Title: Biostatistics in Pharmaceutical Research Statistics and Information Sciences at Eli Lilly
1Biostatistics in Pharmaceutical Research
Statistics and Information Sciences at Eli
Lilly CompanySenior StatisticianYun-Fei
Chen, Ph.D.
2My Background
- I was always interested in mathematics
- B.A.and M.S., National Taiwan Normal University
- Major Math
- Wanted to teach high school math
- Ph.D., Univ. of Wisconsin-Madison
- Major Statistics
- Wanted to be statistics professor in Taiwan
- Why became a statistician in pharmaceutical
research? The variety of statistical challenges
3What Applied Statisticians Do
- Learn about their application area
- Pharmaceutical research - Biostatisticians
- Manufacturing / Quality control
- Government statistics / Surveys
- Technology
- Banking / Investment
- Social sciences
- Business / Economics
- Design experiments / Analyze data
- Research new methods for analyzing data
- Teach
4Eli Lilly and Company
- Headquarters Indianapolis, Indiana
- Over 35,000 employees worldwide
- 6,900 (19) in RD
- 11,000 in Indianapolis
- Products sold in 159 countries
- Net sales 11 Billion (2002)
- 6.7 Billion from 6 products
- 19 of sales invested in RD
- 30 compounds expected to reach clinical testing
- Named one of the 100 Best Companies to Work for
in America by Fortune magazine (2002)
5Statistics and Information Sciences
300 Global Statisticians in 10 Countries
(U.S., Canada, Belgium, England, France,
Austria, Germany, Japan, Singapore,
Australia) Supporting Clinical Health
Economics Toxicology Discovery Clinical
Pharmacology Animal Health Product
Development Manufacturing Clients Clinical
Physicians, Analytical Chemists, Biologists,
Management, Process Engineers, Worldwide
Regulatory Authorities, Industry Groups,
Formulation Scientists, Physical Chemists,
Packaging Engineers, Medical Writers ...
6Three Most Fundamental Stages in Drug Development
Discovery
Clinical Trials
MD, Statistician
MD, Chemist Biologist
Toxicology/ADME
VM, Pathologist, Toxicologist, Pharmacist
10,000?1000 ? 100 ? 1
Researchers test thousands of molecules to
isolate one molecule (a lead candidate) that has
the potential to treat a specific disease or
condition
7Developing a New Drug
Hypothesis generation
Candidate development
Commercialization
Target identificationandvalidation Assay
development Lead generation
Globaloptimization
Globallaunch
PhaseIII
Submit
PhaseIB/II
PhaseIA
FHDPreparation
Leadoptimization
Product phase
Program phase
Project phase
Project must meet critical success factors to
move to next phase
Average Cost gt800 Million Dollars per
Drug Average Time 10-15 Years from Discovery to
Commercial Availability
8The Role of Clinical Physician
Discovery
Clinical Trials
MD
MD
- Provide clinical perspective to early drug
development - Translate the pre-clinical results to the
clinical setting - Work with statisticians to design clinical
trials - Oversee the execution of clinical trials,
especially with regards to toxicity of a
compound - Work with investigators and regulatory
scientists on the appropriate study design so
that it can be approved and accepted by the
health care community -
9The Roles of Chemist and Biologist
Discovery
- Chemist Synthesize molecules and SAR
- Play essential role for smaller molecules
- Does the change that I made to this molecule
make - it significantly more soluble, potent and less
toxic? -
-
Chemist, Biologist
- In vitro BiologistDevelopment assay and address
biology of therapeutic candidate and targets - Is the assay optimized/ reproducible?
- Is the biology addressed properly?
-
-
- In vivo Biologist Conduct animal study
- Is the animal model valid?
- Does the effect of this drug in animals really
increase - with dose?
10The Roles of Toxicology/ADME
Discovery
Toxicology/ADME
Clinical Trials
VM, Pathologist, Toxicologist, Pharmacist
Absorption
Distribution
Absorption phase
Excretion (kidneys)
Metabolism (liver)
Elimination phase
concentration
Elimination
Time (hours)
11Phase I/II Statistics
- Development Phase
- Traditional Topics
- Safety
- Pharmacokinetics (blood levels of drug and
metabolites over time) - Optimal dosing
- Efficacy
- All using small sample studies
- Area Undergoing Major Transformation
- High attrition rates (80 candidate drugs fail at
this point) - New emphasis on biomarkers
- Predict safety/efficacy outcomes (surrogate
markers) - Target patient profiles
12What I do A Typical Day?
- Answer questions from scientists
- How should these data be analyzed?
- Should I use SD or SE when I plot my data?
Whats the difference? - Im developing a new assay. Can you design an
experiment to help me optimize the assay
conditions? - Design experiment /Analyze data
- work with others / teams to understand their
studies - Determining the safety and effectiveness of
compounds before going into clinical testing.
Also, determining the validity of the data and
analyses that support these decisions. - Present my analysis results to the team
- In/Formal presentation to 5-50 people
13What I do A Typical Day?
- Discuss/Coach issues with other statisticians
- What is the best way to analyze these data?
- Teach scientists
- Most biologists and chemists are not aware of
statistical - experimental design (factorial design)
- With factorial design it is possible to change
more than - one variable at a time and still determine the
effect of - each one on the outcome
- Statistical packages/webtools
- Statistics Ethics
- Professional activities
- Attend conferences
- Write papers for publication
14Basic Statistics Mean, SD, SE
- Sample Mean
- average of the sample
- estimate the population mean
- the larger the sample size, the more accurate
- Standard Deviation (SD)
- measures how spread out the sample is from its
mean - Use when you describe the patient population
- Standard Error (SE)
- measures the variation of the sample mean
- Use when you describe the mean response
15SD and SE
Mean10, SD5, n4, SE2.5
15
12.5
7.5
5
16Basic Statistics Testing and p-value
Test Treatment A and B to prove A better than
B Start by assuming "No Effect" (null
hypothesis) Gather data Calculate the probability
of getting data at least that rare, assuming "no
effect" (i.e. by chance) The lower this
probability (smaller the p-value), the less
believable the null hypothesis
17Example - Blood Pressure
Blood Pressure
Blood Pressure
Placebo Treated
Before Trt After Trt
Design Single group, measure before and
after treatment Analysis paired
t-test Significant? Depends on SD of the paired
differences. Different p 0.01 Not
Different p 0.21 125 - 80 45 125
- 105 20 145 - 105 40 145 - 130
15 160 - 130 30 160 - 80 80
- Design Parallel Groups
- Analysis 2-sample t-test
- Significant? No
- Example data, p 0.10 Placebo
Treated 125 80
145 105 160
130
18Interesting Fact about SE bars
Number of standard errors required so that
non-overlapping error bars are statistically
significant
Blood Pressure
Placebo Treated
- People tend to judge statistical significant by
whether not the error bars overlap. - Usually, we plot /- 1 standard error (SE).
- Unfortunately, this is not the correct error bar
for doing this.
19Biology - Dose-response Curve
- Does the biological effect increase significantly
with dose? - Fit a sigmoidal dose-response curve using
non-linear regression analysis - Estimate an ED50, the dose required to achieve a
50 effectED50 3.8 mg/kg95 CI (1.8,
8.1) mg/kg
0.1 0.3 1.0 3.0 10
30 100 mg/kg
Actual Dose
20Chemistry - Is one compound more potent?
- Is compound 2 more potent than compound 1?
- Measure potency with EC50 (like ED50)
- Compound 1EC50 9.595 CI (6.6, 13.6)
- Compound 2EC50 3.095 CI (2.6, 3.6)
- The compounds have significantly different
potencies
21Example - Survival Analysis
- Compare survival time (time to death) between
cancer - patients on treatment A to treatment B
22Example Kaplan-Meier Survival Curve
1
0.9
0.8
0.7
0.6
of Patients Free of Relapse
0.5
0.4
0.3
0.2
0.1
0
0
25
50
75
100
125
150
175
200
225
250
Time (days)
Olanzapine
Quetiapine
Log-Rank test p 0.45
23Example - Correlation
The correlation coefficient gives a measure of
the linear relationship between two random
variables -1 correlation coefficient 1 Usually
denoted by r
r -0.95
r -0.03
r 0.95
24Example - Correlation
Correlation is only useful for identifying linear
relationships
Example the correlation coefficient between X
and Y estimated from the data is 0.03 (true
correlation is zero)
Yet the strong (nonlinear) relationship is
obvious True relationship Y (X-0.5) 2
noise Estimated from the data as Y 0.0012
0.9843 (X - 0.5009)2
25Example - Correlation
- If X1 and X2 are independent then they are
uncorrelated - If X1 and X2 are uncorrelated they may be
independent - Common misunderstandings/mistakes
- Low correlation ? no relationship
- consider Y(X-0.5)2 case previously shown
- High correlation ? line is a good fit
- Not Plotting Data
r0.85 r20.73
26Example - Correlation
r 0.7 in all examples
From pages 78-79 of Graphical Methods for Data
Analysis by Chambers, Cleveland, Kleiner, and
Tukey
27Statistics as a Career Why?
Q There are hundreds of careers available to me.
Why should I be a statistician?
- Advancements in technology, genetics, medicine
and - manufacturing have created strong demand of
- statisticians
- Statistician was one of the top 10 most
desirable jobs - and first in the category of Best Working
environment - in the 2002 Jobs Rated Almanac.
- Statisticians play in policy-making roles in
government, - development of pharmaceuticals, market
research, - quality control, and educational programs.
-
28Statistics as a Career How?
Q What do I need?
- MS or PhD has a solid foundation in statistics
and - mathematics and computer skills
- Having knowledge in another subject matter is a
plus - Communication skills, both oral and written
- Teamwork skills
- Professional flexibility
-
-
AMSTAT NEWS, September 2002
29Acknowledgement
- Philip Iversen, Research Scientist, Statistics
and - Information Sciences
- Brian Eastwood, Senior Research Scientist,
Statistics and - Information Sciences
- Michael Lahn, Clinical Research Physician
- Joe Heuer, Research Scientist, BIOTDR in vivo
group - Ho Yeong Song, Research Scientist, BIOTDR in
vitro group - Thomas Tan, Senior Biologist, IIRGRB