ISSN: 1941-7632

Copyright 2009 American Heart Association. All rights reserved. Print ISSN: 1941-7640. Online
Greenville Avenue, Dallas, TX 72514
Circulation: Cardiovascular Interventions is published by the American Heart Association. 7272
DOI:10.1161/CIRCINTERVENTIONS.108.840066
2009;2;366-375; originally published online Sep 15, 2009; Circ Cardiovasc Interv
Martin Fahy, W. John Boscardin and for the AMIHOT-II Trial Investigators
Barbara S. Lindsay, Bonnie H. Weiner, Alexandra J. Lansky, Mitchell W. Krucoff,
Bramucci, James C. Blankenship, D. Christopher Metzger, Raymond J. Gibbons,
Gregg W. Stone, Jack L. Martin, Menko-Jan de Boer, Massimo Margheri, Ezio
Coronary Intervention in Acute Myocardial Infarction
Effect of Supersaturated Oxygen Delivery on Infarct Size After Percutaneous
840066/DC1
http://circinterventions.ahajournals.org/cgi/content/full/CIRCINTERVENTIONS.108.
Data Supplement (unedited) at:

http://circinterventions.ahajournals.org/cgi/content/full/2/5/366
located on the World Wide Web at:
The online version of this article, along with updated information and services, is
http://www.lww.com/reprints
Reprints: Information about reprints can be found online at

journalpermissions@lww.com
410-528-8550. E-mail:
Fax: Kluwer Health, 351 West Camden Street, Baltimore, MD 21202-2436. Phone: 410-528-4050.
Permissions: Permissions & Rights Desk, Lippincott Williams & Wilkins, a division of Wolters

http://circinterventions.ahajournals.org/subscriptions/
at
Subscriptions: Information about subscribing to Circulation: Cardiovascular Interventions is online
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
Original Articles
Effect of Supersaturated Oxygen Delivery on Infarct Size
After Percutaneous Coronary Intervention in Acute
Myocardial Infarction
Gregg W. Stone, MD; Jack L. Martin, MD; Menko-Jan de Boer, MD; Massimo Margheri, MD;
Ezio Bramucci, MD; James C. Blankenship, MD; D. Christopher Metzger, MD;
Raymond J. Gibbons, MD; Barbara S. Lindsay, RN; Bonnie H. Weiner, MD;
Alexandra J. Lansky, MD; Mitchell W. Krucoff, MD; Martin Fahy, MSc; W. John Boscardin, PhD;
for the AMIHOT-II Trial Investigators
BackgroundMyocardial salvage is often suboptimal after percutaneous coronary intervention in ST-segment elevation
myocardial infarction. Posthoc subgroup analysis from a previous trial (AMIHOT I) suggested that intracoronary
delivery of supersaturated oxygen (SSO
2
) may reduce infarct size in patients with large ST-segment elevation
myocardial infarction treated early.
Methods and ResultsA prospective, multicenter trial was performed in which 301 patients with anterior ST-segment
elevation myocardial infarction undergoing percutaneous coronary intervention within 6 hours of symptom onset were
randomized to a 90-minute intracoronary SSO
2
infusion in the left anterior descending artery infarct territory (n222)
or control (n79). The primary efficacy measure was infarct size in the intention-to-treat population (powered for
superiority), and the primary safety measure was composite major adverse cardiovascular events at 30 days in the
intention-to-treat and per-protocol populations (powered for noninferiority), with Bayesian hierarchical modeling used
to allow partial pooling of evidence from AMIHOT I. Among 281 randomized patients with tc-99m-sestamibi
single-photon emission computed tomography data in AMIHOT II, median (interquartile range) infarct size was 26.5%
(8.5%, 44%) with control compared with 20% (6%, 37%) after SSO
2
. The pooled adjusted infarct size was 25% (7%,
42%) with control compared with 18.5% (3.5%, 34.5%) after SSO
2
(P
Wilcoxon
0.02; Bayesian posterior probability of
superiority, 96.9%). The Bayesian pooled 30-day mean (SE) rates of major adverse cardiovascular events were
5.01.4% for control and 5.91.4% for SSO
2
by intention-to-treat, and 5.11.5% for control and 4.71.5% for SSO
2
by per-protocol analysis (posterior probability of noninferiority, 99.5% and 99.9%, respectively).
ConclusionsAmong patients with anterior ST-segment elevation myocardial infarction undergoing percutaneous
coronary intervention within 6 hours of symptom onset, infusion of SSO
2
into the left anterior descending artery infarct
territory results in a significant reduction in infarct size with noninferior rates of major adverse cardiovascular events
at 30 days.
Clinical Trial Registrationclinicaltrials.gov Identifier: NCT00175058 (Circ Cardiovasc Intervent. 2009;2:366-375.)
Key Words: myocardial infarction

infarct size

angioplasty

reperfusion
M
yocardial salvage is frequently suboptimal despite
successful reperfusion in ST-segment elevation myo-
cardial infarction (STEMI), resulting in left ventricular dys-
function and increased mortality.
1,2
Although delays to reper-
fusion contribute to irreversible myonecrosis,
3
additional
causal mechanisms include microcirculatory dysfunction and
reperfusion injury.
4,5
Most prior studies of pharmacological
and mechanical interventions to reduce infarct size after
fibrinolysis and primary percutaneous coronary intervention
(PCI) have been negative.
6,7
The delivery of supersaturated
Received December 1, 2008; accepted July 30, 2009.
From the Columbia University Medical Center (G.W.S., A.J.L., M.F.), New York-Presbyterian Hospital and the Cardiovascular Research Foundation,
New York, NY; Sharpe-Strumia Research Foundation of the Bryn Mawr Hospital (J.L.M.), Main Line Health, Bryn Mawr, Pa; Isala Clinics
Weezenlanden (M.J.D.), Zwolle, the Netherland; Universitaria di Careggi (M.M.), Florence, Italy; Policlinico San Matteo (E.B.), Pavia, Italy; Geisinger
Medical Center (J.C.B.), Danville, Pa; Wellmont Holston Med Center (D.C.M.), Kingsport, Tenn; Mayo Clinic Foundation (R.J.G.), Rochester, Minn;
TherOx, Inc. (B.S.L.), Irvine, Calif; Worcester Medical Center (B.H.W.), Worcester, Mass; Duke Clinical Research Institute (M.W.K.), Durham, NC; and
University of California (W.J.B.), San Francisco, San Francisco, Calif.
The online-only Data Supplement is available at http://circinterventions.ahajournals.org/cgi/content/full/CIRCINTERVENTIONS.108.840066/
DC1.
Correspondence to Gregg W. Stone, MD, Columbia University Medical Center, New York-Presbyterian Hospital, the Cardiovascular Research
Foundation, 111 E 59th St, 11th Floor, New York, NY 10022. E-mail gs2184@columbia.edu
2009 American Heart Association, Inc.
Circ Cardiovasc Intervent is available at http://circinterventions.ahajournals.org DOI: 10.1161/CIRCINTERVENTIONS.108.840066
366 by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
oxygen (SSO
2
) with a PaO
2
of 760 to 1000 mm Hg in the
infarct-related artery immediately after successful reperfusion
markedly reduces infarct size in porcine coronary occlusion
models,
8
possibly by decreasing capillary endothelial cell
swelling, reducing formation of lipid peroxide radicals, alter-
ing nitric oxide synthase expression, and/or inhibiting leuko-
cyte activation and adherence.
913
Following promising pilot
study results,
1416
269 patients with anterior or large inferior
STEMI undergoing successful PCI within 24 hours of symp-
tom onset (the preclinical parameters for which SSO
2
suc-
cessfully reduced infarct size) were randomly assigned to
SSO
2
or control in the Acute Myocardial Infarction With
Hyperoxemic Therapy (AMIHOT)-I trial. In this study, in-
farct size measured by technetium (tc)-99m-sestamibi single-
photon emission computed tomography (SPECT) imaging at
14 days was not significantly different between the 2 treat-
ment groups.
17
However, the 105 patients with anterior
infarction reperfused within 6 hours assigned to SSO
2
had a
smaller median infarct size, less post-PCI residual ischemic
burden measured by ST-segment Holter monitoring, and im-
proved echocardiographic regional wall motion at 3 months.
17
Editorial see p 363
Clinical Perspective on p 375
As a post hoc subgroup analysis, these findings from the
AMIHOT-I trial are not definitive. We therefore performed a
second, prospective, randomized trial of SSO
2
therapy, this
time confined to patients with large anterior infarction under-
going PCI within 6 hours of symptom onset (AMIHOT II).
The study was powered with a Bayesian approach using
hierarchical modeling to allow partial borrowing of evidence
from the AMIHOT-I trial.
Methods
The AMIHOT-I Trial
The AMIHOT-II protocol intentionally preserved core design ele-
ments from AMIHOT I, the details of which have been previously
described.
17
In brief, from January 2002 to December 2003, 269
patients with anterior or large inferior STEMI and baseline Throm-
bolysis In Myocardial Infarction (TIMI) 0 to 2 flow undergoing
primary or rescue PCI within 24 hours after symptom onset were
enrolled in AMIHOT I. Patients in whom TIMI 2 to 3 flow was
achieved were randomly assigned to receive a 90-minute infusion of
SSO
2
in the infarct artery versus control (standard of care without
intracoronary infusion). Three primary efficacy end points were
prespecified: infarct size at 14 days measured by tc-99m-sestamibi
SPECT; total ischemic burden within 3 hours after reperfusion
measured by continuous Holter monitoring as the area under the
ST-segment level versus time curve (an end point distinct from
ST-segment resolution); and echocardiographic infarct zone regional
wall motion at 3 months. The primary safety end point was the
composite incidence of major adverse cardiovascular events
(MACE), defined as death, reinfarction, target vessel revasculariza-
tion, or stroke at 30 days. The baseline characteristics and outcomes
in the treatment groups have been previously reported.
17
The trial
prespecified examination of subgroups based on infarct location
(anterior versus nonanterior) and by time to reperfusion (less than or
greater than 6 hours).
The AMIHOT-II Trial, Study Population
Patients aged 18 years or older with anterior MI (ST-segment
elevation 1 mm in 2 contiguous precordial leads [V1V4] or new
left bundle-branch block with confirmation of left anterior descend-
ing coronary artery occlusion) and symptom onset within 6 hours
were considered for enrollment. Also for eligibility, angiographic
documentation of baseline TIMI 0 to 2 flow in a native coronary
artery and intended intracoronary stent placement were required.
Principal clinical and angiographic exclusion criteria included abso-
lute contraindications to anticoagulant therapy; hemorrhagic stroke
within 6 months; intra-aortic balloon pump counterpulsation or
cardiogenic shock; coronary artery bypass graft surgery within 30
days; severe valvular stenosis or insufficiency, pericardial disease,
nonischemic cardiomyopathy, ventricular septal defect, pseudoaneu-
rysm, or papillary muscle rupture; cardiopulmonary resuscitation for
10 minutes; expected survival 6 months due to noncardiac
comorbidities; current participation in other investigational device or
drug trials; inability or unwillingness to provide informed consent or
to agree to all follow-up study procedures; systemic arterial PO
2
80 mm Hg despite supplemental oxygen; severe target vessel
calcification or tortuosity; coronary stenosis 40% proximal to the
infarct lesion, unprotected left main stenosis 60%, or a significant
nonstented coronary dissection; total symptom-to-balloon time of
6 hours, or TIMI 0 to 1 flow at the end of procedure; or surgery or
additional PCI planned within 30 days after procedure. A screening
log was kept at all participating sites to document reasons for patient
ineligibility. The study was approved by the institutional review
board at each participating center, and consecutive eligible patients
signed informed written consent.
AMIHOT-II Protocol Procedures
and Randomization
Before angiography, an ECG was performed, a 24-hour 12-lead
continuous digital electrocardiographic monitor (180, Northeast
Monitoring, Maynard, Mass) was placed, cardiac biomarkers were
drawn (creatine phosphokinase [CK], CK MB fraction [CK-MB] or
troponin), and 325 mg of aspirin was administered. A clopidogrel
loading dose of 300 or 600 mg was recommended before procedure,
but in no case 4 hours after the procedure. Left ventriculography,
coronary arteriography, and PCI were performed with standard
techniques and commercially available devices. Anticoagulation
during PCI was achieved with intravenous unfractionated heparin.
Glycoprotein IIb/IIIa inhibitor and stent selection decisions were per
investigators discretion. After PCI, cardiac biomarkers were drawn
every 8 to 12 hours, aspirin was continued indefinitely, and 75 mg of
clopidogrel was administered daily for at least 1 month depending on
the stent type.
Eligible patients were randomized at the completion of the PCI
procedure in an open-label and unbalanced fashion (as described
later) to either an intracoronary infusion of SSO
2
or standard of care
without infusion. Randomization was performed using an automated
voice response system, in blocks of 19 (14 SSO
2
patients for each 5
control patients2.8:1) stratified by time to reperfusion (0 to 3 hours
or 3 to 6 hours) and lesion location (proximal or nonproximal left
anterior descending), accomplished using an adaptive scheme with a
biased coin randomization.
18
Device Description and Study Procedures
SSO
2
was delivered for 90 minutes using an extracorporeal circuit
(TherOx, Inc, Irvine, Calif), as previously described.
14,17
Blood is
withdrawn either from the side port of a single femoral sheath sized
2F larger than the PCI guide catheter (coaxial configuration), or
alternatively through a second 5F sheath placed in the contralateral
femoral artery and is oxygenated in a polycarbonate chamber to
achieve a PO
2
of 760 to 1000 mm Hg. Hyperoxemic blood is then
returned to the patient at 75 mL/min for 90 minutes through an
intracoronary infusion catheter placed in the infarct artery proximal
to the stent, during which the guide catheter is disengaged from the
left main coronary ostium. At the beginning of the protocol, the only
infusion catheter available was the 5.3F Tracker-38 (Target Thera-
peutics, Fremont, Calif), which in the coaxial configuration required
a 7F guide catheter and 9F sheath. During the latter phases of
enrollment the lower profile 4.6F MI-Cath infusion catheter
Stone et al Supersaturated Oxygen and Infarct Size 367
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
(TherOx, Inc) was introduced, which required a 6F guide catheter
and 8F sheath, allowing for a smaller arteriotomy.
The SSO
2
infusion was initiated in all patients in the cardiac
catheterization laboratory immediately after the final coronary an-
giogram, after which the patient could remain in this setting or be
transferred either to a holding area or the coronary care unit for
completion of the infusion. The systemic arterial PO
2
was measured
every 30 minutes during the 90-minute infusion, and nasal oxygen
adjusted to maintain the PO
2
80 mm Hg. The activated clotting
time was checked every 30 minutes and supplemental heparin
boluses administered as necessary to maintain the activated clotting
time 250 seconds during the active infusion.
Data Management
Independent study monitors verified 100% of case report form data
onsite. All adverse cardiac events were adjudicated by an indepen-
dent committee blinded to treatment allocation after review of
original source documentation. A data safety and monitoring com-
mittee periodically reviewed blinded safety data, each time recom-
mending the study continue without modification. Independent core
nuclear, electrocardiographic and angiographic laboratory analyses were
performed by technicians blinded to treatment assignment and clinical
outcomes using validated methods as previously described.
1921
Endpoints and Definitions
The primary efficacy end point was infarct size measured by
tc-99m-sestamibi SPECT at 14 days, powered for superiority. The
primary safety measure was composite MACE (as defined in
AMIHOT I), measured at 30 days, powered for noninferiority. Death
was defined as all-cause death. Reinfarction was defined as recurrent
ischemic symptoms lasting 20 minutes with new ST-segment
elevation and/or CK-MB re-elevation if occurring 96 hours after
the index event. Target vessel revascularization was defined as any
repeat PCI or bypass graft surgery of the study coronary artery.
Stroke was defined as a neurologic deficit lasting 24 hours, or 24
hours with a brain imaging study showing infarction.
Power and Statistical Analysis
A prespecified Bayesian hierarchical modeling approach was used
for analysis of the primary end points to allow partial borrowing of
evidence from the previously performed AMIHOT-I trial. Direct
specification of a previous distribution using the favorable results of
the AMIHOT-I trial would have inflated frequentist type I error
beyond acceptable levels.
22
The hierarchical Bayesian approach
allows inference for the AMIHOT-II trial to borrow a certain amount
of evidence from the AMIHOT-I trial, with the amount of borrowing
determined by the similarity of the data between the 2 trials.
23
This
may be conceptualized as precision weighted averages of the various
information sources for the parameter where the weights are data-
determined. Estimates are pulled in varying degrees toward the
overall mean, a behavior known as shrinkage.
24
Moreover, the model
was tuned such that the AMIHOT-II trial would have frequentist type
I error of no 5% level on the boundaries of the null hypotheses for
the primary end points.
25
A posterior probability of 95% for both
end points was required for success. The treatment versus control
effect differences for the primary efficacy end point were determined
using a pooled analysis adjusted for the study-specific medians.
Probability statements as to the strength of evidence for the efficacy
end point are made with respect to the Bayesian model, however.
Unbalanced randomization (2.8:1) was used to maximize power for
the safety end point for a given sample size; power gains for
unbalanced randomization in standalone noninferiority designs can
be substantial,
26
and the advantage is amplified in the Bayesian
hierarchical design. Further details of the power calculations and
Bayesian modeling for the primary efficacy and safety end points are
provided in the online-only Data Supplement.
Regarding the primary efficacy end point, assuming an absolute
reduction in infarct size of 5% of the left ventricle, randomizing 304
patients in a 2.8:1 ratio between SSO
2
and control in AMIHOT II
using Bayesian hierarchical modeling to incorporate the AMIHOT-I
findings, provided 85.4% power to demonstrate superiority of SSO
2
.
Regarding the primary safety end point, assuming 30-day rates of
MACE of 7% in both randomized arms, with a noninferiority of
6% (a margin agreed on with the Food and Drug Administration),
80.7% power was present to declare noninferiority between the
2 groups. This approach is distinctly different from simple pooling of
the AMIHOT-I subgroup data in patients with anterior MI reperfused
within 6 hours with the AMIHOT-II results. Simple pooling (which
greatly inflates type I error) would have provided significantly
greater power for the primary efficacy and safety end points (93%
and 86%, respectively). Conversely, it is also important to note that
AMIHOT II was intentionally underpowered as a standalone study
for the primary efficacy and safety end points (73% and 64% power,
respectively); statistical testing of AMIHOT II alone was therefore
not prespecified for primary end point analysis. Rather, some degree
of borrowing from the AMIHOT-I data would be necessary for either
primary end point to be satisfied (with the degree of borrowing
depending on the similarity of the datasets), with the principal
statistical analysis planned only on the blended Bayesian dataset.
Use of Bayesian analysis was restricted to assessment of the
primary end points. Secondary and subgroup analyses were con-
ducted using standard (frequentist) methods. Categorical variables
were compared by Fisher exact test. Continuous variables are
presented as meanSD or median (interquartile range), and were
compared by the nonparametric Wilcoxon rank-sum test. Exact
Wilcoxon 2-sample tests were used to compare infarct size data
between the SSO
2
and control groups. Linear regression analysis was
used to adjust for differences between the groups in age, gender,
prior MI, diabetes, infarct location, time to reperfusion, post-PCI
ST-segment resolution prerandomization, and major bleeding. All
primary and secondary analyses were performed in the intent-to-treat
population. A per-protocol subset was used as a coprimary analysis
for the primary (noninferiority) safety end point, consisting of
randomized patients not excluded due to a major protocol deviation
likely to impact the primary safety end point. Formal interaction
testing was used to assess the impact of baseline left ventricular
ejection fraction (LVEF) on the relative reduction in infarct size with
SSO
2
. Smoothed medians of infarct size distributions were computed
by kernel density estimation to de-emphasize the effects of discrete-
ness in smaller subgroups. Statistical evaluations using frequentist
methods in the AMIHOT-II study patients were performed using a
2-sided significance level of 0.05. Non-Bayesian statistical analyses
were performed by SAS version 9.1.3, Cary, NC. Bayesian inference
was conducted using Markov chain Monte Carlo computation by the
R2WinBUGS interface to WinBUGS 1.4.1 (Data Supplement).
Results
Patients
Between September 13, 2005, and May 26, 2007, a total of
2517 consecutive patients with STEMI were screened at 20
sites in 4 countries for enrollment in AMIHOT II, of whom
317 (12.6%) were enrolled (Figure 1). Of the 2200 excluded
patients, 1244 (56.5%) had a nonanterior MI and 388 (15.4%)
presented 6 hours after symptom onset. Among the remain-
ing 934 patients with anterior STEMI presenting within 6
hours of symptom onset, 617 (66.1%) were not eligible for
randomization, the most common reasons being cardiogenic
shock or intra-aortic balloon counterpulsation use, baseline
TIMI-3 flow, procedural complications, inability to obtain
consent, or physician discretion (Figure 1). Of the 317
eligible patients who provided informed consent, 13 received
SSO
2
as part of a nonrandomized training cohort, and 3
patients who were prematurely randomized in error were
subsequently deregistered before any treatment was delivered
after it was recognized that major exclusion criteria were
present (ventricular septal defect in 1 patient, prolonged
368 Circ Cardiovasc Intervent October 2009
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
cardiopulmonary resuscitation in 1 patient, and cardiogenic
shock in 1 patient). Thus, the intention-to-treat study cohort
consisted of 301 randomized patients, 222 of whom were
assigned to SSO
2
and 79 to control. Of the 301 randomized
patients, tc-99m-sestamibi SPECT infarct size assessment
was completed in 281 patients (93.4%), and no patient was
lost to follow-up at 30 days.
The baseline characteristics of the randomized AMIHOT-II
groups were well matched (Table 1). The enrolled
AMIHOT-II patients were also comparable with the cohort of
AMIHOT-I patients with anterior STEMI reperfused within 6
hours of symptom onset, except that the AMIHOT-I patients
were less likely to have hypertension and baseline TIMI-2
flow, had a lower baseline LVEF, were more likely to receive
glycoprotein IIb/IIIa inhibitors but less likely to undergo
thrombectomy (Table 1). Drug-eluting stents were also not
available during AMIHOT I.
AMIHOT-II SSO
2
Procedure
Among patients randomized to SSO
2
, the coaxial approach
was most commonly used, which required a 9F sheath to
accommodate the Tracker-38 infusion catheter (Table 2). To
avoid the 9F sheath, contralateral femoral arterial access was
used for the draw sheath in 40.1% of patients in whom the
Tracker-38 was used. In contrast, the coaxial (single sheath)
approach was used in almost all cases when the smaller
MI-Cath infusion catheter became available (Table 2).
A 90-minute or longer SSO
2
infusion was delivered in
84.7% of patients (Table 2). The infusion most commonly
took place in the cardiac catheterization laboratory, though
30% of patients received the infusion in other settings. Vital
signs and arterial blood gas measurements (assessed every 30
minutes) were stable during the infusion; neither the systemic
arterial PO
2
(140 mm Hg on supplemental low-flow nasal
cannula oxygen) nor oxygen saturation (98%) changed
during intracoronary SSO
2
infusion (data not shown).
Infarct Size
As shown in Figure 2, among 101 patients with anterior
STEMI reperfused within 6 hours in AMIHOT I in whom
infarct size was measured, the median (interquartile range)
infarct size (measured as the percentage of the left ventricle)
was 23% (5%, 37%) with control therapy compared with 9%
(0%, 30%) after SSO
2
(smoothed medians, 24% versus
17.5%, respectively). Among 281 randomized patients with
tc-99m-sestamibi SPECT data in AMIHOT II, infarct size
was 26.5% (8.5%, 44%) with control therapy compared with
20% (6%, 37%) after SSO
2
(unadjusted P0.10, adjusted
P0.03). The pooled study-level adjusted infarct size from
the AMIHOT I and II trials was 25% (7%, 42%) with control
therapy compared with 18.5% (3.5%, 34.5%) after SSO
2
(P
Wilcoxon
0.02; Bayesian posterior probability of superior-
ity, 96.9%). Figure 3 depicts the histogram of infarct sizes in
the pooled treatment and control cohorts, demonstrating a
shift to smaller infarcts across the range of the skewed
distribution. Among 154 patients with a baseline LVEF of
40%, infarct size was reduced from 33.5% (17.5%, 43.5%)
with control to 23.5% (7.5%, 38.5%) with SSO
2
, an absolute
reduction of 10% (0%, 14%), whereas the absolute decrease
in infarct size was less marked in the 196 patients with an
LVEF of 40% (16.5% [4.5%, 31.5%] with control versus
12.5% [2.5%, 30.5%] with SSO
2
, a reduction of 4% [1%,
7%], P for interaction, 0.60).
Cardiac Biomarker and ST-Segment Assessment
Among patients randomized in AMIHOT II to SSO
2
versus
control, there were no significant differences in the post-PCI
peak levels of CK-MB (299257 versus 289175 IU/L
respectively, P0.39) or troponin (96136 versus 128161
ng/mL respectively, P0.27). Nor were there significant
differences between the groups in post-PCI total ischemic
burden measured by the Holter monitor cumulative ST-
elevation time trend curve area at 3 hours (11782596 versus
13693684 V min, respectively, P0.69).
Clinical Outcomes
As shown in Table 3, by intention-to-treat analysis, the
Bayesian pooled 30-day mean (SE) rates of MACE were
5.01.4% for control and 5.91.4% for SSO
2
(posterior
Figure 1. Patient ow and follow-up in the
AMIHOT-II trial. See text for details. IABP
indicates intra-aortic balloon pump.
Stone et al Supersaturated Oxygen and Infarct Size 369
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
probability of noninferiority, 99.5%). By per-protocol analy-
sis, the Bayesian pooled 30-day rates of MACE were
5.11.5% for control and 4.71.5% for SSO
2
(posterior
probability of noninferiority99.9%).
Other adverse events in AMIHOT II appear in Table 4.
Hemorrhagic complications and access site-related events,
mostly hematomas, were more frequent in patients random-
ized to SSO
2
. Among SSO
2
patients, use of the smaller
MI-Cath infusion catheter compared with the Tracker-38 was
associated with a reduction in access site-related adverse
events (from 27.2% to 13.3%, P0.03), due mainly to fewer
access site reacted complications in the SSO
2
group with use
Table 1. Baseline Characteristics of the AMIHOT-I and AMIHOT-II Study Populations and Procedural Results Before Randomization
AMIHOT I
Anterior MI,
6 hours* (N105)
AMIHOT II
All patients
(N301) P
AMIHOT II
SSO
2
(N222) Control (N79)
Age, y 58.412.3 60.412.0 0.15 60.912.2 59.211.3
Male 80 (76.2) 242 (80.4) 0.40 173 (77.9) 69 (87.3)
Diabetes mellitus 10 (9.5) 47 (15.6) 0.14 36 (16.2) 11 (13.9)
Hypertension 33 (31.4) 140 (46.5) 0.008 104 (46.8) 36 (45.6)
Hyperlipidemia 36 (34.3) 134 (44.5) 0.09 100 (45.0) 34 (43.0)
Current smoking 50 (47.6) 119 (39.5) 0.17 85 (38.3) 34 (43.0)
Prior MI 9 (8.6) 27 (9.0) 1.0 20 (9.0) 7 (8.9)
Rescue PCI 4 (3.8) 18 (6.0) 0.47 11 (5.0) 7 (8.9)
Serum creatinine 1.5 mg/dL 3 (2.9) 8 (2.7) 1.0 6 (2.7) 2 (2.5)
Symptom onset to PCI, min 20373 20876 0.81 20974 20584
LVEF, % 37.810.9 40.58.7 0.005 40.28.6 41.39.1
Infarct artery location
Proximal LAD N/A 143 (47.5) 106 (47.7) 37 (46.8)
Mid-LAD N/A 150 (49.8) 109 (49.1) 41 (51.9)
Distal LAD N/A 5 (1.7) 5 (2.3) 0 (0.0)
Diagonal N/A 3 (1.0) 2 (0.9) 1 (1.3)
Initial TIMI flow grade (pre-PCI), site
0 73 (69.6) 206 (68.4) 155 (69.8) 51 (64.6)
1 18 (17.1) 25 (8.3) 0.008 19 (8.6) 6 (7.6)
2 14 (13.3) 70 (23.3) 48 (21.6) 22 (27.9)
Initial TIMI flow grade (pre-PCI), core lab
0/1 N/A 214/289 (74.0) 163/216 (75.5) 51/73 (69.9)
2 N/A 47/289 (16.3) 37/216 (17.1) 10/73 (13.7)
3 N/A 28/289 (9.7) 16/216 (7.4) 12/73 (16.4)
PCI performed 105 (100.0) 301 (100.0) 1.0 222 (100) 79 (100)
Stent implanted 105 (100.0) 297 (98.7) 0.36 220 (99.1) 77 (97.5)
Bare metal stent 105 (100.0) 134/297 (45.1) 0.0001 97/220 (44.1) 37/77 (48.0)
Drug-eluting stent 0 (0) 163/297 (54.9) 123/220 (55.9) 40/77 (52.0)
Thrombectomy 3 (2.9) 67 (22.3) 0.0001 50 (22.5) 17 (21.5)
Glycoprotein IIb/IIIa inhibitor used 100 (95.2) 202 (67.1) 0.0001 151 (68.0) 51 (64.6)
Final TIMI flow grade (post-PCI), site
2 7 (6.6) 17 (5.7) 0.81 11 (5.0) 6 (7.6)
3 98 (93.3) 284 (94.3) 211 (95.0) 73 (92.4)
Final TIMI flow grade (post-PCI), core lab
0/1 N/A 5/286 (1.7) 3/215 (1.4) 2/71 (2.8)
2 N/A 25/286 (8.7) 22/215 (10.2) 3/71 (4.2)
3 N/A 256/286 (89.5) 190/215 (88.4) 66/71 (93.0)
ST-segment resolution post-PCI, % N/A 59.025.1 65.225.3
Discharge medications
Aspirin 98 (93.3) 284 (94.4) 0.81 209 (94.1) 75 (94.9)
Thienopyridine 102 (97.1) 296 (98.3) 0.43 217 (97.8) 79 (100.0)
Data are presented as meanSD or n (%). LAD indicates left anterior descending; N/A, not available.
*Subgroup of patents in the AMIHOT-I trial with anterior MI reperfused within 6 hours of symptom onset.
370 Circ Cardiovasc Intervent October 2009
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
of a single unilateral compared with dual bilateral femoral
artery sheaths (from 45.2% versus 13.8%, P0.0001).
Discussion
The principal finding of this study, representing the net
results of a prespecified Bayesian analysis from 2 consecutive
randomized trials, is that in patients with anterior STEMI
undergoing successful PCI within 6 hours of symptom onset,
a 90-minute post-PCI infusion of SSO
2
significantly reduces
infarct size, with noninferior rates of MACE at 30 days. As
such, SSO
2
represents the first adjunctive therapy demon-
strated in a pivotal trial to reduce infarct size when used in
concert with a mechanical reperfusion strategy in STEMI.
SPECT imaging using tc-99m-sestamibi is the most widely
studied technique to evaluate myocardial salvage and infarct
size, having been shown to correlate with global and regional
left ventricular function and volumes after STEMI,
2731
clin-
ical outcomes after reperfusion therapy (including early and
late mortality and heart failure),
3136
and pathologically with
the extent of fibrosis in human hearts.
32,37
In this study, the
absolute reduction in infarct size with SSO
2
was greatest in
patients with the greatest clinical needthose with the largest
infarctions in whom the prognosis is known to be poor
despite successful PCI.
2,38
Although the median reduction in
infarct size with SSO
2
was 6.5% in the entire study popula-
tion, the median infarct size in patients with a baseline LVEF
of 40% was decreased from a median of 33.5% with control
to 23.5% with SSO
2
, representing incremental salvage of
10% of the left ventricular myocardium. Although the abso-
lute reduction in infarct size was less in smaller anterior
infarcts (a median 4% decrease in infarct size), the relative
reduction was comparable, as evidenced by the negative
interaction effect. Moreover, as the SSO
2
infusion is not
initiated until after PCI is completed, no delays to reperfusion
are required to deliver this therapy, representing a procedure
that can readily be incorporated into current reperfusion
treatment pathways in which minimizing door-to-balloon
time is an imperative.
39
Although this study was not designed to address the
mechanisms underlying the decrease in infarct size with
SSO
2
, neither post-PCI ischemic burden (representing resid-
ual or recurrent ischemia)
,
20
nor peak cardiac biomarker
Figure 2. Infarct size among patients with acute anterior myo-
cardial infarction in whom PCI was performed within 6 hours of
symptom onset randomized to SSO
2
versus control in the
AMIHOT-I and AMIHOT-II trials, with the adjusted pooled infarct
size estimates shown. Compared with control, SSO
2
was asso-
ciated with a signicant reduction in infarct size, as evidenced
by the Bayesian posterior probability of 95%. The thick black
lines represent the median, with the vertical limits of the boxes
representing the interquartile (25% to 75%) ranges. The limit
lines represent the 95% CIs. LV indicates left ventricle.
Figure 3. Histogram of infarct sizes (each vertical bar represent-
ing a 5% increment) in the pooled AMIHOT-I and AMIHOT-II
treatment and control cohorts, demonstrating a reduction in
infarct size with SSO
2
across the naturally skewed distribution
of infarct size. LV indicates left ventricle; IQR, interquartile
range.
Table 2. SSO
2
Procedure in 222 Randomized AMIHOT-II
Patients
Largest sheath size
7F 55 (24.8)
8F 48 (21.6)
9F 118 (53.2)
10F 1 (0.5)
Draw sheath approach
Coaxial 160 (72.1)
Contralateral femoral artery 62 (27.9)
Sheath size
5F 30 (13.5)
6F 32 (14.4)
Infusion catheter
Tracker-38 147 (66.2)
Coaxial draw sheath configuration 88/147 (59.9)
Contralateral femoral draw sheath 59/147 (40.1)
MI-cath (INCA-1) 75 (33.8)
Coaxial draw sheath configuration 72/75 (96.0)
Contralateral femoral draw sheath 3/75 (4.0)
SSO
2
infusion duration, min 81.225.5
60 min 25 (11.3)
6089 min 9 (4.1)
90 min 165 (74.3)
90 min 23 (10.4)
SSO
2
therapy delivery location
Catheterization laboratory 152/216 (70.4)
Holding area 5/216 (2.3)
Coronary care unit 53/216 (24.5)
Other 6/216 (2.8)
Data are presented as m (%) or meanSD.
Stone et al Supersaturated Oxygen and Infarct Size 371
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
levels (an important prognostic signal after primary PCI)
40
were improved with SSO
2
. However, recurrent ischemia was
uncommon in both groups, and varying biomarkers were
collected at different hospitals, assessed infrequently (every 8
to 12 hours), and not measured by a central core laboratory,
precluding reliable quantification. In experimental models,
improved microcirculatory function and reduction in reper-
fusion injury has been hypothesized to underlie many of the
beneficial effects of SSO
2.
813
Conceptually, these findings
also suggest that reperfusion injury continues to have an
important reversible component after epicardial flow restora-
tion, making possible beneficial therapies such as SSO
2,
which can be implemented without necessitating delay to
reperfusion.
Randomization to SSO
2
was associated with an increase in
hemorrhage-related adverse events, mostly access site hema-
Table 3. MACE Through 30 Days
SSO
2
Control P
Posterior Probability
of Noninferiority*
Intention-to-treat population
AMIHOT-I N134 N135
MACE, n (%) 9 (6.7) 7 (5.2) 0.62
AMIHOT-II N222 N79
MACE, n (%) 12 (5.4) 3 (3.8) 0.77
Death 4 (1.8) 0 (0) 0.58
Reinfarction 4 (1.8) 2 (2.5) 0.65
Target vessel revascularization 8 (3.6) 3 (3.8) 1.0
Stroke 0 (0) 0 (0)
MACE blended, adjusted Bayesian
meanSE
5.91.4% 5.01.4% 99.5%
Per-protocol population
AMIHOT-I N119 N124
MACE, n (%) 9 (7.6) 7 (5.6) 0.61
AMIHOT-II N186 N78
MACE, n (%) 7 (3.8) 3 (3.8) 1.0
Death 2 (1.1) 0 (0) 1.0
Reinfarction 3 (1.6) 2 (2.6) 0.63
Target vessel revascularization 5 (2.7) 3 (3.8) 0.70
Stroke 0 (0) 0 (0)
MACE blended, adjusted Bayesian
meanSE
4.71.5% 5.11.5% 99.9%
*Posterior probability that the SSO
2
therapy group MACE rate is not more than 6% greater than the control group
rate.
Table 4. Other Adverse Events Among Randomized Patients in AMIHOT-II
SSO
2
(N222) Control (N79) P
Stent thrombosis 9 (4.1) 2 (2.5) 0.73
Any access site related adverse event 50 (22.5) 10 (12.7) 0.07
Hematoma 39 (17.6) 8 (10.1) 0.15
Any hemorrhagic adverse event 55 (24.8) 10 (12.7) 0.03
Access site related* 41 (18.5) 9 (11.4) 0.16
Mild 34 (15.3) 8 (10.1) 0.34
Moderate 6 (2.7) 0 (0) 0.35
Severe 1 (0.5) 1 (1.3) 0.46
Nonaccess site related 16 (7.2) 1 (1.3) 0.05
Hemoglobin baseline, g/dL 14.3 (13.4, 15.5) 14.7 (13.7, 15.5) 0.27
Hemoglobin 24 hours, g/dL 12.9 (12.0, 13.8) 13.6 (12.6, 14.6) 0.0005
Transfusion 14 (6.3) 1 (1.3) 0.13
Data are presented as n (%) or median (interquartile range).
*Mild, does not require transfusion or result in hemodynamic compromise; moderate, requires transfusion; *Severe,
intracranial bleed or hemodynamic compromise requiring treatment.
372 Circ Cardiovasc Intervent October 2009
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
tomas due to use of contralateral femoral artery access to
avoid the 9F sheath required for the Tracker-38 infusion
catheter. The rates of access site-related complications and
bleeding were reduced to control levels with the introduction
of the lower profile MI-Cath infusion catheter, which facili-
tated use of a single smaller sheath, thus obviating contralat-
eral femoral artery access. As major bleeding can increase
mortality in STEMI,
41
minimizing hemorrhagic complica-
tions is essential if the benefits of infarct size reduction with
SSO
2
are to be realized.
A novel aspect of this investigation was specification of the
primary end point based on Bayesian hierarchical analysis,
allowing partial pooling of data from 2 consecutive random-
ized trials. Such methodology is well established,
4244
in-
creasingly used for randomized trials,
45,46
and described by
the Food and Drug Administration as an underutilized ap-
proach to reduce sample size, allowing pivotal trials to be
completed more rapidly and efficiently.
22
This study was
planned in concert with the Food and Drug Administration as
the US approval trial for SSO
2
in patients with anterior
STEMI undergoing PCI within 6 hours of symptom onset,
including selection of the primary efficacy and safety end
points. Bayesian hierarchical modeling allow data to be
borrowed from prior studies, with the extent of borrowing
depending on how closely the results from the new study
reflect the previous experience. Thus, if the results of AMIHOT II
varied greatly from AMIHOT I, little evidence would be
borrowed from the prior experience and the AMIHOT-II
results would be minimally changed (or could even be
adversely affected). The present Bayesian model avoids bias
from knowledge of the prior subgroup by ensuring that type
I conditional error is 5%, exactly the same type I error that
a standalone frequentist trial would have. In this study, a
reduction in infarct size was present in both trials in patients
with anterior STEMI reperfused within 6 hours of symptom
onset, allowing sufficient borrowing such that the pooled
Bayesian posterior probability for superiority was 96.9%,
signifying a significant reduction in infarct size with SSO
2
compared with control. Of note, the smoothed median differ-
ences in infarct size were similar in both AMIHOT I and
AMIHOT II. Thus, the finding of efficacy in the AMIHOT II
Bayesian model does not represent regression to the mean
had regression to the mean been present to a significant
degree, the posterior probability would not have been 95%.
Similarly, the posterior probability of noninferiority for
safety (30-day MACE) with SSO
2
compared with control was
99.5% and 99.9% in the intention-to-treat and per-protocol
populations, respectively, both highly statistically significant.
Use of Bayesian methodology in this investigation thus
allowed statistically valid study conclusions to be reached
with randomization of only 304 patients in AMIHOT II,
whereas 458 patients would have been required for 80%
power had traditional frequentist statistics been used. The
Bayesian approach is thus consistent with the US statutory
least burdensome pathway for clinical investigation and
device approval.
22
Several limitations of this investigation should be acknowl-
edged. No significant differences in survival at 30 days
between the control and treatment groups were present.
However, although the 6.5% median (4.5% mean) reduction
in infarct size with SSO
2
represents a greater improvement in
myocardial recovery than with tPA compared with streptoki-
nase,
47
or with primary PCI compared with tPA,
34
much
larger studies than AMIHOT II would be required to detect an
improvement in survival given the currently achieved low
mortality rates with contemporary primary PCI. The current
trial was also underpowered for a robust analysis of sub-
groups. The for noninferiority for the safety end point may
also be considered broad, although such a safety margin is
typical for regulatory device approval trials, and the Bayesian
estimates for noninferiority between SSO
2
and control were
highly significant by both intention-to-treat and per-protocol
analyses. Serial echocardiographic measures of regional wall
motion recovery, which correlate closely with tc-99m-
sestamibi infarct size and which improved in AMIHOT 1
with SSO
2
, were not measured in AMIHOT II as they are
more load dependent and technique sensitive than infarct size,
requiring a larger sample size. Finally, the safety and efficacy
results demonstrated for SSO
2
in the present study apply only
to those patients enrolled in AMIHOT II, and should not be
extrapolated to other patient cohorts, such as those with
nonanterior STEMI, patients reperfused beyond 6 hours after
symptom onset and those in cardiogenic shock.
In summary, in patients with anterior STEMI undergoing
successful PCI within 6 hours of symptom onset, a post-PCI
infusion of SSO
2
for 90 minutes safely reduces infarct size,
an effect which is most pronounced in patients with the
greatest amount of myocardium at risk, with noninferior rates
of MACE at 30 days.
Appendix
For AMIHOT-I trial organization and list of participating investiga-
tors, see reference 17.
AMIHOT-II Trial Organization and List of
Participating Investigators
Principal Investigator: G.W. Stone, Columbia University Medical
Center, New York Presbyterian Hospital and the Cardiovascular
Research Foundation, New York City, NY.
Co-Principal Investigator: J.L. Martin, Sharpe-Strumia Research
Foundation of the Bryn Mawr Hospital, Main Line Health, Bryn
Mawr, Pa.
Bayesian Statistician: W.J. Boscardin, University of California,
San Francisco, San Francisco, Calif.
Study Sponsor: TherOx, Inc, Irvine, Calif; B.S. Lindsay (Vice
President, Clinical Programs).
Site and Data Monitoring: TherOx, Inc, Irvine, Calif.
Data Management and Biostatistical Analysis: Boston Biomedical
Associates, Northborough, Mass.
Clinical Events Adjudication Committee: B.W. Weiner (Chair),
M.J. Schweiger, and S. Waxman.
Data and Safety Monitoring Board: D.W. Holmes (Chair), E.
Bates, J. Ferguson, W. Gaasch, and K. Freeman.
Nuclear Core Laboratory: The Mayo Clinic Nuclear Cardiology
Laboratory, Rochester, Minn; R.J. Gibbons (Co-Director), T. Miller,
P. Chareonthaitawee, and A. Lapeyre.
Angiographic Core Laboratory: The Cardiovascular Research
Foundation, New York, NY: A.J. Lansky (Director) and E. Cristea.
ECG and Holter Core Laboratory: Duke Clinical Research Insti-
tute, Durham, NC; M.W. Krucoff (Director) and C. Green.
Study Sites, Principal Investigators, and Primary Study Coordina-
tors: Saint Paul Hospital, Vancouver, BC: J.G. Webb, E. Grieve;
Mercy Heart Institute, Sacramento, Calif: M. Chang, W. Marquardt,
Stone et al Supersaturated Oxygen and Infarct Size 373
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
S. Bordash; Saint Agnes Hospital, Fresno, Calif: R. Plenys, C.
Okamoto; Mercy Hospital, Miami, Fla: M. Mayor, I. Mariota; Mercy
Medical Center, Des Moines, Iowa: M.A. Tannenbaum, C. Noyes;
Spedali Civili, Brescia, Italy: F. Ettori, C. Fiorina; Ospedaliera
Universitaria di Careggi, Florence, Italy: R. Margheri, G. Spaziani;
Policlinico San Matteo, Pavia, Italy: E. Bramucci, B. Marinoni, U.
Canosi; Harper University Hospital, Detroit, Mich: R. Spears, M.
Fathy; Henry Ford Health System, Detroit, Mich: A. Kugelmass, J.
Longlade; William Beaumont Hospital; Royal Oak, Mich, S. Dixon,
D. Richardson; Isala Klinieken Weezenlanden, Zwolle, Netherlands:
M.J. De Boer, D. Beuving; Allegheny General Hospital, Pittsburgh,
Pa: D. Lasorda, C. Harter; Geisinger Medical Center, Danville, Pa: J.
Blankenship, K. Skelding, D. Zimmerman; Sharpe-Strumia Research
Foundation of the Bryn Mawr Hospital, Main Line Health, Bryn
Mawr, Pa: J.L. Martin, A. Pratsos, C. Pensyl; Tri-State Medical
Center, Beaver, Pa: J. Rich, M. Kilhof; Jackson-Madison County
Medical Hospital, Jackson, Tenn: H.K. Lui, A. Hysmith; Wellmont
Holston Valley Medical Center, Kingsport, Tenn: C. Metzger, A.
Armstrong; Scott and white Hospital, Temple, Tex: S. Gantt, J. Asea;
East Texas Medical Center, Tyler, Tex: S.M. Lieberman, J. Crump.
Sources of Funding
The study was sponsored and funded by TherOx, Inc. The sponsor
was involved in study design and in data collection, analysis, and
interpretation, along with the principal investigators. The corre-
sponding author had full access to all the data in the study. The
manuscript was prepared by the corresponding author and revised by
all coauthors. The authors controlled the decision to submit the
manuscript for publication. The sponsor was provided the opportu-
nity for a nonbinding review of the manuscript before its submission.
Disclosures
Dr Stone reports having received research support from TherOx,
Abbott Vascular, Boston Scientific, and The Medicines Company.
Dr Martin reports having equity interests and serving as a consultant
to TherOx. Dr Blankenship reports serving on a speakers bureau for
Sanofi-Aventis. Dr Gibbons reports having received research support
from TherOx. Ms Lindsay is a full-time employee of and owns
equity in TherOx. Dr Weiner reports serving as a consultant to
TherOx and having received research support from Medtronic,
Boston Scientific, and Abbott Vascular. Dr Krucoff reports having
served as a consultant to and received research grants and consul-
tancy fees from TherOx. Dr Boscardin reports having served as a
consultant to TherOx. Drs de Boer, Margheri, Bramucci, Metzger,
Lansky, and Fahy report no conflicts of interest.
References
1. Bolognese L, Neskovic AN, Parodi G, Cerisano G, Buonamici P, Santoro
GM, Antoniucci D. Left ventricular remodeling after primary coronary
angioplasty: patterns of left ventricular dilation and long-term prognostic
implications. Circulation. 2002;106:23512357.
2. Halkin A, Stone GW, Dixon SR, Grines CL, Tcheng JE, Cox DA, Garcia
E, Brodie B, Stuckey TD, Mehran R, Lansky AJ. Impact and determinants
of left ventricular function in patients undergoing primary percutaneous
coronary intervention in acute myocardial infarction. Am J Cardiol.
2005;96:325331.
3. Gersh BJ, Stone GW, White HD, Holmes DR Jr. Pharmacological facil-
itation of primary percutaneous coronary intervention for acute myo-
cardial infarction: is the slope of the curve the shape of the future? JAMA.
2005;293:979986.
4. Yellon DM, Hausenloy DJ. Myocardial reperfusion injury. N Engl J Med.
2007;357:11211135.
5. Ito H. No-reflow phenomenon and prognosis in patients with acute
myocardial infarction. Nat Clin Pract Cardiovasc Med. 2006;3:499506.
6. Bolli R, Becker L, Gross G, Mentzer R Jr, Balshaw D, Lathrop DA.
Myocardial protection at a crossroads: the need for translation into
clinical therapy. Circ Res. 2004;95:125134.
7. Stone GW. Angioplasty strategies in ST-segment-elevation myocardial
infarction. II. Intervention after fibrinolytic therapy, integrated treatment
recommendations, and future directions. Circulation. 2008;118:552566.
8. Spears JR, Prcevski P, Jiang A, Brereton GJ, Vander Heide R. Intra-
coronary aqueous oxygen perfusion, performed 24 hours after the onset of
postinfarction reperfusion, experimentally reduces infarct size and
improves left ventricular function. Int J Cardiol. 2006;113:371375.
9. Buras J. Basic mechanisms of hyperbaric oxygen in the treatment of
ischemia-reperfusion injury. Int Anesthesiol Clin. 2000;38:91109.
10. Thom SR, Elbuken ME. Oxygen-dependent antagonism of lipid peroxi-
dation. Free Radic Biol Med. 1991;10:413426.
11. Hills BA. A role for oxygen-induced osmosis in hyperbaric oxygen
therapy. Med Hypotheses. 1999;52:259263.
12. Sirsjo A, Lehr HA, Nolte D, Haapaniemi T, Lewis DH, Nylander G,
Messmer K. Hyperbaric oxygen treatment enhances the recovery of blood
flow and functional capillary density in postischemic striated muscle.
Circ Shock. 1993;40:913.
13. Zamboni WA, Roth AC, Russell RC, Graham B, Suchy H, Kucan JO.
Morphologic analysis of the microcirculation during reperfusion of is-
chemic skeletal muscle and the effect of hyperbaric oxygen. Plast
Reconstr Surg. 1993;91:11101123.
14. Dixon SR, Bartorelli AL, Marcovitz PA, Spears R, David S, Grinberg I,
Qureshi MA, Pepi M, Trabattoni D, Fabbiocchi F, Montorsi P, ONeill
WW. Initial experience with hyperoxemic reperfusion after primary
angioplasty for acute myocardial infarction. Results of a pilot study
utilizing intracoronary aqueous oxygen therapy. J Am Coll Cardiol.
2002;39:387392.
15. Warda HM, Bax JJ, Bosch JG, Atsma DE, Jukema JW, van der Wall EE,
van der Laarse A, Schalij MJ, Oemrawsingh PV. Effect of intracoronary
aqueous oxygen on left ventricular remodeling after anterior wall
ST-elevation acute myocardial infarction. Am J Cardiol. 2005;96:2224.
16. Trabattoni D, Bartorelli AL, Fabbiocchi F, Montorsi P, Ravagnani P, Pepi
M, Celeste F, Maltagliati A, Marenzi G, ONeill WW. Hyperoxemic
perfusion of the left anterior descending coronary artery after primary
angioplasty in anterior ST-elevation myocardial infarction. Catheter Car-
diovasc Interv. 2006;67:859865.
17. ONeill WW, Martin JL, Dixon SR, Bartorelli AL, Trabattoni D, Oem-
rawsingh PV, Atsma DE, Chang M, Marquardt W, Oh JK, Krucoff MW,
Gibbons RJ, Spears JR; AMIHOT-Investigators. Acute myocardial
infarction with hyperoxemic therapy (AMIHOT): a prospective, ran-
domized trial of intracoronary hyperoxemic reperfusion after percutane-
ous coronary intervention. J Am Coll Cardiol. 2007;50:397405.
18. Efron B. Forcing a sequential experiment to be balanced. Biometrika.
1971;58:403417.
19. Gibbons RJ, Miller TD, Christian TF. Infarct size measured by SPECT
imaging with techne-tium-99m sestamibia measure of the efficacy of
therapy in acute myocardial infarction. Circulation. 2000;101:101108.
20. Krucoff MW, Croll MA, Pope JE, Pieper KS, Kanani PM, Granger CB,
Veldkamp RF, Wagner BL, Sawchak ST, Califf RM. Continuously
updated 12-lead ST-segment recovery analysis for myocardial infarct
artery patency assessment and its correlation with multiple simultaneous
early angiographic observations. Am J Cardiol. 1993;71:145151.
21. Lansky A, Popma J. Qualitative and quantitative angiography. In: Topol
EJ, ed. Textbook of Interventional Cardiology. Philadelphia, PA: WB
Saunders; 1999:725747.
22. Guidance for the use of Bayesian statistics in medical device clinical
trials. Draft guidance for industry and FDA staff. Available at http://
www.fda.gov/cdrh/osb/guidance/1601.pdf. Accessed August 29, 2008.
23. Pennello G, Thompson L. Experience with reviewing Bayesian Medical
Device Trials. J Biopharm Statistics. 2008;18:81115.
24. Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian Data Analysis. New
York: Chapman & Hall; 2003.
25. Inoue LY, Thall PF, Berry DA. Seamlessly expanding a randomized
phase II trial to phase III. Biometrics. 2002;58:823831.
26. Hilton JF. Designs of superiority and noninferiority trials for binary
responses are noninterchangeable. Biometric J. 2006;48:934947.
27. Christian TF, Behrenbeck T, Pellikka PA, Huber KC, Chesebro JH,
Gibbons RJ. Mismatch of left ventricular function and infarct size dem-
onstrated by technetium-99m isonitrile imaging after reperfusion therapy
for acute myocardial infarction: identification of myocardial stunning and
hyperkinesia. J Am Coll Cardiol. 1990;16:16321638.
28. Christian TF, Behrenbeck T, Gersh BJ, Gibbons RJ. Relation of left
ventricular volume and function over one year following acute myo-
cardial infarction to infarct size determined by technetium-99m-
sestamibi. Am J Cardiol. 1991;68:2126.
29. Behrenbeck T, Pellikka PA, Huber KC, Bresnahan JF, Gersh BJ, Gibbons
RJ. Primary PTCA in myocardial infarction: assessment of myocardial
salvage with Tc-99m-sestamibi. J Am Coll Cardiol. 1991;17:365373.
374 Circ Cardiovasc Intervent October 2009
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
30. Chareonthaitawee P, Christian TF, Hirose K, Gibbons RJ, Rumberger JA.
Relation of initial infarct size to extent of ventricular remodelling in the
year after acute myocardial infarction. J Am Coll Cardiol. 1995;25:
567573.
31. Burns RJ, Gibbons RJ, Yi Q, Roberts RS, Miller TD, Schaer GL,
Anderson JL, Yusuf S; CORE Study Investigators. The relationships of
left ventricular ejection fraction, end-systolic volume index and infarct
size to six-month mortality after hospital discharge following myocardial
infarction treated by thrombolysis. J Am Coll Cardiol. 2002;39:3036.
32. Miller TD, Christian TF, Hopfenspirger MR, Hodge DO, Gersh BJ,
Gibbons RJ. Infarct size after acute myocardial infarction measured by
quantitative tomographic technetium-99m sestamibi imaging predicts
subsequent mortality. Circulation. 1995;92:334341.
33. Miller TD, Hodge DO, Sutton JM, Grines CL, OKeefe JH, DeWood
MA, Okada RD, Fletcher WO Jr, Gibbons RJ. Usefulness of technetium-
99m sestamibi infarct size in predicting post hospital mortality following
acute myocardial infarction. Am J Cardiol. 1998;81:14911493.
34. Schomig A, Kastrati A, Dirschinger J, Mehilli J, Schricke U, Pache J,
Martinoff S, Neumann FJ, Schwaiger M. Coronary stenting plus platelet
glycoprotein IIb/IIIa blockade compared with tissue plasminogen activa-
tor in acute myocardial infarction: stent versus thrombolysis for occluded
coronary arteries in patients with acute myocardial infarction study inves-
tigators. N Engl J Med. 2000;343:385391.
35. Kastrati A, Mehilli J, Dirschinger J, Schricke U, Neverve J, Pache J,
Martinoff S, Neumann FJ, Nekolla S, Blasini R, Seyfarth M, Schwaiger
M, Scho mig A. Myocardial salvage after coronary stenting plus
abciximab versus fibrinolysis plus abciximab in patients with acute myo-
cardial infarction: a randomised trial. Lancet. 2002;359:920925.
36. Ross AM, Gibbons RJ, Stone GW, Kloner RA, Alexander RW. A ran-
domized, double-blinded, placebo-controlled multicenter trial of adeno-
sine as an adjunct to reperfusion in the treatment of acute myocardial
infarction (AMISTAD-II). J Am Coll Cardiol. 2005;45:17751780.
37. Medrano R, Lowry RW, Young JB, Weilbaecher DG, Michael LH, Afridi
I, He ZX, Mahmarian JJ, Verani MS. Assessment of myocardial viability
with technetium-99m sestamibi in patients undergoing cardiac transplan-
tation: a scintigraphic-pathologic study. Circulation. 1996;94:
10101017.
38. Halkin A, Singh M, Nikolsky E, Grines CL, Tcheng JE, Garcia E, Cox
DA, Turco M, Stuckey TD, Na Y, Lansky AJ, Gersh BJ, ONeill WW,
Mehran R, Stone GW. Prediction of mortality after primary percutaneous
coronary intervention for acute myocardial infarction: the CADILLAC
risk score. J Am Coll Cardiol. 2005;45:13971405.
39. Antman EM, Hand M, Armstrong PW, Bates ER, Green LA,
Halasyamani LK, Hochman JS, Krumholz HM, Lamas GA, Mullany CJ,
Pearle DL, Sloan MA, Smith SC Jr; 2004 Writing Committee Members,
Anbe DT, Kushner FG, Ornato JP, Jacobs AK, Adams CD, Anderson JL,
Buller CE, Creager MA, Ettinger SM, Halperin JL, Hunt SA, Lytle BW,
Nishimura R, Page RL, Riegel B, Tarkington LG, Yancy CW. 2007
Focused update of the ACC/AHA 2004 guidelines for the management of
patients with ST-elevation myocardial infarction. Circulation. 2008;117:
296329.
40. Halkin A, Stone GW, Grines CL, Cox DA, Rutherford BD, Esente P,
Meils CM, Albertsson P, Farah A, Tcheng JE, Lansky AJ, Mehran R.
Prognostic implications of creatine kinase elevation after primary percu-
taneous coronary intervention for acute myocardial infarction. J Am Coll
Cardiol. 2006;47:951961.
41. Stone GW, Witzenbichler B, Guagliumi G, Peruga JZ, Brodie BR, Dudek
D, Kornowski R, Hartmann F, Gersh BJ, Pocock SJ, Dangas G, Wong
SC, Kirtane AJ, Parise H, Mehran R. HORIZONS-AMI Trial Investi-
gators. Bivalirudin during primary PCI in acute myocardial infarction.
N Engl J Med. 2008;358:22182230.
42. Goodman SN. Toward evidence-based medical statistics, Part 2: The
Bayes factor. Ann Intern Med. 1999;130:10051013.
43. Spiegelhalter DJ, Myles JP, Jones DR, Abrams KR. Bayesian methods in
health technology assessment: a review. Health Technol Assess. 2000;4:
1130.
44. Diamond GA, Kaul S. Prior convictions: Bayesian approaches to the
analysis and interpretation of clinical megatrials. J Am Coll Cardiol.
2004;43:19291939.
45. Cuffe MS, Califf RM, Adams KF Jr, Benza R, Bourge R, Colucci WS,
Massie BM, OConnor CM, Pina I, Quigg R, Silver MA, Gheorghiade M.
Short-term intravenous milrinone for acute exacerbation of chronic heart
failure: a randomized controlled trial. JAMA. 2002;287:15411547.
46. Holmes DR Jr, Teirstein P, Satler L, Sketch M, OMalley J, Popma JJ,
Kuntz RE, Fitzgerald PJ, Wang H, Caramanica E, Cohen SA. SISR
Investigators. Sirolimus-eluting stents vs vascular brachytherapy for
in-stent restenosis within bare-metal stents: the SISR randomized trial.
JAMA. 2006;295:12641273.
47. Chareonthaitawee P, Gibbons RJ, Roberts RS, Christian TF, Burns R,
Yusuf S. The impact of time to thrombolytic treatment on outcome in
patients with acute myocardial infarction. Heart. 2000;84:142148.
CLINICAL PERSPECTIVE
Primary percutaneous coronary intervention (PCI) in patients with acute ST-segment elevation myocardial infarction has
become widely accepted as the preferred reperfusion modality because of its high success rate in restoring patency of the
occluded infarct artery, with resultant low rates of death, reinfarction, recurrent ischemia, and stroke. Nonetheless,
myocardial salvage is often suboptimal in many patients after primary PCI, in part because of late presentation and also
because of microcirculatory dysfunction and reperfusion injury. The intracoronary delivery of supersaturated oxygen with
a PaO
2
of 760 to 1000 mm Hg into the coronary artery supplying the myocardial infarct zone for 90 minutes after
successful primary PCI has been shown in preclinical models to markedly enhance myocardial recovery. In the randomized
AMIHOT-I and AMIHOT-II trials, this therapy was compared with primary PCI without intracoronary infusion in a total
of 406 patients with anterior ST-segment elevation myocardial infarction reperfused by successful PCI within 6 hours of
symptom onset. Compared with control, SSO
2
resulted in a significantly smaller infarct size at 14 days as measured by
tc-99m-sestamibi single-photon emission computed tomography imaging, with noninferior rates of major adverse
cardiovascular events at 30 days. The benefit in terms of infarct size reduction was particularly profound in patients with
the largest infarctions (baseline left ventricular ejection fraction 40%), in whom an incremental 10% salvage of the left
ventricular myocardium was noted. As such, supersaturated oxygen represents the first adjunctive therapy demonstrated in
a pivotal trial to reduce infarct size when used in concert with a mechanical reperfusion strategy in ST-segment elevation
myocardial infarction.
Stone et al Supersaturated Oxygen and Infarct Size 375
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from
Supplemental Material

Bayesian methodology
PRIMARY EFFECTIVENESS ENDPOINT ANALYSIS
The AMIHOT and AMIHOT-II effectiveness data were analyzed jointly in a hierarchical Bayesian
model. Primary effectiveness analysis was performed on an intent-to-treat basis.
It was necessary to develop a transformation of the primary endpoint infarct size that would, to a degree,
remove the skewness of the data (which includes a number of exactly zero values). Transformations of
the form Y=log(X + c) were considered, where log denotes the natural logarithm. Consideration of a
wide variety of choices for c gave reasonable approximations of normality, and extremely similar
statistical inference, and a value of c=10 was chosen based on the distribution of the AMIHOT-I data in
designing the AMIHOT-II trial. Using this transformation in the subgroup of patients with anterior
STEMI undergoing PCI in less than 6 hours (the anterior/LT6 subgroup), the mean (SD) for Control
subjects was 3.30 (0.66), and for SSO
2
subjects the mean (SD) was 3.06 (0.72). Thus, a difference of
almost one quarter of a log in these groups was observed. The data distributions are not exactly normal
after transformation, and the use of smearing
4
could be used to describe the treatment versus control
differences in medians on the original scale. More directly, the treatment effect has been reported using a
pooled analysis for the subgroup of interest adjusted for the study-specific medians.
As noted, the current randomized trial was entirely based on anterior patients with times to reperfusion of
less than or equal to 6 hours (LT6). Of the 269 patients randomized into the completed AMIHOT trial,
240 subjects had both infarct size data and time to reperfusion recorded. The anterior LT6 subgroup
represented a total of 101 patients. Thus, the original study can be regarded as being composed of 4
mutually exclusive subgroups: anterior/LT6, anterior/greater than (GT)6, non-anterior/LT6, non-
anterior/GT6. The Bayesian hierarchical model was used to analyze the AMIHOT data and the results
from the current study. In this model, the posterior mean effect size in any particular subgroup is shrunk
towards the overall mean. This procedure gives a formal methodology to account for the reduction of the
target population. The model also includes random study effects. As a result, if the difference between
the SSO
2
Therapy and Control means for the proposed study was different from that seen in the
AMIHOT trial, the analysis would not permit a high degree of borrowing. This allows the type I error to
be kept to a reasonable level even though the prior information is centered over a favorable outcome. In
particular, the following model was assumed. The log-transformed values (as discussed above) in each
subgroup were assumed to be normally distributed with a mean and standard deviation specific to the
subgroup-treatment (SSO
2
Therapy versus Control) combination.
The mean values for study i=1,2 (1=AMIHOT, 2=proposed study) and subgroup j
(j=1,2,3,4 in study i=1 and j=1 only for study i=2) are parameterized as:

C
ij
=Mean Control grouP=
0
+ e
C
j
+
C
i


T
ij
=Mean SSO
2
Therapy grouP=
C
ij
+
0

+ e
T
j
+
T
i
,
where
0
is the grand mean for the Control group, o
0
describes the overall SSO
2
Therapy versus Control
difference, e
C
j
represents the subgroup effect and
C
i
describes the study effect in the Control arm, and
1

by on November 17, 2009 circinterventions.ahajournals.org Downloaded from

e
T
j
represents the subgroup effect and
T
i
describes the study effect in the SSO
2
Therapy minus Cont
differences.
rol
i



The full model is written as

y
C
ijk
~ N (
C
ij
, o
C
2
)
y
T
ijk
~ N (
T
ij
, o
T
2
)

C
ij=

0
+ e
C
j
+
C
i


T
ij
=
C
ij
+
0

+ e
T
j
+
T

e
C
j
~ N(0, |
e
2
);
C
i
~ N(0, |

2
)
e
T
j
~ N(0,
e
2
);
T
i
~ N(0,

2
)
|
e
~ U(0.01, 0.67); |

~ U(0.01, 0.10)
t
e
~ U(0.01, 0.67); t

~ U(0.01, 0.10)
o
C
~ U(0.01, 2.0); o
T
~ U(0.01, 2.0)

0


~ N(3.2, 0.7
2
);
0
~ N(0, 0.7
2
)

where y
C
ijk
denotes the response of the kth Control subject in study i subgroup j (k runs from 1 to n
C
ij
),
and, similarly, y
T
ijk
denotes the response of the kth SSO
2
Therapy subject in study i subgroup j (k runs
from 1 to n
T
ij
). Specific choices of prior distributions were made to balance three goals: (i) be as non-
informative as possible, (ii) capture first-order substantive considerations, and (iii) allow the model to
have acceptable frequentist type I error in the case of no treatment effect. Some specific considerations
were as follows:

(i) The transformed response, log(y+10), is constrained to lie in the interval 2.3 to 4.7. Thus,
standard deviations of 0.5 or 1.0 are quite large relative to this range, and a standard
deviation of 2.0 is essentially non-informative.
(ii) The grand mean for Control infarct size (
0
) is centered around a prior mean that
corresponds to 15 on the original scale. The standard deviation of 0.7 then gives a 99.7%
prior probability that the grand mean can be between 0 and 100, a very broad range on the
original scale.
(iii) The overall mean for the SSO
2
Therapy minus Control difference (
0
) can be anywhere
from 0 to 100 on the original scale.
(iv) |
e
~ U(0.01, 0.67); |

~ U(0.01, 0.10). This is an extremely vague prior for the subgroup
effects in the Control group (which can easily span the full range of the data). Study
random effects for the Control group have a standard deviation no larger than 0.10 on the
log scale. This suggests that if the mean in a particular Control subgroup is 15 on the
original scale, then study-specific means for that subgroup could possibly vary from 8 to
23. The stronger prior on study effects is used to combat the problem of having only two
study effects for which we are attempting to estimate a variance component.
(v) t
e
~ U(0.01, 0.67); t

~ U(0.01, 0.10). The same comments noted above for the Control
random effects standard deviations, except that in the treatment case, the parameters refer
to differences between the SSO
2
Therapy and Control means.
(vi) o
C
~ U(0.01, 2.0); o
T
~ U(0.01, 2.0). The Control and SSO
2
Therapy group standard
deviations for individual measurements should be very precisely determined by the data,
and thus we specify only that these values are certain to be less than 2.0 on the logarithmic
scale.

Inference is based on the posterior distribution of the SSO
2
Therapy minus Control mean difference in
study 2, subgroup1, i.e. (
T
21

C
21
)=
0

+ e
T
2
+
T
1
, given the previous and current study data.
2

by on November 17, 2009 circinterventions.ahajournals.org Downloaded from

Consideration is given to the posterior probability that this combination of parameters is less than zero,
indicating that the SSO
2
Therapy group tends to have lower infarct sizes. Should this posterior
probability be 95% or higher, the effectiveness endpoint would have been fulfilled. That is, the
effectiveness endpoint would be satisfied if P(
T
21

C
21
< 0 | data from all subgroups in both trials) >
0.95. Inference was conducted using Markov chain Monte Carlo using an R front end
1
to WinBUGS
1.4.1
2
(called R2WinBUGS
3
). Sufficient run lengths were generated to ensure the accuracy of all decimals
reported in the Bayesian model results tables (more detail is given later in this section).

PRIMARY SAFETY ENDPOINT ANALYSIS
A hierarchical Bayesian model was also be used to evaluate the major adverse cardiovascular events
(MACE) rates in the two studies using the intention to treat population. The following set-up of the
model is assumed (similar to Warn, Thompson, and Spiegelhalter, 2002
5
). As above the study numbers
are i=1,2 (i=1 is AMIHOT, i=2 is proposed incremental study), the subgroup numbers are j=1,2,3,4 (j=1
is anterior/LT6, j=2 is anterior/GT6, j=3 is non-anterior/LT6, j=4 is non-anterior/GT6). For study 2, j=1
only. Then we denote r
C
ij
/n
C
ij
=# of Control MACE events/# of Control subjects in study i, subgroup j,
and r
T
ij
/n
T
ij
=# of SSO
2
Therapy MACE events/# of SSO
2
Therapy subjects in study i, subgroup j. The
full model is specified as follows:




r
C
ij
~ Bin (n
C
ij
,
C
ij
)
r
T
ij
~ Bin (n
T
ij
,
T
ij
)
logit(
C
ij
)=
C
ij

C
ij
=
0
+ e
C
j
+
C
i

T
ij
=
C
ij
+
0
+ e
T
j
+
T
i
(truncated to [0,1])
e
C
j
~ N(0, |
e
2
);
C
i
~ N(0, |

2
)
e
T
j
~ N(0,
e
2
);
T
i
~ N(0,

2
)
|
e
~ U(0.01, 0.30); |

~ U(0.01, 0.30)
t
e
~ U(0.001, 0.10); t

~ U(0.001, 0.033)

0
~ N(-2.6, 0.5
2
); o
0


~ N(0, 0.2
2
);

where
0
is the grand mean of MACE rates in the Control group (on logit scale), o
0
is the overall SSO
2

Therapy versus Control difference in safety rates (i.e. on risk difference scale), e
C
j
represents the
subgroup effect and
C
i
describes the study effect in the Control arm (these parameters are both on the
logit scale), and e
T
j
represents the subgroup effect and
T
i
describes the study effect in the SSO
2
Therapy
minus Control differences (these parameters are both on the risk difference scale). The basis for the
Bayesian model of these various parameters is as follows:

Treatment rates are modeled as an additive deviation from the Control rate in order to maximize
efficiency in testing risk differences.

(i)
0
~ N(-2.6, 0.5
2
). The grand mean of MACE rates in the Control group should be on the
order of 7%; the logit of 0.07 is equal to -2.6. The choice of standard deviation gives a 99.7%
prior probability that the average Control MACE rate is between 0% and 25%.
(ii) o
0


~ N(0, 0.2
2
). The overall SSO
2
Therapy versus Control difference in safety rates is
assumed to be normally distributed around zero with a standard deviation of 20 percentage
points. This is an extremely vague prior.
(iii) |
e
~ U(0.01, 0.30); |

~ U(0.01, 0.30). The random effects standard deviations for the logit
of Control rates between subgroups and between studies are assumed to be no larger than
0.30. If the mean rate is 7%, this translates into a 99.7% prior probability that subgroup or
3

by on November 17, 2009 circinterventions.ahajournals.org Downloaded from

of a trial.
n
ons.
(iv) t
e
~ U(0.001, 0.10); t

~ U(0.001, 0.033). The random effects standard deviations in the
differences between SSO
2
Therapy and Control means among the subgroups and between
studies are assumed to be no larger than 10 percentage points and 3.3 percentage points,
respectively. This in turn suggests that these differences between subgroups could vary as
much as 30 percentage points around the mean, and differences between the two studies
could be as large as 10 percentage points.

A second analysis similar to that described above for the intention to treat analysis (ITT) was performed
using the per protocol (PP) safety sample. Because it is unclear for a non-inferiority hypothesis whether
the ITT or PP subset is more conservative, both were considered primary for the primary safety endpoint
analysis. Again, the same Bayesian hierarchical model was employed with the original AMIHOT trial
data limited to the PP patient sample using the same limiting criteria as was used for the AMIHOT-II PP
patient sample.

EFFECTIVENESS BASED SAMPLE SIZE CALCULATION
The power of the current study with n
T
21=
224 patients in the SSO
2
Therapy arm and n
C
21=
80 in the
Control arm was calculated using the defined Bayesian hierarchical model. Computations were based o
2000 simulated data sets, with each data set analyzed by MCMC with 20,000 iterati

Table 1. Power to reach the conclusion of effectiveness for various true SSO
2
Therapy versus Control
differences.

A (Transformed
Scale)
Original
Scale
Posterior
Power
Frequentist
Power
Complete
Pooling
0.00 0.0% 5.0% 5% 17%
-0.20 -5.0% 85.4% 73% 93%
-0.25 -6.5% 96.0% 88% 97%
Prior Predictive 64.0%







The middle two lines in Table 1 demonstrate that there is 85.4% power to detect an effect size of 5.0%
infarct size reduction, and 96% power to detect a difference similar to that seen in the AMIHOT-I trial of
6.5%. In the other direction, as shown by the first line of the table, we found that the hierarchical model
was able to achieve an appropriate type I error rate when, in fact, the SSO
2
Therapy mean and the Control
mean are both equal. Finally, the bottom line of the table gives the prior predictive power for the study as
64%.

SAFETY BASED SAMPLE SIZE CALCULATION
The power of the current study with n
T
21=
224 patients in the treatment arm and n
C
21=
80 in the Control arm
was calculated using the defined Bayesian hierarchical model. Computations were performed using
MCMC (with 15000 iterations) for the full grid of possible pairs of MACE counts in the two arms of the
new trial.

4

by on November 17, 2009 circinterventions.ahajournals.org Downloaded from

In addition, for comparison, the frequentist power was calculated using the program PASS 2002 (Number
Cruncher Statistical Systems, Kaysville, Utah, 2002) and assumed no use of the AMIHOT-I data.
Table 2. Power table and Type I error calculations for the MACE safety endpoint.

C
21

T
21
n
C
21
=80, n
T
21
=224
Posterior Power Frequentist Power Complete Pooling
0.05 0.05 96.3% 75% 97%
0.06 0.06 89.8% 69% 92%
0.07 0.07 80.7% 64% 86%
0.08 0.08 70.5% 60% 79%
0.09 0.09 59.8% 56% 72%
0.07 0.13 4.7% 5% 8%
0.07 0.14 2.5% 2% 4%
0.07 0.15 2.0% 1.5% 2%
Prior Predictive 61.2%

Type I error calculations are shown in italics for a true Control rate of 0.07 and true SSO
2
Therapy rates
that are bigger than this rate by d=0.06 or more.
At the expected MACE rate of 0.07, it is evident that the proposed sample size of 80 Control and 224
SSO
2
Therapy patients would have achieved adequate power to conclude equivalence using the Bayesian
hierarchical model.

RUN LENGTHS FOR MCMC ANALYSIS OF PRIMARY EFFICACY ENDPOINT
We generated enough posterior simulations using the MCMC technique discussed above to ensure that
the first decimal point in the posterior probability of satisfying the efficacy or safety endpoints is correct.
We generated three chains, each with 2,000,000 iterations. The first 400,000 simulations were regarded
as burn-in and the resulting 1,600,000 simulations for each of the three chains were thinned, saving only
every 5
th
iteration. The resulting set of posterior simulations for analysis is thus of size 3 x
320,000=960,000. Convergence diagnostics indicated that the autocorrelation in this thinned set was
extremely low, but we conservatively regard this as an effective sample size of N
eff=
480,000. The Monte
Carlo standard error for estimating a probability of 95.0% is thus sqrt(0.950 * 0.005 / 480000)=0.0001,
ensuring that the first decimal place is correct with almost 100% confidence.


BAYESIAN SENSITIVITY ANALYSES FOR PRIMARY EFFICACY ENDPOINT
A secondary analysis similar to that described above for the ITT analysis was performed for the PP
analysis. The same Bayesian hierarchical model was employed with the original AMIHOT trial data
limited to the PP patient sample using the same criteria used for the AMIHOT-II PP patient sample.
Sensitivity analyses on the Bayesian primary effectiveness endpoint evaluation in both the ITT and PP
analyses were conducted by varying two of the assumptions for prior distributions on the
hyperparameters. The first model of these alternative models (M2) changes the prior distribution for the
standard deviation of the study random effects in the Control group from |

~ U(0.01, 0.10) to |

~
U(0.01, 0.20). This change allows the average Control infarct size to be more variable between
5

by on November 17, 2009 circinterventions.ahajournals.org Downloaded from

6

nd II.
AMIHOT-I and II. Because the AMIHOT-II trial featured larger infarct size results than the AMIHOT-I
study, the M2 value for this parameter is more appropriate than the setting in the pre-specified M1 value.
In particular, the prior distributions used in M1 put very little weight on the possibility of a Control group
infarct size for AMIHOT-II of higher than 25, whereas the M2 model allows for this amount of variation
in the two studies. The second of these alternative models (M3) makes a similar change to the prior
distribution for the standard deviation of the study treatment effects from t

~ U(0.01, 0.10) to t

~
U(0.01, 0.20). This change allows the treatment effect to be more variable between AMIHOT-I a
In order to gauge the appropriateness of models M1, M2, and M3 for the ITT analyses, we calculated the
Deviance Information Criterion.
6
The Deviance Information Criterion (DIC) is a measure of model fit,
with smaller values corresponding to better fit of the model to the data. Model M2 had the strongest
support in the data (DIC=974.8), followed by Model M1 (DIC=975.1), and then Model M3 (DIC=975.2).
Thus, the model fit is slightly preferable in M2, and this is also the model that gives reasonable a priori
possibility for the Control group infarct size to be above 25 in AMIHOT-II. All three models gave very
similar results with respect to the posterior probability of effectiveness.


References
1. R version 2.3.1. R Development Core Team. R: A language and environment for statistical computing.
R Foundation for Statistical Computing, Vienna, Austria, 2006.
2. Spiegelhalter DJ, Thomas A, Best NG, Lunn D. WinBUGS Version 1.4 Users Manual. MRC
Biostatistics Unit, Cambridge, 2003.
3. Sturtz S, Ligges U, Gelman A. R2WinBUGS: A Package for Running WinBUGS from R. Journal of
Statistical Software 2005;12:116.
4. Duan, N. Smearing estimate: a nonparametric retransformation method. Journal of the American
Statistical Association 1983;78:605-610.
5. Warn, D, Thompson, S, Spiegelhalter, D. Bayesian random effects meta-analysis of trials with binary
outcomes: methods for the absolute risk difference and relative risk scales. Statistics in Medicine
2002;21:16011623.
6. Spiegelhalter, DJ, Best, NG, Carlin, BP, and van der Linde, A. Bayesian measures of model complexity
and fit (with discussion). Journal of the Royal Statistical Society, Series B. 2002;64:583639.
by on November 17, 2009 circinterventions.ahajournals.org Downloaded from