AZT, Early Intervention, and the Concorde Controversy
The widespread overreaction to the April 2 report ofpreliminary results from the UK-Irish-French Concorde trial
of AZT vs. placebo for early HIV treatment has confused and
upset patients and physicians. Seldom if ever have U.S. and
British mainstream AIDS researchers been as far apart as they
are over the Concorde study. And seldom has information
affecting patient care been so widely communicated to the
public with so little regard about how it is received.
For patients and physicians now, the bottom line is:
* For asymptomatic patients, Concorde found no benefit from
early use of AZT compared to starting AZT later. But this
conclusion contradicts that of all three other major
controlled trials of AZT treatment of asymptomatics (1,2,3).
For this reason, and because only preliminary and sketchy
results are available (eight paragraphs and one table), and
because the Concorde study found no reason not to use AZT,
U.S. physicians at least are not changing their treatment
practices until more information is available. The U.S.
National Institute of Allergy and Infectious Diseases (NIAID)
said, in an update published April 1, "At this time we see no
basis for changing the current recommendation to initiate
antiretroviral therapy for HIV-infected persons whose CD4+ T
cell count [T-helper count] falls below 500 per cubic
millimeter (mm3) of blood."
Even in the UK, where unpublished information about the
Concorde results has been quietly shared with physicians and
patients (information not available to U.S. researchers), and
where the Concorde results are widely accepted as correct,
there seems to be little rush to change medical practice at
this time. (In part this is due to the fact that early use of
AZT has been rare in the UK, compared to the U.S. and France;
the Concorde results have only reinforced what has already
been the prevailing medical strategy.)
* The Concorde trial only studied AZT for early intervention,
in asymptomatic patients. No one has suggested that these
results should affect the treatment of persons with AIDS or
with other symptomatic HIV disease.
* Concorde only studied the use of AZT as a single-drug
treatment. The benefits of AZT were already widely believed
to be of limited duration. Concorde has no bearing on
combination antiviral treatment, which is likely to have more
lasting benefit than AZT alone.
* Recent press coverage of the Concorde announcement has been
misleading. In part this is due to the April 2 news release
of the British Medical Research Council, which carried the
story to the world, but did not even summarize the eight
available paragraphs correctly. Ultimately, the garbled
message was due to the time pressures of a race against news
leaks, international failures of cooperation which prevented
consultation, and public-relations mistakes by Wellcome, the
British developer of AZT.
* The conclusion that Concorde shows that T-helper count does
not work as a marker for testing new antiviral drugs is, we
believe, open to serious challenge (see below). In any case,
nothing in the trial discourages use of the T-helper count
for monitoring individual patients.
Hopefully, a more detailed report of the Concorde results
will be available by the time of the International AIDS
Conference in Berlin, June 7-11. Concorde may well change
medical practice, after enough information is released to
allow the worlds' scientists and physicians to evaluate it.
Meanwhile, we hope that Concorde serves as a wake-up call to
encourage more balanced AIDS treatment research strategies,
instead of continuing to rely excessively on a single drug or
class of drugs.
The rest of this article looks in more detail at:
* The Concorde trial itself;
* The preliminary results, and how the news was released;
* T-helper count for testing antivirals -- and why we believe
the Concorde conclusions may be a statistical artifact.
The Concorde Trial
Concorde, the largest and longest-running controlled trial
comparing AZT to a placebo, enrolled over 1700 patients
between October 1988 and October 1991, in 65 medical centers
in the UK, Ireland, and France. It recruited asymptomatic
HIV-positive patients, regardless of T-helper cell counts.
The average time on the study has been about three years. The
current information release resulted from a planned analysis
of the data through December 1992. Followup data is still
being collected on the patients.
The study itself must be distinguished from the issues
concerning its recent news release. Concorde asked the right
question for 1988, and it was well designed to answer that
question. It appears to have been conducted in an ethical
way, with the high survival rate in both arms of the study
suggesting that the volunteers received good medical care.
The Concorde trial was co-sponsored by the British Medical
Research Council (MRC) and the French National AIDS Research
Agency. Decisions were made by the Concorde Co-ordinating
Committee. Most of the data analysis was done in the UK.
The trial was originally designed to find out whether it is
better to begin AZT treatment immediately for symptom-free
HIV-positive persons, or to wait until HIV-related symptoms
develop and begin treatment then. Patients were randomly
assigned to receive AZT immediately (250 mg four times a
day), or to receive a placebo until symptoms developed, at
which time they were switched to AZT.
But one year after the trial began, in October 1989, the
Concorde study was modified for ethical reasons. A major U.S.
trial (ACTG 019), which also compared AZT to placebo in
asymptomatic patients, was terminated for patients with T-
helper counts under 500, after finding that those given
placebo were about twice as likely to progress to AIDS as
those given AZT.
As a result, Concorde was modified to allow patients with T-
helper counts of 500 or fewer, as shown by at least two
different blood tests at least one month apart, to switch to
AZT if they wanted to. Under this rule, 282 of the 872
patients in the deferred treatment group (the group starting
with the placebo) eventually switched to AZT "before
developing symptomatic disease." But under the statistical
rule by which the study was analyzed (called "intent to
treat"), these patients were still counted with those taking
the placebo, no matter how long they had in fact been taking
AZT.
Many experts believe that this October 1989 modification
changed the nature of what the study was measuring. Concorde
started as a comparison between giving AZT to every
asymptomatic person with HIV (no matter what their T-helper
count), vs. waiting and giving AZT only when HIV-related
symptoms developed; we could call that study Concorde I. It
changed to comparing AZT for everyone, vs. something closer
to a widespread de facto standard of care, which is to
evaluate patients individually when their T-helper count
drops below 500, and decide on a case by case basis when to
start AZT. We could call this second study Concorde II. Since
this change occurred only one year into a three-year
recruiting period, it is possible that most of the study
volunteers were never in Concorde I at all -- and it is
certain that the great bulk of patient years in the study
occurred in Concorde II. Therefore, instead of saying that
the Concorde trial found no benefit of using AZT in
"asymptomatic" patients -- the message given to the world's
press -- it would be more accurate to say that it found no
benefit (during an average followup of three years) in using
the drug in all asymptomatic patients, vs. using it
selectively in those asymptomatic patients whose T-helper
counts drop below 500.
A practical difference in these two statements is that the
latter does not imply that asymptomatic patients now taking
AZT in accordance with U.S.-standard medical practice should
stop. But the message that was in the press does. It may
needlessly interfere with medical care, and make it more
difficult to convince persons at risk of HIV that they should
be tested and receive treatment if necessary.
The April 2 Report
The only Concorde results generally released so far are the
eight-paragraph letter published in the April 2 Lancet(4),
an April 2 news release from the MRC, and quotes from an
April 2 MRC news conference. The letter and other materials
were not peer reviewed by The Lancet.
The letter reported:
* That there were no significant differences between the
immediate-treatment (AZT) group and the deferred-treatment
group, either in survival or in progression to AIDS. The
three-year survival rates were fairly good for both: 92
percent for the group that started AZT immediately, 93
percent for those that started on placebo (and often switched
to AZT later). The rate of progression to AIDS or death was
18 percent in both groups. The rate of progression "to
'minor' ARC, AIDS, or death" was 29 percent in the immediate-
treatment group, and 32 percent among those who started on
the placebo. None of these differences is statistically
significant.
* That comparing the groups was valid, despite the change in
the study which allowed 282 asymptomatic patients to switch
from placebo to AZT if their T-helper counts went under 500,
because of the very different amounts of AZT use in the two
treatment arms.
* That the toxicity of AZT in this study was relatively low
-- despite the high dose of the drug used, and the long time
that patients were taking it, compared to other studies.
There were no new or unexpected toxicities.
* That there was a significant difference in T-helper counts
between the groups. Those who began AZT immediately had
gained a median of 20 cells by three months. Those who began
with placebo had lost a median of ten by that time. This
difference of about 30 in T-helper counts persisted; at six
and at 12 months it was slightly larger than 30, and it was
observed for up to three years.
* That the difference in T-helper counts, without a
difference in disease progression or survival, casts doubt on
the value of T-helper counts to measure whether an antiviral
is working.
The final paragraph of the Lancet letter states the main
conclusions:
"In conclusion, Concorde has not shown any significant
benefit from the immediate use of zidovudine compared with
deferred therapy in symptom-free individuals in terms of
survival or disease progression, irrespective of their
initial CD4 count. The discrepancy between this result and
the significant effect of immediate zidovudine on CD4 cell
counts casts doubt on the value of using changes over time in
CD4 count as a predictive measure for effects of antiviral
therapy on disease progression and survival. Concorde
provides a valid comparison of the strategies of immediate
versus deferred treatment with zidovudine since there was a
striking difference between the two groups in the amount of
zidovudine taken on study before progression to ARC or AIDS.
A more detailed analysis is being prepared and follow-up
continues in all patients."
The most widely distributed message was not the letter but
the April 2 news release from the MRC Press Office. This
release, written by press experts not AIDS experts, changed
the letter and made the results appear more definitive. For
example:
* The opening sentence of the news release reads, "The large
Anglo-French Concorde randomized trial of zidovudine in
asymptomatic HIV-infected individuals shows that there is no
significant clinical benefit in terms of survival or disease
progression to AIDS or AIDS-related complex (ARC) in those
who started zidovudine immediately rather than those who
waited for the onset of symptomatic disease." Note that the
Lancet letter only claims that "Concorde has not shown any
significant benefit," while the news release claims it does
show there is none.
* The news release clearly implies that it is just as good to
wait for symptomatic disease as to start AZT early. But the
letter only says that Concorde "was designed in 1988" to
determine that; as we have seen, what the study was designed
to do in 1988 and what it did do are not be the same.
* In reporting the amount of time those assigned to placebo
were actually on AZT instead due to the October 1989 change
in the study, the news release only mentioned a figure of 7
percent at 18 months; it gave no hint that the Lancet letter
also included a figure of 14 percent at three years. Also,
the news release cited "the large number of patients in the
study" as part of the reason a valid comparison was still
possible; the letter, written for a professional instead of
lay audience, does not suggest that a large number of
patients overcomes a selection bias, which is the issue here.
* And where the letter "casts doubt" on using T-helper counts
to measure the effectiveness of an antiviral, the news
release "casts serious doubt."
The news release, in short, pushes beyond the letter in
discouraging early treatment with AZT.
Top U.S. government AIDS experts had only two days' notice
before the public release of the Concorde results. U.S.
experts have questioned why the release could not have waited
a short time (as there was no emergency in terms of patient
care, because the two groups in the study did equally well)
so that it could include an executive summary to give
physicians and researchers more guidance and information.
To address these questions, AIDS TREATMENT NEWS interviewed
Tim Peto, M.D., of Oxford University. Dr. Peto is the
scientific secretary to the British Medical Research Council
AIDS Trials Committee, and also a member of the Concorde Co-
Ordinating Committee. On April 14, Dr. Peto explained the
timetable of the information release.
After the data cut-off on December 31, 1992, it took over two
months to collect the data from the 65 sites and do
preliminary checking and analysis. The data was unblinded in
mid March; the CCC got the results only about two weeks
before the Lancet letter. It had to share the results with
the physicians at the trial sites, who would tell their
patients. Clearly the story would soon reach the news media;
to prevent garbled information from going out, it was decided
that there should be a basic statement of results released to
everybody. Also, it was feared that the inside information
would be used in the financial markets if not released
quickly.
So on the morning of Thursday, April 1 (the day before the
Lancet letter was published), the trial physicians were told
the results, and on that afternoon the press was told. As for
NIAID, which first heard on Wednesday, March 31, Dr. Peto
noted that, "When we completed our analysis, we told them."
Dr. Peto also noted that the Concorde result was no great
surprise, since the trial had gone well beyond the time it
would have been stopped if a clear difference had been seen.
As far as we can determine, almost no details of the study
have left the borders of the three Concorde countries.
Therefore, AIDS experts elsewhere were only bystanders while
a strong conclusion was inserted into the world's press. They
were locked out from effective dialog due to lack of data. At
issue is the standard of care for tens of thousands of
people, current drug approvals (especially in Europe),
procedures for approving drugs in the future, questions
around the developing theories of AIDS pathogenesis, and the
kinds of drug-development strategies and trials which are
most appropriate in the current epidemic. As one researcher
put it, to come to such far-reaching conclusions likely to
affect widespread medical treatment would normally be done
with exquisite care, after opening the data to scrutiny and
professional critique, and responding to questions, concerns,
and objections. She noted "the enormity of the conclusions,
without allowing any of us to access the whole body of data."
Others have commented that the Concorde researchers are very
reputable scientists, well respected in the UK and in the
U.S., and clearly acting with good intentions in getting the
bottom-line information out quickly.
Among the information researchers most want to see now is the
standard Kaplan-Meier curves, showing when the deaths,
progressions, treatment changes, or other events occurred.
They also want a subset analysis showing how the patients who
elected to switch from placebo to AZT compared with those who
did not make that choice -- at the time they switched, and
later.
However, Dr. Peto told AIDS TREATMENT NEWS that the MRC never
plans to do this analysis. Following the "intent to treat"
philosophy, he explained that it would be invalid to analyze
subsets which were not determined at the time of
randomization. Self-selection into those groups, he said,
would cause biases which could make the analysis invalid.
We did not have time to run this statement by U.S. AIDS
researchers. We expect that some of them would point out that
self selection had already affected the study, and they
needed the analysis in order to understand what had happened
and make any possible corrections.
As for the news release, Dr. Peto told us that the MRC Press
Office wrote it, saying that they were the experts in
communication with the lay news media. "We planned to make
them mean the same," he said, speaking of the Lancet article
and the press release. "The differences were not meant to be
substantive." In the U.S., NIAID did not know about the press
release until over a week later, when AIDS TREATMENT NEWS
happened to fax them a copy.
This is not the only communication problem between U.S. and
British AIDS researchers. We have heard complaints from
others that British scientists and physicians have had
trouble obtaining executive summaries and other information
from NIAID; in one case, only a leaked copy got the
information to British physicians. We have also been told of
letters from leading British researchers which have been
repeatedly ignored by their U.S. counterparts. We could not
investigate further by press time. But these complaints, as
well as the U.S. complaints about the handling of the
Concorde news, suggest that behind the surface solidarity,
there is often a failure to communicate. Dr. Peto noted that
there is no intentional communication problem, and "When we
meet in person, everything is fine."
Conspiracy Concerns
Because of the atmosphere of confusion and lack of
information, the sudden death of one of a leading Concorde
statisticians, who was hit by a car while riding his bicycle
to work in the UK about two weeks before the news release,
led to disquiet among researchers. Hundreds of millions of
dollars have changed hands in the financial markets as a
result of the news; if there had been a conspiracy to drive
down the price of Wellcome stock, a single technician who
broke ranks and released contradictory or ambiguous
information could have deflated it. This is because AZT's
defenders were desperately looking for any scraps of data on
which to build their case, but finding none, due to the tight
control of the information released. If they had anything,
even the ambiguities which must always exist in a study of
this size, the story would have been changed from "AZT
doesn't work" to a technical dispute among experts -- of
little interest to the press.
Reports of a second death of another Concorde statistician,
this time in France, only increased the concern. But this
death may not have happened; the rumor may only reflect an
erroneous report that the UK death had occurred in France.
We are aware of no evidence of foul play, and we believe it
is very unlikely. But completeness requires reporting what
concerns have arisen and why.
A different concern is that the MRC Press Office faxed its
news release to anti-AZT activists several hours before the
financial markets had the news. One copy left the MRC Press
Office at 15:40 on April 1, London time; the routing
information added by various fax machines or computer cards
traced its path, which included arrival at AIDS TREATMENT
NEWS about an hour later. It was still morning in New York,
yet the price of Wellcome PLC American depository receipts
closed unchanged on the New York Stock Exchange that day. The
issue is not financial misuse of the information -- there is
no evidence of that -- but whether the MRC favored the anti-
AZT view -- as was already suggested by how it restated the
Lancet letter in its press release. We have been assured that
there was no favoritism, that the MRC sent the same news
release that afternoon to mainstream publications which
respect its press embargo. We believe this is true; still, it
seems unusual that news of such financial value would
circulate widely in newspaper offices without the stock
markets finding out. The next morning, Wellcome stock fell 12
percent in London on the strength of the same press release;
it partly recovered, but had lost over $500,000,000 within a
few days of the Concorde news.
T-Helper Conclusion:
A Statistical Artifact?
The Lancet letter (in the paragraph quoted above) said that
the lack of clinical benefit from early AZT, in contrast with
the substantial T-helper count difference between the two
treatment groups, cast doubt on using the T-helper count as a
"surrogate marker" (to substitute for death or disease
progression in trials) when testing new antivirals. This
issue is important, because without a surrogate marker, any
trial of a new AIDS treatment is likely to take years; since
HIV develops slowly, it would take that long to achieve a
statistically significant number of deaths or disease
progressions in the comparison group, even if the new
treatment worked perfectly. The press amplified the doubts;
for example, the one Lancet sentence became a major article
questioning surrogate markers in THE NEW YORK TIMES, April 6.
The FDA's accelerated approval system is based entirely on
surrogate markers (so far, only on the T-helper count), and
any move away from faster testing could be disastrous for
people with AIDS -- not only because of delays, but because
many drugs would not be tested at all. While researchers hope
for better surrogate markers soon, especially high-tech viral
tests, the tests themselves may need to be validated by
clinical endpoints in lengthy trials before they can be used
for drug approvals (unless the tests can use frozen blood
from patients whose future progression is known).
But does Concorde really show what it claims to show, that
there are doubts about T-helper counts? Most U.S. researchers
believe that this question must be addressed by case-by-case
analysis of whether individual patients whose T-helper counts
rose after starting the drug seemed to show improved
prognosis as a result. It is not enough to simply compare
groups medians, as the Concorde preliminary analysis did.
For different but related reasons, we believe, there is a
strong possibility that both Concorde conclusions (that AZT
does not work for early intervention, and that the T-helper
count does not work as a surrogate marker) may be statistical
artifacts -- illusions resulting from how the study was done,
not reflecting what is happening in reality. Some of the
following concerns, so far as we know, have not been raised
before.
The analysis below is complex, and readers who get lost can
skip to the summary at the end.
Concorde compared two treatment strategies for asymptomatic
HIV infection: treating everyone immediately with AZT, vs.
treating patients selectively. The selective group all
started on placebo; later, many but not all of them began AZT
for one of two very different reasons: because they reached
an endpoint as defined by the protocol, or because they
voluntarily switched to open-label AZT (only available for
those with T-helper count under 500, and only after the
protocol change of October 1989). Those switching voluntarily
were still counted with the placebo group, however, under the
"intent to treat" analysis rule, which groups patients
according to what drug they were assigned to, not what drug
they actually took. 282 of the 872 patients who started on
placebo took this option and switched to AZT voluntarily
without reaching an endpoint.
No data has been released about who decided to switch to AZT,
but we can reasonably assume that those doing poorly must
have been the most likely to do so. And one little-known fact
is that Concorde used its own definition of "asymptomatic,"
meaning that certain minor HIV-related symptoms could occur
without constituting an endpoint. Patients and physicians,
therefore, had actual symptoms potentially available, as well
as blood work, to help them decide when they needed to choose
AZT.
What Concorde showed is that a selective treatment strategy
could work as well, on the whole, as treating everyone with
too high a dose of AZT.
But notice that the mere fact that a selective strategy
worked equally well as a universal one automatically
guarantees the "discrepancy" that Concorde noted between T-
helper count and clinical benefit -- whether or not T-helper
count really works as a marker of drug effectiveness. This is
because (as has long been known) AZT treatment causes median
T-helper counts to rise, compared to where they would have
been with no treatment, and to remain higher than they would
have been for long periods. Therefore, in the group where
everyone gets the AZT, the median T-helper count will rise
above that of the other group -- while clinically, the two
groups are equivalent by definition -- creating the
"discrepancy."
Basically, the patients likely to get sick within the time-
frame of the study were likely to start AZT at the earliest
sign of problems, creating the clinical equivalence between
the selective-treatment group and the other group where
everyone started with AZT. But the more stable patients, who
were unlikely to contribute endpoints in any case whether
treated or not, got AZT in one group and placebo in the
other, creating the T-helper difference.
It gets worse. The selection bias caused by the voluntary
switches to AZT biased the clinical and the T-cell results
differently, because patients with higher T-helper counts to
begin with tend to have a higher numerical rise than those
with lower counts. Clinically, the switch to AZT brought the
two groups closer together, because those likely to develop
endpoints moved toward similar treatment. But the same
targeting meant that the patients most likely to be treated
differently were the very ones likely to develop the biggest
T-helper count differences as a result.
The Lancet letter addressed the concern about selection bias
by noting that, at three years, the group which switched
voluntarily to AZT had used AZT for only 14 percent of its
time on study. If AZT were helpful, then wouldn't the
majority of time off the drug cause greater disease
progression than in the group where everyone started with AZT
and most stayed on it? We are not convinced of that, because:
* The 14 percent of time spent on AZT was probably critical
time, since both physicians and patients were highly
motivated to notice any signs of impending trouble and start
treatment then.
* All the selective treatment needed to do was to push the
endpoints (death or disease progression) out past the
December 1992 end of the study, in order to avoid them
counting as endpoint. The three-year figure, above, of 14
percent of time on AZT implies that most of those who chose
to switch to AZT did so relatively late in the study (or the
percentage would have been higher). AZT causes an initial
boost in the T-helper count, lasting for an average of about
six months, and this boost might well have been enough to
hold off trouble, in many cases, until past the end of the
study. Even if the earlier lack of treatment had allowed more
immune damage to occur, it might not catch up with the
patient until after Concorde was over.
* Similarly, the benefits of early intervention may not show
themselves early. A difference of 30 does not matter much is
it is 500 vs. 530; but it does matter if it is 20 vs. 50. In
many if not most patients, disease progression due to lack of
treatment would not be expected to show itself until past the
December 1992 end of the Concorde study.
* The AZT dose, 250 mg four times a day, was about twice the
dose now in common use. Previous studies have suggested that
too high a dose not only increases toxicity, but actually
reduces efficacy compared to the correct dose. This may be
because side effects force more people to drop treatment
entirely -- especially those with most advanced illness who
may need treatment most -- or for other reasons. Whatever the
reason, the excessive dose would suppress the performance of
the universal-treatment group, making it easier for the
selective-treatment group to equal it, than if a better dose
had been used.
* The table published in The Lancet does show a trend toward
more early progression in the selective-treatment group
source: AIDS Treatment News
