a b s t r a c t

Introduction and methods: Early and accurate risk stratification of patients with Atrial fibrillation in the emer- gency department (ED) could aid the physician in determining a timely treatment strategy appropriate to the se- verity of disease. We conducted a retrospective review of 243 adult patients who presented to a Tertiary ED with AF in 2017. Primary outcome studied was 30-day adverse event (a Composite outcome of repeat visit to the ED, Cardiovascular complications, and all-cause mortality).

Secondary outcome studied was 90-day all-cause mortality. We compared the performance of the RED-AF, AFTER and CHA2DS2-VASc score by plotting receiver operating characteristic curves and estimating the areas under curves (AUC), and assessed the potential to further improve the tools with their incorporation of new variables.

Results: Existing scoring tools had poor predictive value for 30-day adverse events, with the RED-AF score performing comparatively better, followed by the AFTER and CHA2DS2-VASc score. All scores performed collec- tively better to predict 90-day mortality, with the AFTER score performing the best, followed by the RED-AF and CHA2DS2-VASc score. By incorporating heart rate at initial presentation to the ED as a variable into the AFTER Score, we generated a Modified AFTER Score with superior Predictive performance above existing scores for 90-day mortality.

Conclusion: Existing scores collectively performed poorly to predict 30-day adverse outcomes, but the AFTER and Modified AFTER score showed good predictive value for 90-day mortality. Further studies should be done to val- idate their use in guiding clinician’s disposition of patients with AF in the ED.

(C) 2020

Introduction

Atrial fibrillation is the most common sustained cardiac arrhythmia [1], with evidence of progressive increases in overall burden and AF-associated mortality between 1990 and 201 and significant public health implications worldwide [2]. While the prevalence of AF is lower in Asia compared to Europe and the United States [3], its prev- alence and associated healthcare cost are currently underestimated, and are expected to increase due to greater awareness, population ageing and increasing prevalence of associated Risk factors and comorbidities

? This research was presented at the 26th Asia Pacific Symposium on Critical Care and Emergency Medicine, Bali on 02 August 2019.

* Corresponding author.

E-mail addresses: [email protected], [email protected] (C.F.C. Yeo), [email protected] (H. Li), [email protected] (Z.X. Koh), [email protected] (N. Liu), [email protected] (M.E.H. Ong).

[4]. Atrial fibrillation in itself is associated with increased risk of major cardiovascular events, heart failure, ischaemic heart disease, chronic kidney disease, and sudden cardiac death, as well as stroke and all- cause mortality [5].

Each patient presenting to the emergency department (ED) with atrial fibrillation has varying Severity of disease and risk of complica- tions which may not be clinically apparent at time of presentation. Early and accurate risk stratification of these patients will aid the physi- cian in determining a timely treatment strategy appropriate to the se- verity of disease. To date, neither the 2014 American Heart Association (AHA), The American College of Cardiology Foundation (ACA), the Heart Rhythm Society (HRS) guidelines, nor the 2016 European Society of Cardiology (ESC) guidelines for the management of atrial fibrillation specifically address ED treatment and disposition [6]. On this note, Barrett et al. designed 2 separate prediction models, the Atrial Fibrilla- tion and Flutter Outcomes & Risk Determination (AFFORD) score [7] and Risk Estimator decision aid for Atrial Fibrillation (RED AF) score

https://doi.org/10.1016/j.ajem.2020.06.018

0735-6757/(C) 2020

[8] to predict risk of adverse events at 30-days in patients presenting with AF to the ED. Atzema et al. also created the Atrial Fibrillation in the Emergency Room (AFTER) [9] score which had a complex version as well as a pragmatic version known as TrOPs-BAC, to predict 30-day all cause mortality. Each of these studies was done at a single academic institution, lacking further validation in a variety of ED settings to date. This present report aims to compare the performance of the RED AF and pragmatic AFTER score, as well as the well-established CHA2DS2- VASc score [10] in their ability to accurately risk-stratify patients with AF in the ED, while assessing the potential to further improve the tools with their incorporation of new variables. The AFFORD and com- plex AFTER score were not included in this study due to the complexity of the score which makes it difficult to calculate without a web-based calculator. The objective of this study is to validate a simple, easy-to- recall scoring tool to guide ED physicians on their management and dis-

position of AF patients.

Methods

Study design and patient recruitment

We conducted a retrospective, non-randomized, observational co- hort validation study of 243 adult patients who presented to the ED of a tertiary hospital between January to December 2017. Patients were el- igible if they were at least 18 years of age and had AF as either a primary or secondary diagnosis in the ED. Patients who presented with supra- ventricular tachycardia, atrial flutter, or whose diagnosis of AF was un- certain were excluded from the study. If a patient made multiple visits to the ED for AF, their First visit in the year of 2017 was taken as the pri- mary visit, while subsequent visits were considered under outcomes.

The study was approved by the institution’s Centralised Institutional Review Board (CIRB) and waiver of informed consent was obtained for this study. As this study is based on retrospective review of medical re- cords, patients’ baseline characteristics and clinical parameters were ex- tracted from the ED’s electronic medical records system onto an excel spreadsheet. All variables used in the CHA2DS2-VASc, RED-AF and AFTER scores are depicted in Table 1, and these scores were calculated and applied retrospectively to the cohort for the present analysis. Base- line characteristics collected included patient demographics (age,

Table 1

Variables in the CHA2DS2-VASc, RED-AF and pragmatic AFTER score.

Variables CHA2DS2-VASc RED-AF Pragmatic score score AFTER score

Age X X X

Gender X X

Presence of other acute ED diagnosis X

Presenting symptoms

Shortness of breath X

Palpitations X

Past medical history

Smoking X

Hypertension X X

Diabetes mellitus X

Chronic obstructive pulmonary disease X X

Congestive heart failure X X X

Stroke/transient ischemic X attack/thromboembolism

Vascular disease X

bleeding risk X

Home use of diuretic X

Home use of beta-blocker X

Findings at ED

Peripheral edema on examination X

Inadequate ventricular Rate control at 2 h X

Raised troponin levels X

gender and race), Presenting complaints at the ED, past medical history, vital signs at triage and 2-hour heart rate, as well as investigation re- sults. Cut-off values of troponin, serum creatinine and Hemoglobin levels were determined by the institution’s guidelines.

Outcomes

Primary outcome studied was the occurrence of one or more 30-day composite adverse event which consists of repeat visit to the ED for AF, cardiovascular complications (ventricular arrhythmias, cardiac arrest, stroke, transient ischemic attack, acute coronary syndrome, new onset heart failure, acute decompensating heart failure and unscheduled pacemaker insertion), AF and non-AF related mortality. Secondary out- come studied was the presence of 90-day all-cause mortality. The out- comes of each patient were traced through the hospital’s electronic medical health records system. All information possibly indicating that an adverse event was met were further investigated by examining med- ical records from outpatient clinics and different specialties within the hospital. If no repeat visits or medical notes after first visit to the ED were found in the system and there was no recorded date of demise, it was assumed that the patient did not satisfy any of the outcomes.

Intervention - the different scores used

Given that our objective is to validate a simple, easy-to-remember risk Prediction tool, we excluded the AFFORD score and complex AFTER score from this study due to the large number of variables in- volved and complexity in calculating these scores in the absence of an available web-based calculator. Furthermore, the large number of vari- ables in these scores meant that certain variables were not well- documented in the ED medical records, making the calculation of the complete scores difficult. Instead, we chose to focus this study on the comparison of the predictive performance of RED-AF score, pragmatic AFTER score, and CHA2DS2-VASc score.

The RED-AF score is a recently proposed score based on the AFFORD cohort assembled from the Vanderbilt University Medical Center, a university-affiliated, tertiary care ED in the United States. It assigns points for 12 clinical variables in ED patients with atrial fibrillation or flutter (see Table 1), and predicted the risk of an adverse event at 30 days (see Outcomes) based on total points [11]. It showed modest pre- dictive discrimination in the derivation cohort, and was later prospec- tively validated with similar performance as the original cohort. The AFTER score was based on a retrospective study of 3510 patients from 24 hospital EDs in Ontario, Canada with 30-day all-cause mortality as the main outcome studied. The pragmatic model of the AFTER score in- cludes only 6 variables - a positive Troponin result, other acute ED diag- nosis, pulmonary disease (chronic obstructive pulmonary disease), bleeding risk, aged 75 years or older, and congestive heart failure (see Table 1). Even without the extra variables present in the complex model, it showed good discriminatory ability in its derivation and vali- dation cohort [9]. The CHA2DS2-VASc score was included in this study as it is the most widely used scoring tool to guide clinicians on long- term anticoagulation use in patients with AF, based on their risk of stroke or thromboembolism. It is the score currently recommended by the current 2016 ESC guidelines [12] and the 2014 AHA/ACC/HRS guide- lines [13] as the main score to assess stroke risk in patients with non- valvular AF.

Statistical analysis

Statistical analysis was done using IBM SPSS Statistics version 23.0 software. Descriptive statistics were performed for the baseline charac- teristics of patients, and all categorial variables were analysed using chi-square test. Univariate logistic regression and multivariate logistic regression analysis were done for the primary and secondary outcomes. We compared the predictive performance of the CHA2DS2-VASc, RED-

AF and pragmatic AFTER score by examining their Receiver Operating Characteristic (ROC) curves and calculating the areas under the ROC curve (AUCs), also known as the c-statistic, and its corresponding 95% confidence intervals (CI).

While the primary focus of our analyses was to compare the perfor-

mance of the CHA2DS2-VASc, RED-AF and pragmatic AFTER score, we also examined if adding new variables to the best-performing scoring tool could improve its predictive performance. Using the variables found to be statistically significant through univariate and multivariate logistic regression, we input each single variable into the scores follo- wed by a combination of variables. We then examined the c-statistics of these modified scores in comparison with the existing scoring tools using the full range of point scores, followed by the cut-off points for supposedly “low” risk of 90-day mortality corresponding with a cut- off of at least 95% sensitivity.

Results

243 patients presented to the ED with a primary or secondary diag- nosis of AF in January to December 2017. The mean age of these patients was 62.3 years +-12.9, and 50.2% were male. Of these 243 patients, 207 patients (85.2%) were admitted, and mean length of stay in the hospital was 5.12 days +- 5.09. Univariate analysis of patient characteristics in re- lation to occurrence of our primary and secondary outcomes is shown in Table 2. A total of 110 patients met our primary outcome of at least one 30-day adverse event, while 21 patients met our secondary outcome of 90-day all-cause mortality. The breakdown of frequencies of 30-day ad- verse events and 90-day cause of death are shown in Table 3. Of note, heart failure collectively (new onset heart failure and acute decompen- sated heart failure) was the most common complication, making up 60.9% (n = 67) of all 30-day adverse events occurrences. The majority cause of death was unknown (42.6%, n = 9), as out of hospital demise or demise in other institutions outside of the hospital of study were un- traceable as limited by our study’s IRB.

In Fig. 1, the ROC curves of the CHA2DS2-VASc, RED-AF and prag- matic AFTER score for predicting 30-day adverse events are shown. Col- lectively, all 3 scores had poor predictive performance for 30-day adverse event. Using the entire point score range, the RED-AF score had a C-index of 0.682 (95% CI = 0.575-0.712) and performed compar- atively better than the pragmatic AFTER score (C-index = 0.644; 95% CI

= 0.613-750) and CHA2DS2-VASc score (C-index = 0.565; 95% CI =

0.493-0.637). Variables associated with 30-day adverse events through multivariate logistic regression were symptoms of heart failure at ED, history of heart failure, raised BNP and chest radiograph changes. This is in keeping with heart failure being the most commonly experienced 30-day adverse event.

It was found that all 3 scores collectively performed better in predicting 90-day all-cause mortality. Of note, the pragmatic AFTER score had superior performance with C-statistic of 0.836 (95% CI = 0.764-0.907), followed by the RED-AF score (C-statistic = 0.721; 95% CI = 0.620-0.823) and the CHA2DS2-VASc score (C-statistic = 0.662; 95%CI = 0.536-0.788). Using multivariate logistic regression, it was found that variables associated with 90-day all-cause were presence of other Acute diagnosis at the ED, history of heart failure, raised tropo- nin levels, and tachycardia with heart rate N150 beats per minute. Out of these 4 variables, the first 3 variables made up 3 out of 6 variables in the pragmatic AFTER score.

To see if we could further improve the Predictive ability of the prag- matic AFTER score for 90-day all-cause mortality, we added several fac- tors which showed association with 90-day all-cause mortality either through univariate or multivariate logistic regression (tachycardia with heart rate N150 beats per minute, history of ischemic heart disease, blood pressure, and hemoglobin level) individually to the score, followed by a combination of these variables. It was found that adding tachycardia with heart rate N150 beats per minute to the pragmatic AFTER score, with 2 points assigned to a positive finding and 0 points

to a negative finding, gave the best improvement in its predictive value (C-statistic = 0,878, 95% CI = 0.811-0.945; P-value = 0.002). We named this the Modified AFTER score. In Fig. 2, the ROC curves of the CHA2DS2-VASc, RED-AF, pragmatic and Modified AFTER score for predicting 90-day all-cause mortality are shown.

Table 4 shows the comparison of performance of CHA2DS2-VASc, RED-AF, pragmatic AFTER and Modified AFTER scores at cut-offs for “low-risk” of 90-day mortality. We defined “low risk” as the absolute level of safety that will miss no more than 5% of all patients with 90- day mortality, by taking the lowest score corresponding to at least 95% sensitivity on their ROC curves. At these cutoffs, the CHA2DS2-VASc score classified 16 patients as “low risk”, of whom 0 had a demise within 90 days. The RED-AF score classified 67 patients as “low risk” with 1/67 (1.5%) patients suffering a demise within 90 days. The pragmatic AFTER score classified 125 patients as “low risk” with 1/125 (0.8%) suffering a 90-day demise. Lastly, the modified AFTER score identified 118 patients as “low risk” with 1/118 (0.8%) suffering a 90-day demise.

In 2017, our ED admission rates for AF stood at 85.4%. Using the cut- offs for low risk of 90-day mortality for the above scores, the pragmatic AFTER and Modified AFTER scores had the potential to reduce admission rates to 48.6% and 51.4% respectively. This is assuming that all patients who fell under the “low risk” categories had no other concomitant med- ical diagnoses requiring admission. On the other hand, the RED-AF score would only reduce admission rates to 72.4%, and the small number of patients who fell into the low risk category in the CHA2DS2-VASc score actually increases admission rates to 93.4%.

Discussion

Atrial fibrillation is a disease with high heterogeneity in terms of presentation, severity and complications amongst different individuals. Accurate risk stratification of these patients in the ED is crucial to guide the allocation of limited Hospital resources and manpower. High-risk patients can be worked-up and treated earlier with more aggressive treatment strategies and monitoring, while identified low-risk patients can be managed without excessive investigations and unnecessary hospitalisations. Our high admission rate (85.2%) accompanied by the relatively low incidence of 30-day (4.5%) and 90-day mortality (7.8%) suggest that there is room to discharge many more of these patients home after observation in the ED. An ideal risk stratification tool in the ED is one that is accurate, user friendly, easy to remember, and com- posed of variables that are easily and routinely obtained in the ED set- ting. In addition, it should identify the largest number of patients at truly low risk of adverse outcomes without compromising safety. How- ever, no guidelines address the use of a Risk scoring tool for ED disposi- tion of AF patients to date [1,12,14]. To our knowledge, our study is the first study to externally validate the RED-AF and pragmatic AFTER score, and compare the performance of these scores against each other.

There were two overlaps of variables in all 3 scores: age and conges- tive cardiac failure. Our study had similar findings to associate presence of history of congestive cardiac failure with poorer outcomes. This was found likewise across multiple studies. In the EVEREST Trial, it was found that patients hospitalized for heart failure with atrial fibrilla- tion or flutter on initial ECG was associated with increased mortality (hazard ratio = 1.23; 95% CI,=1.04-1.46) as well as cardiovascular mortality/heart failure hospitalization (hazard ratio = 1.26; 95% CI = 1.07-1.47) [15]. Similarly, in the study of Wang, it was found that indi- viduals with AF or congestive heart failure who subsequently develop the other condition have a poor prognosis for mortality [16]. Several of these studies focused on the temporal relationship of the develop- ment of heart failure and AF, where it was found that the development of AF after heart failure conferred the largest increased risk of death compared to heart failure patients without AF [17,18]. Nonetheless, there is strong evidence of the bi-directional relationship shared be- tween heart failure and AF, and their joint influence on mortality [16,19,20]. Surprisingly, age was not found to be associated with 30-

Table 2

Baseline characteristics of study cohort and rate of 30-day adverse events and 90-day all-cause mortality.

Absence of	Presence of	OR (95% CI)	P value	Absence of	Presence of	OR (95% CI) P value
30-day	30-day			90-day	90-day
adverse	adverse events			all-cause	all-cause
events				mortality	mortality

Age - - 1.003 (0.984,

1.023)

Length of stay - - 1.046 (0.993, 1.101)

0.7415 - - 1.047 (1.007,

1.089)

0.0913 - - 1.113 (1.042,

1.188)

0.0199

0.0014

Age

<=64 47 36 Reference 78 5 Reference

1.343 (0.711,

65-74	35	36
>=75	51	38

2.537)

0.973 (0.532,

1.78)

Gender, male 66 55 1.015 (0.613,

1.682)

0.3639 67 4 0.931 (0.240,

3.610)

0.9286 77 12 2.431 (0.818,

7.229)

0.9535 108 13 1.715 (0.684,

4.301)

0.9180

0.1101

0.2499

Race

Chinese 96 84 Reference 162 18 Reference

Malay	23	15	0.745 (0.365, 1.521)	0.4194	36	2
Indian	1	7	8 (0.964, 66.362)	0.0541	8	0
Others	13	4	0.352 (0.11, 1.12)	0.0770	16	1

0.5 (0.111, 2.252) 0.3666

0 0.9912

10.244 (2.928,

Presence of other acute diagnoses at the ED	82	18
Disposition from the ED
Discharged	90	53
EOWa	43	57
Warded AORb	14	2
HDUc	11	7
ICUd	101	81

35.835)

0.563 (0.07,

4.494)

0.0003	82	18	10.244 (2.928, 35.835)	0.0003
	16	0	Reference
0.0023	18	0	-	1.0000
	165	17	-	0.9920
	2	0	-	1.0000
0.0959	13	1	-	0.9922
0.0251	8	3	-	0.9914
0.2249
0.0005	79	5	Reference

0.5874

Reference

2.251 (1.336,

3.792)

Reference

4.455 (0.767,

25.86)

5.614 (1.24,

25.417)

Type of AF 1 1 7 (0.302, 162.211)

42 (5.111,

Newly diagnosed AF	2	12
Known paroxysmal AF Chronic/persistent AF	4	7
resenting complaints Shortness of breath 48 70
Palpitations	50	29
Chest pain	32	28
Symptoms of heart failure	5	37
ast medical history Smoking Non-smoker/quit N1y 122 100
Active/quit b1y	11	10
Diagnosed hypertension	96	83
Diagnosed diabetes mellitus	72	4
Prior stroke/TIA	19	25
Prior hemorrhage/bleeding	14	17
Heart failure/left ventricular dysfunction	46	12
Myocardial infarction/ischemic heart disease	74	12
Vascular disease	80	12
Aortic plaque	21	7
Malignancy	25	6
Chronic respiratory disease	26	3
COPD	6	0

345.115)

12.25 (1.788,

83.948)

1.901 (0.67,

5.389)

0.2271

0 0.9906

0.0107	133	16
	10	0
b0.0001	107	11
0.0639	77	2
0.8019	58	2
b0.0001	38	4
	205	17
0.8209	17	4
0.5643	163	16
0.2141	72	4
0.0913	38	6
0.2542	24	7
0.0005	18	40
0.0343	42	44
0.0626	48	44
0.0024	18	10
0.0296	19	12
0.7285	18	11
0.9884	4	2

P

3.099 (1.832,

5.241)

0.594 (0.343,

1.031)

1.078 (0.6, 1.934)

12.975 (4.884,

34.473)

P

Reference

1.109 (0.453,

2.718)

1.185 (0.666,

2.109)

1.182 (0.483,

2.896)

0.198 (0.045,

0.873)

0.298 (0.067,

1.317)

1.139 (0.363,

3.576)

Reference 2.837 (0.858,

9.385)

1.158 (0.406,

3.301)

0.7142

0.0324

0.1103

0.8231

0.0875

0.7833

0.49 (0.159, 1.51) 0.49 (0.159, 1.51) 0.2141

1.765 (0.913,

3.412)

1.554 (0.728,

3.314)

5.101 (2.027,

12.842)

2.667 (1.075,

6.613)

2.367 (0.956,

5.86)

4.786 (1.739,

13.173)

3.152 (1.121,

8.866)

1.937 (0.706,

5.313)

4.125 (1.515,

11.228)

3.651 (1.943,

6.859)

1.444 (0.852,

2.45)

1.181 (0.702,

1.986)

0.639 (0.282,

1.448)

0.735 (0.34,

1.589)

0.1992

0.0055

0.0001

0.1726

0.5318

0.2831

0.4335

1.256 (0.346,

4.559)

0.71 (0.32, 1.575) 0.3993

0 (0, Inf)

0.597 (0.107,

3.324)

0.5561

Table 2 (continued)
Absence of 30-day adverse events	Presence of 30-day adverse events	OR (95% CI)	P value	Absence of 90-day all-cause mortality	Presence of 90-day all-cause mortality	OR (95% CI) P value

Renal disease 37 4 1.176 (0.374,

3.697)

Valvulopathy 27 2 0.76 (0.168,

3.447)

Cardiac ablation 7 1 1.536 (0.18,

13.116)

0.7809 21 20 1.185 (0.605,

2.321)

0.7223 10 19 2.568 (1.14,

5.786)

0.6950 5 3 0.718 (0.168,

3.073)

0.6204

0.0229

0.6549

electrical cardioversion 10 1 1.06 (0.129, 8.71) 0.9568 7 4 0.679 (0.194,

2.384)

0.1174	4	4
0.4105	22	30
0.5723	80	68
0.0414	25	43
	39	35
0.8575	5	3
0.1208	38	36
0.0340	51	36

0.5459

Dementia	6	2
Warfarin with history of labile INR or INR N3	49	3
Beta blocker use	134	14
Diuretic use	58	10

Systolic blood pressure/mmHg

Normal (SBP = 90-120)	63	11
Hypotensive (SBP b90)	7	1
Pre-hypertensive (SBP = 121-140)	69	5
Hypertensive (SBP N140) Heart rate (/min)	83	4

3.789 (0.715,

20.083)

0.588 (0.166,

2.08)

1.313 (0.51,

3.384)

2.571 (1.038,

6.368)

Reference 0.818 (0.091,

7.318)

0.415 (0.137,

1.261)

0.276 (0.084,

0.908)

1.217 (0.297,

4.982)

1.892 (1.017,

3.52)

1.073 (0.639,

1.801)

2.773 (1.553,

4.95)

Reference 0.669 (0.149,

3.003)

1.056 (0.554,

2.012)

0.787 (0.421,

1.469)

Reference

2.522 (0.335,

18.984)

2.969 (1.646,

5.355)

2.751 (1.064,

7.114)

0.7848

0.0441

0.7909

0.0006

0.5994

0.8693

0.4513

0.3691

0.0003

0.0368

Reference

Normal (HR = 60-100)	79	2	Reference		58	23
Bradycardia (HR b60)	4	0	0	0.9914	2	2
Tachycardia (HR = 101-150)	124	11	3.504 (0.757, 16.229)	0.1089	62	73
Extreme tachycardia (HR N150) Respiratory rate (/min)	15	8	21.067 (4.066, 109.137)	0.0003	11	12
Normal (RR = 12-20)	193	15	Reference		119	89
Tachypnea (RR N20) Oxygen saturation	29	6	2.662 (0.956, 7.413)	0.0609	14	21
Normal (SpO2 N/= 95%)	207	21	Reference		125	103
Hypoxia (SpO2 b95%)	15	0	0	0.9881	8	7

2.006 (0.967,

4.161)

Reference

1.062 (0.373,

3.027)

Temperature
Afebrile	209	17
Febrile	13	4

Reference 122 104 Reference

0.0617

0.9105

3.783 (1.111,

12.875)

0.0333 11 6 0.64 (0.229, 1.79) 0.3949

Adequate ventricular rate control (2 h heart rate b100 bpm)

110 6 0.407 (0.152,

1.088)

0.0732 73 43 0.527 (0.316,

0.881)

0.0146

Peripheral edema on examination 60 7 1.35 (0.52, 3.506) 0.5377 15 52 7.053 (3.664,

13.575)

b0.0001

HB b14.0 g/dL 150 19 4.497 (1.019,

19.836)

Serum creatinine N101 U/mol 79 11 1.963 (0.799,

4.827)

0.984 (0.399,

BNP raised	96	9
CXR result Normal	76	1

2.431)

0.0471 87 82 1.606 (0.915,

2.819)

0.1416 38 52 2.321 (1.362,

3.956)

0.9728 34 71 5.301 (3.053,

9.204)

0.0991

0.0020

b0.0001

Reference 58 19 Reference

0.5158	19	12
0.0066	32	20
0.0371	24	59
0.0001	38	49

2.533 (0.153,

Acute changes seen	30	1
Chronic changes seen	42	10
Both acute and chronic changes seen	74	9

41.822)

18.095 (2.238,

146.287)

9.243 (1.143,

74.781)

Troponin raised 70 17 9.229 (2.995,

28.437)

1.928 (0.792,

4.691)

1.908 (0.891,

4.088)

7.504 (3.717,

15.151)

2.008 (1.18,

3.418)

0.1479

0.0966 b0.0001 0.0102

Bolded if P value b 0.05 indicating it is statistically significant.

^a EOW - emergency observation ward.

^b AOR - at own risk.

^c HDU - high dependency unit.

^d ICU - intensive care unit.

Table 3

Breakdown of 30-day adverse events and cause of death for 90-day all-cause mortality.

No. (%) (n = 243)

was 65.2, significantly lower than Framingham cohort (mean age = 75) [23], ORBIT-AF cohort (mean age = 71) [24], and slightly lower than the AFFORD (mean age = 68) and AFTER (mean age = 69) cohort, which

30-Day adverse events Repeat visit to the ED for AF	21 (8.642%)	collectively identified age as an important predictor of adverse out- comes. The reason for this discrepancy is unclear at this point of time.
Ventricular arrhythmia	0 (0%)	It may be a reflection of a younger demographic of patient presenting
Cardiac arrest	1 (0.412%)	to this particular ED, or an interplay of other sociodemographic factors
Stroke Transient ischemic attack Acute coronary syndrome New onset heart failure	13 (5.350%) 2 (0.823%) 9 (3.704%) 41 (16.872%)	that requires further investigation. As expected, the RED-AF score performed best in the study of our primary outcome, given that our primary outcome was modelled
Acute decompensating heart failure	26 (10.700%)	after the 30-day composite outcome used in the derivation and val-

Unscheduled pacemaker insertion 4 (1.646%)

AF related mortality 2 (0.823%)

Non-AF related mortality 9 (3.704%)

More than 1 adverse event 18

90-Day all-cause mortality

AF-related mortality 2 (0.823%)

Ischemic heart disease 2 (0.823%)

Malignancy 3 (1.235%)

Pneumonia 4 (1.646%)

Urinary tract infection 1 (0.412%)

Unknown^a 9 (3.704%)

^a Unknown= out of hospital demise/COD untraceable within SGH medical records/COD unknown for coroner’s investigation.

day adverse events, and only showed statistically significant association with 90-day all-cause mortality in our study. This is in spite of the widely accepted association of age as both a risk factor and predictor of poor prognosis in AF [5,21,22]. The mean age of our study population

idation studies for the RED-AF score. Despite performing the best, the c-index of the RED-AF score was only modest (c-index = 0.682) and falls below a c-index of 0.8 that is routinely used as a benchmark value to determine if a scoring tool has good discrimina- tive value. The score placed the highest weightage on smoking, assigning 52 points to recent history of smoking in the past 1 year. However, there remains conflicting evidence regarding the associa- tion of smoking and adverse outcomes, as reflected by the recent studies by Kwon et al. [25], Zhu [26] and Pawar [27]. Our study also did not find significant association between smoking and the occur- rence of adverse events. This may be in part due to the small propor- tion of declared active smokers within our study population (8.6%). The RED-AF score was the most complex score out of all 3 scores, composing of 12 variables with a wide range of point allocation for each variable. Without the adaptation of this score into an online cal- culator, it will be difficult to advocate its use in the ED given the large number of variables one has to subject to memory and the tedious, time-consuming calculation of the score.

Fig. 1. ROC curves and AUCs of the CHA2DS2-VASc, RED-AF and pragmatic AFTER score in predicting 30-day adverse events.

Fig. 2. ROC curves and AUCs of the CHA2DS2-VASc, RED-AF, pragmatic AFTER and Modified AFTER score in predicting 90-day all-cause mortality.

Table 4

Comparison of performance of the CHA2DS2-VASc, RED-AF, pragmatic and Modified AFTER score at a cut-off of at least 95% sensitivity.

	At a cut-off of at least 95% sensitivity
		CHA2DS2-VASc score	RED-AF score	Pragmatic AFTER score	Modified AFTER score
	Corresponding cut-off for “low risk” of 90-day all-cause mortality	0	b/=110	b/=2	b/=2
	No. of pts. classified as “low risk” according to cut-off (%)	16 (6.6%)	67 (27.6%)	125 (51.4%)	118 (48.6%)
	No. of 90-day all-cause mortality in “low-risk” group (%) Negative predictive value (NPV)	0 (0.0%) 1 (0.760-1.000)	1 (1.5%) 0.985 (0.909-0.999)	1 (0.8%) 0.992 (0.950-1.000)	1 (0.8%) 0.992 (0.941-1.000)

Our comparison of the CHA2DS2-VASc, RED-AF, pragmatic AFTER and Modified AFTER scores in 243 patients with AF in the ED showed that at no lower than 95% sensitivity, the pragmatic AFTER score identi- fied the most number of patients at low-risk of 90-day all-cause mortal- ity. Apart from the Modified AFTER score, it was also the score that showed best predictive performance for 90-day mortality. It was unsur- prising that the pragmatic AFTER score performed well to predict 90- day all-cause mortality, given that potential variables considered in the design of the score was based on the previous work of Atzema et al. to identify factors associated with 30-day and 90-day mortality [9,28]. Unique to the pragmatic AFTER score were the inclusion of car- diac biomarkers (troponin), bleeding risk and presence of other acute ED diagnosis, with good evidence supporting their inclusion. Of note, these 3 variables were found to be likewise statistically significant in our multivariate analysis of 90-day mortality. Bleeding risk is an impor- tant prognostic factor in mortality, as not only are patients more prone to Life-threatening hemorrhage, they are also less likely to receive anticoagulation and are at increased risk of severe, life-threatening

stroke [29]. As identified in the original AFTER study, the association of a secondary acute ED diagnosis with mortality had good face validity, given that several studies have shown that atrial fibrillation worsens prognosis in the setting of other Acute diseases, such as acute myocar- dial infarction [30] and sepsis [31]. The association of cardiac troponin concentrations with stroke and other cardiovascular events was initially reported by the Randomized Evaluation of Long Term Anticoagulant Therapy (RE-LY) biomarker substudy [32], and was later confirmed and further expanded in the Apixaban for the Prevention of Stroke in Subjects with Atrial Fibrillation (ARISTOTLE) study. In the ARISTOTLE study, it was shown that the concentrations of High-sensitivity troponin I or T, even within the normal range, were associated with increased risk of stroke, major bleeding events, cardiac and all-cause mortality [33,34]. The design of the pragmatic AFTER score was targeted at being easy to recall and simple to calculate, without the need of a web-based calcula- tor. Its superior performance over the CHA2DS2-VASc and RED-AF score shows that more variables do not always result in better stratification of patients.

In the study of both outcomes, the CHA2DS2-VASc score consistently performed the worst, which could be explained by the fact that its de- sign was to predict stroke risk rather than composite adverse events or mortality. However, there have been several studies proposing the use of the CHA2DS2-VASc score to predict mortality and other adverse events apart from stroke or thromboembolism [35,36]. In the study by Larsen, it was found that at 5-year follow-up, the c-index of the CHA2DS2-VASc score for predicting mortality was 0.63 (95% CI = 0.59-0.67) [37], similar to that of our study findings for 90-day mortal- ity. Corresponding to 95% sensitivity, “low risk patients” were defined to have a CHA2DS2-VASc score of 0, in keeping with ESC guidelines which restricted low stroke risk to patients with CHA2DS2-VASc score of 0. However, at this cutoff score of 0, the CHA2DS2-VASc score included only 16 patients into the “low risk” group. This strict cutoff value re- sulted in majority of patients falling out of the “low risk” group, which could result in a larger proportion of patients receiving extensive workup and unnecessary hospitalisations. On the contrary, the prag- matic AFTER score was able to include almost 8 times the number of pa- tients (n = 125) into their “low risk” group, with incidence of 90-day mortality within the “low-risk” group only increasing by 1.

In our multivariate analysis, it was found that tachycardia with heart rate N150 bpm during initial triage at the ED was significantly associated with 90-day mortality. By adding this variable to the pragmatic AFTER score to derivea Modified AFTER score, we were able to significantly im- prove its predictive performance for 90-day mortality. There are cur- rently limited studies associating extreme tachycardia in atrial fibrillation with increased mortality but physiologically, we are able to understand that patients in fast AF are likely to have reduced cardiac output due to poor ventricular filling, and are at higher risk of develop- ing tachycardia-induced cardiomyopathy [38]. Also, patients in AF with existing tachycardia are not achieving adequate rate or Rhythm control, predisposing them to increased risk of cardiovascular complications and mortality. In a subgroup analysis on 30-day all-cause mortality, the Modified AFTER score was also able to improve the predictive perfor- mance of the pragmatic AFTER score in predicting 30-day all-cause mor- tality, as c-statistic improved from 0.860, 95% CI = 0.776-0.944 in the pragmatic AFTER score to 0.891, 95% CI = 0.838-0.944, showing the po- tential of both scores to predict shorter-term mortality in AF patients that may be more relevant for ED decision-making. In addition, both the pragmatic and Modified AFTER score had the best negative predic- tive values of 0.992. The False negative rate of 0.8% for both scores fall within the acceptable range determined by most clinicians of 1-2% [39,40].

The combination of being accurate, user friendly, and safe makes the pragmatic and Modified AFTER score the favoured tool in our study to predict short-term mortality in AF patients in the ED, although further studies will be needed to validate the Modified AFTER score in other cohorts.

Limitations

Our study is not without its limitations. Given that this is a retrospec- tive study of a relatively small number of patients, it is subject to hidden confounding bias and may not be representative of the local AF popula- tion. We recommend external validation of the RED-AF, pragmatic and Modified AFTER scores in a larger, prospective trial. In addition, given that only this hospital’s medical records were accessible, outcomes of patients may be under reported if they followed up with care at other healthcare institutions. To overcome this, we suggest the establishment of a local AF registry targeted at information sharing between institu- tions, which will allow a nationwide, multicenter study to be conducted. Finally, it is also unclear at this point whether 90-day mortality is a useful outcome to acutely risk-stratify patients in the emergency de- partment. Even though the pragmatic and Modified AFTER score performed well to predict 30-day mortality, the number of patients sat- isfying that outcome is small (n = 11). Future studies should aim to

validate these scores more extensively in its predictive ability for 72- hour or 30-day adverse events or mortality.

Conclusions

In summary, all 3 scores performed poorly to predict 30-day com- posite adverse events of AF patients in the ED, but the pragmatic AFTER score performed well to predict 90-day all-cause mortality in our study population. Adding heart rate at initial presentation to derive a Modified AFTER score improved its predictive performance without compromising on safety. Further studies are needed to assess the feasi- bility of incorporating these proposed tools into clinical guidelines to guide ED physicians’ management and disposition of AF patients.

Sources of support

This project was conducted under the Department of Emergency Medicine, Singapore General Hospital. It did not receive any equipment or grants from external parties.

CRediT authorship contribution statement

Chloe F.C. Yeo: Investigation, Formal analysis, Methodology, Data curation, Writing - original draft, Writing - review & editing, Visualization. HuiHua Li: Formal analysis. Zhi Xiong Koh: Project ad- ministration, Data curation, Resources. Nan Liu: Supervision, Conceptu- alization, Formal analysis, Resources. Marcus E.H. Ong: Supervision, Conceptualization, Writing - review & editing, Resources.

References

1. January CT, et al. 2014 AHA/ACC/HRS guideline for the management of patients with atrial fibrillation: executive summary: a report of the American College of Cardiol- ogy/American Heart Association Task Force on Practice Guidelines and the Heart Rhythm Society. Circulation. 2014;130(23):2071-104 Dec.
2. Chugh SS, et al. Worldwide Epidemiology of atrial fibrillation: a global burden of dis-

   ease 2010 study. Circulation. 2014;129(8):837-47 Feb.

   Kodani E, Atarashi H. Prevalence of atrial fibrillation in Asia and the world. J Arrhyth-

   mia. 2012;28:330-7.

   Omar R, et al. Atrial fibrillation in Singapore and Malaysia: current trends and future prospects; 2011.

3. Staerk L, Sherer JA, Ko D, Benjamin EJ, Helm RH. Atrial fibrillation: epidemiology, pathophysiology, and clinical outcomes. Circ Res. 2017;120(9):1501-17 Apr.
4. Kirchhof P, et al. 2016 ESC guidelines for the management of atrial fibrillation devel- oped in collaboration with EACTS. Eur J Cardio-Thoracic Surg. 2016;50(5):e1-88 Nov.
5. Barrett TW, et al. The AFFORD clinical decision aid to identify emergency depart- ment patients with atrial fibrillation at low risk for 30-day adverse events. Am J Cardiol. 2015;115(6):763-70 Mar.
6. Barrett TW, Jenkins CA, Self WH. Validation of the Risk Estimator Decision Aid for Atrial Fibrillation (RED-AF) for predicting 30-day adverse events in emergency de- partment patients with atrial fibrillation. Ann Emerg Med. 2015;65(1):13-21.e3 Jan.
7. Atzema CL, et al. A clinical decision instrument for 30-day death after an emergency department visit for atrial fibrillation: the atrial fibrillation in the emergency room (AFTER) study presented at the Canadian Cardiovascular Congress, October 2014, Vancouver, British Co. Ann Emerg Med. 2015;66(6):658-668e6.
8. Olesen J, Torp-Pedersen C, Hansen M, Lip G. The value of the CHA2DS2-VASc score for refining stroke risk stratification in patients with atrial fibrillation with a CHADS2 score 0-1: a nationwide cohort study. Thromb Haemost. 2012;107(06): 1172-9 Nov.
9. Barrett TW, et al. A clinical prediction model to estimate risk for 30-day adverse events in emergency department patients with symptomatic atrial fibrillation. Ann Emerg Med. 2011;57(1):1-12 Jan.
10. Kirchhof P, et al. 2016 ESC guidelines for the management of atrial fibrillation devel- oped in collaboration with EACTS. Russ J Cardiol. 2017;147(7):7-86 Werner Budts.
11. Alpert JS, et al. 2014 AHA/ACC/HRS guideline for the management of patients with atrial fibrillation: a report of the American College of Cardiology/American Heart As- sociation Task Force on Practice Guidelines and the Heart Rhythm Society; 2014.
12. Cairns JA, Connolly S, McMurtry S, Stephenson M, Talajic M, CCS Atrial Fibrillation Guidelines Committee. Canadian Cardiovascular Society Atrial Fibrillation Guidelines 2010: prevention of stroke and systemic thromboembolism in atrial fibrillation and flutter. Can J Cardiol. 2011;27(1):74-90 Jan.
13. Mentz RJ, et al. Atrial fibrillation or flutter on initial electrocardiogram is associated with worse outcomes in patients admitted for worsening heart failure with reduced ejection fraction: findings from the EVEREST Trial. Am Heart J. 2012;164(6): 884-892.e2 Dec.
14. Wang TJ, et al. Temporal relations of atrial fibrillation and congestive heart failure and their joint influence on mortality the Framingham Heart Study; 2003.
15. Chamberlain AM, Redfield MM, Alonso A, Weston SA, Roger VL. Atrial fibrillation and mortality in heart failure: a community study. Circ Heart Fail. 2011;4(6):740-6 Nov.
16. Mountantonakis SE, Grau-Sepulveda MV, Bhatt DL, Hernandez AF, Peterson ED, Fonarow GC. Presence of atrial fibrillation is independently associated with adverse outcomes in patients hospitalized with heart failure. Circ Heart Fail. 2012;5(2): 191-201 Mar.
17. Anter E, Jessup M, Callans DJ. Atrial fibrillation and heart failure. Circulation. 2009; 119(18):2516-25 May.
18. Lombardi F. Risk stratification in atrial fibrillation patients - a review focused on mortality. Arrhythmia Electrophysiol Rev. 2012;1(1):8-11 Sep.
19. Schnabel RB, et al. 50 year trends in atrial fibrillation prevalence, incidence, risk fac- tors, and mortality in the Framingham Heart Study: a cohort study. Lancet. 2015;386 (9989):154-62 Jul.
20. Zulkifly H, Lip GYH, Lane DA. Epidemiology of atrial fibrillation. Int J Clin Pract. 2018; 72(3):e13070 Mar.
21. Wang TJ, et al. A risk score for predicting stroke or death in individuals with new- onset atrial fibrillation in the community. JAMA. 2003;290(8):1049 Aug.
22. Fox KAA, et al. Improved risk stratification of patients with atrial fibrillation: an in- tegrated GARFIELD-AF tool for the prediction of mortality, stroke and bleed in pa- tients with and without anticoagulation. BMJ Open. 2017;7(12):e017157 Dec.
23. Kwon Y, et al. Association of smoking, alcohol, and obesity with cardiovascular death and ischemic stroke in atrial fibrillation: the Atherosclerosis Risk in Communities (ARIC) Study and Cardiovascular Health Study (CHS); 2016.
24. Zhu W, Guo L, Hong K. Relationship between smoking and adverse outcomes in pa- tients with atrial fibrillation: a meta-analysis and systematic review. Int J Cardiol. 2016;222:289-94 Nov.
25. Pawar PP, et al. Association between smoking and outcomes in older adults with atrial fibrillation. Arch Gerontol Geriatr. 2012;55(1):85-90.
26. Atzema CL, Austin PC, Chong AS, Dorian P. Factors associated with 90-day death after emergency department discharge for atrial fibrillation. Ann Emerg Med. 2013;61(5): 539-548.e1.
27. Lip GYH, Banerjee A, Lagrenade I, Lane DA, Taillandier S, Fauchier L. Assessing the risk of bleeding in patients with atrial fibrillation. Circ Arrhythm Electrophysiol. 2012;5(5):941-8 Oct.
28. Jabre P, et al. Mortality associated with atrial fibrillation in patients with myocardial infarction. Circulation. 2011;123(15):1587-93 Apr.
29. Walkey AJ, Wiener RS, Ghobrial JM, Curtis LH, Benjamin EJ. Incident stroke and mor- tality associated with new-onset atrial fibrillation in patients hospitalized with se- vere sepsis. JAMA. 2011;306(20):2248-54 Nov.
30. Hijazi Z, et al. Cardiac biomarkers are associated with an increased risk of stroke and death in patients with atrial fibrillation: a randomized evaluation of long-term anticoagulation therapy (RE-LY) substudy. Circulation. 2012;125(13):1605-16 Apr.
31. Hijazi Z, et al. high-sensitivity troponin I for risk assessment in patients with atrial

    fibrillation. Circulation. 2014;129(6):625-34 Feb.

    Hijazi Z, et al. high-sensitivity troponin T and risk stratification in patients with atrial

    fibrillation during treatment with apixaban or warfarin. J Am Coll Cardiol. 2014;63

    (1):52-61 Jan.

    Kundu A, O’Day K, Lessard D, Gore J, Goldberg R, McManus D. The CHA2DS2VASC score can predict mortality in patients with atrial fibrillation following hospitaliza- tion for an acute myocardial infarction: insights from the Worcester heart attack study. J Am Coll Cardiol. 2018;71(11):A470 Mar.

32. Apiyasawat S, Tangcharoen T, Wisaratapong T, Yamwong S, Wiboonpolprasert S, Sritara P. CHA2DS2-VASc scores predict mortality after hospitalization for atrial fi- brillation. Int J Cardiol. 2015;185:293-6 Apr.
33. Larsen TB, Lip GYH, Skjoth F, Due KM, Overvad K, Hvilsted Rasmussen L. Added pre- dictive ability of the CHA ₂ DS ₂ VASc risk score for stroke and death in patients with atrial fibrillation. Circ Cardiovasc Qual Outcomes. 2012;5(3):335-42 May.
34. Umana E, Solares CA, Alpert MA. Tachycardia-induced cardiomyopathy. Am J Med. 2003;114(1):51-5 Jan.
35. Than M, et al. What is an acceptable risk of major adverse cardiac event in chest pain patients soon after discharge from the Emergency Department?: a clinical survey. Int J Cardiol. 2013;166(3):752-4 Jul.
36. Kline JA, et al. Pretest probability assessment derived from attribute matching. BMC Med Inform Decis Mak. 2005;5(1):26 Dec.