External Validation of the HELT-E2S2 Score in Japanese Patients With Nonvalvular Atrial Fibrillation　― A Pooled Analysis of the RAFFINE and SAKURA Registries ―

Takahiro Yamauchi; Yasuo Okumura; Koichi Nagashima; Ryuta Watanabe; Yuki Saito; Katsuaki Yokoyama; Naoya Matsumoto; Katsumi Miyauchi; Sakiko Miyazaki; Hidemori Hayashi; Yuya Matsue; Yuji Nishizaki; Shuko Nojiri; Tohru Minamino; Hiroyuki Daida

doi:10.1253/circj.CJ-23-0318

Abstract

Background: The HELT-E₂S₂ score, which assigns 1 point to Hypertension, Elderly aged 75–84 years, Low body mass index <18.5 kg/m², and Type of atrial fibrillation (AF: persistent/permanent), and 2 points to Extreme Elderly aged ≥85 years and previous Stroke, has been proposed as a new risk stratification for strokes in Japanese AF patients, but has not yet undergone external validation.

Methods and Results: We evaluated the prognostic performance of the HELT-E₂S₂ score for stroke risk stratification using 2 large-scale registries in Japanese AF patients (n=7,020). During 23,241 person-years of follow-up (mean follow-up 1,208±450 days), 287 ischemic stroke events occurred. The C-statistic using the HELT-E₂S₂ score was 0.661 (95% confidence interval [CI], 0.629–0.692), which was numerically higher than with the CHADS₂ score (0.644, 95% CI 0.613–0.675; P=0.15 vs. HELT-E₂S₂) or CHA₂DS₂-VASc score (0.650, 95% CI, 0.619–0.680; P=0.37 vs. HELT-E₂S₂). In the SAKURA AF Registry, the C-statistic of the HELT-E₂S₂ score was consistently higher than the CHADS₂ and CHA₂DS₂-VASc scores across all 3 types of facilities comprising university hospitals, general hospitals, and clinics. However, in the RAFFINE Study, its superiority was only observed in general hospitals.

Conclusions: The HELT-E₂S₂ score demonstrated potential value for risk stratification, particularly in a super-aged society such as Japan. However, its superiority over the CHADS₂ or CHA₂DS₂-VASc scores may vary across different hospital settings.

Atrial fibrillation (AF) is a prevalent arrhythmia in Japan, and as in Western countries, the proportion of AF has increased over the past few decades.¹ AF is strongly associated with strokes, heart failure, and death.²^–⁴ Several real-world observational studies have demonstrated that those AF-related events increase with aging, and the increase is more pronounced in very elderly patients especially.¹^,⁵^,⁶ Therefore, the identification of patients with a stroke risk and instituting stroke prevention are urgent issues for a “super-aged” society such as in Japan. Currently, the CHADS₂ and CHA₂DS₂-VASc scores are recommended by guidelines and are used in clinical practice for stroke risk stratification.⁷^–⁹ However, these scores lack information on very elderly or low body weight patients. The J-RHYTHM registry failed to find body mass index (BMI) as an independent stroke risk factor,¹⁰ although several recent reports have consistently reported its significance.⁵^,⁶^,¹¹ The CHADS₂ score has limited ability to stratify thromboembolic risk in the Japanese AF patients,¹² as do novel risk factors that have been identified in several Japanese cohorts with different characteristics including BMI <18.5 kg/m²,¹³ persistent AF,¹⁴ renal dysfunction,¹³^,¹⁵^–¹⁷ anemia,¹³ and elevated B-type natriuretic peptide (BNP) or NT-proBNP levels.¹⁸^,¹⁹ To overcome such issues and address these challenges, a new risk stratification system called the HELT-E₂S₂ score has recently been proposed for the Japanese population.²⁰ It consists of H: hypertension, E: elderly aged 75–84 years, L: low BMI (<18.5 kg/m²), T: persistent/permanent AF, E2: extremely elderly >85 years, and S2: previous stroke. These factors have been weighted according to their relevance to the Japanese population, which has a significant proportion of elderly individuals.²⁰^,²¹ However, there has been no external validation thus far,²² and it is unclear whether this score can be generalized to other Japanese cohorts. Therefore, in this study, we aimed to validate the HELT-E₂S₂ score in Japanese patients with AF enrolled in an integrated analysis of 2 Japanese registries: the RAFFINE Registry²³^,²⁴ and the SAKURA AF Registry.¹⁷^,²⁵

Methods

Study Patients

Individual data from patients enrolled in the RAFFINE Study (n=3,889) (approval no. M11-0799 [November 11, 2014]) and the SAKURA AF Registry (n=3,267) (approval no. RK-130111-2 [February 1, 2013]) were combined and analyzed. The details of the study design and baseline characteristics of the subjects in these registries have been reported elsewhere.¹⁷^,²³^–²⁵ Briefly, in the RAFFINE Study, patients aged ≥20 years with documented AF by 12-lead ECG or 24-hour Holter ECG were enrolled. All patients were followed up annually for at least 3 years and up to 5 years. In the SAKURA AF Registry, patients were eligible for inclusion if they had nonvalvular AF diagnosed on a 12-lead ECG, 24-hour Holter ECG, or event-activated ECG recording, were aged 20 years, or had started on, or were already being treated with, any oral anticoagulants (OACs) for stroke prophylaxis. Therefore, none of the study patients were not taking OACs. All patients were followed up for at least 1 year and up to 3 years. Of the 7,156 patients, 22 with an unknown BMI, 73 with an unknown type of AF, and 41 with an unknown stroke/transient ischemic attack (TIA) were excluded, thus leaving 7,020 patients for the analysis in this study. This study was approved by the Ethics Committee of Nihon University (RK-210413-11) as well as the Institutional Review Board of Juntendo University Hospital (H20-0394) and was conducted according to the ethical principles of the Declaration of Helsinki.

Data Collection

We collected common baseline data from the RAFFINE Study and SAKURA AF Registry as follows: age, sex, weight, and height; type of AF (paroxysmal AF [AF lasting ≤7 days], persistent AF [AF lasting >7 days and ≤1 year], or long-standing persistent AF [lasting >1 year]); currently used medications, including antiarrhythmic, anticoagulation, and antiplatelet drugs; comorbidities and/or risk factors, including hypertension, diabetes, congestive heart failure, stroke/TIA, ischemic heart disease (IHD), smoking and alcohol consumption at enrolment, and prior major bleeding. The presence of previous major bleeding was also recorded; the CHADS₂ and CHA₂DS₂-VASc scores were calculated and recorded.

A history of a stroke/ TIA included symptomatic/asymptomatic cerebral infarction, TIA, and intracranial hemorrhage in the RAFFINE Study, whereas only symptomatic stroke and TIA were included in the SAKURA AF Registry. In addition, the RAFFINE Study defined IHD as angina, myocardial infarction (MI), silent myocardial ischemia, and a history of coronary intervention or coronary artery bypass grafting, whereas the SAKURA AF Registry defined IHD as only angina or MI. A low hemoglobin was defined as <13 g/dL for males and 12 g/dL for females.

Statistical Analysis

Data are expressed as the mean±standard deviation, number (%), or events/number of patients (%). Differences in the clinical characteristics between groups were analyzed by an unpaired Student’s t-test, Man-Whitney test, or chi-square test, as appropriate. For ≥3 group comparisons, analysis of variance (ANOVA), Kruskal-Wallis test, or chi-square test was conducted, as appropriate. Kaplan-Meier curves were generated to depict the cumulative incidence of ischemic stroke for categorical variables (0, 1, ≥2) of the CHADS₂ and CHA₂DS₂-VASc scores and for those (0, 1, 2, 3, ≥4) of the HELT-E₂S₂. The discrimination ability was evaluated using a concordance-statistic (C-statistic) and 95% confidence intervals (CI), and the C-statistics between the two groups were compared using the DeLong test. A Cox proportional hazards model was used to calculate the hazard ratios (HRs) and 95% CIs for the incidence of ischemic stroke events. Multivariate Cox proportional hazards models were conducted using a forced entry method to determine any significant risk factors of stroke events. Factors in the multivariate model included all components of the HELT-E₂S₂ score, as well as diabetes, vascular disease, creatinine ≥1.0 mg/dL, low hemoglobin level, and no use of OACs. Finally, the estimated probability of developing a stroke and the observed proportion of strokes stratified by the risk score were calculated and compared between patients with and without OACs by an exact Poisson test. Statistical analysis was performed using JMP 16 software (SAS, Cary, NC, USA), and P<0.05 was considered statistically significant.

Results

Clinical Characteristics of the Study Patients

The characteristics of the patients enrolled are summarized in Table 1. The mean age was 72.1±9.5 years, and 2,042 (29%) were female. The mean weight was 63.2±13 kg and the mean CHADS₂, CHA₂DS₂-VASc, and HELT-E₂S₂ scores were 1.9±1.2, 3.1±1.6, and 2.2±1.3, respectively (hypertension, 72%; previous stroke, 13%; persistent/long-standing persistent AF, 62%).

Table 1. Characteristics of Patients in the Total Study Group and in the RAFFINE Study and SAKURA AF Registry

	Total (n=7,020)	RAFFINE Study (n=3,828)	SAKURA AF Registry (n=3,192)	P value*
Age (years)	72.1±9.5	72.2±9.6	71.9±9.4	0.26
<75	4,026 (57)	2,147 (56)	1,879 (59)	0.045
75–84	2,424 (35)	1,370 (36)	1,054 (33)
≥85	570 (8)	311 (8)	259 (8)
Female sex (%)	2,042 (29)	1,204 (31)	838 (26)	<0.001
Height (cm)	162.1±9.6	161.9±9.8	162.5±9.5	0.006
Weight (kg)	63.2±13	62.6±13	63.9±13	<0.001
BMI (kg/m²)	23.9±3.7	23.8±3.7	24.1±3.7	0.001
<18.5	363 (5)	213 (6)	150 (5)	0.10
Type of AF (%)
Paroxysmal	2,660 (38)	1,468 (38)	1,192 (37)	<0.001
Persistent	1,069 (15)	360 (9)	709 (22)
Long-standing persistent	3,291 (47)	2,000 (52)	1,291 (40)
Congestive heart failure	1,618 (23)	905 (24)	713 (22)	0.20
Hypertension	5,062 (72)	2,782 (73)	2,280 (71)	0.25
Diabetes mellitus	1,898 (27)	1,163 (30)	735 (23)	<0.001
Stroke/TIA	940 (13)	579 (15)	361 (11)	<0.001
Vascular disease	603 (9)	354 (9)	249 (8)	0.031
OAC	462 (7)	462 (12)	0 (0)	<0.001
CHADS₂ score	1.9±1.2	2.0±1.3	1.8±1.2	<0.001
0	742 (11)	390 (10)	352 (11)
1	2,065 (29)	1,034 (27)	1,031 (32)
≥2	4,213 (60)	2,404 (63)	1,809 (57)
CHA₂DS₂-VASc score	3.1±1.6	3.2±1.6	3.0±1.5	<0.001
0	242 (3)	135 (4)	107 (3)
1	829 (12)	429 (11)	400 (13)
≥2	5,949 (85)	3,264 (85)	2,685 (84)
HELT-E₂S₂ score	2.2±1.3	2.3±1.3	2.1±1.2	<0.001
0	428 (6)	227 (6)	201 (6)
1	1,746 (25)	908 (24)	838 (26)
2	2,294 (33)	1,155 (30)	1,139 (36)
3	1,422 (20)	812 (21)	610 (19)
≥4	1,130 (16)	726 (19)	404 (13)
Creatinine ≥1.0 mg/dL^†	1,988 (28)	1,048 (27)	940 (30)	0.047
Low hemoglobin^‡	1,615 (23)	935 (25)	680 (22)	0.012

Data are shown as the mean±SD or number (%). *Per Student’s t-test, Mann-Whitney test, or chi-square test, as appropriate. ^†28 patients and ^‡109 patients were excluded from the analysis due to missing data on creatine and hemoglobin levels. AF, atrial fibrillation; BMI, body mass index; OAC, oral anticoagulant [drug]; TIA, transient ischemic attack.

Clinical Outcome and Prediction of Ischemic Stroke Events by the CHADS₂, CHA₂DS₂-VASc and HELT-E₂S₂ Scores

In this pooled analysis, 287 ischemic stroke events occurred during 23,241 person-years of follow-up (mean follow-up, 1,208±450 days). When stratified by the CHADS₂ score, patients with a score ≥2 (high risk) had a significantly higher incidence of strokes (HR 4.04, 95% CI 2.20–7.39; P<0.001) than those with a score of 0 (low risk) as a counterpart (Figure 1A, Table 2). The C-statistic for the CHADS₂ score (total score: 0–6) was 0.644 (95% CI, 0.613–0.675) (Table 3). As for the CHA₂DS₂-VASc score stratification, patients with a score ≥2 had a significantly higher incidence of strokes (HR 12.73, 95% CI, 1.79–90.68; P=0.011) as compared with those with a score 0 (Figure 1, Table 2). The C-statistic for the CHA₂DS₂-VASc score (total score: 0–9) was 0.650 (95% CI, 0.619–0.680) (Table 3).

Figure 1.

Kaplan-Meier curves for the incidence of ischemic stroke stratified by the CHADS₂ (A) and CHA₂DS₂-VASc scores (B). The hazard ratio (HR) and 95% confidence interval (CI) are calculated for scores of 1 and ≥2 as compared with a score of 0 as a reference.

Table 2. Ischemic Stroke Events Stratified by the Risk Scoring System and OACs or No OACs

	Total	RAFFINE	SAKURA AF
Total no. of patients	7,020	3,828	3,192
Ischemic stroke events	287 (4.1)	166 (4.3)	121 (3.8)
CHADS₂ score
0	11/742 (1.5)	5/390 (1.3)	6/352 (1.7)
1	46/2,065 (2.2)	20/1,034 (1.9)	26/1,031 (2.5)
≥2	230/4,213 (5.5)	141/2,404 (5.9)	89/1,809 (4.9)
CHA₂DS₂-VASc score
0	1/242 (0.4)	0/135 (0.0)	1/107 (0.9)
1	10/829 (1.2)	3/429 (0.7)	7/400 (1.8)
≥2	276/5,949 (4.6)	163/3,264 (5.0)	113/2,685 (4.2)
HELT-E₂S₂ score
0	9/428 (2.1)	5/227 (2.2)	4/201 (2.0)
1	29/1,746 (1.7)	15/908 (1.7)	14/838 (1.7)
2	76/2,294 (3.3)	36/1,155 (3.1)	40/1,139 (3.5)
3	76/1,422 (5.3)	46/812 (5.6)	30/610 (4.9)
≥4	97/1,130 (8.6)	64/726 (8.8)	33/404 (8.2)
OAC
OAC	265/6,558 (4.0)	144/3,366 (4.3)	121/3,192 (3.8)
No OAC	22/462 (4.8)	22/462 (4.8)	–

Values are expressed as events/number of patients (%). OAC, oral anticoagulant [drug].

Table 3. C-Statistics and Comparison of Risk Stratification Scores

	C-statistic (95% CI)	P value (Comparison with HELT-E₂S₂ score)
All patients (n=7,020)
HELT-E₂S₂ score	0.661 (0.629–0.692)	Ref.
CHADS₂ score	0.644 (0.613–0.675)	0.15
CHA₂DS₂-VASc score	0.650 (0.619–0.680)	0.37
RAFFINE Study (n=3,828)
HELT-E₂S₂ score	0.669 (0.627–0.709)	Ref.
CHADS₂ score	0.666 (0.625–0.704)	0.85
CHA₂DS₂-VASc score	0.674 (0.634–0.712)	0.74
University hospitals (n=2,668)
HELT-E₂S₂ score	0.679 (0.630–0.725)	Ref.
CHADS₂ score	0.666 (0.618–0.710)	0.49
CHA₂DS₂-VASc score	0.685 (0.638–0.728)	0.78
General hospitals (n=599)
HELT-E₂S₂ score	0.652 (0.542–0.748)	Ref.
CHADS₂ score	0.626 (0.498–0.739)	0.59
CHA₂DS₂-VASc score	0.623 (0.504–0.730)	0.53
Clinics (n=561)
HELT-E₂S₂ score	0.636 (0.520–0.739)	Ref.
CHADS₂ score	0.696 (0.586–0.788)	0.15
CHA₂DS₂-VASc score	0.664 (0.560–0.754)	0.54
SAKURA AF Registry (n=3,192)
HELT-E₂S₂ score	0.648 (0.597–0.695)	Ref.
CHADS₂ score	0.612 (0.563–0.659)	0.029
CHA₂DS₂-VASc score	0.615 (0.566–0.662)	0.07
University hospitals (n=1,155)
HELT-E₂S₂ score	0.645 (0.568–0.715)	Ref.
CHADS₂ score	0.603 (0.522–0.679)	0.05
CHA₂DS₂-VASc score	0.609 (0.534–0.680)	0.17
General hospitals (n=1,296)
HELT-E₂S₂ score	0.657 (0.579–0.726)	Ref.
CHADS₂ score	0.617 (0.551–0.679)	0.18
CHA₂DS₂-VASc score	0.643 (0.571–0.709)	0.67
Clinics (n=741)
HELT-E₂S₂ score	0.702 (0.561–0.812)	Ref.
CHADS₂ score	0.631 (0.476–0.763)	0.08
CHA₂DS₂-VASc score	0.588 (0.438–0.724)	0.003
OAC (n=6,558)
HELT-E₂S₂ score	0.664 (0.630–0.696)	Ref.
CHADS₂ score	0.646 (0.613–0.678)	0.16
CHA₂DS₂-VASc score	0.648 (0.615–0.679)	0.22
No OAC (n=462)
HELT-E₂S₂ score	0.657 (0.536–0.760)	Ref.
CHADS₂ score	0.659 (0.559–0.746)	0.96
CHA₂DS₂-VASc score	0.701 (0.595–0.789)	0.27

C-statistic was evaluated by the following classification: score 0–7 in the HELT-E₂S₂ score, score 0–6 in the CHADS₂ score, and score 0–9 in the CHA₂DS₂-VASc score. CI, confidence interval; OAC, oral anticoagulant [drug].

The patient characteristics for the HELT-E₂S₂ score categories (scores 0, 1, 2, 3, and ≥4) are presented in Supplementary Table 1. The number of ischemic stroke events during the entire follow-up period was 9/428 (2.1%), 29/1,746 (1.7%), 76/2,294 (3.3%), 76/1,422 (5.3%), and 97/1,130 (8.6%) for HELT-E₂S₂ scores 0, 1, 2, 3, and ≥4, respectively (Table 2). Figure 2 shows the Kaplan-Meier curves for each score. Score 1 had the lowest cumulative incidence of ischemic strokes, followed by a score of 0, but a score ≥1 was associated with incremental cumulative incidence of ischemic strokes with increasing scores. Patients with HELT-E₂S₂ scores ≥3 had a significantly higher incidence of stroke (HR, 2.68, 95% CI, 1.34–5.36; P=0.005 for those with a score of 3 and HR 4.60, 95% CI, 2.32–9.11; P<0.001 for those with scores ≥4) than those with a score of 0 (Figure 2). The C-statistic using the HELT-E₂S₂ score (total scoring: 0–7) was 0.661 (95% CI, 0.629-0.692), which was numerically (but not significantly) higher than that using the CHADS₂ score (0.644, 95% CI 0.613–0.675; P=0.15 vs. HELT-E₂S₂) or CHA₂DS₂-VASc score (0.650, 95% CI, 0.619–0.680; P=0.37 vs. HELT-E₂S₂) (Table 3).

Figure 2.

Kaplan-Meier curves for the incidence of ischemic stroke stratified by the HELT-E₂S₂ score. The hazard ratio (HR) and 95% confidence interval (CI) are calculated for scores of 1, 2, 3, and ≥4 as compared with a score of 0 as a reference.

Comparison Between the RAFFINE Study and SAKURA AF Registry

Table 1 shows the baseline patient characteristics for those enrolled in the RAFFINE Study (n=3,828) and in the SAKURA AF Registry (n=3,192). The incident rates of ischemic stroke events for each, the CHADS₂, CHA₂DS₂-VASc, and HELT-E₂S₂ scores in both studies are summarized in Table 2, and the Kaplan-Meier curves for each HELT-E₂S₂ score are shown in Figure 3A,B. In both studies, patients with a score of 1 had the lowest cumulative incidence of ischemic strokes, followed by those with a score of 0, but those with a score ≥1 had higher cumulative incidences of ischemic strokes as the score increased. In the RAFFINE Study, the C-statistic using the HELT-E₂S₂ score was 0.669 (95% CI 0.627–0.709), which was numerically higher than that using the CHADS₂ score (C-statistic 0.666; 95% CI 0.625–0.704) and moderately lower than that using the CHA₂DS₂-VASc score (C-statistic 0.674; 95% CI, 0.634–0.712), without any statistically significant differences (Table 3). In the SAKURA AF Registry, the C-statistic using the HELT-E₂S₂ score was 0.648 (95% CI, 0.597–0.695), which was significantly higher than that with the CHADS₂ score (C-statistic 0.612; 95% CI, 0.563–0.659) and more likely to be higher with the CHA₂DS₂-VASc score (C-statistic 0.615; 95% CI, 0.566–0.662) (Table 3).

Figure 3.

Kaplan-Meier curves of ischemic stroke incidence stratified by the HELT-E₂S₂ score in patients enrolled in the RAFFINE Study (A) and in those in the SAKURA AF Registry (B). Hazard ratios (HR) and 95% confidence intervals (CI) were calculated for scores of 1, 2, 3, and ≥4 as compared with a score of 0 as a reference.

Table 4 shows the patient characteristics in the RAFFINE Study and the SAKURA AF Registry divided by private clinics, university hospitals, and general hospitals, as previously reported.²³^,²⁶ The C-statistic using the HELT-E₂S₂ score in the RAFFINE Study was numerically higher than other traditional scores in general hospitals, but the C-statistic using the CHA₂DS₂-VASc score and the C-statistic using the CHADS₂ and CHA₂DS₂-VASc scores higher than that using the HELT-E₂S₂ score in university hospitals and clinics, respectively. In contrast, in the SAKURA AF Registry, the C-statistic using the HELT-E₂S₂ score was consistently higher than other traditional scores across all types of facilities (Table 3).

Table 4. Patients’ Characteristics in the Different Facilities in the RAFFINE Study, SAKURA AF Registry and the J-RISK AF Study

	RAFFINE Study (n=3,828)				SAKURA AF Registry (n=3,192)				J-RISK AF**
	University Hospital (n=2,668)	General Hospital (n=599)	Clinic (n=561)	P value*	University Hospital (n=1,155)	General Hospital (n=1,296)	Clinic (n=741)	P value*	(n=12,289)
Age (years)	72.2±9.6	72.8±9.2	71.4±10.0	0.06	70.2±9.8	72.6±8.9	73.4±9.2	<0.001	70.2±11
<75	1,480 (55)	338 (56)	329 (59)	0.41	762 (66)	735 (57)	382 (52)	<0.001	7,665 (63)
75–84	978 (37)	210 (35)	182 (32)		323 (28)	452 (35)	279 (38)		3,735 (30)
≥85	210 (8)	51 (9)	50 (9)		70 (6)	109 (8)	80 (11)		889 (7)
Female sex	830 (31)	201 (34)	173 (31)	0.48	270 (23)	337 (26)	231 (31)	<0.001	3,758 (31)
Height (cm)	162.1±9.7	161.1±9.7	161.5±10.2	0.034	164.1±9.3	162.0±9.3	160.8±9.7	<0.001
Weight (kg)	62.7±13	61.5±13	63.4±12	0.028	65.1±13	63.6±13	62.5±13	<0.001
BMI (kg/m²)	23.7±3.7	23.6±3.9	24.2±3.4	0.008	24.1±3.6	24.1±3.9	24.0±3.6	0.92
<18.5	155 (6)	41 (7)	17 (3)	0.011	47 (4)	73 (6)	30 (4)	0.12	895 (7)
Type of AF (%)
Paroxysmal	1,065 (40)	209 (35)	194 (35)	0.002	544 (47)	394 (30)	254 (34)	<0.001	5,504 (45)
Persistent	224 (8)	75 (13)	61 (11)		231 (20)	279 (22)	199 (27)		6,785 (55)
Long-standing persistent	1,379 (52)	315 (53)	306 (55)		380 (33)	623 (48)	288 (39)
Congestive heart failure	668 (25)	145 (24)	92 (16)	<0.001	199 (17)	381 (29)	133 (18)	<0.001	3,258 (27)
Hypertension	1,940 (73)	440 (73)	402 (72)	0.79	794 (69)	932 (72)	554 (75)	0.016	8,971 (73)
Diabetes mellitus	797 (30)	181 (30)	185 (33)	0.35	276 (24)	326 (25)	133 (18)	<0.001	2,541 (21)
Stroke/TIA	457 (17)	73 (12)	49 (9)	<0.001	119 (10)	165 (13)	77 (10)	0.11	1,729 (14)
Vascular disease	263 (10)	58 (10)	33 (6)	0.012	95 (8)	102 (8)	52 (7)	0.63	1,703 (14)
No OAC	292 (11)	63 (11)	107 (19)	<0.001	–	–	–	–	3,197 (26)
CHADS₂ score	2.1±1.3	2.0±1.2	1.8±1.1	<0.001	1.6±1.1	2.0±1.2	1.8±1.2	<0.001	1.7±1.3
0	272 (10)	57 (10)	61 (11)		171 (15)	108 (8)	73 (10)		1,965 (16)
1	689 (26)	169 (28)	176 (31)		396 (34)	375 (29)	260 (35)		3,920 (32)
≥2	1,707 (64)	373 (62)	324 (58)		588 (51)	813 (63)	408 (55)		6,404 (52)
CHA₂DS₂-VASc score	3.3±1.6	3.2±1.5	2.9±1.5	<0.001	2.8±1.5	3.2±1.5	3.1±1.5	<0.001	2.9±1.7
0	98 (4)	14 (2)	23 (4)		57 (5)	28 (2)	22 (3)		890 (7)
1	290 (11)	56 (9)	83 (15)		189 (16)	126 (10)	85 (11)		1,830 (15)
≥2	2,280 (85)	529 (88)	455 (81)		909 (79)	1,142 (88)	634 (86)		9,569 (78)
HELT-E₂S₂ score	2.3±1.4	2.3±1.3	2.2±1.3	0.06	1.9±1.2	2.2±1.2	2.2±1.2	<0.001
0	163 (6)	27 (5)	37 (7)		108 (9)	58 (4)	35 (5)		1,077 (9)
1	626 (23)	139 (23)	143 (25)		377 (33)	287 (22)	174 (23)		3,301 (27)
2	771 (29)	202 (34)	182 (32)		383 (33)	487 (38)	269 (36)		3,983 (32)
3	581 (22)	122 (20)	109 (19)		170 (15)	286 (22)	154 (21)		2,190 (18)
≥4	527 (20)	109 (18)	90 (16)		117 (10)	178 (14)	109 (15)		1,738 (14)
Creatinine ≥1.0 mg/dL^†	702 (26)	217 (36)	129 (23)	<0.001	353 (31)	426 (33)	161 (22)	<0.001	3,821 (31)
Low hemoglobin^‡	728 (27)	159 (27)	98 (18)	<0.001	269 (24)	307 (24)	152 (21)	0.24	3,256 (27)

Data are shown as the mean±SD or number (%). *Per ANOVA, Kruskal-Wallis test, or chi-square test, as appropriate. **Cited by reference no. 21. ^†28 patients and ^‡109 patients were excluded from the analysis due to missing data on creatine and hemoglobin levels. Abbreviations as in Table 1.

Risk Factors for Strokes in the Entire Study Group and 3 Different Facilities in the RAFFINE Study and SAKURA AF Registry

In the entire study group, the components of the HELT-E₂S₂ score, with the exception of hypertension, were independently associated with the incidence of ischemic stroke events. Diabetes mellitus and vascular disease were also significant risk factors for ischemic strokes even after multivariate Cox proportional hazards models (adjusted HR for diabetes mellitus 1.40 95% CI 1.09–1.80; P=0.001 and adjusted HR for vascular disease 1.48, 95% CI, 1.05–2.09; P=0.002) (Table 5). Although the significant stroke risk factors varied among the 3 different facilities, a consistent stroke risk factor across both the RAFINE Study and the SAKURA AF Registry was age ≥75 years. In the RAFFINE Study, diabetes was identified as a significant risk factor for ischemic strokes in general hospitals and clinics, and, specifically in university hospitals, all components of the CHA₂DS₂-VASc scores were significantly or tended to be associated with the incidence of ischemic stroke events (Table 5).

Table 5. Univariate and Multivariate Analyses of the Entire Study Group, and Univariate Analyses by Cox Proportional Hazards Model in the 3 Types of Facility in the RAFFINE Study and SAKURA AF Registry

Characteristic	Entire Study				RAFFINE Study						SAKURA AF Registry
	Univariate		Multivariable		Clinics		University Hospitals		General Hospitals		Clinics		University Hospitals		General Hospitals
	HR (95% CI)	P value	Adjusted HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Age (<75 years)	Ref.		Ref.		Ref.		Ref.		Ref.		Ref.		Ref.		Ref.
75–84 years (Elderly)	2.11 (1.64–2.71)	<0.001	1.77 (1.36–2.31)	<0.001	2.02 (0.84–4.87)	0.12	2.26 (1.53–3.36)	<0.001	2.80 (1.02–7.70)	0.046	1.16 (0.35–3.80)	0.81	1.63 (0.93–2.86)	0.09	3.20 (1.71–6.00)	<0.001
≥85 years (Extreme elderly)	3.66 (2.58–5.17)	<0.001	2.94 (2.03–4.26)	<0.001	7.36 (2.77–19.59)	<0.001	4.60 (2.72–7.77)	<0.001	1.29 (0.15–10.71)	0.81	3.96 (1.11–14.14)	0.034	1.20 (0.37–3.93)	0.76	4.76 (2.08–10.90)	<0.001
Female sex (%)	1.23 (0.96–1.57)	0.10	1.15 (0.89–1.51)	0.28	0.87 (0.37–2.05)	0.75	1.72 (1.20–2.47)	0.003	0.81 (0.28–2.30)	0.69	0.56 (0.16–1.98)	0.37	0.91 (0.48–1.73)	0.78	1.34 (0.74–2.41)	0.34
Congestive heart failure	1.16 (0.89–1.52)	0.28	0.97 (0.73–1.29)	0.83	1.51 (0.61–3.75)	0.37	1.44 (0.98–2.12)	0.06	0.70 (0.20–2.43)	0.57	1.16 (0.33–4.13)	0.81	0.85 (0.40–1.80)	0.67	0.90 (0.49–1.66)	0.74
Hypertension	1.41 (1.06–1.86)	0.015	1.21 (0.90–1.62)	0.21	1.16 (0.49–2.74)	0.74	1.65 (1.05–2.60)	0.030	0.66 (0.24–1.78)	0.41	2.12 (0.48–9.39)	0.32	1.36 (0.73–2.54)	0.34	1.48 (0.76–2.88)	0.25
Diabetes mellitus	1.49 (1.17–1.90)	0.001	1.40 (1.09–1.80)	0.001	6.64 (2.80–15.71)	<0.001	1.37 (0.94–1.98)	0.10	2.60 (1.00 –6.74)	0.049	1.74 (0.55–5.48)	0.34	1.15 (0.62–2.12)	0.65	0.78 (0.40–1.52)	0.46
Stroke/TIA	2.62 (2.01–3.38)	<0.001	2.28 (1.75–2.98)	<0.001	2.07 (0.72–6.00)	0.18	3.10 (2.14–4.49)	<0.001	1.46 (0.42–5.10)	0.55	2.29 (0.64–8.11)	0.20	3.31 (1.82–6.01)	<0.001	1.86 (0.96–3.63)	0.07
Vascular disease	1.86 (1.34–2.58)	<0.001	1.48 (1.05–2.09)	0.002	0.74 (0.10–5.51)	0.77	1.65 (1.00–2.73)	0.049	4.95 (1.83–13.40)	0.002	0.95 (0.12–7.23)	0.96	0.91 (0.33–2.53)	0.86	3.52 (1.85–6.71)	<0.001
Type of AF (persistent/permanent)	1.53 (1.18–1.97)	<0.001	1.38 (1.06–1.80)	0.002	1.30 (0.57–2.97)	0.54	1.48 (1.01–2.17)	0.047	4.11 (0.94–17.99)	0.06	3.12 (0.70–13.83)	0.13	1.57 (0.90–2.72)	0.11	1.31 (0.70–2.45)	0.40
Low BMI <18.5 kg/m²	1.71 (1.10–2.64)	0.002	1.68 (1.07–2.63)	0.002	2.13 (0.29–15.81)	0.46	0.94 (0.41–2.14)	0.89	0.95 (0.13–7.13)	0.96	NA^†		3.13 (1.34–7.33)	0.008	3.47 (1.63–7.38)	0.001
Creatinine ≥1.0 mg/dL*	1.39 (1.08–1.77)	0.001	1.13 (0.87–1.48)	0.37	1.56 (0.68–3.60)	0.29	1.43 (0.98–2.09)	0.06	1.46 (0.54–3.93)	0.45	1.02 (0.28–3.66)	0.98	1.53 (0.88–2.65)	0.13	1.07 (0.61–1.90)	0.80
Low hemoglobin**	1.56 (1.22–2.00)	<0.001	1.11 (0.85–1.45)	0.44	1.14 (0.43–3.02)	0.79	1.71 (1.18–2.47)	0.005	2.10 (0.80–5.53)	0.13	0.68 (0.15–3.05)	0.62	1.71 (0.97–3.02)	0.06	1.37 (0.75–2.50)	0.30
No OAC	1.03 (0.66–1.60)	0.89	0.79 (0.50–1.24)	0.30	1.04 (0.39–2.75)	0.94	1.17 (0.66–2.08)	0.59	0.37 (0.12–1.13)	0.08	–		–		–

*28 patients and **109 patients were excluded from the analysis due to missing data on creatine and hemoglobin levels, respectively. ^†None of the patients suffered from a stroke/TIA. CI, confidence interval; HR, hazard ratio. Other abbreviations as in Table 1.

Comparison Between Patients Without and With OACs

The clinical characteristics of patients with OACs (n=6,558) and without OACs (n=462) are summarized in Supplementary Table 2 and the incident rates of ischemic stroke events in Table 2. Incremental increases in the ischemic stroke incident rates were observed with increasing HELT-E₂S₂ scores in both the patients with and without OACs, but the increments were particularly notable in patients without OACs (Supplementary Figure). The C-statistic using the HELT-E₂S₂ score in patients without OACs was 0.657 (95% CI, 0.536–0.760), which was lower than that using the CHADS₂ (C-statistic 0.659; 95% CI, 0.559–0.746, P=0.96 vs. HELT-E₂S₂) or CHA₂DS₂-VASc score (C-statistic 0.701; 95% CI 0.595–0.789, P=0.27 vs. HELT-E₂S₂), and did not reach statistical significance (Table 3).

Discussion

Main Findings

The present analysis of the RAFFINE-SAKURA studies aimed to validate the use of the HELT-E₂S₂ score in Japanese AF patients. The main ﬁndings are described. (1) When stratified by the HELT-E₂S₂ score, a score ≥1 showed that the cumulative incidence of ischemic stroke increased with an increasing score. The C-statistic using the HELT-E₂S₂ score was 0.661, which was numerically higher than with the CHADS₂ and CHA₂DS₂-VASc scores. (2) A similar tendency was observed in the SAKURA AF Registry especially, whereas in the RAFFINE Registry the C-statistic using the HELT-E₂S₂ score was numerically higher than that using the CHADS₂ score but moderately lower than that using the CHA₂DS₂-VASc score. Among the 3 types of facilities (university hospitals, general hospitals, and clinics), the HELT-E₂S₂ score consistently demonstrated better discrimination ability for strokes in the SAKURA AF Registry, but this was seen only in general hospitals in the RAFFINE Study.

Generalizability of the HELT-E₂S₂ Score for Stroke Risk Stratification

Similar to the original study reported by the J-RISK AF research group,²⁰ the C-statistic for the overall HELT-E₂S₂ score (0.661) was higher (but not significant) than that using the CHADS₂ score (0.644) or the CHA₂DS₂-VASc score (0.650). The major difference was that the C-statistic was lower than that seen in the J-RISK AF study (C-statistic 0.681). Overall, the patient characteristics were similar between studies, but the present study patients were slightly older (72.1 vs. 70.2 years), had a higher prevalence of diabetes (27% vs. 21%) and persistent or permanent AF (62% vs. 55%), so they had higher CHADS₂ (1.9 vs. 1.7) and CHA₂DS₂-VASc (3.1 vs. 2.9) scores. As a result, the ischemic stroke event rate was higher than that in the J-RISK AF study (4.1% vs. 2.0%). These trends were particularly evident among the patients in the RAFFINE Study. Also, multivariate analysis revealed that the presence of diabetes and vascular disease were significant factors for ischemic strokes as indicated by adjusted HRs of 1.40 (P=0.001) and 1.48 (P=0.002), respectively (Table 5), but not significant in the J-RISK AF study. Diabetes and vascular disease are well-established lifestyle disease risk factors for ischemic strokes worldwide,²⁷^–³⁰ but the Japanese guidelines suggest they are not significant risk factors for a stroke,⁹ based on a pooled analysis of 3 large registries in Japan¹² that were used to create the HELT-E₂S₂ score.²⁰^,²¹ These differences might have affected the lower C-statistic in the present study population more than in the J-RISK AF study, and the lower C-statistic of this score than that of the CHA₂DS₂-VASc score in the RAFFINE Study. As expected, we found a consistently better discrimination ability of the HELT-E₂S₂ score for strokes than the traditional scores among all three types of facilities in the SAKURA AF Registry, but notably this trend was only evident in general hospitals in the RAFFINE Study. In contrast to those in the SAKURA AF Registry and J-RISK AF study, patients in the university hospitals included the RAFFINE Study exhibited a higher prevalence of multicomorbidity, as indicated by the highest CHADS₂ and CHA₂DS₂-VASc scores. Consequently, most components of those traditional risk scores were found to be significant risk factors for strokes (Table 5). Although this may confer a slightly lower discrimination ability of the HELT-E₂S₂ score than the CHA₂DS₂-VASc score, the C-statistic for the HELT-E₂S₂ score in the university hospitals in the RAFFINE Study was numerically similar to that for the CHA₂DS₂-VASc score. These findings suggest that the HELT-E₂S₂ score has the potential to be useful for stroke risk stratification even for patients receiving care at university hospitals. Despite clinics with the same facility criteria, the C-statistic of the HELT-E₂S₂ score was predominantly higher in the SAKURA AF Registry but lower than that of the other traditional risk scores in the RAFFINE Study. Although most comorbidities such as heart failure, stroke/TIA, and vascular disease were less frequent in patients attending clinics in both the RAFFINE Study and SAKURA AF Registry than in the J-RISK AF study, a notable difference was the highest prevalence of diabetes in the clinics of the RAFFINE Study and the oldest age in the SAKURA AF Registry, surpassing those in other facilities and the J-RISK AF study (Table 4). Therefore, because clinics typically examine patients with fewer comorbidities, the discrimination ability of the HELT-E₂S₂ score in clinics may depend on the prevalence of elderly AF patients and diabetes.

Role of the HELT-E₂S₂ Score for Indication of OAC Initiation

In this study, the annual incident rate of strokes was numerically higher in the no OAC group than in the OAC group for a score 1 on the HELT-E₂S₂ score, and it was even higher for a ≥2 score on the HELT-E₂S₂ score (Supplementary Figure). This study showed that the HRs for scores ≥2 tended to be higher, whereas the J-RISK AF study showed that the HRs for scores ≥2 were significantly higher than those of score 0 (reference) and that increase was significantly greater in no OAC than OAC patients for an HELT-E₂S₂ score ≥2.²⁰ Together with our results this suggests that patients with a score ≥2 should be considered as candidates for OACs, although a larger number of patients is needed to confirm this point.

Study Limitations

First, although this was a large, prospective observational study, enrolment was limited to a subset of the Japanese population. Second, from the statistical viewpoint, the registry did not have a large enough number of patients without OACs (n=462). This limited sample size may have affected the analysis and should be taken into careful consideration when interpreting the results. Finally, even in one of the largest combined registries in Japan, it remains uncertain whether the HELT-E₂S₂ score can be adapted to other cohorts, particularly in Asian people outside of Japan.

Conclusions

Our external validation study revealed that the HELT-E₂S₂ score demonstrated a slightly improved, albeit not statistically significant, performance compared with the CHADS₂ and CHA₂DS₂-VASc scores in predicting the incidence of ischemic strokes among Japanese AF patients. This superior discrimination ability for stroke risk was observed across different types of healthcare facilities (i.e., university hospitals, general hospitals, and clinics) in the SAKURA AF Registry, as well as in general hospitals in the RAFFINE Study. These results suggested that the HELT-E₂S₂ score has greater value in most hospitals in a super-aged society such as Japan. However, it is worth noting that in the RAFFINE Study we did observe a slightly lower discrimination ability of the HELT-E₂S₂ score as compared with the CHA₂DS₂-VASc score in university hospitals, and the CHA₂DS₂-VASc and the CHADS₂ scores in clinics, which may be attributed to the higher prevalence of multiple comorbidities, particularly diabetes.

Acknowledgments

We express our gratitude to all study participants and all supporting staff. We also thank Mr. John Martin for English language editing.

Design

Multicentre, prospective, observational study.

Disclosures

Y.O. has received research funding from Bayer Healthcare and Biosense Webster, Inc., scholarship donation from Boston Scientific Japan, Endowed Courses from Boston Scientific Japan, Japan Lifeline, Fukuda Denshi, Abbott Japan, BIOTRONIK Japan, Medtronic Japan, and has accepted remuneration from Bayer Healthcare, Daiichi-Sankyo, Bristol-Meyers Squibb, AstraZeneca K.K., Ono Pharmaceutical, and Medtronic Japan. K.N. has received research funding and accepted remuneration from Johnson & Johnson K.K. N.M. received lecture fees from Nihon Medi-physics and PDRadio Pharma, and a scholarship donation from PDRadio Pharma. Y.M. received honoraria from Otsuka Pharmaceutical Co, Novartis Japan, and Bayer Japan, and a collaborative research grant from Pfizer Inc. H.D. received speakers’ bureau/honoraria from Novartis Pharma K.K., Bayer Yakuhin, Ltd., Sanofi K.K., Kowa Company, Limited, Taisho Pharmaceutical Co., Ltd., Abbott Medical Japan LLC, Otsuka Pharmaceutical Co., Ltd., Amgen K.K., MSD K.K., Daiichi-Sankyo Company, Limited, Pfizer Japan Inc., FUKUDA DENSHI CO., LTD., Tsumura & Co., and TOA EIYO LTD, and trust research/joint research funds from Philips Japan, Ltd., FUJIFILM Holdings Corporation, Asahi Kasei Corporation, Inter Reha Co., Ltd., TOHO HOLDINGS Co., Ltd, GLORY LTD., BMS K.K, Abbott Japan LLC, and Boehringer-Ingelheim Japan, Inc., as well as scholarship funds from Eisai Co., Ltd., Bayer Yakuhin, Ltd., and Daiichi-Sankyo Company, Limited. T.M. and H.D. are members of Circulation Journal’s Editorial Team. The other authors have no conflicts of interest.

Funding

This work was supported by scholarship funds as follows (in alphabetical order): Abbott Japan, Astellas Pharma, AstraZeneca, Bayer Healthcare, Boehringer-Ingelheim, Boston Scientific Japan, Bristol-Meyers Squibb, Crosswill Medical, Daiichi-Sankyo, Eisai, Fukuda Denshi, FUJIFILM RI Pharma, Japan Lifeline, Kowa Pharmaceutical, Kyowa Hakko Kirin, Mitsubishi Tanabe Pharma, Medical Hearts, Medtronic Japan, Mochida, MSD, Nippon Shinyaku, Otsuka Pharmaceutical, Pfizer, Philips Respironics, Roshe Diagnostics, Sanwa Kagaku Kenkyusho, Sanofi, Shionogi, Sumitomo Dainippon Pharma, Takeda Pharmaceutical, and Nihon Medi-Physics. This study was conducted as investigator-initiated research based on the contract with and financially supported by Bayer Yakuhin Ltd. The funding sources had no roles in the study design, collection, analysis, and interpretation of the data, writing of the report, or decision to submit the article for publication.

IRB Information

The Ethics Committee of Nihon University (RK-210413-11) and the Institutional Review Board of Juntendo University Hospital (H20-0394).

Clinical Trial Registration

UMIN Clinical Trials Registry (URL: http://www.umin.ac.jp/ctr/) Unique identifier: UMIN000009617, UMIN000014420.

Data Availability

The deidentified participant data will not be shared.

Supplementary Files

Please find supplementary file(s);

https://doi.org/10.1253/circj.CJ-23-0318

References

責任著者(Corresponding author)

J-STAGEへの登録はこちら（無料）