Back to MEANS Homepage Notes on a visit to Zeneca. June 1996
Larry Furlong
The company was historically 80% toxicology and 20% research.
Statistics in the toxicology has become largely automated which
has freed up time for other areas. Now 20% toxicology, 30%
research and 50% development and production. There is an
increasing understanding in the company of what the statistician
has to offer and more work to do. LF is concerned with
identifying what in practice they can do and in selling it to
other members in the company.
The emphasis is on the value added by the statistician. They
need to continually re-evaluate the best way of adding value. We
want our client to adopt a continuous improvement culture and we
need to adopt it also.
There are many basic standard practices used in the company, t-test,
sample size , but much more important is the need for insight
into experimental design and compromise. Such questions as 'is
this assay still re-usable ñ is it good enough?' The
statistician needs to be able to work with variation, to follow
quality and have an opening questioning attitude.
Planning, design and rightly interpreting data are important,
not 'did they use the right test?' It is rare for them to carry
out tests. The statistician is looking to increase efficiency,
are they doing silly things, is the process stable enough to
compare today with a week ago. One recent development is that LF
has been introducing factorial changes into experimental design.
All the statisticians need to be comfortable with the techniques
so they can answer questions from others; they need to be able to
understand new techniques.
Recently they have developed a half-day course teaching
package to teach factorial design. (I saw this later, it is
presenter dependent with quite a lot of hands-on work by those
attending using a mixture of real data and simple data to make a
theoretical point). The interactive and diagrammatic approach
makes it accessible to those coming on the course. Feedback from
scientists has been very positive.
Understanding of variation is crucial - sources and level.
More often than occasional ñ main point is their analysis is a
small point of the overall.
Anyone coming to Zeneca needs a deep understanding of ANOVA
even if they can not do the calculations. This understanding
becomes deeper with use. The statistician needs to be comfortable
from different viewpoints. A good grasp of theory needs to be
there in the background.
In the pharmaceutical industry there is a regulatory
requirement for a statistician. The statistician has to model the
reality and may need a huge toolkit and the ability to learn as
you go. They need to know about blocking, factorial design.
Working from first principles is important. They have taken 'quality'
on board and the need to bring things out into the open.
They need to be able to deal with people. What are the people
issues? Identify resistance and what truly motivates them for
change. Understand their issues. two-way communication skills,
listening, reflective questioning, clarification. Handle people
well, recognise their skills and be able to challenge effectively.
Within the company there needs to be a better understanding
and use of data in production including SPC, but graphs are
better than words. There are reviews of historical data in
response to a particular problem, they work with specs for
regulatory authorities, some complex statistical analysis when
there is not a designed experiment and not a stable relationship.
Cusums are used in diagnosis. They continually aim to move from
firefighting to something more fundamental and this has proved
successful. This is often reducing variability, identify major
sources first.
There is sensitivity about SPC in a regulatory authority. They
need to be able to work on any problem and come up with a
practical solution; technical solutions may be very assumption
dependent. Presenting the data pictorially is important to see if
things are OK. They need to understand the limitations of
modelling and also understand the broader business problem ñ
getting a good solution to the right question on time and with an
appropriate amount of effort! There is some use of simulation
which need some programming skills. They need to be able to
defend their actions. The issue here is that it is an excellent
solution to the business problem ñ it may only appear a
good solution from a ëstatisticalí perspective.
Penny Evans
Has been with Zeneca only a few years. Finds herself very much
dealing with people, the consulting process. Variability is the
key and this did not come out of her undergraduate course. Her
recollection was that undergraduate statistics was very dry. She
was told how to solve once the thing was set out as a statistical
problem but here she needs to establish what the problem is in
the first instance. The process is much more open-ended. Her role
is to help identify variability.
She is largely involved with experimental design and SPC ñ
but has to explain to others why using them so needs to be able
to talk to the user to help them understand structures.
PSI runs courses for people who have been employed for 1-3
years. The last one they had a 'growing potatoes' exercise which
required careful thinking into experimental design (not analysis).
In many ways the statistical calculations are at the end of a
long line.
Penny needs to ask questions leading to how to solve the
problem. e.g. What are you trying to achieve? Why are you
collecting these data? The statistician has an advantage of being
able to ask stupid questions. They help the others to think in a
structured way and put structure on what they are doing.
She recommended that an undergraduate course should give
students problems to solve and to fit into a real environment. .
Help them to clarify their ideas, or even push for consulting
skills. Places like Zeneca could help with this. The statistician
needs to be able to communicate 'This is what the statistician
could do for you'.
I asked about assessment at university. She felt that exams
were too much regurgitating lectures ñ needs to be embedded. We
must assess 'can they apply it?'. People do need to know
statistical theory. Could the undergraduate be encouraged to talk
to/work with the engineering, chemistry student etc.
Alan Irving
Has had a wide experience with Rolls Royce, lecturer at
Bradford University, now at Zeneca. His view is that
undergraduate courses have not changed much. There are no
communication skills, no consideration of ill-defined problems,
no development of listening skills. Statisticians need to be able
to follow up a loose description of a problem and work in teams.
They need apply common statistical techniques to different
circumstances. They need more practical courses. There is a need
for simple graphical skills which are not always there even
though the ideas are now in the National Curriculum for schools.
The ability to interpret data and presentation skills are
important. He described in more detail the potato study used at
the recent PSI course.
Robert Shaw
Described the PSI course. It would be worthwhile getting hold
of the notes of this course - Chrissie Bennett, Smith Kline
Beecham is the current chair of the subgroup who organises this
01279 622000.
Robert works across Production and Development ñ as do all
the statisticians on this site ñ developing new drugs, bulk
production, manufacturing, chemical problems, lotions, creams and
tablets. They are involved with problem solving, interaction with
FDA. They are beginning to introduce SPC. They work on simulation
of process development
When developing new drugs the statistician is involved with
parameter setting, optimisation, sensitivity and robustness. They
feel they can offer a structured approach, facilitation, the
involvement of someone not too close to the process. There is a
danger that other departments feel that they are losing control.
He has been involved with the in-house in-service courses.
Resistance comes from background culture and difficult
personalities. One strategy they use is to find a champion in the
department and pilot examples, train for awareness, introduce
packages, create a user group, maintain the changes implemented
and try to change the culture as part of a Quality Improvement
Programme.
Overall role is broadly encompassing, general consultancy.
statistical/strategic thinking, problem solving and training. The
challenges are to work as a part of project team to introduce
change and improvement.
In retrospect as an undergraduate he would have found it
helpful to be exposed to practical problems to get at underlying
issues. Needs help in learning to think round a problem. He found
his course dry.
Some final thoughts
Apply some of this approach to change to the MeaNs project.
What motivates university lecturers in statistics. How can we
get them to change, what are their hopes and fears. How can we
apply Quality principles to the process of changing lecturing,
improve the quality of the product, satisfying the customer.
Peter Holmes
June 21, 1996