Issues in the Early Termination of the Aspirin Component of the Physicians’ Health Study Data Monitoring Board of the Phys;cians’ Health Study: John Cairns, MD, Lawrence Cohen, MD, Theodore Coltor., ScD, David L. DeMets, PhD, Daniel Deykin, MD, Lawrence Friedman, MD, Peter Greenwald, MD, George B. Hutchison, MD, and Bernard Rosner, PhD The Physicians’ He&h Study is a rundomized, dot&-blind, placebo-controlled prevention trial of 22,071 US physicians, using a factorial design to evulttute the role of aspirin in the prevention of cardiovascular mortality and beta curotene in the reduction ofcancerincidence. After approximately 5 years of follow-up, rhe aspirin component was terminated, 3 years ahead of schedule. Sewerul factors were considered in the decision to terminate, inch&g a curdiouusatlar mortulity rate markedly lower than expected in both aspirin and placebo subjecrs, precluding the evaluation of the primary aspirin hypothesis, and a highly significant ( P < .OOOOl) and impressive (44%) reduction in the &k ofjirst myocurdial infarction in the aspirin ,pup. Issues in the decision to terminate ure described in this report. Ann Epidemiol 199 1; 1:395-405. KEY WORDS:
Cknrcul
trials,
early
terminutlon.
INTRODUCTION
On December 18, 1987, the Data Monitoring Board (DMB) of the Physicians’ Health Study (PHS) recommended early termination of the aspirin component of this randomized, double-blind, placebo-controlled trial. The beta carotene component was to be continued without interruption. The results of the aspirin component were published in a preliminary report on January 28, 1988 (1) and a final report on July 20, 1989 (2). As stated in these reports, it was unlikely that the PHS would observe sufficient cardiovascular (CV) deaths to evaluate this hypothesis until at least the year 2000, well beyond the planned end of the trial. The PHS also observed a statistical extreme ( P < .OOOOl)and 44% reduction (99 events in aspirin, 17 1 events in placebo groups) in the risk of a first myocardial infarction (MI) in the aspirin group. A brief rationale for the decision to terminate the aspirin arm early is provided in the two published reports. Since the issues were complex and discussed at length by the DMB, they are presented here in more detail.
STUDY
DESIGN
The design of the PHS has been described (3, 4) previously and we shall summarize here the aspects relevant to this report. Briefly, the PHS is a randomized, doubleFrom Harvard School of Pubhc Health, Boston, MA (J.C., J.B.H.); Yale Unwersity School of Medlclne, New Haven, CT (L.C.); Boston University School of Public Health, Boston, MA (T.C.); Umverslty of Wisconsin Biostatlstics Center, Madison, WI (D.L.D.); Boston Veterans Administration Medical Center, Boston, MA (D.D.); National Heart, Lung, and Blood Institute, Bethesda, MD (L.F.); Nattonal Cancer Institute, Bethesda, MD (P.G.); and Harvard MedIcal School, Boston, MA (B.R.). Address reprint requests to: David L. DeMets, PhD, Director, Biostatistics Center, Department of Statistics, University of Wisconsin Medical School, 5th & 6th Floors, Medical Sciences Center, 1300 University Avenue, Madison, WI 53706. Recetved March 28, 1990; revised October 12, 1990. 0 1991 Elsevier SciencePuhhahmg Co.. Inc.
1047.2797/91/$3.50
396
AEP Vol. 1, No. 5 AugustI991: 395-405
Cairns et al. EARLY TERMINATION OF ASPIRIN IN PHS
blind, placebo-controlled a 2 X 2 factorial design. taken
every other
placebo.
day would decrease
The other question,
on alternate their
outcomes,
As stated
and their
no previous effects
on total
confirmed
stroke.
at this dose.
aspirin
tative
mortality
aim of the aspirin
to determine
high compliance
whether
including
could
therefore,
total
all eligible
tolerate
mortality
study population.
this
select
mortality
a sample
mortality
aspirin
mortality
estimate
1983 and, design,
experience
DMB
consisted
physicians
aspirin
arm and
of clinicians,
to the DMB was to examine
effects
requiring
alteration
the DMB adopted recognized
that
monitoring
the key secondary
interim
end points
policies
that served as guidelines
analyses
and consideration
using a conventional
analyses corresponds significance
to guard against this requires a standardized of any
primary
outcome
CV
lower mortality in 1982 and
the 2 x The
2 factorial
follow-up
the course
was
of the trial. The
of beneficial
primary
or adverse
of one or both parts of the trial. every 6 months.
.05) at each analysis would lead to false-positive of that significance level (6). For these reasons, boundary procedure
assumption
population
statisticians.
the data for early evidence
approximately
interim
into
during and
have
and 75% thereafter.
to the placebo. results
or early termination
to meet
frequent
11,034
subjects/ significance
arbitrary
general
year,
were randomized
epidemiologists,
charge
DMB was scheduled
of the
to 7.5 years based on anticipated
scheduled to end in late 1990. A DMB was appointed to review The
from a largely 25%
and as the
for 5.0 years would
aged 40 to 84 years, were screened
22,071
to the
of feasibility
with a one-tailed
50% in the second
male physicians,
ultimately,
11,037
to
trials (5)) the quantiby about 20% and,
of 4.5 deaths/l000
followed
was derived
would
period was revised
260,000
were
well enough
For reasons
in CV mortality
in the first year after entry,
Over
of side effects physicians
no prior MI, were selected
estimate
size of 22,000
a 20% reduction
population
Later, the follow-up rates,
10% was proposed.
40 to 84 years old with
Based on an initial
power of .95 to detect that
by about
groups,
level of .05. This
the
MI and
in a trial of long duration.
US male physicians,
y in the placebo
men with
nonfatal
the frequency
Based on previously reported results in secondary prevention hypothesis that aspirin would reduce total CV mortality
cost,
is to assess the healthy
The first is to determine
was used in which subjects
design
in a single study.
component
aim is to determine
period
a factorial
of apparently
aims are stated.
taken
that the two diseases,
related,
and on CV morbidity
The other secondary A run-in
Given
to two hypotheses
in a population
MI. Two groups of secondary
of aspirin ensure
the primary
on total CV mortality
of aspirin
of cancer.
the answers
with that seen with
50 mg of beta carotene
are not appreciably
to obtain
in the protocol,
as compared
is whether
the incidence
treatments
opportunity
effect of aspirin
CV mortality,
still under study,
days would decrease
offers a unique
given
primary prevention trial that addresses two questions using The question of interest here is whether 325 mg of aspirin
The
Early in the follow-up, for the trial. The DMB of both
the primary
level of significance
(e.g.,
and P =
results with a frequency far in excess the DMB adopted the Haybittle-Peto
excess likelihood of false-positive results (7). This or normal t statistic of 3.0 or greater during interim before
consideration
of early
termination.
This
before claiming statistical to a nominal P value of .0013 ( one-tailed) at an intermediate observation. With this conservative approach, signifi-
cance levels (e.g., P < .05) only slightly below the conventional level can still be used at the final scheduled analyses. The DMB also recognized, as had others (8-10)) that factors bearing on the decision to terminate any study were complex and that simple rules or boundaries should not be used alone but should be incorporated with
AEP Vol. I, No. 5 August 1991: 395-405
EARLY TERMINATION
TABLE 1 Date’
Cairns et al. OF ASPIRIN IN PHS
397
Confirmed deaths Aspirin
Outcome
Placebo
RR
P
6186
cv
28
33
.a3
.52
1187
Total cv
58 37
75 42
.76 ,116
.I4 .57
12187
Total CV
9I 44
102 44
.88 .99
.43 .99
I15
.95
.74
llil
Total
other information of evidence.
into guidelines.
The final decision would be based on the totality
INTERIM RESULTS AND DELIBERATIONS As described by the Coronary Drug Project Research Group (8), a number of factors must be considered in reviewing interim data before making any decision for early termination. In the PHS, the DMB examined the issues of: (1) initial comparability of treatment groups with respect to known risk factors; (2) compliance with treatment; (3) completeness of follow-up; (4) confirmation of the specified primary study outcomes; (5) secondary study outcomes; and (6) data on known complicating effects of aspirin use, chiefly bleeding phenomena. Randomization produced highly comparable treatment groups so that any outcome differences might be attributed to active treatment and not any imbalance in risk. Compliance with assigned treatment, either aspirin or placebo, as measured by proportion of assigned treatments taken was high throughout and was over 85% in each group at the time of termination. Use of other concomitant therapies was evenly distributed as well. Follow-up was 100% complete with respect to mortality and 99.7% for morbidity. Confirmation of all outcomes was sought by an end point committee blinded as to treatment assignment and independent of the DMB. Summary results for the last three interim analyses before termination are provided in Tables 1 through 3, which show total and CV mortality, confirmed nonfatal and total MI, and confirmed ischemic and hemorrhagic stroke. The data in these tables were given special attention by the DMB relative to possible changes in conduct of the study. Bleeding problems, including bruising, hematemesis, melena, other gastrointestinal hemorrhage, and epistaxis, were all highly significantly increased in the aspirin group. The absolute rates of these problems, however, were lower than reported in previous trials, presumably reflecting the low dose of aspirin in this trial and the selection at enrollment, including the run-in period. Gastrointestinal ulcers and other gasrointestinal symptoms were also increased in the aspirin groups, though, for most of these morbidity categories, the differences between aspirin and placebo were not statistically significant. The DMB did not consider any of these findings sufficient reason to change the conduct of the study. Between July 1986 and December 1987, the DMB held three formal meetings and numerous informal discussions centered around five Issues: 1.
The CV mortality
rate and time trend.
398
Cairns et al. EARLY TERMINATION
TABLE
2
Confirmed
myocardial
Outcome
Date”
infarctions Aspirin
10185 7186 1187 12187
46 56 7i 75 85 89 9Y 104
2.
The CV mortality The statistical
CV mortality
study years already The
5.
The stroke
available process
rate and conditional
MI rate difference
the investigators
to terminate
CV, showed
between
rate difference
to the DMB.
As is evident
between
.47
.0003 -
.5Y .61 .61 .60 .56 .56 .53
aspirin
of the trial and length
P
.OOlY ,003 .0007 .0004 < .OOOl
Cknrcul
trials,
early
terminutlon.
INTRODUCTION
On December 18, 1987, the Data Monitoring Board (DMB) of the Physicians’ Health Study (PHS) recommended early termination of the aspirin component of this randomized, double-blind, placebo-controlled trial. The beta carotene component was to be continued without interruption. The results of the aspirin component were published in a preliminary report on January 28, 1988 (1) and a final report on July 20, 1989 (2). As stated in these reports, it was unlikely that the PHS would observe sufficient cardiovascular (CV) deaths to evaluate this hypothesis until at least the year 2000, well beyond the planned end of the trial. The PHS also observed a statistical extreme ( P < .OOOOl)and 44% reduction (99 events in aspirin, 17 1 events in placebo groups) in the risk of a first myocardial infarction (MI) in the aspirin group. A brief rationale for the decision to terminate the aspirin arm early is provided in the two published reports. Since the issues were complex and discussed at length by the DMB, they are presented here in more detail.
STUDY
DESIGN
The design of the PHS has been described (3, 4) previously and we shall summarize here the aspects relevant to this report. Briefly, the PHS is a randomized, doubleFrom Harvard School of Pubhc Health, Boston, MA (J.C., J.B.H.); Yale Unwersity School of Medlclne, New Haven, CT (L.C.); Boston University School of Public Health, Boston, MA (T.C.); Umverslty of Wisconsin Biostatlstics Center, Madison, WI (D.L.D.); Boston Veterans Administration Medical Center, Boston, MA (D.D.); National Heart, Lung, and Blood Institute, Bethesda, MD (L.F.); Nattonal Cancer Institute, Bethesda, MD (P.G.); and Harvard MedIcal School, Boston, MA (B.R.). Address reprint requests to: David L. DeMets, PhD, Director, Biostatistics Center, Department of Statistics, University of Wisconsin Medical School, 5th & 6th Floors, Medical Sciences Center, 1300 University Avenue, Madison, WI 53706. Recetved March 28, 1990; revised October 12, 1990. 0 1991 Elsevier SciencePuhhahmg Co.. Inc.
1047.2797/91/$3.50
396
AEP Vol. 1, No. 5 AugustI991: 395-405
Cairns et al. EARLY TERMINATION OF ASPIRIN IN PHS
blind, placebo-controlled a 2 X 2 factorial design. taken
every other
placebo.
day would decrease
The other question,
on alternate their
outcomes,
As stated
and their
no previous effects
on total
confirmed
stroke.
at this dose.
aspirin
tative
mortality
aim of the aspirin
to determine
high compliance
whether
including
could
therefore,
total
all eligible
tolerate
mortality
study population.
this
select
mortality
a sample
mortality
aspirin
mortality
estimate
1983 and, design,
experience
DMB
consisted
physicians
aspirin
arm and
of clinicians,
to the DMB was to examine
effects
requiring
alteration
the DMB adopted recognized
that
monitoring
the key secondary
interim
end points
policies
that served as guidelines
analyses
and consideration
using a conventional
analyses corresponds significance
to guard against this requires a standardized of any
primary
outcome
CV
lower mortality in 1982 and
the 2 x The
2 factorial
follow-up
the course
was
of the trial. The
of beneficial
primary
or adverse
of one or both parts of the trial. every 6 months.
.05) at each analysis would lead to false-positive of that significance level (6). For these reasons, boundary procedure
assumption
population
statisticians.
the data for early evidence
approximately
interim
into
during and
have
and 75% thereafter.
to the placebo. results
or early termination
to meet
frequent
11,034
subjects/ significance
arbitrary
general
year,
were randomized
epidemiologists,
charge
DMB was scheduled
of the
to 7.5 years based on anticipated
scheduled to end in late 1990. A DMB was appointed to review The
from a largely 25%
and as the
for 5.0 years would
aged 40 to 84 years, were screened
22,071
to the
of feasibility
with a one-tailed
50% in the second
male physicians,
ultimately,
11,037
to
trials (5)) the quantiby about 20% and,
of 4.5 deaths/l000
followed
was derived
would
period was revised
260,000
were
well enough
For reasons
in CV mortality
in the first year after entry,
Over
of side effects physicians
no prior MI, were selected
estimate
size of 22,000
a 20% reduction
population
Later, the follow-up rates,
10% was proposed.
40 to 84 years old with
Based on an initial
power of .95 to detect that
by about
groups,
level of .05. This
the
MI and
in a trial of long duration.
US male physicians,
y in the placebo
men with
nonfatal
the frequency
Based on previously reported results in secondary prevention hypothesis that aspirin would reduce total CV mortality
cost,
is to assess the healthy
The first is to determine
was used in which subjects
design
in a single study.
component
aim is to determine
period
a factorial
of apparently
aims are stated.
taken
that the two diseases,
related,
and on CV morbidity
The other secondary A run-in
Given
to two hypotheses
in a population
MI. Two groups of secondary
of aspirin ensure
the primary
on total CV mortality
of aspirin
of cancer.
the answers
with that seen with
50 mg of beta carotene
are not appreciably
to obtain
in the protocol,
as compared
is whether
the incidence
treatments
opportunity
effect of aspirin
CV mortality,
still under study,
days would decrease
offers a unique
given
primary prevention trial that addresses two questions using The question of interest here is whether 325 mg of aspirin
The
Early in the follow-up, for the trial. The DMB of both
the primary
level of significance
(e.g.,
and P =
results with a frequency far in excess the DMB adopted the Haybittle-Peto
excess likelihood of false-positive results (7). This or normal t statistic of 3.0 or greater during interim before
consideration
of early
termination.
This
before claiming statistical to a nominal P value of .0013 ( one-tailed) at an intermediate observation. With this conservative approach, signifi-
cance levels (e.g., P < .05) only slightly below the conventional level can still be used at the final scheduled analyses. The DMB also recognized, as had others (8-10)) that factors bearing on the decision to terminate any study were complex and that simple rules or boundaries should not be used alone but should be incorporated with
AEP Vol. I, No. 5 August 1991: 395-405
EARLY TERMINATION
TABLE 1 Date’
Cairns et al. OF ASPIRIN IN PHS
397
Confirmed deaths Aspirin
Outcome
Placebo
RR
P
6186
cv
28
33
.a3
.52
1187
Total cv
58 37
75 42
.76 ,116
.I4 .57
12187
Total CV
9I 44
102 44
.88 .99
.43 .99
I15
.95
.74
llil
Total
other information of evidence.
into guidelines.
The final decision would be based on the totality
INTERIM RESULTS AND DELIBERATIONS As described by the Coronary Drug Project Research Group (8), a number of factors must be considered in reviewing interim data before making any decision for early termination. In the PHS, the DMB examined the issues of: (1) initial comparability of treatment groups with respect to known risk factors; (2) compliance with treatment; (3) completeness of follow-up; (4) confirmation of the specified primary study outcomes; (5) secondary study outcomes; and (6) data on known complicating effects of aspirin use, chiefly bleeding phenomena. Randomization produced highly comparable treatment groups so that any outcome differences might be attributed to active treatment and not any imbalance in risk. Compliance with assigned treatment, either aspirin or placebo, as measured by proportion of assigned treatments taken was high throughout and was over 85% in each group at the time of termination. Use of other concomitant therapies was evenly distributed as well. Follow-up was 100% complete with respect to mortality and 99.7% for morbidity. Confirmation of all outcomes was sought by an end point committee blinded as to treatment assignment and independent of the DMB. Summary results for the last three interim analyses before termination are provided in Tables 1 through 3, which show total and CV mortality, confirmed nonfatal and total MI, and confirmed ischemic and hemorrhagic stroke. The data in these tables were given special attention by the DMB relative to possible changes in conduct of the study. Bleeding problems, including bruising, hematemesis, melena, other gastrointestinal hemorrhage, and epistaxis, were all highly significantly increased in the aspirin group. The absolute rates of these problems, however, were lower than reported in previous trials, presumably reflecting the low dose of aspirin in this trial and the selection at enrollment, including the run-in period. Gastrointestinal ulcers and other gasrointestinal symptoms were also increased in the aspirin groups, though, for most of these morbidity categories, the differences between aspirin and placebo were not statistically significant. The DMB did not consider any of these findings sufficient reason to change the conduct of the study. Between July 1986 and December 1987, the DMB held three formal meetings and numerous informal discussions centered around five Issues: 1.
The CV mortality
rate and time trend.
398
Cairns et al. EARLY TERMINATION
TABLE
2
Confirmed
myocardial
Outcome
Date”
infarctions Aspirin
10185 7186 1187 12187
46 56 7i 75 85 89 9Y 104
2.
The CV mortality The statistical
CV mortality
study years already The
5.
The stroke
available process
rate and conditional
MI rate difference
the investigators
to terminate
CV, showed
between
rate difference
to the DMB.
As is evident
between
.47
.0003 -
.5Y .61 .61 .60 .56 .56 .53
aspirin
of the trial and length
P
.OOlY ,003 .0007 .0004 < .OOOl
