Identifying Key Opinion Leaders Using Social Network Analysis

Transcription

1 Cognizant Insights Identifying Key Opinion Leaders Using Social Network Analysis By applying a more scientific approach to finding and engaging opinion leaders, pharma companies can more effectively scale and accelerate their R&D efforts, thereby gaining an edge over their rivals. Executive Summary Key opinion leaders (KOLs) undeniably play a very important role in the life sciences industry. Pharma companies typically engage them during the later stages of clinical trials and during drug promotions. However, as regulatory approval challenges mount, many drug manufacturers are seeking deeper collaboration with opinion leaders throughout the drug development process. Identifying KOLs requires broad and up-to-date experience with these individuals, something pharma companies lack. In fact, most life science companies delegate this exercise to third parties. In turn, these providers make use of tried-andtrue methods of engaging KOLs, including literature database search and survey methods. In an age where consumers hold sway over purchase considerations and channel preferences across industries, these traditional approaches are increasingly ineffective and exacerbate the risk of experience bias. This white paper offers a different approach to identifying key opinion leaders social network analysis (SNA). SNA is based on social networks that are derived from interactions between individuals and thereby model meaningful social relationships. Besides highlighting the benefits of SNA over traditional methods, this paper demonstrates a simple application of SNA on a co-author or author-collaboration network for authors of scientific publications in the field of medical genetics. The focus of this demonstration is to identify thought leaders who could be sought for advice and research collaboration. Many of the authors shortlisted as opinion leaders (based on our analysis) have made significant contributions in the field of medical genetics. Besides having led key research projects, many of the shortlisted authors are members of advisory boards and have held leadership or executive positions in important organizations such as the National Society of Genetic Counselors and the European Board of Medical Genetics. Why Key Opinion Leaders Are Critical to Life Sciences Companies KOLs, commonly referred to as thought leaders, play a very important role in the life sciences industry, especially in the adoption and usage of new products. KOLs are typically healthcare professionals (HCPs) who hold senior positions in the medical community. However, they may also be members of the academic community who have contributed significantly to or have advanced knowledge in specific therapeutic areas. By virtue of their position or expertise, KOLs tend to have an asymmetric influence on physicians, and hence on prescribing behavior and treatment guidelines. cognizant insights june 2015

2 Engaging with KOLs on Drug Development Typical KOL Activities in Drug Development Process Research & Drug Discovery Clinical Trials Phase II and Phase III Approval Phase IV (Post-approval) Identify research areas. Provide insights on disease areas. Undertake research on drug company s behalf. Consult on clinical program design. Provide advice on product profile. Advice on regulatory/ approval process. Provide information on drug efficacy or benefits. Give presentations at congresses. Publish articles or papers. Build product awareness. Figure 1 Although involvement of KOLs in drug promotion is a contentious issue, 1 they play a vital role in other phases of drug development as well (see Figure 1). KOLs collaborate with companies on Phase III trials, influence basic research, conduct Phase IV outcomes research and assist in the identification of new research areas or product lines. They may also be involved in a drug s regulatory review process and provide information on comparative benefits and side effects of a drug. Post approval (Phase IV), most major pharmaceuticals companies deploy teams of medical science liaisons (MSLs) to respond to medical inquiries from KOLs in specific therapeutic areas. Identifying Key Opinion Leaders KOL management, which includes segmenting opinion leaders and managing relationships with them, has well-structured practices. However, KOL identification, which is the precursor to KOL management, provides a lot of scope for improvement. Most pharmaceuticals companies depend on third parties to provide KOL databases that, in turn, make use of traditional methods like literature searches and survey methods. These methods have their own characteristic limitations. Figure 2 lists some of the commonly followed traditional methods and their limitations. Traditional KOL Methods Observation Method An outsider, who is familiar with a subject area or community, identifies who are the most influential members in that community. Limitations Observer may be more biased towards well known or high profile members. Observer has limited ability to track all interactions. Hence, such methods are well-suited only to small communities for identifying regional opinion leaders. Figure 2 Survey Methods Individuals working in a given subject area are asked to provide names of others who they rely on for advice or who they think are the thought leaders in the given subject area. Alternatively, individuals may be asked to rate their peers in terms of various criteria like quality of education, degree of contribution to the subject area or quality of contributions. Limitations Expense increases as sample size increases. Suffers from sampling bias i.e., the responses and outcomes of the survey are highly dependent on the chosen sample. Surveyors need to ensure that the sample size is sufficiently large and the members in the sample are reasonably representative of the intended population. Results may also get skewed due to nonresponses. Both methods also suffer from volunteer response bias. Literature Searches Companies employ researcher(s) to search through various publication databases and other similar sources and identify and profile prominent thought leaders in a research area. The researcher typically uses search engines and other tools in order to perform an exhaustive search and go through huge volumes of data. Limitations Time-consuming as a lot of the search and analysis work is manual. Results depend on the search parameters as well as the researcher s knowledge of the research area. The criteria used for identifying opinion leaders do not take into account social information. For example, number of publications or number of conferences attended are often used as metrics. However, such metrics cannot account for relevant information like whether the publication was well acknowledged by peers or whether the author made any significant contribution to the conference. Scalability is constrained by the limited ability of researcher to search for data.

3 Using SNA for KOL Identification The limitations of the traditional methods have pushed the industry toward innovative methods such as SNA to help identify opinion leaders. In SNA, the interactions between participating actors are tracked as relationships or connections and are used to construct a social network. These relationships, which have developed over time, form the basis of collaboration and information sharing within and across communities and thus provide a collective value referred to, by Putnam, as social capital. 2 SNA leverages this information by analyzing the social structure of a network and the position of actors within a network. Consider, for example, a research community collaborating on a specific therapeutic area. This could typically involve a large community of researchers ranging in the thousands and spanning multiple geographies. The network could include various types of actors including pharmaceuticals companies, educational institutions, advisory boards and regulatory bodies. In this case, a social network could be created to capture community members collaborating on research. Analyzing this network could then provide rich contextual information. For example, analyzing the position of members in the network could provide insights into how important they are within the research community. Structural analysis of the network could also provide additional insights such as the presence of hidden cohesive groups. Such insights are generally not possible with traditional methods. A research report found that bibliometric, or literature-based, methods fail to identify over 50% of thought leaders compared to SNA. 3 Another key tenet of SNA is that it is scalable, and thus can be used on an entire community rather than on just a sample of members. Moreover, SNA is typically free from volunteer response bias as well as other bias associated with survey and observation methods (see sidebar below). Understanding SNA To apply SNA, organizations need to first understand what they are getting into. Below are the definitions we use: Social network: A social network is a special type of graph that depicts relationships between actors. The actors of interest (individuals, groups, companies, etc.) are the nodes and relationships between the actors are the arcs or ties. In 1953, Moreno first introduced the idea Why Social Network Analysis Social networks are derived from social interactions like relationships, collaboration, etc. and therefore model social prestige. It s a natural tendency in a social group to get associated to respected members of the groups, and hence respected members of a community will tend to be connected to the majority of the community members either directly or indirectly via others. Thus, social networks could provide valuable insights on thought leaders. People generally prefer to work with others who share similar interests and values. Structural analysis of a network could provide additional insights like the presence of hidden cohesive groups or clusters. Scalable: SNA can be applied to large networks, and thus can accommodate an entire community rather than just a sample of members. SNA is free from responder bias as well as other bias associated with survey and observation methods. cognizant insights 3

4 to represent social interactions in a network structure, called a sociogram. 4 Facebook is a classic example of a social network where the nodes represent the Facebook users or groups and friendships are represented as ties. Social network analysis: As explained by McGuire, this is the process of applying analysis techniques to a social network to answer specific questions about that network. Often, these questions focus around who key actors are in the social network. Other questions may be looking for groups of actors with strong ties to one another or how best to improve the communications or productivity of the group being analyzed. 5 Applying SNA: General Methodology A typical methodology for applying SNA is shown in Figure 3. Based on the end purpose that the analysis intends to serve, data is collected from various sources; it is then cleansed and processed to remove duplicates and converted into a usable format. Concept extraction: This refers to the process of extracting the nodes and ties from the data. Selection of nodes depends on the actors on which the analysis needs to focus and ties are chosen based on the nature of interaction that needs to be represented. Complex networks may include attributes or properties for nodes or ties (e.g., the age or qualification of each person or node or strength of a tie based on frequency of interaction between the participating nodes). Once the relevant data has been extracted, a network is created. Applying network measures: The objective of the analysis, as well as the nature of the network, will determine the specific measures that need to be applied to derive insights from the network. This can vary from applying simple centrality measures to complex measures that analyze how the network is structured or how it evolves with time. (For a detailed explanation on centrality measures, refer to the Appendix on page 10.) Identifying KOLs from a Coauthor s Network Using SNA This section explains the use of the centrality measures on a coauthors or author-collaboration network to identify key profiles. Data was collected for all publications in the field of medical genetics 6 during the last ten years ( ) from PubMed. 7 From this data, a network based on author-collaboration was created. The network represents a social network of who has collaborated with whom (i.e., the nodes or vertices are the authors and a tie or link between a pair of nodes implies that they have collaborated on at least one publication). The focus of our analysis is to identify thought leaders for the purpose of advocacy or research collaboration rather than for drug promotion; hence, our decision to use scientific publications (in the chosen field) from PubMed. For the purpose of drug promotion, other factors such as number of publications, willingness to deliver seminars or media presence of the individual may be taken into account. As stated above, the purpose of the analysis will inform the data to be collected and the measures to be applied. A Typical Social Network Analysis Methodology Profile Information Literature Databases Data Cleansing and Data Conversion Concept Extraction & Network Creation Data Visualization & Network Measures Derive Insights Other Sources Figure 3 cognizant insights 4

5 Frequency Distribution of Number of Coauthors per Publication Frequency Distribution of Number of Authors per Publication Cumulative (%) Distribution of Number of Authors per Publication Frequency 400 Percent Number of Authors per Publication Number of Authors Figure 4 The author-collaboration network was created using igraph library 8 for R. 9 For each publication, the authors have been extracted and each coauthor combination pair is represented as a tie. Publications that have only one author were ignored. Duplicate author names have been replaced manually. Some articles in the data set have a large set of coauthors, which tends to produce a bias in centrality measures, especially degree (an article with a large number of coauthors will result in a high degree value for each of the coauthors). To remove such bias, we may consider only those authors who make significant contributions to an article or publication. However, as such information is rarely available, we have assumed that only a certain number of authors mentioned in a publication make significant contributions. A distribution of authors per publication for our data is shown in Figure 4. In our research, 98% of publications or articles have 12 or fewer authors. For our analysis, we considered only the first 12 coauthors of any publication as significant coauthors and used them for network construction. The overall network comprises 6,151 authors and 20,908 ties among them. This network is not completely connected and has multiple isolated groups of authors (i.e., components). The igraph library in R (which has been used in our analysis to calculate centrality) makes an approximation when calculating closeness values of nodes that are not connected to one or more nodes. For the sake of simplicity, we focus our analysis only on the largest connected component of the network which has 2,517 authors and 10,375 ties (see Figure 5). Note that this component accounts for about 41% of nodes and 50% of ties of the overall network. The centrality measures degree, closeness, eigenvector centrality and betweenness were applied to the largest component subnetwork, and the authors were sorted and ranked based on each of the measures. The top 15 authors across each measure were picked Author-Collaboration Network Note: Opinion leaders are in red. Figure 5 cognizant insights 5

6 Top 15 Ranked Authors Degree Closeness Eigenvector Betweenness Author ID common to two measures Author ID common to three measures Note: Authors are represented by IDs across each centrality measure. Figure 6 (see Figure 6) and profile information for them was collected from publicly available data. For the sake of confidentiality, all authors have been identified using network ID (or author ID) and their names have been removed. It may be noted here that in our case, the top 15 rankings of authors remain the same even if the centrality measures are applied to the full network. Results As highlighted in Figure 6, some listed authors are among the top-ranked across two or three measures. However, no author is ranked in the top 15 across all four measures. The degree of overlapping across measures may vary across networks. A total of 40 unique author profiles were identified in the union set of the top 15 across all the centrality measures. Profile information was manually collected for each of the ranked authors from publicly available data on the Web. The list includes eminent people who are regarded as leaders in the field of genetics. The top authors have an average work experience of 26 years. A significant majority (60%) of the ranked authors hold a doctorate degree and many of the authors have multiple qualifications. These authors have held leadership positions in important organizations related to medical genetics, such as the National Society of Genetic Counselors and the NCI Board of Scientific Advisors. Most also head educational departments or programs in the field of genetics within their academic institutions. Many of the authors have led important research programs or have worked as principal investigators in research projects. Their contributions to the field of genetics, especially hereditary diseases (which is a major focus of medical genetics) and genetic counseling, have been highly recognized; 24 authors (i.e., 60%) have received one or more awards for their contribution to research and academics. Based on an analysis of the collected information, we found that 34 authors (i.e., 85%) could be considered to be opinion leaders in the area of medical genetics. Note: We could not ascertain with absolute certainty whether the remaining profiles could be considered to be thought leaders or not, due to lack of information. For example, one author was an assistant professor of genetics and genomic sciences, but no other information on work experience or research focus was available. Similarly, we found two authors who were certified genetic counselors, but additional information was not available. Figure 7 offers a snapshot of this profile information. cognizant insights 6

7 Ranked Author Profile Snapshots Profile Snapshot of 40 Authors Identified as Opinion Leaders Qualifications* At Least 24 MD PhD Authors Have Won One or More Awards Certified Genetic Counselor * Many authors have multiple qualifications Country-wide Distribution Average Years of Experience* * Work-ex data available for 33 of the 40 authors 7% Australia 10% Canada 2% France 3% Italy 5% The Netherlands 8% UK 65% U.S. Work Experience Summary Majority of the authors hold very senior positions in the academic community heading research groups or departments within the institutions they work for e.g. Director of the Graduate Program of Study in Genetic Counseling or Head, Genetic Services Unit and Director, Genetic Counseling Training or Senior Researcher, Team Leader. A number of authors have led or worked on research projects funded by federal or private organizations. Many of them worked as Principal Investigators or Co-investigators. Advisory Board Memberships A number of authors are key members of scientific or medical advisory boards like: Scientific Advisory Board, Lynch Syndrome, Australia Advisory Panel, Center for Disease Control, U.S. National Human Genome Research Institute, U.S. Advisory Committee for New Genetics Services, Ontario Ministry of Health and Long Term Care Medical Advisory Committee Birmingham, Alabama Key Contributions & Achievements Listed below are some of the key contributions and achievements of the 40 authors. We have deliberately removed dates and specifics to keep the authors anonymous. Organizational Affiliations The majority of the authors are affiliated with key societies or organizations in the area of genetics. Besides being members, some of the authors have held leadership roles in these bodies, including: Two past-presidents and one founding member of National Society of Genetic Counselors. One member of board of directors and one president of American Board of Genetic Counseling. Board members of Scientific Counselors, National Human Genome Research Institute. Inaugural chair of European Board of Medical Genetics. One past-president and one founding member of American College of Medical Genetics and Genomics. One of the first and key contributors to hereditary cancer research. Has a cancer related syndrome named after him. Played a major role in isolating the genes for both Huntington s disease and myotonic dystrophy and built comprehensive genetic counseling service for the whole of Wales. Played a major role in the development of national guidelines in genetics (American College of Medical Genetics). Part of the team that discovered BRCA1 and BRCA2 one of the most important breakthroughs in cancer research. Involved in the EuroGentest2 European project to improve the standard of genetic counseling, particularly regarding genetic testing. Spearheaded a multi-million-dollar genomic medicine initiative, known as the Genomedical Connection. International Award, National Society of Genetic Counselors (U.S.) Advanced Medical Leader Award from the British Association of Medical Managers Strategic Leader Award, National Society of Genetic Counselors (NSGC) Natalie Weissberger Paul Lifetime Achievement Award from the National Society of Genetic Counselors American Cancer Society s Distinguished Service Award Canadian Gene Cure Foundation Champion of Genetics American Academy on Physician and Patient Award for outstanding research contribution Figure 7 cognizant insights 7

8 Looking Forward Next Steps A simple application of SNA to a scientific publications network yielded highly relevant results for us. It is possible to apply the same approach to other therapeutic areas. We recommend that life sciences companies consider the following before embarking on their SNA journeys: Besides being empirically sound and objective, SNA offers the potential for life sciences companies (in fact, any organization) to generate a great deal of KOL insights. Many industries are already exploring how they can apply SNA to numerous opportunities as varied as ferreting out malinvestment (including fraud as well as funding for terrorist networks), to identifying key influencers in an online social network for targeted digital advertising. Going forward, we may see stronger adoption of SNA within life sciences, especially for identifying KOLs (see sidebar, page 9). cognizant insights Network attributes: Our analysis focused on a simple collaboration relationship among the authors or researchers. More comprehensive and complex analysis could involve various attributes of the network. For example, analysis may include attributes of the network members such as experience, location, academic qualifications and number of publications. Weighted networks: Our analysis did not assign any weights to the relationships between members. In other words, all relationship ties were considered equally important. However, weighted networks are possible based on variables such as geographical location of members, frequency of collaboration or strength of a relationship. Dynamic networks: Much recent study related to SNA has focused on dynamic networks (i.e., how a network evolves with time and accounting for time-based attributes). For example, considering our network, we could have assigned weights to the participation of network members with recent contributions being assigned higher value than older ones. 8

9 Applying Social Network Analysis The following hypothetical scenarios are designed to provide a taste of how SNA can be leveraged by organizations. Scenario 1: A large pharma company wants to leverage external research to develop a novel drug for obesity. The company hopes that the identified experts will contribute toward the entire drug development process from predevelopment to post-approval feedback analysis. Challenge(s): Identifying the best experts in the given therapeutic area is a huge challenge. While prior research work and contributions to publications are important criteria, the pharma company is well aware of the fact that quality of contribution is more important than quantity when it comes to research. Considering only numerical criteria such as the number of publications or conferences attended won t suffice. Approach: It is a well-accepted observation that people prefer to work with their reputed peers. As mentioned above, social networks model social prestige. The company collected scientific publications that focused on research specific to its drug area. A network of coauthors could be constructed, with additional weights assigned based on frequency as well as recency of collaboration. SNA measures could then be applied to the network to identify KOLs. This could be followed by further scrutiny of the prior work of the shortlisted profiles, and interviews could be conducted to narrow down the list of KOLs selected for research collaboration. Scenario 2: A large pharma company wants to improve the adoption rate for a recently launched drug in a particular market. The drug is used to treat a particular chronic condition, and two competing drugs have been available for the same treatment for about a year. The company believes that engaging KOLs will increase the rate of adoption of the drug. Challenge(s): Given that the drug category relates to a chronic condition, drug efficacy is not easily observable. As such, there is uncertainty in the physician community regarding the best treatment option. The lack of comparative data makes the role of KOLs in adoption of a drug even more significant and the task of identifying the KOLs even more relevant. Approach: Typically, the level of detailing required for a physician is proportional to his or her prescription volume. However, this approach is not always effective. The company could take an alternative approach by identifying physicians in the target market who have prescribed drugs that belong to the specific drug category. A market survey could be conducted among these physicians, requesting that they identify other physicians with whom they consulted for guidance or with whom they discussed treatment-related issues. Based on collected responses, a social network could be created that represented who influenced whom. SNA could be used to analyze this network and identify highly influential physicians. This could then be crossreferenced with product detailing data which could tell if influential physicians received appropriate detailing. The company could then move quickly to remedy the situation. cognizant insights 9

10 Appendix Social Network Analysis Measures Once a network has been constructed, various types of analysis can be applied. This ranges from simple network metrics, such as graph density and centrality, to complex ones related to identifying communities within a network. Widely used centrality measures include: Centrality measures: One of the most popular measures used in SNA to identify key players is centrality. It is used to determine how important a node is to a network. The four most popular centrality measures are degree, closeness, betweenness and eigenvector centrality. Of these, the first three were formalized by Freeman 10 and the fourth measure, eigenvector, by Bonacich. 11 Degree of a node is the number of edges incident on that node. For a network with n nodes, the maximum number of edges to a node is n-1. The normalized degree for a node i is explained by the following formula: h h Betweenness centrality of an actor or node is a measure of how often an actor lies on the shortest path between two other actors or nodes. It measures the degree to which a particular node controls the flow of information or resources between other nodes in a network. This measure assumes that information flows in a network along the shortest paths. If p kj represents the number of possible shortest paths between two nodes k and j and p kij the number of shortest paths between k and j that include node i, then the normalized value of betweenness centrality for node i is explained by the following formula: ( ) Closeness is a measure of being as close to as many actors as possible. McGuire explains closeness as, in this position, an actor needs to rely on the fewest number of people to send or receive a resource flowing in the network. 5 The normalized measure of closeness centrality for a node i is given by: h h h For a network that is not fully connected, i.e., if nodes i and j are not connected, d ij cannot be defined and hence closeness centrality cannot be calculated. 5 Some tools and libraries take the distance between two unconnected nodes as a large finite value (e.g., n, the number of nodes in the network). Opsahl 12 proposed an alternative closeness measure as: Eigenvector centrality is defined as the principal eigenvector of the sociomatrix of the social network. 11 This measure signifies how well connected a node is to other nodes that are themselves highly connected. Therefore, even if a node may not have many adjacent nodes, if the nodes adjacent to it are well connected then the node will have a high Eigen centrality value. The eigenvector V of a network is defined as: h h h cognizant insights 10

11 Footnotes 1 S. Sismondo, Key Opinion Leaders and the Corruption of Medical Knowledge: What the Sunshine Act Will and Won t Cast Light On, The Journal of Law, Medicine and Ethics, pp , 2013 Fall. 2 R. D. Putnam, Bowling alone: America s declining social capital, Journal of Democracy, pp , Lnx Pharma, Quantity Does Not Equal Quality in Evaluating a Scientist s Real Importance as a Key Opinion Leader, J. Moreno, Who Shall Survive? A New Approach to the Problem of Human Inter-relations, R. M. McGuire, Weighted Key Player Problem for Social Network Analysis, Air Force Institute of Technology, Medical genetics is the application of genetics to medical care. It focuses on diagnosis, management and counseling of individuals with genetic disorders including hereditary diseases. Medical_genetics. 7 PubMed comprises over 24 million citations for biomedical literature from MEDLINE, life science journals and online books. 8 igraph is a library for network analysis. 9 R is a free software environment for statistical computing and graphics L. Freeman, Centrality in networks: Conceptual clarification, Social Networks, pp. 1: , P. Bonacich, Factoring and weighting approaches to status scores and clique identification, Journal of Mathematical Sociology, pp. 2: , T. A. F. S. J. Opsahl, Node centrality in weighted networks: Generalizing degree and shortest paths, Social Networks 32 (3), pp , References E. Rogers and D. G. Cartano, Methods of Measuring Opinion Leadership, Public Opinion Quarterly, pp , S. Wasserman and K. Faust, Social Network Analysis: Methods and Applications, Cambridge University Press, S. Borgatti, Centrality and network flow, Social Networks, pp. 27:55-71, cognizant insights 11

12 About the Author Amoy Kumar Dutta is a Consultant within Cognizant Business Consulting s Enterprise Applications Services (EAS) Practice. He has over six years of industry experience and specializes in business strategy, customer relationship management consulting, sales and marketing process consulting and business development across the life sciences, travel, telecommunications and entertainment industries. He can be reached at Amoykumar.Dutta@cognizant.com. About Cognizant Cognizant (NASDAQ: CTSH) is a leading provider of information technology, consulting, and business process outsourcing services, dedicated to helping the world s leading companies build stronger businesses. Headquartered in Teaneck, New Jersey (U.S.), Cognizant combines a passion for client satisfaction, technology innovation, deep industry and business process expertise, and a global, collaborative workforce that embodies the future of work. With over 100 development and delivery centers worldwide and approximately 217,700 employees as of March 31, 2015, Cognizant is a member of the NASDAQ-100, the S&P 500, the Forbes Global 2000, and the Fortune 500 and is ranked among the top performing and fastest growing companies in the world. Visit us online at or follow us on Twitter: Cognizant. World Headquarters 500 Frank W. Burr Blvd. Teaneck, NJ USA Phone: Fax: Toll Free: inquiry@cognizant.com European Headquarters 1 Kingdom Street Paddington Central London W2 6BD Phone: +44 (0) Fax: +44 (0) infouk@cognizant.com India Operations Headquarters #5/535, Old Mahabalipuram Road Okkiyam Pettai, Thoraipakkam Chennai, India Phone: +91 (0) Fax: +91 (0) inquiryindia@cognizant.com Copyright 2015, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the express written permission from Cognizant. The information contained herein is subject to change without notice. All other trademarks mentioned herein are the property of their respective owners. Codex 1234