How to set up ai agents for eligibility pre screening cha…

• Gather your eligibility documents. The agent can only prescreen against what you give it. Pull together the trial protocol, the informed‑consent inclusion‑exclusion list, any screening checklists, and a plain‑language explanation of each criterion (age range, diagnosis, lab values, prior treatments, etc.). The more operational detail you include – like “history of severe adverse reaction to X,” not just “no allergies” – the sharper the agent’s questions will be.
• Own the content. Ensure you have the site’s institutional clearance to use protocol information in a patient‑facing chatbot. Redact any protected health information before uploading.
• Prepare for tone. Eligibility pre‑screening is a conversation, not a form. Decide how the agent should speak: a warm, empathetic nurse‑coordinator tone works better than a clinical‑trial bullet‑list. You’ll set that when you configure the agent.
• Know your hand‑off path. While Chatref’s AI agent handles initial eligibility conversations, have a clear plan for when a candidate needs to speak with a coordinator. Even without automated hand‑off, you can include a message that says “A study coordinator will follow up” for cases the agent cannot resolve.

1. Create your knowledge base

In your Chatref workspace, open Knowledge Base and add your eligibility documents. Chatref accepts PDFs, plain text, and URLs. For prescreening, we recommend:

• One file per trial, with the full protocol and a separate simplified “participant‑facing” version of the inclusion‑exclusion criteria.
• Avoid mixing documents for different trials in one knowledge base – keep the agent focused.

Chatref reads and indexes the content so answers stay grounded in your own documents, not general medical knowledge.

2. Build an AI agent

Go to Agents and create a new agent. Give it a name like “Diabetes Trial Pre‑screening”. Choose the knowledge base you just built. The agent will now answer only from those docs.

3. Write the agent’s instructions

This is where you turn a generic assistant into an eligibility pre‑screener. In the agent’s Instructions field, tell it exactly how to behave. A strong instruction set includes:

• Goal: “Your job is to determine if a visitor may be eligible for the ABCD‑123 diabetes trial by asking questions based on the trial’s inclusion‑exclusion criteria. Do not give medical advice or make a final eligibility determination.”
• Flow: “Start by asking why they reached out. If they want to be screened, ask one question at a time, covering age, diagnosis date, current medications, HbA1c range, prior treatments, and any exclusion criteria. Only move to the next question after the answer is clear.”
• Guardrails: “If a visitor reports a condition that is a clear exclusion, stop the screening and kindly say they do not appear to meet the criteria for this trial, and suggest speaking with their doctor. Do not ask further medical questions.”
• Escalation: “If you cannot determine eligibility or the visitor asks about enrollment, say ‘A study coordinator will contact you to discuss the next steps’ and collect their name and preferred contact method.”

Chatref’s agent follows these instructions while pulling precise criteria from the knowledge base – so a question like “What is the maximum HbA1c allowed?” will be answered from your own protocol, not from a web search.

4. Set the welcome message

In the agent’s Greeting, write the first thing a visitor sees. For a clinical trial site, something like:

“Interested in the ABCD‑123 diabetes trial? I can walk you through a few quick questions to see if you may be eligible. This is not a final determination and no information is stored.”

This sets expectations and discourages casual chat.

5. Deploy the widget

Back in your dashboard, grab the embed snippet from Widget. Add it to your trial‑specific landing page or your general patient portal. Restrict the origin to your domain so the widget only works on your site.

Use the Playground inside Chatref to test before publishing anywhere.

• Simulate a candidate who clearly meets criteria: start with “Is the diabetes trial still recruiting?” and see if the agent asks about age, diagnosis, etc., and confirms likely eligibility without overpromising.
• Simulate a clear exclusion: have the test conversation mention a recent heart attack that is an exclusion. The agent should stop screening and refer out.
• Test edge cases: ambiguous lab values (“my A1c was around 8”) or partial information. The agent should ask clarifying follow‑ups, not guess.
• Confirm grounding: ask the agent “What is the HbA1c cutoff?” and verify it pulls the exact number from your protocol document.

If it doesn’t follow your instructions, tighten the Instructions field – shorter, numbered directives often work best.

The agent doesn’t ask enough questions and jumps to a conclusion. Tweak the instructions to be explicit: “Ask no more than one question at a time. Wait for the visitor’s answer before continuing.” Also ensure the exclusion criteria in your knowledge base include clear statements like “excluded if” – not just a table.

The agent gives vague answers about eligibility. The underlying documents may lack operational detail. Add a separate plain‑language document that spells out each inclusion criterion in a conversational sentence, for example “You must be 18 to 75 years old to participate.” The agent mirrors that clarity.

Visitors ask about trials not covered by your knowledge base. The agent will answer only from what you uploaded. If you support multiple trials, create a separate knowledge base and agent for each, then route visitors on your website based on which trial they’re interested in.

The screening conversation feels cold or robotic. Adjust the agent’s instructions with tone guidance: “Respond in a warm, empathetic tone. Use the visitor’s first name if they provide it. Thank them for their patience.” But still keep the medical focus.

What causes eligibility pre screening chatbot problems for Clinical Trial Sites & CROs?

Most failures come from three areas: incomplete or ambiguous inclusion‑exclusion documents that leave the agent unsure which details matter; overly generic agent instructions that don’t force a step‑by‑step questioning flow; and lack of robust testing against real candidate scenarios before going live. When the knowledge base says “no serious cardiovascular disease” without defining “serious,” the agent cannot ask the right follow‑up.

How do I improve eligibility pre screening chatbot for Clinical Trial Sites & CROs?

Start by enriching your knowledge base with a dedicated plain‑language eligibility guide that translates the protocol into participant‑friendly sentences and adds explicit exclusion triggers (“do not proceed if”). Then refine the agent’s instructions to specify the exact order of questions, when to stop, and exactly what to say at each branching point. Finally, test with mock conversations until the agent reliably asks all critical questions and correctly stops on disqualifying answers.