Chapter 1

Statistics in Drug
Development
Christy Chuang-Stein
Ralph D’Agostino

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Statistical Support to Non-Clinical Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Statistical Support to Clinical Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Battling a High Phase III Failure Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Do Statisticians Count? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Emerging Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.1 Introduction
In the past 50 years, the value of medicine has been clearly demonstrated by a longer life
expectancy, a lower infant mortality rate, and the higher quality of life many of our senior
citizens have been enjoying. Since the introduction of stomach-acid-blocking H2 antagonist
drugs in the late 70’s, the number of surgeries to treat ulcer has been greatly reduced.
Childhood vaccination has literally wiped out diphtheria, whooping cough, measles, and
polio in the U.S. Deaths from heart disease have been cut by more than half since 1950 and
continue to decline. Even though we still face great challenges in combating cancer, great
strides have been made in treating childhood leukemia. Early detection has led to
successful treatment of some types of cancer such as breast cancer. Treatments for
schizophrenia and bipolar disorder have allowed many patients to live almost normal lives.
A report (2006) on the value of medicine can be found at the Pharmaceutical Research and
Manufacturers of America (PhRMA) website.
The use of statistics to support discovery and testing of new medicines has grown
exponentially since the Kefauver-Harris Amendments, which became effective in 1962. The
Kefauver-Harris Amendments required drug sponsors to prove a product’s safety and
efficacy in controlled clinical trials in order to market the product. Since the Amendments,
the number of statisticians working in the pharmaceutical industry has greatly increased.
This increase took another jump when the manufacturing process came under close
scrutiny. As we move into the 21st century, the lure and the promise of genomics and

Christy Chuang-Stein is Site Head, Midwest Statistics, Pfizer, USA. Ralph D’Agostino is Professor, Department of Math-
ematics and Statistics, Boston University, USA.
2 Pharmaceutical Statistics Using SAS: A Practical Guide

proteomics will further intensify scientists’ reliance on statistics. The need to enhance our
overall knowledge about diseases, the need to insert more points into the decision-making
process, and the need to bring economics into development strategy considerations will
undoubtedly present new opportunities for statisticians.
Even in the face of new opportunities, there are many well-established roles for
statisticians in the pharmaceutical industry. The term “well-established” is a relative term
since new roles will become more established over time. For example, trial simulation and
modeling, viewed as new advancements a decade ago, have now become common practice
to help design better trials across the pharmaceutical industry.
Concerned that the current medical product development path may have become
increasingly challenging, inefficient, and costly, the U.S. Food and Drug Administration
(FDA) issued a document in March 2004 entitled “Challenge and Opportunity on the
Critical Path to New Medical Products”. The document attempts to bridge the
technological disconnect between discovery and the product development process. The
disconnect is thought to be largely due to the fact that the pace of development work has
not kept up with the rapid advances in product discovery. The document addresses three
major scientific and technical dimensions in the critical path of product development. The
three dimensions relate to safety assessment, demonstration of a product’s medical utility
(benefit or effectiveness), and the product’s industrialization (scaling up). In addition to
understanding the challenges, establishing the right standards and developing better
toolkits for each dimension will be key to our ultimate success in overcoming the perceived
stagnation in getting new drugs and biologics to the market. Statisticians, with their
training in quantification and logics, can play a major role in the preparation and the
execution of the action plan.
The call for innovation is nothing new for the pharmaceutical industry. The industry as
a whole has made great strides in its basic science research in recent years. Cutting edge
techniques are being developed on a daily basis to probe into the biologic origin and genetic
connection of diseases. The research on microarrays and genomics has produced more data
than could be perceived just a few years ago. With the race to unlock the mysteries of
many diseases and finding cures for them, statistical support needs to be broadened in
dimensions and increased in depth. Time has never been more right for statisticians to
work alongside with their colleagues, being discovery scientists, clinical personnel,
manufacturing engineers, or regulatory colleagues. The collaboration should not only help
transform data to knowledge, but also help use knowledge for better risk-based decisions.
In this chapter, we will briefly cover some traditional statistical support to show how
statistics has been used in many aspects of drug development. Our coverage is by no means
exhaustive. It is simply an attempt to illustrate how broad statistical applications have
been. We will also highlight some areas where a statistician’s contribution will be crucial in
moving forward, in view of the FDA’s Critical Path initiative and the pharmaceutical
industry’s collective effort to take advantage of the FDA’s call for innovation.

1.2 Statistical Support to Non-Clinical Activities

In an eloquent viewpoint article, Dennis Lendrem (2002) discussed non-clinical statistical
support. Traditionally, statistical thinking and approaches are more embraced in areas
where regulators have issued guidelines. Examples are pre-clinical testing of cardiac
liability, carcinogenicity, and stability testing. Recently, Good Manufacturing Practice has
also become a subject of great regulatory interest. The latter captured public attention
when manufacturing problems created a shortage of the flu vaccines for the 2004–2005
season. By comparison, statistical input in areas such as high-throughput screening,
chemical development, formulation development, drug delivery, and assay development is
being sought only when the scientists feel that statisticians could truly add value. This
mentality could limit statisticians’ contributions since researchers will not know how
 Chapter 1 Statistics in Drug Development 3

statisticians could help unless they have previously worked with statisticians or have been
referred to statisticians by their grateful colleagues. For example, scientists who are used to
experimenting with one factor at a time won’t know the value of factorial experiments.
Similarly, even though statisticians well versed in Six Sigma and Design for Six Sigma are
well aware of the many applications of the Six Sigma principles, they need to actively sell
the applications to potential clients.
The non-clinical support model differs from that in the clinical area because of the
usually large client-to-statistician ratio. As a result, after a statistician completes a
particular job, he/she often looks for opportunity to consolidate the techniques and
institutionalize the tools for the client to use on a routine basis. The automation allows
statisticians to focus on opportunities for new collaboration and developing new
methodologies for applications.
Non-clinical statisticians often work individually with their clients. Lendrem (2002)
described them as “pioneers” because of the frequent needs to venture into unknown areas
of new technology. Quantifying gene expression via the microarray technology is one such
example. Another is industry’s (and government’s alike) investment in identifying
biomarkers for testing mechanism of action of new molecular or biologic entities. In both
cases, the findings will have great clinical implications, but the work starts in the research
laboratories and our non-clinical statisticians are the first to deal with the need to measure,
to quantify, and to validate the measurements from the technical perspective.
Because of the small number of non-clinical statisticians in many pharmaceutical
companies, it is useful for non-clinical statisticians to form an inter-company network to
benefit mutual learning. Some of this networking has been in existence for some time. In
the U.S., a CMC (Chemistry, Manufacturing, and Control) Statistical Expert Team was
formed in the late 60’s to focus on the chemistry and control issues related to the
manufacturing of pharmaceutical products. Another example is the Pharmacogenomics
Statistical Expert Team that was formed in the fall of 2003. Both teams are sanctioned by
PhRMA and consist of statisticians from major pharmaceutical companies.

1.3 Statistical Support to Clinical Testing

Clinical testing is typically conducted in a staged fashion to explore the effect of
pharmaceutical products on humans. The investigation starts with pharmacokinetic and
pharmacodynamic studies, followed by proof-of-concept and dose-ranging studies. Some
specialty studies such as drug effect on QT/QTc prolongation and drug-drug interactions
studies are conducted at this early stage. Common adverse reactions and early signs of
efficacy are the objectives of such trials. The early testings, if satisfactory, lead to the
confirmatory phase where the efficacy and safety of the product candidate are more
thoroughly investigated in a more heterogeneous population.
Despite common statistical principles, different stages of clinical testing focus on
different statistical skill sets. For early proof-of-concept and dose-ranging efforts, study
designs could be more flexible and the goal is to learn as efficiently and effectively as
possible. Adaptations, in terms of dose allocation, early termination, and study population
give great flexibility to such trials. Extensive modeling that incorporates accumulated
learning on a real-time basis can lead a sponsor to decision points in a more expedited
fashion. Because the purpose of this phase of development is primarily to generate
information to aid internal decisions, the developers are freer to use innovative approaches
as long as they can successfully defend the decisions that become the basis for later
development.
By comparison, statistical approaches for the confirmatory phase need to be carefully
pre-planned, pre-specified, and followed in order to give credibility to the results. A
pharmaceutical sponsor needs to decide a priori study designs, primary endpoints, primary
analysis population, success criteria, handling of missing data, multiple comparisons, plus
4 Pharmaceutical Statistics Using SAS: A Practical Guide

many others. ICH E9 (1998) gives a very detailed description of all aspects of trial design
and analysis that a statistician should consider at this stage. When adaptation is planned,
the rule needs to be clearly specified in advance. When interim analysis is anticipated, a
sponsor’s access to the interim results needs to be tightly controlled.
The confirmatory phase is the place where knowledge about a new molecular or biologic
entity is solidified to support a target label. The knowledge, along with the approved label,
becomes the basis for recommendations to prescribing physicians and the medical
community. The confirmatory trials are also the place where the risk-benefit and
cost-effectiveness of a new pharmaceutical product are first delineated. The greater number
of subjects studied at this stage gives a sponsor a decent chance to study adverse actions
that have a rate between 0.1% and 1%. This phase overlaps somewhat with the life cycle
management phase where new indications are being explored and drug differentiation is
being sought. If there is a post-marketing study commitment, additional studies will be
initiated to fulfill the conditions for approval.
Increasingly, statisticians are participating in promotion review and educational
communications to the general public. In addition, many statisticians contribute to
activities related to pharmacovigilance and epidemiology.

1.4 Battling a High Phase III Failure Rate

The attrition rate of compounds in the pharmaceutical industry is extremely high. Setting
aside compounds that fail the preclinical testing, it is generally recognized that less than
12% of compounds entering into the human phase testing will eventually makes to the
market place. The rate is a composite figure formed as the product of the success rates of
passing the Phase I testing, passing the Phase II testing, passing the Phase III testing, and
passing the regulatory review. Among failures at the various stages, Phase III attrition has
the greatest impact. This is so not only because of all the accumulated resources expended
up to this point, but it is also because Phase III failure represents a great disappointment
to the sponsor, leaving the sponsor short of a defendable marketing application.
In a recent article, Chuang-Stein (2004) conducted a root cause analysis of the Phase III
failure rate that was reported to be running at the 50% level. This most recent figure is
higher than the 32% rate reported in DiMasi, Hansen, and Grabowski (2003).
Chuang-Stein attributed the cause to three major factors: the candidate factor, the sponsor
factor, and the environmental factor. While we can’t dismiss the pipeline problem, and we
have admittedly very little control over the behaviors of some corporate decision-makers at
the highest level, many of the causes indeed relate to how clinical development is being
conducted and how decisions are made to move compounds through different phases of the
development. Chuang-Stein discussed what statisticians could do to help reduce the
attrition rate at the late stage. One area where the methodology is well developed and
statisticians could make immediate contributions is the judicious use of adaptive designs,
or at least group sequential designs, in Phase III trials. The goal of such designs is to give
the Phase III trials the best chance for success or to terminate them early if the trials are
not likely to meet their objectives. Implicit in such designs is the inclusion of more decision
points based on interim results to allow evidence-based decisions. The need to incorporate
regular decision points is not limited to Phase III testing. It should be part of every stage
of the drug development continuum. These decision points serve as reality checks on the
long and costly development journey.
The industry is at a crossroad, and changes are critically needed. Statisticians should
take advantage of the challenges and fully engage themselves in looking for better ways to
support clinical development of pharmaceutical products.
 Chapter 1 Statistics in Drug Development 5

1.5 Do Statisticians Count?

In a soul-searching article, Andy Grieve (2002) asks whether statisticians count. Even
though the number of statisticians working in the pharmaceutical industry has increased
by 50-fold since the late 70’s, Grieve felt that the influence statisticians had in their
respective companies had not increased proportionally. Grieve looked at the barriers that
prevented statisticians from contributing as much as they could and offered some solutions.
Particularly noteworthy is the assertion that it is the statistician, and not statistics,
that is important. Statistics, as a discipline, does not influence, does not persuade, does not
design studies, does not analyze data, does not interpret findings, and does not report
results. Statisticians are the ones who make the discipline meaningful by doing all of the
above. In other words, statisticians, through their own behavior and communication,
spread the statistical principles and promote the statistical thinking. So, when we discuss
the successful use of statistics in drug development, we need to bear in mind that as
statisticians working in the pharmaceutical industry, we need to be the champions for such
causes through our passion for the statistics profession.

1.6 Emerging Opportunities

The Critical Path initiative document describes many opportunities to improve the
efficiency of product development. We will mention just a few here. On better tools to
assess a compound’s safety, FDA states the need for new techniques to evaluate drug liver
toxicity, new methods to identify gene therapy risk, better predictors of human immune
responses to foreign antigens, new methods to further enhance the safety of transplanted
human tissues, and efficient protocols for qualifying biomaterials. On better tools to
demonstrate the medical utility of a compound, FDA shares the agency’s successful
experience with biomarkers in HIV infection and duodenal ulcer healing. FDA states the
need for more biomarkers and surrogate markers that can guide product development. In
addition, FDA discusses the need for better animal models to combat bioterrorism, more
clinically relevant endpoints, better imaging technologies, more innovative designs and
analysis methods, and the need for implementing the concept of model-based drug
development. The latter involves building mathematical and statistical characterization of
the time course of the disease and the drug effect, using available clinical data.
The Critical Path initiative document also discusses the need for better methods to
characterize, standardize, control, and manufacture medical products on the commercial
scale. Since manufacturing expenses could exceed the research and development
investment, there is a need for a better validation process that follows the risk-based
inspection paradigm advocated by the FDA in recent months. The latter includes more
attention to setting specifications and shifting from detailed data analysis to overall process
quality assessment. The same philosophy suggests moving toward acceptance of a
probabilistic definition, rather than a pass or fail on the manufacturing process. Most
important, FDA wants to encourage the manufacturers to integrate state-of-the-art science
and technology into their manufacturing processes.
Following the issuance of the Critical Path initiative document, different centers within
the FDA have further identified areas for innovations and have presented opportunities to
the FDA’s Science Board on November 5, 2004. Since May 2004, many workshops have
directed at least part of their agenda towards more efficient and effective ways to test and
develop new treatments. Common to many of the discussions are the needs to apply
quantitative thinking and techniques. Taking the clinical phase of product development as
an example, we see that a major emphasis is to use mathematical and statistical models to
help guide drug development and approval. The central idea is to pool data from multiple
trials to augment our knowledge base and actively incorporate such knowledge in
subsequent studies. Interestingly enough, the concept of pooling data has now been
6 Pharmaceutical Statistics Using SAS: A Practical Guide

extended to pooling data of drugs that belong to the same class. The pooling of
information across companies, while challenging, will undoubtedly facilitate the collective
learning of the pharmaceutical industry.
The opportunities for statisticians to make substantial contributions at the strategic
level are beyond what one could have imagined 20 years ago. Along with the opportunities
come expectations that statisticians will help solve the puzzle faced by modern-day
scientists in the pharmaceutical industry.

1.7 Conclusion
Statistics, as a discipline, has broadened its scope significantly over the past 20 years.
Wherever there is a need for quantification, statistics has a role. The ability to think in
terms of variability, to separate signals from noise, to control sources of bias and variation,
and to optimize under given conditions, makes statisticians a valuable partner in the
development of new pharmaceutical and biological products.
Mining historical data to add to our cumulative knowledge is a low-cost and high-yield
activity. Many companies realize the value of this activity and are actively pursuing it. For
example, Bristol-Myers Squibb (Pink Sheet, December 13, 2004) formed a discovery
toxicology group and retrospectively analyzed approximately 100 development compounds
that failed during a 12-year period. Bristol-Myers Squibb hoped to use the acquired
knowledge to decide what assays and technology to implement early to reduce compound
attrition. Bristol-Myers Squibb concluded that a combination of in vitro, in vivo, and in
silico techniques was needed to improve productivity and reduce attrition. According to the
same report in the Pink Sheet (December 13, 2004), other pharmaceutical companies have
reached similar conclusions.
Data mining is also expected to help us look for better predictors for hepatotoxicity and
cardiovascular toxicity such as Torsade de pointes. Data mining examples can go on and
on. I can’t think of any scientists who are more poised and qualified to lead this data-based
learning endeavor than statisticians!
The challenge is definitely on us, statisticians!

References
Chuang-Stein, C. (2004). “Seize the opportunities.” Pharmaceutical Statistics. 3, 157–159.
DiMasi, J., Hansen, R.W., Grabowski, H.G. (2003). “The price of innovation: new estimates of drug
development costs.” Journal of Health Economics. 22, 151–185.
“FDA ‘Critical Path’ May Lead To Changes In Dosing, Active-Control Trials.” December 13, 2004.
Pink Sheet. 31.
Grieve, A. (2002). “Do statisticians count? A personal view.” Pharmaceutical Statistics. 1, 35–43.
International Conference of Harmonization of Pharmaceuticals for Human Use (ICH) E9. (1998).
“Statistical Principles for Clinical Trials.” Available at http://www.ich.org.
Lendrem, D. (2002). “Statistical support to non-clinical.” Pharmaceutical Statistics. 1, 71–73.
Pharmaceutical Research and Manufacturers of America (PhRMA) Report. (2006). Value of
Medicines. Available at http://www.phrma.org/education/.
The Critical Path to New Medical Products. (March 2004). Available at
http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.html.