Examination of Witnesses (Questions 429
- 439)
WEDNESDAY 24 OCTOBER 2007
PROFESSOR JOHN
BOURNE CBE, PROFESSOR
PROFESSOR CHRISTL
DONNELLY AND
PROFESSOR ROSIE
WOODROFFE
Q429 Chairman: I do not know who
wants to start but it might be quite interesting to take the views
of the former ISG members about what has been said. This may not
be peer review in the accepted sense but it is the best that we
can manage.
Professor Bourne: Would it be
helpful if I put some structure onto what has gone before because
I can understand why you are confused?
Q430 Chairman: Yes. Structure is
always helpful.
Professor Bourne: First of all
let me make the point that the comment was made that we did an
experiment. That is not true. What we did was carry out a trial
of future policy options, one of which was reactive culling, the
other was proactive culling, with an expectancy at the time that
these would be rolled into the field. Government subsequently
determined that this was not going to happen. They eliminated
the Wildlife Unit and if there was going to be any culling in
future farmers had to do it. However, this was not just an experiment.
The whole trial was designed by us to provide a range of epidemiological
evidence and, my goodness, it is a good job we did it that way
because that epidemiological evidence is so critically important.
That is point number one. Point number two is that it is suggested
that the evidence is confusing. I do not think it is confusing
at all. I think we have shown very clearly that localised culling
makes the situation worse and we provide the very plausible hypothesis
why that happens, namely, perturbation, but since reactive culling
has finished the negative effect of reactive culling has gone.
I think that is quite an interesting observation. If you stop
culling, stop perturbation; the negative effect of reactive culling
goes. We have shown that reactive culling as carried out makes
the situation worse, and we associate that with greater perturbation,
but all the issues around that lead to an increased incidence
in cattle and we are suggesting there is a link between the two.
With proactive culling, which in our hands led to something like
an 80% culling efficiency which resulted at the end of the day
in something like 70 to 75% removal of badgers, and a large proportion
of those that were left came in from the outside; they were immigrant
badgers, we did find overall in 100 square kilometres, the area
mentioned in the King report, something like a 23% reduction.
There is no question that badgers contribute to cattle TB. In
the very centre of the area it was higher than that. The modelling
suggested overall that the impact badgers were making on the cattle
TB problem in the areas in which we were working was about 30%,
30%, 40%, it was about that. The rest was down to cattle. We put
forward ways in which we believed the cattle element could be
tackled, and Sir David has accepted all of that; not in his report
but certainly in media reports he accepted all of that. The issue
comes down to how do you deal with the badger component. There
is no doubt that culling has an impact. The problem is that although
you have a positive impact in culled areas it is this perturbation
issue that creates a problem. This, I think, is where it would
be useful to spend a few moments discussing this. I am aware that
in addition to the King report Defra sought in-house reports from
the Central Science Laboratory (CSL), the Central Veterinary Laboratory
(CVL) and Natural England. I have not seen these but I know they
were commissioned, and within Natural England and CSL there are
high-class ecologists, but I am sure you would find these reports
of value too. I suggest you ask for these and study them; they
do provide you with further information which I believe could
be helpful to you. What we have shown is that perturbation is
an issue. I am bound to say that when we started this work I certainly
did not believe that perturbation was a problem. We knew about
it but I did not think it would be an issue, but the data we have
suggest it is an issue. In the King report I think they have underplayed
perturbation. I will ask Rosie to talk about this in a moment.
It would appear that they have not read all the literature. There
are a number of papers that have been published on this which
they do not appear to have read and I think they are overestimating
the impact of geographical boundaries. They are making a number
of statements on the value of one kilometre of arable land, for
which there is absolutely no data whatsoever, and they are really
saying, "There is a problem there but we can overcome it
in some way by believing that we can". With respect to the
size of the areas, we did extrapolate to 300 square kilometres
or more. Sir David said we found significant impact of badger
culling by that extrapolation. That is not true. We reported a
modest difference between the positivity and the negativity. We
then questioned the economics of the thing. You asked me earlier
about comments that Lord Rooker had made and I think he criticised
us with respect to the economic analysis. If you look at our first
report, when it was suggested that a cost analysis be done, it
was Lord Rooker who asked us to put an economist on the committee,
so it was quite clear that we were going to provide some economic
information, which we did. We did not do a cost benefit analysis;
there were no benefits, and it is fully explained in the report.
The other thing we found was that if one were to get the same
effect that we did culling had to be done over the piece simultaneously,
it had to be done in a co-ordinated manner and it had to be carried
out for a very long period of time. In other words, it was not
a trivial thing to do. We discussed this at length with respect
to licensed culling that Defra firmly put on the agenda in our
final report, and our advice was that it was totally inappropriate
to expect farmers to do this. It could only be done using expert
fieldsmen who would require an immense organisation and logistical
activity to do it. Perhaps Rosie could pick it up from there and
discuss the perturbation issue.
Professor Woodroffe: Certainly.
Q431 Chairman: Just before you do,
and I am sorry if this is going to be a little bit tedious but
it is very central to the question of this perturbation effect,
going back to what Sir David's report said, in paragraph 27 it
says, "The results are hard to interpret but it would appear
that there is some evidence for an increase in cattle TB outside
the removal area". They then go on to discuss the two-kilometre
zone and its statistical impact, and they then say, "The
increase was not statistically significant", so that is an
interpretive point on the data that you had, which I would be
grateful for your comment on. They then go on to make a third
point, that there was an apparent small beneficial effect in the
area up to half a kilometre from the removal area, so they are
saying that things get better just across the boundary, which
is an interesting observation. Then later on, after some more
commentary, they say that perhaps perturbation—and I hope
I have not misunderstood paragraph 31—"indicates that
the detrimental effect seen outside the removal area may well
be transient". In other words, you saw something happening
but it goes away after a while. I guess that must refer to their
subsequent observations about the comings and goings of badgers,
but then they say in paragraph 37, "However, there is a reasonable
possibility that the disruption is transient. The data do not
discount this theory. If the disruption is transient the single
regression model for the effect of perturbation used by the ISG
is not adequate", so that is, if you like, an attack on your
methodology. Perhaps you could, bearing in mind we are not statisticians
and not experts, help us to address that critique of your information.
Professor Woodroffe: I think you
have absolutely put your finger on a very important point. Whilst
Sir David King and Professor Woolhouse have repeatedly reiterated
that they think there is almost no disagreement between the ISG
and themselves, you have put your finger on the key area of disagreement,
and I think that the reason you were seeing shaking heads behind
them is that this particular issue of detrimental effects which
they considered unconvincing and hard to interpret we considered
very strong. Perhaps Professor Donnelly would be best placed to
address the statistical issues and then I can come back to the
ecological issues.
Professor Donnelly: As a statistician
I feel I should declare one interest, which is that I am going
to use "significant" in its statistical sense, which
is that if it is a significant effect the 95% confidence interval
does not include zero. I think that is a bit different from the
way John was just referring to it when he was saying, "David
King said it was significant, I say it is modest".
Q432 Chairman: You talk about margins
for errors. We watch opinion polls. Those are the crude numbers
we understand.
Professor Donnelly: There are
a few things that are going on. One is this issue of between culls
one and two, and so let us address that first, and then we need
to talk about what we are using to test the detrimental effect.
Between culls one and two the breakdowns that will be detected
will include breakdowns that resulted from infection events that
happened before the first cull and infection events that happened
between those two culls, because testing was every year and so
it includes over that one-year time period. Keep in mind that
we are always comparing like with like in terms of proactive areas
with survey-only areas, and so we are comparing those same time
periods. What will happen is that in that first year there will
be a bias in the estimate towards no effect because some of what
we are observing will be from the no effect period before any
culling took place. Therefore, rather than there being any invalid
period when we might get something spurious, we are actually getting
an estimate that is biased towards the null because some of the
differences will not be detected until the period between the
second and third culls.
Professor Woodroffe: So it is
conservative.
Professor Donnelly: Yes. We did
not place any more emphasis on that particular usually year period
than any other year period, but we considered them all in sequence.
When we did that we found a non-significant trend towards a decreasing
detrimental effect outside. We did address that. We discussed
it both in the report and also in the IJID paper which David King
cited. However, one of the things that we want to consider is
that when doing sub-group analyses you have less precision, and
I draw people's attention, if you have the final report, to page
99, which shows estimates stratified by which culls they were
between, and I have the IJID paper which has the same graph as
figure 2 right in the middle, so I will distribute it. It is either
page 99 of the final ISG report or figure 2, which is right in
the middle of the paper I am handing round. These figures show
the variation in our estimated effects—
Q433 Mr Drew: Are you looking at
table 2 or figure 2?
Professor Donnelly: Figure 2.
Q434 Chairman: Figure 2, page 304,
top right hand corner.[6]
Just take us through the construction of the picture that we are
going to see before you explain it to us because some of my colleagues—and
this is not attacking their ability to understand—may not
quite be keeping up, and the audience behind us are not where
we are, so you will just have to bear with us whilst we have this
explained to us, and if anybody does not understand perhaps they
would indicate and ask as we go along. We will do our best to
understand.
Professor Donnelly: The top figure
shows divided-up estimates for inside the targeted culling area
and outside the culling area, and so the estimates are divided
up in the top figure by cull sequence between culls one and two,
two and three, three and four and so on; four and more get lumped
together. What you see is that consistently in all four of these
time periods outside trial areas in this two-kilometre zone there
is a negative effect of culling which is detrimental above the
line, and below the line it is a beneficial effect of culling.
The beneficial effect of culling appears, based on the point estimates,
to be greater after more culls. I wanted to point this out because
there was this point that a simple regression is not an adequate
way of looking at this. We very specifically presented this data
so that people could see exactly what the estimates were. A regression
will assume, say, that there is a linear effect, so you can put
a straight line through. Here we are not assuming that. There
is a "connect-the-dots" between them so that you can
connect them up, but there is no assumption of linearity there;
it is purely the data, so you can see wobbles and variations cull
to cull. In each of these there is a non-significant but general
trend towards a decreasing detrimental effect with more culls
outside and increasing benefits, so both things move towards a
better situation after more culls.
Q435 Mr Gray: So what you are saying
is, the more badgers you kill the less TB there is?
Professor Donnelly: After you
have culled for a longer time.
Mr Gray: All this thing goes all the
way across this and I cannot understand what you are talking about.
Chairman: Do not attack the messenger.
She is doing her best to explain it in her language to us as lay
people.
Mr Gray: I do not think so.
Chairman: You have drawn a conclusion
now, so you want to test out whether you have got the right message.
Q436 Mr Gray: I am trying to understand
what she is saying. Would you mind expressing it in language that
a thick fellow like me can understand? I am sorry about that.
Professor Donnelly: More annual
culls, better results.
Chairman: Now you have got it. He is
in line for a degree already.
Q437 Lynne Jones: You said non-significant.
Professor Donnelly: I was talking
about the point estimates. We looked at what the net effects were,
and, balancing these two different areas out, pluses and minuses,
the net effect between the first and second culls is consistently
detrimental overall, looking at those two counteracting forces,
but in all the ways that we looked at it with different sorts
of linear effects in different databases we get a net beneficial
effect after four culls. That still means that that is a net beneficial
effect judged over the whole area, and of course that still involves
some areas experiencing a detrimental effect and some areas experience
a beneficial effect. The next thing I want to point out is on
the graph below that, where this is all time periods taken together
but we have now sub-divided by how far from the boundary, and
this is probably more difficult to see. Think of the line that
goes vertically in the middle of the graph as the boundary between
the targeted area for culling and the untargeted area. As you
move to the right you are getting deeper and deeper into a trial
area and you will see that the estimates move towards a more beneficial
effect. That was a non-significant trend but you do see quite
a lot of consistency there. That is where David King was quoting
about 40% benefit by the time you get deeply inside these trial
areas. Outside the area targeted for culling there are four estimates
dividing up the zone up to two kilometres out. You see an estimate
of beneficial and then three detrimental effects. In that first
zone between nought to 0.5 kilometres outside the culling area
some of that area was actually culled, and the reason for that
was that the trial area boundaries were drawn up and then after
that areas were targeted for culling based on Rosie looking at
maps and approving boundaries so that whole social group territories
were targeted, and Rosie can talk about why that is. Some of that
area between nought to 0.5 actually received culling and it is
quite plausible that that is why there was a beneficial effect
there, but beyond that you see three detrimental effects.
Q438 Chairman: Whoa, whoa. In the
report which Sir David King did he talked about, if you like,
doing some work in the margin, in the area which would be the
perturbed area. Is that what you are effectively saying you subsequently
did, which was to go back and when you have got the extent of
the social groups of the badgers do a bit more culling but outside
the boundary?
Professor Donnelly: It was not
going back. It was done from the outset.
Professor Woodroffe: What we did
was, we started out with trial areas which we defined on the basis
of property boundaries. The entire area plus land up to two kilometres—usually
not as far as two kilometres but a maximum of two kilometres—outside
was then surveyed for badgers. For all of the areas before they
were allocated to treatments they were surveyed in that way, and
then we went and inferred from the field data where we thought
the boundaries of the social group territories fell, and we did
that because we were concerned about perturbation and we thought
we could minimise perturbation by trying to remove complete social
groups, so we wanted to make sure that every property that fell
inside the trial area was going to get the full culling treatment.
We did not want incomplete culling on the outer area occurring
because perhaps the main setts of the badgers that lived on this
farm were just outside.
Q439 Chairman: So basically you refined
the cull areas to be as complete as possible in terms of the social
areas for the badger setts you had identified, and therefore in
any commentary which subsequently comes about, if you like, the
statistics of perturbation take that refinement of the boundaries
into account?
Professor Woodroffe: Yes. The
culling areas were slightly bigger than the trial areas and that
is made very clear in all sections of the report. That is probably
why you see that potentially beneficial effect just outside the
trial area boundary, because a lot of those farms were receiving
culling even though they were not inside the trial area.
Professor Donnelly: Could I make
one more comment to discuss these graphs? The boundary you see
on each one of these points, to give you an indication of uncertainty,
are 95% confidence intervals, but we had a concern in terms of
the interpretation of these 95% confidence intervals in the David
King report looking and saying, "Okay, outside trial areas
three of the four include zero; therefore this does not look like
such a secure result". If you look on the right hand side,
all five of those include zero and that does not mean that the
beneficial effect inside is insecure. Those confidence intervals
are in a sense your level of uncertainty if you are just looking
at one of these points. What we were doing here was showing how
well each one of those points was estimated individually to justify
where we thought we could get benefit by fitting a linear regression,
and that is what we did for the cull number effect for the figure
above. For the figure below we fit a linear effect looking at
distance from the border inside trial areas. Outside trial areas
there was no evidence, looking at this, of any linear trend.
6 Donnelly, C et al (2007). "Impacts
of widespread badger culling on cattle tuberculosis: concluding
analyses from a large-scale field trial", International
Journal of Infectious Diseases, Volume 11, Issue 4, July 2007,
p 300-308. Back
|
