Ischemic Stroke and Bleeding – Clinical Benefit of Anticoagulation in Atrial Fibrillation after Intracerebral Hemorrhage

Abstract

Background and Purpose:

Patients with intracerebral hemorrhage (ICH) and atrial fibrillation (AF) are at risk for ischemic events. While risk calculators (CHA₂DS₂-VASc and HAS-BLED) have been validated to assess risk for ischemic stroke and major bleeding in AF patients, decisions about anticoagulation must consider the net clinical benefit (NCB) of anticoagulation. Furthermore, stroke and bleeding risk are highly correlated, making decisions more difficult.

Methods:

We examined patients in The Genetic and Environmental Risk Factors for Hemorrhagic Stroke (GERFHS III) Study, a population-based retrospective study of spontaneous ICH patients without a structural or traumatic cause in the Greater Cincinnati/Northern Kentucky region between 7/2008 and 12/2012. CHA₂DS₂-VASc and HAS-B(L)ED (minus “L” because “labile INR” was unavailable) scores were calculated for ICH patients with AF. Using a Markov state transition model, we estimated NCB of anticoagulation relative to no treatment in quality-adjusted life years (QALYs). We defined minimal clinically relevant benefit as 0.1 QALYs.

Results:

Among 1186 cases of spontaneous ICH, 95 cases had AF and met our survival criteria. Within one year, 8/95 (8%) would be expected to have a major bleeding event on anticoagulation, and 5/95 (5%) of patients would be expected to have an ischemic stroke off anticoagulation. 68/95 (71%) patients would have higher risk for major bleeding than for ischemic stroke. Anticoagulation with direct-acting anticoagulants (DOAC) would result in no clinically significant gain or loss in 73%. Roughly 12% would gain more than 0.1 QALYs and 15% would lose greater than 0.1 QALYs. Among patients receiving aspirin, most have no significant NCB or loss. Overall, anticoagulation of the entire cohort would result in an aggregate loss of 0.92 QALYs.

Conclusion:

Our analysis suggests that universal anticoagulation after ICH would be associated with a net loss of QALY. Additional factors should be considered before anticoagulating patients with AF after ICH.

Clinical Trial Registration Information:

NCT00930280

URL:

https://clinicaltrials.gov/ct2/show/NCT00930280

Introduction

Intracerebral hemorrhage (ICH) is a devastating disease, associated with a 30-day mortality ranging from 25% to 50%.^{1, 2} Patients with ICH may have risk factors for ischemic stroke as well, including atrial fibrillation (AF), which increases the risk of cardioembolic ischemic stroke 5-fold.³ Although the use of anticoagulation has been well-established to substantially lower ischemic event rates, the safety of starting or resuming anticoagulation after an ICH for thromboembolic prophylaxis is controversial.^{2, 4, 5}

Some authors have found the risk of recurrent bleeding, including ICH, with anticoagulation to be relatively low.⁶ For example, one meta-analysis suggested that resuming anticoagulation after ICH was associated with a decreased thromboembolic event rate, with no significant increase in recurrent ICH.⁷ As a result, some investigators have called for more widespread use of anticoagulation after ICH.⁸ However, other studies have found a significantly increased risk of recurrent ICH with anticoagulation.⁴

Several risk calculators have been developed and validated to assess risk for ischemic stroke and for major bleeding in AF patients with anticoagulation. The most widely used and well-validated scores include the CHA₂DS₂-VASc score, for predicting ischemic events, and the HAS-BLED score, for risk of bleeding with anticoagulation.^9–13 Yet some risk factors (age, hypertension, and prior stroke) contribute to both scores, making decisions about the best course of action challenging, especially for patients with both AF and ICH.

Electronic decision-support applications have been developed to assist clinicians in complex situations.^14–17 These tools use information from the patient’s electronic health record to evaluate the net clinical benefit or loss associated with alternative treatment decisions for individual patients. The Atrial Fibrillation Decision Support Tool (AFDST)^{14–16, 18–20} was developed by one of the current authors to guide decision making about anticoagulation therapy in patients with AF. In 2003, Eckman et al performed a decision analysis specifically focused on the question of anticoagulation with warfarin in AF patients with a prior history of ICH.²¹ The results of that analysis generally recommended against anticoagulation, particularly in patients with a history of prior lobar hemorrhage. Results were less clear for patients with a prior deep hemorrhage. However, clinical circumstances have changed significantly since 2003. Direct acting anticoagulants (DOACs) are now available with comparable efficacy to warfarin, but lower bleeding risk. Furthermore, new algorithms have become available to better stratify risk of both AF-related stroke and bleeding (CHA₂DS₂-VASc, HAS-BLED).

Therefore, we sought to describe the distribution of the CHA₂DS₂-VASc and HAS-BLED risk scores among a retrospective group of primary ICH patients with AF within a large, biracial population, to better understand their predicted risks of ischemic vs. hemorrhagic events. Furthermore, we used the decision analytic model forming the calculational engine of the AFDST to assess the net clinical benefit of anticoagulation with a DOAC by estimating gains or losses in quality-adjusted life years (QALYs) of anticoagulation strategies compared with no thromboprophylaxis in this same retrospective cohort.^{14, 20}

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request. The source dataset for the present analysis was drawn from the Genetic and Environmental Risk Factors in Hemorrhagic Stroke (GERFHS) III study, which identified all adult (age ≥20 years) cases of spontaneous primary ICH (excluded cases with trauma, brain tumor or vascular malformation) among residents of the five-county Greater Cincinnati/Northern Kentucky region who presented to 14 area hospitals from July 2008 through December 2012.²² Informed consent was not obtained, all data for the analysis was obtained from the hospital charts via HIPAA waiver. The definition for ICH used in this study is identical to that used for previous epidemiological stroke studies by our investigative group: non-traumatic abrupt onset of severe headache, altered level of consciousness, or focal neurological deficit associated with a focal collection of blood within the brain parenchyma as observed on CT or at autopsy and is not due to hemorrhagic conversion of a cerebral infarction.

Potential cases were identified by retrospective review of primary and secondary discharge ICD-9 codes 430–438.9 at all 16 adult hospitals in the study region, as well as by prospective admission diagnosis surveillance (“hot pursuit”) of emergency and radiology departments. Research nurses abstracted data for each potential case, including demographics, patient medical history, medications taken within two weeks of the ICH, stroke evaluation, diagnostic tests, and neuroimaging. The abstractions were reviewed in detail by study physicians, who made the final decision about case eligibility and assignment of location of ICH. ICH due to hemorrhagic transformation of ischemic stroke, trauma, brain tumor, or vascular malformation did not qualify for the GERFHS III study. The institutional review board for each participating hospital system approved the study.

Because the focus of the present analysis was to evaluate anticoagulation decisions in ICH patients with history of AF, cases were included only if (1) the patient had a history of AF prior to the ICH or developed AF during the acute hospitalization for ICH and (2) the patient was discharged alive, was not discharged to hospice, and survived at least 30 days after discharge. Several patients had more than one ICH during the study period; only the first event in the study period was included in the analysis.

CHA₂DS₂-VASc and HAS-BLED scores were calculated for all cases included in the analysis.^{23, 24} The variables used to compute CHA₂DS₂-VASc, from the patient’s medical history, include congestive heart failure (CHF; 1 point), hypertension (HTN; 1 point), age (65–74, 1 point; and ≥75, 2 points), diabetes (1 point), stroke/TIA/thromboembolism (2 points), vascular disease (1 point), and sex category (1 point if female). Variables used to compute HAS-BLED include poorly controlled HTN (1 point), abnormal renal (1 point), abnormal liver function (1 point), prior stroke (1 point), prior major bleeding or predisposition (1 point), elderly (age>65; 1 point), and concomitant drugs with antiplatelet effect (anti-platelets or NSAIDs, 1 point), and excessive alcohol usage (1 point). Data on labile INR, defined as <60% of time within goal INR range, was not available and therefore was not included in our calculations of HAS-BLED; the adjusted HAS-B(L)ED score, without labile INR, has been shown to have a similar predictive capability for bleeding risk.⁶ Published risk coefficients for each score were used to calculate predicted annual ischemic and bleeding events.^{10, 23} The AFDST also computes a separate estimate for the risk of ICH in patients not receiving anticoagulant therapy, based on a multivariate regression model developed on a Swedish registry of 90,490 untreated patients with AF.²⁵ It then calculates the annual rate of ICH while receiving anticoagulation by multiplying this base rate times the relevant relative risks for warfarin and the various DOACs.²⁶

The computational engine for the AFDST is a 29-state Markov state transition model that considers strategies of 1) no antithrombotic therapy, 2) aspirin, 3) warfarin, 4) dabigatran, 5) apixaban, 6) rivaroxaban, and 7) edoxaban. The decision model was developed at the University of Cincinnati, using a standard computer program (Decision Maker, Boston, MA) for model construction and analysis. Details regarding the AFDST and the underlying decision model have been described previously.¹⁴

We analyzed model recommendations for all members of the final cohort by generating a batch file containing all necessary values for clinical and demographic parameters used by the AFDST. We then used Decision Maker’s remote control function to run a script file containing patient-level information through the decision model. Results were stored to a text file which was loaded into a data base for further analysis. The AFDST recommends the strategy resulting in the largest gain in expected utility in QALYs. We used 0.1 QALYs as a minimum clinically significant gain to consider one strategy better than another. Predicted gain or loss was based on the oral anticoagulant with the greatest expected utility compared with no thromboprophylaxis. It is important to note, the event rates and risks we identify were not observed in the current study, but are only predicted.

Results

We identified 1,186 cases of spontaneous ICH between July 2008 and December 2012 among 1,151 residents of our population, of whom 232 had history of AF or new-onset AF. Among 664 individuals who met our survival criteria (discharged alive, not to hospice, and survived at least 30 days post discharge), 95 with AF were available for analysis.

Table 1 compares demographics, medical history, and anticoagulant/antiplatelet use between the 95 ICH survivors with AF versus the 569 ICH survivors with no history of AF. The ICH survivors with AF were significantly older than survivors without AF (mean age 77 vs. 67). In addition, higher proportions of ICH/AF survivors were on warfarin prior to their ICH (54.7% vs. 5.4%), and higher proportions of ICH/AF survivors had history of stroke, HTN, hyperlipidemia, CHF, or vascular disease, compared with those without AF. Lower proportions of ICH/AF patients were black, heavy users of alcohol, or users of street drugs, compared with those without AF. Of the 52 survivors on warfarin prior to admission, 25 (2 of whom had mechanical heart valves) had INR > 3.0.

Table 1.

Population Demographics and Comorbidities

	ICH pts with history of AF	ICH pts with no history of AF	p
N	95	569
Black, n (%)	12 (12.6%)	165 (29.0%)	0.01
Age, mean (SD)	76.9 (10.5%)	66.1 (15.2)	<0.01
Female, n (%)	43 (45.3%)	276 (48.5%)	0.56
History of stroke/TIA, n (%)	34 (35.8%)	112 (19.7%)	<0.01
Diabetes, n (%)	26 (27.4%)	134 (23.6%)	0.42
Hypertension, n (%)	85 (89.5%)	457 (80.3%)	0.03
Hyperlipidemia, n (%)	45 (47.4%)	190 (33.4%)	0.01
CHF history, n (%)	33 (34.7%)	56 (9.8%)	<0.01
Vascular disease history, n (%)	34 (35.8%)	115 (20.2%)	<0.01
Alcohol use, n (%)	(n=70)	(n=419)
Heavy (>2 servings per day)	1 (1.4%)	48 (11.5%)	0.03
Street drug use, n	2 (2.1%)	53 (9.3%)	0.02
Anticoagulant use prior to ICH
Warfarin, n (%)	52 (54.7%)	31 (5.4%)	<0.01
Heparin or LMWH, n (%)	4 (4.2%)	11 (1.9%)	0.25
Antiplatelet use prior to ICH
Aspirin, n (%)	40 (42.1%)	207 (36.4%)	0.29
Other antiplatelet, n (%)	6 (6.3%)	39 (6.9%)	0.85
Insurance status, n (%)
Medicare	79 (83.2%)	313 (55.0%)	<0.01
Abnormal renal function (GFR<15), n (%)	2 (2.1%)	11 (1.9%)	1.00
Abnormal liver function, n (%)	4 (4.2%)	32 (5.6%)	0.57

LMWH=low molecular weight heparin. CHF= Congestive heart failure. TIA= Transient ischemic attack. GFR= Glomerular filtration rate.

Table 2 shows a cross-tabulation of CHA₂DS₂-VASc and HAS-B(L)ED scores for the 95 ICH survivors with AF. According to the published risk estimates associated with each CHA₂DS₂-VASc and HAS-B(L)ED score, 68 of the 95 (71%) would have a higher risk for major bleeding than for ischemic stroke if on anticoagulation. It must be realized, however, that the long-term consequences of most major bleeds (other than ICH) are not as significant as those for ischemic strokes. Thus, this comparison alone is not enough to determine net clinical benefit.

Table 2.

Joint distribution of CHA₂DS₂-VASc and HAS-B(L)ED scores.

CHA2DS2-VASc score (ischemic risk)		HAS-B(L)ED score (bleeding risk)				Total	Predicted ischemic events within 1 year
		2	3	4	≥5
		1.88%	3.74%	8.70%	12.50%
0	0%		1			1	0.00
1	1.3%			2		2	0.03
2	2.2%	1	2	3	1	7	0.15
3	3.2%	1	4	10	3	18	0.58
4	4.0%		1	12	11	24	0.96
5	6.7%		8	9	5	22	1.47
≥6	9.5%*		9	9	3	21	2.00
Total		2	25	45	23	95	5.19
Predicted major bleeding events within 1 year		0.04	0.94	3.92	2.88	7.76

Bleeding risk is the expected percentage of patients associated with the HAS-B(L)ED score who will have major bleeding within one year (from Lip et al., Stroke 2010²³). Ischemic risk is the expected percentage of patients associated with the CHA₂D2_S-VASc score who will have ischemic strokes within one year (from Pisters et al., Chest 2010¹⁰). Predicted major bleeding is calculated as the total count times the bleeding risk for the score. Predicted ischemic strokes are calculated as the total count times the ischemic risk for the HAS-B(L)ED score. Yellow cells denote higher risk for ischemic stroke than for major bleeding. Red cells denote higher risk for major bleeding than for ischemic stroke.

^*Denotes an adjusted predicted thromboembolic rate to take into account the lower number of patients with CHA₂D2_S-VASc scores >6 (the total number of events for CHA₂D2_S-VASc scores 6–8 was divided by the person years and then divided by 0.36 to account for the assumption that warfarin reduces risk by 64%).

Figure 1 is a histogram showing the distribution of predicted gain/loss of QALYs across the cohort due to anticoagulation compared to no treatment. Based on the 0.1 QALY threshold for clinical relevance, 73% would have no clinically significant gain or loss, roughly 12% would gain more than 0.1 QALYs, and 15% would experience a loss of greater than 0.1 QALYs. It also must be noted that the maximal net gain or loss achievable in this population of patients is limited by their age and the competing mortality risks of their significant comorbidities (see Table 1). Figure 2 depicts a similar representation as Figure 1, although the comparison is based on aspirin for thromboprophylaxis, compared to no treatment. Very few patients achieve any net benefit from aspirin, and the majority receive either no clinically significant benefit or have a loss of QALYs.

Figure 1.

The histogram reports QALY scores of the population, with higher values denoting more patients. The comparison was between oral anticoagulant vs no thromboprophylaxis. n= 95. 12/95 (~ 13%) have a net gain of >0.1 QALYs.

Figure 2.

The histogram reports QALY scores of the population, with higher values denoting more patients. The comparison was between aspirin vs no thromboprophylaxis. n= 95. 4/95 (~ 4%) have a net gain > 0.1 QALYs.

Table 3 displays results of the decision model estimates for aggregate net clinical benefit or loss for the entire cohort, for each combination of CHA₂DS₂-VASc and HAS-B(L)ED scores, for anticoagulation with a DOAC compared with no thromboprophylaxis. The largest aggregate gain (2.31 QALYs) was estimated among patients with a CHA₂DS₂-VASc of ≥6 and a HAS-B(L)ED of 4. The second highest gain (1.67 QALYs) was among patients with a CHA₂DS₂-VASc of 5 and a HAS-B(L)ED of 3. Overall, an aggregate loss of 0.92 QALYs (336 quality-adjusted life days), was estimated over the entire cohort were they to receive anticoagulation therapy. Among the 95 patients in the analysis, 43 would potentially benefit from anticoagulation, with an aggregate gain of 6.06 QALYs (mean gain of 0.141 QALYs per patient, as shown in supplemental table I), whereas the remaining 52 patients would have an aggregate loss of 6.98 QALYs (mean loss of 0.134 QALYs, supplemental table I) were they to receive anticoagulant therapy.

Table 3.

Aggregate Gain/Loss from Best Antithrombotic Therapy Compared with No Thromboprophylaxis.

	HAS-B(L)ED
CHA2DS2VASc	2	3	4	≥5	Total
0		−0.8168			−0.8168
1			−0.7138		−0.7138
2	−0.3467	−0.5561	−0.8591	−0.2199	−1.9818
3	−0.0464	−0.1827	−0.6259	−0.4717	−1.3267
4		0.0453	−0.4336	−0.9134	−1.3017
5		1.6657	0.1069	0.0569	1.8295
≥6		0.8455	2.3139	0.2277	3.3871
	−0.3931	1.0009	−0.2116	−1.3204	−0.9242

Heat map showing gains or losses in QALYs. Green denotes a gain in QALY, darker green meaning higher gains. Yellow denotes gains or losses of smaller magnitude, while orange denotes loss of QALYs, darker orange meaning higher losses.

The supplemental tables are attached for reference. Supplemental table I shows the average gain/loss of QALY with each CHA₂DS₂-VASc/HAS-B(L)ED combination. Supplemental table II shows the risk of ICH based on the associated CHA₂DS₂-VASc/HAS-B(L)ED value.

Discussion

Within our population of ICH survivors, we predict that a majority of patients with AF would likely not gain any clinically significant net benefit from anticoagulation therapy. Although in aggregate the cohort lost slightly more than 0.92 QALYs from anticoagulation therapy, there were subgroups of patients who had a predicted clinically significant gain (> 0.1 QALYs), namely those with a CHA₂DS₂-VASc score of ≥6 and a HAS-B(L)ED score less than 5 (see supplement). It is important to note that the morbidity and mortality associated with recurrent ICH is the major determinant of net clinical benefit or loss associated with resuming anticoagulation. Since the HAS-BLED has only modest predictive power for ICH²⁷, the AFDST separately predicts ICH risk based on a separate cohort of AF patients. As shown in supplemental table II, for a large proportion of ICH survivors in our cohort, the risk of ICH exceeded the risk of AF-related ischemic stroke. The cells in which ICH risk is less than that of AF-related stroke, are the same cells in which the average gain from anticoagulation is greater than 0.1 QALYs.

Our findings highlight the importance of discovering additional clinical markers, be they radiologic, serologic, or genomic^{28, 29} that better predict major hemorrhage with anticoagulation, particularly ICH. Our results are fully concordant with current American Stroke Association guidelines for the management of intracerebral hemorrhage, which do not recommend anticoagulation post-ICH.³⁰ While our previously published decision analysis focusing specifically on patients with a past history of ICH found that anticoagulation was not warranted, that analysis did not explore specific subgroups of patients stratified by either CHA₂DS₂-VASc or HAS-BLED scores.²¹ That analysis however, did explore separately patients with a history of prior lobar versus deep ICH. Because the AFDST is a general tool for decision support of AF patients and not specifically focused on patients with prior ICH, it does not have the same granularity regarding location of prior ICH, and is thus unable to differentiate ICH risk based on location of past ICH. Furthermore, the multivariate regression model by Friberg et al we used to determine ICH risk, does not contain location of prior ICH as a covariate.²⁵

The decision to resume or initiate anticoagulation for AF following an ICH must depend on the balance of risk versus benefit for those patients. Subjective decisions by physicians are likely driving current practice. Unfortunately, descriptive studies of outcomes following resumption of oral anticoagulation in survivors of ICH likely have been subject to selection bias, since healthier patients tend to be better candidates for anticoagulation, likely skewing results towards better outcomes in those receiving treatment. Using objective risk prediction instruments such as the CHA₂DS₂-VASc and HAS-B(L)ED can help guide and objectify the complex decision regarding which patients should resume anticoagulation. However, it must be kept in mind that the long-term consequences of major extracranial bleeding events are generally more favorable than those of AF-related ischemic stroke. Furthermore, the HAS-B(L)ED algorithm is not a good predictor of ICH.²⁵ Finally, bleeding risk and ischemic stroke risk are highly correlated, thus many patients with elevated CHA₂DS₂-VASc scores also have high HAS-B(L)ED scores. Thus, in our analysis many of the patients at highest risk for ischemic stroke, who one might expect to benefit most from anticoagulation, were also at high risk of major hemorrhage and ICH, limiting the net benefit of anticoagulation.

Another challenge for analyses of ICH outcomes is the wide variation in rates of ICH recurrence quoted in the literature. Some investigators have posited rates as low as 1–2%/yr^{4, 6, 31} while others note rates as high as 15%/yr.²¹ We calculated each patient’s risk of recurrent ICH based on a multivariate regression model developed on a population of more than 90,000 non-anticoagulated AF patients and then applied relative hazards of ICH for each anticoagulant based on best evidence in the literature.²⁵ There are numerous confounders that can account for the wide variation in estimates for recurrent ICH, including location of hemorrhage (lobar vs deep), age and prior microbleeds (likely markers for cerebral amyloid angiopathy [CAA]), hypertensive status and apolipoprotein E epsilon 2 or 4 carrier status. Studies have shown that lobar hemorrhages have the highest risk of recurrence, and are most often associated with CAA^{22, 32}, making these patients a group to avoid when resuming anticoagulation. However, given uncertainties in the quoted recurrence rates for ICH, population-based estimates are needed that do not have survival or selection biases inherent in single-center or academic institutional reports.

Our analysis had several limitations. We were unable to ascertain INR values prior to the index event, potentially limiting the accuracy of the HAS-BLED score. However, a recent study demonstrated the validity of the HAS-B(L)ED score that does not contain information about labile INR.⁶ The CHA₂DS₂-VASc and HAS-B(L)ED are both limited by fewer numbers of patients at the highest risk levels, and therefore may be less accurate at extreme scores. Our analysis predicted net benefit/loss for anticoagulation vs. no therapy and aspirin vs. no therapy but we did not present comparisons of anticoagulation vs. aspirin. In light of the poor efficacy of aspirin in preventing AF-related stroke, and almost comparable rates of major bleeding while taking aspirin or the DOACs, we did not calculate these results. As mentioned previously, the variability of ICH recurrence rates in the literature could mean that we have over- or under-estimated bleeding risk. Additionally, the reported values for outcomes in our study are only predicted outcomes, and not actually observed events. Therefore, they should be interpreted with caution. Finally, our analyses and results do not account for the benefit aspirin may provide for other indications such as coronary artery disease.

Our study benefited from having 95 ICH patients with AF for whom we had sufficient clinical information to calculate CHA₂DS₂-VASc and HAS-B(L)ED scores. Thus, we were able to perform individual decision analyses on each of these patients using the AFDST. To our knowledge, this is the first study to characterize AF patients post ICH by CHA₂DS₂-VASc and HAS-B(L)ED scores, and is certainly the first study to perform individualized decision analyses on such patients to estimate the potential net benefit or loss associated with resumption of oral anticoagulation. An NIH-funded prospective trial plans to evaluate this question, but until then, decision analysis may be the best tool clinicians can use to guide decisions for this complex clinical dilemma. Including patients’ values and preferences in these calculations, and weighing individual patient’s willingness to accept certain risks for possible gains, may add further benefit to shared decision-making in this complicated clinical scenario.