Research Data 123

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/265024648
Statistics in Pharmaceutical Development and Manufacturing
Article in Journal of Quality Technology · April 2009

DOI: 10.1080/00224065.2009.11917764
CITATIONS READS
41 25,505
5 authors, including:
Ron Snee Paul R Mcallister

Snee Associates, LLC GlaxoSmithKline
432 PUBLICATIONS 8,072 CITATIONS 25 PUBLICATIONS 1,040 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Statistical Thinking View project
Working to Bring Analytics to Construction Industry Safety Data View project
All content following this page was uploaded by Ron Snee on 16 October 2015.
The user has requested enhancement of the downloaded file.

Statistics in Pharmaceutical
Development and Manufacturing
JOHN J. PETERSON
GlaxoSmithKline Pharmaceuticals, Collegeville, PA 19426
RONALD D. SNEE
Tunnell Consulting, King of Prussia, PA 19406
PAUL R. McALLISTER
GlaxoSmithKline Pharmaceuticals, Upper Merion, PA 19406
TIMOTHY L. SCHOFIELD
Merck & Co., West Point, PA 19486
ANTHONY J. CARELLA
Pfizer Inc., Groton, CT 06340
The pharmaceutical industry is undergoing rapid change and facing numerous challenges, including the
demands of global competition, the need to speed up the drug-development process, and the Food and
Drug Administration’s (FDA’s) expectations for the incorporation of the principles of quality by design
(QbD) and process analytical technology (PAT) in process and analytical development. Statistical thinking
and methods play a significant role in addressing these issues. This article provides an overview of the use of
statistical thinking and methods in the R&D and manufacturing functions of the pharmaceutical industry.
The exposition includes the history of pharmaceutical quality and regulation, phases of pharmaceutical
development and manufacturing and the basic quality and statistical tools employed in each, emerging sta-
tistical methods, the impact of statistical software and information technology, and the role of statisticians
in pharmaceutical development and manufacturing. Four case studies are included to illustrate how these
issues play out in actuality. A summary provides a succinct synopsis of those issues and concludes that the
complex, technical nature of pharmaceutical development and manufacturing offers many opportunities for
the effective use of statistical thinking and methods and that those who use these methods can become
catalysts for both process-development understanding and product-quality improvement. Additional details
can be found in our technical report (Peterson et al. (2009)).
Key Words: Design Space; Drug Manufacturing; FDA; ICH; Opportunity; PAT; Pharmaceutical Industry;
Quality by Design; R&D; Statistical Methods; Statistical Thinking; Technical Change.
North America. He is a Member of ASQ. His email address is

Dr. Peterson is Director in the Research Statistics Unit. He [email protected].
is a Senior Member of ASQ and an ASA Fellow. He is the corre-
Mr. Schofield is currently a Senior Consultant with the
sponding author. His email address is [email protected].
Biologics Consulting Group, Inc. He is a member of the ASA.
Dr. Snee is Principal of Performance Excellence and Lean His email address is tschofi[email protected].
Sigma Initiative Leader. He is a recipient of the ASQ Shewhart
Mr. Carella is a Director in the Nonclinical Statistics
and Grant Medals, and is an ASQ Fellow. His email address
Group, Global Research and Development Division, Pfizer,
is [email protected].
Inc. He is a Senior Member of ASQ. His email address is
Dr. McAllister is Director of the Statistical Sciences Unit, anthony.j.carella@pfizer.com.
Vol. 41, No. 2, April 2009 111 www.asq.org

112 JOHN J. PETERSON, ET AL.
S TATISTICAL concepts and tools have been suc-

cessfully applied for decades in such sectors as
chemicals, automobile manufacturing, and computer-
mance. In pharmaceuticals, however, it is diffi-
cult to tightly connect, say, ‘tablet dissolution’
rates to drug efficacy and safety over a vast ar-
chip manufacturing, but their use in the far more ray of potential product users, each with differ-
regulated pharmaceutical industry presents some ent body size, age, lifestyle, genetics, and drug-
unique challenges. For example, many pharmaceu- metabolism chemistries. Furthermore, pharma-
tical companies hesitate to invest heavily in large- ceutical companies must maintain quality in a
scale manufacturing-process quality before a drug is many-step production process that creates a
approved for marketing because failure in the clinic complex molecule that must have the proper
means product failure. Because time from inception molecular structure and be free of serious chem-
to clinical approval may span 12–15 years, 60–75% of ical impurities or biological contaminants. In
product patent life may have expired by the time the addition, up to now, there has been a lack of
Phase III (confirmatory) trials have been completed. incentive for continuous improvement in phar-
Even after a successful set of Phase IIA (dose deter- maceutical manufacturing after regulatory ap-
mination) and Phase IIB (proof of concept) clinical proval. This is due primarily to the fact that
trials are completed, 40–50% of product patent life substantial changes in the manufacturing pro-
may have expired. In addition, one in three drugs is cess or recipe required formal regulatory ap-
expected to fail in Phase I (first time in man) trials proval. But new regulatory guidance has been
for assessing safety, tolerability, and drug blood lev- recently introduced to provide more flexibil-
els. Pharmaceutical companies are therefore under ity with regard to continuous improvement in
considerable economic pressure to file for approval manufacturing. However, pharmaceutical man-
of the new drug application (NDA) with regulatory ufacturers will have to show clear process un-
authorities as soon as possible after completing suc- derstanding and prediction ability in order to
cessful Phase III clinical trials, when a considerable be granted such flexibility. To meet all these
amount of money must be invested in product launch. complex challenges, pharmaceutical companies
As a result, quality and process understanding ini- need more, not less, statistical thinking and
tiatives must compete for time and money with po- practice.
tential losses from considerable ‘at-risk’ development
• Economic Pressures. Many companies, faced
activities.
with thin product pipelines, major patent expi-
Nevertheless, a number of factors are converg- rations, and downward pressure on pricing, now
ing to increase the need for sophisticated, statistics- need to cut their manufacturing costs, improve
driven approaches to quality and process understand- yield and productivity, and generate bottom-
ing in the pharmaceutical industry, including: line savings that can be used to drive growth
and innovation. Statistically driven improve-
• Regulatory Trends. Recent regulatory guide- ment methodologies found in QbD are critical
lines from the Food and Drug Administra- for success in these efforts.
tion (FDA), the European Medicines Agency
• Increased Need for Effective Technology Trans-
(EMEA), and the International Conference
fer. Virtually every drug at some stage of its de-
on Harmonisation (ICH) encourage scientifi-
velopment or manufacture must be transferred
cally based approaches to quality and compli-
from one site to another. Furthermore, mergers,
ance. Implementing the concepts embodied in
acquisitions, the rise of “global” generics, the
those guidelines—process analytical technology
ongoing rationalization of manufacturing, and
(PAT), quality by design (QbD), and design
other factors have increased the frequency with
space—will require new, more statistically rig-
which pharmaceutical manufacturing organi-
orous and risk-based ways of doing things.
zations must effectively and efficiently trans-
• Inherent Characteristics of Pharmaceutical fer products and manufacturing processes from
Manufacturing. Many of these inherent char- one location to another. Cost pressures, mar-
acteristics—and the challenges they present— ket needs, government regulations, tax benefits,
clearly call for the increased use of sophisticated and logistic issues have also greatly magnified
statistics-driven approaches. For example, in the importance of efficient, compliant, and cost-
most other manufacturing industries, product effective technology transfer—whether it’s to a
specifications are clearly tied to product perfor- nearby plant or a site a world away. Success-
Journal of Quality Technology Vol. 41, No. 2, April 2009

STATISTICS IN PHARMACEUTICAL DEVELOPMENT AND MANUFACTURING 113
ful transfer requires a degree of understanding process understanding, promises to be markedly dif-
of products and processes that can be greatly ferent in terms of industrial statistical focus. Table 1
improved by statistical techniques. provides a summary of the key regulatory events.
As these trends continue and converge, the role The 20th Century: Regulation and Reaction
of statistics and statisticians will only grow larger in
In the United States, the precedent for federal reg-
the industry. To provide readers with a wider per-
ulation of biological products was first established in
spective on the status and use of statistical tools and
the Biologics Control Act of 1902, which came in
methods in the pharmaceutical industry—now and
response to the deaths of 13 children caused by a
in the future—the sections that follow treat these
contaminated diphtheria vaccine and the deaths of 9
key issues:
children caused by a contaminated smallpox vaccine.
• The history of pharmaceutical quality and reg- The act created the Center for Biologics Evaluation
ulation and Research (CBER, which became one of the cen-
• The phases of pharmaceutical development and ters of the FDA in 1972).
manufacturing and the basic quality and statis- In 1906, the United States enacted the Food and
tical tools employed in each Drugs Act, known as the “Wiley Act.” Although
• Emerging statistical methods the law focused largely on adulterated food, it also
• The impact of statistical software and informa- sought to prevent false claims on product labels and
tion technology to force the acknowledgement of product ingredients
• The role of the statistician in pharmaceutical such as alcohol, opium, and morphine.
development and manufacturing In 1927, the precursor to the current agency
dropped all research functions and the current name,
The article concludes with four case studies that
FDA, was applied in 1930. In response to 107 deaths
illustrate how these issues play out in actuality, and
from a poisonous solvent used in the manufacture
an article summary provides a succinct synopsis of
of a sulfa drug, the 1938 Federal Food, Drug, and
the issues and their implications for the future. This
Cosmetic Act (FDCA) required sellers to prove that
article will not address statistical methods or quality
their products were safe. The Act also authorized
in the conduct of clinical trials. The interested reader
factory inspections and supplemented penalties to in-
should see Cleophas et al. (2006) for statistical meth-
clude injunctions as well as seizures of product and
ods in clinical trials and Griffin and O’Grady (2006)
criminal prosecution. The early 1940s saw the death
for a review of quality in the conduct of clinical tri-
of some 300 people from a contaminated sulfa drug,
als. Two books that review statistics in the pharma-
leading to stricter manufacturing standards, which
ceutical industry more broadly are by Millard and
would eventually come to be known as the Good
Krause (2001) and Buncher and Tsay (2005). A list
Manufacturing Practice (GMP) standards. Also dur-
of key abbreviations and acronyms are given in the
ing this period, the FDA enacted batch certification
Appendix at the end of this article.
for some products. This required producers to sub-
mit a sample of every batch of product to the FDA
History for testing.
To understand the forces that have shaped the use
The 1960s witnessed the birth of some 10,000 de-
of statistics in pharmaceutical development today, it
formed infants in Europe from a compound called
is essential to understand the history of the regula-
thalidomide. The compound was not marketed in the
tion of the industry. Although bills to regulate food
United States, but resulted in the Kefauver–Harris
and drugs were introduced in the U.S. Congress as
Drug Amendments of 1962 passed to ensure drug effi-
early as 1879, the modern history of pharmaceutical
cacy and greater drug safety. For the first time, drug
regulation may be divided into two distinct eras: (1)
manufacturers were required to prove to the FDA
the period of largely reactive legislation that lasted
the effectiveness of their products before marketing
through most of the 20th century and (2) the pe-
them. In addition, testing in animals was required
riod of science-based regulatory initiatives that be-
before any human dosing and the FDA’s power to
gan with the dawn of the 21st century and contin-
inspect manufacturing facilities was expanded.
ues today. Though the two periods overlap to some
extent, the new era, with its emphasis on increased The Prescription Drug Marketing Act (PDMA)
Vol. 41, No. 2, April 2009 www.asq.org

TABLE 1. Key Regulatory Events
Year Event Purpose
1906 Pure Food and Drug Act Prevent false claims

1930 FDA takes its current name Agency is purely regulatory—no research functions
1938 Federal Food, Drug, and Cosmetic Act Require proof of safety before marketing
1949 First publication of FDA “Guidance Address the appraisal of toxic chemicals in foods
to Industry”
1962 Kefauver–Harris Drug Amendments Require proof of efficacy and safety before marketing
1987 Prescription Drug Marketing Act Ensure that pharmaceutical products purchased by
consumers are safe and effective, and free from counterfeit,
adulterated, misbranded, subpotent, or expired drugs
2004 Pharmaceutical cGMPs for the 21st Emphasize risk-based approaches to development and
Century—A Risk-Based Approach manufacturing
2004 PAT—A Framework for Innovative Achieve greater understanding of drug development and
Pharmaceutical Development, manufacturing processes. Data acquisition and multivariate
Manufacturing, and Quality analysis cited as important tools
Assurance
2005 ICH Harmonized Tripartite Guideline: Foster quality by design and the understanding of design
Pharmaceutical Development, Q8 space—emphasis on design of experiments to define
interactions and work in multidimensions
2005 ICH Harmonized Tripartite Guideline: Encourage the use of quality risk-management tools in all
Quality Risk Management, Q9 phases of a product’s lifecycle
2007 ICH Harmonized Tripartite Guideline: Enhance science- and risk-based regulatory approaches
Pharmaceutical Quality System, Q10
of 1987 was designed to halt the sale of counterfeit, knowledge about products and processes, technology
adulterated, and expired drugs. The FDA moved to used to manufacture and control these processes, and
expedite the approval of drugs for life-threatening the underlying foundation of a robust quality system
diseases such as AIDS and eased access to drugs for at the manufacturing site. This represented a clear
patients with limited possibilities for treatment. shift away from a long-standing position of rigid reg-
ulation and inspection to achieve quality standards.
This history of pharmaceutical regulation in the
20th century could be viewed as a succession of qual- The knowledge base to support these changes
ity regulations, with heavy legislative involvement begins in research and development and continues
deriving from the potential for pharmaceuticals to though technology transfer and commercial manu-
cause harm to human life. In parallel, the quality of facturing. Information related to the active pharma-
products and processes in nonpharmaceutical indus- ceutical ingredient (API) and drug product formula-
tries has evolved largely as a result of market forces. tion, manufacturing processes and analytical meth-
ods, critical-to-quality parameters and attributes,
The 21st Century: The Rise of Science- and
and product specifications are all key elements of the
Risk-Based Approaches
knowledge base. With this knowledge, the firm and
The focus of the FDA and regulatory agencies in the FDA can determine the potential for events to
many other parts of the world is currently expand- affect fitness for use (i.e., risk). A product’s risk po-
ing to include a greater emphasis on fundamental un- tential can be assessed through a mutually developed
derstanding of manufacturing processes as the basis risk classification system. By sharing risk-mitigation
for a knowledge-driven, risk-based approach to qual- strategies with the FDA, a manufacturer may have
ity. In the early 2000s, the FDA began a program to a product reclassified to a lower risk class. This is
focus on what it called manufacturing science. The one of the key benefits of a science- and risk-based
FDA defined manufacturing science as encompassing approach to GMPs.

These new approaches to regulation, compliance, a definitive list and we apologize for any inadvertent
and quality were embodied in a series of guidelines omissions of names or contribution areas.)
issued to the industry by the FDA and The Interna-
tional Conference on Harmonisation of Technical Re-
quirements for Registration of Pharmaceuticals for Shein-Chung Chow Product stability
Human Use (ICH). The ICH is a unique project that Joseph Ciminera Control charting, process
brings together the regulatory authorities of Europe, monitoring
Japan, and the United States. The last five rows of Mike Free Product stability
Table 1 summarize some of the key documents. They John R. Murphy Screening designs, content
are readily accessible on the FDA and ICH websites. uniformity
Earl Nordbrock Product stability
While regulatory submissions relating to drug Charles B. Pheatt Product dissolution
product manufacturing have traditionally focused on Steven Ruberg Product stability
prescriptive validation and product testing, and less David Salsburg Methods comparison
on process design, the increased process understand- Charles B. Sampson Quality control
ing pointed to by these 21st century guidelines could Jun Shao Product stability
lead to increased regulatory flexibility for drug man- Wayne A. Taylor Sampling plans
ufacturing. In an effort to streamline costs and im- Lynn Torbeck Process validation
prove product quality, the pharmaceutical industry
has become increasingly interested in working with
regulators to modernize drug development and qual-
ity programs. Science- and risk-based regulatory pro- Drug Development,
cesses will ensure that FDA resources are focused on Manufacturing Processes,
the highest risk areas and firms are encouraged to use and Analytical Methods
innovative technology to mitigate risk. However, it is
incumbent upon the firms to ensure that low- and Following the discovery of an API, pharmaceuti-
medium-risk areas remain in an appropriate state of cal development takes place along five parallel paths:
control because these risk classes will receive less reg- (1) clinical trials, (2) preclinical assessment, (3) API
ulatory attention. development, (4) drug product development (final
dosage form), and (5) analytical method develop-
Additional historical evidence of the growing im- ment (Figure 1). The objective of this work is the
portance of statistics to the pharmaceutical indus- submission and approval of an NDA. Once clinical
try may be found in the eventual creation of a Bio- development begins, it usually drives the time lines
pharmaceutical Section within the American Statis- for the other four development paths.
tical Association (ASA). Initially, a subsection of Except in the smallest companies, no individ-
the more general Biometrics Section, full section sta- ual statistician is responsible for supporting all five
tus was driven by the need to have a professional paths. This section treats API development, drug
organization to bring together industry, academia, product formulation, and analytical development, in-
and regulators to help meet the specialized needs cluding the statistical methods and tools that are ap-
of the industry. The early leaders in the creation of plied in pharmaceutical development and manufac-
the subsection were Joe Dresner, Charlie Dunnett, ture. It also demonstrates how these activities fit into
Mike Free, Joe Ciminera, Ron Gauch, Marti Hearron, a QbD paradigm. Although API development, drug
and Joe Meyer. Early in the life of the Biopharma- product formulation, and analytical development are
ceutical Section, the primary preclinical topics were presented sequentially here for the sake of clarity,
those related to toxicology—especially carcinogenic- they usually proceed in parallel and they also sup-
ity studies. Process development and manufacturing port the manufacture of clinical trial supplies. Fur-
issues have only come to the forefront with recent thermore, all three development paths are periodi-
FDA initiatives. Professional organizational homes cally affected by how the drug product is faring in
for these topics are currently spread across ASA and the clinical trials. This section, and indeed the entire
the American Society for Quality (ASQ). article, focus primarily on the activities in the gray
arrows in the box in Figure 1.
Below is a short summary of some of the early
contributors to the area of “industrial” statistics in Table 2 shows how API, drug product develop-
the pharmaceutical industry. (This is not meant to be ment and manufacturing, and analytical-method de-

FIGURE 1. Schematic of Pharmaceutical Development and Manufacturing Processes.
TABLE 2. Phases of Development for API, Drug Product, and Analytical Methods
API Drug product Analytical methods
Definition The active ingredient in the The medicine as Procedures for quantifying the
medicine administered to the amounts of active ingredient
patient (e.g., a pill or impurities in either the
or injection) API or drug product
Stages:
Preclinical Molecule selection, synthetic Animal studies Determine measurement needs,
route development, select analytical methods,
creation of supplies for develop standards
animal testing
Phase I Start GMP manufacturing Consider formulation, Analytical-method optimization
process for clinical trial investigate supplies,
investigation of physical stability requirements
properties
Phase II Scale-up of chemical Drug-product formulation Analytical-method validation
synthesis optimized, shelf-life
studies begin
Phase III Clinical-trial supply Manufacture of clinical Product and process
manufacture, identification trial supplies, scale-up specifications finalized,
of critical process of drug product process, process monitoring (e.g.,
characteristics, process shelf-life determination, control charts) established
validation process validation
Phase IV Technology transfer from Technology transfer from Product and process monitoring
R&D to commercial R&D to commercial for statistical control
manufacturing facility manufacturing facility and quality improvement

velopment for small-molecule therapeutics interlink lot in common, with the principal differences being
with the phases of clinical studies. The relationships that large molecules are associated with many raw
are detailed and discussed in this section of the arti- materials, numerous upstream and downstream pro-
cle. cess steps, a variety of operating conditions, and nu-
merous types of equipment. Because large-molecule
API Development: R&D, Tech Transfer, and drugs are produced from living organisms, variability
Manufacturing is also higher and viral contamination can be an is-
The API in a medication or vaccine is the sub- sue. Nevertheless, the associated statistical methods
stance or organism that fights the symptom or ill- and approaches used are similar to those used for
ness that is being treated or induces immunity to the small-molecule drug production. Additional discus-
pathogen. After the discovery of a molecule that has sion of drug products from small and large molecules
pharmaceutical activity, work is done on this candi- is included in this section.
date API to further refine the material, to create the
Drug Product Development and Commercial
manufacturing process, and to scale it up to the de-
Manufacture (Small Molecules)
sired production levels for full-scale manufacturing.
In the case of small-molecule pharmaceuticals, the Table 3 describes the important formulation and
first step is the chemical synthesis of the API and the process development activities that must be per-
testing of candidate synthetic routes to determine formed and the milestones that must be achieved in
the most effective (chemistry) route from a manufac- preclinical assessment and in clinical trials to demon-
turing perspective. Next, work is done to verify what strate acceptable drug product safety and efficacy to
was made and how to make it at larger scale—for ex- regulatory authorities.
ample, a move from 2–3-gm to 50-gm batches, which The last column of Table 3 lists standard statisti-
are needed to supply preclinical studies in animals. cal techniques and methodologies that are routinely
Work is then done to move to bigger batches (1–5 applied in support of the corresponding stage of drug
kg) using small-scale commercial equipment or scal- product development.
able equipment (not beakers). With Phase III clini-
cal studies, API manufacturing eventually grows to Figure 2 illustrates a typical tablet manufacturing
its largest scale to date, at least one tenth of com- process. The API is blended with excipients to form
mercial batch size. the tablet (i.e., the drug product). The excipients
are ingredients that help to keep the tablet intact in
Much of the statistical thinking and many of the storage and then to dissolve at a particular rate after
methods and tools used in the development and man- ingestion by the patient.
ufacture of an API are the same as those used for
chemical processes. Also, much of process and prod- Drug Product Development and Commercial
uct development involves experimentation—the key Manufacture (Large Molecules and Vaccines)
tool, of course, being design of experiments (DoE).
The primary differences in drug product develop-
Experimentation includes screening experiments,
ment between small molecules and large molecules
product/process understanding studies, regression
(biologics and vaccines) relate to the route of admin-
modeling, process optimization, and robustness stud-
istration of the product and to drug product stabil-
ies. In manufacturing, statistical process control
ity. Most biologics and vaccines are injectables and
(SPC) is used extensively to monitor and improve
must be formulated to cause a minimum amount of
processes. On the improvement side, approaches us-
discomfort to the patient. Some formulation com-
ing statistical techniques such as Six Sigma, Lean
ponents, such as salts, produce stinging. These are
Manufacturing, PAT, Design for Six Sigma, and QbD
necessary, however, to help stabilize the molecule
are increasingly being used.
throughout product shelf-life. Some vaccines, such
It should be noted that the preceding discussion of as live attenuated-virus vaccines, must be stored re-
API applies to drugs developed from synthetic chem- frigerated or frozen, and many undergo a formula-
icals, the so-called “small molecules”. Biologics and tion process called lyophilization, which is freeze dry-
vaccines that are produced in biological systems are ing under carefully controlled time, temperature, and
referred to as “large-molecule” drugs (e.g., proteins), pressure conditions. Similar to the development of
as they are typically much larger molecules. Large- a small-molecule formulation, a large molecule or a
and small-molecule drug production methods have a vaccine formulation is expedited through the use of

TABLE 3. Drug Product Development and Statistical Support
Milestones Activities Statistical support
Nominate an API for Discover the API and perform Multiple comparison techniques for
clinical development various preclinical studies combinatorial chemists; analysis of genomic
data; design and analysis of animal safety
studies, etc.
Perform Phase I Determine Phase I dosage type Analysis of historical data; statistical thinking
clinical studies (e.g., liquid, capsule, tablet (design and analyze experiments)
[or new technology])
Excipient compatibility studies Design and analyze experiments
Accelerated stability studies Regression analysis
Perform Phase IIA Determine Phase II dosage type Analysis of historical data; statistical thinking
(dose ranging) and (new technology) (design and analyze experiments)
IIB (proof of concept)
clinical studies
Evaluate excipient compatibility Design and analyze experiments
(if not performed previously)
Develop Phase II dosage Design and analyze factorial and/or mixture
formulation experiments
Develop Phase II manufacturing Design and analyze factorial and/or response
process surface experiments
Stability studies Regression analysis
Perform Phase III (If necessary, determine Phase (Design and analyze experiments to
clinical studies III dosage type) investigate scalability and/or economic
concerns with Phase II dosage type)
Develop and/or scale Phase III Design and analyze factorial and/or
dosage formulation mixture experiments
Develop and/or scale Phase III Design and analyze factorial, mechanistic,
manufacturing process and/or response surface experiments
Develop PAT applications Multivariate analysis
Transfer technology to Write reports and consult
commercial manufacturing
division
Submit new-drug Develop and/or scale commercial Design and analyze factorial, mechanistic,
application formulation and process mixture, and/or response surface
experiments
Define design and knowledge Design and analyze product- and process-
spaces for DP formulation understanding experiments
and process
Conduct ICH campaign Analyze ICH stability studies (set expiry)

TABLE 3. Continued
Milestones Activities Statistical support
Produce commercial Establish QA procedures Assess process capability and establish quality systems
product to control the process (SPC, PAT, establish sampling
plans, etc.)
Monitor DP stability Analyze data from annual stability lots
Improve the process Data mining, DoE, Six Sigma techniques, Lean
techniques, JIT manufacturing, etc.
multifactor design of experiments. Another key area Analytical Methods

of statistical support is stability study design and
manufacturing modeling. For some of the less stable Analytical methods are used by pharmaceutical
vaccines, release specifications and manufacturing manufacturers and regulatory authorities to help
targets must be supported by strategic, and some- guide development and to characterize and control
times innovative, stability and manufacturing stud- production and distribution of drugs, biologics, and
ies, in order to obtain reliable data to establish a re- vaccines to the public. Most methods are used to help
lease potency that guarantees safe and effective prod- demonstrate the quality of a product, its potency,
uct through expiry and to establish a robust manu- and purity. Methods evolve over the course of de-
facturing process. velopment in correspondence to need and regulatory
requirements. Statisticians are involved throughout
Another feature unique to large-molecule phar- the method lifecycle. This includes development of
maceuticals and vaccines is related to the complex- processing strategies for complex methods, such as
ity of their structure. These products can not usu- bioassay and near infrared spectroscopy, employment
ally be comprehensively characterized and, there- of multifactor DoE to optimize the method, valida-
fore, process or facility changes require carefully de- tion design and analysis to demonstrate the method
signed comparability studies. The industrial statisti- is fit-for-use, and control strategies for maintain-
cian may work with engineers to design an efficient ing satisfactory method performance. Methods range
and effective study. from measurement of physical characteristics of a
FIGURE 2. Tablet Manufacturing Process and Its Variables.

drug, to chemical composition, to biological activity (QC) laboratory. Various approaches are utilized to
in vivo. help assure that the method is performing properly.
An equivalence approach correctly addresses the hy-
Methods used to control pharmaceutical prod- potheses of interest. The change in the average re-
ucts are required to undergo validation in order sponse between laboratories and the increase in vari-
to demonstrate that they are reliable measures of ability are held to restrictions that must be met using
product quality. Validation standards are outlined in methods similar to those used to establish bioequiva-
the United States Pharmacopeia (USP (2005)) and lence (Schuirmann (1987)). Another approach using
the ICH guidelines. These guidelines specify the pa- an β-expectation tolerance interval, similar to its use
rameters requiring validation, as well as validation in method validation, has been advocated to demon-
methodology. Design considerations and other meth- strate that measurements made in the QC laboratory
ods of analysis of validation studies are described in conform to an acceptable shift from the distribution
Schofield (2003) and Boulanger et al. (2007). in the development lab (Dewé et al. (2007)).
Pharmaceutical substances and pharmaceutical
A key element of the quality assessment of prod-
products are controlled and monitored through spec-
uct during development is the determination of prod-
ifications and process capability or control limits.
uct shelf-life. ICH Q1E provides a formulation for
Specifications are defined in ICH (1999) as “a list of
the design and analysis of development stability
tests, references to analytical procedures, and appro-
data (ICH (2003)). Studies are usually performed for
priate acceptance criteria, which are numerical lim-
each product image, which includes different dosages
its, ranges, or other criteria for the tests described.”
and container/closure profiles. A fixed-effects anal-
The limits may be specified in regional compendia
ysis is performed on stability measurements, which
such as the United States Pharmacopeia and the
are designed over time to capture changes in im-
British Pharmacopeia, or developed by the manu-
portant properties of the drug, such as potency and
facturer. Once defined, these are used to assure the
degradants. Rules for pooling the slopes and inter-
quality of products or their intermediates.
cepts of the batches use hypothesis testing at an
The ideal pharmaceutical quality system is com- increased significance level, such as α = 0.25, in
posed of specification limits, release limits, and order to improve the power of the test to detect
process-capability limits. We denote by LSL, LRL, meaningful differences among lots. After pooling has
and LCL the lower specification limit, the lower re- been assessed, shelf-life is estimated as the inter-
lease limit, and the lower capability limit, respec- section of the one-sided 95% confidence interval on
tively. USL, URL, and UCL denote the repective up- the mean and the product specification. Other ap-
per limits. The specification limits (LSL, USL) reflect proaches based on prediction interval (Carstensen
restrictions within which product is fit for use, and and Nelson (1976)) and tolerance-interval concepts
must conform throughout shelf-life. The release lim- (Kiermeier et al. (2004)) have been proposed but do
its (LRL, URL) assure these limits are met at release not appear to be much used in practice as yet.
and throughout product shelf-life and may be calcu-
lated per Apostol et al. (2008), while the control or Statisticians have contributed to development-
process capability limits (LCL, UCL) describe both stability studies by offering design strategies, such
process and assay variability. See Figure 3 for an ex- as bracketing and matrixing of stability time points,
ample. which greatly improve efficiency with little impact
on the effectiveness of the study (Nordbrock (1992)).
Methods that have been developed and validated Bracketing is a strategy wherein extremes of a prod-
in the research laboratories must be transferred for uct image are tested rather than all images. Thus, if
implementation in the manufacturing quality-control product is to be sold as 20-, 50-, and 100-mg doses, a
bracketing study might study only the 20- and 100-
mg doses. Matrixing involves testing only a subset
of the batches at selected time points (a type of in-
complete block design). The combination of bracket-
ing and matrixing leads to a parsimonious evaluation
of product stability. Mixed-effects modeling of “ran-
dom batches” has been proposed by others (Chow
FIGURE 3. Assay Quality-Control Limits. and Shao (1991)). Efficient strategies for monitoring

stability of commercial product have been described sire to implement continuous, real-time quality assur-
in Fairweather et al. (2003). ance. As previously noted, the FDA considers PAT
to be a system for designing, analyzing, and con-
Quality-control testing of manufactured product
trolling manufacturing through timely measurements
is carried out to help assure safe and effective product
(i.e., during processing) of critical quality and per-
and to monitor the process for shifts or trends. Statis-
formance attributes of raw and in-process materials
ticians work with the QC labs to establish test strate-
and processes, with the goal of ensuring final product
gies designed to obtain reliable estimates of product-
quality (FDA (2004a, b)). PAT is expected to pro-
quality attributes related to dose, potency, and pu-
duce gains in quality, safety, and efficiency by (1) re-
rity. Replication may be utilized to increase the pre-
ducing product-cycle times by using on-, in-, and/or
cision of the reportable value, and thereby reduce
at-line measurements and controls, (2) enabling real-
the risks to the customer of an out-of-specification
time release, (3) increasing automation, and (4) fa-
(OOS) result. Some laboratories employ sequential-
cilitating continuous processing.
test plans in which product is tested using a vari-
able number of assay measurements. The statistician Many PAT tools tend to be multivariate in nature
works with the quality laboratory to establish ac- (in both the independent- and dependent-variable
ceptable quality criteria with minimum risk to the sense). Pharmaceutical products and processes in-
manufacturer and customer alike. volve complex, multifactorial systems. The under-
Regulatory guidelines prescribe a rigorous inves- standing of these systems is achieved through the
tigation of the QC laboratory and the manufactur- use of multivariate mathematical approaches, such
ing process when an OOS result is obtained (FDA as statistical design of experiments, response sur-
(2006)). Statisticians may contribute to the process face methodologies, process simulation, and pattern
by providing the laboratory with retest and resam- recognition tools, in conjunction with knowledge-
pling strategies. management systems.
Over time, methods may change or a new method Such data is often collected by way of process an-
may be developed to measure a particular quality alyzers. Such analyzers include those that take uni-
attribute of a product. When this is done, the labo- variate process measurements (e.g., pH, temperature,
ratory performs a concordance analysis. Samples are pressure) and those that nondestructively measure
tested across a range of responses, and a concordance biological, chemical, and physical attributes. These
correlation coefficient is determined as a measure of measurements may be taken at line, in line, or on line.
the agreement of the two methods (Lin (1989)). A Process analyzers typically generate large volumes
related measure is the concordance slope (Schofield of data. Multivariate statistical methods are often
(2003)). This can be obtained from an eigenanalysis needed to extract critical process knowledge for real-
of the two assays, and it measures the slope of the time control and quality assurance (e.g., principal-
relationship between the two methods. components analysis, projection to latent structures,
time series, batch modeling). Sensor-based measure-
Emerging Statistical Methods ments may provide a useful process signature.
Satisfying the FDA’s encouragement of the phar- One of the chief technologies associated with the
maceutical industry to achieve better understand- PAT initiative is near-infrared (NIR) spectroscopy
ing of their manufacturing processes and to quantify (Skibsted (2006)). This technology is a spectroscopic
the risks associated with an OOS product requires method utilizing the near-infrared region of the elec-
a wider set of statistical tools than that commonly tromagnetic spectrum (from about 800 nm to 2500
used in pharmaceutical manufacturing. This need is nm). NIR spectroscopy produces a spectral trace that
in large part driven by new technologies supported may characterize subtle physical and chemical prop-
by the PAT initiative and by the need for better tech- erties of a substance (e.g., chemical intermediate or
nology transfer from development to manufacturing tablet excipient) (see Figure 4).
plant.
Such a trace is a high-dimensional vector response.
Conventional pharmaceutical manufacturing is As the number of experimental units (n) will be much
generally accomplished using batch processing with smaller than the dimension of the vector (p), we have
laboratory testing conducted on collected samples to a “large p/small n” statistical inference situation.
evaluate quality. A key component of PAT is the de- Such situations are typically dealt with by latent-

FIGURE 4. An NIR Trace.
variable methods. Such methods are popular in the Another technology being considered by some
field of chemometrics, which typically uses partial pharmaceutical companies is quasi- (or semi-) contin-
least squares (PLS) to analyze and reduce the di- uous manufacturing (Leuenberger (2001)). The idea
mensionality of such data. here is to produce many small batches, similar in size
to those produced for clinical-trial supplies. This ob-
Spectroscopy, or other technologies that produce viates the need for scale-up and also produces rich
traces of output across space or time, produce what batch-to-batch variation information, which invites
is known as functional data, that is, an observation the use of Bayes and empirical Bayes techniques. A
viewed from the perspective of a ‘profile’ rather than recent overview of Bayesian approaches to process
a point in a one- or low-dimensional space. (See Ram- monitoring, control, and optimization can be found
sey and Silverman (2002) for examples of functional in Colosimo and del Castillo (2007).
data and associated statistical-analysis techniques.)
Because of the increase in functional data due to new Despite a predominance of batch manufactur-
technologies, the field of quality and industrial statis- ing, a case can be made for utilization of contin-
tics is adjusting by starting to create new methods for uous manufacturing, at least for some situations
statistical process control based on such data (Jeong (Kossik (2002)). As such technology becomes in-
et al. (2006)) and for process optimization (Nair et tegrated where needed in the pharmaceutical in-
al. (2002)) involving functional data. dustry, it will generate a corresponding need for
statistical feedback-control procedures (Box and
The use of data-mining statistical methods may Luceño (1997)) and for more sophisticated statistical
help provide a better fundamental understanding of process-monitoring methodologies (e.g., Alt (2007)).
processes measured by functional response traces.
However, practitioners need to understand that data It should be kept in mind that data-intensive
mining is not magic and that the acquisition of new methods may benefit from better data-gathering
knowledge (or law-like relationships) among factors techniques (Steinberg et al. (2008)). Data mining as
affecting a process requires the purposeful variation a practice is oriented toward analyzing a given col-
of factors according to well-designed experiments. lection of data. This alone may be acceptable for
However, the careful use of data-mining techniques those industries for which it may be difficult to con-
to detect high-dimensional special causes may help duct designed experiments (e.g., credit-card compa-
here (Pamias (2005)). nies). However, for an industry such as pharmaceuti-

cals, where most data is obtained from experiments, pointingly small value. What is needed is a process-
concepts of experimental design should be used for optimization approach tied directly to the probabil-
improving data gathering in the presence of data- ity of meeting specifications.
intensive technologies.
Design space, a key concept associated with risk
In the various QbD regulatory documents, the in ICH Q8 (2005), actually is not directly related
word ‘risk’ turns up often, usually in reference to to experimental design. Rather, it is related to a re-
the fact that regulatory authorities see new quality gion of process capability. The ICH Q8 guidance de-
initiatives for pharmaceutical manufacturing from a fines design space as the “multidimensional combi-
‘risk-based’ perspective, with an eye to low risk to nation and interaction of input variables (e.g., ma-
the patient. From a quality and industrial statisti- terial attributes) and process parameters that have
cal viewpoint, this brings up two primary statistical been demonstrated to provide assurance of quality.”
methodologies, more directly those related to risk as- The issue of risk is embodied in the phrase “assur-
sessment and quantification (see ICH Q9 (2005)) and ance of quality.” The guidance goes on to state that
more indirectly those related to variation determina- “working within the design space is not considered
tion. as a change.” Of course this begs the question: “How
much assurance?” Further, the ability to modify the
ICH Q9 stresses identification and assessment of production recipe without regulatory approval (if one
associated risk likelihoods and consequences of risk stays within the design space) means that manu-
events. Classical quality risk-analysis identification facturers could pursue continuous improvement with
procedures, such as failure modes and effects analysis their manufacturing processes.
(FMEA) and fault tree analysis (FTA), are recom-
mended. Probabilistic risk assessment is only men- Currently, however, pharmaceutical manufactur-
tioned briefly as associated with “supporting statisti- ers may have little business incentive for continuous
cal tools.” However, it does appear that some regula- improvement in their manufacturing processes after
tors recognize the important role of probability mod- regulatory approval—for two reasons. First, typical
els for risk assessment. See, for example, Claycamp manufacturing processes require costly and lengthy
(2008). While sophisticated risk-assessment method- regulatory approval for any change in the specified
ologies may be new to the pharmaceutical industry, manufacturing conditions. Second, there is the per-
the changing regulatory atmosphere appears to wel- ceived risk of product batch failure due to changing
come any sound procedure for improving risk assess- the manufacturing conditions (i.e., process parame-
ment. Probabilistic risk assessment and systems reli- ters and input variables).
ability are areas where quality professionals can make
However, a valid design space, approved by regu-
important contributions to pharmaceutical manufac-
lators, would provide pharmaceutical manufacturers
turing.
with the ability to make small-to-moderate changes
Clearly, the determination of risk depends on vari- in their manufacturing conditions (within the design
ation. For example, a probabilistic reliability compu- space) without the time-consuming process of regu-
tation involving a normally distributed endpoint re- latory approval. This would also allow manufactur-
quires knowledge of variation as well as knowledge ers to safely experiment within the design space and
of the mean. Therefore, a proper reliability quan- thereby gain important information about their man-
tification may require a careful variance-components ufacturing process after approval by regulators.
quantification. However, a more insidious issue is The ICH Q8 Annex (2007) appears to suggest
that both quality engineers and statisticians may the classical approach of “overlapping mean response
simply follow statistical methods designed primar- surfaces” as a way to construct a design space. But,
ily for statistical inference on the means as a way to as previously noted, such an approach does not quan-
address assurance of quality. One example is the use tify “how much assurance” of acceptable product,
of “overlapping mean response surfaces” as a way to and the resulting “sweet spot” may possess factor
find the “sweet spot” (Lind et al. (1960), Anderson combinations associated with poor process reliabil-
and Whitcombe (1998)) of a multiple-response pro- ity.
cess with regard to meeting process-response specifi-
cations. Peterson (2004) shows that the probability A Bayesian approach to the ICH Q8 definition of
of meeting specifications when operating well within design space has been given by Peterson (2008), al-
such a sweet spot can be associated with a disap- though such a region for an HPLC assay had been

suggested earlier by Peterson (2004, Fig 6). Two case various technicians to do their own analyses. Such
studies are also illustrated in Stockdale and Cheng software is useful not only for statisticians but for
(2009). Here, design space is defined as teaching and the enablement of (nonstatistical) en-
gineers and scientists. It can also help to empower a
{x : Pr(Y ∈ A | x, data) ≥ R}, (1)
trained statistical “champion.”
where Pr(Y ∈ A | x, data) is the posterior predictive
The pharmaceutical industry is inundated with so-
probability that a new vector of relevant production
phisticated assay equipment and measuring devices
responses, Y , will fall within the specification (i.e.,
(e.g., for image or spectroscopic analyses). Some of
acceptance) region, A, given the experimental data
this equipment has built-in proprietary algorithms
and a vector of controllable process factors, x. Here,
that are statistical in nature. It can be difficult to
R would be some prechosen level of reliability. (See
ascertain how well these algorithms perform or what
Peterson (2008) for some comments on the choice
“fine print” assumptions are required for reliable sta-
of a value for R.) For the design space in (1), the
tistical analyses. Some statistical packages have their
input variables (e.g., raw material attributes) could
algorithms well documented, while others do not.
be treated as noise variables.
On the other hand, a design space could be defined Along with data mining, opportunities in the field
as of high-dimensional information visualization arise
{(x, z ) : Pr(Y ∈ A | x, z , data) ≥ R}, (2) with PAT as well. Shop-floor technicians and chemi-
cal/biochemical engineers will have a strong desire to
where z is a vector of input-material measure- view information in their data whether or not it may
ments. The design space in (2) would be useful be high dimensional. The visual package Spotfire®
for feed-forward control procedures where movement (Spotfire Inc.) can help here, but much more is pos-
within the design space would be motivated primar- sible in theory (see, e.g., Fayyad et al. (2002)).
ily by input-material measurements (MacGregor and
Bruwer (2008)). A further discussion of other appli- Further opportunities for information technology
cations of the Bayesian approach can be found in the to contribute lie in the field of database and knowl-
authors’ technical report, Peterson et al. (2009). edge management. A pharmaceutical-industry sur-
vey (Morris (2005)) has indicated that only 10%
The Impact of Statistical Software of database information has been leveraged to im-
and Information Technology prove overall competitiveness and compliance. So
good opportunities exist for database and knowledge-
For drug-discovery research, the pharmaceutical management professionals to impact the pharmaceu-
industry has already witnessed the rapid increase in tical industry from both competitiveness and regu-
software for number crunching and information stor- latory perspectives. As mentioned above, the PAT
age and retrieval. Part of this has been due to the initiative will only increase these needs due to data
genomic revolution and the associated ‘omic’ techno- complexity and volume.
logical platforms for generating such data. However,
with the advent of the PAT initiative, technology In addition to communication among repositories
will again be driving the need for more sophisticated of collected data, there is a need for related databases
software and information technology to store data, of information that allow for easy storage and re-
retrieve it, and manage it. Further, emerging statis- trieval of data analyses (graphs, point estimates, con-
tical methods, as outlined in the previous section, fidence intervals, etc.) along with the language code
will also require, to some degree, additional com- and software that generated such entities from the
puting support or application development. Mean- data. Again, all of this is more pressing for the highly
while, increased computing speed, storage, and in- regulated pharmaceutical industry.
teractive capability are also affecting quality and in-
For all of the foregoing reasons, the pharmaceu-
dustrial statistics in the pharmaceutical industry. On
tical industry has become interested in “enterprise”
one hand, need will, in some cases, drive software cre-
statistical and information systems. Such systems al-
ation. On the other hand, software availability will
low users to store and track information across vari-
guide what statisticians and quality engineers are
ous groups within an organization. Examples of such
willing to do with statistics and graphics.
systems are the Statistica Enterprise wide SPC sys-
The increased availability of easy-to-use statistical tem (StatSoft, Inc.) , the SAS ® Enterprise Guide®
commercial software allows scientists, engineers, and 4.1 for Statistical Analysis (SAS Institute Inc.), and

the Minitab Quality Companion (Minitab Inc.). Such But what of statisticians themselves? Certainly,
systems provide electronic venues for developing add- they have played a major role in providing the in-
on modules for analyses particular to the pharmaceu- frastructure as well as in solving important problems
tical industry. But such systems also call for greater and expanding the use of statistics in the pharmaceu-
due diligence in software validation across a large tical industry. However, as statistical thinking and
group of company subunits that are in a regulated methods become even more critical for success in
environment. the industry, statisticians will need to acquire new
skills, particularly leadership skills, to move from
Despite the need for reliable software validation, their traditional role as passive advisor to a more
increasing sophistication in technology is driving the dynamic role as creator of value. Statisticians in the
need for increasing sophistication in statistical anal- pharmaceutical industry can provide leadership in
yses and algorithms (Steinberg et al. (2008)). Many three basic, and synergistic, areas: (i) within their
younger statisticians (particularly recent graduates) own company, (ii) within the pharmaceutical indus-
in both the pharmaceutical industry and associated try (e.g., the Pharmaceutical Research and Manufac-
regulatory agencies (e.g., the FDA) have learned turers of America (PhRMA)), and (iii) in collabora-
their applied statistics using the R (open source) tion with statisticians at the FDA (e.g., the annual
computing environment (Bell et al. (2006)). Conse- FDA/Industry Statistics Workshop).
quently, these statisticians want to use R because
it is familiar and often has relatively recent statisti- From Consultant to Collaborator to Leader
cal methods that are not yet available in commercial
As statistics originally evolved with scientific
packages. In fact, the Drug Information Association
management, statisticians became highly specialized
and the FDA cosponsor an “Open Toolbox Initiative
functionaries in large organizations. They analyzed
Forum” to support vendor-neutral software products
data that other people had created and passed along
in an integrated environment.
their analysis to engineers and business leaders who
Interestingly, there are no federal regulations pro- used the analysis in making decisions. In the phar-
hibiting the use of open-source statistical software maceutical industry, as in other industries, statisti-
(Bell et al. (2006)). Instead, the FDA rightly sup- cians first served in that consultative role, typically
ports good software validation that produces a “level working one-on-one with internal “clients,” design-
of confidence” in the results. Further, the choice of ing studies, analyzing data, and providing training.
statistical software should not alter the results (Bell They provided methods for data collection, strategies
et al. (2006)). Given the growing complexity of sta- for design of experiments, and provided guidance on
tistical software, this appears to imply that identical the most effective use of the concepts, methods, and
(or possibly similar) analyses using two or more dif- tools of statistics. When statistical software emerged,
ferent software applications may be needed to help they added that expertise to their portfolios.
provide such a level of confidence. Of course, such A confluence of several powerful trends in recent
“level 1” validation exercises may not be considered years is requiring new roles for the statistician. First,
sufficient for many statistical analyses subject to reg- the internet now makes many data sets accessible
ulatory oversight. For a discussion of unmet software to anyone instantly, undermining the statistician’s
needs for industrial statistics, see the authors’ tech- “ownership” of data. Second, the ubiquitous com-
nical report, Peterson et al. (2009). mercial statistical software described above enables
nonstatisticians to perform many of the statistical
The Changing Role of Statisticians operations that were formerly the province of spe-
cialists. Third, statistics classes are now common-
As the foregoing discussion makes clear, the uti-
place in academia, including in business, engineer-
lization of statistical thinking and methods has made
ing, economics, and social-science curricula; and the
significant contributions to the success of the phar-
widespread use of statistics-intensive methodologies,
maceutical industry. Undeniably, statistical methods
like Six Sigma, has provided mass statistical training
now play an integral part in the industry—in discov-
for nonstatisticians in many organizations.
ery by chemists and biologists, in drug development
and manufacturing by chemists, engineers, and phar- But more positively, and perhaps most impor-
maceutical scientists, and in regulatory compliance tantly, pharmaceutical industry scientists and engi-
and quality control. neers began to recognize the value that statisticians

TABLE 4. Expanded Role of Statisticians
Consultant (Old) Leader (New Expanded)
Analyze data and design experiments Determine the appropriate strategy and approach
Teach statistical tools Design training systems, coach, mentor, and train
Work with technical people Work with managers and technical personnel
Consult on other people’s projects Lead cross-functional projects
Narrow expertise and accountability Broad expertise and accountability
Follow simple regulatory guidelines Collaborate with regulatory agencies to influence new guidance
Reactive Proactive
could provide. As statisticians were assigned to more their expanded roles, they will need to expand their
and more project teams, their role changed from con- skills dramatically (See Table 4).
sultant to collaborator. They made unique contribu-
Many of these leadership skills are new to statis-
tions in their capacity as the primary interpreters
ticians, and they will have to work to acquire them
of data, guiding interpretation and determining the
(Snee and Hoerl (2004)). Other new skills may play
most effective use of tools and methods. Today, in
into some of the traditional strengths of statisticians.
many industries, statisticians serve as team leaders,
Among the most important leadership skills are the
providing guidance and oversight of programs from
following:
beginning to end. While this role is less common in
the pharmaceutical industry, it is a model that can • Effective leaders not only know how to lead,
work effectively and should be considered. they also have substantial business and regula-
New Skills Needed tory knowledge. In their new role, statisticians
will need to understand how business works in
One index of the increasing value that the indus- general, how it works in the pharmaceutical in-
try places on statistical methods may be found in the dustry, and how it works specifically for their
number of industry organizations devoted to address- companies. They will therefore need to under-
ing statistical issues. Those organizations include: stand strategic planning (the process for arriv-
• PhRMA Chemistry, Manufacturing, and Con- ing at change objectives) and strategic deploy-
trol (CMC) Statistics Expert Team comprised ment (the process of implementing strategy).
of ∼30 statisticians from ∼20 PhRMA member • Leaders are comfortable with process and sys-
companies; tems thinking. In many ways, this is the eas-
• Informal Nonclinical Statistics Forum, a collo- iest aspect of leadership for statisticians. Af-
quium of nonclinical statistical managers who ter all, improving the way work gets done in-
meet yearly to share experiences; evitably entails improving processes. Statisti-
• Midwest Biopharmaceutical Statistics Work- cians, steeped in analytic rigor, are uniquely
shop and the FDA/Industry Statistics Work- positioned in that regard. Because leaders are
shop, which has been including an increas- also familiar with the proven, structured im-
ing number of CMC-related statistics topics on provement methods, such as Six Sigma or
their program list. Lean Sigma, that provide the practical means
for improving processes, statisticians have a
The role of statistics and statisticians will grow head start in these statistics-intensive methods.
even more central as regulators move toward the risk- Statisticians are also most strongly positioned
based approach to compliance embodied in PAT and to develop sound statistical approaches and co-
QbD. With the door thus open for even broader and gent quantifications relative to ideas put forth
deeper use of statistical thinking and methods, in- in new regulatory guidance documents, such
dustry statisticians and their employers will need not as the “design space” concept. But it should
only to rethink their technical skills but also to learn be remembered that those methods also in-
to deploy PAT and QbD. As statisticians take on clude such management and leadership skills

as project selection, project management, and lightening regulatory burdens. The challenge is con-
results-tracking with clear metrics and mile- vincing others that statisticians are capable of being
stones. in these leadership positions. In a sense, current lead-
ers are being asked to allow some newcomers into the
• Leaders possess the so-called soft skills in abun- game. This is not easy.
dance (Snee (1998)). To be real leaders of crit-
ical projects, statisticians must know how to Pharmaceutical Case Studies
create stakeholders in a project, including those
executive stakeholders and champions who are The four case studies discussed below provide a
necessary for success. They must become adept small sample of the types of problems encountered
at reviewing, coaching, and communicating. in the development and manufacturing of pharma-
They must understand group dynamics, know ceuticals.
how to lead teams, and know how to design
Case 1: Improving Yield for a Pharmaceutical
and facilitate meetings. Projects, no less than
Synthesis
grand strategy, require vision and direction for
success. Statisticians must not only know how As described in Aggarwal (2006), the yield from
to set such direction but also communicate it the synthesis of a small-molecule pharmaceutical (the
concisely and clearly to their teams. In addi- API) was lower than desired (∼40%). The labora-
tion, they should be adept at removing barri- tory decided to optimize the process using DoE. In
ers, like insufficient resources, lack of training, a sequential experimental design, the initial design
and inadequate time, that impede success. And (a full factorial) was used to screen for significant
they should remember some of those barriers factors. A follow-up experiment (a response surface
lie in people’s psyches—especially in their fear study) was performed over the appropriate region to
of change—and that the best way to overcome optimize the process, using factors suggested by the
them is through coaching and counseling, not initial experiment. The factors studied in the screen-
criticism. ing experiment are illustrated in the following half
normal plot of experimental results (Figure 5).
With these leadership skills, coupled with tech-
Solvent volume, the catalyst loading, and reaction
nical expertise, statisticians should be ideally posi-
temperature were clearly identified as having an im-
tioned to help deploy new risk-based compliance ini-
pact on yield. Interaction plots (Figure 6) revealed
tiatives, to show their organizations where statistical
some of the nature of the effects.
methods can create value, and to create and sustain
statistically driven continuous improvement with the The labs used this information to generate a re-
aim of cutting costs, speeding time to market, and sponse surface design (a face-centered central com-
FIGURE 5. Half-Normal Plot of Effects from Pharmaceutical Synthesis Experiment.

FIGURE 6. Plots for Volume–Temperature and Catalyst–Temperature Interactions.
posite design, to avoid setting levels in excess of the design-space construction within their company. See-
capability of the reaction), using the factors identi- ing this as an opportunity, some statisticians within
fied in the screening design. The labs also hoped to the company decided to develop a probabilistic risk-
discover how to manage catalyst load using some of based approach to design-space construction to ad-
the other factors because the catalyst was expensive. dress, in particular, the need to demonstrate “assur-
The resulting response surface revealed that yield ance of quality” as required by the ICH Q8 guidance.
could be optimized with increases in temperature and
The statisticians involved (in both the United
solvent volume, while the amount of catalyst could
States and Europe) worked together as an infor-
be decreased due to the synergistic impact of these
mal team and separately as individual consultants
two factors. Final yields were confirmed to be in ex-
and members of matrix teams to formulate and
cess of 90% compared with average yields of approx-
build example Bayesian design spaces as defined in
imately 40% that were observed prior to employing
(1) in the Emerging Statistical Methods section.
DoE. Thus, the lab doubled its capacity, while saving
This approach was applied to problems in both
in cost by using less catalyst.
(small-molecule) pharmaceutical and biopharmaceu-
tical projects. Some prototype programs were written
Case 2: Leadership for Design-Space Issue
in R and SAS/IML, and then data from experiments
were analyzed and presented to scientists and their
As stated previously, the new regulatory concept
management. Some external presentations and pa-
of “design space” involves creating a multidimen-
pers were generated. Eventually many in-roads were
sional process capability region that has been demon-
made to acceptance by the company scientists of this
strated to provide “assurance of quality”. Recently,
creative formulation of a design space. A simplified
in a large, multinational pharmaceutical company,
example of one of their reliability-based design spaces
various scientists (process chemists, pharmaceutical
is shown in the probability surface plot in Figure 7.
scientists, chemometricians, etc.) were meeting to
formulate ideas about how to develop an approach to Further progress toward company-wide accep-

FIGURE 7. Reliability Contour Plot (See Level Scale on Right). The elliptical region in white is a prototype design space.
This region has higher relative probability of simultaneously meeting all product-release specifications as a function of the
temperature and catalyst factor levels. (Plot obtained from Stockdale and Cheng (2009). Used with permission from the
Association for Quantitative Management.)
tance is still ongoing. Some future hurdles remain factured across a range of processing conditions, in
(e.g., addressing functional data). A definitive mea- order to best simulate the product distribution. Each
sure of success would be for such a design space to be lot was assayed twice in the bioassay, in a manner
submitted by the company, approved by regulatory that would allow for independent estimates of inter-
authorities, and implemented in the manufacturing run variability throughout the experiment using the
plant. design shown in Figure 8.
Case 3: Validation of a Complex Tedious A reference (Ref) was included in each run to cal-
Biological Assay ibrate bioassay results, while a clinically relevant lot
(Clin) was included in order to set a specification
Validation of biological assays, particularly assays for commercial materials tested in the bioassay. The
using large numbers of animals, is a tedious and ex- reference was tested in duplicate in order to derive
pensive prospect. It’s particularly important, as well, a criterion for “parallelism” in the bioassay accord-
to make sure that a minimum number of animals are ing to Hauck et al. (2005). The validation results are
utilized for these purposes. An animal potency as- illustrated in Figure 9.
say was developed for a vaccine product that uses
120 animals per run and takes 6 weeks to perform. Excellent precision was observed between replicate
The challenge was to obtain the maximum amount runs of the bioassay for all lots, while commercial ma-
of information with minimum use of mice and time. terials performed as well as or better than lots tested
The laboratory chose to include an assessment of in clinical trials (POC Lots 1–3 in Figure 9). The pre-
the product distribution together with the assay val- cision estimated from these data was used together
idation and thereby gain valuable information about with a minimum threshold to derive a release po-
both assay and product variability. Ten lots of prod- tency limit for lots of commercial material according
uct were strategically identified, which were manu- to Apostol (2008).

Run
1 2 3 4 5 6 7 8 9 10
Ref Ref Ref Ref Ref Ref Ref Ref Ref Ref
Lot 1 Lot 1 Lot 2 Lot 4 Lot 4 Lot 5 Lot 7 Lot 7 Lot 8 Lot 9
Lot 2 Lot 3 Lot 3 Lot 5 Lot 6 Lot 6 Lot 8 Lot 9 Lot 10 Lot 10
Clin Ref Clin Ref Clin Ref Clin Ref Clin Ref
FIGURE 8. Design for Biological Assay Validation.
Case 4: Regulatory Action and Use of determined to use its ultimate stricture—seizure of
Statistical Methods product at the manufacturing site. Federal marshals
entered the manufacturing site and seized all lots of
A brief conceptualized version of a recent incident
the particular product. Manufacturing of the prod-
illustrates the leverage exercised by regulatory au-
uct was halted. The FDA appointed a third party
thority. A pharmaceutical company manufactures a
“monitor” to oversee the investigation of root cause,
tablet with two active ingredients. One of the ingre-
the implementation of process changes, and the as-
dients is present in a very small absolute amount
sessment of new manufacture postimprovement. The
by weight. Small absolute deviations in the actual
third-party monitor was hired at the expense of the
weight of this ingredient produce large relative (per-
product manufacturer. Additionally, the manufac-
cent) deviations from the targeted dosage. As with
turer was required to post a substantial bond that
most drug products, the specifications for this active
would be forfeited in the event of insufficient cooper-
ingredient are expressed as an allowable range of per-
ation and progress.
centage of target dose. These ranges are typically 90
to 110% of target dose. Where were the statisticians? The ultimate ques-
More important for this discussion, the manufac- tions in this incident regarded patient safety. One
turer is required to report all OOS events to the slightly underdosed tablet may not be harmful, but
FDA. These events are monitored and the agency has when a medicine is taken daily to treat a chronic
a series of escalating remedies to enact. Early among ailment (e.g., asthma or diabetes), the potential
these is the right to send inspectors to the manu- long-term effects are of chief concern. Therefore,
facturing facility at any time, announced or unan- the statisticians were asked to calculate probabili-
nounced. The inspectors must have a stated purpose, ties such as, “How many defective tablets were likely
but once an inspection begins, their audit trail may to occur in a monthly supply of 30?” or “How many
lead in many directions, particularly if irregularities defective tablets might a patient obtain in a year’s
are encountered along the way. In the example here, supply?” These probabilities depend on the failure
a series of OOS lots of product occurred over the rate in the manufacturing process. The underlying
course of a number of months—not every lot failed, statistical questions then become “the best estimate
but there were enough failures to cause notice. The of the tablet failure rate” and “the upper bound on
facility was inspected; the OOS reports were audited, the failure rate.” The data suggest that the tablet
and the conclusion was reached by the FDA that the failure rate, π, varies from batch to batch. Binomial
manufacturer did not understand the root cause of mixture models are suggested, where π is a random
the failures and was therefore presenting an unmiti- effect associated with variations in the set-up, inputs,
gated risk to the public. Note that this is not a case and conditions of each manufacturing run.
of direct contamination or of an immediately harmful
In terms of uncovering root causes for the fail-
level of active ingredient. The regulator was princi-
ures, a primary question is “When did the problem
pally concerned about the lack of apparent under-
begin?” This means extensive examination of time-
standing demonstrated by the manufacturer.
series manufacturing data. An element of statisti-
The manufacturer’s responses to the regulatory cal interest is that the batch sizes vary and range
concerns were deemed unsatisfactory and the FDA from hundreds of thousands to millions of tablets,

FIGURE 9. Results for Biological Assay Validation Study. (Note: run 4 of the design was repeated due to a performance
issue related to the duplicate of the reference; thus, three replicates of lots 4 and 5 appear in the figure.)
and the number of tablets sampled from batches also and quality and the use of statistical methods for
varies—ranging from 20 to a few hundred. Control- both. These regulatory trends, as well as the in-
chart techniques were heavily utilized. In addition, a herent complexity of pharmaceutical manufacturing,
variety of graphical techniques, including box plots economic pressures, and the increased need for tech-
and cusums, were helpful in explaining statistical nology transfer, will continue to accelerate the in-
conclusions to both regulators and company senior dustry’s need for sophisticated, statistics-driven ap-
managers. Identification of a starting point allowed proaches to quality and process understanding as
the manufacturer to identify process changes or inci- well as the statisticians to apply them.
dents that occurred in the specified time frame. This
generated hypotheses regarding ultimate causes. The Today, pharmaceutical development, following
statisticians were again involved in helping design ex- the discovery of an active pharmaceutical ingredi-
periments that would test those hypotheses. Resolu- ent (API), typically proceeds along several paral-
tion and prevention were largely engineering consid- lel paths, each requiring particular statistical tech-
erations. However, an extensively revised monitoring niques. These tools and statistical methods in-
system was put in place—again with heavy emphasis clude Design of Experiments (DoE), screening ex-
on the use of control-charting techniques, including periments, optimization studies, regression mod-
appropriate training and interpretation. Ultimately, eling, process optimization, and robustness stud-
the manufacturer was successful in reinstating their ies. Analytical-method development employs such
product in the marketplace. tools as analysis of variance (ANOVA), variance-
component studies, method ruggedness studies, and
Summary basic statistical techniques, including graphics. In
manufacturing, statistical process control (SPC) is
Throughout most of the 20th century, pharmaceu- used extensively to monitor and improve processes.
tical regulation was largely a matter of reactive leg- Statistical techniques found in Six Sigma, Lean Man-
islation, but this century has seen the rise of science- ufacturing, PAT, Design for Six Sigma, and QbD are
based regulatory initiatives, including ICH guide- also increasingly being used to improve processes.
lines on pharmaceutical development, pharmaceuti-
cal quality, the pharmaceutical quality system, and The FDA’s encouragement of the pharmaceutical
the FDA’s guidance on Good Manufacturing Prac- industry to achieve better understanding of manufac-
tice (GMP), Process Analytical Technology (PAT), turing processes and to quantify the risks associated
and Quality by Design (QbD). These initiatives have with out-of-specification product has widened the set
brought a new emphasis on process understanding of statistical tools to include the use of multivari-

ate mathematical approaches, such as response sur- Key Abbreviations and Acronyms (Continued )
face methodologies, process simulation, and pattern
recognition tools, in conjunction with knowledge- GMP Good Manufacturing Practice
management systems. These emerging statistical ICH International Conference on
methods are also driving the development of more Harmonisation
powerful commercial statistical software, which is LC Label claim
enabling nonstatisticians to do their own statisti- LOA Limit of agreement
cal analyses. Information technology is also making NDA New drug application
contributions to data mining, database management, OOS Out of specification
and knowledge management, increasing the indus- PAT Process analytical technology
try’s interest in “enterprise” statistical software that PhRMA Pharmaceutical Research and
can be used across an organization. Manufacturers Association
As statistical thinking and methods become even QbD Quality by design
more critical for success in the industry, statisticians QC Quality control
have begun to move from their traditional role as RSD Relative standard dSeviation
passive advisor to a more dynamic role as creator USP United States Pharmacopeia
of value. Their role is becoming even more central
as regulators move toward the risk-based approach
to compliance embodied in PAT and QbD, requiring
References
industry statisticians to rethink their technical skills,
learn to deploy PAT and QbD, and acquire new lead- Aggarwal, V. K.; Staubitz, A. C.; and Owen, M. (2006).
ership skills commensurate with the increased impor- “Optimization of the Mizoroki–Heck Reaction Using Design
of Experiment (DoE)”. Organic Process Research and De-
tance of their discipline within the industry. With velopment 10, pp. 64–69.
those new skills and methods, they can become cat- Alt, F. B. (2007). “A Bayesian Approach to Monitoring the
alysts for both process-development understanding Mean of a Multivariate Normal Process”. In Bayesian Pro-
and product-quality improvement. cess Monitoring and Optimization, Colosimo, B. M. and del
Castillo, E. (eds.), pp. 139–166. Boca Raton, FL: Chapman
and Hall.
Acknowledgments
Anderson, M. J. and Whitcomb, P. J. (1998). ‘’Find
First, we would like to thank the Discussants the Most Favorable Formulations”. Chemical Engineering
for their valuable input and review of our paper. Progress April, pp. 63–67.
Apostol, I., I.; Schofield, T. L.; Koeller, G.; Powers,
We would also like to thank Professor Enrique del
S.; Stawicki, M.; and Wolfe, R. A. (2008). “A Rational
Castillo (JQT Editor for 2006–2008) for inviting us to Approach to Setting and Maintaining Specifications for Bio-
develop such an expository article. Finally, we thank logical and Biotechnology-Derived Products.” Biopharm In-
Professor Joe Pigeon and Dr. Darryl Downing for ternational 21(6), 21(7), 27(8).
their review of the (expanded) technical report ver- Bell, S. B.; Morrish, K.; Harrison, F.; Petullo, D.;
sion of this paper. Thompson, L. and Gray, G. (2006). “Times ‘R’ a Chang-
ing: FDA Perspectives on ‘Open Source’ ”. Presented at the
Joint Statistical Meetings, August 6th, Seattle, WA.
Appendix Boulanger, B.; Dewé, W.; Gilbert, A.; Govaerts, B.; and
Maumy, M. (2007). “Risk Management for Analytical Meth-
Key Abbreviations and Acronyms
ods Based on the Total Error Concept: Conciliating the Ob-
jectives of the Pre-Study and In-Study Validation Phases”.
Abbreviations/ Chemometrics and Intelligent Laboratory Systems 86, pp.
acronyms Meaning 198–207.
Box, G. E. P. and Luceño, A. (1997). Statistical Control by
API Active pharmaceutical ingredient Monitoring and Feedback Adjustment. New York, NY: John
cGMP Current Good Manufacturing Wiley & Sons.
Practice Buncher, C. R. and Tsay, J.-Y. (2005). Statistics in the
Pharmaceutical Industry, 3rd edition. Boca Raton, FL: CRC
CMC Chemistry, manufacturing,
Press.
and control Carstensen, J. T. and Nelson, E. (1976). “Terminology Re-
DCU Dose-content uniformity garding Labeled and Contained Amounts in Dosage Forms”.
EMEA European Medicines Agency Journal of Pharmaceutical Sciences 65, pp. 311–312.
FDA Food and Drug Administration Chow, S. C. and Shao, J. (1991). “Estimating Drug Shelf-
Life with Random Batches.” Biometrics 47, pp. 1071–1079.

Claycamp, H. G. (2008). “Room for Probability in ICH Q9: Leuenberger, H. (2001). “New Trends in the Production of
Quality Risk Management”. Presented at the Pharmaceuti- Pharmaceutical Granules: Batch Versus Continuous Process-
cal Statistics 2008 Confronting Controversy conference or- ing”. European Journal of Pharmaceutics and Biopharma-
ganized by the Institute of Validation Technology, March ceutics 52, pp. 289–296.
18–19, Arlington, VA. Lin, L. I. (1989). “A Concordance Correlation Coefficient to
Cleophas, T. J.; Zwinderman, A. H.; and Cleophas, T. F. Evaluate Reproducibility”. Biometrics 45(1), pp. 255–268.
(2006). Statistics Applied to Clinical Trials, 3rd edition. New Lind, E. E.; Goldin, J.; and Hickman, J. B. (1960). “Fitting
York, NY: Springer. Yield and Cost Response Surfaces”. Chemical Engineering
Colosimo, B. M. and del Castillo, E. (2007). Bayesian Pro- Progress 56, p. 62.
cess Monitoring and Optimization. Boca Raton, FL: Chap- MacGregor, J. and Bruwer, M.-J. (2008). “A Framework
man and Hall/CRC. for the Development of Design and Control Spaces”. Journal
Dewé, W.; Govaerts, B.; Boulanger, B.; Rozet, E., of Pharmaceutical Innovation 3, pp. 15–22.
Chiap, P.; and Hubert, Ph. (2007). “Using Total Error as Millard, S. P. and Krause, A. (2001). Applied Statistics in
Decision Criterion in Analytical Method Transfer”. Chemo- the Pharmaceutical Industry. New York, NY: Springer.
metrics and Intelligent Laboratory Systems 85, pp. 262–268. Minitab, Inc., 1829 Pine Hall Rd, State College, PA.
Fairweather, W. R.; Mogg, R.; Bennett, P. S.; Zhong, J.; Morris, K. (2005). “Making the Most of Drug Development
Morrisey, C.; and Schofield, T. L. (2003). “Monitoring Data”. Pharmaceutical Manufacturing April, p. 73.
the Stability of Human Vaccines”. Journal of Biopharma- Nair, V.; Taam, W.; and Ye, Q. (2002). “Analysis of Func-
ceutical Statistics 13(3), pp. 395–413. tional Responses from Robust Design Experiments”. Journal
Fayyad, U.; Grinstein, G. G.; and Wierse, A. (2002). In- of Quality Technology 34, pp. 355–370.
formation Visualization and Knowledge Discovery. London: Nordbrock, E. (1992). “Statistical Comparison of Stability
Academic Press. Study Designs”. Journal of Biopharmaceutical Statistics 2,
FDA. (2004). “Guidance for Industry: PAT—A Framework pp. 91–113.
for Innovative Pharmaceutical Development, Manufacturing, Pamias, J. (2005). “PAT Success and Speed Hinge on Data
and Quality Assurance”. Mining”. PharmaManufacturing.com, at http://www.pharm
FDA. (2004). “PAT—A Framework for Innovative Pharma- amanufacturing.com/articles/2005/276.html.
ceutical Development, Manufacturing, and Quality Assur- Peterson, J. J. (2004). “A Posterior Predictive Approach to
ance”. Multiple Response Surface Optimization”. Journal of Qual-
FDA. (2006) “Guidance for Industry: Investigating Out-of- ity Technology 36, pp. 139–153.
Specification (OOS) Test Results for Pharmaceutical Pro- Peterson, J. J. (2008). “A Bayesian Approach to the ICH Q8
duction”. Definition of Design Space”. Journal of Biopharmaceutical
Griffin, J. P. and O’Grady, J. (2006). Textbook of Phar- Statistics 18, pp. 959–975.
maceutical Medicine, 5th edition. Malden, MA: Blackwell Peterson, J. J.; Snee, R. D.; McAllister, P. R.;
Publishing. Schofield, T. L.; and Carella, A. J. (2009). “Statistics
Hauck, W. W.; Capon, R. C.; Callahan, J. D.; Muth, J. in Pharmaceutical Development and Manufacturing”. DDS
E. D.; Hsu, H.; Lansky, D.; Sajjadi, N. C.; Seaver, S. S.; Technical Report, at www.biometrics.com (Statistical and
Singer, R. R.; and Weisman, D. (2005). “Assessing Paral- Quantitative Sciences).
lelism Prior to Determining Relative Potency”. Journal of Ramsey, J. O. and Silverman, B. W. (2002). Applied Func-
Pharmaceutical Science and Technology 59(2), pp. 127–137. tional Data Analysis. New York, NY: Spriger-Verlag.
ICH. (1999). “ICH Harmonised Tripartite Guideline Q6A, SAS Institute Inc., Cary, NC.
Specifications: Test Procedures and Acceptance Criteria for Schofield, T. L. (2003). “Contributions on ‘Assay Develop-
New Drug Substances and New Drug Products: Chemical ment’ and ‘Assay Validation’ ” in Encyclopedia of Biophar-
Substances”. maceutical Statistics, S.-C. Chow (ed.). New York, NY: Mar-
ICH. (2003). “ICH Harmonised Tripartite Guideline Q1E cel Dekker, Inc.
Evaluation of Stability Data”. Schuirmann, D. J. (1987). “A Comparison of the Two One-
ICH. (2005). “ICH Harmonized Tripartite Guideline: Pharma- Sided Tests Procedure and the Power Approach for Assess-
ceutical Development, Q8”. ing the Equivalence of Average Bioavailability”. Journal of
ICH. (2007). “ICH Harmonized Tripartite Guideline: Pharma- Pharmacokinetics and Pharmacodynamics 15, pp. 657–680.
ceutical Development: Annex to Q8”. Skibsted, E. (2006). “Near Infrared Spectroscopy: The
ICH. (2005). “ICH Harmonized Tripartite Guideline: Quality Workhorse in the PAT Toolbox”. At www.spectroscopy.com,
Risk Management, Q9”. vol. 18, no. 5.
Jeong, M. K.; Lu, J.-C.; and Wang, N. (2006). “Wavelet- Snee, R. D. (1998). “Non-Statistical Skills that Can Help
Based SPC Procedure for Complicated Functional Data”. Statisticians Become More Effective”. Total Quality Man-
International Journal of Production Research 44, pp. 729– agement Journal 9(8), pp. 711–722.
744. Snee, R. D. and Hoerl, R. W. (2004). “Statistical Leader-
Kiermeier, A.; Jarrett, R. G.; and Verbyla, A. P. (2004). ship—As Traditional Workplace Roles Change, Learn to
“A New Approach to Estimating Shelf Life”. Pharmaceutical Transition from Consultant to Leader”. Quality Progress Oc-
Statistics 3, pp. 3–11. tober, pp. 83–85.
Kossik, J. (2002). “Think Small: Pharmaceutical Facility Snee, R. D. and Hoerl, R. W. (2005). Six Sigma Beyond the
Could Boost Capacity and Slash Costs by Trading in Cer- Factory Floor—Deployment Strategies for Financial Ser-
tain Batch Operations for Continuous Versions”. Pharma- vices, Health Care and the Rest of the Real Economy. Upper
mag.com, article ID/DDAS-SEX 52B/ http://www.pharma Saddle River, NJ: Pearson Prentice Hall.
manufacturing.com. Spotfire, Inc. 212 Elm Street, Somerville, MA.

StatSoft, Inc., 2300 East 14th Street, Tulsa, OK. Stockdale, G. W. and Cheng, A. (2009). “Finding Design
Steinberg, D. M.; Bisgaard, S.; Doganaksoy, N.; Fisher, Space and a Reliable Operating Region Using a Multivari-
N.; Gunter, B.; Hahn, G.; Keller-McNulty, S.; Ketten- ate Bayesian Approach with Experimental Design”. Quality
ring, J.; Meeker, W. Q.; Montgomery, D. C.; and Wu, C. Technology and Quantitative Management, to appear.
F. J. (2008). “The Future of Industrial Statistics—A Panel United States Pharmacopeia 28 (2005). “Validation of
Discussion”. Technometrics 50(2), pp. 103–127. Compendial Methods,” chapter 1225.
View publication stats

Research Data 123

Uploaded by

Copyright:

Available Formats

Research Data 123

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Research Data 123

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Statistics in Pharmaceutical Development and Manufacturing

Article in Journal of Quality Technology · April 2009

Ron Snee Paul R Mcallister

SEE PROFILE SEE PROFILE

Statistical Thinking View project

Working to Bring Analytics to Construction Industry Safety Data View project

The user has requested enhancement of the downloaded file.

North America. He is a Member of ASQ. His email address is

Vol. 41, No. 2, April 2009 111 www.asq.org

S TATISTICAL concepts and tools have been suc-

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

TABLE 1. Key Regulatory Events

Year Event Purpose

1906 Pure Food and Drug Act Prevent false claims

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

FIGURE 1. Schematic of Pharmaceutical Development and Manufacturing Processes.

API Drug product Analytical methods

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

TABLE 3. Drug Product Development and Statistical Support

Milestones Activities Statistical support

Journal of Quality Technology Vol. 41, No. 2, April 2009

Milestones Activities Statistical support

multifactor design of experiments. Another key area Analytical Methods

FIGURE 2. Tablet Manufacturing Process and Its Variables.

Vol. 41, No. 2, April 2009 www.asq.org

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

FIGURE 4. An NIR Trace.

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

TABLE 4. Expanded Role of Statisticians

Consultant (Old) Leader (New Expanded)

Journal of Quality Technology Vol. 41, No. 2, April 2009

FIGURE 5. Half-Normal Plot of Eﬀects from Pharmaceutical Synthesis Experiment.

Vol. 41, No. 2, April 2009 www.asq.org

FIGURE 6. Plots for Volume–Temperature and Catalyst–Temperature Interactions.

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

FIGURE 8. Design for Biological Assay Validation.

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

Journal of Quality Technology Vol. 41, No. 2, April 2009

Vol. 41, No. 2, April 2009 www.asq.org

Journal of Quality Technology Vol. 41, No. 2, April 2009

View publication stats

You might also like