Extracranial Hemorrhage and Stroke in Atrial Fibrillation

Abstract

Background and Purpose

Anticoagulation therapy reduces the risk for ischemic stroke in atrial fibrillation (AF) but also predisposes patients to hemorrhagic complications. There is limited knowledge on the risk of first-ever ischemic stroke in AF patients after extracranial hemorrhage (ECH).

Methods

We conducted a retrospective study using the California State Inpatient Database (SID) including all non-federal hospital admissions in California from 2005–2011. The exposure variable was hospitalization with a diagnosis of ECH with a previous diagnosis of AF. The outcome variable was a subsequent hospitalization with acute ischemic stroke. We excluded patients with stroke prior to or at the time of ECH diagnosis. We calculated adjusted hazard ratios (HRs) for ischemic stroke during follow up and at 6-month intervals using Cox regression models adjusted for pertinent demographics and co-morbidities. In subgroup analyses, subjects were stratified by primary ECH diagnosis, severity/type of ECH, age, CHA₂DS₂-VASc score, or the presence/absence of a gastrointestinal or genitourinary (GI/GU) cancer.

Results

We identified 764,257 AF patients (mean age 75 years, 49% women) without a documented history of stroke. Of these, 98,647 (13%) had an ECH-associated hospitalization, and 22,748 patients (3%) developed an ischemic stroke during the study period. Compared to patients without ECH, subjects with ECH had an approximately 15% higher rate of ischemic stroke (overall adjusted HR, 1.15 [95% CI, 1.11—1.19]). The risk appeared to remain elevated for at least 18 months after the index ECH. In subgroup analyses, the risk was highest in subjects with a primary admission diagnosis of ECH, severe ECH, gastrointestinal-type ECH, with GI/GU cancer, and age ≥ 60 years.

Conclusion

AF patients hospitalized with ECH may have a slightly elevated risk for future ischemic stroke. Particular consideration should be given to the optimal balance between the benefits and risks of anticoagulation therapy and the use of non-anticoagulant alternatives such as left atrial appendage closure in this vulnerable population.

Introduction

Atrial fibrillation (AF) is associated with an increased risk of ischemic stroke which is significantly attenuated with anticoagulation therapy, albeit with an increased risk of hemorrhagic complications.¹ Conversely, oral anticoagulation is routinely held in the setting of acute hemorrhage which potentially predisposes patients to an increased risk of thromboembolic complications. Indeed, it has been shown that patients with AF who develop intracranial hemorrhage while on anticoagulation therapy are at an increased risk of recurrent ischemic stroke^{2, 3}, and this has been attributed to cessation of anticoagulation therapy and other stroke prevention strategies.⁴ There is limited data, however, on the risk of ischemic stroke in patients with AF who develop an extracranial hemorrhage (ECH).

In the current study, we sought to determine the risk of ischemic stroke in AF patients with or without ECH utilizing the California State Inpatient Database (SID). In secondary analyses, we sought to compare the time course of risk of ischemic stroke after ECH in 6-month epochs up to 2 years after the index hospitalization.

Methods

Study cohort

Data used in this study is publicly available (www.hcup-us.ahrq.gov/sidoverview.jsp). We obtained approval to analyze data from the SID, which is an administrative dataset provided by the Healthcare Cost and Utilization Project (HCUP) (HCUP State Inpatient Databases (SID). Healthcare Cost and Utilization Project (HCUP). 2005–2011. Agency for Healthcare Research and Quality, Rockville, MD. www.hcup-us.ahrq.gov/sidoverview.jsp). The California SID contains publicly available de-identified information on all non-federal inpatient hospitalizations in California. Each entry in the dataset represents a hospitalization for a single patient and includes information on up to 25 ICD-9-CM (International Classification of Diseases, Ninth Revision, Clinical Modification) diagnosis codes. Each patient in the dataset is assigned a unique but anonymous identifier number, which allows for tracking of each patient’s hospitalizations longitudinally. This study used de-identified data and therefore was considered exempt by the Institutional Review Board at NYU Langone Health. Our study population was defined as all patients within the California SID dataset with AF or atrial flutter (ICD-9-CM codes 427.3x) hospitalized at non-federal hospitals in California from 2005 to 2011. All patients were followed from the time that the first diagnosis of AF appeared in the dataset.

Exposure variable

The exposure variable was hospitalization with ECH, defined as any major extracranial bleeding episode and identified in the dataset by previously validated ICD-9-CM codes for gastrointestinal bleeding and other major bleeding (Supplemental Table 1).^{5, 6} Subjects who did not have ICD-9-CM codes for ECH in any diagnostic position were considered unexposed controls. We excluded subjects who had a history of ischemic stroke (identified by ICD-9-CM codes 433.x1, 434.x1, and 436) or intracranial hemorrhage (identified by ICD-9-CM codes 430, 431, and 432) prior to the index ECH^{7, 8} and those who were diagnosed with an ischemic stroke during their index hospitalization.

Outcome

The primary outcome of interest was re-hospitalization with an acute ischemic stroke after the index hospitalization for ECH and after the first hospitalization for the control patients. Ischemic stroke was defined as the presence of ICD-9-CM codes 430, 433.x1, 434.x1, and 436 in any diagnostic position.^{7, 8} The sensitivity and specificity of these diagnostic codes for acute ischemic stroke have been previously reported as 86% and 95%, respectively.^{8, 9}

Covariates

Demographic factors:

Age and sex.

Clinical variables:

Hypertension, diabetes mellitus, hyperlipidemia, congestive heart failure, coronary heart disease, peripheral vascular disease, chronic kidney disease, and smoking status as identified using previously validated ICD-9 codes.^10–12 Demographic factors and clinical variables were used to calculate individuals’ CHA₂DS₂-VASc scores at the time of their index hospitalization.

Analytical plan

We fitted Cox regression models to compare the risk of ischemic stroke between the exposed and unexposed groups. We used the following pre-specified models: Model 1: unadjusted; Model 2: adjusted for age and sex; Model 3: model 2 + hypertension, diabetes, hyperlipidemia, chronic kidney disease, coronary heart disease, congestive heart failure, smoking status; and Model 4: adjusted for CHA₂DS₂-VASc score. The Cox proportional hazards assumption was tested and met for all models. We also calculated the risk of ischemic stroke in patients with ECH across 6-month epochs over a two-year period. To determine the effect of residual confounding on our results, we calculated the E-value, defined as the minimum strength of association that an unmeasured confounder must have to explain away a specific exposure-outcome association.^{13, 14}

We conducted separate secondary analyses within subgroups that were dichotomized by age (< 60 and ≥ 60), CHA₂DS₂-VASc score (< 2 and ≥ 2), presence and absence of a primary ECH diagnosis during the index hospitalization (defined as ECH diagnosis within first three diagnostic positions of dataset), presence and absence of severe ECH (defined as concomitant blood transfusion procedure using presence of ICD-9 procedure codes 99.03 and 99.04 during the index hospitalization¹⁵), classification of ECH as gastrointestinal (GI) bleed and non-GI bleed, and classification based on the presence or absence of a history of a GI or genitourinary (GU) cancer at the time of ECH (Supplementary Table I).

All analyses were performed using SPSS version 25.0 (Chicago, IL) and a two-sided p < 0.05 was considered statistically significant throughout.

Results

We identified 764,257 patients with AF, with a mean age of 75.2 years (SD = 12.7 years); 48.5% were women. In the entire cohort, ECH occurred in 13.3% (n= 98,647) of patients and 3.4% (n= 22,748) patients were re-admitted for an acute ischemic stroke after their index admission. Overall, patients with ECH and ischemic stroke were more likely to be older and had higher rates of co-morbid conditions (Table 1 and Supplementary Table II).

Table 1.

Characteristics of patients stratified by presence of extracranial hemorrhage.

Characteristics	ECH (n= 98,647)	No ECH (n= 665,610)
Age, yrs; mean (SD)	77.5 (11.0)	74.9 (12.9)
Women (n(%))	44,799 (45.4)	325,769 (48.9)
Race/Ethnicity (n(%))
White	69,610 (70.6)	486,050 (73.0)
Black	5,193 (5.3)	29,826 (4.5)
Hispanic	11,386 (11.5)	73,507 (11.0)
Asian/Pacific Islander	8,103 (8.2)	42,980 (6.5)
Congestive heart failure (n(%))	60,235 (61.1)	295,756 (44.4)
Hypertension (n(%))	42,083 (42.7)	181,522 (27.3)
Coronary heart disease (n(%))	56,773 (57.6)	304,479 (45.7)
Peripheral vascular disease (n(%))	4,991 (5.1)	19,430 (2.9)
Diabetes mellitus (n(%))	37,837 (38.4)	205,843 (30.9)
Chronic kidney disease (n(%))	37,210 (37.7)	152,188 (22.9)
Hyperlipidemia (n(%))	48,195 (48.9)	293,075 (44.0)
Smoking (n(%))	29,607 (30.0)	147,585 (22.2)
CHA2DS2-VASc score; median (IQR)	3 (2)	3 (2)

Association between extracranial hemorrhage and incident ischemic stroke

Kaplan-Meier analysis indicated that subjects with ECH had lower stroke-free survival probabilities overall than those who did not have ECH (Figure 1). In unadjusted Cox regression, ECH was associated with an increased risk of incident ischemic stroke during follow up (Overall HR, 1.15 [95% CI, 1.11—1.19]). This association persisted after adjusting for potential confounders and the composite CHA₂DS₂-VASc score, respectively (Table 2, models 2 to 4). Furthermore, the effect size of ECH on ischemic stroke risk remained stable in 6 months epochs over the first 18 months from the index hospitalization in ECH patients (Supplementary Table III). For our adjusted HR, the E-value was calculated to be 1.56 [Lower 95% CI 1.29].

Figure 1.

Stroke-free survival probabilities for patients with: a) any ECH b) ECH as principal diagnosis c) ECH with history of gastrointestinal/genitourinary cancer and d) ECH without history of gastrointestinal/genitourinary cancer.

Table 2.

Overall hazard ratios for ischemic stroke in ECH patients compared to non-ECH patients in the four different Cox regression models, adjusting for pertinent co-variates.

Model:	Overall Hazard Ratio (95% CI)
1	1.15 (1.11—1.19)
2	1.12 (1.09—1.17)
3	1.15 (1.11—1.19)
4	1.09 (1.05—1.13)

Model 1: unadjusted

Model 2: Adjusted for age and sex

Model 3: Adjusted for age, sex, smoking status, CHF, CHD, HLD, HTN, DM, PVD, CKD

Model 4: Adjusted for CHA₂DS₂-VASc score

Additional subgroup analyses

In our sub-analyses, the overall risk of ischemic stroke was particularly elevated among subjects with primary ECH (adjusted HR, 1.18 [95% CI, 1.13–1.24]), severe ECH (adjusted HR, 1.21 [95% CI, 1.14—1.27]), GI hemorrhage (adjusted HR, 1.18 [95% CI, 1.13—1.24]), history of GI/GU cancer (adjusted HR, 1.40 [95% CI, 1.20—1.64]) and those 60 years of age or older (adjusted HR, 1.16 [95% CI, 1.12—1.20]) (Table 3). Survival curves for subgroups are displayed based on ECH characteristics (Figure 2) and patient characteristics (Figure 3).

Table 3.

Hazard ratios of subgroup analysis for ECH patients stratified by primary ECH diagnosis, presence of severe or non-severe ECH, presence/absence of gastrointestinal-type (GI) ECH, history of GI/GU cancer, age, and CHA₂DS₂-VASc score.

	Overall HRs
	Model 1	Model 2	Model 3	Model 4
Primary ECH diagnosis (n= 710,693)	1.22 (1.16–1.27)	1.16 (1.11–1.21)	1.18 (1.13–1.24)	1.17 (1.12–1.22)
Severe ECH (n= 708,532)	1.23 (1.17–1.30)	1.17 (1.11–1.24)	1.21 (1.14–1.27)	1.12 (1.06–1.18)
Non-severe ECH (n= 721,335)	1.10 (1.06–1.15)	1.09 (1.05–1.14)	1.11 (1.06–1.16)	1.07 (1.02–1.12)
GI ECH (n= 732,383)	1.23 (1.17–1.28)	1.16 (1.11–1.21)	1.18 (1.13–1.24)	1.15 (1.10–1.20)
Non-GI ECH (n= 667,827)	1.01 (0.88–1.17)	1.05 (0.91–1.21)	1.09 (0.94–1.26)	0.96 (0.83–1.11)
History of GI/GU cancer (n= 43,117)	1.46 (1.25–1.71)	1.39 (1.19–1.62)	1.40 (1.20–1.64)	1.40 (1.20–1.63)
No history of GI/GU cancer (n= 718,666)	1.15 (1.11–1.19)	1.12 (1.08–1.16)	1.14 (1.10–1.19)	1.09 (1.05–1.13)
Age < 60 years (n= 91,959)	1.01 (0.83–1.22)	1.00 (0.82–1.20)	0.93 (0.77–1.13)	0.94 (0.78–1.14)
Age >= 60 years (n= 671,807)	1.13 (1.09–1.17)	1.13 (1.09–1.17)	1.16 (1.12–1.20)	1.09 (1.06–1.14)
CHA2DS2-VASc score < 2 (n= 130,424)	1.21 (1.05–1.41)	1.18 (1.02–1.37)	1.17 (1.00–1.36)	1.20 (1.03–1.39)
CHA2DS2-VASc score ≥ 2 (n= 633,833)	1.11 (1.07–1.15)	1.12 (1.08–1.16)	1.15 (1.10–1.19)	1.08 (1.04–1.12)

Model 1: unadjusted

Model 2: Adjusted for age and sex

Model 3: Adjusted for age, sex, smoking status, CHF, CHD, HLD, HTN, DM, PVD, CKD

Model 4: Adjusted for CHA₂DS₂-VASc score

ECH characteristic subgroups may not add up to total N of study (N= 764,257) because the unexposed (no ECH) group is unchanged in these subgroups.

Figure 2.

Stroke-free survival probabilities for subgroup analyses based on extracranial hemorrhage (ECH) characteristics: a) Severe ECH b) non-severe ECH c) Gastrointestinal (GI) hemorrhage and d) non-GI hemorrhage.

Figure 3.

Stroke-free survival probabilities for subgroup analyses based on patient characteristics: a) Age < 60 years b) age ≥ 60 years c) CHA₂DS₂-VASc score < 2 and d) CHA₂DS₂-VASc score ≥ 2.

Discussion

The most important finding of our study was that ECH is associated with increased risk for incident ischemic stroke in patients with AF. This association was stronger in subgroups of patients with advanced age, more severe hemorrhage, and GI type hemorrhage, and concomitant GI/GU cancer history. Furthermore, the effect size remained relatively stable for at least 18 months from the initial ECH hospitalization suggesting that the elevated risk continues over time which is also supported by the survival curves continued separation over time.

A possible explanation for this relates to the fact that AF patients at highest risk for hemorrhagic complications are typically also at high risk for stroke given shared risk factors such as older age.^{16, 17} Nevertheless, our associations persisted after adjusting for a wide range of pertinent risk factors including age. An alternative explanation may be that ECH prompted withholding of anticoagulant therapy. Prior retrospective and observational studies suggested that resuming warfarin after a GI bleed was associated with reduced risk of thromboembolic events and death, without a significant increase in recurrent GI bleeding.^18–20 However, because the SID does not include information on medication therapy, we were unable to test this hypothesis. Furthermore, it is also possible that the hemorrhage²¹ and related treatments may be a direct cause of increased risk of ischemic stroke in the short-term.²¹ Moreover, it has been repeatedly shown that packed red blood cell transfusions increase the risk for thromboembolic complications including cardiovascular events.^22–25 In our study, patients who had a severe ECH requiring blood transfusions were among subgroups with the highest risk of ischemic stroke. Lastly, it is possible that a subset of ECH patients had concurrent anticoagulation reversed such as with prothrombin complex concentrate, vitamin K, or fresh frozen plasma, which has been associated with a modest increase in the risk of thrombotic complications.^26–28 These may explain the early risk of ischemic stroke after ECH but the continued increased risk of ischemic stroke after ECH makes blood product and coagulation factor administration unlikely to be the sole mechanism of increased ischemic stroke risk.

It is important to note that our patient cohort was stroke free at the time of each patient’s index hospitalization and thus our results may not extend to patients with history of cerebrovascular disease and stroke. In fact, the risk of stroke after ECH may be more pronounced among patients with prior stroke history, and this deserves further exploration in future studies. In addition, our exclusion of patients with ischemic stroke and GI bleeding occurring in the same hospitalization may have underestimated the effect size in our study. Arguably, however, anticoagulation in the acute setting of an ECH may be contraindicated in most patients. Accordingly, our findings are of importance as they highlight the longer-term risk for thromboembolic complications and the critical need to better define patients at risk and approaches to stroke prophylaxis. Furthermore, the risk of death has been shown to be increased in patients with GI bleed²⁹ and thus death may have constituted a competing risk in these patients.

Our data indicates that in addition to the management of modifiable risk factors, anticoagulation should be restarted as soon as possible after the bleeding source has been secured; an approach that is supported by observations that bleeding recurrence after anticoagulation resumption was not significantly increased.^{18, 20, 30} Nevertheless, the presumed risk for major bleeding complications still needs to be carefully weighed as the 30-day mortality with anticoagulation related GI bleeding has been reported to be as high as 8%.³¹ For high-risk patients, left atrial appendage closure may be considered to avoid long-term anticoagulation.³² Although left atrial appendage closure requires peri-procedural anticoagulant therapy, emerging data indicates that this may not be uniformly required and could even be used in the setting of a GI bleed with concurrent use of octreotide.^33–35 Nevertheless, because the overall risk in our cohort was modest and we lack information on anticoagulant therapy it remains to be determined whether these approaches truly improve overall outcomes and stroke risk and randomized clinical trials are needed to establish the safety and efficacy of left atrial appendage closure in patients with ECH.

Strengths and Limitations

Our study has several strengths including using a large administrative dataset with real world data, making it sufficiently powered to adjust for potential confounders and increasing its generalizability. Limitations include those inherent to observational studies such as selection, indication, and follow-up bias. In addition, we lacked data on anticoagulation status in patients both before and after the ECH, though it is important to note that in patients of younger age (< 60 years) the ECH may have not occurred in the setting of anticoagulation as these patients may have not been prescribed anticoagulation with the absence of other risk factors in CHA₂DS₂-VASc stroke risk assessment. While we adjusted for risk factors for ischemic stroke in AF, we lacked data on other predictors, including serum and imaging markers. In particular, the association between ischemic stroke and ECH may have been in part caused by INR lability, as variations in INR due to suboptimal adherence to anticoagulation and/or drug interactions may have affected the risk of both ECH and ischemic stroke. Further, we used administrative data that only included inpatient hospitalizations and thus patients with hemorrhagic complications managed on an outpatient basis were not captured. In addition, the competing risk of death may have led to underestimating our ischemic stroke in patients with ECH. Although we used well-validated and commonly employed ICD-9-CM diagnostic and procedural codes to identify our exposure and outcome of interest, there is still a possibility of information bias through miscoding. Lastly, this study was performed between 2005 and 2011, which is prior to the advent of direct oral anticoagulants and left atrial appendage occlusion and thus we are unable to extrapolate whether these associations hold true in an era where direct oral anticoagulant use is more common than warfarin use. Given these limitations, our results should be interpreted cautiously, considered hypothesis generating only, and not guide clinical decision making at this point pending confirmation by future prospective studies.

Conclusion

Inpatients with AF and ECH may be at increased risk for ischemic stroke, though our calculated effect size is small. Although the clinical impact of our findings needs further exploration, the results may have significant implications for future research to test the safety and efficacy of stroke prevention strategies, such as anticoagulation resumption and left atrial appendage closure to reduce the risk of ischemic stroke in this patient population.

Non-standard Abbreviations and Acronyms

AF

Atrial fibrillation

ECH

Extracranial hemorrhage

GI

Gastrointestinal

GU

Genitourinary

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

SID

State Inpatient Database