Medical Search Strategy Builder

Build structured medical literature search strategies from natural language research questions using PICO/PEO frameworks. Generates database-specific search strings for PubMed, Embase, Cochrane Library, Web of Science, and arXiv.

When to Use

• User asks to construct search strings or build search strategies
• User mentions PICO, PEO, systematic review, or literature search
• User needs PubMed, Embase, Cochrane, Web of Science, or arXiv search syntax
• User wants to convert a research question into database queries
• User mentions MeSH terms, search strategy, or search string construction
• Input can be in Chinese or English

Workflow

Step 1: Analyze the Research Question

1. Identify the research question from the user's input
2. Detect the language (Chinese or English)
3. Determine the appropriate framework:
• PICO (Population, Intervention, Comparison, Outcome) → for interventional/therapeutic research
• PEO (Population, Exposure, Outcome) → for observational/etiological research

Framework selection keywords:

Step 2: Extract PICO/PEO Concepts

For each dimension, identify:

For each concept, collect three types of search terms:

1. MeSH/Subject Headings — standardized controlled vocabulary (highest precision)
2. Free-text terms — natural language synonyms (highest recall)
3. Truncated variants — wildcard forms using (e.g., diabet)

Step 3: Validate Concepts

Each concept MUST have:

• A non-empty name
• At least one search term (MeSH, free-text, or truncated)
• Ideally 3+ search terms total for adequate recall

Step 4: Generate Database-Specific Strategies

Use the syntax rules for each target database. See references/database-syntax.md for detailed syntax guides.

Quick reference:

Step 5: Format Output

Generate a structured Markdown report containing:

1. Research question analysis (original question, framework, research type)
2. PICO/PEO concept breakdown table
3. Search term table (MeSH, free-text, truncated terms per concept)
4. Database-specific search strategies (in code blocks)
5. Validation and optimization tips

Output Format

## Research Question Analysis
**Original question**: [user's question]
**Research type**: [interventional/observational]
**Framework**: [PICO/PEO]

### PICO/PEO Breakdown
| Dimension | Content |
|-----------|---------|
| P (Population) | [concept name]: [key terms] |
| I (Intervention) | [concept name]: [key terms] |
| C (Comparison) | [concept name]: [key terms] |
| O (Outcome) | [concept name]: [key terms] |

## Search Strategies

### PubMed
\```pubmed
("Diabetes Mellitus, Type 2"[MeSH] OR "type 2 diabetes"[Title/Abstract])
AND
("Metformin"[MeSH] OR metformin[Title/Abstract])
\```

### Embase
\```embase
#1 P: ('diabetes mellitus, type 2'/exp OR 'type 2 diabetes':ti,ab)
#2 I: ('metformin'/exp OR 'metformin':ti,ab)
#Final: #1 AND #2
\```

Common MeSH Terms Quick Reference

See references/common-mesh-terms.md for the full lookup table.

Important Notes

• MeSH terms improve precision; free-text terms improve recall. Always use both.
• Phrase searching: Multi-word terms should be quoted (e.g., "type 2 diabetes")
• Truncation: Use to capture word variants (e.g., diabet matches diabetes, diabetic, diabetology)
• Boolean operators must be UPPERCASE in all databases
• arXiv does not support MeSH — all terms are treated as free-text
• Embase uses Emtree (not MeSH) — terms may differ from PubMed
• For systematic reviews, consider adding methodology filters (e.g., "Randomized Controlled Trial as Topic"[MeSH])
• Always validate search strategies by checking result counts and relevance in the target database