parallel-literature-search

What this skill does

# Parallel Literature Search

**All sources at once.** This skill searches PubMed, web, and your RAG knowledge base in parallel, then synthesizes the findings into a single coherent summary with citations.

---

## WHAT IT DOES

| Source | What It Searches | Output |
|--------|------------------|--------|
| **PubMed** | Academic literature, trials, reviews | PMIDs, abstracts, citations |
| **Perplexity** | Web, recent news, guidelines | Summaries with sources |
| **RAG (AstraDB)** | Your curated knowledge base | Guideline excerpts, textbook refs |

---

## THE DIFFERENCE

| Approach | Sources | Time | Depth |
|----------|---------|------|-------|
| Sequential search | One at a time | 5+ min | Deeper but slow |
| **Parallel search** | All at once | 30-60 sec | Fast overview |
| Manual search | You do it | 20+ min | Variable |

---

## TRIGGERS

Use this skill when you say:
- "Search for evidence on [topic]"
- "What does the literature say about [topic]?"
- "Find research on [topic]"
- "Quick literature review on [topic]"
- "Evidence for [clinical question]"

---

## USAGE

### In Claude Code (Recommended)

```
"Parallel search: SGLT2 inhibitors in HFpEF"

"Find all evidence on GLP-1 and cardiovascular outcomes"

"What does literature say about statin discontinuation?"
```

### CLI Mode

```bash
# Basic search
python scripts/parallel_search.py --query "SGLT2 inhibitors heart failure"

# Specify sources
python scripts/parallel_search.py --query "GLP-1 cardiovascular" --sources pubmed,perplexity

# Save output
python scripts/parallel_search.py --query "CAC scoring" --output ~/research/
```

---

## OUTPUT FORMAT

```markdown
# Literature Search: SGLT2 Inhibitors in HFpEF

**Query:** SGLT2 inhibitors heart failure preserved ejection fraction
**Searched:** 2025-01-01 09:30:45
**Sources:** PubMed, Perplexity, RAG

---

## SYNTHESIS

SGLT2 inhibitors have demonstrated significant benefit in HFpEF based on
EMPEROR-Preserved and DELIVER trials. Key findings:

1. **EMPEROR-Preserved (PMID: 34449189)**: Empagliflozin reduced composite
   endpoint of CV death/HHF by 21% (HR 0.79, 95% CI 0.69-0.90)

2. **DELIVER (PMID: 36027570)**: Dapagliflozin showed 18% reduction in
   worsening HF/CV death (HR 0.82, 95% CI 0.73-0.92)

3. Current guidelines (ACC/AHA 2022) recommend SGLT2i as Class 2a for HFpEF.

---

## PUBMED RESULTS (5 most relevant)

| # | Title | PMID | Year | Type |
|---|-------|------|------|------|
| 1 | Empagliflozin in HFpEF | 34449189 | 2021 | RCT |
| 2 | Dapagliflozin in HFpEF | 36027570 | 2022 | RCT |
| 3 | Meta-analysis SGLT2i HF | 37654321 | 2023 | MA |
| 4 | Real-world SGLT2i outcomes | 38765432 | 2024 | Obs |
| 5 | SGLT2i mechanism review | 39876543 | 2024 | Rev |

---

## WEB RESULTS (Perplexity)

- **ACC 2024 Update**: New data on SGLT2i in cardiorenal syndrome
- **ESC Guidelines 2023**: Updated recommendations for SGLT2i
- **Clinical Practice**: Real-world prescribing patterns

---

## RAG RESULTS (Your Knowledge Base)

- **Braunwald Ch. 27**: Heart failure classification and treatment
- **ACC/AHA HF Guidelines**: Class recommendations for SGLT2i
- **ESC HF Guidelines**: European perspective on SGLT2i use

---

## EVIDENCE QUALITY

| Source | Strength | Notes |
|--------|----------|-------|
| EMPEROR-Preserved | High | Large RCT, well-conducted |
| DELIVER | High | Large RCT, confirmatory |
| Meta-analyses | High | Consistent findings |
| Real-world | Moderate | Observational limitations |

---

## KEY CITATIONS

1. Anker SD, et al. N Engl J Med. 2021;385:1451-1461. (PMID: 34449189)
2. Solomon SD, et al. N Engl J Med. 2022;387:1089-1098. (PMID: 36027570)
3. Vaduganathan M, et al. Lancet. 2022;400:757-767. (Meta-analysis)

---

## GAPS & CONSIDERATIONS

- Limited data in specific HFpEF phenotypes
- Long-term safety data still accumulating
- Indian-specific data limited (consider local studies)
```

---

## ARCHITECTURE

```
User Query
     │
     ├──────────────────┬──────────────────┐
     │                  │                  │
     ▼                  ▼                  ▼
[PubMed Agent]   [Perplexity Agent]  [RAG Agent]
     │                  │                  │
     ▼                  ▼                  ▼
  PMIDs &           Web sources       Guideline
  Abstracts         & summaries       excerpts
     │                  │                  │
     └──────────────────┴──────────────────┘
                        │
                        ▼
               [Synthesis Agent]
                        │
                        ▼
              Unified Report with
              Citations & Evidence
```

---

## INTEGRATION

### Works With:
- `quick-topic-researcher` - Quick overview
- `deep-researcher` - Comprehensive review
- `youtube-script-master` - Evidence for scripts
- `cardiology-editorial` - Literature for editorials

### Feeds Into:
- Content creation pipeline
- Video script research
- Editorial writing
- Newsletter content

---

## DEPENDENCIES

```python
# Core
anthropic>=0.18.0
python-dotenv>=1.0.0
rich>=13.0.0

# Already have these via your setup
# PubMed MCP - configured in .mcp.json
# Perplexity - via OpenRouter or MCP
```

---

## API KEYS NEEDED

| Key | Purpose | Status |
|-----|---------|--------|
| ANTHROPIC_API_KEY | Synthesis | Already have |
| NCBI_API_KEY | PubMed (via MCP) | Already have |
| PERPLEXITY_API_KEY | Web search | Already have |

---

## HOW CLAUDE SHOULD USE THIS SKILL

When user asks for literature/evidence:

### Step 1: Parse the Query
Extract:
- Main topic
- Specific aspects (population, intervention, outcome)
- Time frame (if mentioned)

### Step 2: Launch Parallel Searches

```python
# PubMed (via MCP)
pubmed_search_articles(queryTerm="SGLT2 inhibitors heart failure", maxResults=10)

# Perplexity (via MCP or API)
perplexity_ask(messages=[{"role": "user", "content": "Latest evidence on SGLT2 inhibitors in heart failure 2024"}])

# RAG (if available)
# Query AstraDB for relevant guidelines
```

### Step 3: Synthesize Results
Combine findings from all sources into:
- Key takeaways
- Evidence quality assessment
- Complete citation list
- Gaps and considerations

### Step 4: Format Output
Structured report with:
- Executive synthesis
- Source-by-source findings
- Full citations
- Actionable insights

---

## CLINICAL QUESTION OPTIMIZATION

The skill recognizes PICO format:

| Component | Example | How It's Used |
|-----------|---------|---------------|
| **P**atient | "elderly patients with HFpEF" | Filters PubMed |
| **I**ntervention | "SGLT2 inhibitors" | Primary search term |
| **C**omparison | "vs placebo" | Narrows to RCTs |
| **O**utcome | "mortality" | Focuses results |

---

## SAMPLE QUERIES

```
# Basic clinical question
"SGLT2 inhibitors in heart failure"

# PICO format
"In elderly patients with HFpEF, do SGLT2 inhibitors reduce mortality compared to placebo?"

# Specific trial
"What are the key findings from EMPEROR-Preserved?"

# Guideline-focused
"Current ACC/AHA recommendations for SGLT2i in heart failure"

# Comparative
"SGLT2i vs GLP-1 for cardiovascular outcomes in diabetes"
```

---

## NOTES

- **Speed**: Parallel search takes 30-60 seconds vs 5+ minutes sequential
- **Depth**: Good for overview, not exhaustive systematic review
- **Citations**: Always includes PMIDs for verification
- **Updates**: Perplexity provides most recent web data

---

*This skill gives you evidence from multiple sources in under a minute - perfect for content preparation and quick clinical questions.*