Literature Review

---
name: literature-review
description: Conduct systematic literature reviews — search strategy, screening, quality appraisal, synthesis, and gap identification. Use when mapping what a field knows on a question.
category: ai-research
---

# Literature Review

A literature review answers: what does the field know, how well does it know it, and what's 
missing. Done systematically, it's research in its own right — not a reading list.

## Overview

The method: define a focused question, design a reproducible search, screen studies against 
explicit criteria, appraise their quality, extract findings into a structured form, and synthesize 
— narratively or quantitatively. The review's value is in the synthesis: patterns across studies, 
contradictions explained, and gaps named. Every step is documented so another researcher could 
repeat it.

## When to use

- Starting a thesis, paper, or project: establishing what exists.
- Answering "what does the evidence say about X?" rigorously.
- Identifying research gaps worth pursuing.
- Writing the related-work or background section of a paper.

## Core concepts

- **Review question**: narrow and answerable — population, intervention/exposure, comparison, 
outcome. Vague questions produce unmanageable reviews.
- **Search strategy**: databases, keywords, synonyms, date ranges, documented exactly. Snowball 
from key papers' references and citations.
- **Screening**: title/abstract then full-text, against pre-defined inclusion/exclusion criteria. 
Two independent screeners when rigor matters; record exclusions.
- **Quality appraisal**: assess each study's methods — design, sample, measures, analysis, bias 
risks. Weight findings by quality, not by count.
- **Data extraction**: structured forms — study design, sample, methods, key findings, 
limitations. Consistency enables comparison.
- **Synthesis**: narrative (themes, patterns), tabular (comparison matrices), or quantitative 
(meta-analysis). End with: what's established, what's contested, what's missing.

## Practical workflow

1. Frame the question precisely; write inclusion/exclusion criteria before searching.
2. Run the documented search across databases; deduplicate; snowball key citations.
3. Screen in two passes (title/abstract, then full text); log every exclusion reason.
4. Appraise quality with a checklist appropriate to the study designs; extract findings into 
structured forms.
5. Synthesize: group by theme or method, compare findings, explain contradictions, assess the 
strength of evidence.
6. Write up: search flow, included studies table, synthesis, gaps, and implications — with the 
protocol documented for reproducibility.

```text
Review protocol (write before searching):
QUESTION:  <focused, answerable>
CRITERIA:  include <...> / exclude <...>
SOURCES:   <databases + search strings + date range>
SCREENING: <two-pass process, who screens>
APPRAISAL: <quality checklist per study design>
EXTRACTION:<fields captured per study>
SYNTHESIS: <narrative / tabular / meta-analytic plan>
```

## Common pitfalls

- **Question too broad**: "AI in healthcare" is a library, not a review. Narrow until the corpus is 
manageable.
- **Undocumented search**: a review nobody can reproduce. Record every string, database, and date.
- **Cherry-picking**: citing supportive studies, ignoring the rest. The criteria — written first 
— protect against this.
- **Vote counting**: "5 studies say yes, 3 say no" without weighing quality. Appraise before 
synthesizing.
- **No gap analysis**: summarizing without identifying what's missing. Gaps are the review's main 
contribution.
- **Stale by publication**: fields move fast. Note the search date; update before submitting.