Tried-and-true methods of identifying and managing KOLs do not always yield optimal results. They tend to lean toward high-profile experts who are “top of mind” for the professionals entrusted by pharma and biotech companies with spotting and engaging them, and the hidden risk in this is that some of those prominent KOLs are in fact LOLs – “loud” but unimpactful opinion leaders in the clinical research ecosystem who produce more noise than signal.
The traditional approaches to KOL identification, which rely primarily on observation, surveys methods, and literature searches based on standard publishing metrics, suffer from observer and response biases and are far from unearthing the most effective potential collaborators, while at the same time overlooking social information that could turn up powerful insights. Because they tend to prize quantity – in the form of peer-reviewed journal articles, conference appearances, clinical trials management, advisory board participation, or other quantifiable scientific activity – over quality, they risk failing to distinguish between purely prolific producers of output and truly influential advocates whose work has meaningful repercussions beyond publications and the echo chamber of conference halls.
Indeed, research has shown that relying on bibliographic searches or considering only the number of publications or conferences attended as indicators of thought leadership fails to spot more than 50% of influential actors compared to social network analysis (SNA) – and this is research dating back to 2010, when SNA did not have the capabilities it has today. (The study, which questioned the dominant KOL identification methodology even then, can be consulted here.)
The “secret” to identifying KOLs using SNA is that this approach, which makes explicit the webs of relationships between thought leaders and derives insights from their centrality and connectedness to others in the network, allows for selection of criteria that are not easily mappable using traditional methods. Such criteria may include identifying KOLs with more recent publications rather than with a greater number of publications, with recency being assigned higher value than sheer volume, thus potentially reducing the impact of age or career length on KOL selection; the frequency of collaboration within a network of experts, which can be interpreted as evidence of authors´ preferred working relationships; or the social reverberation of publications, keynotes or other appearances related to a given therapeutic area, measured as the sum of substantive comments and engagements on social media.
An example of a KOL identified through SNA who may not necessarily have been captured using a literature search would be an author of a publication that is well received on social media and commented by Twitter or LinkedIn members as having an impact on clinical practice or real-world outcomes research. By contrast, an example of a LOL identified through conventional bibliographic search would be a highly visible, prolifically published expert whose work falls short of meaningfully impacting clinical practice, physicians´ prescribing practices, or community conversations.
Of course, using SNA for KOL identification should be done discerningly because it has its own dark side: the presence of hyperactive social media commentators and self-proclaimed experts who “talk the talk” and have all the bearings of – and perhaps even the networks typically associated with – KOLs, but whose engagement with and impact on others remains at the level of metaphorical back-patting without translating into concrete actions (e.g. evangelizing new research ideas, driving market adoption of a new drug or treatment).
In the era of AI and Big Data, it is paramount for pharmaceutical and biotech companies of all sizes to get smarter about their KOL strategies and upgrade their tactics. If you have already leveraged SNA in your practice or are considering doing so after reading this blog, we invite you to share your experience in the comments or contact us directly with feedback. Here at PeakData we are always interested in learning how we can best help our clients and are constantly tweaking, fine-tuning, and adjusting our methodologies so we can be of service to you.
Artificial Intelligence and machine learning are making ever more sophisticated forays into healthcare and pharmaceutical R&D through applications ranging from processing medical images, unstructured EHR notes, and biosensor-equipped patient wearables to pattern-finding in synthetic control arms of clinical trials. However, beyond their usefulness in such controlled environments, little is known about how AI-powered data analytics can support pharmaceutical companies and biotech startups in their market access and commercialization efforts.
One scenario where AI´s role in pharmaceutical innovation is still poorly understood and under-utilized is in identifying global key opinion leaders (KOLs) such as physicians, hospital pharmacists, community healthcare providers, and others on the frontlines of patient care. These professionals are crucial in accelerating adoption of new drugs and treatments in local markets, as they can impact the presentation of research in professional forums, serve as phase III clinical trial investigators, or participate in the collection of real-world data through post-marketing surveillance.
Traditionally, addressing this need has been challenging for medical affairs managers, because identifying KOLs in an ocean of unstructured data floating in different formats and languages on the internet is a Herculean task. The conventional way to approach it is through consultancies that manually reach out and conduct interviews with potential leads – a high-touch, time-consuming process whose output is often limited when compared with the resources expended and the client´s outreach ambitions. The conventional way to approach this is by having medical science liaisons or other members of the medical affairs team identify, reach out to, and network with KOLs in an effort to understand how a drug in development compares with current treatments they are using with patients or how it may be improved to make a difference and create value.
AI offers a more efficient and less personal relationship-dependent approach to identifying KOL’s: by leveraging advanced capabilities, AI professionals can develop algorithms that comb through structured and unstructured data across the web and extract relevant information from millions of websites. Mapping potential KOLs in this way has less human bias because it is agnostic to consultants’ top-of-mind roster of contacts, casts a wider net than website rankings in search results can ever do, and is scalable to a degree that is simply not feasible when using human labour.
The “magic” making this novel approach possible is recent advancements in natural language processing (NLP). NLP is a well-established field of machine learning, which has experienced vast development of late. Its functionalities allow data technologists to acquire high-resolution, per-client data by creating scripts that match predefined criteria against publicly available information. In the context of identifying KOLs for , these criteria may include target country, practice area, specialty, services offered, level of care, and even academic affiliation. Whenever those criteria are reflected on the websites or other online materials of potential partners, the algorithm will detect it, extract the relevant details, and compile them into a client-specific database sorted by degree of relevance to the expressed criteria.
Another advantage of using AI to get a comprehensive list of KOLs who can provide guidance in the late stages of pharma and biotech companies’ R&D journey is the capability to capture up-to-date contact data and replace previous versions with it. This is important when considering that KOLs who are often most worth engaging with are the ones who ascend quickly in their careers and frequently move between institutions, change affiliations, participate in multiple projects, and give keynote addresses in locations around the world. As such, their contact details may overlap or become quickly outdated, which makes them harder to identify through conventional country- or institution-based search criteria.
NLP-powered algorithms can make sense of the labyrinth of publicly available KOL details on the web and convert them into an organized, updated resource thanks to their capability to monitor petabytes of messy web data in real time. These “super powers” allow software engineers to design scripts that send out alerts as information corresponding to clients´ KOL search criteria changes, assuring that medical affairs professionals are kept abreast of key details important to their outreach efforts.
The combined challenge of manually identifying a critical mass of KOLs and accounting for unpredictable changes in their contact details is compounded further by the different languages of the web domains where these data often reside. For marketers wanting to reach local KOLs and prescribers this can be yet another hindrance since deciding whether to engage with a particular person can depend on the context in which they identify themselves online. Multilingual NLP coupled with real-time data capture and updating that can dynamically convert to English (or another preferred language) the relevant information means the AI toolbox has a solution for this too.
Sustaining market adoption once a new drug or treatment has been approved by regulators means companies remain invested in maintaining relationships with KOLs who can speak to the effectiveness and value of their products throughout their life cycle. This is especially relevant in cases where the comparative effectiveness of therapeutic innovations – which takes into account subgroups of patients who benefit differentially from treatment with the same drug – surpasses that of established products, since physicians and other patient-facing professionals are in the best position to attest to the impact of novel drugs to heterogeneous patient populations.
For this reason, just as important as identifying currently recognized professionals is creating channels to map and connect with up-and-coming KOLs or “rising stars.” One way AI can help in this area is by automating client searches whose criteria may include progressively higher publishing frequency in scientific journals compared to the length of a young KOL´s career, the type of journals in which such publications appear, grants received, simultaneous or evolving expertise in more than one field, speaking engagements, and consulting gigs.
Regardless of the stage at which companies find themselves in their medical affairs endeavors, the opportunities afforded by proprietary AI and machine learning algorithms to dive into the ocean of publicly available big data on the web and come out with pearls of information tailored to the most demanding search criteria will increasingly mark the difference in outcomes of KOL identification and retention efforts. The sooner pharmaceutical and biotech organizations incorporate this approach into their existing strategies, the more rewarding their global and local network-building will be.