Introduction

In patients with chronic heart failure (CHF), the activation of the sympathetic nervous system and the renin–angiotensin–aldosterone system induces a hypercoagulable state, increased aggregation of thrombocytes, and reduced fibrinolysis [1, 2]. Moreover, CHF patients have endothelial dysfunction, malfunction of cerebral autoregulation and rheological alterations consistent with flow abnormality via low cardiac output and aberrant flow through dilated cardiac chambers [3–5]. Thus, CHF is associated with an increased risk of thrombus formation [6] and is accompanied by a two to threefold increased risk of stroke [7]. In the clinical setting, stroke in CHF patients is associated with poor outcomes and higher mortality [7]. Current guidelines recommend anticoagulation for CHF patients with concomitant atrial fibrillation (AF) but not for those in sinus rhythm [8]. It is clinically relevant to stratify the risk of stroke in CHF patients in sinus rhythm.

The CHADS2 score and CHA2DS2-VASc score are proposed as a useful way to stratify the risk of ischemic stroke or transient ischemic attack (TIA) in patients with AF [9, 10]. These scores were reported to predict ischemic stroke in the absence of atrial fibrillation among patients with coronary heart disease [11]. However, it remains unclear whether the CHADS2 score could predict ischemic stroke or TIA in CHF patients without AF. Recently, the new stroke risk score has been proposed to identify CHF patients without AF who were at high risk of ischemic stroke from 2 large and contemporary HF trials, the controlled rosuvastatin in multinational trial heart failure (CORONA) and the Gruppo Italiano per lo Studio della Sopravvivenza nell’Insufficienza cardiac-heart failure trial (GISSI-HF) [12]. This new “stroke risk score” included the following independent predictors of stroke: age, New York Heart Association (NYHA) functional class, diabetes mellitus with insulin, body mass index and prior stroke. The aim of the present study is to evaluate the prognostic power of the CHADS2 score for ischemic stroke or TIA in CHF patients without AF in comparison to CHA2DS2-VASc score and the new “stroke risk score”.

Methods

We retrospectively studied 127 consecutive stable CHF outpatients with a radionuclide left ventricular ejection fraction (LVEF) <40 % and without a history of AF, who had been enrolled in our previous prospective cohort study from October 1995 to October 1998. CHF was diagnosed by clinical signs and symptoms according to the Framingham Heart Study criteria. These criteria require the presence of at least two major criteria or one major criterion in addition to two minor criteria [13] to confirm heart failure. To be included in the present study, all patients who had experienced at least one episode of decompensated heart failure were required to be stable for 3 months on conventional therapy. Patients were excluded from the present study if they had significant renal (serum creatinine level >3.0 mg/dl) or hepatic dysfunction (aspartate transaminase or alanine transaminase >three times of upper normal limits). The mean patient age was 64 ± 12 years. Of the 127 patients, 97 were men and 30 were women. CHF was due to ischemic heart disease in 76 patients and idiopathic dilated cardiomyopathy in 51. The average NYHA functional class was 2.0 ± 0.7, with 25 % of patients categorized as class I, 54 % of patients categorized as class II, and 20 % categorized as class III. The mean radionuclide LVEF was 30 ± 7 %. All patients gave a written informed consent for their participation in this study, which was approved by the Osaka General Medical Center’s Review Committee.

At entry, we calculated the CHADS2 score, CHA2DS2-VASc score and “stroke risk score” from baseline clinical characteristics. The CHADS2 score is derived from the sum of point values of individual stroke risk factors [CHF, hypertension, age ≥75, diabetes mellitus (1 point each), and prior stroke or TIA (2 points)]. The CHA2DS2-VASc score is derived from the sum of point values of individual stroke risk factors [CHF, hypertension, age 65–75, diabetes mellitus, vascular disease and female sex (1 point each), and prior stroke or TIA and age 75 or older (2 points)]. Hypertension was defined by either self-report or systolic blood pressure ≥140 mmHg; blood pressure was measured in all participants in the sitting position after 5 min of rest. Diabetes mellitus was defined as self-reported diabetes, use of a diabetes medication, or hemoglobin A1c ≥6.5 %. Prior stroke and TIA were determined by self-report. Vascular disease included prior myocardial infarction, peripheral artery disease and aortic plaque in patients’ history.

The “stroke risk score” was obtained by 5 individual stroke risk factors (age, NYHA class, diabetes mellitus on insulin, body mass index and prior stroke), which were significant predictors of stroke in Cox multivariate proportional hazard regression analysis among patients included in CORONA and GISSI-HF trials. The “stroke risk score” was calculated by multiplying age (per 10 years increase) by 3.31, BMI (per 5 kg/m2 increase up to 30) by −3.01 and then adding 4.72 in the case of NYHA III or IV, 6.26 when patients had diabetes mellitus treated with insulin and 5.91 when patients had the prior stroke [12].

All patients underwent ECG-gated blood-pool scintigraphy with a conventional rotating gamma camera equipped with a low-energy, high-resolution, parallel-hole collimator. Patients were given 740 MBq of technetium-99 m-labeled human serum albumin (Nihon Medi-Physics, Nishinomiya, Japan). LVEF was calculated with a standard program [14]. In addition, all patients underwent echocardiography and 24-h ambulatory electrocardiographic monitoring, and a venous blood sampling. In echocardiography, left ventricular end-diastolic dimension, left ventricular end-systolic dimension, and left atrial dimension were measured by standard techniques [15]. In 24-h ambulatory electrocardiographic monitoring, ventricular arrhythmias were classified according to the Lown’s grade. Blood sampling for assessment of serum creatinine, sodium, albumin levels, and plasma noradrenaline concentration, was drawn from an intravenous cannula after the patients had rested for more than 30 min in the supine position. Estimated glomerular filtration rate was calculated by the Modification of Diet in Renal Disease formula. Plasma norepinephrine concentration was determined in ethylenediaminetetraacetic acid-plasma by high-performance liquid chromatography [16] at Shionogi Biomedical Laboratories (Osaka, Japan). A duplicate determination in the laboratory showed a coefficient of variation of 0.4–5.5 %.

All study patients were followed in the heart failure unit of our hospital at least every 1 or 2 months. The primary endpoint of this study was the incidence of ischemic stroke or TIA, which was determined by the review of medical records performed by 2 independent and blinded physicians. If the 2 physicians agreed on the outcome classification, their classification was binding. In the event of a disagreement, a third blinded physician was consulted.

Stroke was defined as a new neurologic deficit not known to be secondary to brain trauma, tumor, infection, or other cause, based on the WHO MONICA criteria [17]. All stroke outcomes were subtyped as hemorrhagic, ischemic, or procedure-related, based on physician diagnosis, which was confirmed by computed tomography or magnetic resonance imaging. Stroke outcomes in this study were restricted to patients with non-procedure-related ischemic strokes. TIA outcomes were based on the clinical judgment of the physicians who were blinded to the medical records, guided by the definition of TIA as a focal neurologic deficit (in the absence of head trauma) lasting more than 30 s and no longer than 24 h, with rapid evolution of the symptoms to the maximal level of deficit in less than 5 min and with subsequent complete resolution [11, 18].

The Student’s t test was used to compare differences in continuous variables, and the data were presented as mean ± standard deviation, if the data was normally distributed. If the data was not normally distributed, Mann–Whitney U test was used and the data were presented as median (interquartile range). The Fisher’s exact test was used to compare differences in categorical variables. One-way analysis of variance followed by a post hoc Scheffé test was applied to find differences among low, intermediate, and high CHADS2 score groups. Cumulative rates of events were calculated using the Kaplan–Meier method. Comparison of event-free survival rates between groups was assessed with a 2-sided log-rank test. Receiver-operating characteristic (ROC) curves and areas under the curve (AUC) analysis was performed to further explore the discriminatory ability of CHADS2 score, CHA2DS2-VASc score and “stroke risk score” for ischemic stroke or TIA. Cox proportional hazards regression models were used to identify patients at risk of ischemic stroke or TIA and to calculate the multivariate-adjusted hazard ratios (HRs) and 95 % confidence interval (CI) of the parameters. The C-index was used to measure how well the model discriminated between patients with high and low risk of ischemic stroke or TIA. A value of 0.5 for C-index indicates no discrimination and a value equal to 1 indicates perfect discrimination. All data were statistically analyzed using StatView version 5 (SAS Institute, Cary, North Carolina), except for ROC analysis and the measurement of C-index, which was analyzed using EZR version 1.03 (Saitama Medical Center, Jichi Medical University, Saitama, Japan) [19]. A p value <0.05 was considered statistically significant.

Results

At the baseline, there were 38, 51, 30, 2, 5, and 1 patients with a CHADS2 score of 1, 2, 3, 4, 5, and 6, respectively. The mean value of CHADS2 score and CHA2DS2-VASc score were 2.1 ± 1.0 and 3.7 ± 1.4, respectively. The mean “stroke risk score” was 4.8 (1.8–10.9). During a follow-up period of 8.4 ± 5.1 years (range 0–18), ischemic stroke or TIA was observed in 21 of 127 CHF patients (stroke in 17 and TIA in 4 patients), which was confirmed by computed tomography or magnetic resonance imaging in 86 % (18/21) of cases.

Baseline characteristics in CHF patients with and without ischemic stroke or TIA are shown in Table 1. Patients with ischemic stroke or TIA had a significantly higher incidence of hypertension and diabetes mellitus, a greater CHADS2 score, CHA2DS2-VASc score and a higher medication rate with anticoagulation therapy than those without stroke or TIA, while there were no significant differences in baseline characteristics such as sex, age, NYHA class, “stroke risk score”, LVEF, anti-platelet or anti-heart failure drug therapy, or estimated glomerular filtration rate between them.

Table 1 Baseline clinical and study characteristics in chronic heart failure patients with and without ischemic stroke or transient ischemic attack
Full size table

When we demonstrated the discriminatory ability for ischemic stroke or TIA, CHADS2 score had the highest AUC [0.805 (95 % CI 0.719–0.892)]. CHADS2 score was statistically superior to “stroke risk score” [AUC 0.598 (95 % CI 0.465–0.731), p = 0.003] and tended to be superior to CHA2DS2-VASc score [AUC 0.739 (95 % CI 0.616–0.863), p = 0.098] (Fig. 1).

Fig. 1
figure 1

Receiver operating characteristics (ROC) curves for CHADS2 score, CHA2DS2-VASc score and “stroke risk score” in discriminating ischemic stroke or TIA, TIA transient ischemic attack

Full size image

Univariate Cox proportional hazard analysis showed that the CHADS2 score, CHA2DS2-VASc score, “stroke risk score”, age ≥75 years, prior stroke or TIA, diabetes mellitus and hypertension were significantly associated with stroke or TIA, while there were no significant association between stroke or TIA and the other components of 3 risk scores such as gender, age ≥65 years, vascular disease, NYHA class (III or IV) and BMI. Adjusted in the model with anticoagulation therapy, CHADS2 score (C-index 0.794, 95 % CI 0.663–0.925) and CHA2DS2-VASc score (C-index 0.740, 95 % CI 0.605–0.875), but not “stroke risk score” (C-index 0.625, 95 % CI 0.488–0.762), were still significantly associated with ischemic stroke or TIA (Table 2).

Table 2 Cox proportional hazard analysis for the identification of patients at risk of ischemic stroke or transient ischemic attack
Full size table

According to CHADS2 score, the study patients were classified into three groups: low CHADS2 score (=1), intermediate CHADS2 score (=2), and high CHADS2 score (≥3) groups. Baseline characteristics in CHF patients with low, intermediate, and high CHADS2 scores are shown in Table 3. There were no significant differences in baseline characteristics such as age, NYHA class, LVEF, anticoagulation or anti-platelet therapies or anti-heart failure drug therapy among the three groups, except for hypertension, diabetes mellitus, prior stroke, systolic blood pressure and renal function. The estimated glomerular filtration rate was significantly lower in the high CHADS2 score group than in the other 2 groups.

Table 3 Baseline clinical and study characteristics in chronic heart failure patients with low, intermediate and high CHADS2 score
Full size table

Ischemic stroke or TIA was observed in none of the 38 patients in the low CHADS2 score group, 7 of 51 (1.6 per 100 person-years) in the intermediate CHADS2 score group, and 14 of 38 (4.7 per 100 person-years) in the high CHADS2 score group. The higher the CHADS2 scores, the more stroke or TIA occurred (high vs intermediate, p = 0.01; intermediate vs low, p = 0.01) (Fig. 2). Compared to patients with an intermediate CHADS2 score, those with high CHADS2 scores had a significantly higher risk of stroke or TIA (adjusted HR, 3.2; 95 % CI 1.3–7.9).

Fig. 2
figure 2

Ischemic stroke- or TIA-free rate curves in CHF patients with low, intermediate, and high CHADS2 scores, Low CHADS2 score = 1, intermediate = 2, and high = 3–6, CHF chronic heart failure, TIA transient ischemic attack

Full size image

Although no study patients had the diagnosis of AF at entry, 22 of 127 CHF patients (17 %) developed new non-valvular AF documented by electrocardiography during the follow-up period; 6 of 38 patients (16 %) in the low CHADS2 score group, 10 of 51 (20 %) in the intermediate CHADS2 score group, and 6 of 38 (16 %) in the high CHADS2 score group. We continuously followed all study patients despite the development of AF. Even if we stopped following these 22 patients at the time of AF development, at multivariate analysis, CHADS2 score, but not CHA2DS2-VASc score or “stroke risk score”, was significantly and independently associated with ischemic stroke or TIA (p = 0.006, C-index 0.776, 95 % CI 0.635–0.917).

When we divided all study patients into 2 groups according to the presence of anti-coagulation therapy at baseline, ischemic stroke or TIA was observed in 8 of 73 patients without anti-coagulation therapy and 13 of 54 with that. In patients with anti-coagulation therapy, CHADS2 score, CHA2DS2-VASc score, diabetes mellitus, and hypertension were significantly associated with stroke or TIA while the stroke risk score showed no significant association with stroke or TIA. In patients without anti-coagulation therapy, only CHADS2 score was also significantly associated with stroke or TIA (p = 0.009, C-index 0.806, 95 % CI 0.769–0.843).

Discussion

This study revealed that the CHADS2 score could also be useful in stratifying the risk for ischemic stroke or TIA in CHF patients without AF, and that the prognostic power of the prediction for ischemic stroke of CHADS2 score was greater than “stroke risk score”.

Some factors could cause the superiority of CHADS2 score to the “stroke risk score”. First, this study had lower proportion of CHF patients with severe symptom (NYHA III or IV) and with diabetes mellitus treated with insulin than patients enrolled in CORONA and GISSI-HF, and so we could not reflect exactly the stroke risk in these patients. Secondly, patients had the lower medication rate of anti-platelet therapy and higher medication rate of anticoagulant therapy in this study than those enrolled in CORONA and GISSI-HF. It might result in the difference in the mechanism of stroke, atherothrombotic or cardioembolic, and might affect stroke risk in study patients. Finally, this study had quite long period of follow-up. Patients in this study might acquire increased stroke risk during the longer follow-up period, even when their stroke risk had been regarded as low at the entry.

Several mechanisms might explain the relationship between a high CHADS2 score and stroke in CHF patients without AF. Even in the absence of AF, CHF patients develop a hypercoagulable state and endothelial dysfunction [3–5]. Other components of the CHADS2 score such as hypertension, diabetes mellitus and prior stroke have an association with hypercoagulability and endothelial dysfunction [20–22]. Therefore, CHF patients with a high CHADS2 score have activated thrombus formation.

The components of the CHADS2 score such as CHF, hypertension and diabetes mellitus could contribute to left atrial remodeling, characterized by dilation and mechanical dysfunction of the left atrium [23, 24]. A dilated left atrium could result in blood stasis, and thus an increased risk of thromboembolism [25, 26], independent of cardiac rhythm, although there was no significant difference in left atrial dimension among low, intermediate high CHADS2 scores in the present study.

The use of antithrombotic treatments remains an important question in the care of patients with CHF. In the present study, the event rate in non-AF CHF parents with high CHADS2 scores (3–6) was comparable to the rate in AF patients with moderate-to-high CHADS2 scores (2–3), a population known to derive benefit from stroke prevention therapies such as anticoagulation. Studies that have examined the role of anticoagulation therapy for reducing thromboembolic risk have been inconclusive [27]. The multicenter, randomized double-blind and placebo-controlled Heart Failure Long-term Antithrombotic Study (HELAS) and the unblinded randomized Warfarin/Aspirin Study in Heart failure (WASH) did not find a benefit with antithrombotic therapy [28]. Although the clinical outcome studies [the prospective randomized warfarin and anti-platelet therapy in chronic heart failure (WATCH) study and the multicenter, double-blinded and randomized warfarin versus aspirin in patients with reduced cardiac ejection fraction (WARCEF) study] suggested that warfarin may reduce stroke risk compared with anti-platelet therapy [27, 29], the lack of a placebo group and lower-than-projected enrollment prevents definitive conclusions from being made [29]. Thus, current evidence does not support the routine use of anticoagulation for preventing thromboembolic events in CHF patients who remain in sinus rhythm [29].

There are some limitations to our study. First, the small and empirically chosen population sample size is a major limitation. Second, medications such as anticoagulation therapy at baseline and during the follow-up period, might affect the incidence of ischemic stroke. At baseline, the rate of anticoagulation therapy tended to decrease as CHADS2 score increased with no significant difference. The incidence of stroke or TIA was higher in patient with than without anticoagulation at baseline, although the statistical difference was not significant (24 vs 11 %, p = 0.08, adjusted HR 2.170 95 % CI 0.897–5.249). In the present study, anticoagulation therapy was performed more in CHF patients with than without ischemic origin (old myocardial infarction) [61 %(46/76) vs 16(8/51), p < 0.0001], possibly according to the usefulness of antithrombotic therapy in the secondary prevention of myocardial infarction [30]. So, the higher incidence of stroke or TIA in patients with anticoagulation therapy might be due to selection bias. During the follow-up period, an anticoagulation drug was newly administered in 13 of 73 patients without anticoagulation therapy at entry. There was no difference in the medication rate of anticoagulation at the last follow-up visit among low, intermediate, and high CHADS2 score groups (63, 49, and 47 %, respectively). Third, we evaluated the presence of AF according to past history and electrocardiography at entry. However, we could not eliminate the possibility that CHF patients with paroxysmal AF could be misclassified into the sinus rhythm group if they have no symptoms such as palpitations and no evidence of AF on baseline electrocardiography [31]. Such silent AF might have an influence on the incidence of ischemic stroke in our study. Fourth, we could not investigate the mechanism of ischemic stroke, whether atherothrombotic or cardioembolic. Thus, we could not clarify which mechanism was mainly associated with a high stroke risk in CHF patients with high CHADS2 scores. Identifying the mechanism of ischemic stroke may be useful in determining appropriate stroke prevention therapy in patients with CHF. Finally, we included only CHF patients with LVEF less than 40 %, and so our results could not be applied to CHF patients with preserved ejection fraction either.

In conclusion, we found that CHADS2 score could stratify the risk of ischemic stroke in CHF patients with the absence of AF, with greater prognostic power than the “stroke risk score”.