Using Users: When Does External Knowledge Enhance Corporate Product Innovation?

Transcription

1 Using Users: When Does External Knowledge Enhance Corporate Product Innovation? Prior research on corporate innovation highlights the importance of accessing external knowledge from other firms and universities. However, survey evidence indicates that product users are perhaps the most important source of external knowledge. We build on existing theory to identify the conditions under which user knowledge contributes to corporate innovation and when the benefits will be greatest. Using a panel dataset of medical device companies and their collaborative efforts with innovative physicians, we find evidence that inventive collaborations with users enhance corporate product innovation, and that the benefits are greatest in new technology areas and in the generation of radical innovations. Keywords: innovation strategy, knowledge sourcing, open innovation, health care strategy, intellectual property strategy, R&D management 1

2 Introduction Collaboration between medical device firms and the physicians that use their products is economically significant and controversial. For example, according to the U.S. inspector general, four leading orthopedic device makers spent $800 million on physician consultants between 2002 and Firms claim that physicians are critical sources of knowledge and feedback that enable the development of innovations, while regulators and policymakers worry that these arrangements create conflicts of interest where physicians are compensated in return for using or recommending a particular medical device. These concerns led to a Department of Justice investigation of the U.S. orthopedic industry in and spurred new rules governing transparency of these relationships in the 2010 Affordable Care Act. 3 We examine this important phenomenon through the lens of the academic literature on knowledge management and innovation, with special emphasis on user innovation. In doing so, we explore not only whether these collaborations with physicians increase innovative outcomes for medical device firms, but also the conditions under which the benefits are the greatest. These collaborations are just one example of the broad set of strategies firms employ to source external knowledge. A considerable amount of academic research has examined how firms manage innovation, focusing on both internal and external sources of new ideas (Arora and Gambardella 1990; Grant 1996a, 1996b; Chesbrough 2003; Karim and Mitchell 2004; Cassiman and Veugelers 2006; Phene et al. 2006; Bercovitz and Feldman 2007; Sampson 2007). Due to the limitations of developing new knowledge internally (Thompson 1965; Nelson and Winter 1982; Levitt and March 1988; Christensen and Bower 1996), accessing and integrating external knowledge is paramount (Cohen and Levinthal 1994; Rosenkopf and Almeida 2003; Laursen and Salter 2006). While most prior literature focuses on extramural knowledge from other firms and universities, and has found evidence that sourcing this knowledge is beneficial to the focal firm s innovative performance, Cohen, Nelson, and Walsh (2002) identify customers as the most important source of information for suggesting new projects. A substantial literature on user innovation (von Hippel 1988; Riggs and Von Hippel 1994; Lilien et al. 2002) explains why this 1 Testimony of Gregory Demske, Assistant Inspector General for Legal Affairs, February 27th, (Last accessed January 11th, 2013.) 2 For a summary of this investigation and its impact, see Healy and Peterson (2009). 3 For a summary of these provisions from the American Medical Association, please see: (Last accessed February 26 th, 2013.) 2

3 variety of external knowledge is unique and potentially valuable. However, few studies have estimated the impact of external knowledge from product users on corporate innovation, an important gap in the extant literature. Moreover, we lack theory and systematic empirical evidence about the conditions under which sourcing external knowledge from users will be most beneficial for a firm. To address these gaps in the literature, we integrate theoretical insights from the literature on knowledge management and innovation to generate predictions about how and when knowledge from product users increases firms innovative performance. We specify conditions under which external knowledge from users may be especially beneficial, establishing important contingencies to guide future work on sources of innovation. To test our predictions, we examine the impact of U.S. medical device firms inventive collaborations with product users on firms innovative performance, in the form of new products. The key product users, physicians, often have valuable and unique knowledge that can help medical device firms develop new products. As summarized by Marybeth Thorsgaard, a spokeswoman for Medtronic, [t]he products we develop and manufacture cannot be invented by trying a new formula in a lab like in the pharma industry. They must be designed and produced in close collaboration with the men and women who will use them: the world's most highlyskilled and innovative doctors and surgeons. 4 Despite the contentious debate over the benefits and risks of relationships between medical device firms and physicians, we have little evidence on the value of these collaborations. We utilize a new dataset covering an unbalanced panel of 128 publicly owned medicaldevice firms in the United States from 1985 to We examine the effect of prior collaborations with physicians, in the form of co-invented patents, on the number of products approved by the U.S. Food and Drug Administration (FDA), a proxy for innovations. Our results demonstrate that firm collaboration with physicians is associated with an increase in firm innovation, as expected. A one-standard-deviation increase in firm-physician collaborations is associated with 17 percent more innovations. In addition, we find that the benefit of collaboration depends on the maturity of the technology area; collaborations in new technology areas are associated with performance benefits, while those in established technology areas are 4 Moore, Janet. Medical Device Payments to Doctors Draw Scrutiny. Minneapolis Star Tribune, September 8th,

4 not. Further, we find evidence that collaborations with physicians are especially valuable for developing radical innovations. Our results are robust to additional analyses that account for potential endogeneity in the timing and level of firm-physician collaborations and potential complementarities between physician and non-physician inventions, factors that are typically not addressed in the literature in this domain. In the next section, we review the relevant prior theory and develop three testable hypotheses. We then introduce our dataset and empirical context and conclude with a discussion of the implications of this research. Theory and Hypotheses Sources of knowledge for innovation While firms certainly benefit from an established base of internally developed knowledge, this same knowledge and the existing organizational practices within the firm can inhibit product innovation (Nelson and Winter 1982; Anderson and Tushman 1990). Scholars have concluded that an organization s prior experience may constrain internal development of substantial novel inventions and innovations (Nelson and Winter 1982), corporate bureaucracy (Thompson 1965), competency traps (Levitt and March 1988), and existing customer preferences (Christensen and Bower 1996). As a consequence, established firms may struggle to identify and develop new ideas internally (Henderson 1993; Dushnitsky and Lenox 2005). Increasingly, scholars and practitioners are documenting that valuable knowledge may reside outside of the firm (Cohen and Levinthal 1990), and that accessing and integrating this knowledge is critical to firms innovative performance (Rosenkopf and Almeida 2003). Innovations, especially the most novel and important innovations such as drug-eluting stents or a bone cement to treat spinal fractures, are formed from the recombination of diverse knowledge (Fleming 2001; Rosenkopf and Nerkar 2001), which often requires new knowledge from outside the firm. How firms access extramural ideas and combine knowledge across organizational boundaries, whether drawing from regional networks, other firms, or universities, has been the subject of a substantial recent literature (Mowery 1983; Saxenian 1990; Mowery et al. 1996; Powell et al. 1996; Almeida and Kogut 1999; Stuart 2000; Ahuja and Katila 2001; Cohen, Nelson and Walsh 2002; Grant and Baden-Fuller 2004). This literature has been rightfully influential on studies on corporate innovation, but there are opportunities to extend this work. Most importantly, this literature has not fully incorporated 4

5 product users as an important and unique source of external knowledge, despite considerable evidence that users generate valuable knowledge (von Hippel 1988; Lilien et al. 2002; von Hippel 2005). Below, we attempt to synthesize key ideas advanced by user innovation scholars that are directly relevant to the conditions under which external knowledge is most beneficial for corporate innovative performance. We first review selected prior literature that documents the extent and importance of innovation from professional users and hobbyists in order to draw out the factors that make knowledge generated by users valuable and distinct from corporate knowledge. Based on this theory, we discuss important contingencies that influence the value of user knowledge to corporations, the key contribution of this paper. Then we describe the new product development process in the medical device industry, highlighting the role of physician innovation with a brief case study. Users and the product development process: prior evidence Early scholarly work on innovation generally assumed that producers would generate innovations and benefit from commercializing them (Schumpeter 1934). Eric von Hippel (1976, 1986, 1988) and co-authors identified that users, individuals or firms that benefit from using a product, could also be an important source of innovation. Their subsequent studies have confirmed that user innovation is widespread and significant (von Hippel 1988, 1998). This stream of research has found that percent of important inventions across a wide variety of industries were generated by users (von Hippel 1988). Chatterji and Fabrizio (2012) find that patented corporate inventions that include contributions from users have different attributes, in terms of quality and breadth, than other corporate patented inventions, implying that unique and valuable knowledge resides with product users. In addition, survey evidence indicates that customers often provide important insights for new R&D projects, and contribute substantially to the completion of existing R&D projects (Cohen, Nelson and Walsh 2002). Why can users generate ideas that are so distinctive and valuable? Prior work indicates that this phenomenon arises because communities of users have different motivations and knowledge than incumbent firms (von Hippel, 1986; Riggs and Von Hippel, 1994; von Hippel, and Krogh 2003; Luthje et al., 2005; Shah, 2006; Gächter et al., 2010). Users are motivated by trying to meet unmet needs they have identified through experience and enhancing their reputation in the community of users (Shah 2006). For example, users are more likely to focus on 5

6 improving product functionality for themselves and others in the community of users rather than on selecting projects for commercial viability (von Hippel, 2005). These communities can also encourage the development of prototypes, provide feedback on early inventions, and facilitate the wider adoption of a new product (Hienerth and Lettl, 2011). Incumbent manufacturers, on the other hand, are more likely to invest in innovations that can quickly be brought to the mass market. Thus, distinct motivations drive substantial differences between the inventions generated by users and those generated by established firms. Product users also possess knowledge that is fundamentally different from the knowledge developed by researchers within firms. Users experience a product s functions and limitations firsthand. These experiences may uncover problems that manufacturers did not anticipate and may also suggest potential solutions or improvements that are relevant to other users. Indeed, Cohen et al. (2002) found that customers are the most important source of information suggesting new projects, more so than a firm s own manufacturing operations. While the standalone value of user innovations has been well explored, we know far less about the contribution of user knowledge to corporate invention and innovation. Select papers have documented the contributions of product users to corporate innovation process (von Hippel et al. 1999; Jeppesen and Frederiksen 2006) and suggested that firms can create strategies to encourage user contributions to corporate innovation (Jeppesen and Molin 2003). These papers lay a foundation for future work by explaining what motivates individuals to participate in firmsponsored user communities (Jeppesen and Frederiksen, 2006) and providing practical insights for managers looking to identify promising users (von Hippel et al., 1999). Other papers explore how the work of users can benefit corporations more indirectly. For example, Baldwin et al. (2006) argue that users often initiate production processes that require little capital and have high variable costs, setting the stage for entry by incumbent manufacturers later on in the technology life cycle. However, there is no systematic evidence regarding the impact of user contributions on corporate innovation outcomes or when collaborations with users are most beneficial for firms. Furthermore, despite the suggestion that users are sources of valuable knowledge for corporations, an influential literature suggests that relying on the typical customer s experience will actually inhibit innovation (Hamel and Prahalad 1991). Rather than being constrained by tyranny of the served market (Hamel and Prahalad, 1991:83), this research argues that 6

7 managers seeking to develop innovations must nudge their customers toward the new products, not just cater to existing customer preferences. This argument is consistent with Christensen (1997), who argues that firms that rely on existing customers are more likely to develop sustaining innovations than disruptive innovations. This literature reinforces the notion that understanding when to engage users is paramount. These two views of the value of working with product users can be reconciled somewhat by consulting studies that emphasize the type of user that is valuable to firm innovation. The diverse user innovation literature has explored both professional users, who innovate in their main occupation (Riggs and von Hippel, 1994; Jeppesen and Frederiksen, 2006; Chatterji and Fabrizio, 2012), and avid hobbyists (e.g. Dahlin et al. 2004). Even within these groups, scholars have tried to identify lead users, who experience needs ahead of the rest of the population, are most likely to benefit from innovation, and are most likely to generate innovations (von Hippel 1986; Urban and von Hippel 1988; Luthje and Herstatt 2004; Lettl et al. 2006; Hoffman et al. 2010). These lead users are also the least likely to be constrained by functional fixedness (Lilien et al. 2002) that inhibits creative problem solving. The users in our empirical setting are all professionals and share many characteristics with the lead users described in prior research, because we examine inventive collaborations between practicing physicians and medical device companies. There is limited empirical evidence about the role that professional users play in the medical device industry. Lettl et al. (2006) describe four case studies where inventive users make significant contributions to radical innovations in the medical equipment industry. These authors find evidence that users develop valuable networks to broadly disseminate their innovations, a boon for established manufacturers. Their findings also suggest that professional users, specifically practicing physicians, are in a stronger position to develop valuable innovations, relative to the typical hobbyist in another industry setting. These findings are consistent with Chatterji et al. (2008), who demonstrate that 20 percent of patented inventions in the medical device industry come from practicing physicians, the key product users in this industry. Chatterji and Fabrizio (2012) find that these patents are broader and more significant in terms of the pattern and number of citations received than non-user patents. While these papers imply that practicing physicians have important ideas, no systematic evidence on their 7

8 contribution to corporate innovation is yet available, nor have possible contingencies been considered. There are a variety of mechanisms by which users can contribute to the firm innovative performance. First, in their role as customers, users can provide insights about what features they desire and the most effective sales and marketing strategies. Firms that incorporate these insights into their product development plans may develop more (or different) products than they would have otherwise. Second, influential lead users can play a special role in certifying products and recommending them to others, potentially leading to more sales of existing products but not necessarily the development of innovations. Finally, users can engage with companies to codevelop inventions. Because these inventions reflect users valuable and distinct knowledge and insight, they are more likely to be robust to the remaining hurdles in the product development cycle, leading to innovations that the firm would not otherwise have developed. While this list of user contributions is neither exhaustive nor mutually exclusive, we focus this paper on establishing the impact of upstream, invention-focused collaborations between firms and product users. These inventive collaborations are much more likely to involve valuable knowledge transfer from physicians to firms, as opposed to collaborations focused on marketing and sales. In adopting this narrow lens, we likely understate the full value of users to the corporate innovation process. However, the benefit of this approach is that we can examine the impact of identifiable user knowledge contributions on firm innovative outcomes. Taken together, our argument is that user knowledge, like other knowledge external to the firm, can enhance the ability of firms to generate innovations. The ideas conceived by product users spring from particular motivations, experience, and knowledge sets that are difficult for firms to replicate. A firm s own innovations are conditioned by its own accumulated experience with research, development, and existing products. Attempts to bring users in-house would likely destroy the value of their contributions because it would undermine the users distinctiveness. Therefore, in order to exploit potentially valuable user knowledge, firms manage these collaborations across the firm boundary, akin to alliances and corporate venture capital investments. When firms and product users collaborate on inventions, they can conceive and develop valuable and novel ideas, which are commercialized as innovations. Thus, we propose: H1: Inventive collaborations with product users will increase corporate innovative performance. 8

9 When is user knowledge most beneficial? The issues raised in prior work underscore that important contingencies dictate the value of usergenerated knowledge to corporate innovation (Baldwin, Hienerth and Von Hippel 2006; Hienerth and Lettl 2011). We move beyond the prior literature to identify the conditions under which sourcing external knowledge from users will be most beneficial. Given that organizing innovative activity across the firm boundary increases the difficulty of coordination, communications, and knowledge integration (Grant 1996b), it is critical to examine when the benefits of accessing external knowledge are more likely to exceed these costs. We focus on two instances where we expect user knowledge to be beneficial: (1) in the early phases of technology development, and (2) in the process of developing radical innovations. In both cases, we argue that users are likely to possess knowledge that facilitates corporate innovation and that is difficult for firms to replicate. New technology areas According to the prior literature on industry and product life cycles, 5 in the early period of the cycle, knowledge is distributed unequally. This era of ferment (Anderson and Tushman 1990:604) or entrepreneurial regime (Winter 1984:295) is marked by investigations of various ideas and new entrants as the industry seeks to converge around a particular standard. In this period, the knowledge required to develop new ideas is typically not embedded in existing routines (Agarwal and Gort 2002), and it is not yet clear which characteristics are most important to product users (Dosi 1982). Building on this work, we propose that external knowledge is most valuable at the beginning of the product life cycle, when a technology area is new. Because new technology areas are characterized by greater uncertainty about ideal product attributes and consumer preferences, users can represent isolated pockets of valuable knowledge. In this stage, product users might have insights about what will become the most salient product characteristics, helping the firm to understand the nature of consumer demand and potential customer 5 Some studies refer to product life cycle while others use the term industry life cycle. Our focus is on differences in the impact of user collaborations across the life cycle phases of the medical device sector, and we use the term product life cycle, consistent with Klepper (1996). 9

10 preferences, and to estimate the market size. At the earliest stages of the industry and product life cycle, it will also be the most difficult for firms to replicate outside knowledge in-house (Shah and Tripsas 2007; Tripsas 2008). Therefore, in the early stages, users have knowledge that is useful in the innovation process that is not available either within the firm or from other sources. This is consistent with existing evidence that user inventions in the medical device sector tend to occur earlier in the product life cycle (Chatterji and Fabrizio, 2012). As the cycle progresses, a dominant design emerges (Utterback and Abernathy 1975; Anderson and Tushman 1990), firms and customers develop a stronger and more uniform sense of the market, and knowledge diffuses across the industry. Next, in the retention stage (Anderson and Tushman 1990), firms develop complementary assets to support commercialization and new inventions become less frequent. During these later stages, we expect external knowledge to become comparatively less valuable to corporate innovation because knowledge is widely diffused, more standardized and codified, and more easily replicable. As product characteristics become standardized and widely adopted, and reviews of existing products diffuse knowledge of consumer reactions to product features, firms comes to understand the market, existing technology, and customers desires. In these later stages, either firms will have developed the knowledge internally, through experience and research, or such knowledge will be accessible through patents, trade publications, and other means. Thus, we expect that user knowledge will be less beneficial for firm innovation in older technology areas. H2: Inventive collaborations with product users will increase corporate innovative performance more in newer technology areas than in older technology areas. Development of radical innovations Firms are less likely to possess the knowledge required to generate radical innovations, as opposed to incremental innovations, and thus user knowledge will likely be more beneficial in the former case. We will use the terms radical and incremental according to the definitions provided by Henderson and Clark (1990:9): Incremental innovation introduces relatively minor changes to the existing product, exploits the potential of the established design, and often reinforces the dominance of established firms, and Radical innovation in contrast is based on a 10

11 different set of engineering and scientific principles and often opens up whole new markets and potential applications. Innovation based on the recombination of established, familiar knowledge components is less likely to be a significant breakthrough (Henderson 1995; Fleming 2001; Rosenkopf and Nerkar 2001). By drawing on areas of established expertise within the firm, firm researchers exploit the learning from prior research activities (both successes and failures) and develop innovations that are technologically proximate to prior innovations (Helfat 1994; Stuart and Podolny 1996). The incremental innovation process exploits and reinforces the accumulated knowledge within the firm and fits within the firm s established organizational routines. At the other end of the spectrum, innovations that require new recombinations of diverse knowledge components are more likely to be either breakthroughs or failures (Cyert and March 1963; Fleming 2001; Katila and Ahuja 2002). These novel recombinations may require different internal processes and incentive structures that diverge from the firm s established practices. Such organizational changes are difficult for an established firm and may be inconsistent with other firm activities (Henderson and Clark 1990). Most importantly for our case, novel recombinations require access to diverse and divergent knowledge (Nelson and Winter 1982; March 1991), which may not exist within the firm (Cohen and Levinthal 1990). It is important to note that if we only observe successful innovations those that reach at least some lower threshold of expected value such that firms pursue their development we will not observe the failures that result from new combinations of diverse knowledge. Instead, if the line of reasoning above holds, we will observe that radical innovations are associated with combinations of distinct sets of knowledge and that incremental improvements are associated with combinations of established, local knowledge. Since users knowledge is distinct from what the firm can develop internally, we predict that incorporating knowledge from product users is more beneficial for the production of radical innovations than incremental innovations. At first glance, this prediction may appear to contrast with prior literature that has suggested that user innovators most often generate incremental innovations (Luethje et al., 2005; von Hippel, 2005), but our argument is not inconsistent with these findings. First, inventive physicians, similar to professional users in other settings, possess characteristics that make their contributions particularly useful for generating radical innovations (Lettl et al., 2006). Second, our arguments are about when collaborating with users will benefit corporate innovation most 11

12 significantly. If user-inventors working with firms are mostly contributing incremental ideas, the firm likely has the same overlapping knowledge, so collaboration will not provide a substantial increase to corporate innovative performance. However, in the (perhaps small fraction of) cases when the users provide the kind of insights that leads to radical innovations, the firm would not have had access to this valuable and unique knowledge without the user. In these cases, the impact on corporate innovation will be substantial. Therefore, regardless of whether the typical user innovation is incremental or radical in nature, we argue that the contribution of user knowledge to firm innovation is larger for radical innovation. H3: Inventive collaborations with product users will increase corporate innovative performance more with respect to radical innovations than incremental innovations. Empirical context: collaborations between physicians and medical device firms The medical device industry is R&D-intensive and characterized by several well-established incumbent firms and thousands of small ventures that manufacture a wide variety of medical devices, instruments, and diagnostics across numerous medical specialties. 6 While similar to the pharmaceutical industry in some ways, a key difference is that the product cycles are generally much shorter in the medical device industry. 7 Furthermore, intellectual property rights to inventions in the medical device industry are especially important; according to a recent survey (Cohen et al., 2002), the importance of patents for securing rent appropriation is even greater in medical devices than in pharmaceuticals. This industry is well suited for studies about innovation since the intermediate knowledge relevant for the innovation process is clearly defined and identifiable using records of patented inventions. Although patents do not capture all knowledge used in innovation development, the high degree of appropriability via patents in the medical device industry strongly supports our use of patents to create measures of knowledge and collaboration. Finally, the data on FDAapproved medical devices provides a reliable record that is a reasonable approximation of innovation outcomes. 6 Advanced Medical Technological Association, AdvaMed Website, ( Last accessed June 5, The Food and Drug Administration Website, ( Last accessed June 5,

13 Most importantly, the medical device industry provides an ideal setting in which to study users contributions to industrial innovation. First, there is a well-identified set of users (physicians) who are part of a professional community of practice. Users in this industry are highly trained, participate interactively in user communities, and develop experience with products through use. These attributes create an environment in which users are keenly aware of existing problems, possess the knowledge to generate potential solutions, and have insights into future market needs that, taken together, can support the generation of potentially valuable inventions and innovations. 8 Innovation in the medical device industry involves interaction between physicians and device companies at all stages of development (Gelijns and Rosenberg 1994). From product conception to clinical testing to dissemination, medical device companies devise strategies to tap the knowledge of their most important customers: practicing physicians (Chatterji et al. 2008). In this paper, we focus exclusively on physician-firm inventive collaboration at the earlier stages of product research and development, and not on product dissemination, marketing, endorsement, or other activities. There are two general scenarios through which physicians and companies collaborate on inventions (Carlin 2004). In the first scenario, a physician or team of physicians will patent a new invention that generates interest from a medical device company. If both parties agree, a license or a transfer of patent rights from inventor to company can be arranged. In a famous example of this case, Dr. Thomas Fogarty, a prolific medical device inventor, licensed the patent for his revolutionary balloon catheter to Edwards Life Sciences (White 2006). This example has parallels to previous academic work on star scientists (Zucker and Darby 1996). While these arrangements are quite common in the industry, they do not always represent collaborative interactions between physicians and companies, because the physician may have developed the idea independently before ever engaging the firm. Under the second scenario, the physician inventor will consult or co-develop an idea with a medical device company, resulting in a patented invention (Carlin 2004). According to our discussions with industry experts, these collaborations can be connected to a longer term 8 It is important to note that these conditions are not specific to the medical device industry. Other studies, including Riggs and von Hippel (1994) and Shah and Tripsas (2007), have demonstrated the value of user inventions in industries as diverse as juvenile products, sports equipment, and scientific instruments. 13

14 consulting agreement whereby a contracted physician automatically assigns any resulting intellectual property to the firm. In these cases, any resulting patents will list the physician as an inventor but will be assigned to the medical device company. Co-invention without a long term consulting agreement is also common, though these arrangements will typically have a narrower scope of work and concrete deliverables. Industry experts agree that the appearance of a physician on a company patent has significant ramifications and would thus not be observed without substantial contributions from the individual in the development of the idea. It is this type of inventive collaboration, where the company is listed as the assignee and a physician appears as an inventor on the patent, that we examine in our paper. This allows us to systematically track inventive collaborations with observable data and provides a measure that is closest to the phenomenon of sourcing external knowledge described in the prior literature. The case of stent development at ACS To illuminate the context further, we offer a brief example of a fruitful physician-firm collaboration involving Dr. Richard Stack, a renowned cardiologist, and Advanced Cardiovascular Systems/Guidant on the bio-absorbable stent. 9 Stents are medical devices used to prop open blocked arteries, often during percutaneous coronary intervention. In contrast to openheart surgery, where a large incision is made through the patient s breastbone, stents offer a minimally invasive alternative procedure by allowing the device to be threaded through a small incision in the femoral artery and navigated to the blocked vessel. Dr. Stack was the founder of the interventional cardiology department at Duke University and invented an early version of the bio-absorbable stent in At the time, there were two competing stent designs, the Sigwart self-expanding stent and the Palmaz balloon expandable stent. Stack s design was arguably ahead of its time. It was made of a polymer that eventually dissolved into carbon dioxide and water after effectively propping open the affected artery. This feature would prove to be crucial years later, but Stack s initial stent was not as strong and malleable as the metal alternatives. Johnson & Johnson used Palmaz s design to develop its blockbuster bare metal stent in 1994, and later generations of stents added a drug-eluting coating intended to reduce the relatively high rate of re-blockage of the artery. 9 This section is based on a March 2010 interview with Dr. Stack and information from secondary sources. 14

15 During this period, Stack collaborated with a California-based company, Advanced Cardiovascular Systems (ACS), to further develop his bio-absorbable stent. (He later left Duke as a professor emeritus to found a string of successful medical device companies.) As part of this collaboration, Stack worked with ACS researchers to develop patented inventions that list both Stack and ACS employees as inventors. 10 Stack noted that ACS funded some of his earlier research through Duke and that physician-firm collaborations could be initiated by either side. On one hand, physicians may generate an idea and shop it around to medical device companies or venture capital firms. On the other hand, medical device firms send representatives to medical research conferences and are often well aware of promising research in academic institutions. Moreover, company sales representatives often build strong relationships with doctors and can easily list the most innovative doctors in their region. Collaborations could originate from the efforts of a firm, a physician, or even an intermediary at a medical device incubator. Stack possessed years of clinical experience in cardiology and a deep understanding of the unmet need namely, to provide a minimally invasive solution without leaving a foreign object inside the patient. He was motivated by his conviction that using existing stents was akin to placing a cast on a broken arm, but not removing the cast when the bone healed. In addition, he had seen that leaving a foreign object inside the body led to increased risks of clotting, generating considerable controversy over the safety of bare metal and drug-eluting stents. Stack s unique knowledge and motivation to address an unmet need allowed him to envision and build a potential solution: the bio-absorbable stent. In addition, the foresight that he accumulated through clinical experience led to later innovations in stent delivery. As a leader in the cardiology community, he knew that a stent that performed just as well or better than those currently on the market, but reduced the risk of clotting (by removing the foreign object from the body), would be popular with physicians, patients, and hospital administrators. In the subsequent section, we describe a large dataset that will help us understand if collaborations like this benefit medical device firms on average, and when these benefits are most apparent. Data and Methods The goal of our empirical analysis is to estimate the impact of firm inventive collaborations with physicians on new product innovation at the firm level. Specifically, we explore the variation in 10 ACS was later acquired by Eli Lilly, spun out as Guidant in 1994, and sold to Abbott as part of the Boston Scientific acquisition of Guidant in

16 levels of firm co-inventions with physicians over time for a given firm and compare the associated change in new product approvals. The null hypothesis is that such collaborations with physicians will not substantially influence innovative outcomes of the firm, which would be the case if co-inventions with physicians only served as a means to incentivize physicians to use or promote the firm s products (as often alleged by industry critics concerned about conflicts of interest), or if user involvement in the innovation process was a hindrance to innovation, as claimed in some of the literature reviewed above. For Hypotheses 2 and 3, the null hypothesis is that the impact of firms collaborations with physicians on innovative performance does not vary by the age of the technology area or the radicalness of the innovation. This is a viable alternative if one believes that users contribute only incremental ideas, relating primarily to established technologies, or contribute equally across vintages of technologies. The ideal experiment to test our predictions would involve randomly assigning firms to collaborate with inventive physicians in randomly selected years, and observing the change in innovative performance in the subsequent period, relative to before the collaboration. Obviously, such an experiment is not possible. Because firms may choose whether and when to work with physicians, and physicians can select which firms to work with, collaborations are not random. A pooled OLS analysis would therefore provide correlations reflecting both the patterns of selection into collaboration and the effects of collaboration. As described below, we use panel data so that we can compare the innovative performance of each firm over time, and examine how innovative performance changes as the firm collaborates with physicians (or does not). However, the collaboration itself is a choice of the firm. This creates an empirical challenge: firms will elect to collaborate with physicians when it is beneficial for them to do so, and will collaborate more when they benefit more from doing so. In other words: collaboration with physicians is endogenous. We discuss the various dimensions of this challenge, mapping to specific sources of endogeneity, and provide analyses to address it in the Empirical Challenges section below. First, however, we discuss our baseline empirical model. As described in detail below, we make use of patent data and also compile FDA data on products approved through the 510(k) and PMA processes. Following Hausman et al. (1984) and Griliches (1990), we employ a production function model to estimate the elasticity of innovation (output) to knowledge inputs including R&D investment, accumulated knowledge stock, and firm employment. To test our hypotheses, we also include physician collaborations, i.e., the 16

17 count of patents co-invented with a physician, as a separate knowledge input. Our maintained assumption is that after controlling for these knowledge inputs and characteristics of the firm, as well as a firm-specific fixed effect, 11 there are no unobserved, time-varying factors that drive innovative performance that are also correlated with physician collaborations. As in Jaffe (1989), the Cobb-Douglas production function is as follows: where I i is the innovative output of firm i, DrPats is the count of patents co-invented with a physician, R&D is the firm s current expenditures on research and development, Knowl.Stock is the firm s accumulated stock of knowledge, and Employ is the number of employees of the firm, a measure of firm size. Taking the natural log of both sides yields the equation to be estimated: We estimate this model using Poisson quasi-maximum likelihood estimation. The primary hypothesis, that collaboration with physicians increases innovative performance, would be supported if is positive and statistically significantly different than zero. To test the prediction that collaboration with physicians in newer technology areas is more beneficial for firm innovation (Hypothesis 2), we categorize each physician co-invention according to the age of the technology class to which the patent is assigned, as described below. To test the prediction that physician collaboration is more useful for development of radical, rather than incremental, innovations, we separate the FDA-approved innovations ( ) into those requiring PMA approval (radical technologies) and those approved through the 510(k) process (incremental technologies), as described in detail below. We estimate the model separately for these two outcome measures. Sample We constructed a large, unbalanced panel of public medical device companies in the United States with data from 1985 to The sample includes all public firms in the primary medical device Standard Industrial Classification (SIC) code that were granted at least 10 patents between 1980 and This approach purposefully excludes large conglomerates and firms that 11 A Hausman test rejected the appropriateness of the Random Effects model at the 1% level. 17

18 are primarily pharmaceutical firms in order to focus on medical device firms. 12 This method also intentionally excludes firms with only rare patented inventions in order to narrow our study to firms that are pursuing an innovation-based strategy. In practice, because patenting is so critical to appropriability in this industry and innovations drive revenue, all firms patent heavily. Firms with very few patents tend to be small and private, and therefore would not be in the sample of public firms in any case. In order to avoid right truncation associated with not observing patents that have been applied for but not yet granted, we only include patents through 1997 in the analysis. Likewise, our identification of physician inventors extends only back to 1980; in order to include controls for knowledge and product stock, based on the prior five years, we cannot use data before 1985 in the analysis. The resulting dataset includes 128 firms and 803 firm-year observations. The dataset includes information on the firms FDA-approved products from the Center for Devices and Radiological Health (CDRH), granted patents from the Hall et al. (2001) dataset, and firmyear-level data from the Standard & Poor s Compustat database. In order to identify collaborations with physicians, we relied on the American Medical Association (AMA) Masterfile data, which includes biographical information for all licensed physicians in the U.S. We constructed our dataset as follows. For the firms meeting the criteria described above, we identified all successful U.S. medical device patents applied for between 1980 and 1997 using the NBER patent data (Hall et al. 2001). We used that data to identify all inventors on each patent, including inventors first, middle, and last names and their locations by cities and states. The AMA Physician Masterfile contains the name, demographics, address history, practice type, and medical school information for all licensed U.S. physicians. Using this data, we can identify which inventors in our sample of medical device patents were physicians. The matching process involved several steps. We initially identified any physicians with the same first and last name and state location as an inventor in our sample. To do this, we relied on the physicians historic and current locations in the AMA data and the inventors addresses in the patent data, matching the timing of the patent application to the years of licensure at each 12 Including conglomerate firms would necessitate using firm-level data on number of employees and R&D expenditures. Assuming that a significant portion of employees and R&D spending was allocated to activities not related to the medical device industry, including these firms would bias the coefficients on these control variables down and inflate the error terms for the focused medical device firms. Depending on the correlation between physician collaboration and the significance of non-medical device industry activity, this could bias the coefficient on physician collaborations either up or down. In order to avoid these problems, we limit the empirical analysis for firms whose primarily industry is medical devices. 18

19 address. After identifying potential matches, we examined them more closely to assure a legitimate match. For each record, if there was a middle name or initial available from both the patent data and the AMA data, we either confirm that these records are complete matches or eliminate them from our list. When one or both of the middle initial observations were not present, we cross-checked records by city location in both files. In those instances where observations did not have middle name data and did not match on city, we explored these records more closely to assess whether it was a legitimate match, and eliminated any that we could not confirm. There are 5437 unique physician-inventors on the co-invented patents of medical device firms in our sample. Note that the physician co-inventors represent a diversity of employment roles. For example, 38 percent were in group practice, 24 percent in solo practice, and 5 percent in medical schools. The physician co-inventors also represent a diverse set of specialties. Physicians in various surgery specialties were well represented, collectively forming the largest contingent in the data (about 15%). Measures Innovations. Many studies of innovation rely on counts of patents or citation-weighted patents because they lack information on actual product introductions. We are fortunate to have systematic and reliable data on innovations. We constructed a firm-year count of the number of FDA-approvals (NumInnov) based on the CDRH data. To date the innovations, we used the year that the application was received by the FDA. Our base measure is the aggregated count of product approvals via the 510(k) and PMA process, including all classes of products. We also considered separately the count of 510(k) and PMA product approvals to test Hypothesis 3. Because our empirical approach relies on identifying radical innovations, it is important to briefly summarize how regulators classify medical devices. To bring a new medical device to market, the FDA s Center for Devices and Radiological Health (CRDH) must approve an application. 13 New medical device products are reviewed and approved through one of two processes: pre-market notification and pre-market approval. Approval via pre-market notification, commonly referred to as the 510(k) process, requires demonstration of substantial equivalence to a device currently on the market, called a predicate device. This is intended for less risky devices that are similar to devices proven safe based on a history of sales and use. 13 Although it is technically correct to say that the application rather than the product itself has been approved, we will refer to product approvals throughout the paper for simplicity. 19

20 Devices approved through the 510(k) process are thus intended to be incremental and are often modifications to existing products. The approval process is simplified and often takes less than three months (Singh 2007). As a result, products on the 510(k) track can sometimes be brought to market in as little as a year from conception (Lawyer et al. 2007). The total investment required ranges between $10 million and $20 million. 14 The most novel devices, for which no equivalent device exists, undergo the much more rigorous pre-market approval (PMA) process, which involves animal testing and human clinical trials (Singh 2007). Recent research has found the average review time for a PMA application in was approximately 409 days, and significantly longer for orthopedic devices (Singh 2007). The total investment required to complete the PMA process ranges from $30 million to $100 million. 15 In 2006, the FDA granted 39 PMAs and (k)s, a ratio of PMAs to 510(k)s roughly similar to approvals over the last several years (Lawyer et al. 2007). A recent study by the U.S. Government Accountability Office found that between 2003 and 2007, the 510(k) process had a 90 percent approval rate and the PMA process had a 78 percent approval rate. 16 Physician co-invention. Our primary independent variables of interest are firm-physician inventive collaborations in the current and prior year, which we proxy for with the count of patents assigned to the firm that are co-invented with a physician in a firm-year observation, where the year is the application year (DrPats t and DrPats t-1 ). 17 If at least one inventor on a firm s patent is a physician, we counted this as a physician co-invented patent. In practice, 83 percent of these physician co-invented patents also include non-physician inventors, presumably company employees. One potential concern is that we are missing collaborations between firms and university-based physicians if the university, rather than the partner firm, is listed as the assignee on such co-inventions. However, we found that only 7 percent of all physician-invented medical device patents were invented by physicians at medical schools. An examination of these 14 Drug Eluting Stents: A Paradigm Shift in the Medical Device Industry, Stanford Graduate School of Business Case-OIT-50, 02/13/06, Lyn Denend and Stefanos Zenios. 15 Drug Eluting Stents: A Paradigm Shift in the Medical Device Industry, Stanford Graduate School of Business Case-OIT-50, 02/13/06, Lyn Denend and Stefanos Zenios. 16 Medical Devices: FDA Should Take Steps to Ensure That High-Risk Device Types Are Approved through the Most Stringent Premarket Review Process. U.S. Government Accountability Office, January We experimented with models including up to five years of lagged physician co-inventions. Results indicate that it is co-inventions in the most recent two years that have an effect on firms innovative outcomes, a reasonable result given that product cycles in the industry can be as short as 18 months. 20