European Heart Journal (2009) 30, 2029–2037 CLINICAL RESEARCH

doi:10.1093/eurheartj/ehp222 Arrhythmia/electrophysiology

A history of atrial fibrillation and outcomes

in chronic advanced systolic heart failure: a
propensity-matched study
Mustafa I. Ahmed 1, Michel White 2, O. James Ekundayo 1, Thomas E. Love3,

Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023

Inmaculada Aban 1, Bo Liu 1, Wilbert S. Aronow4, and Ali Ahmed 1,5*
1
University of Alabama at Birmingham, 1530 3rd Avenue South, CH-19, Ste-219, Birmingham, AL 35294-2041, USA; 2Universite de Montreal, Montreal, Canada; 3Case Western
Reserve University, Cleveland, OH, USA; 4New York Medical College, Valhalla, NY, USA; and 5Veterans Affairs Medical Center, Birmingham, AL, USA

Received 15 October 2008; revised 12 April 2009; accepted 29 April 2009; online publish-ahead-of-print 16 June 2009

Aims Atrial fibrillation (AF)-associated poor outcomes in heart failure (HF) are often attributed to older age, advanced
disease, and comorbidity burden of HF patients with AF. Therefore, we examined the effect of AF on outcomes
in a propensity-matched study in which patients with and without AF were well balanced on all measured baseline
characteristics.
.....................................................................................................................................................................................
Methods Of the 2708 advanced chronic systolic HF patients in the Beta-Blocker Evaluation of Survival Trial, 653 had a history
and results of AF. Propensity scores for AF were calculated for each patient and were used to assemble a cohort of 487 pairs of
patients with and without AF who were balanced on 74 baseline characteristics. Matched Cox regression analyses
were used to estimate associations of AF with outcomes during 23 months of mean follow-up. All-cause mortality
occurred in 187 (rate, 2046/10 000 person-years of follow-up) and 181 (rate, 1885/10 000 person-years) matched
patients with and without AF, respectively [matched hazard ratio (HR) when AF was compared with no-AF 1.03,
95% confidence interval (CI) 0.79–1.33; P ¼ 0.84]. Heart failure hospitalization occurred in 215 (rate, 3171/
10 000 person-years) and 184 (rate, 2405/10 000 person-years) matched patients with and without AF, respectively
(matched HR when AF was compared with no-AF 1.28, 95% CI 1.00–1.63; P ¼ 0.049). Hazard ratios and 95% CIs for
AF-associated HF hospitalization for bucindolol and placebo groups were, respectively, 1.08 (0.81–1.43) and 1.54
(1.17– 2.03; P for interaction ¼ 0.09).
.....................................................................................................................................................................................
Conclusion A history of AF had no intrinsic association with mortality but was associated with HF hospitalization in chronic sys-
tolic HF.
-----------------------------------------------------------------------------------------------------------------------------------------------------------
Keywords Heart failure † Atrial fibrillation † Mortality † Hospitalization

concern for residual bias and procedural transparency.7,8

Introduction
Atrial fibrillation (AF) is common in heart failure (HF) and has been
associated with poor outcomes.1 – 3 However, to what extent a
history of AF is independently associated with poor outcomes
has not been studied in a propensity-matched population of
chronic advanced systolic HF patients.3 – 6 Traditional regression-
based multivariable risk adjustment models are limited by strong
model assumptions that may not always be appropriate and
However, propensity-score matching can be used to assemble
two groups of patients balanced on all measured baseline covari-
ates.9 – 12 Furthermore, because outcomes data are not needed
during study design, it allows investigators to design observational
studies while remaining blinded to study outcomes, a key feature of
randomized clinical trials. Therefore, in the current study, we
examined the association of a history of AF and long-term out-
comes in a propensity-matched cohort of HF patients.

* Corresponding author. Tel: þ1 205 934 9632, Fax: þ1 205 975 7099, Email: aahmed@uab.edu
Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2009. For permissions please email: journals.permissions@oxfordjournals.org.
2030
Methods
Source of study data
The Beta-Blocker Evaluation of Survival Trial (BEST) was a multicentre
randomized clinical trial of bucindolol, a beta-blocker, in advanced sys-
tolic HF, methods and results of which have been previously pub-
lished.13 Briefly, 2708 patients with advanced chronic systolic HF
were enrolled from 90 different sites across the USA and Canada
between May 1995 and December 1998. These patients had a mean
left ventricular ejection fraction of 23% and all patients had
New York Heart Association class III– IV symptoms. Over 90% of all
patients were receiving angiotensin-converting enzyme inhibitors,
diuretics, and digitalis. We used a public-use copy of the BEST
dataset obtained from the National Heart Lung and Blood Institute
that included 2707 patients (one patient did not consent to be
included in the public-use copy). The current analysis focuses on a
subset of 974 propensity-matched patients.
History of atrial fibrillation
Overall, 653 (24%) patients had a history of AF. Data on AF were col-
lected by study investigators and were not centrally adjudicated. Of the
653 patients with AF, 293 (45%) had electrocardiographic (ECG) evi-
dence of AF at baseline. Considering the similar unadjusted all-cause
mortality between patients with a history of AF who had and did
not have ECG evidence of AF at baseline, we chose to combine
these patients into one group. Of the 2054 patients without AF,
only 10 patients had ECG evidence of new AF at baseline.
Study outcomes
Primary outcomes for this propensity-matched study were all-cause
mortality and HF hospitalization. Secondary outcomes were cardiovas-
cular and HF mortality and all-cause hospitalization. Investigators
blinded to the outcomes ascertained study outcomes. Patients were
followed for a mean of 23 months (median follow-up, 22 months;
range: 1 – 50 months).
Assembly of a balanced cohort
Because of significant imbalances in baseline characteristics between
patients with and without AF before matching (Table 1), we used
propensity-score matching to assemble a cohort of patients whereby
those with and without AF would be well balanced on all measured
baseline characteristics.8,14 We began by estimating propensity
scores for AF for each of the 2707 participants using a non-
parsimonious multivariable logistic regression model. In the model,
AF was used as the dependent variable, and all clinically relevant base-
line characteristics (n ¼ 74) displayed in Figure 1 were included as
covariates.
Because propensity-score models are sample-specific adjusters and
are not intended to be used for out-of-sample prediction or estimation
of coefficients, measures of fitness and discrimination are not impor-
tant for the assessment of the model’s effectiveness.9 – 12 The efficacy
of propensity-score models is best assessed by estimating post-match
absolute standardized differences between baseline covariates that
directly quantifies the bias in the means (or proportions) of covariates
across the groups, expressed as a percentage of the pooled standard
deviations. We therefore calculated pre- and post-match absolute
standardized differences and presented those findings as Love plots.9
An absolute standardized difference of 0% indicates no residual bias
and , 10% is considered of inconsequential bias.
We used a greedy matching protocol to match patients with and
without AF who had similar propensity scores to five, four, three,
M.I. Ahmed et al.
two, and one decimal places in five repeated steps.9 – 12 In the first
step, we multiplied the raw propensity scores by 100 000. For
example, propensity scores of 0.57520576 and 0.57520374 for a pair
of patients with and without AF, respectively, were converted to
57520.58 and 57520.37. Then we rounded both the numbers to the
nearest value divisible by 0.25 (e.g. 57520.50) and matched. All patients
matched by propensity scores to the five decimal places were then
removed from the file. In the second step, we multiplied the raw pro-
pensity scores by 10 000, rather than 100 000, and repeated the above
process. This was then repeated three more times, each time, multiply-
ing by 1000, 100, and finally 10. In all, we were able to match 487 of the
653 history of AF patients with 487 patients who had no history of AF,
but had similar propensity for AF.
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
Statistical analysis
For descriptive analyses, we used Pearson x2 and Wilcoxon rank-sum
tests for the pre-match, and McNemar’s test and paired sample t-test
for the post-match comparisons of baseline covariates between
patients with and without AF, as appropriate. Kaplan– Meier and
matched Cox regression analyses were used to determine the associ-
ations of AF with various outcomes during 23 months of mean
follow-up. Log-minus-log scale survival plots were used to check pro-
portional hazards assumptions. To assess the effect of loss of partici-
pants during matching, we repeated our analysis in all 2707
pre-match patients using three different Cox regression models: (i)
unadjusted, (ii) multivariable-adjusted, using all covariates used in the
propensity-score model, and (iii) propensity score adjusted. To elimin-
ate potentially inflated pre-match associations due to larger sample
size, we assembled a pre-match cohort of 487 pairs of patients by
pairing our matched AF patients with a randomly selected subset of
487 non-AF patients from the pool of 2054 pre-match non-AF
patients.
Sensitivity analysis
Even though our matched cohort achieved excellent balance in all
measured covariates between the two groups, we do not know if
there was bias due to imbalances on a hidden covariate (a baseline
characteristic that was not measured and thus not used to estimate
propensity scores). We therefore conducted a formal sensitivity analy-
sis to quantify the degree of hidden bias that would need to be present
to invalidate any conclusions based on significant association between
AF and primary outcomes among matched patients.15 Specifically, we
used Rosenbaum’s model and related mathematical equations to
place bounds on the significance levels that would have been appropri-
ate for our primary comparisons between patients with and without
AF had a hidden bias of a particular size occurred.15
Subgroup analysis
Select subgroup analyses of matched patients based on demographics
(age  65 years, sex, and race), clinically relevant comorbidities and
marker of disease severity (coronary artery disease, hypertension, dia-
betes mellitus, chronic kidney disease, and left ventricular ejection frac-
tion , 25%), and randomization factor (bucindolol use) were
conducted to determine heterogeneity of the associations of AF
with HF hospitalization. We first calculated the absolute risk differ-
ences, and then estimated the effects of AF on HF hospitalization in
each group using Cox regression models. Because of a lack of a signifi-
cant overall association between AF and mortality, we did not perform
a formal subgroup analysis for this outcome. However, considering
that patients were randomized to bucindolol, we examined the associ-
ation of AF and all-cause mortality in subgroups of patients receiving
 Atrial fibrillation and heart failure outcomes
Table 1 Baseline patient characteristics by history of atrial fibrillation before and after propensity matching

Before propensity matching After propensity matching

Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023

........................................................................................ .....................................................................................
No history of atrial History of atrial P-value No history of atrial History of atrial P-value
fibrillation (n 5 2054) fibrillation (n 5 653) fibrillation (n 5 487) fibrillation (n 5 487)
.............................................................................................................................................................................................................................................
Age (years) 58.8 (+12.4) 64.8 (+10.9) ,0.0001 62.8 (+11.7) 63.2 (+11.2) 0.55
Female 511 (25) 82 (13) ,0.0001 68 (14) 70 (14) 0.92
African American 515 (25) 112 (17) ,0.0001 92 (19) 90 (19) 0.94
Current smoker 378 (18) 96 (15) 0.03 70 (14) 78 (16) 0.52
.............................................................................................................................................................................................................................................
Past medical history
Duration of heart failure (months) 46 (+46) 60 (+54) ,0.0001 56 (+54) 55 (+48) 0.68
Coronary artery disease 1176 (57) 417 (64) 0.003 321 (66) 313 (64) 1.00
Coronary artery bypass surgery 552 (27) 230 (35) ,0.0001 161 (33) 157 (32) 0.84
Percutaneous coronary intervention 325 (16) 98 (15) 0.62 78 (16) 75 (15) 0.86
Angina pectoris 1078 (53) 322 (49) 0.16 253 (52) 249 (51) 0.84
Hypertension 1205 (59) 390 (60) 0.63 271 (56) 286 (59) 0.37
Diabetes mellitus 747 (36) 217 (33) 0.15 178 (37) 176 (36) 0.95
Hyperlipidaemia 918 (44) 252 (39) 0.006 194 (40) 195 (40) 1.000
.............................................................................................................................................................................................................................................
Clinical findings
Body mass index (kg/m2) 36.6 (+8.5) 36.4 (+8.2) 0.59 36.5 (+8.5) 36.6 (+8.4) 0.88
Systolic blood pressure (mmHg) 117 (+18) 117 (+18) 0.85 116 (+18) 117 (+18) 0.34
Diastolic blood pressure (mmHg) 71 (+11) 70 (+11) ,0.0001 70 (+11) 70 (+11) 0.34
Heart rate (b.p.m.) 82 (+13) 79 (+13) , 0.0001 80 (+13) 80 (+13) 0.92
Jugular venous distension 1.63 (+0.85) 1.91 (+1.01) ,0.0001 1.84 (+0.95) 1.87 (+1.0) 0.59
S3 gallop 904 (44) 274 (42) 0.36 213 (44) 215 (44) 0.95
Pulmonary râles 252 (12) 107 (16) 0.007 75 (15) 77 (16) 0.93
Lower extremity oedema 516 (25) 214 (33) , 0.0001 153 (31) 155 (32) 0.94
NYHA class III 1900 (93) 581 (89) 0.005 438 (90) 433 (89) 0.68
.............................................................................................................................................................................................................................................
Medications
ACE-inhibitors 1890 (92) 594 (91) 0.40 447 (92) 445 (91) 0.91
Digitalis 1879 (92) 615 (94) 0.026 459 (94) 459 (94) 1.00
Diuretics 1898 (92) 626 (96) 0.002 464 (95) 463 (95) 1.00
Vasodilators 891 (43) 293 (45) 0.50 218 (45) 218 (45) 1.00
Anti-arrhythmic drugs 38 (2) 36 (6) ,0.0001 23 (5) 22 (5) 1.00
Anti-coagulants 1063 (52) 507 (78) ,0.0001 344 (71) 347 (71) 0.87

Continued

2031
 2032
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
Table 1 Continued

Before propensity matching After propensity matching

........................................................................................ .....................................................................................
No history of atrial History of atrial P-value No history of atrial History of atrial P-value
fibrillation (n 5 2054) fibrillation (n 5 653) fibrillation (n 5 487) fibrillation (n 5 487)
.............................................................................................................................................................................................................................................
Chest X-ray findings
Pulmonary oedema 222 (11) 86 (13) 0.10 72 (15) 70 (14) 0.93
Cardiomegaly 1644 (80) 560 (86) 0.001 408 (84) 403 (83) 0.73
.............................................................................................................................................................................................................................................
Echocardiography findings
LV ejection fraction (%) 22.9 (+7.2) 23.5 (+7.3) 0.05 22.6 (+7.4) 23.0 (+7.2) 0.38
RV ejection fraction (%) 34.9 (+12.1) 33.9 (+10.5) 0.047 33.2 (+11.6) 33.4 (+10.6) 0.83
.............................................................................................................................................................................................................................................
Laboratory findings
Haemoglobin (g/dL) 14.0 (+1.6) 14.0 (+1.8) 0.98 13.9 (+1.6) 14.0 (+1.7) 0.52
White blood cell (103/mL) 7.5 (+2.2) 7.4 (+2.1) 0.17 7.4 (+2.6) 7.5 (+2.1) 0.75
Serum creatinine (mg/dL) 1.22 (+0.40) 1.33 (+0.42) ,0.0001 1.31 (+0.43) 1.29 (+0.41) 0.38
Blood urea nitrogen (mg/dL) 23 (+14) 29 (+18) ,0.0001 27 (+16) 27 (+16) 0.54
Serum sodium (mEq/L) 139 (+3) 139 (+3) 0.53 139 (+4) 139 (+3) 0.58
Serum potassium (mEq/L) 4.32 (+0.47) 4.28 (+0.51) 0.06 4.32 (+0.51) 4.29 (+0.50) 0.29
Serum magnesium (mg/dL) 1.7 (+0.24) 1.8 (+0.27) 0.003 1.8 (+0.24) 1.8 (+0.26) 0.63
Serum glucose (mg/dL) 136 (+77) 129 (+67) 0.035 134 (+74) 132 (+69) 0.66
Plasma thromboplastin time (s) 30.6 (+9.4) 34.0 (+8.98) , 0.0001 33.6 (+15.1) 32.8 (+7.83) 0.28
Plasma norepinephrine (pg/mL) 500 (+296) 562 (+331) ,0.0001 550 (+366) 551 (+345) 0.98

Values for categorical variables are given with % values in parentheses. Values for continuous variables are given with +standard deviation in parentheses. ACE, angiotensin-converting enzyme; LV, left ventricular; RV, right ventricular.

M.I. Ahmed et al.

Atrial fibrillation and heart failure outcomes
Figure 1 Love plots for absolute standardized differences for baseline covariates between patients with and without a history of atrial fibrilla-
tion, before and after propensity score matching (ACE, angiotensin-converting enzyme; ARB, angiotensin-receptor blocker).
bucindolol and placebo. Finally, we formally tested for first-order inter-
actions using Cox proportional hazards models, entering interaction
terms for each subgroup (e.g. gender*AF for the gender subgroups).
All statistical tests were two-sided, and tests with P , 0.05 were con-
sidered significant. All statistical analyses were done using SPSS for
windows version 15.16
Results
Baseline characteristics
Matched patients had a mean age of 63 (+11) years, 14% were
women and 19% were African American. Before matching, patients
with AF were older, had a longer duration of HF and higher
comorbidity burden than those without AF. These and other sig-
nificant imbalances in baseline characteristics before matching
and the balances achieved on all baseline characteristics after
matching are displayed in Table 1 and Figure 1. After matching,
absolute standardized differences for all measured covariates
were ,10% (most were ,5%), suggesting substantial covariate
balance across the groups (Figure 1).
Atrial fibrillation and mortality
Overall, 368 (38%) matched patients died, including 314 (32%) due
to cardiovascular causes and 110 (11%) due to HF during 1874
patient-years of follow-up. In the 487 matched-pair cohort, all-
cause mortality occurred in 187 (rate, 2046/10 000 person-years)
and 181 (rate, 1885/10 000 person-years) patients with and
2033
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
without AF, respectively [matched hazard ratio (HR) when AF
was compared with no AF 1.03, 95% confidence interval (CI)
0.79 –1.33; P ¼ 0.84; Figure 2A and Table 2]. No formal sensitivity
analysis was conducted due to lack of any significant association
between AF and all-cause mortality. Atrial fibrillation-associated
mortality due to all causes was of borderline significance among
patients in the placebo group (HR for AF 1.30, 95% CI 0.98–
1.73; P ¼ 0.073) but not among those in the bucindolol group
(HR for AF 0.90, 95% CI 0.67–1.21; P ¼ 0.48; P for interaction ¼
0.08; data not shown).
In the full pre-match cohort of 2707 patients, all-cause mor-
tality occurred in 39% (rate, 2088/10 000 person-years of
follow-up) and 29% (rate, 1430/10 000 person-years of follow-up)
of patients with and without AF, respectively (unadjusted HR for
AF 1.47; 95% CI 1.27–1.70; P , 0.0001). To determine if the sig-
nificant pre-match association between AF and total mortality
may have been due to its larger sample size, we assembled 487
pairs of pre-match patients by pairing 487 matched AF patients
with 487 random non-AF patients from the 2054 pre-match
patients without AF. Among the 487 random-pair pre-match
patients, all-cause mortality occurred in 38 and 29% of patients,
respectively, with and without AF (unadjusted HR 1.44; 95% CI
1.16 –1.80; P ¼ 0.001; data not shown). Multivariable-adjusted
and propensity-adjusted HR’s for AF were, respectively, 1.18
(95% CI 1.01–1.38; P ¼ 0.040) and 1.12 (95% CI 0.94– 1.34;
P ¼ 0.21).
In matched patients CV mortality occurred in 156 (rate, 1706/
10 000 person-years) and 158 (rate, 1645/10 000 person-years)
2034
Figure 2 Kaplan –Meier plots for (A) all-cause mortality, (B) all-
cause hospitalization, and (C ) heart failure hospitalization by a
history of atrial fibrillation (HxAF) in propensity score matched
pairs (CI, confidence interval; fib, fibrillation; HR, hazard ratio).
M.I. Ahmed et al.
patients with and without AF, respectively (matched HR when AF
was compared with no AF 0.97, 95% CI 0.74–1.28; P ¼ 0.83;
Table 3). Heart failure mortality occurred in 60 (rate, 656/10 000
person-years) and 50 (rate, 521/10 000 person-years) patients
with and without AF, respectively (matched HR when AF was
compared with no AF 1.10, 95% CI 0.67 –1.82; P ¼ 0.70;
Table 3). Pre-match associations of AF with mortality due to
cardiovascular causes and HF are displayed in Table 3.
Atrial fibrillation and hospitalization
Overall, 630 (65%) patients were hospitalized including 399 (41%)
due to HF. Heart failure hospitalizations occurred in 215 (rate,
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
3171/10 000 person-years) and 184 (rate 2405/10 000 person-
years) patients with and without AF, respectively (matched HR
when AF was compared with no AF 1.28, 95% CI 1.00– 1.63;
P ¼ 0.049; Figure 2B and Table 2). In the absence of hidden bias,
a sign-score test for matched data with censoring provides evi-
dence (P ¼ 0.048) that HF patients with a history of AF clearly
had more hospitalizations due to HF than those without a
history of AF. Findings from our sensitivity analysis indicate that a
hidden covariate, which is a near-perfect predictor of HF hospital-
ization, may potentially explain away the association between AF
and HF hospitalization, if that would increase the odds of AF by
only 0.205%. The associations of AF with HF hospitalization
among various subgroups of patients are displayed in Figure 3.
Atrial fibrillation-associated hospitalization due to HF was signifi-
cant only among patients in the placebo group (HR for AF 1.54,
95% CI 1.17–2.03; P ¼ 0.002) but not among those in the bucin-
dolol group (HR for AF 1.08, 95% CI 0.81–1.43; P ¼ 0.61; P for
interaction ¼ 0.09; Figure 3). Association of AF with all-cause hos-
pitalization is displayed in Table 3.
Overall 918 (94%) matched patients were receiving digoxin and
among these patients, AF was associated with increased HF hospi-
talization (HR 1.29, 95% CI 1.06–1.58; P ¼ 0.012; data not shown).
However, this association varied by the use of bucindolol. Among
patient receiving digoxin without bucindolol (n ¼ 451), AF was
associated with increased HF hospitalization (HR 1.56, 95% CI
1.17–2.06; P ¼ 0.002). On the other hand, among patients receiv-
ing both digoxin and bucindolol (n ¼ 467), AF had no association
with HF hospitalization (HR 1.08, 95% CI 0.81– 1.45; P ¼ 0.58; P
for interaction ¼ 0.09; data not shown).
In the pre-match cohort of 2707 patients, HF hospitalization
occurred in 45% (rate, 3209/10 000 person-years of follow-up)
and 37% (rate, 2243/10 000 person-years of follow-up) of patients
with and without AF, respectively (unadjusted HR for AF 1.38; 95%
CI 1.21–1.58; P , 0.0001; Table 2). Multivariable-adjusted and
propensity-adjusted HR’s were, respectively, 1.26 (95% CI 1.09–
1.45; P ¼ 0.002) and 1.22 (95% CI 1.04– 1.44; P ¼ 0.016).
All-cause hospitalizations occurred among 328 (rate, 6271/
10 000 person-years) and 302 (rate 5216/10 000 person-years)
matched patients with and without AF, respectively (matched HR
when AF was compared with no AF 1.22, 95% CI 1.00– 1.50;
P ¼ 0.06; Figure 2B and Table 2). Pre-match associations of AF
with all-cause hospitalizations are displayed in Table 3.
Atrial fibrillation and heart failure outcomes 2035

Table 2 Association of atrial fibrillation with all-cause mortality and heart failure hospitalization in Beta-Blocker
Evaluation of Survival Trial

Rate per 10 000 person-years (events/ Absolute rate Matched hazard ratio P-value
total follow-up years) difference (per 10 000 (95% confidence
.................................................. person-years)a interval)
No history of History of atrial
atrial fibrillation fibrillation
...............................................................................................................................................................................
Before matching n ¼ 2054 n ¼ 653
All-cause mortality 1430 (603/4216) 2088 (256/1226) þ658 1.47 (1.27–1.70) , 0.0001
HF hospitalization 2243 (750/3344) 3209 (294/916) þ967 1.38 (1.21–1.58) , 0.0001
...............................................................................................................................................................................
After matching n ¼ 487 n ¼ 487

Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023

All-cause mortality 1885 (181/960) 2046 (187/914) þ161 1.03 (0.79–1.33) 0.84
HF hospitalization 2405 (184/765) 3171 (215/678) þ766 1.28 (1.00–1.63) 0.049

a
Absolute differences in rates of events per 10 000 person-years of follow-up were calculated by subtracting the event rates in the atrial fibrillation group from the event rates in
the no atrial fibrillation group (before values were rounded).

Table 3 Association of atrial fibrillation with other outcomes in Beta-Blocker Evaluation of Survival Trial

Rate per 10 000 person-years (events/total Absolute rate difference Matched hazard ratio P-value
follow-up years) (per 10 000 (95% confidence
........................................................ person-years)a interval)
No history of atrial History of atrial
fibrillation fibrillation
...............................................................................................................................................................................
Before matching n ¼ 2054 n ¼ 653
Cardiovascular mortality 1219 (514/4216) 1762 (216/1226) þ543 1.45 (1.24–1.70) , 0.0001
Heart failure mortality 403 (170/4216) 750 (92/1226) þ347 1.88 (1.46–2.43) , 0.0001
All-cause hospitalization 5020 (1255/2500) 6493 (448/690) þ1473 1.25 (1.13–1.40) , 0.0001
...............................................................................................................................................................................
After matching n ¼ 487 n ¼ 487
Cardiovascular mortality 1645 (158/960) 1706 (156/914) þ61 0.97 (0.74–1.28) 0.83
Heart failure mortality 521 (50/960) 656 (60/914) þ136 1.10 (0.67–1.82) 0.70
All-cause hospitalization 5216 (302/579) 6271 (328/523) þ1056 1.22 (1.00–1.50) 0.06

a
Absolute differences in rates of events per 10 000 person-years of follow-up were calculated by subtracting the event rates in the atrial fibrillation group from the event rates in
the no atrial fibrillation group (before values were rounded).

Discussion haemodynamic abnormalities leading to worsening HF and hospi-

Significant bivariate associations of AF with increased risk for both
all-cause, cardiovascular, and HF mortality and all-cause and HF
hospitalizations in patients with advanced chronic systolic HF
suggest that AF remains an important prognostic marker in these
patients. However, data from propensity-matched patients, who
were well balanced in 74 baseline characteristics, demonstrate
that AF had no intrinsic association with total or cause-specific
mortality, but had an independent association with HF hospitaliz-
ation. These findings are important as hospitalization due to
worsening HF is a major cause of hospitalization for patients
with HF, and in the US it is the leading cause for hospitalization
for Medicare beneficiaries 65 years and older. Further, the lack
of an intrinsic association of AF with mortality may in part
explain the lack of superiority for rhythm control strategy relative
to rate control in these patients.17
Atrial fibrillation is characterized by irregular cardiac rhythm,
which along with an uncontrolled heart rate, may cause significant
talization. Atrial fibrillation is also associated with structural and
functional pathologies such as a dilated left atrium with impaired
contraction, impaired left ventricular filling, and activation of vaso-
constrictive neurohormones.4,18,19 However, little is known about
the individual contribution of an abnormal cardiac rhythm or an
uncontrolled heart rate to the haemodynamic instability in HF
patients with AF that may lead to HF or all-cause hospitalizations.
An irregular cardiac rhythm alone, in the presence of controlled
ventricular rate, may cause acute adverse haemodynamic events
in AF.20 However, whether an irregular rhythm alone, in the pres-
ence of normal heart rate, may also lead to increased HF hospital-
ization is unknown. Findings from a recent large randomized clinical
trial of rhythm vs. rate control for AF in systolic HF suggest that
fewer patients in the rate control group required hospitalization,
highlighting the importance of a rate control strategy.17
Although uncontrolled heart rate may also lead to adverse
haemodynamic consequences, heart rate in matched patients
with AF was well controlled and was well balanced. Therefore,
2036
Figure 3 Association of a history of atrial fibrillation with hospitalization due to heart failure in subgroups of propensity score-matched
patients in the BEST (CI, confidence interval; HR, hazard ratio).
the observed association of AF and HF hospitalization cannot be
explained by baseline uncontrolled heart rate or baseline differ-
ences in heart rate. However, it is plausible that heart rate
varied between patients with and without AF during follow-up.
Heart failure patients with AF may have an inappropriate tachycar-
dic response to physical activities or other stresses which could
lead to acute decompensation with resultant increases in HF hos-
pitalization. This notion is supported by our subgroup analyses that
demonstrated that AF-associated increased HF hospitalization was
only observed among patients not receiving bucindolol, a beta-
blocker (Figure 3). In the BEST, patients receiving bucindolol
were less likely to have tachycardia.13 Beta-blockers are currently
recommended as the drug of choice for control of ventricular rate
in HF patients with AF.21 This is particularly important as beta-
blockers, when added to digitalis, may have a synergistic effect in
controlling ventricular rate both at rest and during physical activi-
ties.22 – 24 Over 90% of patients in our study were receiving digoxin,
and among these patients, AF-associated increase in HF hospitaliz-
ation was only seen in patients who were not receiving bucindolol.
Interestingly, although AF had no overall association with mortality,
in the subgroup of patients not receiving bucindolol, AF seemed to
be associated with increased mortality, but had no significant
association among those receiving bucindolol.
Despite an increased unadjusted bivariate association, findings
from our study suggest that AF may not have an intrinsic associ-
ation with mortality and that the bivariate association was likely
due to differences in prognostically important baseline covariates.
This is consistent with the findings from a recent large randomized
clinical trial of rhythm vs. rate control for AF in systolic HF, where
M.I. Ahmed et al.
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
rhythm control did not reduce the primary endpoint of cardiovas-
cular mortality.17 The authors of that study speculated that the
failure of the rhythm control strategy may have been due to the
fact that AF in HF may not have an intrinsic association with mor-
tality,17 a contention which is supported by findings from our
study. It was further speculated that the benefit of rhythm
control may have been cancelled by harmful effects of anti-
arrhythmic therapies. The proportion of patients receiving anti-
arrhythmic drugs was low in our study and was well balanced in
our matched cohort.
Previous studies have demonstrated a lack of an independent
association between AF and mortality.4,25 However, our study is
distinguished from those studies by the use of propensity-score
matching in which patients with and without AF were well
balanced in 74 important demographic, clinical, subclinical, and bio-
chemical covariates. Further, our subgroup analysis provided
insight into the role of beta-blockers in HF patients with AF.
Limitations
There are several limitations in the current study. Despite an
excellent post-match balance in 74 baseline covariates, it is poss-
ible that there were imbalances in unmeasured covariates. In fact,
our sensitivity analysis suggests that our findings may be rather sen-
sitive to an unmeasured confounder. However, sensitivity analysis
cannot determine whether an unmeasured covariate exists or
not. Furthermore, for an unmeasured covariate to be a confoun-
der, in addition to being associated with the exposure (AF), it
must also be a near-perfect predictor of outcomes (HF hospitaliz-
ation) and not be strongly correlated with any of the 74 measured
Atrial fibrillation and heart failure outcomes
covariates used in our propensity model. We were able to match
75% of patients with AF and any effect due to loss of participants
during matching would be minimal. Further, we were able to
reproduce our key findings in the pre-match dataset adjusting for
propensity score.
Interestingly, despite same sample size, we observed discordant
findings after risk adjustment using a traditional multivariable model
and propensity score. Multivariable regression adjustments may
not ensure that the distribution of the confounders is balanced
between groups, which may lead to extrapolations beyond the
data.7 Results of this study based on advanced systolic HF patients
may not be generalizable to patients with mild to moderate HF and
those with diastolic HF, who constitute half of all HF patients. It is
possible that some patients without baseline history of AF may
have developed AF during the follow-up period. However, this
regression dilution is known to underestimate true associations.26
Conclusions
Atrial fibrillation is a marker of increased mortality and hospitaliz-
ation in advanced systolic HF and remains a useful tool to identify
patients at risk of poor outcomes. Atrial fibrillation seems to have
an intrinsic association with hospitalization due to worsening HF,
which was worse in patients not receiving bucindolol. Despite
the overall lack of an intrinsic association of AF with mortality, it
appears to increase mortality in those not receiving bucindolol.
These data provide additional arguments that HF patients with
AF should be treated with beta-blockers in the absence of an
absolute contraindication.
Acknowledgements
‘The Beta-Blocker Evaluation of Survival Trial (BEST) is conducted
and supported by the NHLBI in collaboration with the BEST Study
Investigators. This manuscript was prepared using a limited access
dataset obtained from the NHLBI and does not necessarily reflect
the opinions or views of the BEST or the NHLBI.
Funding
A.A. is supported by the National Institutes of Health through a grant
from the National Heart, Lung, and Blood Institute (R01-HL085561),
and a generous gift from Ms Jean B. Morris of Birmingham, Alabama.
Conflict of interest: none declared.
References
1. Ahmed A, Thornton P, Perry GJ, Allman RM, DeLong JF. Impact of atrial fibrilla-
tion on mortality and readmission in older adults hospitalized with heart failure.
Eur J Heart Fail 2004;6:421 – 426.
2. Ryder KM, Benjamin EJ. Epidemiology and significance of atrial fibrillation. Am J
Cardiol 1999;84:131R –138R.
3. Dries DL, Exner DV, Gersh BJ, Domanski MJ, Waclawiw MA, Stevenson LW.
Atrial fibrillation is associated with an increased risk for mortality and heart
failure progression in patients with asymptomatic and symptomatic left ventricular
systolic dysfunction: a retrospective analysis of the SOLVD trials. Studies of Left
Ventricular Dysfunction. J Am Coll Cardiol 1998;32:695 –703.
4. Crijns HJ, Tjeerdsma G, de Kam PJ, Boomsma F, van Gelder IC, van den Berg MP,
van Veldhuisen DJ. Prognostic value of the presence and development of atrial
fibrillation in patients with advanced chronic heart failure. Eur Heart J 2000;21:
1238– 1245.
5. Middlekauff HR, Stevenson WG, Stevenson LW. Prognostic significance of atrial
fibrillation in advanced heart failure. A study of 390 patients. Circulation 1991;
84:40–48.
2037
6. Olsson LG, Swedberg K, Ducharme A, Granger CB, Michelson EL, McMurray JJ,
Puu M, Yusuf S, Pfeffer MA. Atrial fibrillation and risk of clinical events in chronic
heart failure with and without left ventricular systolic dysfunction: results from the
Candesartan in Heart failure-Assessment of Reduction in Mortality and morbidity
(CHARM) program. J Am Coll Cardiol 2006;47:1997 –2004.
7. Fitzmaurice G. Confounding: regression adjustment. Nutrition 2006;22:581 –583.
8. Rubin DB. Using propensity score to help design observational studies: appli-
cation to the tobacco litigation. Health Services and Outcomes Research Methodology
2001;2:169 –188.
9. Ahmed A, Husain A, Love TE, Gambassi G, Dell’Italia LJ, Francis GS,
Gheorghiade M, Allman RM, Meleth S, Bourge RC. Heart failure, chronic diuretic
use, and increase in mortality and hospitalization: an observational study using
propensity score methods. Eur Heart J 2006;27:1431 –1439.
10. Ahmed A, Rich MW, Love TE, Lloyd-Jones DM, Aban IB, Colucci WS, Adams KF,
Gheorghiade M. Digoxin and reduction in mortality and hospitalization in heart
failure: a comprehensive post hoc analysis of the DIG trial. Eur Heart J 2006;27:
Downloaded from https://academic.oup.com/eurheartj/article/30/16/2029/631270 by guest on 27 August 2023
178 –186.
11. Ahmed A, Rich MW, Sanders PW, Perry GJ, Bakris GL, Zile MR, Love TE, Aban IB,
Shlipak MG. Chronic kidney disease associated mortality in diastolic versus systo-
lic heart failure: a propensity matched study. Am J Cardiol 2007;99:393 – 398.
12. Ahmed A, Zannad F, Love TE, Tallaj J, Gheorghiade M, Ekundayo OJ, Pitt B. A
propensity-matched study of the association of low serum potassium levels and
mortality in chronic heart failure. Eur Heart J 2007;28:1334 –1343.
13. Investigators B. A trial of the beta-blocker bucindolol in patients with advanced
chronic heart failure. N Engl J Med 2001;344:1659 –1667.
14. Rosenbaum PR, Rubin DB. The central role of propensity score in observational
studies for causal effects. Biometrika 1983;70:41 –55.
15. Rosenbaum PR. Sensitivity to hidden bias. In: Rosenbaum PR (ed.), Observational
Studies. 2nd ed. New York: Springer-Verlag; 2002. p110 –124.
16. SPSS Inc. SPSS for Windows, Rel. 15 [computer program]. VersionChicago, IL: SPSS
Inc.; 2008.
17. Roy D, Talajic M, Nattel S, Wyse DG, Dorian P, Lee KL, Bourassa MG, Arnold JM,
Buxton AE, Camm AJ, Connolly SJ, Dubuc M, Ducharme A, Guerra PG,
Hohnloser SH, Lambert J, Le Heuzey JY, O’Hara G, Pedersen OD, Rouleau JL,
Singh BN, Stevenson LW, Stevenson WG, Thibault B, Waldo AL. Rhythm
control versus rate control for atrial fibrillation and heart failure. N Engl J Med
2008;358:2667 – 2677.
18. Van den Berg MP, Tuinenburg AE, Crijns HJ, Van Gelder IC, Gosselink AT, Lie KI.
Heart failure and atrial fibrillation: current concepts and controversies. Heart
1997;77:309 –313.
19. Tuinenburg AE, Van Veldhuisen DJ, Boomsma F, Van Den Berg MP, De Kam PJ,
Crijns HJ. Comparison of plasma neurohormones in congestive heart failure
patients with atrial fibrillation versus patients with sinus rhythm. Am J Cardiol
1998;81:1207 – 1210.
20. Clark DM, Plumb VJ, Epstein AE, Kay GN. Hemodynamic effects of an irregular
sequence of ventricular cycle lengths during atrial fibrillation. J Am Coll Cardiol
1997;30:1039 – 1045.
21. Hunt SA, Abraham WT, Chin MH, Feldman AM, Francis GS, Ganiats TG,
Jessup M, Konstam MA, Mancini DM, Michl K, Oates JA, Rahko PS, Silver MA,
Stevenson LW, Yancy CW, Antman EM, Smith SC Jr, Adams CD, Anderson JL,
Faxon DP, Fuster V, Halperin JL, Hiratzka LF, Jacobs AK, Nishimura R,
Ornato JP, Page RL, Riegel B. ACC/AHA 2005 Guideline Update for the Diagnosis
and Management of Chronic Heart Failure in the Adult: a report of the American
College of Cardiology/American Heart Association Task Force on Practice
Guidelines (Writing Committee to Update the 2001 Guidelines for the Evaluation
and Management of Heart Failure): developed in collaboration with the American
College of Chest Physicians and the International Society for Heart and Lung
Transplantation: endorsed by the Heart Rhythm Society. Circulation 2005;112:
e154 – e235.
22. Matsuda M, Matsuda Y, Yamagishi T, Takahashi T, Haraguchi M, Tada T,
Kusukawa R. Effects of digoxin, propranolol, and verapamil on exercise in patients
with chronic isolated atrial fibrillation. Cardiovasc Res 1991;25:453 –457.
23. David D, Segni ED, Klein HO, Kaplinsky E. Inefficacy of digitalis in the control of
heart rate in patients with chronic atrial fibrillation: beneficial effect of an added
beta adrenergic blocking agent. Am J Cardiol 1979;44:1378 –1382.
24. Farshi R, Kistner D, Sarma JS, Longmate JA, Singh BN. Ventricular rate control in
chronic atrial fibrillation during daily activity and programmed exercise: a cross-
over open-label study of five drug regimens. J Am Coll Cardiol 1999;33:304–310.
25. Stevenson WG, Stevenson LW, Middlekauff HR, Fonarow GC, Hamilton MA,
Woo MA, Saxon LA, Natterson PD, Steimle A, Walden JA, Tillisch JH. Improving
survival for patients with atrial fibrillation and advanced heart failure. J Am Coll
Cardiol 1996;28:1458 –1463.
26. Clarke R, Shipley M, Lewington S, Youngman L, Collins R, Marmot M, Peto R.
Underestimation of risk associations due to regression dilution in long-term
follow-up of prospective studies. Am J Epidemiol 1999;150:341–353.