文献综述

---
name: lit-review
description: "When to activate: literature review, systematic review, meta-analysis, academic research, evidence synthesis, PRISMA, citation management, bibliographic search"
origin: zakr
metadata:
  namespace: zakr/devtools-pack
  version: 1.0.0
  category: business
  domain: research
  platforms: [claude-code, codex, gemini-cli, cursor, aider, windsurf]
---

# Literature Review Patterns

## Systematic Review Methodology

### PRISMA Checklist (Preferred Reporting Items)
- [ ] Define research question using PICO/PICOS framework
- [ ] Register protocol (PROSPERO, OSF) before data collection
- [ ] Document search strategy (databases, date ranges, keywords)
- [ ] Define inclusion/exclusion criteria a priori
- [ ] Screen titles/abstracts independently (2 reviewers)
- [ ] Full-text screening with reasons for exclusion
- [ ] Data extraction using standardized form
- [ ] Assess risk of bias for each study
- [ ] Synthesize results (narrative or meta-analytic)
- [ ] Report PRISMA flow diagram

### PICO Framework
| Element | Question |
|---------|----------|
| **P**opulation | Who are the participants/subjects? |
| **I**ntervention | What is being done/tested? |
| **C**omparison | What is the alternative/control? |
| **O**utcome | What are the measured results? |
| **S**tudy type | What research designs qualify? |

## Database Search Strategy

### Primary Databases by Domain
| Domain | Databases |
|--------|-----------|
| Medicine/Health | PubMed, Cochrane, Embase, CINAHL |
| Social Sciences | PsycINFO, Sociological Abstracts, SSRN |
| Business | Business Source Complete, ABI/INFORM, JSTOR |
| Engineering | IEEE Xplore, ACM Digital Library, Scopus |
| Multidisciplinary | Web of Science, Scopus, Google Scholar |
| Preprints | arXiv, bioRxiv, SSRN, OSF Preprints |

### Search String Construction
```
# Boolean operators
("machine learning" OR "deep learning" OR "neural network")
AND
("clinical decision" OR "medical diagnosis" OR "healthcare")
AND
("systematic review" OR "meta-analysis" OR "randomized controlled trial")

# MeSH terms (PubMed)
"Machine Learning"[MeSH] AND "Clinical Decision-Making"[MeSH]

# Proximity operators (Scopus)
TITLE-ABS-KEY("machine learning" W/5 "healthcare")
```

### Search Documentation Template
```
Database: PubMed
Date: 2024-01-15
Query: ("machine learning"[Title/Abstract]) AND ("diagnosis"[MeSH Terms])
Filters: Publication date 2019-2024, English
Results: 1,847 records
```

## Screening Process

### Two-Stage Screening
**Stage 1 — Title/Abstract:**
- Apply inclusion/exclusion criteria quickly
- Mark: Include / Exclude / Uncertain
- Resolve uncertain with second reviewer or team discussion

**Stage 2 — Full-text:**
- Review complete paper against criteria
- Record specific reason for each exclusion
- Track with Rayyan, Covidence, or Abstrackr

### Inclusion/Exclusion Criteria Template
```
INCLUSION:
- Peer-reviewed empirical studies
- Published 2015-2024
- Study population: adults 18+
- Reports quantitative outcome measures
- Available in English or Spanish

EXCLUSION:
- Conference abstracts without full papers
- Editorials, commentaries, letters
- Gray literature (unless specified)
- Duplicate publications
- Studies without control group
```

## Data Extraction

### Extraction Form Fields
```
Study ID:
Author(s):
Year:
Country:
Study design:
Sample size:
Population characteristics:
Intervention/exposure:
Comparison/control:
Outcome measures:
Follow-up duration:
Key findings:
Limitations noted by authors:
Risk of bias assessment:
Notes:
```

### Risk of Bias Tools
| Study Type | Tool |
|------------|------|
| RCTs | Cochrane RoB 2 |
| Observational | Newcastle-Ottawa Scale |
| Diagnostic | QUADAS-2 |
| Qualitative | CASP Checklist |
| Economic | Drummond Checklist |

## Evidence Synthesis

### Narrative Synthesis Structure
1. **Overview of included studies** — table with key characteristics
2. **Thematic grouping** — cluster by intervention type, outcome, population
3. **Direction of effect** — consistent vs contradictory findings
4. **Strength of evidence** — quality assessment summary
5. **Gaps identified** — what remains unanswered

### GRADE Evidence Quality
| Level | Description | Downgrade for |
|-------|-------------|---------------|
| High | Further research unlikely to change estimate | — |
| Moderate | Further research likely to impact estimate | Risk of bias, inconsistency |
| Low | Further research very likely to impact | Indirectness, imprecision |
| Very Low | Any estimate uncertain | Publication bias |

## Citation Management

### Recommended Tools
- **Zotero** — free, browser integration, group libraries
- **Mendeley** — free, collaboration features
- **EndNote** — paid, institutional license common
- **Rayyan** — screening-specific, free tier available

### Export Formats
```
RIS format:
TY  - JOUR
AU  - Smith, J.
TI  - Title of paper
JO  - Journal Name
VL  - 45
IS  - 2
SP  - 123
EP  - 134
PY  - 2023
DO  - 10.xxxx/xxxxx
ER  -
```

## Writing the Literature Review

### Structure Template
```
1. Introduction (10%)
   - Why this topic matters
   - Scope of the review
   - Review objectives

2. Methods (20%)
   - Search strategy
   - Inclusion/exclusion criteria
   - Data extraction procedure
   - Quality assessment approach

3. Results (40%)
   - PRISMA flow diagram
   - Study characteristics table
   - Thematic findings

4. Discussion (25%)
   - Synthesis of evidence
   - Contradictions explained
   - Gaps and future directions
   - Limitations of review

5. Conclusion (5%)
   - Key take-aways
   - Practical implications
```

### Common Pitfalls
- Confirmation bias — only searching to support hypothesis
- Language bias — excluding non-English studies
- Publication bias — only finding positive results
- Outdated search — not updating before submission
- Missing gray literature — guidelines, reports, theses