Before meeting with a physician, a Medical Science Liaison (MSL) needs to know the relevant evidence: the key studies, the strongest data points, and a way to discuss risk on the same terms the clinician will. Done the traditional way, that preparation can take at least half a day, and it often means waiting on a clinical or regulatory team to surface the validated evidence.
In our recent webinar, DistillerSR + Claude: A Practical Playbook for Medical Affairs and Commercial Teams, I walked through how an MSL can do that prep themselves, in a matter of minutes, using DistillerSR’s Agentic AI to query a validated repository of literature evidence in plain language through the integration with Claude.
Here is the step-by-step walkthrough.
The scenario
The MSL is preparing for an on-site meeting with a physician at a major academic center, focused on Cardiac Implantable Electronic Devices (CIEDs). The goal for the session is to find relevant evidence on the treatment in question, understand the key findings, define risk measurement, and produce a safety dossier.
Everything that follows runs against a DistillerSR project, queried conversationally. The evidence that the AI is basing its answers on has already been reviewed and validated by a team of experts from the MSL’s organization. It is not searching the open web, and it is not just relying on model memory.
Step 1: Choose the project to query
The MSL starts by pointing the integration at the right project, in this case one called Cardiac Stent CER Update.
If you do not know which project to begin with, you can start even broader. You can ask to see all of your DistillerSR projects, or all of your projects on a specific topic, and drill down from there. The point is that you begin at a high level and keep narrowing, rather than needing to know exactly where everything lives before you start.
Step 2: Narrow to high-confidence evidence
With the right project selected, the next move is to filter to the most reliable studies. The MSL asks to see all the references in the project flagged as randomized controlled trials (RCTs).
Starting from RCTs means starting from a high confidence level. The data has not only been validated by the team, but the studies themselves are considered high quality, which means more reliable results to build on. In the demonstration, this query returned around 1,700 references flagged as RCTs in the cardiac stent project.
Step 3: Refine
While 1,700 references is an accurate count, it is far more than any person can work through rapidly. So the MSL asks a follow-up question to narrow the set further, this time to the references specifically about CIEDs.
It is important to note that the MSL is not writing query logic. They are asking questions in natural language, getting results, and refining from there. The system carries the context from one question to the next, so each follow-up builds on the last. And because the underlying data is structured in a tabular format, it is easy for the AI to work through and return more specific, pointed answers.
Step 4: Surface the talking points with evidence-backed citations
Now the MSL gets to the evidence that will actually anchor the conversation. The query surfaces a specific study covering CIEDs with a large patient population, just over 5,000 patients, focused on the device in question, showing a reduction in adverse events of roughly 25 percent.
Here, the result is drawn from validated evidence, flagged with the right context, and presented as a summary the MSL can dig into and verify. The first objective, finding and summarizing relevant evidence on the specific device, is met. That is a concrete, evidence-backed talking point, and it links directly to the reference behind it. This is the difference between querying something like all of PubMed and getting a list of papers that might be relevant.
Step 5: A shared, validated framework to measure risk
The next objective is accurately measuring risk. The MSL asks how many references in the same project include a cohort with infection risk.
This is a second stage of the same line of inquiry. At this point, the MSL has an answer to the first question, has collected some data, and now keeps going. Because the context flows, the system favors references related to CIEDs as it looks for studies that also include an infection-risk cohort.
One of the first results includes a PADIT (Prior procedure, Age, Depressed renal function, Immunocompromised, Type of procedure) score, a standardized measure of infection risk. That stood out as a strong speaking point, because it gives the MSL and the physician a shared, validated framework to assess risk rather than approaching it from different angles. By summarizing the infected-cohort studies, the MSL can also confirm details like the size and design of each study, building trust in the data rather than taking the output at face value.
Step 6: Generate the safety dossier
With the evidence and the risk framework in hand, the MSL pulls it all together. The prompt asks to create a safety dossier on the extracted data related to CIEDs that includes an infection-risk cohort.
DistillerSR’s Agentic AI goes into the project, finds the relevant references, and generates a safety dossier. Because the dossier is rooted in human-validated data and directly cites its references, it is reliable from the start. It accomplishes in minutes what typically would take hours, while ensuring that every point is fully traceable to its source.
Summary
In the space of a few prompts, the MSL went from a project to a meeting-ready, fully cited safety dossier. Along the way they found the evidence that mattered, linked it back to specific validated studies rather than a general database, and established a shared way to measure risk with the physician.
Key drivers for this workflow’s success:
- The AI was grounded in literature evidence that was extracted and validated by experts from the MSL’s organization.
- The process was a natural language conversation, not a technical query. The MSL could explore, ask follow-ups, and refine in plain language, without specialized skills or query syntax. Each question built on the last because the system carried the context through.
- The human remained the critical thinker. The AI surfaced and summarized, but the MSL confirmed the study quality and the data before relying on it.
- Human experts were not the bottleneck. The MSL queried a repository of validated literature evidence directly without adding to the workload of the people who build and maintain it.
A Living Evidence Base
The meeting does not have to be the end of the loop. Follow-up points from the conversation with the physician, new studies to look into, or areas to prioritize, can be fed back into the DistillerSR project or routed to the team, so the evidence base keeps improving.
This is what evidence management looks like in day-to-day practice. Validated evidence the whole organization can leverage, with every answer traceable to its source.
