Transcript The Role of the Pharmaceutical Statistician What can be

Role of Pharmaceutical Statistician
March 10, 2009
The Role of the
Pharmaceutical
Statistician
What can be improved?
Per Larsson
Head of Biostatistics Knowledge Management
& Competency Development
Novo Nordisk
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 2
Stephen Senn (from Dicing with Death)
“Statisticians are engaged in an exhausting but
exhilarating struggle with the biggest challenge that
philosophy makes to science: how do we translate
information into knowledge?”
Are pharmaceutical statisticians today really translating information
into knowledge?
Or are they just struggling with producing more and more information?
What can be improved?
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 3
Roles and responsibilities for statisticians
• Statistics heavily relies on consistency in the choice of
methods and presentations
• Statistics and data presentations should be done in a
consistent way within projects
• Statisticians in a company or project cannot act
independently of each other
• There is a need for strong Project Statisticians
• That are fully accountable for all statistics within their
project
Role of Pharmaceutical Statistician
March 10, 2009
AstraZeneca Statistical Roles
• Global Product Statisticians
• Accountable for all statistics for their product
• Coordinate statistics worldwide for the product
• Therapeutic Area Statistical Experts
• Coordinate statistics worldwide for a therapeutic area
• e.g. Respiratory medicines
• Chief Statistical Expert
• Has an overall statistical responsibility within the company
• Coordinates Therapeutic Area Statistical Experts and Global
Product Statisticians
• Member of the Project Review Board
• Statistics is a rather strong function in AstraZeneca
Slide no 4
Role of Pharmaceutical Statistician
Novo Nordisk Statistical Roles
March 10, 2009
Slide no 5
• ? – July 2008
• Project Statisticians with unclear mandate
• Unclear with regard to
• different phases in drug development
• global responsibility
• manager
• Worked differently in different projects
• Many small projects, large number of project statisticians
• A title rather than a role, mainly used for promotion
• Statistics was a rather weak function in Novo Nordisk
• From July 2008
• International Project Statistician
• Accountable for all statistics within the Key Project Area globally
• A role, not a title
• Requires both statistical skills, project management skills, and good
communication skills
• Currently under implementation
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 6
Novo Nordisk Statistical Roles (cont.)
• The new International Project Statistician role, together
with a corresponding role for statistical programmers,
will be used to standardise statistics within projects
• Regular meetings with all International Project
Statisticians to
• Share knowledge
• Standardise between projects
• Will build a stronger statistical function
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 7
Are we doing statistics in the right way?
• We still have problems with the basics
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 8
Analysis populations
• Lots of people in drug development still believe that
one should make the statistical analyses for both the
ITT and the PP populations, and that these two
analyses correspond to two different questions:
• ITT – how effective is the drug work if prescribed
• PP – how effective is the drug work if taken according to
instructions
• A clinical trial cannot answer any of these questions
• Clinical trials do not mimic real life at all
• In addition the PP analysis is severely biased
Role of Pharmaceutical Statistician
Analysis populations (cont.)
• And sometimes it gets even worse
•
•
•
•
PK Population
PD Population
Alternative PP Population
...
March 10, 2009
Slide no 9
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 10
Analysis populations (cont.)
• There is only one question a clinical trial can answer
• Does the drug have any effect given the conditions in the
trial
• We should answer this question with as little bias as
possible (and use a conservative approach)
• ICH E9, Full Analysis Set
• Use it in the correct way!
• All data for each individual endpoint
• And who wants two results anyway???
• Use one analysis set (FAS) and report one result
• All other investigations should be referred to as stability
analyses
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 11
Non-inferiority trials
• Lots of people in drug development still believe that a
non-inferiority trial can show that treatment A is at
least as effective as treatment B
• Even more scary, some still believe that a p-value of
0.1 means that there is no effect
• EMEA guidance 2005
• “The objective of a non-inferiority trial is sometimes
stated as being to demonstrate that the test product is
not inferior to the comparator. However, only a
superiority trial can demonstrate this.”
Role of Pharmaceutical Statistician
March 10, 2009
Non-inferiority trials (cont.)
• Statisticians have caused much confusion and
contributed to many bad trial designs by not
understanding the difficulties with the non-inferiority
concept
• David Brown, MHRA
• GET RID OF NON-INFERIORITY TRIALS
And bring back active control trials
• The difference between A and B was x, with a 95%
confidence interval. (Thus A is likely to be superior to
placebo, with difference z etc).
Now make your judgements…
Slide no 12
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 13
What information do we present in tables
• Some statisticians still believe that it is ALWAYS useful to
present SDs or SEMs, and that the median should always be
included
Treatment A
Treatment B
Treatment C
• Is the SD or SEM EVER useful?
• We should focus more on USEFUL data presentations
Total
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 14
Error bars
• Lots of people in drug development still like to see error
bars
• Lots of statisticians still provide them
• The reader either ignore the error bars or misinterpret
them
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 15
Use of error bars, Example
Primary endpoint
P= 0.0000144
• A truly bad data presentation
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 16
The Statistical Analysis Plan
• Some people in drug development believe that you can
vaguely describe the statistical analyses in the protocol
and write a SAP later
• ICH E9, 1998
• “Only results from analyses envisaged in the protocol
(including amendments) can be regarded as
confirmatory.”
• The SAP is more of an exception, and can introduce minor
changes based on a blind review of data
• Without the statistical methods sufficiently well
described in the protocol, one cannot judge whether the
design is appropriate
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 17
The Statistical Analysis Plan
• Many protocols are unclear, and some designs are bad,
because the statistician did not carefully plan the
statistics when the trial was planned
• There is usually no reason not to include the full details
of statistical analyses in the trial protocol
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 18
Pre-specification
• Some people in drug development believe that it is important
to pre-specify descriptive statistics and graphs in the trial
protocol or the SAP
• Analyses should be pre-specified, not data presentations
• Some people believe that is important to pre-specify the
significance level
• There is no multiplicity problem associated with the choice of
significance level
• Pre-specification should not have any impact on the reviewers
interpretation of the results
• There is no value in pre-specifying a higher significance level
than 5% (two-sided)
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 19
Has “statistically significant” became a buzzword?
• Is this better or worse than just “better”?
• Statisticians have a responsibility to ensure that their terms
are used in a correct way
Role of Pharmaceutical Statistician
March 10, 2009
Slide no 20
One of the most important improvement
we can make
• Write statistical reports
•
•
•
•
Explain analyses and results in non-technical terms
Put the different analyses in their right perspective
Interpret the analyses
Suggest conclusions
• Translate the information into knowledge
• Thanks!