The document discusses building an integrated early clinical development platform to improve the path to proof of concept for new drug candidates. It argues that traditional single-site phase I studies are evolving into a networked model where learning is maximized through connections to patient populations, biomarkers, adaptive trial designs, and data integration across sites. This approach seeks to address fundamental pharmacologic questions earlier and improve the probability of success in phase II trials, which dominates the cost of drug development.

1. http://dij.sagepub.com/
Therapeutic Innovation & Regulatory Science
http://dij.sagepub.com/content/48/5/546
The online version of this article can be found at:
DOI: 10.1177/2168479014526600
2014 48: 546 originally published online 19 March 2014Therapeutic Innovation & Regulatory Science
Oren Cohen and Frederic Sax
Building an Integrated Early Clinical Development Platform to Improve the Path to Proof of Concept
Published by:
http://www.sagepublications.com
On behalf of:
Drug Information Association
can be found at:Therapeutic Innovation & Regulatory ScienceAdditional services and information for
http://dij.sagepub.com/cgi/alertsEmail Alerts:
http://dij.sagepub.com/subscriptionsSubscriptions:
http://www.sagepub.com/journalsReprints.navReprints:
http://www.sagepub.com/journalsPermissions.navPermissions:
What is This?
- Mar 19, 2014OnlineFirst Version of Record
- Sep 10, 2014Version of Record>>
by guest on September 23, 2014dij.sagepub.comDownloaded from by guest on September 23, 2014dij.sagepub.comDownloaded from

2. Commentary
Building an Integrated Early Clinical
Development Platform to Improve
the Path to Proof of Concept
Oren Cohen, MD1
, and Frederic Sax, MD1
Abstract
Probability of success in phase II dominates the drug development cost calculus, with phase I/II as the critical juncture for proof of
concept. Failure to address fundamental pharmacologic questions in early development is alarmingly frequent and a strong
predictor of failure. Safety, manufacture, formulation, and commercialization issues are also vital. Systems biology provides a
framework to analyze genomic, proteomic, and metabolomic data and construct complex network models of molecular
pathophysiology. Biomarkers offer the largest learning opportunity, and combined adaptive protocol designs provide a lean but
scientifically robust path to proof of concept. The traditional model of phase I study execution in a clinical pharmacology unit is
evolving to a networked model of an integrated early clinical development platform. The power of this platform is enhanced with a
proactive multidisciplinary approach to quality and safety, including lean 6 sigma tools and simulations.
Keywords
proof of concept, early clinical development, phase I/II study, research and development productivity
Declining Productivity
Over the past decade, increasing research and development
(R&D) costs, longer development times, and diminished
R&D productivity have dominated the biopharmaceutical
industry landscape. These trends, coupled with patent expiries
and ongoing pricing pressures, are threatening the traditional
biopharmaceutical R&D model.
The most striking feature of declining productivity is the
decreasing rate of success at early phases of the drug develop-
ment process (Table 1). Prior to 2007, many biopharmaceutical
companies favored a ‘‘shots on goal’’ approach to address the
R&D productivity challenge.3
This strategy assumed static suc-
cess rates and encouraged entry of more drug candidates into
clinical trials under the simple quantitative hypothesis that
‘‘more in ¼ more out.’’ However, in the period between
2002 and 2007, there was no uptick in the number of drug
approvals despite a dramatic increase in the number of com-
pounds entering clinical development. Project teams across the
industry were incentivized to get drug candidates to the next
phase of development. The approach was costly and unsustain-
able. The unintended consequence of this strategy was that too
many R&D teams focused too heavily on cost and time lean-
ness and not enough on the fundamental pharmacologic charac-
teristics of drug candidates and the possibility that they would
have the safety and efficacy profiles required to succeed in the
marketplace.
The Importance of Proof of Concept
Portfolio modeling has shown that the cost of drug develop-
ment is most sensitive to the probability of success in phase
II.4
At Eli Lilly, where the capitalized cost of development is
about US$1.8 billion per drug launch, an increase in the prob-
ability of success in phase II from 34% to 50% would decrease
this cost by about $500 million. Conversely, a decrease in suc-
cess rate from 34% to 25% would increase the cost by about
$500 million.4
The fact that the probability of success in phase
II dominates the development cost calculus is consistent with
the importance of proof of concept (PoC). Phase I/II is the crit-
ical juncture in the drug development process in which PoC is
sought. A PhRMA position paper describes PoC as ‘‘the
1
Quintiles, Durham, NC, USA
Submitted 23-Oct-2013; accepted 06-Feb-2014
Corresponding Author:
Oren Cohen, Quintiles, 4820 Emperor Blvd, Durham, NC 27703, USA.
Email: oren.cohen@quintiles.com
Therapeutic Innovation
& Regulatory Science
2014, Vol. 48(5) 546-551
ª The Author(s) 2014
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/2168479014526600
tirs.sagepub.com
by guest on September 23, 2014dij.sagepub.comDownloaded from

3. earliest point in the drug development process at which the
weight of evidence suggests that it is ‘reasonably likely’ that
the key attributes for success are present and the key causes
of failure are absent.’’5
Sheiner’s classic characterization of
early clinical development as the ‘‘learn’’ phase of the ‘‘learn
and confirm’’ continuum of the drug development process is
particularly instructive in this regard.6
Early clinical develop-
ment (phases I and II) is the period during which drug develo-
pers have the opportunity to maximize learning about the
compound under study. A robust PoC will be achieved only
if the pharmacokinetic and pharmacodynamic properties of the
compound are clearly delineated. From a scientific perspective,
the following fundamental pharmacologic questions must be
answered during this period:
Does the compound have the right pharmacologic prop-
erties to allow for human use? Does it distribute to the
target space or tissue?
Does it have an appropriate therapeutic index for
human use?
Does it interact with its target in humans as anticipated?
Does this interaction result in a biological effect that
may reasonably be expected to have the intended clini-
cal impact?
Although these principles seem straightforward, failure to
adequately address these questions in early clinical develop-
ment is alarmingly frequent and is a strong predictor of failure.
An analysis of 44 phase II programs from Pfizer between 2005
and 2009 found that for many cases (n ¼ 12, 27%), there were
data showing only binding to target.7
There were no data
regarding downstream pharmacologic effects, and
pharmacokinetic-pharmacodynamic relationships were not
well established. Among these 12 compounds, all failed in
phase II. In the case of 18 other compounds (41%), there were
supportive data for either target binding and tissue exposure or
downstream pharmacologic effects—but not both. Among
these 18 compounds, 2 compounds survived into phase III. For
the remaining 14 compounds (32%) for which pharmacologic
fundamentals were established early on, 12 achieved PoC, and
8 advanced to phase III. Swinney and Anthony go on to suggest
that high attrition rates may be due in part to the persistently
high percentage of compounds entering clinical development
that were discovered with phenotypic assays rather than
target-based approaches.8
This could be consistent with the Pfi-
zer data, as PoC is generally more straightforward when the
molecular mechanism of action is known. Support for this con-
cept can also be found in the higher success rates among biolo-
gic compounds, where mechanisms are generally well
delineated, compared with small molecules.9
Improving the ‘‘Concept’’ in PoC
One fundamental lever in improving the path to PoC is the con-
cept itself. Our ability to discover and design therapeutic agents
that target a disease process depends heavily on our under-
standing of the underlying disease pathogenesis. For example,
cholesteryl ester transfer protein inhibitors are effective at low-
ering low-density lipoprotein cholesterol and raising high-
density lipoprotein cholesterol. The prediction that this drug
class would revolutionize the treatment of hypercholesterole-
mia and its complications has been shattered by high-profile
phase III failures, and it begs the question about whether the
simplistic notion of lowering ‘‘bad’’ cholesterol and raising
‘‘good’’ cholesterol is an incomplete, if not flawed, therapeutic
approach.10-12
A similar conundrum exists for antidiabetic
agents. Long regarded as the gold standard surrogate marker
for diabetes, hemoglobin A1c levels are embroiled in contro-
versy. In the case of rosiglitazone, the disconnect between its
potent ability to lower blood sugar (as reflected by hemoglobin
A1c levels) and the possible increase in cardiovascular events
raised the possibility of off-target effects13
and prompted the
FDA to require large-scale cardiovascular outcomes studies for
candidate antidiabetic drugs.14
Recent failures of antibody drugs directed against amyloid-
beta for Alzheimer disease present similar issues. In this case, it
is imperative to demonstrate that candidate drugs reach the tar-
get tissue (ie, central nervous system) in sufficient concentra-
tion to be able to exert the intended pharmacodynamic effect.
Such evidence may depend on sophisticated imaging and/or
cerebrospinal fluid sampling in early-phase studies. If antiamy-
loid drugs fail despite robust pharmacokinetic properties, then
the underlying hypothesis regarding the importance of amyloid
to the pathogenesis of Alzheimer disease would have to be
modified or even questioned. In December 2012, Lilly
announced that it planned a large new trial of solanezumab, its
treatment for Alzheimer disease, among those with mild symp-
toms, instead of seeking US approval of the product based on
prior trials in which the drug failed to help a wider group of
patients.15
The traditional paradigm of clinical research links clinical
observations to biochemical and molecular abnormalities that
can be targeted by drugs. Counterintuitive results (some of
which are highlighted above) demonstrate the weakness of this
paradigm—namely, that human cognition allows for a fairly
Table 1. Success rates across phases of drug development (%).
Development Stage 20041
20122
Phase I 81 64
Phase II 58 32
Phase III 57 60
Cohen and Sax 547
by guest on September 23, 2014dij.sagepub.comDownloaded from

4. limited assessment of possible cause-effect relationships. The
reality is that when a drug is introduced into a biological system,
there are often myriad effects other than the few that we happen to
be looking for based on our simplistic assumptions and hypoth-
eses. Systems biology provides a framework that recognizes this
complexity, allowing a systematic analysis of vast numbers of
data points from genomic, proteomic, and metabolomic
approaches.16,17
The large-scale generation and integration of
these data are increasingly allowing the construction of complex
network models that provide a new framework for understanding
the molecular basis of physiologic or pathophysiologic states.
Network-based drug discovery aims to harness this knowledge
toinvestigate and understandthe impactof candidatedrugs onthe
molecular networks that define these states.
Higher Learning: The Integrated Early
Clinical Development Platform
As our understanding of disease pathogenesis expands, so too
does our ability to target novel pathways and molecules. In the
mid-1980s, a half-dozen interleukins were known, providing
rich new targets for the treatment of inflammatory diseases.
Today, revolutionary therapies targeting IL-1 and IL-6 (as well
as TNF-alpha) are available even as we recognize the bluntness
of these ‘‘targeted’’ drugs. The list of interleukins alone has
expanded to 36, and there is considerable excitement about tar-
geting IL-23 and IL-17 in certain systemic inflammatory dis-
eases. One hundred years ago, most any disease involving
joints, ligaments, and tendons was called rheumatism. Today,
textbooks of rheumatologic diseases recognize scores of dis-
tinct clinical entities, yet the classification remains somewhat
crude and empiric. How many different diseases does rheuma-
toid arthritis (still crude even in its nomenclature) actually
encompass? Will differences in signaling pathway abnormal-
ities, such as the JAK-STAT pathway, describe distinct mole-
cular forms of the disease? Can dysregulated expression of
certain genes be silenced or corrected with small interfering
RNA targeted for delivery to the relevant tissue?
Maximizing learning during early clinical development
decreases the probability of advancing poor drug candidates
through the development process. This can be a far more effec-
tive means of saving cost and time compared with lean develop-
ment of too many poor drug candidates. Fortunately, the early
clinical development environment is changing rapidly in
response to scientific progress. The complexity of some areas,
such as inflammation and oncology—both of which involve
multiple pathways, disease types, therapeutic targets, and dosage
regimens—means that early clinical development is akin to
3-dimensional chess. The chess board must account for different
disease targets (eg, tumor types, immunologic mediators/recep-
tors, signaling pathways), subpopulations of patients (sometimes
indicating different disease types), treatment with different com-
binations of drugs, and different doses of these drugs. There is a
place for new mathematical approaches to address this complex-
ity, developing positive or negative predictive values to drive
decision making based on clinical data from small numbers of
participants. Such models can incorporate the sensitivity and
specificity of assays and the chance of type I or II decision errors.
An environment for deliberate and precise intervention and
observation (ie, clinical pharmacology units) remains critical.
Increasingly, however, maximizing learning requires connec-
tions outside the 4 walls of a clinical pharmacology unit. Critical
components of such an integrated early clinical development
platform are illustrated in Figure 1.
Access to patient populations, for example, is now often
required in early-phase studies of diabetes and rheumatologic
diseases. Biomarkers provide the largest opportunity for learn-
ing in early-phase clinical trials. Oncology serves as the best
example. In this regard, the classification of breast cancer was
historically based on traditional histology, size, and stage para-
meters. Today, these traditional parameters can be combined
with assessment of receptor expression and other genetic mar-
kers to yield a broad spectrum of subclasses of disease that we
clinically recognize as simply ‘‘breast cancer.’’ Indeed, com-
mercially available assays exist that are able to categorize
breast cancer into disease subclasses based on molecular anal-
ysis of multiple genes.18
The disease subclasses have distinct
prognoses and patterns of responsiveness to different che-
motherapeutic agents. The I-SPY (investigation of serial stud-
ies to predict your therapeutic response with imaging and
molecular analysis) studies are collaborative trials that utilize
imaging and molecular biomarkers, adaptive design, and bioin-
formatics to rapidly test candidate drugs for breast cancer.19
This integrated approach, which seeks to maximize learning
about candidate drugs in early development, has the potential
to radically decrease clinical development time.
Similar approaches will speed development of new thera-
pies in areas such as immunology and rheumatology, in which
better models of disease pathogenesis are revealing promising
new pharmacologic targets. Combined adaptive protocol
designs provide a lean but scientifically robust path to PoC.
In these studies, a multicenter approach can be used to encom-
pass single-and multiple-ascending dose cohorts in normal
healthy volunteers as well as PoC in patients. Such aggressive
multicenter study designs are most successful when data
‘‘noise’’ can be minimized. This requires cross-site planning
and training, a bioinformatics platform that can operate across
clinical sites, and use of highly objective endpoints (eg, bio-
marker assay performed on a common validated platform).
Predicting results with statistical and pharmacologic model-
ing and simulation can be a powerful tool for managing the
increasing complexity in the early development arena.20,21
The
548 Therapeutic Innovation Regulatory Science 48(5)
by guest on September 23, 2014dij.sagepub.comDownloaded from

5. combination of modeling and simulation software with
high-performance computing enables construction of in silico
models of physiology and disease states.22
Pharmacokinetic-
pharmacodynamic and exposure-response models, as well as
trial execution models, can yield important insights into study
design and decision making. The value and potential of these
tools are maximized when a project team incorporates the
expertise from biostatisticians, statistical programmers, clinical
pharmacologists, and therapeutic experts. Some of the specific
areas where modeling and simulation can be particularly useful
are listed in Table 2.
Sophisticated modeling and simulation tools are no longer
rate limiting with regard to computational time, and results can
often be available in near real time for decision making. In this
regard, we recently utilized analytics software and high-
performance cloud computing to simulate a clinical trial and
inform the design of a Bayesian adaptive trial design. Up to
1000 computational cores were implemented simultaneously,
reducing modeling and simulation computational time from 2
months to approximately 1 hour.
Specialist techniques are also increasingly required to study
pharmacodynamic properties of drugs in patients in early-phase
studies. Digital electrocardiography capabilities are necessary
to conduct thorough QT studies that are frequently required by
regulatory mandate. Spirometry, whole body plethysmography,
Table 2. Areas where modeling and simulation are especially useful.
Maximum tolerated dose—first in human
Between cohort modeling and simulation
Simulating multiple-dose exposures
Formulation/dosing design
Immediate release to extended release
Absorption nonlinearity
General vs Loading doses
Population pharmacokinetic/pharmacodynamic
Covariate-based dose adjustments (eg, weight)
Drug-drug interaction identification
Special populations
Predict renal / hepatic failure exposures
Safety studies
Thorough QT study to evaluate the drug’s potential to delay cardiac
repolarization (E14 and concentration-QT modeling)
Figure 1. The integrated early clinical development platform.
Cohen and Sax 549
by guest on September 23, 2014dij.sagepub.comDownloaded from

6. and bronchoscopy can maximize learning about drugs for asthma
or chronic obstructive pulmonary disease.23-25
Joint imaging,
synovial fluid analysis, and synovial biopsy represent the nascent
tools to more intelligently assess drugs that target rheumatoid
arthritis and other inflammatory arthropathies.26
The 21st-century clinical pharmacology unit is a complex
environment that often resembles a hospital or acute care clinic
rather than a dormitory. In this setting, where increasingly
complex studies are being conducted with drugs acting on
novel targets, the need to ensure participant safety and well-
being is greater than ever. Though rare, adverse outcomes
including death have occurred at academic and commercial
clinical pharmacology units27-29
and serve as reminders that all
stakeholders in the clinical research enterprise must always put
the interests of patients and volunteers first. Minimizing the
opportunity and probability for human error in early-phase
studies is an area that requires more discipline and attention.
A proactive approach to safety and quality requires a strong
quality assurance organization, attentiveness and discipline in
operations, and the fortitude to become a learning organization.
Important elements of a learning organization include empow-
erment of staff to question authority, encouragement to report
and learn from errors and ‘‘near misses,’’ and maintenance of
a culture in which a constructive approach to human error is
favored over blame and punishment. A multidisciplinary holis-
tic approach to study planning and preparation should include a
test of study comprehension for staff; a focused failure modes
and effects analysis of high-impact risk points within a study
protocol and proactive mitigation planning; and simulation and
‘‘dry runs,’’ especially where novel equipment or procedures
are being employed.
Conclusions
Going forward, it is essential to restore clinical pharmacology
thinking to a central role in early development. Achieving
higher levels of confidence in PoC will be enabled by harnes-
sing the capabilities of an integrated early clinical develop-
ment platform to address real-world problems. This
platform is composed of state-of-the-art clinical pharmacol-
ogy units; biomarker design, delivery, and analysis; interoper-
able data systems; modeling and simulation capabilities;
planning and design technologies and expertise; therapeutic
and clinical pharmacology expertise; and capabilities to per-
form specialized procedures and techniques. By providing a
more robust characterization of the pharmacologic properties
of early-phase assets—in a ‘‘back to basics’’ approach that
also makes full use of innovative tools and techniques—this
platform will drive sound decision making and result in
improved productivity and success rates.
Declaration of Conflicting Interests
Authors are employed by Quintiles.
Funding
The author(s) received no financial support for the research, author-
ship, and/or publication of this article.
References
1. Abrantes-Metz R, Adams C, Metz A. Pharmaceutical develop-
ment phases: a duration analysis. Washington, DC: Bureau of
Economics; 2004. Working paper 274. http://www.ftc.gov/be/
workpapers/wp274.pdf.
2. Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J.
Clinical development success rates for investigational drugs. Nat
Biotechnol. 2014;32:40-51.
3. Douglas FL, Narayanan VK, Mitchell L, Litan RE. The case for
entrepreneurship in RD in the pharmaceutical industry. Nat Rev
Drug Discov. 2010;9:683-689.
4. Paul SM, Mytelka DS, Dunwiddie CT, et al. How to improve
RD productivity: the pharmaceutical industry’s grand chal-
lenge. Nat Rev Drug Discov. 2010;9:203-214.
5. Cartwright ME, Cohen S, Fleishaker JC, et al. Proof of concept: a
PhRMA position paper with recommendations for best practice.
Clin Pharmacol Ther. 2010;87:278-285.
6. Sheiner LB. Learning versus confirming in clinical drug develop-
ment. Clin Pharmacol Ther. 1997;61:275-291.
7. Dethloff L.Strategies to reduce compound attrition in early devel-
opment. Paper presented at: CHI’s Accelerating Proof of Concept
Conference; October 5, 2011; Philadelphia, PA.
8. Swinney DC, Anthony J. How were medicines discovered? Nat
Rev Drug Discov. 2011;10:507-179.
9. DiMasi JA, Feldman L, Seckler A, Wilson A. Trends in risks
associated with new drug development: success rates for investi-
gational drugs. Clin Pharmacol Ther. 2010;87:272-277.
10. Sofat R, Hingorani AD, Smeeth L, et al. Separating the
mechanism-based and off-target actions of CETP-inhibitors using
CETP gene polymorphisms. Circulation. 2010;121:52-62.
11. Vergeer M, Bots ML, van Leuven SI, et al. Cholesteryl ester
transfer protein inhibitor torcetrapib and off-target toxicity: a
pooled analysis of the rating atherosclerotic disease change by
imaging with a new CETP inhibitor (RADIANCE) trials. Circula-
tion. 2008;118:2515-2122.
12. Hayes E. Dalcetrapib failure raises yet more questions on value of
tinkering with HDL cholesterol. Pink Sheet. 2012;74(20).
13. Matthews L, Berry A, Tersigni M, D’Acquisto F, Ianaro A, Ray D.
Thiazolidinediones are partial agonists for the glucocorticoid
receptor. Endocrinology. 2009;150:75-86.
14. FDA Center for Drug Evaluation and Research. Guidance for
industry: Diabetes mellitus: evaluating cardiovascular risk in new
antidiabetic therapies to treat type 2 diabetes. http://www.fda.gov/
downloads/Drugs/GuidanceComplianceRegulatoryInformation/
Guidances/ucm071627.pdf. Published December 2008.
15. Pierson R. Lilly Alzheimer’s drug gets second chance, plans new
trial. http://news.yahoo.com/lilly-alzheimers-drug-gets-second-
550 Therapeutic Innovation Regulatory Science 48(5)
by guest on September 23, 2014dij.sagepub.comDownloaded from

7. chance-plans-trial-152647164–finance.html. Published Decem-
ber 12, 2012.
16. Schadt EE, Friend SH, Shaywitz DA. A network view of disease
and compound screening. Nat Rev Drug Discov. 2009;8:286-295.
17. Hopkins AL. Network pharmacology: the next paradigm in drug
discovery. Nat Chem Biol. 2008;11:682-690.
18. Sotiriou C, Pusztai L. Gene expression signatures in breast cancer.
N Engl J Med. 2009;360:790-800.
19. Esserman LJ, Woodcock J. Accelerating identification and regu-
latory approval of investigational cancer drugs. J Am Med Assoc.
2011;306:2608.
20. O’Kelly M, Berry S, Walp D, Garrett A. Lifecycle modeling and
simulation part 1: current practice and future impact on healthcare
innovation and delivery. http://www.quintiles.com/Library/Arti-
cle.aspx?articleID¼5403. Published May 31, 2012.
21. O’Kelly M, Berry S, Walp D, Garrett A. Lifecycle modeling and
simulation part 2: applications in clinical development. http://
www.quintiles.com/Library/Article.aspx?articleID¼27728.
22. Claret L, Lu JF, Bruno R, Hsu CP, Hei YJ, Sun YN. Simulations
using a drug-disease modeling framework and phase II data pre-
dict phase III survival outcome in first-line non-small-cell lung
cancer. Clin Pharmacol Ther. 2012;92:631-634.
23. Borrill ZL, Houghton CM, Tal-Singer R, et al. The use of plethys-
mography and oscillometry to compare long-acting bronchodila-
tors in patients with COPD. Br J Clin Pharmacol. 2008;65:
244-252.
24. Diamant Z, Tarasevych S, Clarke GW. New and existing pharma-
cotherapeutic options for persistent asthma and COPD. Neth J
Med. 2011;69:486-499.
25. Diamant Z, Clarke GW, Pieterse H, Gispert J. Disease models of
chronic inflammatory airway disease: applications and require-
ments for clinical trials. Curr Opinion Pulm Med. 2014;20:37-45.
26. van Kuijk AWR, Gerlag DM, Vos K, et al. A prospective, rando-
mised, placebo-controlled study to identify biomarkers associated
with active treatment in psoriatic arthritis: effects of adalimumab
treatment on synovial tissue. Ann Rheum Dis. 2009;68:1303-1309.
27. Suntharalingam G, Perry MR, Ward S, et al. Cytokine storm in a
phase 1 trial of the anti-CD28 monoclonal antibody TGN1412. N
Engl J Med. 2006;355:1018-1028.
28. SoRelle R. Human gene therapy: science under fire. Circulation.
2000;101:e9023-e9024.
29. Savulescu J, Spriggs M. The Hexamethonium Asthma Study
and the death of a normal volunteer in research. J Med Ethics.
2002;28:3-4.
Cohen and Sax 551
by guest on September 23, 2014dij.sagepub.comDownloaded from