Annex F
Faxed statement from Andy Millar re unblinding
of Zacutex study 215 7 May 1998
There is an aspect of my conduct which I always
knew would come into question when on 11 March I went to Kleinwort
Benson with Perpetual to raise the questions about the company
which today still remain unanswered. As I stated to the Sunday
Business on 15 March "Events are going to take place that
will enable everyone to judge appropriately my action. I have
acted in accordance with my (judgement and) conscience over something
I find difficult and I hope I have done the right thing."
It seems time to give my detailed account of the unblinding of
study 215 (US Zacutex pancreatitis), so this can be judged. (These
events are partly written from memory as I do not have access
to all company records at home)
In November 1996 data from study 214 indicated
that Zacutex was effective in pancreatitis, but did not prove
it: While the main analysis was in favour of Zacutex it was not
statistically significant and we had "data-dredged".
This is when unplanned analyses are undertaken. It is well known
that you can burn your fingers. In 214, however, the effects were
so consistent across the different analyses as to be convincing.
However, I realised that study 215 was collecting data from which
I could assess the risk of "burning our fingers". This
seemed important. I decided to unblind a small but vital part
of the data-mortality and considered the following general factors:
Mortality is accepted as the "hardest"
or ultimate measure for trials. It cannot be subjectively altered
(ie death is not defined as "severe" or "mild")
and date of death is an unalterable record (except in rare ITU
circumstances). If all patients are analysed irrespective of treatment
received or protocol violation ("intent to treat") then
the analysis cannot be fudged (by excluding patients because they
"turned out not to have pancreatitis" or "died
from falling out of bed"!). Some non-double blind mortality
studies with intent to treat analyses are therefore entirely acceptable.
Looking at the mortality data cannot possibly invalidate the intent
to treat analysis except in one circumstance which was never intended
nor arose. Namely that the study is stopped early, if the results
look good. (This would be as obviously unfair as stopping a 5,000m
race at 3,000m because your brother is in the lead and declaring
him the winner!).
I considered the following specific factors:
The original analysis for study 215 involved
a composite of organ failure and mortality of which the latter
was the minor part. I would only see total mortality, but nothing
else such as individual characteristics and so would not be in
a position to bias any subsequent categorisation or evaluability
rating. I kept all this data completely secret from all company
personnel and doctors involved in the study. In my view even the
most stringent critics of trial methodology would accept that
the double blind was retained almost entirely intact. Even if
they considered it was broken, intent to treat mortality data
is virtually free of bias and is not harmed.
On 27 November 1996, from a very incomplete
dataset, I noticed that the overall mortality rate was rather
low and there were 4, 2 and 1 deaths respectively on Zacutex 100
mg, 10 mg and placebo respectively. These numbers were much too
small to draw any firm conclusions but, were inconsistent with
the treatment effect in 214 and actually raised a possible toxicity
question. However Zacutex had been non-toxic in all previous studies.
I informed my supervisor Peter Lewis. Because of the toxicity
question, I kept the situation under review and acquired more
data. Over the next few months the figures changed to 7, 12 and
5 on 100 mg, 10 mg and placebo respectively. This was inconsistent
either with toxicity or efficacy. I stopped reviewing the situation
but continued to express my doubts about the total project to
the company. I notified Keith McCullagh in writing of the total
situation including all the data on 15 May because there seemed
to be some misunderstanding between Peter Lewis and himself on
the exact situation and its significance. The differences with
study 214 led me to consider the possibility that US ITU treatment
may result in lower mortality and abolish the treatment effect
of lexipafant. (UK doctors will be relieved to know that this
has been shown not be the case!).
We had a planned data review committee, agreed
in advance with the FDA (showing, by the way, that data review
on an ongoing clinical trial does not completely invalidate it
as I hear has been claimed). I had a problem in that the committee
might stop the study in the absence of efficacy which would produce
the indeterminate position of one positive data set (214) and
one equal sized negative dataset (215). This dilemma would only
be solved by a further, probably larger study. I preferred to
ensure that 215 continued and became dominant. I therefore decided
to give the committee the data as treatments A, B and C rather
than individual dose groups.
The data review committee sat and saw detailed
data from about 300 patients of actual diagnoses, patient details,
organ failure, complication rates, adverse effects, mortality
and cause of mortality (I only saw raw mortality). From 214 we
had realised that mortality was the right end point and the committee
agreed that we make the study an out and out mortality study.
Which we did. It became a 1,500 patient intent to treat mortality
study. I involved Pam Kirby and Keith McCullagh (the only two
remaining executive directors) in this decision and appraised
them of the exact details and pessimistic nature of the mortality
data to hand. When we agreed this with the FDA, I recall, but
do not have a record to hand, that they recommended we undertake
a further data review at an appropriate stage.
The study continued. In September Keith McCullagh
asked me to actually increase the rate of spending on study 215,
which was already well over budget. I simply could not understand
this, however, in the event that the study may have turned around,
I decided to avoid a fight if possible and looked again at the
raw mortality data in exactly the same way as before. There was
no difference between 100 mg lexipafant and placebo and the death
rate on 10 mg was slightly but not significantly higher. Statistical
analysis convinced me that the changes of efficacy or a positive
study were extremely remote. I informed various executive directors
and recommended a further data review committee. This recommendation
was made with increasing desperation but refused up until my suspension
in March.
This put me in a difficult position with regard
to 214 and the outside world. I did not want to stop 215 without
a formal data review, because the data I had seen was very incomplete
and there may be an explanation other than lack of efficacy which
a data review committee could discover. I required a data review
before any statement or action. However, I now find myself in
the position of having to act despite my best efforts to sort
this out behind closed doors. I believe my actions would never
have been called into question if the company had not become so
dysfunctional. You will have to judge yourselves.
| 
