bridging the gap from inventions to innovations ...€¦ · bridge the – so far fairly understood...
TRANSCRIPT
Paper to be presented at
DRUID15, Rome, June 15-17, 2015
(Coorganized with LUISS)
Bridging the Gap from Inventions to Innovations - Increasing Success
Rates in Clinical TrialsAnne Assmus
University of PassauDepartment of Economics and Business
Carolin HaeusslerUniversity of Passau
Chair of Organization, Technology Management and [email protected]
AbstractTurning basic research discoveries into marketable products or technologies is a major challenge, especially in drugdevelopment. We conceptualize the necessary exploitation skills to go beyond the ??the use and development of thingsalready known? (Levinthal & March, 1993, 108) by discussing how basic and applied research skills need to interact inorder to successfully apply inventions to turn them into innovations. Using data on over 4500 clinical trials in thepharmaceutical industry, we find that the likelihood of success increases, when the investigators leading these trialshave a skill set that in terms of quantity and quality balances basic and applied science. Importantly, the positive impactof such "bridging" relies on individuals bridging the basic-applied science divide. Inter-individual splitting of basic andapplied skills in teams does not compensate for intra-individual bridging skills. Breadth of experience in terms of diseasefields reduces success in exploitation and negatively moderates bridging skills.
Jelcodes:O31,J24
な
Bridging the Gap from Inventions to Innovations - Increasing Success Rates in Clinical Trials
ABSTRACT
Turning basic research discoveries into marketable products or technologies is a major
challenge, especially in drug development. We conceptualize the necessary exploitation skills
to go beyond the “‘the use and development of things already known” (Levinthal & March,
1993, 108) by discussing how basic and applied research skills need to interact in order to
successfully apply inventions to turn them into innovations. Using data on over 4500 clinical
trials in the pharmaceutical industry, we find that the likelihood of success increases, when
the investigators leading these trials have a skill set that in terms of quantity and quality bal-
ances basic and applied science. Importantly, the positive impact of such "bridging" relies on
individuals bridging the basic-applied science divide. Inter-individual splitting of basic and
applied skills in teams does not compensate for intra-individual bridging skills. Breadth of
experience in terms of disease fields reduces success in exploitation and negatively moderates
bridging skills.
Keywords: Bridging, Knowledge Recombination, Pharmaceutical Industry, Clinical Trials
JEL Classification: O31, O32, J24
に
1. Introduction
"Translational gap" is a frequently used term to describe the challenge in turning basic scien-
tific discoveries into innovations. The extent to which scientific achievements have been ex-
ploited for practical applications lags far behind expectations, especially in the pharmaceutical
industry, which has spurred the interest of academia, industry and government alike in over-
coming the translational gap (Sung Ns and et al. 2003; Zerhouni 2007). Prior research has
focused on the translational gap between science discoveries and inventions (e.g, Jaffe and
Trajtenberg 1996; McMillan, Narin et al. 2000; Gittelman and Kogut 2003; Lim 2004) but
research on the gap to exploitation – development and testing stage – is lacking. This “transla-
tional gap” is particularly eminent in the pharmaceutical industry, where despite increased
efforts and investments into research and development (R&D) and major scientific advances,
the output of innovative drugs has been declining dramatically over the past years (Hu,
Schultz et al. 2007; Woolf 2008; Pammolli, Magazzini et al. 2011).
An important question is, whether a balanced skillset of individuals or teams in terms of
basic and applied science (“bridging skills”) improves success rates of exploitation projects.
There is an emerging debate in the sociological and economics of science literature on the
relationship between basic and applied science. Basic and applied science are, on one hand,
said to be in logical and normative conflict (Merton 1973; Gittelman and Kogut 2003); on the
other hand found to be complementary in certain instances (Stokes 1997) and applications
(Fleming and Sorenson 2004). Many inventions are derived from basic science discoveries
(McMillan, Narin et al. 2000), basic science skills increase the ability to absorb knowledge
from beyond firm boundaries (Cohen and Levinthal 1990) as well as support information
gathering and problem solving (e.g., (Zucker, Darby et al. 2001).
If basic science is indeed complementary to applied research, four important follow-up
questions emerge in terms of i) whether this applies also for exploitation, when turning inven-
tions into innovations, ii) whether the balance between basic science and applied science
ぬ
skills matter, iii) whether bridging skills interfere with breadth in experience in terms of ap-
plications and whether iv) basic and applied skills need to be intra-individually bundled or
can be split between scientists in a team.
First, we examine an area of development research that has received more limited scholar-
ly attention in comparison to research that analyzes the importance of basic science for explo-
ration output (i.e., invention). While moving from basic science to inventions is generally
seen as an explorative task of learning and creating new knowledge, moving from inventions
to marketable products represents exploitation of the already existing knowledge base (Gupta,
Smith et al. 2006; Liu 2006). Why is the "exploitative" setting interesting? The capabilities
needed for exploitative research are predicted to differ from explorative research (March
1991). Baum, Li, and Usher conceptualize exploitation to refer “to learning gained via local
search, experiential refinement, and selection and reuse of existing routines” (Baum, Li et al.
2000). Brenner and Tushman (2002, 679) highlight that “exploitative innovations involve
improvements in existing components and build on the existing technological trajectory,
whereas exploratory innovation involves as shift to a different technological trajectory”
(Benner and Tushman 2002). While the dominant view in prior literature has stressed that
experience in the relevant appliances is the dominant success factor in exploitation (Levinthal
and March 1993), increasingly voices have raised to reconsider the dominance of applied re-
search in favor of a broader base of insight (FitzGerald 2005). Particularly the term
“translational medicine” born by practitioners in the pharmaceutical industry point to the
importance of bridging skills in exploitation, though we lack empirical insights. More than in
exploration, attributing importance to and applying bridging skills might be particularly
challenging in exploitation, where appropriation by exclusive use and commercial rewards are
the guiding principles. On the other hand the “translational” aspect also illustrates that basic
research embedded in an instutional logic where recognition is the sole property right also is
an important input factor. We aim to address what enhances success under such circumstances
ね
by quantifying the relative importance of either as the input, meaning where the emphasis is
placed between the two institutional logics and their incentive systems.
Second, differentiating between having some knowledge in basic science versus having
profound knowledge or being a star basic scientist might matter. Obviously highly-
accomplished scientists are important, for example, in terms of improving the quality of new
(joiner) scientists (Agrawal, McHale et al. 2013), increasing citations when joining a depart-
ment (Azoulay, Zivin et al. 2011) or providing a quality signal to investors (Zucker and Darby
1998). However, the stars productivity in translating discoveries into inventions has been
questioned. Prior studies report that scientists engaged in highly cited academic research pro-
duce less valuable patents (Gittelman and Kogut 2003; Toole and Czarnitzki 2009) and show
a lower likelihood of patenting (Calderini, Franzoni et al. 2007). A strong orientation towards
academic science might limit its transferability or applicability in a commercial context. This
might well apply to exploiting activities where both institutional logics are present.
Third, we combine the concept of bridging the basic and applied research divide with the
level of breadth in experience on productivity in innovation research (Boh, Evaristo et al.
2014). Breadth could be seen as another form of (horizontally) bridging different fields,
which would point to a general capability of recombining knowledge domains, however, it
might overstrain the scientists’ capacity when combined with (vertical) bridging skills be-
tween basic and applied science.
Lastly, with more and more large teams are becoming the dominant form of organizing
R&D (Wuchty, Jones et al. 2007; Singh and Fleming 2010); the question arises of whether
bridging skills need to be “within” an individual capable of such activities or whether basic
and applied skills can be distributed among team members. The economics of science litera-
ture has highlighted that teams enable to assemble different pieces of knowledge from indi-
viduals (Jones 2009; Singh and Fleming 2010; Bercovitz and Feldman 2011; Jones and
Weinberg 2011) and allow division of labor between team members (Häussler and Sauermann
の
2014). However, we conceptualize bridging skills to rely on single individuals who inhibit
both basic and applied research skills, which cannot be substituted by a split of basic and ap-
plied skills between team members.
We test our hypotheses by using a dataset of more than 4500 clinical trials, which have
been conducted between the year 2005 and 2012. During clinical trials, a patented invention is
tested in its final setting (“the human”) for the first time and in a stepwise process. Clinical
trials constitute an excellent setting for this kind of research on translational science, and rep-
resent an area of major economic relevance facing major hurdles in terms of costs, trial
lengths and high failure rates (DiMasi, Feldman et al. 2010; Pammolli, Magazzini et al.
2011). The term “translational medicine” highlights the difficult transition between the basic
(preclinical) and the clinical stages in the pharmaceutical R&D process. Academia, industry
and government jointly increased initiatives to overcome this gap. In our study, we relate the
skillset of the principal investigators (i.e., their publication activity in terms of basic and ap-
plied science as well as their experience) to the success (i.e., the successful completion of the
trial). While prior literature has mostly interpreted publication activity as basic science, we
exploit the heterogeneity of publications in terms of basic and applied science and thus can
investigate the concept of basic and applied being distinct or along a continuum.
Our findings highlight the importance of human capital bridging the basic and applied sci-
ence divide for the translation of inventions to innovations. For individuals, a balance skillset
is most optimal for taking advantage of the combination of capabilities and knowledge of ei-
ther science as well as to cope with the idiosyncrasies of the academic and commercial incen-
tives. Also within teams, such individuals confer superior performance. In general, excellence
in basic and applied science is not beneficial. Breadth of experience in terms of various dis-
eases lowers success, and negatively moderates bridging.
Our insights contribute to a growing literature on human capital and the production of in-
novation (Fleming and Sorenson 2003; Toole and Czarnitzki 2009). Extending prior literature
は
arguing that the relationship between research and innovation is complex (Gittelman and
Kogut 2003), we add that individuals who are skilled in basic and applied science are able to
bridge the – so far fairly understood – gap from the invention to testing stage. Even in the
exploitation phase, where creativity and the linkage to basic science are proposed to be much
less prominent (Rothaermel and Deeds 2004), bridging scientists with an understanding of
both basic and applied science enhance success. However, in correspondence with (Gittelman
and Kogut 2003) who find that the ability to produce excellent science is detrimental to inno-
vative success, our findings indicate that the right dose of bridging is important as scientists
being too much drawn into the academia are sub-optimal. Hence, there is a problematic dis-
connect between excellence in basic science and in applied science which can only be over-
come by balancing the involvement in the two institutional spheres both in terms of quality
and quantity.
By disentangling intra- versus inter-individual bridging in teams, we also contribute to the
literature on team science (Wuchty, Jones et al. 2007). While prior research has shown that
knowledge in various domains can be distributed among team members (Jones 2009), bridg-
ing is only superior when it is inhibited within individuals and not split between two or more
specialized team members. This finding is novel as it speaks to the limits of specialization.
2. Conceptual Framework
Basic and Applied Research
Basic research has the goal of increasing the scientific knowledge base, which is generally not
appropriable, or "pursuing knowledge for its own sake" as often done in academia (Dasgupta
and David 1994), from which industry can source insights for innovations (Zucker, Darby et
al. 2001; Henderson, Jaffe et al. 2005). Basic research generates a public good, whose appro-
priation is difficult, which’s returns can only be expected in the long run and indirect ways,
and thus requires public investments. Still, public investments into academic and basic sci-
ence ultimately have the raison d'être that they are believed to lead different actors to take
ば
such discoveries on and apply them to generate a benefit for society (Gibbons and Johnston
1974; Pavitt 1991). In contrast, applied science is focusing on specific challenges, whose so-
lutions may represent a product, technology or service that can be commercialized and gener-
ate financial rewards.
There is a long-standing debate on the relationship between basic and applied science. One
strand of proponents argues in favor of a divide between basic and applied as polar phenome-
na that are diametrical points on a line or as two opposing ends of a continuum, (Bush 1945;
Brandl 1998). Bush (1945) states that basic and applied research follow different logics and
thus excellence in one will inhibit the other. The conflicting logic is attributed to different
epistemological communities in which scientists in basic science and commercial applications
are embedded (Dasgupta and David 1994). On the other hand, Stokes in his book titled "Pas-
teur's Quadrant" argues for a different view. He proposes the existence of a third kind of re-
search that obtains both the goal of valuable insights for application (applied science) as well
as increasing the knowledge base (basic science), besides the established purely basic and
purely applied research endeavors (Stokes 1997).
Innovation research also suggests that basic science drives much of the innovative perfor-
mance of firms, industries and nations (Rosenberg 1990; Pavitt 1991; Zucker and Darby
1995; Zucker, Darby et al. 1998). Various mechanisms have been outlined by which basic
science impacts industrial innovation. Many inventions are derived directly from basic sci-
ence discoveries, which is documented by patents citing scientific publications (McMillan,
Narin et al. 2000). For example in life-sciences, firms that publish and developed sophisticat-
ed links to academics achieve greater levels of success (Henderson and Cockburn 1994;
Owen-Smith and Powell 2004). Firms support basic research activities of their scientists in
order to stay well connected to the community and thereby understand and be able to absorb
knowledge from beyond the firm boundaries (Cohen and Levinthal 1990). Human capital well
skilled in basic research support information gathering and problem solving in firms (Zucker
ぱ
and Darby 1996; Zucker, Darby et al. 2001). Rosenberg ascribes basic science a mediating
role in the translation of discoveries (Rosenberg 1990). Basic science research capabilities are
essential for evaluating the outcome of much applied research and for perceiving its possible
implications and can provide valuable guidance to the directions in which there is a high
probability of payoffs to more applied research (Rosenberg 1990). Fleming and Sorenson
postulate that for coupled inventions, like in pharmaceuticals, basic science is necessary to
understand where a specific configuration of components may achieve the highest benefit or
which may be most useful and offer the highest potential (Fleming and Sorenson 2004). Fur-
ther, Cockburn and Henderson (1994, 1998) find that rewards of basic and applied research
act as complements to increase productivity of scientists when analyzing incentive systems
inside pharmaceutical firms.
While prior research has emphasized the importance of both skills in basic as well as in
applied science to turn discoveries into inventions, we examine the area of the exploitation
stage that has received more limited attention. Specifically in the setting of clinical trials, in-
vestigators with skills in basic, bench-side research, and applied research may be better at
understanding the potential, in designing experiments and trials to test hypotheses efficiently
and combine methodologies to create valuable outcomes. Being active in the basic science
paradigm should create incentives for the clinicians to report on such findings as they may be
publishable, even though clinicians generally are not incentivized to report or act on such
findings (DeMonaco, Ali et al. 2006). Thus, to successfully address the translational block
from basic science to improved health, "an interdisciplinary array of clinical investigators" is
essential (Sung Ns and et al. 2003). Hence, we:
H1: Bridging skills, i.e., skills in both basic and applied research, increase success in de-
velopment projects.
Excellence in basic and applied research
ひ
Though, there are compelling reasons for the complementary view of basic and applied re-
search, scientists are still confronted with the conflicting tensions between the basic and the
applied research paradigms. Bridging both sciences may represent a challenge that needs a
very specific skill set. Being too much drawn in one of the two spheres is lowering the skills
to bridge exploitation.
Gittelman and Kogut (2003) find that patents of scientists with highly impact publications
receive fewer patent citations, but concomitantly find investigators with experience in both to
increase innovation outputs. Toole and Czarnitzki (2009) also find that there is a tradeoff be-
tween basic research productivity and a firm’s patent productivity. They differentiate scientif-
ically oriented human capital from commercial oriented human capital gained by firm bio-
scientists during their careers in academe before they joined or founded a firm. Their results
suggest that depth of scientifically oriented human capital is negatively related to a firm pa-
tent productivity but positively to the firm attracting an NIH grant after completion of proof-
of-concept studies.
Hence, we conceptualize bridging skills not only to have a skillset in basic as well as in
applied research but also to carefully balance researchers’ embeddedness in these two
spheres.
H2a: Bridging skills are most effective for the success of development projects if skills
in basic and applied are balanced. (Quantity)
Furthermore, the concept of stars has received increasingly attention, though with conflict-
ing empirical results. While some proponents claim that star scientists are much more prolific,
produce better innovations and have a large impact on the valuation of firms that they advise
or found (Zucker and Darby 1995; Darby, Liu et al. 1999; Zucker, Darby et al. 2001;
Agrawal, McHale et al. 2013), others have voiced concerns of overemphasizing the role of
starts. Rothaermel and Hess find that it is not the intellectual human capital of star scientists
that increases innovative output of a firm, but that of the non-star scientists and though aca-
など
demic scientists may be high-performer in terms of publications and their impact, such scien-
tists produce less innovations (Calderini, Franzoni et al. 2007). Too strong a focus on the aca-
demic career and basic science may thus result in inferior translational capabilities due to a
different motivational or basic science-driven value focus. Bridging scientists that exhibit
excellence in both sciences, accordingly, may decrease the success of their product develop-
ment efforts as they are drawn too much towards the basic science paradigm.
H2b: Excellence in both applied and basic research tilts the balance towards too much
engagement in the academic science community and thus decreases success rates of
translational research projects. (Quality)
2.3 Breadth of Experience and Bridging
While we follow a vertical concept of bridging between basic and applied research, a hori-
zontal concept of bridging would view experience in different research areas as a capability to
linking fields. The latter is more a general capability of recombining knowledge but one,
which might interact with our (vertical) concept of bridging skills. Broad set of experience or
“breadth” is the diversity of experiences and insights that have been accumulated (Fleming,
Mingo et al. 2007). Some scholars argue that creativity requires a recombination of
knowledge, which is best achieved through individuals with broad insights into different
fields (Katila and Ahuja 2002). Too much focus may lead individuals to be trapped within
their thinking and their held beliefs about the matter at hand. Others claim that only specializ-
ing in one or few fields enables to contribute at the knowledge frontier (Sternberg and O'Hara
1999). Jones (2009) propagates the “death of the Renaissance man” as researchers are becom-
ing more and more experts in increasingly narrow knowledge domains as an answer to the
increasing “burden of knowledge”. Researchers require longer learning periods before making
contributions and therefore must specialize.
Further, a strong focus enables experts to search locally to determine what beliefs or rules
to break, whereas less depth of experience may lead to conformist behavior in terms of beliefs
なな
or rules (Weisberg 1999). Studying the company 3M, Boh et al. (2014) report that having in-
depth experience in one field leads to more valuable inventions, while breadth of experience
is associated with higher numbers of inventions, but less influential ones. Macher and Boerner
(2006) in their study of pharmaceutical Contract Research Organizations find that scope in
terms of different technological areas reduces the success rates of exploitation projects when
not paired with in-depth knowledge in a specific area. They recommend that firms should
prioritize development activities.
Breadth in terms of experience in different fields might be advantageous only if recombi-
nation matters and if these benefits exceed the costs of sharing attention to various areas in-
stead of concentrating on one. It is likely challenging to use the experience gained in diverse
fields to excel at exploitative research in a specific field. This might even be more severe in
projects with translational gaps. Because bridging the gap in a specific field is already highly
specific and drawing on a high level of tacit knowledge, it may even be constrained when
researchers are, additionally, active in various fields. Rosenberg reasons that much technolog-
ical knowledge in a field deals with "the specific and the particular” (Rosenberg 1978). Indi-
viduals who are expected to master the gap as well as mastering and sorting impressions and
experiences gained in very diverse applied fields might be overburdened. In translational are-
as, more vertical focused problem-solving strategies may be favored compared to horizontally
broad field knowledge. Bridging scientists are per definition knowledgeable in basic science
which is more general per nature than applied knowledge. Being also broad in terms of the
applied research/ development fields does not effectively and efficiently guide the investiga-
tors in their endeavor. The level of complexity is already too high in order to cope with shar-
ing attention with various fields and being able to make use of the benefit of bridging. The
phenomenon is seen at clinics and research hospitals: while investigators may establish their
own laboratories in which to research the molecular biology behind the disease that they try to
cure, such physicians have to have gained the accreditation for the specialization through
なに
years of training in the disease. Therefore,bridging skills are leveraged if the investigator ex-
perience is focused on one or few fields.
H3: Breadth of experience in applied research fields negatively moderates the effect
of bridging skills on success in exploitation projects.
2.4.Intra- versus inter-individual Bridging
When analyzing the impact of bridging the basic and applied science divide, the extent to
which bridging skills are intra-individually or can be inter-individually split between team
members is important. In general, an increasing amount of knowledge production is per-
formed by teams rather than individuals (Wuchty, Jones et al. 2007; Singh and Fleming
2010). Much of prior studies devote this to the notion that knowledge production involves the
recombination of different knowledge domains which can be achieved by division of labor
among researchers specializing in different fields and recombining their work in teams (Singh
& Fleming, 2010; (Singh and Fleming 2010; Bercovitz and Feldman 2011). On the other
hand, individuals have been shown to be more effective in integrating knowledge diversity
than teams (Taylor and Greve 2006). In contrast to this strand of research, our focus is on
bridging between scientific spheres – which often appears vertically. The translational capa-
bility is likely to have limits in being articulated and codified within teams but rather to be
“sticking” within certain individuals (DeMonaco, Ali et al. 2006). Polanyi (1958, p. 49) states
that "the aim of a skillful performance is achieved by the observance of a set of rules which
are not known as such to the person following them" (Polanyi 1958). Problem solving pro-
cesses and learning that lead to innovation result when different knowledge structures coexist
in the same mind, when at least one head can fit most of the relevant pieces of knowledge
together (Simon 1985). We thus expect individuals to be able to bridge the basic to applied
science paradigms more effectively than teams in which a part of the team members is fo-
cused on basic and the other part on applied research.
なぬ
H4: Intra-Individual bridging mediates the beneficial effect of bridging on the transla-
tion of inventions, even within teams, not the inter-individual bridging.
3. Clinical Research Setting
3.1. From Bench to Bedside
We test our hypotheses with a dataset on clinical trials. The development of new medical in-
terventions to improve health and lives of patients is a complex, costly and lengthy process
with high failure rates. In fact, this process is associated with substantial costs (DiMasi 1991;
DiMasi 2001; DiMasi, Hansen et al. 2003) and the majority of R&D funding of firms is allo-
cated to this process (NSB 2004). Empirical evidence suggests that most of industrial R&D is
exploitative in character with approximately 75% allocated to development activities, 20%
committed to applied research and only 5% targeting basic research (NSB 2004). Especially
in drug development, the attrition rates of novel compounds from discovery to market are
higher than in any other industry, exceeding 90% in the new chemical entity space (Kola and
Landis 2004) and 80% in the biologics space (DiMasi and Grabowski 2007); the time it takes
for a substance from discovery to market is and costs have risen over decades to over
USD800M per marketed drug and more (DiMasi, Hansen et al. 2003; DiMasi, Feldman et al.
2010; DiMasi 2014). Beside its importance in terms of economic impact, the clinical trial set-
ting enables us to analyze the exploitation of inventions or the transfer from an invention into
a product (March 1991; Nijstad and De Dreu 2002; Hoang and Rothaermel 2010). Once a
new substance has been “discovered” at the bench-side, a second set of development-oriented
activities involving bed-side activities begin with the clinical stages. The translation from
bench to bedside is particularly interesting and important with empirical ambiguity in terms of
how best to organize this translation. Early clinical trials are the first time that a treatment
drug is tested on human subjects and thus this form of research differs substantially from oth-
er forms of research. It is strictly regulated and requires approval by different authorities, be-
fore its initiation and is usually conducted at hospitals and medical centers. Developing a new
なね
drug or novel drug combinations to market is a well-defined stepwise process that evolved
over decades as the different stakeholders, among them pharmaceutical firms and regulators,
defined a process that would ensure safety, while maximizing output of novel therapeutics to
achieve commercial goals and benefits for patients.
Principal Investigators of Clinical Trials
Clinical research is conducted at research hospitals and led by principal investigators (PIs)
who are either employed by these hospitals, public funded institutions or research centers. A
principal investigator is responsible for ensuring that an investigation is conducted according
to the signed investigator statement, the investigational plan, and applicable regulations for
protecting the rights, safety, and welfare of subjects under the investigator's care as well as for
the control of drugs under investigation. The investigator has the close oversight over the “ac-
tual clinical research situation on the ground” and is, as part of the clinical management team,
responsible for the “scientific and ethical integrity of clinical trials” (Hoekman, Frenken et al.
2012), p.2). Recently, pharmaceutical firms are criticized for having their clinical trial per-
formed without any further input or insight from investigators, as if it were routine work
(Rasmussen 2005). But, as failures in clinical trials for pharmaceutical firms have been pub-
lished to derive from slow patient recruitment, lack of feasibility of the study protocol, too
high incidences of adverse events among other challenges (Buonansegna, Salomo et al. 2012;
Buonansegna, Salomo et al. 2014), investigators as the responsible doctors applying the drug
to patients and monitoring effects should be the closest to being able to take activities to pre-
vent failures. Investigators of clinical trials should be in a key position “to test biologic hy-
potheses in living patients” and “have the potential to change the standards of care”
(Davidoff, DeAngelis et al. 2001), p.786).
4. Dataset, Measures and Methodology
Dataset
なの
We use the ClinicalTrials.gov database, a comprehensive registry of clinical trials maintained
by the U.S. National Library of Medicine at the National Institutes of Health to obtain de-
tailed information on clinical studies conducted in the U.S. and over 180 other countries.
Mandatory registration of clinical trials has been extended since the establishment of the da-
tabase in the year 2000. Since 2002 sponsors or principal investigators have to register studies
addressing life-threatening diseases in addition to trials funded by the U.S. government before
the enrollment of the first subject. Today, each clinical trial of drugs or biological products
has to be registered if at least one trial site is located in the U.S., drug candidates are manufac-
tured in the U.S., or the trial is conducted under an FDA investigational new drug application.
The data is publicly available and has already been downloaded. Our final dataset covers over
4500 clinical trials conducted between 2005 and 2012 for which we have the names of inves-
tigators and with at least one investigator having published in a peer reviewed journal.1 We
merged information about the publication activity of the PIs on the trials based on a match
with last and first names from the database Scopus. Major work in data preparation has been
done on trial and publication categorization and matching of drug synonyms. Trial indications
have been categorized according the WHO ICD10 system. Each single publication of the PIs
has been classified according to the CHI system (see also (Della Malva, Leten et al. 2013)).
Measures
Dependent Variable
We follow Danzon et al. (2005) and measure success in a given phase as a binary variable
whether the drug, treatment or intervention is advanced to the next phase of testing in the
same indication or not (Danzon, Nicholson et al. 2005). A trial advances to the next phase of
testing if the data generated met the specifications in terms of scope or completeness, and if
the quantitative test results warrant the investment into the next phase of testing. Since drugs
1 Information on the full clinical trials.gov dataset and of our final sample is provided in the Appendix 1.
なは
are often termed in many different ways during the stages, and clinicaltrials.gov does not give
clear guidelines on how information should be provided, we first had to match drug syno-
nyms. To classify the indication of the trials, we made use of the International Classification
of Diseases (ICD) established by the WHO, which we also needed for computing independent
variables on the disease level. With the profound biology background of one co-author, we
were able to design a drug synonym matching and disease and disease fields classification
scheme, which we also advanced with the help of 5 biology and medical experts. Appendix 1
describes the classification and matching scheme as well as the extensive manual coding. This
procedure allows us to identify all trials of a specific drug and their specific disease and dis-
ease field. This information enables us to code trials which show a follow on test phase as
success. We are only able to code a trial as successful, if a follow-on trial is feasible. Hence, we ex-
cluded all phase 3 and phase 4 trials which are most often the final phases of drug testing.2
Independent Variables
Basic, applied and bridging PIs
Publications of investigators are categorized into basic, applied or bridging researchers ac-
cording to their publication record. Our ability to differentiate publications based on the re-
search focus of the journal(s) they publish, whether basic or applied, allows us to control for
different functionality of the underlying research result. While our category "basic research"
represents publications which tend to be "general" in nature (Pavitt 1991) and not appropria-
ble (Rosenberg 1990), publications within the "applied" journal category may be either of
applied research nature, specific to a certain setting, or even intended to position an innova-
tion in the market place (Polidoro and Theeke 2012).
We use the CHI Journal classification system to categorize the PI’s publications
2 The coding details are provided in the Appendix A2.
なば
(Della Malva, Leten et al. 2013). CHI categorizes journals into 5 levels ranging from 0
(very applied) to 4 (very basic). We code publications in journals from level 0 to 2 as applied
and from level 3 and 4 as basic. The cut off points were selected after we presented a sample
of 10 journals to 3 researchers and asked them to attribute them to the basic or applied catego-
ry. For example, Nature as a very strong molecular and mostly basic science journal is cate-
gorized to level 4, as well as Biological Chemistry, whereas the New England Journal of
Medicine belongs to category 2, which our experts denote as an applied research journal. Fur-
ther applied journals are for example Current Opinion in Cardiology or American Surgeon.
Citation rates are considerably higher for level 4 journal publications, with the other levels not
deviating much from each other, as seen in the Appendix 2. We use two approaches to code
the PIs’ scientific orientation, the first highlighting the distinct nature of basic and applied
science and the second following a continuum view.
First, we code three binary variables: the variable d_bridging is one, if the PIs have at least
one publication in the basic applied and in the applied journal category. The binary variable
d_basicpubs_only is 1 if the PIs have only basic research publications and respectively
d_appliedpubs_only is 1 for PIs with only applied research publications.
Second, as an alternative coding, we viewed the basic and applied orientation along a con-
tinuum. In doing so, we coded the binary variable d_balancebridging as being 1, if the PIs
had between 25% and 75% of publications in the basic science journals, 0 otherwise. Hence,
whereas a PI with 1 out of 10 publications would be viewed as bridging researcher in terms of
d_bridging being 1, the same PI would not be classified as “bridging” in terms of the continu-
um view (i.e., d_balancebridging is 0). In addition, we compute the fraction of basic science
publications (share_basic and its square share_basic_squ) to further analyze the quantitative
relationship between bridging and success.
Further, we take excellence in bridging into account and code the binary variable
d_topcitbridging as being 1 if PIs have publications among the top decile cited publications in
なぱ
basic and in an applied journal. We classified a PI or investigator team as in the top decile
according to the 5 year citation rate to their publications. In correspondence, we coded the
binary variable d_topcit_basic for PIs who are among the top 10% of cited PIs for their basic
publications.
Inter- vs intra-individual bridging
While the former defined variables relate to the publications of an individual PI, we also
take into account that trials list a team of PIs. For the hypothesis testing on whether bridging
between basic and applied research needs to be intra-individually versus inter-individually,
i.e., split between members of a team of PIs, we also coded team based variables. We code the
binary variable d_basicPIs_only as being 1 for teams with only PIs listed who publish in basic
scientific journals. Correspondingly, d_appliedPIs_only is coded 1 for teams with only ap-
plied investigators. Further binary variables denote whether they had at least one bridging
investigator on a trial (d_mixofPIsinbridging), whether the team only had bridging investiga-
tors (d_bridgingPIs_only) or whether bridging is achieved through the team by including
basic science investigators and applied science investigator on the team
(d_bridgingthroughteam) but without having a bridging PI.
Breadth of experience
We measure the level of breadth of experience by counting the number of different disease
fields (according to the ICD chapter, e.g., “cancer”) in which the investigator has been listed
in the 5 years prior to the focal field. The variable is denoted ln_fielddivexp. Given the skew-
ness of the experience measures, we use the logarithmic form (1+log).
Control Variables
Several variables are included which are known or expected to influence the success and du-
ration of clinical trials and will allow us to isolate the bridging effect on trial success.
We control for the experience of the investigators in the disease of the focal trial. The vari-
able ln_disexp is measured as the logarithmic form (1+log) of the accumulated number of
なひ
trials all investigators conducted over a time span of five years prior to the trial in the given
disease field such as cancer (according to the ICD chapter).
We include the variable no_investigators to measure the number of PIs on a trial as well as
the accumulated number of publications in the 5 years prior to the trial of all investigators on
a trial (no_pubs).
We also control for institutional differences by including the binary variable d_industry for
the type of institution to differentiate from university or non-profit sponsors of the trials.
Since big pharma firms might act differently, we code the variable d_bigpharma being 1, if a
sponsor belongs to the top 25 pharmaceutical companies according to their sales in 2012. We
control for the experience of the sponsor in clinical trials by sponsor_trialsexp which depicts
the number of trials the sponsor is listed on the 5 years prior to the focal trial and by including
sponsor_disexp to control for the experience in a given disease 5 years prior. We also include
how many trials the drug has been tested in in the same phase priordrugtrials_samephase.
Our notion is that it is likely that if a sponsor has conducted already several trials in the same
phase, it is more experiences than if it is the first trial in a specific phase. As already outlined
above, trial success and duration might differ in regards to the phase of testing. Hence, we
include binary variables for the phase (d_phase_0, d_phase_1, d_phase2, d_phase_12,
d_phase23). The reference class is d_phase_1. As a control for size, the variable enrollment
counts the number of patients enrolled. We code the number of different sites which are re-
cruiting for a given trial location_count. We also control whether the site is one of the 10% of
most prominent sites in the overall trials database (top10site) and whether the trial is conduct-
ed in the USA versus non-USA. Furthermore, prior studies report that success rates differ sig-
nificantly between disease fields. We use the ICD 10 coding system of the WHO to code bi-
nary variables for the most prominent disease fields such as cancer, endocrinedisease, men-
taldisease, circulatorydisease, infectiousdisease. All other disease areas are coded as
otherdisease. Endocrinedisease is used as the reference group in our econometric models.
にど
Descriptive Statistics
Descriptive statistics of the dependent, independent and control variables are reported in
Table 1.3 We first show the descriptive statistics for all observations in Column 2 and then
separate for trials with a single PI (Column 3) and trials with a team of investigators (Column
4). 51% of our trials are led by a single PI. In the following, we briefly comment for the full
sample and for the sake of brevity only mention specifics for the sample split.
Our dependent variable success shows a mean of 43% for the full sample which corre-
sponds with prior findings reporting a success rate of 70% for phase 1 and 30% for phase 2
trials, whereby phase 2 trials account for the majority of trials in our sample (Kola and Landis
2004). The success rate of single led trials is with 45% slightly higher than that of teams led
trials being 40%.
52% of the trials in the overall sample are led by (single or a team of) investigators which
show bridging skills (d_bridging), i.e., have published in a basic and applied journal, com-
pared to 30% being led by pure basic scientists (d_basicpubs_only). The remaining 18% work
in the applied field. 27% are led by well-balanced engaged investigators, i.e., who publish
between 25 and 75% of their output in basic journals (d_balancedbridging). The average share
of basic research articles is 61% (share_basic). Though we have a large number of bridging
investigators, only 7% achieve excellence in both fields, i.e., they are among the 10% of most
cited investigators in terms of their publications in basic as well as applied journals
(d_top_citbridging). Interestingly, 11% of teams with 2 or more PI’s show a top bridging in-
vestigators compared to a share of 7% for trials led by one investigator. Top bridging scien-
tists might prefer to work in a team when they participate as investigator in a clinical trial
Among the sample of team led trials, the majority of teams consist of all members being
bridging investigators (57%, d_bridgingPIs_only), 39% have some bridging PIs among the
3 Correlation tables for all dependent, independent and control variables are provided in the Appendix.
にな
team members (d_mixofPIsinbridging), only 1% of teams consists of pure basic as well as
pure applied investigators (d_bridgingthroughteam), and the remaining 3% consist of only
pure basic or only pure applied investigators.
We include two types of measures for the experience of PIs: The average clinical trial has
been led by a single or team of investigators which had already conducted 0.69 trials in the
same disease in the previous five years (disexp). Not surprisingly, this number is highly
skewed and higher for team led trials. Similarly, our measure of breadth suggests that the av-
erage trial is led by PIs who have an experience in 1.22 different diseases – with this variable
ranging from 0 experience up to experience in 56 different disease fields driven by team trials.
The average trial lists 2.51 investigators, whereby the average team lead trial has 4.09
members. The average trial is conducted by PIs who have published 11.6 publications in peer-
reviewed journals in the 5 years prior to the trial.
36% of trials are sponsored by industry and 5% by a Big Pharma firm. The average spon-
sor has initiated 326 trials and 58.35 trials in the same disease field in the 5 years prior. The
average trial has enrolled 89 patients. The trials are conducted on average at 4.3 sites, which
is about the same as within the team and the individual-led trials and shows that it is rarely
possible to recruit sufficient patient number at one site. 60% of trials have at least one site in
the US, which is the most important market for innovative pharmaceuticals. 9% of trials take
place at a top site. Most trials are in test phase 2 (46%) followed by phase 1 (33%). On aver-
age the drug has been tested in the same phase (but mostly in different indication) in 15.6 tri-
als before. Most drugs are tested for several different indications (diseases) and if successful,
tested in many different specific disease settings and under different formulations. Interesting-
ly, the team led trials show a lower number of previous trials with the same drugs (13 times)
than individual led trials (18 times). Note that we did not include phase 3 or 4 since they usu-
ally do not have a follow on test phase. The largest share of tests is conducted in cancer
(43%), followed by infectious diseases (7.9%), endocrines (7.5%), mental (6.5%).
にに
*** Table 1 about here ***
Econometric Model: Instrumental Variable approach for Selection of investigators
Examining the causal implications of PI characteristics on the success of clinical trials re-
quires a methodology that takes potential selection of specific PIs on a trial into account.
There are two potential sources of selection: (1) the sponsor trying to select the best PIs for a
critical trial and (2) the PI trying to participate in the most interesting trials or the one with a
high probability of success. We consulted prior literature and engaged into in-depth discus-
sions with experts of clinical trials to further our understanding of potential selection. We did
this with the intention to comprehensively control for the types of influences in our models.
Azoulay (2004) suggests that the capabilities of the investigators are affecting success in
trials when hypothesis generation is important (Azoulay 2004). Such hypothesis generation
and testing is especially important in the early phases of the clinical research up to phase 2.
Trials in phase 3 and later focus on statistically proving the effectiveness of the treatment ver-
sus placebo or the then-current standard of care, based on its results in phase 2 trials. In our
study we focus on trials in phase 0 up to phase 2/3 and control for the specific phases. Huck-
man and Zinner (2005, p. 180) argue that the experience of the investigator is believed by
sponsors to be a “significant predictor of enrollment” of patients in a clinical trial. We control
for the number of enrolled patients (Huckman and Zinner 2008).
Our own interviewees expressed that trials are very disease specific. A PI or team of PIs is
preferred with experience in the same disease – but the pool of potential investigators with
disease specific experience strongly determines their possibility to select between investiga-
tors. From these we followed that our variable disexp depicting PIs experience in a specific
disease might be endogenous. In addition, we learned that selection possibility depends on the
pool of potential investigators. We build on these insights and employ an instrumental varia-
ble approach. We run 2 stage least square models and instrument disease experience by the
number of investigators which did run a trial in the same disease the five years prior to a focal
にぬ
trial. Since the increasing number of trials over the years impacts the pool of investigators
with disease specific experience, we normalize the number by dividing it through the number
of total trials. We believe the number of available investigators in a field to be random and
independent of the field and successes of therapeutics in the field. Career choices are made on
a long-term view, also in terms of specialization. Attractiveness of therapies and therapeutic
areas through scientific advancement or developments in therapeutics, though, cannot be pre-
dicted and can change a field of practice dramatically within a few years. The case of meta-
static melanoma is a prominent example, in which no therapeutics achieved better survival
over supportive care until 2011, when the first treatment from the field of immuno-oncology
was approved which showed not only improvement in overall survival, but also >10 year
complete responses, which could mean cure of the disease (Maio 2012). This revolutionary
development not only changed the face of metastatic melanoma, but also of the immuno-
oncology field and has made it much more attractive. Thus, we argue that the number of in-
vestigator in a certain disease area to be independent of the attractiveness of a field and as-
sume that the possibility of selecting a highly experienced investigator is correlated with the
number of investigators that are working in the disease. Previous research and our interviews
did not reveal that PIs with bridging skills are selected consciously so far.
5. Results
5.1. Bridging and Breadth of Experience and the Success
Table 2 reports our results on how bridging skills impact the success of trials for the full
sample of 4,816 trials. We run each model first as a two-stage least square regression with
experience being instrumented and second as an ordinary probit model for which we report
the marginal effects.
We will briefly comment on the appropriateness of the instrumental variable approach be-
fore we turn to the interpretation of the results. The endogeneity test indicates that ln_disexp
is indeed endogenous. The first stage regression of the two-stage least square suggest that the
にね
pool of PIs is a strong predictor of selecting an investigator with prior disease experience (1%
significance). For example, the F-statistic of 144 in Model 2a strongly suggests that our IV is
not weak (Stock-Yogo and Cragg-Donald Wald test).
Model 1a and 1b include the control variables. Our measure for experience in disease
(ln_disexp) is strongly significant in both models. The probit result of Model 1b suggests that
an increase in experience by 1 % increases success by nearly 7%. In the following models we
use various approaches to test our hypotheses on bridging skills of investigators. In Model 2a
and b we add the dummy variables d_bridging and d_basicpubs_only. Results suggest that
trials with bridging investigators are more successfully compared to the reference group
which consists of teams with investigators having solely applied skills. Hence, our results
support our hypothesis 1 and show that bridging is not only relevant for turning discoveries
into inventions (Gittelman and Kogut 2003)but also for turning inventions into innovations. It
is not that the exploitation stage solely requires the use of existing applied knowledge, pro-
cesses and methods (Baum, Li et al. 2000), instead coupling this with knowledge in basic
supports the successful translation.
Models 3 and 4 measure the bridging effect in a quantitative way and thus build on a con-
tinuum view of basic and applied science – the models suggest that a well- balanced skillset is
important. In Models 3a and b, we refine our bridging measure of Models 2a and b to count
only those observations as bridging, where the share of basic publications is between 25 and
75% of all publications. The significant coefficients of d_balancedbridging suggest that the
investigators which have a share of basic publications between 25% and 75% are particularly
successful. Correspondingly, Models 4a and b suggest that the share of basic publications
shows an inverted U-shape relationship with success. Basic research skills are important but
with diminishing return. Hence, there is an optimum in the share of basic science. Insofar our
results build on Toole and Czarnitzki’s (2009) findings that commercial productivity decreas-
にの
es with depth of scientifically oriented human capital but up to a certain level it spurs the suc-
cess rate (Toole and Czarnitzki 2009).
We add also a qualitative aspect to the analysis in Models 5a and b by including a measure
for excellence, which is measured as a dummy variable for the investigators that are the top
10% of cited scientists over a 5 year period in either their basic or applied science, or both.
Models 5a and b provide support for our hypothesis 2b that being excellent in both realms is
too much and has a negative effect on the success rate of a trial. Being among the most cited
individuals in both academic and commercially oriented research may divert the focus or the
investigator from the trial, or may place focus on other aspects that will increase his or her
competitiveness in both worlds of science. This is in line with the findings of Toole and
Czarnitzki (2009) that productivity stems not from the stars, but from the non-star scientists as
academic human capital, at least to the extent that we consider these to be the stars within the
basic research paradigm. Interestingly, being top in applied is positively related to success but
being top in basic is negatively related. Excellence in applied research by itself, though, will
actually increase success. The negative effect of excellence for bridging scientists on success
may result from a kind of “pull effect” towards academe that tilts the balance shown to be
important in previous models, and not on excellence per se, which would need to be further
investigated
*** Table 2 about here ***
Table 3 investigates to what extent breadth of experience in disease fields of the investiga-
tors impacts success. We only display the 2sls-models with experience in diseases instru-
mented since the results are quite robust to the probit model. Model 1 includes our measure
for breadth in field experience with the various forms of bridging variables along with the
interaction with breadth are included in Models 2 to 4. The negative coefficients of
ln_fielddivexp suggest that the broader investigators’ experience in different fields lowers the
success rate of trials. This correspondents with the findings of Macher and Boerner (2006)
には
reporting that companies face diseconomies of scope when working in different therapeutic
areas (Macher and Boerner 2006).
When we add the bridging variables along with the interactions, we find strong support for
H4 claiming that breadth of experience negatively moderates the impact of bridging on suc-
cess. Models 2 to 4 with breadth in terms of field experience reveal that H4 holds for all types
of interactions with bridging and breadth of experience in fields. Hence, if breadth of experi-
ence is coupled with bridging we find a negative impact on success. Presumably, the level of
complexity in coping with sharing attention horizontally in terms of different fields as well as
vertically in terms of basic and applied overstrains the capacity of investigators.
*** Table 3 about here ***
5.2. Trials with one investigator versus trials with a team of investigators
In the following, we split our sample into trials with just one investigator versus trials with a
team of investigators. Our objective is to provide insights whether and how the mechanism of
bridging may differ between individual investigators and teams of investigators.
Trials with one PI and Bridging
In the Appendix 3, we show models for the sample with one investigator. We restrict the table
to report the 2sls models with the instrumental variable approach for disease experience
(ln_disexp) but for reader’s interest re-run Model 1 as a probit. The results with the probit
model are quite similar to the 2sls, though sometimes shows larger significance of coeffi-
cients. Comparing the 2sls-models of the reduced sample in Appendix 3 to the 2sls-models
with the full sample in Table 2, we find that the results are quite robust.
When we include our measure for breadth of experience in field along with the interac-
tions, we do not find any significant result neither for ln_fielddivexp nor for the interactions.
Hence, breadth of experience seems not to matter for trial success. In terms of brevity we only
report but do not show a table with results.
Teams of Investigators and Bridging
にば
Teams of investigators are more likely to horizontally and vertically tap into different areas by
combining specialists in either area within one team. The importance of bridging the basic
and applied science divide may thus derive from very different configurations of individuals
within teams.
Figure 1 depicts the share of basic, applied and bridging investigators across different sizes
of investigator teams. It illustrates that the share of bridging investigators decreases while the
share of applied investigators, i.e., investigators who only publish in applied journals, increas-
es. The larger the team, the higher the fraction of specialized members, which could be a sign
that the diversity of knowledge required for the project requires it to be combined through the
members of the team as indicated by (Jones 2009).
Figure 1 Share of basic, applied and bridging investigators and teamsize
Table 4 reports the econometric models including only the observations with two or more
investigators on the trial. We briefly comment on our “usual” bridging variables but empha-
size Model 5 which splits the bridging measures into inter- and intra-individual bridging.
Models 1 to 4 suggest that the bridging concepts are too a lower extent present in teams.
While all coefficients of the bridging variables go in the same direction as in the model with
the full sample (and as hypothesized), the coefficients for d_bridging in Model 1a and b and
for testing an inverted U-shaped relation for share_basic and the coefficient of top citing in-
vestigators in basic and applied do not show significance. Corresponding to former models,
にぱ
balanced bridging shows a positive and significant impact on success. Comparing the insig-
nificance of Model 1b with the significance of Model 2 suggests that bridging through a min-
imal engagement in both basic and applied research does not influence success of the trial.
The partly insignificance of these measures may be due to the fact that for teams they capture
too many diverse bridging constellations, such as the combination of a basic and an applied
investigator within the team, teams with only bridging investigators or a combination of all
before mentioned.
Our main attention is devoted to Model 5, which splits all team trials into ones with only
basic investigators on the trials (d_basic_only), with only bridging scientists
(d_bridgingPIs_only), teams with a mix of bridging and other PIs (d_mixofPIsinbridging) and
the ones which only achieve bridging by combining PIs focused on either basic or applied
research (d_bridgingthroughteam). The reference category is teams with only applied investi-
gators. Our results suggest that teams with solely bridging scientist or teams with a mix of
bridging and other scientists are more successful than teams with only applied researchers.
This implies that only intra-individual bridging matters. Bridging skills need to be united in
one individual. The bridging concept cannot be achieved by combining pure applied with pure
basic scientists. This is a novel finding that corresponds with the notion of "generalists" who
are found to be driving highly innovative output in teams (Melero and Palomeras 2014). It
suggests that knowledge and skills are best combined, when present within one person
(Simon 1985).
However, teams might have advantages in assembling diverse field experience compared
to single PIs. Models 6 to 10 include breadth of field experience as well as its interactions
with our bridging variables. Breadth of field experience shows a negative significant effect on
success trial suggesting that though teams might be able to acquire diverse field knowledge it
is not paying off. Furthermore Models 8 to 10 suggest that breadth of experience even nega-
にひ
tively moderates the bridging variables. Only Model 10 which splits teams into intra- and
inter-individual bridging does not show any significant interaction effects.
6. Conclusion
We argue in this paper that individuals who possess a balanced set of skills between basic
and applied science can bridge the translational gap between inventions and innovations and
improve success rates in product development, by themselves or as members of a team. This
supports the notion that basic science and applied science are complementary. Our results
from the analysis of the influence of investigator characteristics on clinical trials show that the
value of basic science insight and skills reaches not only into applied research stages but even
into product development. Basic science is not only a source for discoveries that may become
commercial products, its insights, methodologies and problem solving skills of basic science
are important tools even in such "late" activities as final testing and development of the prod-
uct (Pavitt 1991; Zucker, Darby et al. 1998; Zucker, Darby et al. 2001). The translational step
performed in drug development is marked by two different institutional logics and incentive
systems: It incorporates the basic science that enables an understanding of the underlying
functionality of the product or technology, but requires the commercial mindset to optimally
exploit its potential in the market. Investigators exhibit superior performance, if their scien-
tific career and engagement has been marked by both institutional paradigms. Working in the
development stage of the value chain, a main focus is the exploitation of existing knowledge
and the commercialization of the product under testing. Still, the methodologies and problem
solving skills that are gained through insights into the exploration of basic science (Rosenberg
1990) enable both a faster and more successful interrogation of a potential product and its
commercial opportunity.
Essential to the success of bridging is the relative intensity with which both basic and ap-
plied science is conducted by the investigator(s) and serves as input, and where the emphasis
ぬど
is placed between the two systems. Our findings suggest that both activities are complemen-
tary up to a certain degree. Excellence in both will actually decrease success. The values and
reward systems of the highly cited bridging scientists may still lie too much with the scientific
community so as to increase their successes at translating research into product development
(Gittelman and Kogut 2003). As is the case for the balance between exploration and exploita-
tion of (Gupta, Smith et al. 2006), excellence in both basic and applied science causes a trade-
off that inhibits success in exploitation projects, potentially as excellence in either is mutually
exclusive.
The interactive effect of bridging and breadth of experience suggests both benefits and
costs to the increasing diversity of activities. Penalties exist for who attempts to master both
the vertical translational abilities and attempts broadly apply their expertise across fields, sup-
porting findings by Macher and Boehrner (2006). The more general ability of bridging di-
verse fields of exploitation, horizontal bridging, showed negative effects in its interaction
with vertical, or basic to applied bridging. The costs of sharing attention to various areas in-
stead of concentrating on one are higher than the benefit of recombining knowledge from di-
verse fields.
Furthermore, we find intra-individual bridging to be of importance, not inter-individual
bridging within teams. While a lot of research has found specialization (Jones 2009), team
research (Wuchty, Jones et al. 2007) and team size (Milojević 2014) to increase, we find the
generalists in terms of basic and applied science within the teams to moderate the positive
impact of bridging. This finding corresponds with prior research that generalists within teams
are essential for the impact of innovations (Melero and Palomeras 2013; Melero and
Palomeras 2014). Our finding that intra-individual bridging of the basic/applied science
spheres enhances success of turning invention into innovations confirms other findings that
individuals are better at integrating diverse knowledge than teams of specialized members
(Gupta, Smith et al. 2006; Taylor and Greve 2006). Also for teams, we find that breadth of
ぬな
experience negatively moderates bridging. This supports the notion that, though teams should
be able to combine more diverse knowledge, they may face challenges in communication and
coordination, and thus do not perform better at knowledge recombination than individuals
(Williams and O'Reilly 1998).
For the industry, this work adds insights that may aid to alleviate the economic burden of
product development by increasing chances of success through correct selection of key indi-
viduals to lead projects and team members within such projects. It adds supporting evidence
to a broader education of clinical trial investigators in terms of research methodologies
(FitzGerald 2005).Furthermore, the advent of outsourcing of clinical trials to so called clinical
research organizations, which causes an even stronger divide between the scientists and la-
boratories that originally discovered and patented the product and the clinical researchers, that
may not only inhibit learning by the originator company (Smed and Getz 2013), but also de-
crease the likelihood of conducting a successful trial.
ぬに
References Agrawal┸A┻┸J┻Mc(ale┸etal┻ゅにどなぬょ┻Collaboration┸Stars┸andtheChangingOrganizationofScience┺EvidencefromEvolutionaryBiology┻TheChangingFrontier┺RethinkingScienceand)nnovationPolicy┸UniversityofChicagoPress┻Agrawal┸A┻┸J┻Mc(ale┸etal┻ゅにどなぬょ┻gWhystarsmatter┻gUniversityofTorontomimeo┻Azoulay┸P┻ゅにどどねょ┻gCapturingknowledgewithinandacrossfirmboundaries┺evidencefromclinicaldevelopment┻gTheAmericanEconomicReview94ゅのょ┺なのひな┽なはなに┻Azoulay┸P┻┸J┻S┻G┻Zivin┸etal┻ゅにどななょ┻TheDiffusionofScientificKnowledgeAcrossTimeandSpace┺EvidencefromProfessionalTransitionsfortheSuperstarsofMedicine┸NationalBureauofEconomicResearch┻Baum┸J┻A┻┸S┻X┻Li┸etal┻ゅにどどどょ┻gMakingthenextmove┺(owexperientialandvicariouslearningshapethelocationsofchainsfacquisitions┻gAdministrativeScienceQuarterly45ゅねょ┺ばはは┽ぱどな┻Benner┸M┻J┻andM┻Tushmanゅにどどにょ┻gProcessmanagementandtechnologicalinnovation┺Alongitudinalstudyofthephotographyandpaintindustries┻gAdministrativeScienceQuarterly47ゅねょ┺はばは┽ばどば┻Bercovitz┸J┻andM┻Feldmanゅにどななょ┻gThemechanismsofcollaborationininventiveteams┺Composition┸socialnetworks┸andgeography┻gResearchPolicy40ゅなょ┺ぱな┽ひぬ┻Boh┸W┻F┻┸R┻Evaristo┸etal┻ゅにどなねょ┻gBalancingbreadthanddepthofexpertiseforinnovation┺AぬMstory┻gResearchPolicy43ゅにょ┺ぬねひ┽ぬはは┻Brandl┸J┻E┻ゅなひひぱょ┻PasteurfsQuadrant┺Basicscienceandtechnologicalinnovation┸WileyOnlineLibrary┻Buonansegna┸E┻┸S┻Salomo┸etal┻ゅにどなにょ┻Towardsaframeworkofsuccessfactorsforclinicaltrials┻なひth)nternationalProductdevelopmentManagementConference┻Buonansegna┸E┻┸S┻Salomo┸etal┻ゅにどなねょ┻gPharmaceuticalnewproductdevelopment┺whydoclinicaltrialsfail╂gR┃DManagement44ゅにょ┺なぱひ┽にどに┻Bush┸V┻ゅなひねのょ┻gScience┺Theendlessfrontier┻gTransactionsoftheKansasAcademyofScienceゅなひどぬょ┺にぬな┽にはね┻Calderini┸M┻┸C┻Franzoni┸etal┻ゅにどどばょ┻g)fstarscientistsdonotpatent┺Theeffectofproductivity┸basicnessandimpactonthedecisiontopatentintheacademicworld┻gResearchPolicy36ゅぬょ┺ぬどぬ┽ぬなひ┻Cohen┸W┻M┻andD┻A┻Levinthalゅなひひどょ┻gAbsorptivecapacity┺anewperspectiveonlearningandinnovation┻gAdministrativesciencequarterly┺なにぱ┽なのに┻Danzon┸P┻M┻┸S┻Nicholson┸etal┻ゅにどどのょ┻gProductivityinpharmaceutical┽biotechnologyR┃D┻Theroleofexperienceandalliances┻gJournalofhealtheconomics24ゅにょ┺ぬなば┽ぬぬひ┻Darby┸M┻R┻┸Q┻Liu┸etal┻ゅなひひひょ┻Stakesandstars┻Theeffectsofintellectualhumancapitalonthelevelandvariabilityofhigh┽techfirmsfmarketvalues┻Cambridge┸Mass┻┺NationalBureauofEconomicResearch┻Dasgupta┸P.andP┻A┻Davidゅなひひねょ┻gTowardaneweconomicsofscience┻gResearchpolicy23ゅのょ┺ねぱば┽のにな┻Davidoff┸F┻┸C┻D┻DeAngelis┸etal┻ゅにどどなょ┻gSponsorship┸authorship┸andaccountability┻gJAMA┺thejournaloftheAmericanMedicalAssociation286ゅなどょ┺なにぬに┻DellaMalva┸A┻┸B┻Leten┸etal┻ゅにどなぬょ┻gBasicScienceasaPrescriptionforTechnologicalBreakthroughsinthePharmaceutical)ndustry┻gDeMonaco┸(┻J┻┸A┻Ali┸etal┻ゅにどどはょ┻gTheMajorRoleofCliniciansintheDiscoveryofOff̺LabelDrugTherapies┻gPharmacotherapy┺TheJournalof(umanPharmacologyandDrugTherapy26ゅぬょ┺ぬにぬ┽ぬぬに┻
ぬぬ
DiMasi┸J┻ゅにどどなょ┻gRisksinnewdrugdevelopment┺Approvalsuccessratesforinvestigationaldrugs┻gClinicalPharmacology┃Therapeutics69ゅのょ┺にひば┽ぬどば┻DiMasi┸J┻A┻ゅなひひなょ┻gCostofinnovationinthepharmaceuticalindustry┻gJournalofhealtheconomics10ゅにょ┺などば┽なねに┻DiMasi┸J┻A┻ゅにどなねょ┻gPharmaceuticalR┃DPerformancebyFirmSize┺ApprovalSuccessRatesandEconomicReturns┻gAmericanjournaloftherapeutics21ゅなょ┺には┽ぬね┻DiMasi┸J┻A┻┸L┻Feldman┸etal┻ゅにどなどょ┻gTrendsinrisksassociatedwithnewdrugdevelopment┺successratesforinvestigationaldrugs┻gClinicalPharmacology┃Therapeutics87ゅぬょ┺にばに┽にばば┻DiMasi┸J┻A┻and(┻G┻Grabowskiゅにどどばょ┻gThecostofbiopharmaceuticalR┃D┻)sbiotechdiffent╂gManagerialanddecisioneconomics28ゅね【のょ┺ねはひ┽ねばひ┻DiMasi┸J┻A┻┸R┻W┻(ansen┸etal┻ゅにどどぬょ┻gThepriceofinnovation┻Newestimatesofdrugdevelopmentcosts┻gJournalofhealtheconomics22ゅにょ┺なのな┽なぱの┻FitzGerald┸G┻A┻ゅにどどのょ┻gAnticipatingchangeindrugdevelopment┺theemergingeraoftranslationalmedicineandtherapeutics┻gNatureReviewsDrugDiscovery4ゅなどょ┺ぱなの┽ぱなぱ┻Fleming┸L┻┸S┻Mingo┸etal┻ゅにどどばょ┻gCollaborativebrokerage┸generativecreativity┸andcreativesuccess┻gAdministrativeScienceQuarterly52ゅぬょ┺ねねぬ┽ねばの┻Fleming┸L┻andO┻Sorensonゅにどどぬょ┻gNavigatingthetechnologylandscapeofinnovation┻gM)TSloanManagementReview┻Fleming┸L┻andO┻Sorensonゅにどどねょ┻gScienceasamapintechnologicalsearch┻gStrategicManagementJournal25ゅぱひょ┺ひどひ┽ひにぱ┻Gibbons┸M┻andR┻Johnstonゅなひばねょ┻gTherolesofscienceintechnologicalinnovation┻gResearchPolicy3ゅぬょ┺ににど┽にねに┻Gittelman┸M┻andB┻Kogutゅにどどぬょ┻gDoesgoodscienceleadtovaluableknowledge╂Biotechnologyfirmsandtheevolutionarylogicofcitationpatterns┻gManagementScience49ゅねょ┺ぬはは┽ぬぱに┻Gupta┸A┻K┻┸K┻G┻Smith┸etal┻ゅにどどはょ┻gTheinterplaybetweenexplorationandexploitation┻gAcademyofManagementJournal49ゅねょ┺はひぬ┽ばどは┻(äussler┸C┻and(┻Sauermannゅにどなねょ┻gTheAnatomyofTeams┺DivisionofLaborandAllocationofCreditinCollaborativeKnowledgeProduction┻gAvailableatSSRN┻(enderson┸R┻and)┻Cockburnゅなひひねょ┻gMeasuringcompetence╂Exploringfirmeffectsinpharmaceuticalresearch┻gStrategicmanagementjournal15ゅSなょ┺はぬ┽ぱね┻(enderson┸R┻┸A┻Jaffe┸etal┻ゅにどどのょ┻gPatentcitationsandthegeographyofknowledgespillovers┻Areassessment┺comment┻gTheAmericaneconomicreview95ゅなょ┺ねはな┽ねはね┻(oang┸(┻andF┻T┻Rothaermelゅにどなどょ┻gLeveraginginternalandexternalexperience┺exploration┸exploitation┸andR┃Dprojectperformance┻gStrategicManagementJournal31ゅばょ┺ばぬね┽ばのぱ┻(oekman┸J┻┸K┻Frenken┸etal┻ゅにどなにょ┻gThegeographicaldistributionofleadershipinglobalizedclinicaltrials┻gPloSone7ゅなどょ┺eねのひぱね┻(u┸M┻┸K┻Schultz┸etal┻ゅにどどばょ┻gTheinnovationgapinpharmaceuticaldrugdiscovery┃newmodelsforR┃Dsuccess┻g(uckman┸R┻S┻andD┻E┻Zinnerゅにどどぱょ┻gDoesfocusimproveoperationalperformance╂Lessonsfromthemanagementofclinicaltrials┻gStrategicManagementJournal29ゅにょ┺なばぬ┽なひぬ┻Jones┸B┻F┻ゅにどどひょ┻gTheburdenofknowledgeandthe╉deathoftheRenaissanceman╊┺isinnovationgettingharder╂gTheReviewofEconomicStudies76ゅなょ┺にぱぬ┽ぬなば┻Jones┸B┻F┻andB┻A┻Weinbergゅにどななょ┻gAgedynamicsinscientificcreativity┻gProceedingsoftheNationalAcademyofSciences108ゅねばょ┺なぱひなど┽なぱひなね┻
ぬね
Katila┸R┻andG┻Ahujaゅにどどにょ┻gSomethingold┸somethingnew┺Alongitudinalstudyofsearchbehaviorandnewproductintroduction┻gAcademyofmanagementjournal45ゅはょ┺ななぱぬ┽ななひね┻Kola┸)┻andJ┻Landisゅにどどねょ┻gCanthepharmaceuticalindustryreduceattritionrates╂gNatRevDrugDiscov3ゅぱょ┺ばなな┽ばなの┻Levinthal┸D┻A┻andJ┻G┻Marchゅなひひぬょ┻gThemyopiaoflearning┻gStrategicmanagementjournal14ゅSにょ┺ひの┽ななに┻Liu┸W┻ゅにどどはょ┻gKnowledgeexploitation┸knowledgeexploration┸andcompetencytrap┻gKnowledgeandProcessManagement13ゅぬょ┺なねね┽なはな┻Macher┸J┻T┻andC┻S┻Boernerゅにどどはょ┻gExperienceandscaleandscopeeconomies┺trade┽offsandperformanceindevelopment┻gStrategicManagementJournal27ゅひょ┺ぱねの┽ぱはの┻Maio┸M┻ゅにどなにょ┻gMelanomaasamodeltumourforimmuno┽oncology┻gAnnalsofoncology23ゅsupplぱょ┺viiiなど┽viiiなね┻March┸J┻G┻ゅなひひなょ┻gExplorationandexploitationinorganizationallearning┻gOrganizationscience2ゅなょ┺ばな┽ぱば┻McMillan┸G┻S┻┸F┻Narin┸etal┻ゅにどどどょ┻gAnanalysisofthecriticalroleofpublicscienceininnovation┺thecaseofbiotechnology┻gResearchpolicy29ゅなょ┺な┽ぱ┻Melero┸E┻andN┻Palomerasゅにどなぬょ┻TheRenaissanceoftheRenaissanceMan╂TheRoleofBroad)ndividualKnowledgeinTeamsof)nventors┻ぬのthDRU)DCelebrationConference┸Barcelona┸Spain┻Melero┸E┻andN┻Palomerasゅにどなねょ┻gTheRenaissanceManisnotdead┿Theroleofgeneralistsinteamsofinventors┻gResearchPolicy┻Merton┸R┻K┻ゅなひばぬょ┻Thesociologyofscience┺Theoreticalandempiricalinvestigations┸UniversityofChicagopress┻Milojević┸S┻ゅにどなねょ┻gPrinciplesofscientificresearchteamformationandevolution┻gProceedingsoftheNationalAcademyofSciences111ゅななょ┺ぬひぱね┽ぬひぱひ┻Nijstad┸B┻A┻andC┻K┻DeDreuゅにどどにょ┻gCreativityandgroupinnovation┻gAppliedPsychology51ゅぬょ┺ねどど┽ねどは┻NSB┸N┻S┻B┻ゅにどどねょ┻gScienceandEngineering)ndicatorsにどどぬ┻gU┻S┻GovernmentPrintingOffice┺Washington┸D┻C┻┻Owen┽Smith┸J┻andW┻W┻Powellゅにどどねょ┻gKnowledgenetworksaschannelsandconduits┺TheeffectsofspilloversintheBostonbiotechnologycommunity┻gOrganizationscience15ゅなょ┺の┽にな┻Pammolli┸F┻┸L┻Magazzini┸etal┻ゅにどななょ┻gTheproductivitycrisisinpharmaceuticalR┃D┻gNaturereviewsDrugdiscovery10ゅはょ┺ねにぱ┽ねぬぱ┻Pavitt┸K┻ゅなひひなょ┻gWhatmakesbasicresearcheconomicallyuseful╂gResearchPolicy20ゅにょ┺などひ┽ななひ┻Polanyi┸M┻ゅなひのぱょ┻PersonalKnowledge┻Towardsapost┽criticalphilosophy┻┻Chicago┸TheUniversityofChicagoPress┻Polidoro┸F┻andM┻Theekeゅにどなにょ┻gGettingCompetitionDowntoaScience┺TheEffectsofTechnologicalCompetitiononFirmsfScientificPublications┻gOrganizationScience23ゅねょ┺ななぬの┽ななのぬ┻Rasmussen┸N┻ゅにどどのょ┻gThedrugindustryandclinicalresearchininterwarAmerica┺threetypesofphysiciancollaborator┻gBulletinofthe(istoryofMedicine79ゅなょ┺のど┽ぱど┻Rosenberg┸N┻ゅなひばぱょ┻gThediffusionoftechnology┺Aneconomichistorian╆sview┻gTheDiffusionof)nnovations┺AnAssessment┸Centerfor)nterdisciplinaryStudyofScienceandTechnology┸Evanstonゅ)ll┻ょ┺NorthwesternUniversity┻
ぬの
Rosenberg┸N┻ゅなひひどょ┻gWhydofirmsdobasicresearchゅwiththeirownmoneyょ╂gResearchpolicy19ゅにょ┺なはの┽なばね┻Rothaermel┸F┻T┻andD┻L┻Deedsゅにどどねょ┻gExplorationandexploitationalliancesinbiotechnology┺Asystemofnewproductdevelopment┻gStrategicmanagementjournal25ゅぬょ┺にどな┽ににな┻Simon┸(┻A┻ゅなひぱのょ┻gWhatweknowaboutthecreativeprocess┻gFrontiersincreativeandinnovativemanagement4┺ぬ┽にに┻Singh┸J┻andL┻Flemingゅにどなどょ┻gLoneinventorsassourcesofbreakthroughs┺Mythorreality╂gManagementScience56ゅなょ┺ねな┽のは┻Smed┸M┻andK┻A┻Getzゅにどなぬょ┻gUnfulfilledtranslationopportunitiesinindustrysponsoredclinicaltrials┻gContemporaryclinicaltrials35ゅなょ┺ぱど┽ぱは┻Sternberg┸R┻J┻andL┻A┻Of(araゅなひひひょ┻gなぬCreativityand)ntelligence┻g(andbookofcreativity┺にのな┻Stokes┸D┻E┻ゅなひひばょ┻PasteurfsQuadrant┽BasicScienceandTechnological)nnovation┸Brookings)nstitutionPressWashington┻SungNs┸C┻J┻W┻F┻G┻M┻andetal┻ゅにどどぬょ┻gCentralchallengesfacingthenationalclinicalresearchenterprise┻gJAMA289ゅなどょ┺なにばぱ┽なにぱば┻Taylor┸A┻and(┻R┻Greveゅにどどはょ┻gSupermanorthefantasticfour╂Knowledgecombinationandexperienceininnovativeteams┻gAcademyofManagementJournal49ゅねょ┺ばにぬ┽ばねど┻Toole┸A┻A┻andD┻Czarnitzkiゅにどどひょ┻gExploringtherelationshipbetweenscientisthumancapitalandfirmperformance┻ThecaseofbiomedicalacademicentrepreneursintheSB)Rprogram┻gManagementScience55ゅなょ┺などな┽ななね┻Weisberg┸R┻W┻ゅなひひひょ┻gCreativityandKnowledge┺AChallengetoTheories┻g(andbookofcreativity┺にには┻Williams┸K┻Y┻andC┻A┻OfReillyゅなひひぱょ┻gDemographyanddiversityinorganizations┺Areviewofねどyearsofresearch┻gResearchinorganizationalbehavior20┺ばば┽なねど┻Woolf┸S┻(┻ゅにどどぱょ┻gThemeaningoftranslationalresearchandwhyitmatters┻gJama299ゅにょ┺になな┽になぬ┻Wuchty┸S┻┸B┻F┻Jones┸etal┻ゅにどどばょ┻gTheincreasingdominanceofteamsinproductionofknowledge┻gScience316ゅのぱにばょ┺などぬは┽などぬひ┻Zerhouni┸E┻ゅにどどばょ┻gTranslationalresearch┺movingdiscoverytopractice┻gClinicalPharmacology┃Therapeutics81ゅなょ┺なには┽なにぱ┻Zucker┸L┻G┻andM┻R┻Darbyゅなひひのょ┻Virtuouscirclesofproductivity┻Starbioscientistsandtheinstitutionaltransformationofindustry┻Cambridge┸Mass┻┸NationalBureauofEconomicResearch┻Zucker┸L┻G┻andM┻R┻Darbyゅなひひはょ┻gStarscientistsandinstitutionaltransformation┺Patternsofinventionandinnovationintheformationofthebiotechnologyindustry┻gProceedingsoftheNationalAcademyofSciences93ゅにぬょ┺なにばどひ┽なにばなは┻Zucker┸L┻G┻┸M┻R┻Darby┸etal┻ゅにどどなょ┻gCommercializingKnowledge┺UniversityScience┸KnowledgeCapture┸andFirmPerformanceinBiotechnology┻gNationalBureauofEconomicResearchWorkingPaperSeriesNo.8499┻Zucker┸L┻G┻┸M┻R┻Darby┸etal┻ゅなひひぱょ┻g)ntellectualhumancapitalandthebirthofUSbiotechnologyenterprises┻gTheAmericaneconomicreview88ゅなょ┺にひど┽ぬどは┻
ぬは
Table 1: Summary Statistics
Variable Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max
Dependent variablesuccess 0.43 0.49 1 0.45 0 1 0.40 0 1
Independent variablesd_bridging 0.52 0.50 1 0.44 0 1 0.61 0 1
d_basicpubs_only 0.30 0.46 1 0.33 0 1 0.26 0 1
d_appliedpubs_only 0.18 0.39 1 0.23 0 1 0.13 0 1
d_balancedbridging 0.27 0.44 1 0.24 0 1 0.31 0 1
share_basic 61.56 37.70 0 100 59.12 39.87 0 100 64.40 34.80 0 100
d_top_citbridging 0.07 0.26 1 0.04 0 1 0.11 0 1
d_top_citbasic 0.10 0.30 1 0.08 0 1 0.13 0 1
d_top_citapplied 0.11 0.31 1 0.10 0 1 0.12 0 1
d_basic_only 0.03 0.18 1 0.05 0 1 0.02 0 1
d_applied_only 0.01 0.11 1 0.01 0 1 0.01 0 1
d_bridgingPIs_only 0.57 0.50 1 n/a 0.57 0 1
d_mixofPIsbridging 0.39 0.49 1 n/a 0.39 0 1
d_bridgingthroughteam 0.01 0.10 1 n/a 0.01 0 1
fielddivexp 0.98 1.63 0 31 0.59 0.82 0 6 1.43 2.14 0 31
disexp 0.69 1.54 0 27 0.59 1.30 0 11 0.80 1.78 0 27
Control variablesno_investig 2.09 1.65 1 8 n/a 3.35 1.71 2 8
no_pubs 11.61 20.89 1 265 7.75 13.84 1 260 16.10 26.15 1 265
d_industry 0.36 0 1 0.40 0 1 0.31 0 1
d_bigpharma 0.05 0 1 0.05 0 1 0.06 0 1
sponsor_trialexp 326.82 492.31 1 1979 327.61 482.90 1 1979 325.91 503.14 1 1979
sponsor_disexp 59.33 150.40 0 1021 62.95 153.24 0 977 55.12 146.95 0 1021
priordrugtrials_samephase 15.64 34.53 0 252 17.95 38.21 0 252 12.96 29.47 0 252
enrollment 89.49 231.98 1 6600 81.65 223.25 1 6600 98.59 241.47 1 5029
location_count 4.30 8.46 1 50 4.20 8.36 1 50 4.41 8.58 1 50
top10site 0.09 0 1 0.10 0 1 0.08 0 1
USA 0.60 0 1 0.64 0 1 0.56 0 1
d_phase0 0.04 0 1 0.02 0 1 0.05 0 1
d_phase1 0.33 0 1 0.32 0 1 0.35 0 1
d_phase12 0.11 0 1 0.11 0 1 0.11 0 1
d_phase2 0.46 0 1 0.50 0 1 0.41 0 1
d_phase23 0.06 0 1 0.05 0 1 0.07 0 1
endocrinedisease 0.07 0 1 0.08 0 1 0.07 0 1
cancer 0.43 0 1 0.44 0 1 0.42 0 1
circulatorydisease 0.05 0 1 0.04 0 1 0.06 0 1
mentaldisease 0.07 0 1 0.07 0 1 0.06 0 1
infectiousdisease 0.08 0 1 0.08 0 1 0.08 0 1
otherdisease 0.30 0 1 0.29 0 1 0.31 0 1
Full Sample (N=4816) Sample with only one PI (N=2589 Sample with a team of Pis (N=2227)
ぬば
Table 2: Bridging Skills and the Impact on Trial Success
(1a) (1b) (2a) (2b) (3a) (3b) (4a) (4b) (5a) (5b)
VARIABLES success success success success success success success success success success
IV Probit IV Probit IV Probit IV Probit IV Probit
d_basicpubs_only 0.001 -0.021
(0.023) (0.023)
d_bridging 0.045* 0.046*
(0.024) (0.025)
d_balancedbridging 0.065*** 0.077***
(0.017) (0.017)
share_basic 0.002*** 0.003***
(0.001) (0.001)
share_basic_squ -0.000*** -0.000***
(0.000) (0.000)
d_top_citbasic -0.019 -0.054*
(0.030) (0.029)
d_top_citapplied 0.059** 0.083***
(0.028) (0.028)
d_top_citbridging -0.080** -0.081**
(0.040) (0.038)
ln_disexp 0.516*** 0.070*** 0.519*** 0.069*** 0.518*** 0.070*** 0.516*** 0.068*** 0.503*** 0.068***
(0.087) (0.014) (0.087) (0.014) (0.087) (0.014) (0.088) (0.014) (0.087) (0.014)
priordrugtrials_samephas 0.004*** 0.007*** 0.004*** 0.007*** 0.004*** 0.007*** 0.004*** 0.007*** 0.004*** 0.007***
(0.000) (0.001) (0.000) (0.001) (0.000) (0.001) (0.000) (0.001) (0.000) (0.001)
no_investig -0.017*** -0.002 -0.017*** -0.002 -0.018*** -0.003 -0.018*** -0.003 -0.016*** -0.001
(0.006) (0.005) (0.006) (0.005) (0.006) (0.005) (0.006) (0.005) (0.006) (0.005)
ln_no_pubs -0.025*** -0.008 -0.037*** -0.024** -0.029*** -0.014* -0.033*** -0.017* -0.016 0.003
(0.008) (0.008) (0.010) (0.010) (0.008) (0.008) (0.009) (0.009) (0.012) (0.011)
d_industry 0.032* 0.034** 0.032* 0.033* 0.032* 0.034** 0.032* 0.033* 0.031* 0.034*
(0.017) (0.017) (0.017) (0.017) (0.017) (0.017) (0.017) (0.017) (0.017) (0.017)
d_bigpharma -0.186*** -0.068 -0.185*** -0.066 -0.183*** -0.064 -0.184*** -0.066 -0.180*** -0.066
(0.050) (0.043) (0.050) (0.043) (0.050) (0.043) (0.050) (0.043) (0.049) (0.043)
sponsor_trialexp 0.000* 0.000 0.000* 0.000 0.000* 0.000 0.000* 0.000 0.000* 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
sponsor_disexp -0.000 0.000 -0.000 0.000 -0.000 0.000 -0.000 0.000 -0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
enrollment -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
location_count 0.002* 0.003*** 0.002* 0.003*** 0.002* 0.003*** 0.002* 0.003*** 0.002* 0.003***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
top10site -0.010 -0.008 -0.010 -0.008 -0.011 -0.010 -0.011 -0.009 -0.009 -0.005
(0.029) (0.029) (0.029) (0.029) (0.029) (0.029) (0.029) (0.029) (0.029) (0.030)
USA -0.078*** -0.051*** -0.078*** -0.051*** -0.077*** -0.050*** -0.077*** -0.050*** -0.076*** -0.049***
(0.018) (0.017) (0.018) (0.017) (0.018) (0.017) (0.018) (0.017) (0.018) (0.017)
d_phase0 -0.019 -0.057 -0.018 -0.055 -0.016 -0.054 -0.015 -0.051 -0.019 -0.056
(0.043) (0.041) (0.043) (0.041) (0.043) (0.041) (0.043) (0.042) (0.043) (0.041)
d_phase12 0.068*** 0.059** 0.069*** 0.061** 0.068*** 0.059** 0.070*** 0.063** 0.064** 0.054**
(0.026) (0.025) (0.026) (0.025) (0.026) (0.025) (0.026) (0.025) (0.026) (0.025)
d_phase2 -0.055*** -0.080*** -0.056*** -0.082*** -0.058*** -0.083*** -0.057*** -0.083*** -0.058*** -0.085***
(0.018) (0.018) (0.019) (0.018) (0.019) (0.018) (0.018) (0.018) (0.018) (0.018)
d_phase23 0.015 -0.023 0.015 -0.025 0.016 -0.022 0.015 -0.024 0.013 -0.025
(0.035) (0.033) (0.035) (0.033) (0.035) (0.034) (0.035) (0.034) (0.035) (0.034)
cancer 0.069** 0.041 0.066** 0.037 0.067** 0.039 0.065** 0.035 0.070** 0.042
(0.031) (0.030) (0.031) (0.030) (0.031) (0.030) (0.031) (0.031) (0.030) (0.031)
circulatorydisease -0.039 -0.121*** -0.040 -0.123*** -0.041 -0.123*** -0.041 -0.123*** -0.040 -0.120***
(0.046) (0.040) (0.046) (0.040) (0.046) (0.040) (0.046) (0.041) (0.045) (0.041)
mentaldisease 0.052 -0.002 0.050 -0.008 0.047 -0.010 0.046 -0.014 0.049 -0.006
(0.042) (0.039) (0.042) (0.039) (0.042) (0.039) (0.042) (0.039) (0.042) (0.040)
infectiousdisease 0.053 0.026 0.050 0.020 0.043 0.015 0.042 0.008 0.046 0.012
(0.039) (0.038) (0.039) (0.038) (0.039) (0.038) (0.040) (0.038) (0.040) (0.039)
otherdisease 0.046 -0.056* 0.046 -0.057* 0.046 -0.057* 0.046 -0.057* 0.046 -0.055*
(0.036) (0.030) (0.036) (0.030) (0.036) (0.030) (0.036) (0.030) (0.036) (0.030)
Constant 0.272*** 0.275*** 0.267*** 0.280*** 0.259***
(0.042) (0.045) (0.042) (0.044) (0.043)
Observations 4,816 4,816 4,816 4,816 4,816 4,816 4,816 4,816 4,816 4,816
ll ど3661 ど2909 ど3665 ど2903 ど3659 ど2899 ど3655 ど2897 ど3625 ど2896F 29.22 27.02 28.69 27.60 26.78
r2_p 0.115 0.117 0.118 0.119 0.119
Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
ぬぱ
Table 3: Breadth of Experience of Fields and Interactions with Bridging Variables
(1) (2) (3) (4)
success success success success
VARIABLES IV IV IV IV
d_basicpubs_only 0.013
(0.034)
d_bridging 0.264***
(0.069)
d_balancedbridging 0.234***
(0.046)
d_top_citbasic -0.019
(0.039)
d_top_citapplied 0.087**
(0.035)
d_top_citbridging -0.011
(0.065)
ln_fielddivexp -0.551*** -0.499*** -0.540*** -0.535***
(0.123) (0.125) (0.125) (0.120)
int_lnfielddivexp_bridging -0.442***
(0.116)
int_lnfielddivexp_balancedbrid -0.305***
(0.074)
int_lnfielddivexp_topcitbridging -0.106*
(0.061)
ln_disexp 1.109*** 1.432*** 1.230*** 1.088***
(0.240) (0.352) (0.276) (0.238)
ln_no_pubs -0.017* -0.034** -0.024** -0.010
(0.010) (0.014) (0.011) (0.015)
no_investig 0.020** 0.027*** 0.022*** 0.021***
(0.008) (0.010) (0.009) (0.008)
priordrugtrials_samephase 0.004*** 0.004*** 0.004*** 0.004***
(0.000) (0.000) (0.000) (0.000)
d_industry 0.033 0.036 0.038* 0.031
(0.022) (0.025) (0.023) (0.022)
d_bigpharma -0.223*** -0.235*** -0.223*** -0.215***
(0.067) (0.078) (0.071) (0.066)
sponsor_disexp 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
sponsor_trialexp 0.000* 0.000 0.000 0.000*
(0.000) (0.000) (0.000) (0.000)
enrollment -0.000 -0.000 -0.000* -0.000
(0.000) (0.000) (0.000) (0.000)
location_count 0.002* 0.003** 0.003** 0.002*
(0.001) (0.001) (0.001) (0.001)
top10site -0.019 -0.048 -0.035 -0.019
(0.037) (0.044) (0.039) (0.037)
USA -0.081*** -0.075*** -0.078*** -0.077***
(0.023) (0.025) (0.024) (0.023)
Constant 0.266*** 0.104 0.193*** 0.249***
(0.0536) (0.0953) (0.0654) (0.0551)
Observations 4,816 4,816 4,816 4,816
F 17.21 11.88 14.73 15.41
Note: Phase and Field Fixed Effects included.
ぬひ
Table 4: Models with Trials with a Team of Investigators (>1)
(1a) (1b) (2) (3) (4) (5) (6) (7) (8) (9) (10)
success success success success success success success success success success success
VARIABLES Probit IV IV IV IV IV IV IV IV IV IV
d_basicpubs_only -0.030 -0.005 -0.012
(0.039) (0.038) (0.048)
d_bridging 0.025 0.034 0.239***
(0.041) (0.039) (0.077)
d_balancedbridging 0.059** 0.227***
(0.024) (0.051)
share_basic 0.002
(0.001)
share_basic_squ -0.000
(0.000)
d_top_citbasic -0.020 -0.022
(0.041) (0.048)
d_top_citapplied 0.066* 0.108**
(0.039) (0.046)
d_top_citbridging -0.047 0.052
(0.052) (0.079)
d_basic_only 0.202 0.213
(0.130) (0.198)
d_bridgingPIs_only 0.267** 0.402**
(0.104) (0.166)
d_mixofPIsinbridging 0.277*** 0.387**
(0.104) (0.166)
d_bridgingthroughteam 0.220 0.188
(0.148) (0.217)
ln_fielddivexp -0.401*** -0.341*** -0.378*** -0.391*** -0.265
(0.097) (0.090) (0.094) (0.096) (0.251)
int_lnfielddivexp_bridging -0.294***
(0.080)
int_lnfielddivexp_balancedbrid -0.231***
(0.063)
int_lnfielddivexp_topcitbridging -0.121*
(0.065)
int_lnfielddivexp_basic_only -0.109
(0.310)
int_lnfielddivexp_bridging_only -0.182
(0.239)
int_lnfielddivexp_mix -0.150
(0.239)
int_lnfielddivexp_bridging_team 0.177
(0.292)
ln_disexp 0.058*** 0.522*** 0.518*** 0.521*** 0.511*** 0.530*** 0.893*** 1.088*** 0.965*** 0.892*** 0.943***
(0.020) (0.103) (0.102) (0.102) (0.102) (0.103) (0.206) (0.270) (0.227) (0.206) (0.216)
priordrugtrials_samephase 0.012*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005*** 0.005***
(0.002) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) (0.000) (0.001)
no_investig 0.012* -0.008 -0.008 -0.008 -0.007 -0.009 0.013* 0.012 0.013 0.014* 0.013
(0.007) (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) (0.009) (0.008) (0.008) (0.009)
ln_no_pubs -0.031** -0.043*** -0.037*** -0.038*** -0.027* -0.039*** -0.024* -0.046*** -0.029** -0.020 -0.031**
(0.014) (0.013) (0.011) (0.012) (0.016) (0.011) (0.012) (0.017) (0.013) (0.019) (0.013)
d_industry -0.031 -0.029 -0.029 -0.029 -0.028 -0.028 -0.035 -0.030 -0.030 -0.038 -0.035
(0.028) (0.027) (0.027) (0.027) (0.026) (0.027) (0.031) (0.034) (0.032) (0.031) (0.032)
d_bigpharma 0.095 -0.091 -0.084 -0.090 -0.087 -0.099 -0.064 -0.055 -0.053 -0.056 -0.079
(0.079) (0.079) (0.079) (0.079) (0.078) (0.079) (0.089) (0.096) (0.091) (0.089) (0.092)
sponsor_trialexp -0.000 0.000 0.000 0.000 0.000 0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
sponsor_disexp 0.000 0.000 0.000 0.000 0.000 0.000 0.000** 0.000*** 0.000** 0.000* 0.000**
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
enrollment 0.000 0.000 0.000 0.000 0.000 0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
location_count 0.002 0.001 0.001 0.001 0.001 0.001 0.001 0.000 0.001 0.001 0.000
(0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.002)
top10site 0.000 0.001 -0.002 0.000 -0.000 -0.002 0.000 -0.022 -0.015 0.001 -0.001
(0.047) (0.044) (0.044) (0.044) (0.044) (0.044) (0.051) (0.056) (0.053) (0.051) (0.053)
USA -0.048* -0.099*** -0.097*** -0.099*** -0.098*** -0.098*** -0.091*** -0.084** -0.088*** -0.089*** -0.091***
(0.026) (0.027) (0.027) (0.027) (0.027) (0.027) (0.031) (0.033) (0.032) (0.031) (0.032)
Constant 0.159** 0.146** 0.164*** 0.147** -0.099 0.174*** 0.055 0.100 0.160** -0.192
(0.064) (0.059) (0.063) (0.061) (0.116) (0.067) (0.094) (0.075) (0.070) (0.175)
Observations 2,227 2,227 2,227 2,227 2,227 2,227 2,227 2,227 2,227 2,227 2,227
r2_p 0.180
F 16.98 18.18 17.29 16.77 15.82 13.09 9.855 11.64 11.41 9.395
Note: Phase and Field Fixed Effects included.