Session 2 • 30 minutes

Prompt Engineering for Research

Master the art of communicating with AI to get reliable, useful results for your surgical research

Be Specific

Vague: "Tell me about heart surgery complications"

Specific: "List the 5 most common complications following CABG in patients over 70, with incidence rates and prevention strategies"

Provide Context

Tell AI your role, audience, and purpose

"I'm a CT surgeon preparing an NIH grant. Help me write a compelling specific aims page for minimally invasive mitral valve repair..."

Iterate & Refine

Don't expect perfection on the first try

Get initial output → "Make it more concise" → "Add citations" → "Rewrite intro to be more compelling"

AATS-Specific Use Cases
Real prompts surgeons use every day

📚Literature Review

"I'm conducting a systematic review on AI applications in cardiac surgery. Please:
1. Identify key research questions in this field
2. Summarize major findings from the past 5 years
3. Highlight controversies or knowledge gaps
4. Suggest directions for future research

Focus on high-impact journals and well-designed studies."

Result: Structured literature synthesis in minutes instead of days

💰Grant Writing

"Help me write a compelling significance section for an NIH R01 grant investigating [RESEARCH QUESTION].

Include:
- Current state of the field
- Critical knowledge gaps
- Why this research matters now
- Potential impact on patient care

Write for a multidisciplinary review panel. Use clear language while maintaining scientific rigor."

Result: Draft significance section ready for refinement

📋Patient Education

"Generate a patient-friendly explanation of [TAVR procedure] for a consent form. The explanation should:
- Use 8th-grade reading level
- Explain risks and benefits clearly
- Address common patient concerns
- Be approximately 300 words"

Result: Clear, accessible patient materials that improve informed consent

Try It Yourself: AI Playground
Practice prompt engineering with real AI models. Compare GPT-4, Claude, and Gemini side-by-side.

Suggested Exercise:

  1. Start with a vague prompt: "Tell me about CABG surgery"
  2. Refine it: "Summarize current evidence on off-pump vs. on-pump CABG for elderly patients with multiple comorbidities"
  3. Add context: "I'm preparing a case presentation for grand rounds. Focus on decision-making criteria."
  4. Compare how different models respond to the same prompt
Advanced Techniques

🔗 Chain-of-Thought

Ask AI to show its reasoning step-by-step

"Explain your reasoning step-by-step: What are the most important factors when selecting patients for minimally invasive cardiac surgery?"

🎭 Role-Playing

Assign AI a specific expert role

"You are a senior CT surgeon with 30 years of experience in quality improvement. Review this surgical protocol and identify safety concerns."

📝 Few-Shot Learning

Show AI examples of what you want

"Summarize papers in this format: Study: [authors], Methods: [design], Key Finding: [result], Limitation: [issue]. Now summarize: [paste abstract]"

🔄 Iterative Refinement

Improve output through multiple rounds

Get initial → "Make it more concise" → "Add citations" → "Rewrite for lay audience"
Key Takeaways
✓

Specificity matters: More context and detail = better output

✓

Iterate and refine: First drafts are starting points, not final products

✓

Maintain expertise: AI assists your judgment; it doesn't replace it

✓

Verify everything: Check facts, citations, and clinical claims