THE MSL

Journal of the MSL Society

Improving the Quality of Scientific Engagement With Key Experts Using AI


Author(s): and

Claims about AI are everywhere at the moment. LinkedIn seems to have been taken over by it. Co-pilot, Open AI, Claude, and numerous in-app AI options feel like the equivalent of the “talking paperclip” offering help that often does not live up to expectations. So, how do we separate the hype from the tools that can really support time management and, importantly, credibility as an MSL?

The utility of AI for what MSLs do every day is enticing:

Summarization: To support staying up to date with the latest scientific data and clinical practice trends

Connectivity: To keep pace with what HCPs and key experts are interested in,  explore emerging trends, and discover new experts and networks to collaborate with

Efficiency:  To streamline day-to-day activity flows and find efficiencies

Clarity of communication: To enhance communication with internal and external stakeholders through tailored content aimed at individual levels of understanding and focus.

Here, we would like to focus on the first two of those areas of AI application, and how they can support trust-filled MSL / HCP interactions as described by Maister’s “trust equation”(1):

Trustworthiness = ( Credibility plus Reliability plus Intimacy ) divided by Self-Orientation 

Taking time to top-up personal scientific and clinical practice knowledge is a considerable investment. Traditionally, this may have been setting alerts on areas of interest within publication sites and company knowledge databases, and then receiving regular emails with lists of references to click on and read. Typically,  these alerts quickly pile up unread, prompting a feeling of guilt about slipping behind the evidence base and even when time is found, it is usually limited to just reading the abstract or conclusions section. MSLs often ensure that they know the latest treatment guidelines for their therapeutic area by keeping notes on practices and data that support or contradict as they come up, but it is common to find that a personal note repository becoming difficult to navigate (even if it feels reassuringly complete). Not having enough time to dedicate to this part of the role can erode an MSL’s ability to demonstrate credibility and, in some extreme cases, make MSLs avoid opportunities for scientific exchange as they doubt the completeness and recency of their own knowledge.

Untrained or poorly constrained AI chatbots frequently produce citations that are not just irrelevant, but entirely fabricated, complete with convincingly realistic metadata that risks misleading even experienced readers (3,4). One evaluation found that over 60% of references lacked genuine relevance, and nearly half contained false DOIs, author names, or journal titles, exposing a serious vulnerability in the reliability of AI-generated citations, particularly in clinical and academic contexts (2,5).

These findings are unacceptable when patient safety relies on accurate and fully referenced information.  So although these strategies feel like an incremental step forward in this area, often they require extra time to verify data sources and in some Large Language Model (LLM) answers this is not even possible as source data is not easily identified.

The solution in this case is twofold:

  • Isolate the types of data AI extracts from, to exclude unverified sources like non-expert opinion and unreferenced content 
  • Check IA provider disclosures for how LLM has been trained (including how it checks for and minimizes hallucination) before using it for this specialized summary type. 

Some Pharma and Biotech companies are pulling LLMs into their secure company environments and specifying the data sets for the first part, but then finding that the training of the AI is a complex and ongoing commitment that goes beyond just feeding the AI the right data. It’s an intensive, continuous ‘teaching’ process whereby the AI iteratively refines its contextual understanding, chain-of-thought reasoning, and output generation, guided by expert human-in-the-loop (HITL) feedback to ensure accuracy and appropriateness.

When both these things are addressed correctly, these AI platforms can host not only peer-reviewed data curated to the MSLs’ area of interest but also other trusted data sources, including internal Medical Affairs documents and gathered “insights”. This allows the MSL to query across all data sources and prepare for upcoming conversations focusing on their own knowledge gaps and those of their HCPs for just-in-time preparation which improves scientific reliability Minimizing extra validation steps compared to generic LLMs(6).

Until such AI platforms become widely available, a top tip for MSLs using generic LLMs for research is to engineer prompts to suggest relevant, reputable public data sources and request hyperlinked references, always critically verifying all outputs. For non-public data, MSLs must exclusively use secure, company-approved internal tools within the firewall. Never input confidential company data into public AI tools.

Using AI to enhance connectivity involves being able to spot patterns of activity from individual HCPs. The information available online is only one aspect. Understanding the needs of a HCP and where they are already very knowledgeable might include accessing what they publish (peer reviewed and online commentary),  what they say on a podium at a congress vs reveal 1:1 in conversations with Medical Affairs (MSL 1:1 or in groups / advisory settings) and how they request information through med info.  The ever-changing nature of individual experiences makes staying current on this to maintain professional levels of intimacy an ideal application of AI. Layers of data security to protect privileged information that an HCP shares with the MSL and their company is paramount here, as is the appreciation of respectful documentation that is subject to audit and freedom of information requests if stored against the identifiable individual. The linkage of internal data sources is becoming increasingly common practice, and again, medically-trained AI is vital to summarizing insight accurately to avoid “putting words into the mouths of others”.

Ultimately, those who use AI well to understand their HCPs will rise above the harmful effects of self-orientation to instead be able to engage confidently on the HCP (and therefore patient) need agenda. AI accessing these types of data can easily surface new names of individuals based on their interests, who MSLs can then connect into the network of HCPs and Experts they already work with, to increase diversity of thought and practice in their therapeutic area.

A final thought on why this matters to the individual MSL. As MSL-gathered “insight” becomes a part of the data repository that AI draws from, MSLs that are users of AI will have the deepest appreciation of what great source data needs to contain. As a result, they will see their own collected insight being referenced in strategic decision-making like never before. One of the most satisfying parts of the MSL role is being able to make a difference. The future of AI in this space is only truly unlocked by humans with the relationships to take action on the information it summarizes, and those humans are often MSLs.

References: 

  1. Maister, David H., et al. The Trusted Advisor. Simon & Schuster, 2001.
  2. Aljamaan, F., Temsah, M.-H., Altamimi, I., Al-Eyadhy, A., Jamal, A., Alhasan, K., Mesallam, T.A., Farahat, M. and Malki, K.H., 2024. Reference hallucination score for medical artificial intelligence chatbots: development and usability study. JMIR Medical Informatics, 12, p.e54345. https://doi.org/10.2196/54345
  3. Athaluri, S.A., Manthena, S.V., Kesapragada, V.S.R.K.M., Yarlagadda, V., Dave, T. and Duddumpudi, R.T.S., 2023. Exploring the boundaries of reality: Investigating the phenomenon of artificial intelligence hallucination in scientific writing through ChatGPT references. Cureus, 15(4), p.e37432. https://doi.org/10.7759/cureus.37432
  4. Eysenbach, G., 2023. The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers. JMIR Medical Education, 9, p.e46885. https://doi.org/10.2196/46885
  5. Templin, T., Perez, M.W., Sylvia, S., Leek, J., and Sinnott-Armstrong, N., 2024. Addressing six challenges in generative AI for digital health: a scoping review. PLOS Digital Health, 3(5), p.e0000503. https://doi.org/10.1371/journal.pdig.0000503
  6. 2024 White Paper: A Comprehensive Guide to Responsible AI for Life Sciences. Request from: https://www.sorcero.com/a-comprehensive-guide-to-responsible-ai-for-life-sciences

 

About the author:

Vic Ho

Vic Ho has been a Medical Affairs professional for over 20 years, specialising in field medical excellence and capabilities. Vic is passionate about advancing patient-focused strategy, medical impact measurement, and using technology to allow MSLs to be the conduit for patient insight in our industry.

 

Dr Alex Sterling

Alex Sterling, FRSM, MRSB, MBBCH is a SciComms Solutions lead with a background as a critical care physician, medical advisor, and medical scientist, he has extensive experience in applying data-driven approaches and machine learning to clinical research, particularly in areas like respiratory physiology.

Have an opinion on this article? Send a message to the editor.