Data Extraction Forms for Systematic Reviews: A Practical Guide

A data extraction form is the structured document used to systematically pull the same categories of information from every study included in a systematic review. Its design matters more than reviewers new to the process typically expect -- a poorly planned form is the single most common reason a review team ends up re-reading every included study a second time, well after screening is finished and time pressure has increased.

Why extraction forms need to be built before extraction begins

Building your extraction form after you've already started reading full-text studies invites exactly the problem the form exists to prevent: inconsistent capture, where the first several studies get extracted differently than the later ones because your sense of what matters evolved partway through. The form should be piloted on a small handful of included studies before full extraction begins, specifically to catch fields that are missing, ambiguously worded, or unnecessary before you commit to extracting dozens or hundreds of studies against it.

Core fields every extraction form needs

Study identification fields -- author, year, country, funding source, and a unique study ID -- anchor every other data point back to its source and are essential for your PRISMA flow diagram and any later audit of your extraction. Population characteristics need enough granularity to support any planned subgroup analysis; if you intend to examine effects by age group or disease severity later, that data needs to be captured now, not inferred retroactively from a study's discussion section.

Intervention and comparator details should be extracted with enough specificity to judge clinical or methodological comparability across studies later -- dose, duration, delivery method, and comparator type, not just an intervention's generic name, since two studies both testing "cognitive behavioral therapy" can differ meaningfully in session count, delivery format, and therapist qualification in ways that matter for interpreting pooled results.

Outcome data needs to be extracted in a form usable for your planned synthesis method -- means and standard deviations for continuous outcomes intended for a mean-difference meta-analysis, or event counts and total sample sizes per arm for dichotomous outcomes intended for a risk-ratio meta-analysis. Extracting only a reported p-value or a vague "significant improvement" statement, without the underlying numeric data, is a common and costly gap discovered only once statistical analysis begins.

Fields that are easy to forget

Follow-up duration and timing of outcome measurement are frequently under-captured, even though they matter enormously for interpreting and comparing effect sizes across studies with different measurement windows. Study design specifics beyond a simple "RCT" or "cohort" label -- randomization method, blinding, allocation concealment -- feed directly into your risk-of-bias assessment and should be captured on the same form rather than as a separate pass through the same studies. Conflicts of interest and funding source, while sometimes overlooked, are increasingly expected fields given their relevance to risk-of-bias and GRADE assessment of potential reporting bias.

Structuring the form for dual extraction

If you are running dual, independent extraction -- generally expected for the same reasons dual screening is expected -- your form should support a clean comparison step between two extractors' entries, with a defined process for resolving discrepancies. Spreadsheet-based forms with one row per study and one column per field make this comparison straightforward; free-text extraction into a document makes discrepancy comparison considerably harder and more error-prone.

Software options worth knowing

Dedicated systematic review platforms including Covidence and DistillerSR support structured extraction forms directly within the same environment used for screening, and both allow dual extraction with built-in comparison and conflict-resolution features. For teams without access to dedicated software, a well-structured spreadsheet with data validation rules -- restricting entries to specific formats or drop-down options where appropriate -- achieves much of the same consistency benefit at lower cost, though without the automated comparison features.

A quick pre-extraction checklist

Before extraction begins in earnest, confirm your form captures: study identification, population characteristics needed for any planned subgroups, intervention and comparator specifics beyond generic labels, outcome data in the exact numeric form your synthesis method requires, follow-up timing, risk-of-bias-relevant design details, and conflict of interest information. Pilot the form on three to five included studies first, and revise before proceeding to the full set -- this single step prevents the far more expensive problem of discovering a missing field after extraction is otherwise complete.

Version control and audit trail

As your extraction form inevitably evolves during piloting, maintain clear version control -- a dated log of what changed and why, and confirmation of whether earlier-extracted studies were re-checked against later field additions. This matters directly for reproducibility: if a reviewer or committee member asks why one included study's entry looks different in structure from another's, a version log lets you answer precisely rather than reconstructing the history from memory months after extraction was completed.

Handling studies that report data in incompatible units or formats

A frequent, easy-to-underestimate extraction challenge is studies reporting the same outcome in genuinely different units or formats -- one trial reporting a continuous pain score on a 0-10 scale, another using a 0-100 scale, a third reporting only a categorical improvement/no-improvement outcome. Your extraction form should include a field capturing the exact unit or scale used per study, not just the numeric value, since this information is essential later when deciding whether and how these outcomes can be meaningfully combined, standardized, or whether they instead require separate, non-pooled reporting. Missing this at extraction time is one of the more common and most time-costly reasons a review team ends up returning to full-text sources a second time, well after the original extraction pass was believed complete and screening staff have already moved on to other, unrelated project work, making the second pass considerably slower, more expensive in reviewer time, and noticeably more disruptive to project momentum than it would have been if caught during the original extraction pass instead, which is exactly why the piloting step described above earns back far more time than it costs upfront.