European Heart Journal (2003) 24, 172–181
Predictors of 90-day outcome in patients
stabilized after acute coronary syndromes
L.K. Newby a*, M.V. Bhapkar a, H.D. White b, E.J. Topol c,
F.C. Dougherty d, R.A. Harrington a, M.C. Smith d, L.F. Asarch d,
R.M. Califf a, for the SYMPHONY and 2nd SYMPHONY Investigators
a
Duke Clinical Research Institute, Durham, NC, USA
Green Lane Hospital, Auckland, New Zealand
c
Cleveland Clinic Foundation, Cleveland, OH, USA
d
F. Hoffmann-La Roche, Ltd, Basel, Switzerland
b
Received 16 April 2002; accepted 17 April 2002
KEYWORDS
Acute coronary
syndromes;
risk models;
outcomes;
risk stratification
Aims We investigated predictors of 90-day risk among patients surviving the early
period after an acute coronary syndrome (ACS).
Methods and Results The study population included 15 904 stabilized ST-segment
elevation or non-ST-segment elevation ACS patients randomized in SYMPHONY and
2nd SYMPHONY. We developed risk models for death, death or myocardial infarction
(MI), and death, MI, or severe recurrent ischaemia (SRI) using Cox proportionalhazards techniques. Demographic, history, and pre-randomization clinical and medication variables were tested. Validation techniques included development of
individual trial models, backward elimination and bootstrapping. Of 118 variables, 17
independently predicted mortality. The strongest associations included greater age
(2 = 31.1), higher randomization heart rate (2 = 27.4), and heart failure (HF) variables (HF between qualifying event and randomization, 2 = 21.8; history of HF,
2 = 12.2). Higher creatinine clearance (2 = 17.7) and percutaneous coronary intervention between qualifying event and randomization (2 = 11.1) most strongly predicted lower risk. Similar characteristics entered the double and triple composite
models, but HF variables and age less strongly predicted these end-points.
Conclusions In patients stabilized after ACS, those at highest risk over the next 90
days can be identified. Typical clinical markers are better at identifying risk of death
than non-fatal MI or SRI. Novel risk markers are needed for these outcomes.
© 2003 The European Society of Cardiology. Published by Elsevier Science Ltd. All
rights reserved.
Introduction
Quantification of subsequent risk in stabilized
patients after acute coronary syndromes (ACS)
could improve clinical management and also facilitate designing clinical trials of strategies targeted
at preventing recurrent ischaemic events. Although
* Correspondence: L. Kristin Newby, MD, Duke Clinical Research Institute, P.O. Box 17969, Durham, NC, 27715-7969, U.S.A.
models have been developed that identify demographic, historical and clinical characteristics
associated with short-term outcomes after both
ST-segment elevation acute myocardial infarction
(MI) and non-ST-segment elevation ACS,1,2 the
clinical trial populations used have generally
been randomized within 6–24 h of symptom onset.
Thus, these models are limited by reflecting the
high early hazard of ACS patients. For example,
among ST-segment elevation MI patients in Global
0195-668X/03/$ - see front matter © 2003 The European Society of Cardiology. Published by Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0195-668X(02)00325-1
Predictors of outcome in stable post-ACS patients
Utilization of Streptokinase and tPA for Occluded
Coronary Arteries (GUSTO) I, 39% of 30-day deaths
occurred within 24 h; 55% within 48 h.1 Similarly,
among non-ST-segment elevation ACS patients randomized in Platelet glycoprotein IIb/IIIa in Unstable
angina: Receptor Suppression Using Integrilin
Therapy (PURSUIT), 56% of death or MI end-points
occurred within 96 h.3 Less is known about the
characteristics associated with outcome among
patients stable for several days or how clinical management during that time affects later outcome.
To identify characteristics associated with outcomes in patients stabilized after an ACS, we developed predictive models for 90-day death, death or
MI, and death, MI or severe recurrent ischaemia
leading to unplanned revascularization (SRI) using
the 15 904 patients randomized in the Sibrafiban
versus aspirin to Yield Maximum Protection from
ischaemic Heart events post-acute cOroNary
sYndromes (SYMPHONY) and 2nd SYMPHONY trials.
Methods
Patient population
Between August 1997 and August 1999, 15 904
patients from 931 clinical centres in 37 countries
were randomized in SYMPHONY and 2nd SYMPHONY
to evaluate sibrafiban, an oral platelet glycoprotein
IIb/IIIa inhibitor, for secondary prevention. Complete methods and results of these trials have been
published.4–6 Patients with both ST-segment elevation MI and non-ST-segment elevation ACS were
eligible if they presented with ≥20 min of chest pain
or anginal-equivalent symptoms and met electrocardiographic or cardiac marker criteria.4 Randomization occurred within 7 days, and patients were
stable for ≥12 h (without recurrent ischaemia or
haemodynamic instability and with Killip class <II)
at randomization.
SYMPHONY randomized 9233 patients a median
of 3.6 (interquartile range 2.2–5.1) days after ACS
to 90-day treatment with either aspirin 80 mg
twice-daily or one of two dosing strategies (high or
low) of sibrafiban twice-daily without background
aspirin. The 2nd SYMPHONY randomized 6671
patients at a median of 3.7 (2.6–5.5) days to either
aspirin 80 mg twice-daily, aspirin 80 mg plus lowdose sibrafiban twice-daily, or high-dose sibrafiban
alone twice-daily. Median treatment duration was
90 (35–138) days.
End-points
The SYMPHONY primary end-point was the 90-day
incidence of a composite of death, MI, or SRI
173
(death/MI/SRI). The time to this composite was the
2nd SYMPHONY primary end-point. The occurrence
of MI and SRI was determined by a blinded Clinical
Events Classification Committee for each trial using
prespecified criteria.4 SRI was defined as recurrent
ischaemic symptoms for ≥20 min resulting in
unplanned or unscheduled revascularization. The
MI end-point required elevation of CK-MB >upper
limit of normal (ULN) for clinical events, >3×ULN
post-percutaneous coronary intervention (PCI) and
>5×ULN post-bypass surgery, or the development of
new Q-waves in ≥2 contiguous ECG leads.
Statistical methods
General
Extensive information on demographics and
history, clinical characteristics and complications,
procedure and concomitant medication use, and
outcomes was collected in each study. Variables
collected and database structure were common to
SYMPHONY and 2nd SYMPHONY. Baseline characteristics in each study have been previously
published.5,6 After verifying the similarity of the
two populations and ensuring common variable
definitions, the databases were merged to form a
single, combined database. Descriptive statistics
(medians [interquartile ranges] and percentages)
were generated to summarize data.
Modelling
Because 2nd SYMPHONY was terminated early,
duration of follow-up varied widely, median 94
(64–157) days. Therefore, for development of the
outcomes models using the individual or combined
databases, we used Cox proportional-hazards modelling with censoring at 90 days. We developed
predictive models for death, death or MI (death/
MI), and death/MI/SRI.
For each model, univariable associations with
outcomes were assessed for each potential covariate shown in Appendix 1. Continuous variables that
were nonlinear with respect to outcome were
transformed using restricted cubic spline techniques. All variables for which there were fewer
than 200 patients missing data, were entered into
an initial multivariable model using a stepwise
selection process. If a variable with more than
200 patients missing was strongly predictive or considered clinically important it was added into the
initial candidate variables. After initial stepwise
model development, other variables were tested
one at a time in addition to those in the initial
model. Variables were retained if P<0.05. Satisfaction of the proportional-hazards assumption was
verified for each retained variable.
174
L.K. Newby et al.
The final death model included 14 970 patients
and 268 events, the death/MI model, 14 805
patients and 1024 events, and the death/MI/SRI
model, 14 906 patients and 1398 events. There
were only small differences in crude event rates
between the modelled groups and those excluded
from the models due to missing information. Among
patients excluded from the death model, mortality
was 1.71% vs 1.80% in the modelled group. The
difference was slightly larger for death/MI; 8.11% in
the group not included vs 6.92% in the modelled
group. For death/MI/SRI, the event rate was 9.43%
among patients not included compared with 9.38%
among those modelled.
Several strategies were used to validate the
models. First, individual models were developed
for each trial (SYMPHONY and 2nd SYMPHONY)
separately and compared for similarity prior to
the decision to use a final combined-group model.
Further, for models developed in both the separate
trials and the combined database, backward elimination was performed. Finally, bootstrapping (20
samples with replacement per model developed)
was also performed.
Results
The 15 904 patients in SYMPHONY and 2nd
SYMPHONY survived a median 3.6 (2.5–5.4) days
after their qualifying event and prior to randomization. The qualifying event was MI for 73.4% (63.7%
ST-segment elevation and 36.3% non-ST-segment
elevation) and for 26.6%, unstable angina.
Figure 1 displays event-free survival (a) and
hazard function (b) curves for death, death/MI, and
death/MI/SRI for the combined trial populations.
Previous trials of acute ST-segment elevation MI
have demonstrated three phases to these curves, a
high-risk acute period over the first 24–48 h, a more
modest-risk subacute period from 2–10 days, and a
subacute to chronic phase to 30 days. The curves
for each end-point in the current study population
support that the population was enrolled beyond
the initial high-risk period and that the modelling
reflects events occurring in the second and third
risk phases.
Mortality
Baseline characteristics by mortality status and univariable and multivariable associations between
pre-randomization characteristics and mortality for
the 17 independently associated variables are
shown in Table 1. Age and creatinine clearance
(CrCl) had the strongest univariable associations,
Figure 1 Event-free survival (a) and hazard function (b) curves
for death (solid line), death or myocardial infarction (broken
line), and death, MI, or severe recurrent ischaemia leading
to unplanned revascularization (dotted line) in the combined
SYMPHONY and 2nd SYMPHONY trials.
followed by a history of heart failure, heart failure
between qualifying event and randomization, and
diuretic use.
Among the 17 variables independently associated with mortality, greater age, higher randomization heart rate, heart failure between qualifying
event and randomization, history of heart failure,
lower qualifying event systolic blood pressure,
qualifying event MI, and angiotensin converting
enzyme inhibitor use prior to randomization were
most strongly associated with increased 90-day
mortality risk. Better renal function and PCI
prior to randomization were most strongly associated with reduced 90-day mortality risk. As shown
in Fig. 2, better renal function was associated
with lower mortality to a creatinine clearance
of 70 cc . min−1, above which risk remained
constant.
The full model likelihood ratio 2 was 367. A
model with age alone had a likelihood ratio 2 of
Predictors of outcome in stable post-ACS patients
Table 1
175
Pre-randomization characteristics associated with mortality
Wald 2
Dead
Age (years)
Randomization heart rate
(beats . min−1)
Heart failure between QE and
randomization
Creatinine clearance (cc . min−1)
History of heart failure
Percutaneous coronary
intervention between QE and
randomization
QE systolic blood pressure
(mmHg)
QE myocardial infarction
Pre-randomization ACE inhibitor
Pre-randomization thrombolytic
therapy
Prior coronary artery bypass
surgery
Hypertension
Prior myocardial infarction
Prior transient ischaemic attack
Chronic obstructive pulmonary
disease
Anginal symptoms in 6 weeks
before QE
Lateral electrocardiographic
changes at QE
70 (61, 77)
72 (64, 83)
10.6%
64 (49, 85)
16.7%
13.0%
Alive
59 (51, 68)
70 (61, 78)
2.3%
87 (68, 110)
4.1%
25.9%
140 (120, 160) 140 (124, 160)
Univariable
Multivariable
Adjusted hazard ratio (95% CI)
155.4
30.9
31.1
27.5
1.04 (1.03–1.06)
1.03 (1.02–1.03)
81.5
21.8
2.57 (1.73–3.83)
114.1
92.0
27.7
17.7
12.2
12.0
0.97 (0.96–0.99)
1.86 (1.31–2.64)
0.52 (0.36–0.75)
3.3
10.3
0.99 (0.99–1.00)
4.2
29.9
3.7
10.0
10.0
8.7
1.63 (1.20–2.20)
l.49 (1.16–1.90)
0.60 (0.43–0.84)
7.5
0.52 (0.33–0.83)
77.2%
54.0%
17.9%
73.4%
37.5%
26.2%
8.2%
9.1%
63.9%
32.7%
4.6%
7.7%
49.3%
19.9%
1.6%
3.1%
19.5
24.9
20.5
22.3
6.0
5.7
5.5
4.4
l.38 (1.07–1.80)
1.40 (1.06–1.85)
1.96 (1.12–3.44)
1.61 (1.03–2.51)
50.2%
43.9%
4.1
4.3
1.30 (1.02–1.66)
58.0%
48.0%
5.8
3.9
1.29 (1.00–1.65)
0.25
QE = qualifying event.
equation is shown in Appendix 2. Run as a logistic
regression, the model c-index was 0.80.
Death or myocardial infarction
Figure 2 Relationship of creatinine clearance with mortality
(solid line) with 95% confidence intervals (dotted lines) in the
combined SYMPHONY and 2nd SYMPHONY databases.
165, suggesting that age contributed approximately
45% of the full model’s prognostic information.
Together, the six strongest predictors in Table 1
accounted for approximately 81% of the model’s
prognostic information. The mortality model
Covariates in the 90-day death/MI model are displayed by strength of association in Table 2, and
the model equation in Appendix 2. Age and variables describing heart failure were less strong predictors of the double composite than death alone.
Elevated cardiac marker levels at qualifying event,
greater age, diuretic use, and recent anginal symptoms were most strongly associated with increased
risk, while increasing time between qualifying
event and randomization and PCI prior to randomization were most strongly associated with lower
90-day death/MI risk. In a pattern similar to that for
90-day mortality, higher creatinine clearance was
associated with lower 90-day death/MI. Run as a
logistic regression, the model c-index was 0.68.
Death/MI/SRI
The triple composite multivariable model (Table 3,
Appendix 2) was similar to the double composite
176
Table 2
L.K. Newby et al.
Pre-randomization characteristics associated with death/MI
Wald 2
Hours from QE to treatment
Percutaneous coronary intervention
between QE and randomization
Elevated cardiac markers at QE
Age (years)
Creatinine clearance (cc . min−1)
Pre-randomization diuretic use
Anginal symptoms in 6 weeks before
QE
Streptokinase for QE
Hypertension
Heart failure between QE and
randomization
Randomization heart rate
(beats . min−1)
Prior myocardial infarction
High-dose sibrafiban
Diabetes mellitus with end organ
damage
QE Killip class ≥II
Lateral electrocardiographic
changes at QE
Asian race
Never smoked
Pre-randomization heparin
History of heart failure
Recurrent ischaemia between QE
and randomization
Death/MI
No death/MI
Univariable Multivariable Adjusted hazard ratio (95% CI)
84 (56,122)
17.6%
92 (61,131)
26.0%
31.6
34.4
44.9
29.3
0.995 (0.994–0.996)
0.63 (0.53–0.74)
85.8%
64 (54, 73)
79 (60,104)
29.2%
49.5%
82.1%
59 (51, 68)
88 (68,110)
17.8%
43.6%
10.1
89.1
52.6
85.4
13.3
27.3
16.1
15.1
13.0
11.4
1.64 (1.36–1.97)
1.01 (1.007–1.020)
0.985 (0.978–0.993)
1.32 (1.14–1.54)
1.24 (1.10–1.41)
8.9
33.7
53.5
10.6
9.4
9.3
0.67 (0.53–0.85)
1.22 (1.08–1.39)
1.56 (1.17–2.08)
19.9
8.9
1.008 (1.003–1.013)
7.5%
58.6%
5.6%
70 (62, 80)
10.4%
48.9%
2.2%
70 (61, 78)
28.6%
36.5%
2.8%
19.5%
32.5%
1.3%
36.2
8.0
26.1
7.9
7.2
6.7
1.23 (1.07–1.42)
1.19 (1.05–1.35)
1.62 (1.13–2.33)
17.4%
53.5%
10.3%
47.8%
50.2
14.4
6.3
5.9
1.25 (1.05–1.49)
1.17 (1.03–1.32)
5.8%
30.2%
71.0%
9.4%
9.2%
4.4%
30.1%
68.3%
3.9%
7.7%
3.9
0.00
2.90
68.0
4.8
5.7
4.7
4.5
4.4
4.1
1.39
0.86
1.17
1.28
1.25
(1.06–1.82)
(0.75–0.99)
(1.01–1.34)
(1.02–1.62)
(1.01–1.54)
QE = qualifying event.
model. Factors most strongly associated with
90-day occurrence of death/MI/SRI included age,
recurrent ischaemia between qualifying event and
randomization, diuretic use, and elevated qualifying event cardiac markers. Longer time from
qualifying event to randomization, PCI between
qualifying event and randomization, and higher
creatinine clearance were most strongly associated
with lower risk. Run as a logistic regression, the
model c-index was 0.64.
Discussion
Using Cox proportional-hazards modelling we have
identified patient characteristics and variables
describing early clinical course and management
that predict risk for adverse 90-day outcomes
among patients who have survived the initial highrisk post-ACS period and who are clinically stable
after 3–4 days. These findings could be useful not
only in clinical management, but also in designing
clinical trials of new strategies intended to prevent
recurrent events after ST-segment elevation MI or
non-ST-segment elevation ACS.
Mortality
As in previous risk models, age was most strongly
associated with mortality in this cohort, contributing about 45% of the total prognostic information. Age remained strongly predictive despite
co-linearity with other model variables such as
creatinine clearance, not included in modelling
in previous studies. When creatinine clearance
entered the model, the 2 for age was decreased
by 63.
Consistent with previous reports of high cardiac
event rates and cardiovascular mortality in postACS patients with renal impairment,7–10 the
observed strength of association of renal function
with clinical outcomes highlights the high-risk
nature of renal impairment. Importantly, there was
no interaction of sibrafiban treatment with renal
function on the relationship with outcome. Patients
were excluded from the SYMPHONY trials for serum
creatinine >1.5 mg . dl−1, yet median creatinine
clearance was only 87 cc . min−1, and in approximately 25% it was <70 cc . min−1. In the patients
who died, median creatinine clearance was 64
Predictors of outcome in stable post-ACS patients
Table 3
177
Pre-randomization characteristics associated with death/MI/SRI
Wald 2
Death/MI/SRI
Hours from QE to treatment
Percutaneous coronary
intervention between QE and
randomization
Age (years)
Recurrent ischaemia between
QE and randomization
Creatinine clearance
(cc . min−1)
Pre-randomization diuretic use
Elevated cardiac markers at
QE
Heart failure between QE and
randomization
History of hypertension
Never smoked
Prior myocardial infarction
Univariable
Multivariable
Adjusted hazard ratio (95% CI)
85 (57, 124)
20.1%
No death/MI/SRI
92 (61, 131)
25.9%
29.7
25.4
37.8
24.4
0.996 (0.995–0.997)
0.71 (0.62–0.81)
63 (53, 72)
10.1%
59 (51, 68)
7.6%
76.4
15.7
17.4
15.9
1.012 (1.006–1.018)
1.43 (1.20–1.70)
81 (62, 106)
88 (69, 111)
42.7
13.6
0.988 (0.981–0.994)
26.5%
83.6%
17.8%
82.3%
63.0
1.6
11.6
11.3
1.26 (1.10–1.43)
1.29 (1.11–1.50)
4.7%
2.2%
36.0
9.0
1.50 (1.15–1.96)
57.0%
29.5%
27.2%
48.8%
30.2%
19.4%
28.7
0.3
38.3
7.7
7.4
7.3
1.17 (1.05–1.31)
0.85 (0.75–0.96)
1.20 (1.05–1.36)
Randomization SBP (mmHg)
120 (110, 135) 120 (110, 132)
Randomization SBP
Randomization SBP, squared
Randomization SBP, cubed
Anginal symptoms in 6 weeks
before QE
Prior percutaneous coronary
intervention
Streptokinase use for QE
Diabetes mellitus with
end-organ damage
Randomization heart rate
(beats . min−1)
Pre-randomization heparin use
QE Killip class ≥II
5.6
11.9 (3df)
1.27 (1.05–1.54)
0.998 (0.997–1.000)
1.00 (1.00–1.00)
48.3%
43.5%
10.4
6.8
1.15 (1.04–1.28)
14.4%
10.4%
18.1
6.7
1.24 (1.05–1.46)
8.0%
2.4%
10.5%
1.3%
8.2
18.4
6.5
5.9
0.77 (0.64–0.94)
1.53 (1.09−2.14)
7.9
5.4
1.01 (1.001–1.014)
3.3
29.2
4.4
4.3
1.14 (1.01–1.28)
1.18 (1.01–1.38)
70 (61, 80)
71.1%
15.3%
70 (61, 78)
68.2%
10.4%
QE = qualifying event.
(49–85) cc . min−1. Thus, serum creatinine is a suboptimal surrogate for renal function, and it is clear
that even modest degrees of renal dysfunction are
associated with adverse outcomes among ACS
patients.
The association of renal dysfunction with worse
outcome may be due to renal insufficiency itself or
related co-morbidities, but other factors may also
contribute. The Centers for Education and Research
on Therapeutics (CERTs) investigators have shown
failures to prescribe or adjust medications appropriately for renal dysfunction (Donal Reddan, personal communication) which could contribute to
adverse events. Thus, the association of renal dysfunction with worse outcomes should serve to highlight the need for both aggressive risk factor
modification and careful monitoring of prescription
patterns and medication dosing in these patients.
The tendency to exclude this high-risk group from
cardiovascular clinical trials should be carefully
re-examined.
Because most previous MI and ACS studies used
for modelling have randomized patients early after
presentation, it has been difficult to model the
association of early revascularization with mortality. Our results suggest a significant independent
relationship of early PCI with lower 90-day mortality, as well as the double and triple composite
end-points, observations that are consistent with
the findings of recent randomized trials of early
invasive strategies in ACS patients.11,12 Although a
mechanism for this association is not clear, an
antiinflammatory effect of PCI, as suggested
by enhanced benefit of an early invasive strategy among patients with elevated baseline
interleukin-6 levels in the FRISC II study, could be
operative.13 Other studies have suggested that PCI
may mitigate mortality risk associated with an
elevated WBC count (a potential proxy for inflammation).14 Alternatively, given that the hazard for
ischaemic events after PCI occurs largely within the
first 6–9 h after the procedure,15 rather than early
178
PCI resulting in lower risk, these patients may
simply have incurred that risk prior to being
randomized. Therefore, as suggested by Langer,
rather than PCI protecting from events, the association of PCI with reduced risk may simply reflect a
lower-risk population at randomization.16
Similar to models developed in patients randomized early in acute MI and ACS trials, new heart
failure after the qualifying event and higher heart
rate several days out from it predicted higher mortality risk. Such characteristics among ‘stable’
post-ACS patients may identify those relatively less
stable for whom longer hospital stays and more
aggressive inpatient treatment and outpatient
secondary prevention are warranted.
Death/MI and the triple composite
Creating models for the risk of death/MI or other
composite end-points has been more difficult, and
models have had lower discriminative ability and
performed less well across populations than mortality models. In part this may be due to the
heterogeneous MI end-point in clinical trials, which
includes both spontaneous and procedural MI.2
In addition, factors other than demographic and
clinical characteristics directly attributable to
patients or that occur after randomization, which
may be subject to variations in practice pattern,
may be more important in predicting non-fatal
outcomes.
As in previous studies, age and heart failure were
less strongly associated with death/MI or the triple
composite than with death alone. Time from qualifying event to randomization was most strongly
associated with death/MI risk and was nearly as
strongly associated with the triple composite. As
more time elapsed, a subsequent event was less
likely. This is consistent with previous observations
on event timing after MI in which the majority of
cardiac complications occurred within the first
24–48 h and were uncommon after 3 days.17 This
observation may reflect stabilization or healing of
ruptured plaque, use of therapeutic or procedural
interventions during this time, and/or simply a
survivor effect bias.
Diabetes mellitus with end organ damage—a
marker of severe or long-standing diabetes—was
weakly associated with 90-day death/MI and the
triple composite end-point although diabetes mellitus overall was not. This may reflect that other
variables such as creatinine clearance or age carry
much of the prognostic information explained by
diabetes. Regardless, the association of diabetes
with higher likelihood of recurrent ischaemic
L.K. Newby et al.
events emphasizes that this group should be
targeted for more aggressive use of proven secondary prevention measures, and warrants specific
study of this group in future clinical trials.
Presentation with elevated cardiac markers,
unstable symptoms prior to presentation, and prior
MI were among the strongest associations with
death/MI, predicting higher-risk even after stabilization for several days. Targeting these patients
may be particularly important in designing clinical
trials of secondary prevention strategies and in
improving use of proven therapies. However, to do
this most effectively will require development of
better predictive models than those currently
available. Similar to previous studies, the c-indices
of the double and triple composite models in this
study were only 0.68 and 0.64, respectively.
In addition to clinical risk predictors, incorporation of biochemical risk markers or the results of
continuous electrocardiographic ischaemia monitoring may provide additional information to more
effectively identify those at risk for non-fatal
recurrent ischaemic events. In the current models,
presentation marker status added prognostic information to clinical variables even in a selected
population stabilized for several days. Although
serial marker testing over the first 24 h provides
incremental prognostic information to baseline
testing,18 it is unclear if adding later measures
of these markers in patients who are stable
after several days would provide further useful
information.
Markers of inflammation may provide incremental risk stratification above that of clinical variables. Other studies have suggested that WBC count
at presentation may provide prognostic information.14,19 Less is known about WBC count measured
several days after presentation in stable patients.
In our database, 9006 patients had a WBC count
measured at randomization. When forced into our
models, it was an independent predictor of mortality, but not of the double or triple composite
end-points. In previous studies, C-reactive protein
(CRP), a marker of inflammatory system activation,
has also been shown to be predictive of short- and
long-term mortality after ACS,20,21 but like WBC in
our study, the incremental prognostic utility of CRP
for non-fatal events, is less clear.22–24
Several studies have shown that silent ischaemia
detected by continuous electrocardiographic STsegment trend monitoring also predicts cardiac risk
after ACS and may be complementary to cardiac
marker testing.25–28 Therefore, like serum risk
markers, incorporating the results of continuous
ECG monitoring for recurrent ischaemia into
Predictors of outcome in stable post-ACS patients
non-fatal ischaemic event risk models may also
improve prognostic ability.
Strengths and limitations
Overall, the models developed from the SYMPHONY
databases performed similarly to those developed
in other MI and ACS populations. A strength of our
modelling is that we investigated factors associated
with outcome in ACS patients stable for a median of
3.6 days after the index event. Thus, the models
reflect the characteristics and factors associated
with later risk in ACS patients. Also, available for
modelling were clinical, procedural, and medication use variables collected during the period from
qualifying event to randomization that in clinical
trial models have previously not been evaluated for
association with outcome. Many of the same covariates were also found to predict longer-term outcome following discharge among Nottingham Heart
Attack Register patients who had been admitted for
suspected MI, but in whom the diagnosis was not
confirmed.29 Application of these models may be
uniquely suited to risk assessment at or near the
time of discharge to guide further evaluation
and treatment and to counsel patients regarding
prognosis.
A potential limitation of our modelling is that our
population included only patients enrolled in clinical trials who, by selection, may not mirror the
breadth of risk in the general population. Only
future validation of our model in registries or other
unselected populations can confirm its broad
generalizability. We also did not prespecify covariates for inclusion in the modelling. This allowed us
to investigate associations of previously untested
variables with outcomes, but testing of a large
number of candidate variables may have resulted in
some associations due to chance. Use of backward
elimination and bootstrapping to assess the stability of variables included in the initial model and
cross-validation between models from the individual trial databases should reduce this likelihood,
but variables with weaker associations could reflect
spurious associations related to multiple testing.
Further investigation of these variables and their
associations with outcomes in other populations
may be helpful.
Conclusion
Among patients’ stable 3–4 days after an ACS,
multivariable models can identify patients at risk
for death or non-fatal recurrent ischaemic events.
However, typical clinical markers are better at
identifying risk of death than non-fatal MI or SRI.
179
Further investigation into incorporation of novel
laboratory markers of risk may refine the ability
to predict risk for non-fatal ischaemic outcomes.
Application of these risk models should be useful
both in clinical care and in designing clinical trials
to test new strategies for preventing recurrent
ischaemic events after ACS.
Acknowledgements
SYMPHONY and 2nd SYMPHONY were funded by
research grants from F. Hoffmann-La Roche, Ltd,
Basel, Switzerland.
Appendix 1
Candidate variables for multivariable
models
Demographics
Age, height, weight, body mass index, male sex,
race/ethnicity (white, black, Asian, Hispanic,
American Indian).
Risk factors and clinical history
Family history of coronary artery disease, history of
heart failure, hypertension, hypercholesterolaemia, history of angina, diabetes mellitus, diabetes
mellitus with end-organ damage, insulin-treated
diabetes mellitus, prior stroke, prior transient
ischaemic attack, prior myocardial infarction, prior
angiography, prior coronary artery bypass grafting,
severe chronic obstructive pulmonary disease,
cancer, abnormal liver enzymes, chronic renal
insufficiency, peripheral arterial disease, history of
atrial fibrillation, history of supraventricular tachycardia, history of ventricular dysrhythmias, history
of heart block, current smoker, never smoker, past
smoker, angina within 6 weeks before qualifying
event, multiple episodes of angina before qualifying event.
Clinical characteristics
Qualifying event diastolic blood pressure, randomization diastolic blood pressure, qualifying event
systolic blood pressure, randomization systolic
blood pressure, qualifying event heart rate, randomization heart rate, qualifying event Killip class
>II, qualifying event Killip class >III, qualifying
event mitral regurgitation, qualifying event S3
gallop, randomization mitral regurgitation,
randomization S3 gallop.
Qualifying event characteristics
Qualifying event myocardial infarction, qualifying
event ECG abnormalities (left bundle branch
180
block, paced rhythm, T-wave pseudonormalization,
Q wave, ST-segment depression, ST-segment elevation, T-wave inversion), location of ECG changes
(inferior, lateral, posterior, anterior), abnormal
qualifying event ECG, elevated cardiac markers at
qualifying event.
Clinical events between qualifying event and
randomization
Acute mitral regurgitation, atrial fibrillation, heart
failure, shock, pulmonary oedema, recurrent
ischaemia, heart block, supraventricular tachycardia, ventricular fibrillation.
Study treatment characteristics
Time from qualifying event to treatment, treatment assignment (control group, low-dose
sibrafiban, high-dose sibrafiban).
Laboratory measures
Serum creatinine, creatinine clearance (CrCl =
[(140-age) × weight in kilograms/serum creatinine
× 72] × 0.85 if female)
Medication use
Baseline medications (angiotensin converting enzyme inhibitor, antiarrhythmic, aspirin, betablocker, calcium channel blocker, coumadin,
digitalis, diuretic, heparin, intravenous glycoprotein IIb/IIIa inhibitor, lipid-lowering therapy, lowmolecular-weight heparin, nitrates, nonsteroidal
antiinflammatory agents, oral antiplatelet therapy,
statin, vitamins), pre-randomization thrombolysis,
thrombolysis for qualifying event, streptokinase for
qualifying event, rPA for qualifying event, tPA for
qualifying event, APSAC for qualifying event.
Procedure use
Any pre-randomization angiography, angiography
between qualifying event and randomization,
emergency percutaneous coronary intervention,
any pre-randomization stent, stent between the
qualifying event and randomization, any prerandomization percutaneous coronary intervention, percutaneous coronary intervention between
qualifying event and randomization, any prerandomization coronary artery bypass grafting,
coronary artery bypass grafting between qualifying
event and randomization.
Appendix 2
Equations for 90-day models
Mortality
Hazard = exp[(−0.00716 × QE SBP) + (−0.64652 ×
prior coronary bypass surgery) + (0.02453 × ran-
L.K. Newby et al.
domization heart rate) + (0.25094 × lateral ECG
changes at QE) + (0.03995 × age) + (−0.02595 ×
creatinine clearance) + (0.48715 × qualifying event
MI) + (0.25938 × angina in prior 6 weeks) + (0.39708
× pre-randomization angiotensin converting
enzyme inhibitor) + (0.94527 × CHF between QE and
randomization) + (0.32533 × hypertension) +
(0.67301 × prior transient ischaemic attack) +
(0.33639 × prior MI) + (0.47536 × chronic
obstructive pulmonary disease) + (−0.65614 ×
pre-randomization PCI) + (−0.51245 × prerandomization thrombolysis)].
Death or myocardial infarction
Hazard = exp[(0.01356 × age) + (0.00770 × randomization heart rate) + (0.49418 × elevated QE cardiac
markers) + (0.48121 × diabetes mellitus with endorgan damage) + (0.22424 × QE Killip class ≥II) +
(0.15448 × lateral ECG changes at QE) + (−0.00500 ×
hours from QE to treatment start) + (−0.01472 ×
creatinine clearance) + (0.17464 × high-dose
sibrafiban) + (0.33010 × Asian race) + (0.21542 ×
angina in prior 6 weeks) + (0.27811 × prerandomization diuretic) + (0.15235 × prerandomization heparin) + (0.44746 × CHF between
QE and randomization) + (0.21958 × recurrent
ischaemia between QE and randomization) +
(0.24772 × history of CHF) + (0.20136 × hypertension) + (0.20742 × prior MI) + (−0.46333 × prerandomization PCI) + (−0.39597 × streptokinase for
QE) + (−0.15394 × never smoked)].
Death, myocardial infarction, or severe
recurrent ischaemia
Hazard = exp[(0.25604 × elevated QE cardiac
markers) + (0.42182 × diabetes mellitus with endorgan damage) + (0.14207 × angina in prior 6 weeks)
+ (0.01201 × age) + (0.16519 × QE Killip class ≥II) +
(0.00745 × randomization heart rate) + (−0.00390 ×
h from QE to randomization) + (0.22873 × prerandomization diuretic) + (0.12881 × prerandomization heparin) + (0.40749 × CHF between
QE and randomization) + (0.35451 × recurrent
ischaemia between QE and randomization) +
(0.15931 × hypertension) + (0.17923 × prior MI) +
(0.21671 × prior PCI) + (−0.16702 × never smoked) +
(−0.34560 × PCI between QE and randomization) +
(−0.25631 × streptokinase for QE) + (0.23795 ×
randomization SBP) + (−0.00191 × randomization
SBP squared) + (5.01 × 10−6 × randomization SBP
cubed)]
Abbreviations: QE = qualifying event; SBP =
systolic blood pressure; MI = myocardial infarction;
PCI = percutaneous coronary intervention; CHF =
congestive heart failure; ECG = electrocardiogram
Predictors of outcome in stable post-ACS patients
References
1. Lee KL, Woodlief LH, Topol EJ et al. Coronary heart disease/
myocardial infarction: predictors of 30-day mortality in the
era of reperfusion for acute myocardial infarction: results
from and international trial of 41 021 patients. Circulation
1995;91:1659–68.
2. Boersma E, Pieper KS, Steyerberg EW et al. Predictors of
outcome in patients with acute coronary syndromes without
persistent ST-segment elevation: results from an international trial of 9461 patients. Circulation 2000;101:2557–67.
3. The PURSUIT Trial Investigators. Inhibition of platelet glycoprotein IIb/IIIa with eptifibatide in patients with acute
coronary syndromes. N Engl J Med 1998;339:436–43.
4. Newby LK, for the SYMPHONY Steering Committee. Longterm oral platelet glycoprotein IIb/IIIa receptor antagonism
with sibrafiban after acute coronary syndromes: study design of the Sibrafiban versus aspirin to Yield Maximum
Protection from ischemic Heart events post-acute cOroNary
sYndromes (SYMPHONY) trial. Am Heart J 1999;138:210–8.
5. The SYMPHONY Investigators. Comparison of sibrafiban with
aspirin for prevention of cardiovascular events after acute
coronary syndromes: a randomised trial. Lancet 2000;
355:337–45.
6. Second SYMPHONY Investigators. Randomized trial of
aspirin, sibrafiban, or both for secondary prevention after
acute coronary syndromes. Circulation 2001;103:1727–33.
7. Mann J, Gerstein HC, Pogue J et al., for the HOPE
Investigators. Renal insufficiency as a predictor of cardiovascular outcomes and the impact of ramipril: The HOPE
Randomized Trial. Ann Int Med 2001;134:629–36.
8. Krumholz HM, Chen J, Chen Y.-T. et al. Predicting one-year
mortality among elderly survivors of hospitalization for an
acute myocardial infarction: results from the Cooperative
Cardiovascular Project. J Am Coll Cardiol 2001;38:453–9.
9. Al-Suwaidi JM, Reddan D, Pieper KS et al. Prognostic importance of abnormal renal function in patients with acute
coronary syndromes. [Abstr]. Circulation 2000;102:II-793.
10. McCullough PA, Soman SS, Shah SS et al. Risks associated
with renal dysfunction in patients in the coronary care unit.
J Am Coll Cardiol 2000;36:679–84.
11. Cannon CP, Weintraub WS, Demopoulos LA et al. Comparison of early invasive and conservative strategies in patients
with unstable coronary syndromes treated with the glycoprotein IIb/IIIa inhibitor tirofiban. N Engl J Med 2001;
344:1879–87.
12. FRagmin and Fast Revascularisation during InStability in
Coronary artery disease (FRISC II) Investigators. Invasive
compared with non-invasive treatment in unstable coronary artery disease: FRISC II prospective randomised multicentre study. Lancet 1999;354:708–15.
13. Lindmark E, Diderholm E, Wallentin L et al. Relationship
between interleukin 6 and mortality in patients with unstable coronary artery disease: effects of an early invasive
or noninvasive strategy. JAMA 2001;286:2107–13.
14. Bhatt DL, Chew DP, Simoons ML et al. An elevated white
blood cell count is an independent predictor of mortality in
patients with acute coronary syndromes. [Abstr].
Circulation 2000;102:II-776.
15. Thel M, Califf RM, Tardiff BE et al., for the IMPACT-II
Investigators. Timing of and risk factors for myocardial
ischemic events after percutaneous coronary intervention
(IMPACT-II). Am J Cardiol 2000;85:427–34.
181
16. Langer A, Fisher M, Califf RM et al., for the CARS
Investigators. Higher rates of coronary angiography and
revascularization following myocardial infarction may be
associated with greater survival in the United States than in
Canada. Can J Cardiol 1999;15:1095–102.
17. Newby LK, Eisenstein EL, Califf RM et al. Cost effectiveness
of early discharge after uncomplicated acute myocardial
infarction. N Engl J Med 2000;342:749–55.
18. Newby LK, Christenson RH, Ohman EM et al., for the GUSTOIIa Investigators. The value of serial troponin T measurements for early and late risk stratification in patients with
acute coronary syndromes. Circulation 1998;98:1853–9.
19. Barron HV, Cannon CP, Murphy SA et al. Association between
white blood cell count, epicardial blood flow, myocardial
perfusion, and clinical outcomes in the setting of acute
myocardial infarction: a Thrombolysis in Myocardial Infarction 10 substudy. Circulation 2000;102:2329–34.
20. Morrow DA, Nader R, Antman EM et al. C-reactive protein is
a potent predictor of mortality independently of and in
combination with troponin T in acute coronary syndromes: a
TIMI 11A substudy. J Am Coll Cardiol 1998;31:1460–5.
21. Lindahl B, Toss H, Siegbahn A et al. Markers of myocardial
damage and inflammation in relation to long-term mortality
in unstable coronary artery disease. N Engl J Med 2000;
343:1139–47.
22. Biasucci LM, Luizzo G, Grillo RL et al. Elevated levels of
C-reactive protein at discharge in patients with unstable
angina predict recurrent instability. Circulation 1999;
99:855–60.
23. Toss H, Lindahl B, Siegbahn A et al., for the FRISC Study
Group. Prognostic influence of increased fibrinogen and
C-reactive protein levels in unstable coronary artery
disease. Circulation 1997;96:4204–10.
24. de Winter RJ, Bholasingh R, Lijmer JG et al. Independent
prognostic value of C-reactive protein and troponin I in
patients with unstable angina or non-Q-wave myocardial
infarction. Cardiovasc Res 1999;42:240–5.
25. Langer A, Krucoff MW, Klootwijk P et al. Prognostic significance of ST segment shift early after resolution of ST
elevation in patients with myocardial infarction treated
with thrombolytic therapy: The GUSTO-I ST Segment
Monitoring Substudy. J Am Coll Cardiol 1998;31:783–9.
26. Akkerhuis KM, Mass ACP, Klootwijk PAJ et al., for the
PURSUIT ECG-Ischemia Monitoring Substudy Investigators.
Recurrent ischemia during continuous 12-lead ECG ischemia
monitoring in patients with acute coronary syndromes
treated with eptifibatide: relation with death and myocardial infarction. J Electrocardiol 2000;33:127–36.
27. Langer A, Singh N, Freeman MR et al. Detection of silent
ischemia adds to the prognostic value of coronary anatomy
and left ventricular function in predicting outcome in unstable angina patients. Can J Cardiol 1995;11:117–22.
28. Norgaard B, Anderson K, Dellborg M et al., for the TRIM
study group. Admission risk assessment by cardiac troponin
T in unstable coronary artery disease: additional prognostic
information from continuous ST-segment monitoring. J Am
Coll Cardiol 1999;33:1519–27.
29. Packham C, Gray D, Silcocks P et al. Identifying the risk of
death following hospital discharge in patients admitted
with a suspected acute myocardial infarction in whom the
diagnosis is not confirmed. Eur Heart J 2000;21:213–9.