Clinical and neuroimaging risk factors associated with the development of intracerebral hemorrhage while taking direct oral anticoagulants

Abstract

Background and aims

Intracerebral hemorrhage (ICH) associated with direct oral anticoagulant (DOAC) usage confers significant mortality/disability. We aimed to understand the clinical and neuroimaging features associated with developing ICH among DOAC users.

Methods

Clinical and radiological data were collected from consecutive DOAC users with ICH (DOAC-ICH) and age-matched controls without ICH from a single referral center. The frequency/distribution of MRI markers of hemorrhage risk were assessed. Baseline demographics and neuroimaging markers were compared in univariate tests. Significant associations (p < 0.1) were entered into a multivariable regression model to determine predictors of ICH.

Results

86 DOAC-ICH and 94 ICH-free patients were included. Diabetes, coronary artery disease, prior ischemic stroke, smoking history, and antiplatelet usage were more common in ICH patients than ICH-free DOAC users. In the neuroimaging analyses, severe white matter hyperintensities (WMHs), lacunes, cortical superficial siderosis (cSS), and cerebral microbleeds (CMBs) were more common in the ICH cohort than the ICH-free cohort. In the multivariable regression, diabetes [OR 3.53 95% CI (1.05–11.87)], prior ischemic stroke [OR 14.80 95% CI (3.33–65.77)], smoking history [OR 3.08 95% CI (1.05–9.01)], CMBs [OR 4.07 95% CI (1.45–11.39)], and cSS [OR 39.73 95% CI (3.43–460.24)] were independently associated with ICH.

Conclusions

Risk factors including diabetes, prior stroke, and smoking history as well as MRI biomarkers including CMBs and cSS are associated with ICH in DOAC users. Although screening MRIs are not typically performed prior to initiating DOAC therapy, these data suggest that patients of high-hemorrhagic risk may be identified.

Introduction

Due to increasing antithrombotic usage over the last two decades, one-fifth of intracerebral hemorrhage (ICH) cases are associated with oral anticoagulation [1, 2]. Although direct oral anticoagulants (DOACs) may have a favorable safety profile compared to warfarin [3-5], DOAC use carries an annual ICH risk of 0.3–0.6% per year [6-9]. Moreover, recent evidence also suggests that ICH associated with DOAC use (DOAC-ICH) may result in equally devasting functional outcomes as warfarin-related ICH [10-12]. While several major risk prediction tools have been developed for warfarin-associated hemorrhage risk in atrial fibrillation (and applied to warfarin-associated ICH) [13], the risk factors associated with increased ICH risk in DOAC use remain unclear.

In recent years, established neuroimaging markers such as cerebral microbleeds (CMBs) [14, 15] and cortical superficial siderosis (cSS) [16, 17] have been shown to be associated with increased hemorrhage risk in warfarin-related ICH [18]. However, their significance in DOAC-ICH is unknown. In this study, we aimed to understand the clinical and neuroimaging risk factors that portend increased hemorrhagic risk among DOAC users.

Methods

Patients with DOAC-ICH from January 2003 to September 2019 were retrospectively identified from our center’s prospectively entered database of consecutive patients with spontaneous ICH. DOAC users free of ICH from December 2017 to August 2020 were obtained using the Healthcare System Research Patient Data Registry (RPDR), an online query tool that incorporates ICD10-CM diagnoses to search our institution’s electronic health record [19]. We queried for patients aged 18 year or older with atrial fibrillation on DOACs who did not have a past medical history of ischemic stroke or ICH. The patients detailed medical records were subsequently reviewed to confirm whether patients were eligible, e.g., documentation of taking DOAC usage with an available MRI. Patients that did not receive a brain MRI while they were taking DOACs were excluded from the study. The specific details of this search are detailed in the supplemental data. From this cohort, potential subjects were age matched to the DOAC-ICH group.

Clinical and demographic variables

Demographic information and medical history, including vascular risk factors such as hypertension, hyperlipidemia, diabetes mellitus, any history of smoking, alcohol usage, coronary artery disease, atrial fibrillation, and previously reported history of transient ischemic attack/ischemic stroke, were obtained from patient charts of both cohorts (DOAC-ICH and ICH-free DOAC users) as previously described [20]. Pre-hospital antithrombotics and baseline modified Rankin Scale (mRS) scores were also collected from chart review [21].

Neuroimaging analysis

The location of the ICH on computed tomography (CT) scans was classified as either deep, lobar, intraventricular, or cerebellar. Brain MRIs were reviewed by a board-certified neurologist (ASD) for the presence of cSVD markers (as defined by STRIVE criteria [22]) including cerebral microinfarcts [23], WMHs (graded using Fazekas score [24] and classified according to established patterns [25]), lacunes [26, 27], enlarged perivascular spaces (EPVS, graded as previously described [28]), cSS [29], and CMBs. Severe WMHs were defined as Fazekas score ≥ 2, and severe EPVS were defined as a score > 2 [28]. The presence of cerebral amyloid angiopathy (CAA) was diagnosed based on the ICH location and presence and distribution of strictly lobar CMBs and cSS per modified Boston criteria [30]. The SVD score was determined for each patient [31].

Statistical analysis

Median and interquartile range (IQR) were reported for continuous variables and percent and count were reported for categorical variables. Differences were assessed using either t-test or the nonparametric Wilcoxon rank-sum for continuous variables and Fisher’s Exact test for categorical variables. Significant associations (p < 0.1) were entered into a multivariable regression model to determine predictors of ICH development. Two-tailed p values < 0.05 were interpreted as statistically significant.

Analyses were performed with SPSS for Windows, version 23.0 (IBM Corp., Armonk, N.Y., USA). Approval for this study was granted by our hospital’s institutional review board. Informed consent was waived for this study given its retrospective nature and minimal patient risk. Unpublished data are available upon reasonable request by a qualifying investigator.

Results

86 patients out of a total of 1791 patients with spontaneous ICH from 2003 to 2019 were identified as having DOAC-ICH. 94 patients comprised the ICH-free cohort consisting of patients taking DOACs. The baseline characteristics of each cohort are shown in Table 1. The mean age was 77 (± 9) years in the ICH free group and 76 (± 11) in the ICH group (p = 0.52). Within the ICH group, 41 (48%) had deep ICH, 35 (41%) had lobar ICH, 6 (7%) had primary intraventricular hemorrhage, and 4 (5%) patients had cerebellar hemorrhage.

Table 1.

Baseline Characteristics of DOAC Users With and Without ICH

	ICH free (n = 94)	ICH (n = 86)	p value
Age (years), mean (± SD)	77 (± 9)	76 (± 11)	0.52
Female sex	48 (51)	34 (40)	0.14
Race			0.13
White	86 (92)	81 (94)
Asian	1 (1)	1 (1)
Black	5 (5)	4 (5)
Hypertension	86 (92)	77 (90)	0.80
Hyperlipidemia	66 (70)	67 (78)	0.31
Diabetes	13 (14)	34 (40)	< 0.01
Coronary artery disease	21 (22)	30 (35)	0.06
Atrial fibrillation	94 (100)	72 (84)	< 0.01
Prior TIA/ischemic stroke	3 (3)	29 (34)	< 0.01
Smoking	42 (45)	52 (61)	0.03
Alcohol use	45 (48)	38 (44)	0.62
Dementia	7 (7)	8 (9)	0.65
Prior dependence	19 (20)	24 (28)	0.30
Baseline mRS score ≤ 2	75 (80)	62 (72)	0.23
Antiplatelet use	13 (14)	27 (31)	< 0.01
DOAC Type
Apixaban	60 (64)	41 (48)	0.04
Rivaroxaban	32 (34)	39 (45)	0.13
Dabigatran	2 (2)	6 (7)	0.16
Duration of anticoagulation use (weeks), mean (± SD)	90 (± 69)	89 (± 84)	0.93
Baseline creatinine, mean (± SD)	1.04 (± 0.39)	1.23 (± 1.04)	0.11

Data are counts (n) and percentages (%), means and standard deviations (SD), or medians and interquartile ranges (IQR)

DOAC direct oral anticoagulant, ICH intracerebral hemorrhage, TIA transient ischemic attack, mRS modified Rankin Scale

Diabetes was more common in patients with ICH (40% vs. 14%, p < 0.01) as was coronary artery disease (35% vs. 22%, p = 0.06), prior history of transient ischemic attack/ischemic stroke (34% vs. 3%, p < 0.01), and smoking history (61% vs. 45%, p = 0.03). Antiplatelet usage was also more frequent among patients with ICH (31% vs. 14%, p = 0.01). Notably, the mean duration of anticoagulation use was similar between ICH and ICH free patients (90 ± 69 weeks vs. 89 ± 84 weeks, p = 0.93).

In the neuroimaging analyses, MRI was performed in 48 (56%) patients within the DOAC-ICH cohort and was available for all patients in the ICH free group. Severe WMHs (79% vs. 52%, p < 0.01), lacunes (46% vs. 30%, p = 0.06), cSS (23% vs. 1%, p < 0.01), and CMBs (60% vs. 23%, p < 0.01) were more frequent in patients with ICH than patients free of ICH (Table 2). Based on the presence/distribution of CMBs, cSS, and macrobleeds, 16 (33%) patients with ICH were diagnosed with probable CAA compared to 4% of patients without ICH (p < 0.01). Collectively, 98% of patients with ICH had at least one small vessel disease biomarker compared to 66% of patients without ICH (p < 0.01). Furthermore, the median SVD score was higher in the DOAC-ICH group than in the ICH free cohort (2 (1,3) vs. 1 (0,2), p < 0.01).

Table 2.

Imaging Characteristics of DOAC Users With and Without ICH

	ICH free (n = 94)	ICH (n = 48)	p Value
Severe white matter hyperintensities (Fazekas score > 1)	49 (52)	38 (79)	< 0.01
Lacunes	28 (30)	22 (46)	0.06
Severe enlarged perivascular spaces (EPVS)	11 (12)	7 (15)	0.59
Basal ganglia EPVS	5 (5)	5 (11)	0.30
Centrum semiovale EPVS	6 (6)	3 (6)	1.00
Cortical superficial siderosis	1 (1)	11 (23)	< 0.01
Cerebral microbleeds	22 (23)	29 (60)	< 0.01
Any one small vessel disease biomarker	62 (66)	47 (98)	< 0.01
Small vessel disease score	1 (0, 2)	2 (1, 3)	< 0.01
Probable cerebral amyloid angiopathy	4 (4)	16 (33)	< 0.01

Data are counts (n) and percentages (%), means and standard deviations (SD), or medians and interquartile ranges (IQR)

DOAC direct oral anticoagulant, ICH intracerebral hemorrhage

These significant clinical variables (including diabetes, coronary artery disease, prior ischemic stroke history, smoking history, and antiplatelet usage) were entered into a multivariable regression analysis along with the significant neuroimaging variables (WMHs, lacunes, cSS, and CMBs) to determine associations with ICH (Table 3). In the regression model, diabetes [OR 3.53 95% CI (1.05–11.87)], prior transient ischemic attack/ischemic stroke history [OR 14.80 95% CI (3.33–65.77)], smoking history [OR 3.08 95% CI (1.05–9.01)], CMBs [OR 4.07 95% CI (1.45–11.39)], and cSS [OR 39.73 95% CI (3.43–460.24)] were found to be independently associated with ICH.

Table 3.

Risk factors associated with intracerebral hemorrhage among DOAC users

	Univariate		Multivariate
	OR (95% CI)	p value	OR (95% CI)	p value
Diabetes	4.07 (1.97–8.44)	< 0.01	3.53 (1.05–11.87)	0.04
Coronary artery disease	1.86 (0.97–3.59)	0.06	1.53 (0.50–4.71)	0.46
Atrial fibrillation	N/A	N/A	N/A	N/A
TIA/ischemic stroke	15.43 (4.49–53.01)	< 0.01	14.80 (3.33–65.77)	< 0.01
Smoking history	1.89 (1.05–3.43)	0.035	3.08 (1.05–9.01)	0.04
Antiplatelet usage	2.85 (1.36–5.99)	0.01	1.83 (0.56–5.95)	0.31
Severe WMHs	3.49 (1.56–7.81)	< 0.01	2.35 (0.73–7.56)	0.15
Lacunes	2.00 (0.97–4.10)	0.06	0.54 (0.18–1.64)	0.28
CMBs	5.00 (2.36–10.58)	< 0.01	4.07 (1.45–11.39)	< 0.01
cSS	27.65 (3.45–221.81)	< 0.01	39.73 (3.43–460.24)	< 0.01

Data are represented with odds ratios (OR) and 95% confidence intervals (CI)

DOAC direct oral anticoagulant, TIA transient ischemic attack, WMHs white matter hyperintensities, CMBs cerebral microbleeds, cSS cortical superficial siderosis

Discussion

Herein, we demonstrate that certain clinical risk factors such as diabetes, prior stroke history, and smoking history portend an increased risk of ICH while on DOAC therapy. Moreover, we also demonstrate that patients with hemorrhagic neuroimaging markers, such as CMBs and cSS, are associated with an increased risk of developing ICH. To our knowledge, this is the first study that has identified these specific neuroimaging features to be associated with DOAC-ICH.

Regarding clinical risk factors, in our study, although concurrent antiplatelet usage was found to be a risk factor for ICH in univariate analyses, this was not significant in the multivariable regression model. However, in the Randomized Evaluation of Long-term anticoagulant therapY (RE-LY) trial comparing dabigatran use to warfarin use, concomitant aspirin therapy was shown to increase the risk of ICH (relative risk of 1.6) [32], and in the Rivaroxaban versus Warfarin in Nonvalvular Atrial Fibrillation (ROCKET-AF) trial, baseline thienopyridine was associated with increased ICH risk (hazard ratio of 2.57 95% CI 1.30–5.07) [33]. Similar results have been shown with the addition of antiplatelet therapy to warfarin [34, 35]. Interestingly, in a large retrospective trial of DOACs compared to warfarin using the Get With The Guidelines–Stroke registry, Inohara et al. found an increase in in-hospital mortality with concomitant antiplatelet therapy in the warfarin arm; however, this finding was not observed in the DOAC arm [4].

Our study found an association between diabetes and ICH risk. In a recent post hoc analysis of Clinical Relevance of Microbleeds in Stroke (CROMIS-2), a prospective cohort study which examined risk factors for ICH in patients taking oral anticoagulants, diabetes was shown to be a strong risk factor for developing ICH (HR 3.91, 95% CI 1.34–11.4) [36]. Diabetes was also found to be a risk factor for warfarin-related ICH in one older study [37]. Studies linking diabetes as a risk factor for spontaneous ICH have yielded conflicting results, although two meta-analyses incorporating large numbers of patients have shown a modest association [38, 39]. In addition to diabetes, our study demonstrated that a history of transient ischemic attack/ischemic stroke was a risk factor for ICH [OR 14.80 95% CI (3.33–65)]. Other studies have demonstrated a 15-fold increase in ICH in the first year following an ischemic stroke [40]. It is likely that patients with a history of ischemic stroke have a higher burden/severity of vascular risk factors contributing to the increased risk of ICH.

Smoking was also found to be a risk factor for DOAC-ICH in our study. Although this finding has not been reportedly previously in oral anticoagulant-related ICH, smoking is a strong modifiable risk factor for ICH [41]. Smoking is also a risk factor for the development of CMBs, which may in turn, contribute to ICH risk [42]. While not shown in our study, several additional risk factors have been associated with DOAC-ICH in the literature including age, previous stroke/transient ischemic attack, malignancy, fall risk, hyperlipidemia, low creatinine clearance, peripheral arterial disease, Asian race, Black race, reduced serum albumin, and reduced platelet count [32, 33, 43]. Heterogeneity in these study designs as well as ours may account for such differing results.

With respect to the neuroimaging analyses, our study demonstrated that CMBs are a risk factor for DOAC-ICH. This is in agreement with a previous study showing that CMBs are a risk factor for warfarin-related ICH [18]. Given that the duration of anticoagulation use was similar between groups, it is possible that patients in the DOAC-ICH group had CMBs prior to the initiation of anticoagulation which resulted in their hemorrhage. Such a notion is concordant with findings from CROMIS-2, which demonstrated that CMBs were independently associated with symptomatic ICH risk in patients with atrial fibrillation [15]. Our study also demonstrated that cSS, an exclusive marker for CAA [30, 44], was strongly associated with development of an ICH while on DOAC therapy. In accordance with this finding, our study found that nearly one-third of patients that suffered an ICH on DOAC therapy had underlying probable CAA (compared to 4% in the ICH free group, p < 0.01), which carries a specificity of 96% for the diagnosis of CAA based on the validation study of the modified Boston criteria [45]. CAA was also found to be a major contributor to warfarin-associated ICH in an earlier study in which CAA was diagnosed by pathology in 64% of patients with available tissue samples [46]. In contrast to a recent report [43], although WMH were associated with ICH in the univariate analyses, this relationship was not significant after controlling for relevant cofounding risk factors such as lacunes, coronary artery disease, and CMBs.

Our study has several limitations owing to its retrospective nature and small sample size. Due to the high mortality rate among patients with ICH, MRI was not available in all patients. Given the likelihood that the patients who died (and were unable to receive an MRI) may have had a higher burden of vascular risk factors and hemorrhagic biomarkers [47], this potential selection bias would only further support our conclusion to reject the null hypothesis. However, a recent study demonstrated that patients with more severe underlying small vessel disease (represented by CMBs count) may actually have smaller baseline hematoma volumes, potentially leading to reduced mortality [48]. While hematoma volumes were not directly assessed in our study, it can be assumed that patients who died in our study may have had larger hematoma volumes, and therefore, fewer CMBs.

Because MRI studies are not usually performed prior to initiating anticoagulation in clinical practice, our ICH free control group consisted of patients who received an MRI for other neurological reasons (e.g. migraines). Because of this selection bias, it is conceivable that our control group may have an increased amount of neuroimaging findings such as WMHs, that may not have been found if the study was performed in a prospective, randomized fashion. However, such bias would again only further support our conclusion to reject the null hypothesis. On the other hand, these patients may have had better access to health care resources that may have resulted in the inadvertent selection of patients with favorable neurological health profiles. Such selection bias may have resulted in the observed differences in the clinical risk factors of diabetes and smoking between groups. Importantly, although MRIs are not currently performed prior to the development of ICH, our data suggest that patients of high-hemorrhagic risk may be identified.

Our study did not explore the role of APOE ε variants, which are associated with CAA [49, 50], as risk factors for DOAC-ICH. A large, prospective study demonstrated that APOE ε2 and ε4 were risk factors for warfarin-associated ICH, specifically those with ICH in lobar locations [51]. Although we confirmed a high frequency of CAA in our cohort, future studies are needed to explore whether patients carrying these high-risk APOE alleles and receiving DOACs are at an increased risk of ICH.

Conclusion

A recent report has suggested that the HAS-BLED score [52], a widely used hemorrhage risk prediction score for oral anticoagulation, performed poorly at predicting ICH risk in DOAC users [43]. Our study demonstrates several novel risk markers for DOAC-ICH that may be useful in future randomized trials designed to evaluate ICH risk in DOAC users. Creation of accurate prediction models utilizing large numbers of patients are urgently needed as DOACs have now replaced warfarin as the primary oral anticoagulant in atrial fibrillation. Future studies should evaluate the role of nonpharmacological management strategies in conditions that warrant life-long anticoagulation (i.e., atrial fibrillation) in patients that carry the risk markers identified in this study.