Risk of stroke in chronic heart failure patients with preserved ejection fraction, but without atrial fibrillation: analysis of the CHARM-Preserved and I-Preserve trials

Aims The incidence and predictors of stroke in patients with heart failure and preserved ejection fraction (HF-PEF), but without atrial fibrillation (AF), are unknown. We described the incidence of stroke in HF-PEF patients with and without AF and predictors of stroke in those without AF. Methods and results We pooled data from the CHARM-Preserved and I-Preserve trials. Using Cox regression, we derived a model for stroke in patients without AF in this cohort and compared its performance with a published model in heart failure patients with reduced ejection fraction (HF-REF)—predictive variables: age, body mass index, New York Heart Association class, history of stroke, and insulin-treated diabetes. The two stroke models were compared and Kaplan–Meier curves for stroke estimated. The risk model was validated in a third HF-PEF trial. Of the 6701 patients, 4676 did not have AF. Stroke occurred in 124 (6.1%) with AF and in 171 (3.7%) without AF (rates 1.80 and 1.00 per 100 patient-years, respectively). There was no difference in performance of the stroke model derived in the HF-PEF cohort and the published HF-REF model (c-index 0.71, 95% confidence interval 0.57–0.84 vs. 0.73, 0.59–0.85, respectively) as the predictive variables overlapped. The model performed well in the validation cohort (0.86, 0.62–0.99). The rate of stroke in patients in the upper third of risk approximated to that in patients with AF (1.60 and 1.80 per 100 patient-years, respectively). Conclusions A small number of clinical variables identify a subset of patients with HF-PEF, but without AF, at elevated risk of stroke.


Introduction
Up to half of patients with heart failure have a preserved ejection fraction (HF-PEF). [1][2][3] These patients differ from heart failure patients with reduced ejection fraction (HF-REF) in several aspects-they tend to be older, are more often women, and are more likely to have a history of hypertension and atrial fibrillation (AF); they are less likely to have coronary artery disease. Although, mortality rates may not be as high as in patients with HF-REF, the prognosis of HF-PEF patients is considerably worse than that of patients with hypertension, angina pectoris, AF, or diabetes in the same age range and gender distribution. 4 The single most common cause of hospital admission in these patients is worsening heart failure and this, along with death, has been the focus of therapeutic interventions in HF-PEF. 4 However, given the demographic profile and co-morbidity cluster characterizing these patients, stroke may also be a clinically important outcome in HF-PEF. Little is known about the incidence of stroke in HF-PEF, particularly in the absence of AF. To investigate this further, we therefore combined and analysed patient-level data from two large HF-PEF trials, the Candesartan in Heart failure Assessment of Reduction in Mortality and Morbidity-Preserved trial (CHARM-Preserved, ClinicalTrials.gov NCT0 0634712) 5 and the Irbesartan in Heart Failure with Preserved Systolic Function trial (I-Preserve, NCT00095238), 6 to provide a robust estimate of the current incidence of stroke in patients with HF-PEF, with and without AF. We also tested a simple clinical model, developed in HF-REF, 7 for predicting the risk of stroke in patients without AF in this pooled dataset. Easy identification of those at highest risk of stroke coupled with the availability of new oral anticoagulants with a low risk of bleeding might allow for a stroke prevention strategy which has an acceptable benefit/risk balance in patients with HF without AF.

Trial patients
In order to have a sufficiently large number of HF-PEF patients without AF for analysis, we pooled data from the CHARM-Preserved (NCT00634712) and I-Preserve (NCT00095238) trials. Each was a randomized, double-blind, placebo-controlled, multicentre trial and was approved by the appropriate institutional review boards. CHARM-Preserved and I-Preserve enrolled 3023 and 4128 patients, respectively. 5,6 Together, these trials included a broad spectrum of patients with chronic HF-PEF.
CHARM-Preserved enrolled patients aged !18 years in New York Heart Association (NYHA) functional class II-IV with a left ventricular ejection fraction (LVEF) >40% (although for the purposes of this study we included only patients with an LVEF !45%). I-Preserve enrolled patients aged !60 years in NYHA functional class II-IV with an LVEF !45% and corroborating ECG, echocardiographic or radiologic evidence. In addition, patients must have been hospitalized for heart failure in the preceding 6 months or, if not, had to be in NYHA functional class III or IV. N-terminal pro B-type natriuretic peptide (NT-proBNP) was measured at baseline in I-Preserve but not in CHARM-Preserved. In CHARM-Preserved, patients were randomly assigned to candesartan (target dose of 32 mg once daily) or matching placebo. 5 In I-Preserve, patients were randomized to irbesartan (target dose 300 mg once daily) or matching placebo. 6 The primary outcome in CHARM-Preserved was the composite of cardiovascular death or HF hospitalization 5,8 and in I-Preserve it was the composite of all-cause mortality or cardiovascular hospitalization. 6,9 The median follow-up in CHARM-Preserved was 3.1 years and in I-Preserve it was 4.1 years. Study treatment did not reduce the risk of the primary outcome or the risk of stroke in the either trial. 5,6 Incident stroke Incident strokes were centrally adjudicated by an independent endpoint committee in each trial using similar definitions and stroke was part of the primary or secondary composite cardiovascular outcomes in both trials. 5,6,8,9 Stroke in both trials was defined as a persistent (!24 h) disturbance of focal neurological function resulting in symptoms thought to be due to cerebral infarction, evidence of haemorrhage or for which there is no certain aetiology. 5,6,8,9 Incident atrial fibrillation The occurrence of AF was retrospectively collected in CHARM-Preserved during the trial close-out using a specifically designed casereport form. Incident AF was recorded prospectively in I-Preserve, using a specific case-report form.

Statistical methods
We included only patients with an LVEF of !45% (all 4128 patients in I-Preserve and 2573 of the 3023 in CHARM-Preserved). Patients with AF were defined as those with either AF confirmed on their baseline ECG or a history of AF. The remaining patients were defined as those 'without AF'. Descriptive statistics were used to describe the pooled patient population from both trials and to compare these two subgroups, using means (standard deviation) or medians [interquartile range (IQR)] for continuous variables and count (percentage) for categorical variables.
The incidence rate of stroke (per 100 patient-years) was calculated over the trial follow-up period and was compared among the AF and no AF subgroups. We plotted Kaplan-Meier (KM) curves for the occurrence of stroke, according to AF status. To satisfy the assumption of the independence of stroke events, recurrent stroke events in a patient after randomization were not included in the analysis.
Continuous variables [e.g. body mass index (BMI), ejection fraction, and creatinine level] were assessed by visual inspection of restricted cubic splines to identify potential non-linear effects. Uniand multivariable predictors of the risk for stroke were evaluated using Cox proportional hazards regression analysis in patients without AF. Two separate multivariable analyses for stroke were created. First, an 'HF-PEF stroke model' was created using established predictors of ischaemic stroke [10][11][12][13][14][15] with the addition of variables that were significant (P < 0.05) in univariable analysis of our dataset. The final list of variables included was age, sex, LVEF, NYHA class III/IV, BMI, creatinine level, systolic blood pressure, history of stroke, hypertension, and diabetes treated with insulin. Second, we applied a recently published multivariable predictive model for stroke in patients with HF-REF (HF-REF stroke model) in our HF-PEF cohort. 7 The five variables included in this model were age, BMI, NYHA class, history of stroke, and diabetes treated with insulin. There were no data missing for the baseline variables used either model. We calculated the hazard ratio and corresponding 95% confidence intervals (95% CI) to Stroke in HF-PEF express the hazard rate of stroke. The statistical contribution of each variable to the predicted risk of stroke was assessed by the v 2 statistic. In order to be consistent with our previous publication, 7 we compared each model's discrimination ability using estimates of overall c-index for the Cox regression models according to the method of Pencina and D'Agostino, 16 as outlined by Liu et al. 17 We predetermined that we would proceed using only the HF-REF stroke model if the overall c-indexes for the two models were not meaningfully different.
The coefficients from statistically significant variables in the final multivariable model were used to calculate an individual patient's risk score for stroke. The KM curves for occurrence of stroke according to tertiles of risk score were plotted.
Final model calibration and the ability to separate patients into risk groups were assessed by observing predicted compared with observed outcomes in terciles, and by using the Hosmer-Lemeshow goodness-of-fit test. The model's discrimination abilities were evaluated by the overall c-index. 16,17 We also conducted a sensitivity analysis to compare the cumulative incidence function for stroke estimated using competing risk technique (to account for the competing risk of death) 19,20 with the rates of stroke described from the traditional KM curves above. We also compared the overall c-index of the model with the traditional Harrell's c-statistic. [16][17][18] Finally, we validated the preferred risk model in a third HF-PEF trial: the Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist (TOPCAT) (NCT00094302). 21 TOPCAT included patients aged !50 years with at least one symptom and sign of heart failure, an LVEF !45% and either a hospitalization with heart failure in the preceding 12 months or an elevated NT-proBNP or BNP.

Results
Of the 6701 patients with an LVEF !45%, 2025 (30%) had a history of AF or AF on their baseline ECG and 4676 patients (70%) had no AF.

Baseline characteristics
The baseline characteristics of patients with and without AF are shown in the Supplementary material online, Table S1. The baseline characteristics of patients without AF, according whether or not they experienced a subsequent stroke, are shown in Table 1.

Patients with and without atrial fibrillation
Patients without AF were younger and were more likely to have a history of coronary artery disease and hypertension, compared with patients with AF. Patients without AF also had a slightly higher systolic blood pressure but had a lower mean serum creatinine and much lower median NT-proBNP level than patients with AF. There were also notable differences in medical therapy, particularly in use of antiplatelet therapy (69% of patients without AF vs. 39% of those with AF) and anticoagulant treatment (6% vs. 57%, respectively), but also in relation to diuretics, mineralocorticoid receptor antagonists, antiarrhythmic agents, and digoxin.
Patients without atrial fibrillation-with and without incident stroke during follow-up Among patients without AF, those who experienced a stroke (compared with those who did not) were older, more likely to have a history of diabetes, hypertension, and stroke and had worse NYHA functional class. Patients experiencing stroke also had a higher systolic blood pressure, creatinine, and NT-proBNP level. Compared with those not experiencing stroke, those who did were less likely to be treated with lipid lowering therapy but more likely to be taking nitrates, anti-platelet therapy, and insulin. Very few patients in either group were treated with an oral anticoagulant (263 in total, 6%). LVEF did not differ between patients with and without stroke.

Incident AF and risk of stroke
In CHARM-Preserved, 1781 patients did not have AF at baseline. Out of 1781, 59 patients (3.3%) experienced a stroke. Of these 59 patients, 10 (17%) developed new AF before the occurrence of their stroke; the number of patients with a stroke without preceding AF was 49 (83%). Development of AF reported was not reported in any patient following a stroke.
In I-Preserve, 2895 patients did not have AF at baseline. Out of 2895, 112 patients (4%) experienced a stroke. Of these 112 patients, 18 (16%) developed new AF before the occurrence of their stroke; the number of patients with a stroke without preceding AF was 94 (84%). Twenty patients (18%) with an incident stroke had new AF reported before or after their stroke.
Predictors of stroke in patients with heart failure and preserved ejection fraction without AF Figure 2 and Supplementary material online, Table S2 show the relationship between baseline variables and risk of stroke (univariable analysis). Supplementary material online, Table S3 shows an adjusted analysis using the four independent predictors identified in a multivariable stroke model developed in the present HF-PEF cohort   Figure 3 shows the distribution of the risk score for stroke and illustrates the risk of stroke for a given score. A score of approximately 12 predicts a risk of stroke similar to that which was seen among patients with AF in the current cohort. Figure 4 shows KM curves for stroke with patients classified into three equal-sized groups according to risk score. The number of strokes in tertiles 1, 2 and 3 were 37, 45 and 89, respectively. The Figure 5 shows the model's goodness-of-fit by comparing observed and expected probabilities of stroke at 3 years with the patients divided into tertiles. The calibration was also assessed using the Hosmer-Lemeshow test, which was P ¼ 0.761.

Validation of stroke risk model
We tested the predictive model in TOPCAT, which included 1240 patients with and 2205 patients without AF. The mean follow-up was 3.5 years. There were 65 strokes in the patients with AF and 52 strokes in those without AF, giving stroke rates in patients with and without AF 1.64 and 0.71 per 100 patients-years, respectively.
Using the same analytical approach (see Supplementary material online, Table S4, Figures S2 and S3), the 1, 2 and 3 year KM rates of stroke in patients without AF, in the two higher risk tertiles were: ter-    Table S5.

Discussion
In this analysis, HF-PEF patients with AF were at a high risk of stroke, with an average incidence rate of 1.8% per year which is similar to that recently reported in HF-REF patients with AF (1.6% per year). 7 HF-PEF patients without AF in this study had a lower risk of stroke compared with those with AF. However, the overall rate of stroke in HF-PEF patients without AF (1.0% per year) was similar to the rate we recently reported in HF-REF patients without AF (1.2% per year). Moreover, as in HF-REF, a small number of demographic and clinical variables identified a subset of HF-PEF patients without AF who were at greater risk of stroke than the remainder. Specifically, in our pooled analysis, patients in the upper third of the risk score had a rate of stroke (1.6% per year) which was higher than in HF-PEF patients with AF receiving an anticoagulant (1.5% per year), although not as high as in similar patients not treated with an anticoagulant (2.2% per year).
We have been unable to find other reports of the risk of stroke in HF-PEF patients without AF although in patients in the same age range in clinical trials for hypertension (i.e. with a similar co-morbid phenotype to HF-PEF) have a stroke risk of around 1% per year or less. [22][23][24][25][26] In HF-PEF patients with AF randomized to warfarin in ARISTOTLE 27 the rate of stroke was 1.4% per year which was similar to the rate in anticoagulant-treated AF patients in our study (1.5% per year). In AF patients with HF and an LVEF >40% in RELY-AF 28 the rate of stroke or systemic embolism was 2.07% per year in the warfarin group; in ROCKET-AF 29 the rate of the same outcome in similarly defined patients was 2.06% per year. The higher event rates in the latter two trials are due to broader composite outcome (which included noncerebral systemic embolism) and the requirement for patients in these trials to have additional risk factors for stroke.
The similar risk of stroke in patients with HF-PEF and HF-REF, without AF, is also of interest. We previously reported that LVEF was not predictive of stroke in HF-REF patients without AF. Neither was LVEF an independent predictor of stroke risk in this study although we examined only patients with an LVEF !45%. This finding is consistent with observations in three recent trials comparing non-Vitamin K antagonist oral anticoagulants with warfarin in patients with AF. In those trials, the risk of stroke and systemic embolism was similar, irrespective of LVEF category, in patients with AF and concomitant HF. A similar conclusion was reached by the Atrial Fibrillation Clopidogrel Trial with Irbesartan for Prevention of Vascular Events (ACTIVE) in AF patients not treated with an oral anticoagulant where the risk of stroke was similar in patients with concomitant HF-REF or HF-PEF. 30 As in HF-REF, we found that neither systolic blood pressure nor history of hypertension was independent predictor of stroke. Although this contrasts with the findings in other patient cohorts, it is consistent with the 'reverse epidemiology' of heart failure and the known association between higher blood pressure and better outcomes in this condition. [31][32][33] Likewise, we saw an association between lower BMI and higher risk of stroke, another feature of the 'reverse epidemiology' in heart failure. [31][32][33] A particular strength of this study is the validation of our predictive model in another dataset (TOPCAT). Consequently, our findings have clear clinical implications. With a small number of routinely collected clinical variables it is possible to identify patients with HF-PEF, but without AF, who may be at sufficiently high risk of stroke potentially to justify anticoagulation. Clearly, there is as yet no trial evidence to justify such treatment but our findings suggest a means of identifying patients for such a trial. Consistent with this hypothesis, prior trials in patients with heart failure and reduced ejection fraction collectively suggest that anticoagulation can reduce the risk of stroke in patients in sinus rhythm. However, in the largest of these, the Warfarin vs. Aspirin in Reduced Cardiac Ejection Fraction trial (WARCEF), although warfarin was effective in reducing ischaemic stroke this benefit was offset by major bleeding. With non-vitamin K oral anticoagulants, the risk-to-benefit balance might be more favourable, especially as the target International Normalized Ratio (INR) in WARCEF was 2.75 (range 2.0-3.5). [34][35][36][37] Limitations Each of the two trials included had specific inclusion and exclusion criteria and, hence, our findings may not be generalizable to all patients with HF-PEF. Notably, few patients were in NYHA class IV and worse functional class was a predictor of higher risk of stroke. Hence, the risk of stroke may be higher in 'real world' patients than in the cohort studied. Although our data suggest that only the minority of strokes is related to incident AF, systematic detection of new onset AF was insensitive, e.g. continuous ambulatory monitoring was not performed. It is widely recognized that silent AF is frequent in heart failure and undetected AF may have accounted for more strokes than realized. However, waiting for the development of clinically recognized AF before employing anticoagulant therapy may not be the ideal preventive strategy and the best and most cost-effective way to screen for silent AF in HF-PEF is unknown. In addition, these patients may have other reasons to develop thromboembolic and other types of ischaemic stroke, e.g. endothelial dysfunction and blood stasis. Therefore, we believe that our findings support a potential preventive role for anticoagulant therapy in HF-PEF patients in sinus rhythm, particularly as new agents with a lower risk of bleeding are available. Of course, this hypothesis needs to be tested prospectively in a randomized trial and it may be too simplistic to assume that an anticoagulant can substantially reduce the risk of stroke in those with HF-PEF at highest risk.
CHARM-Preserved and I-Preserve were randomized controlled heart failure trials, rather than stroke trials, and used definitions of stroke consistent with other heart failure trials conducted during the same period. Although the definition may not be identical to that used in contemporary stroke trials, it was applied consistently by adjudicators blind to treatment allocation and thus gave an unbiased estimate of treatment effect. Unfortunately, classification of stroke subtype was not carried out in both trials. When the trials were conducted, neuroimaging was not standard in patients with suspected stroke in many, if not most countries, involved. Therefore, we are unable to distinguish between ischaemic and haemorrhagic strokes.
In conclusion, we found that a relatively high-risk subset of a third of HF-PEF patients without AF have a risk of stroke similar to that in HF-PEF patients with AF. This higher-risk subset can be identified using five simple clinical variables. The risk of stroke is similar in HF-PEF and HF-REF patients without AF and is predicted by the same variables. The risk of stroke in these patients might be reduced by treatment with an oral anticoagulant but this hypothesis needs to be tested in a clinical trial. The rate of stroke in the highest risk tertile was not quite as high as in patients with AF not treated with an anticoagulant so it is uncertain what the benefit/risk ratio of such treatment might be.