Designing and Implementing a Data Extraction Form for Systematic Reviews

Introduction

Systematic reviews synthesize evidence across studies to address a focused clinical or research question. However, their reliability hinges on a seemingly simple yet critical step: data extraction. Extracting data from included studies involves far more than copying values—it is a methodologically rigorous process that demands structure, consistency, and foresight.

This article outlines how to create and implement a data extraction form tailored for systematic reviews, highlighting essential planning steps, extraction tools, and content structure.

Planning the Data Extraction Process

Effective data extraction starts with a detailed plan that addresses the who, how, and what of the process.

Who Will Perform the Extraction?

Knowledge requirements: Extractors must have at least basic familiarity with the clinical topic and the review’s objectives.
Double extraction: Ideally, two independent reviewers extract data from each study to minimize bias and detect errors.
Discrepancy resolution: A third-party arbitrator, such as a principal investigator, should be appointed in advance to resolve conflicts in extracted data.

What Is the Operational Plan?

A clear operational workflow is critical. The plan should include:

Standardized extraction forms: These ensure all extractors collect the same data elements using consistent definitions.
A central database: Digital datasheets or structured data entry systems allow for aggregation and analysis.
Training and calibration: Reviewers must be oriented to the form, variable definitions, and extraction rules.
Quality assurance: Build in processes for validation and cross-checking between reviewers.

Example: Before starting extraction for a review on statins and cardiovascular outcomes, a calibration exercise can be conducted using two practice studies to refine outcome definitions and resolve interpretation discrepancies.

Choosing the Right Extraction Format

Manual Methods

Paper forms: Simple but prone to transcription errors and inefficiencies. Generally discouraged except for low-volume reviews.

Digital Tools

Spreadsheets and Forms:

Google Forms: Good for centralized input; links directly to Google Sheets.
MS Excel: Offers flexibility and formulas but requires structured templates.
MS Access: Supports relational databases and custom queries for more advanced users.

Systematic Review Software:

Covidence: Designed for Cochrane workflows; includes built-in screening and extraction functions.
EPPI-Reviewer: Suitable for complex reviews with logic-based form customization.
Systematic Review Data Repository (SRDR): NIH-supported platform offering standardized modules and export functions.

Each platform has trade-offs between customizability, automation, and ease of use. Choose based on team size, review complexity, and available resources.

Structuring the Extraction Content

What should be extracted depends on the review type, but core categories generally include:

1. Extractor Metadata

Reviewer ID or initials
Date of extraction
Notes or flags for verification

2. Study Characteristics

Design type: RCT, cohort, case-control, etc.
Eligibility Criteria: Inclusion and Exclusion Thresholds.
Recruitment setting: Hospital, community, registry, etc.
Geographic region and timeline: Country, year, and duration of data collection.

3. Population Definitions

Description of participants
Inclusion/exclusion criteria applied to the population
Sample size (overall and per group)

4. Intervention or Exposure Details

Type, dosage, frequency, duration
Comparator information (e.g., placebo, active control, standard care)

5. Outcomes

Name and definition of each outcome
Time point of assessment (e.g., 3 months, 1 year)
Measurement tools or scales used (e.g., SF-36, HbA1c, hazard ratios)

6. Statistical Analysis Descriptors

Primary vs. secondary analyses
Adjusted vs. unadjusted estimates
Intention-to-treat vs. per-protocol frameworks

7. Study Results

Numerical data: Event rates, means/SDs, medians/IQRs.
Comparative results: Risk ratios, odds ratios, hazard ratios, mean differences, etc.
Confidence intervals and p-values for effect estimates
Baseline characteristics (e.g., age, sex, comorbidities) to assess clinical comparability between studies

Example: In a meta-analysis of physical therapy interventions for low back pain, outcomes might include pain intensity (on a 0–100 scale), functional status (Roland-Morris Disability Questionnaire), and adverse events.

Conclusion

A well-constructed data extraction form is the backbone of a reliable systematic review. It translates clinical questions into structured variables, ensures uniformity across studies, and sets the stage for robust synthesis. When thoughtfully designed and rigorously applied, it protects against data distortion and enhances the transparency, reproducibility, and credibility of review findings.

Key Takeaways

Plan for at least two extractors and a third arbitrator to resolve discrepancies.
Use electronic tools—paper-based extraction is obsolete and error-prone.
Tailor your form content to the review objectives, including study features, population, interventions, outcomes, and results.
Ensure training and calibration before extraction begins.
Document decisions and maintain traceability, especially for quantitative synthesis.