
|
0.653
|
Arterial thromboembolisms
|
265 (38.6)
|
106 (27.3)
|
0.000
|
86 (35.0)
|
90 (36.6)
|
0.707
|
Venous thromboembolisms
|
29 (4.2)
|
10 (2.6)
|
0.166
|
9 (3.7)
|
9 (3.7)
|
1.000
|
Comorbidity
|
|
|
|
|
|
|
Heart failure
|
333 (48.5)
|
209 (53.9)
|
0.089
|
125 (50.8)
|
125 (50.8)
|
1.000
|
Hypertension
|
483 (70.3)
|
260 (67.0)
|
0.261
|
166 (67.5)
|
171 (69.5)
|
0.628
|
Cerebrovascular disease
|
347 (50.5)
|
121 (31.2)
|
0.000
|
103 (41.9)
|
108 (43.9)
|
0.649
|
Other ischemic heart disease
|
229 (33.3)
|
130 (33.5)
|
0.954
|
80 (32.5)
|
77 (31.3)
|
0.772
|
Dyslipidemia
|
249 (36.2)
|
223 (57.5)
|
0.000
|
105 (42.7)
|
110 (44.7)
|
0.650
|
Diabetes mellitus
|
226 (32.9)
|
106 (27.3)
|
0.057
|
70 (28.5)
|
76 (30.9)
|
0.554
|
Asthma
|
59 (8.6)
|
40 (10.3)
|
0.349
|
17 (6.9)
|
21 (8.5)
|
0.499
|
COPD
|
122 (17.8)
|
91 (23.5)
|
0.025
|
52 (21.1)
|
52 (21.1)
|
1.000
|
Pneumonia
|
109 (15.9)
|
56 (14.4)
|
0.531
|
36 (14.6)
|
37 (15.0)
|
0.899
|
Psychiatric disorders
|
82 (11.9)
|
62 (16.0)
|
0.062
|
35 (14.2)
|
34 (13.8)
|
0.897
|
Fracture
|
62 (9.0)
|
31 (8.0)
|
0.562
|
23 (9.4)
|
18 (7.3)
|
0.415
|
Osteoarthritis
|
134 (19.5)
|
89 (22.9)
|
0.183
|
57 (23.2)
|
55 (22.4)
|
0.830
|
Anemia
|
71 (10.3)
|
40 (10.3)
|
0.990
|
20 (8.1)
|
26 (10.6)
|
0.353
|
Thyroid disease
|
58 (8.4)
|
66 (17.0)
|
0.000
|
29 (11.8)
|
25 (10.2)
|
0.564
|
Cancer
|
31 (4.5)
|
15 (3.9)
|
0.615
|
8 (3.3)
|
8 (3.3)
|
1.000
|
Comedication
|
|
|
|
|
|
|
Diuretics
|
189 (27.5)
|
61 (15.7)
|
0.000
|
53 (21.5)
|
54 (22.0)
|
0.913
|
Angiotensin- converting enzyme inhibitors/ Angiotensin receptor blockers
|
141 (20.5)
|
48 (12.4)
|
0.001
|
36 (14.6)
|
38 (15.5)
|
0.801
|
Beta blockers
|
128 (18.6)
|
49 (12.6)
|
0.011
|
40 (16.3)
|
36 (14.6)
|
0.618
|
Calcium channel blockers
|
155 (22.6)
|
59 (15.2)
|
0.004
|
45 (18.3)
|
51 (20.7)
|
0.495
|
Lipid-lowering agents
|
93 (13.5)
|
36 (9.3)
|
0.039
|
18 (7.3)
|
26 (10.6)
|
0.206
|
Antiarrhythmic Drugs
|
108 (26.2)
|
49 (12.6)
|
0.000
|
45 (18.3)
|
45 (18.3)
|
1.000
|
Mono-antiplatelet therapy
|
204 (29.7)
|
41 (10.6)
|
0.000
|
47 (19.1)
|
40 (16.3)
|
0.408
|
Dual-antiplatelet therapy
|
24 (3.5)
|
5 (1.3)
|
0.032
|
3 (1.2)
|
4 (1.6)
|
0.703
|
Diabetic medications
|
95 (13.8)
|
27 (7.0)
|
0.001
|
23 (9.4)
|
25 (10.2)
|
0.761
|
Steroids
|
62 (9.0)
|
24 (6.2)
|
0.099
|
17 (6.9)
|
19 (7.7)
|
0.729
|
NSAIDs
|
161 (23.4)
|
64 (16.5)
|
0.007
|
53 (21.5)
|
45 (18.3)
|
0.367
|
Proton pump Inhibitors
|
65 (9.5)
|
19 (4.9)
|
0.007
|
16 (6.5)
|
16 (6.5)
|
1.000
|
Abbreviations: NOACs, non-vitamin K oral anticoagulants; SD, standard deviation; no., number; GOT, aspartate aminotransferase; GPT, alanine aminotransferase; APTT, activated partial thromboplastin time; INR, international normalized ratio; NSAID, nonsteroidal anti-inflammatory drug; COPD, chronic obstructive pulmonary disease. |
*All comorbidities, CAD severity indicators, COPD severity indicators, and comedications were measured in the year before the cohort entry date. |
‡ p-value with < 0.05 represents meaningful differences between two groups. |
We analyzed the data of the patients receiving treatments within 0-6 months, 6 months-1 year, and 1-2 years to determine the risks of any major bleeding or thromboembolic event. In addition, we conducted three sensitivity analyses to examine the robustness of our findings. First, we re-evaluated the analyses by an intention-to-treat definition to account for the probability that censorship was associated with the outcome. In the intention-to-treat approach, we assumed that patients continuously used their initial anticoagulant during the entire follow-up period, regardless of any discontinuation or switching that occurred. Second, we excluded patients with a history of any major bleeding or thromboembolic event before cohort entry date. Finally, we performed PS-weighted analysis to maximize the representative cohort in a sensitivity analysis. Furthermore, we calculated the dosage of NOAC therapy and the TTR value for warfarin therapy using the Rosendaal method to determine whether an ideal therapeutic range was achieved [19]. We defined low-dose NOACs as dabigatran 110 mg twice daily or rivaroxaban 10 mg/day [20, 21]. A TTR level of ≥0.65 was considered to represent good clinical outcomes to minimize possible bleeding or thromboembolic complications in warfarin users [22].