Evidence Extraction Method

#Define the scope and identify each source

This is a proposed method, not a statement that an evidence review has been completed. Before extracting information, define the question, eligible study types, population, intended use and outcomes. Record how sources would be found and selected. Do not describe a collection as comprehensive unless the search and selection process support that claim.

For each source, capture its title, authors, publication date, publication type and stable identifier or direct link. Record the version and any identified correction or retraction. Link multiple reports from the same study so that participants and findings are not mistakenly counted as separate evidence.

#Capture findings with their context

Use a structured form covering design, setting, recruitment, sample size, participant characteristics and follow-up. Record the method evaluated, comparator or reference standard, outcomes and analysis approach. Note whether analyses were prespecified, whether evaluation data were separate from development and how missing observations or failed tests were handled.

Extract numerical findings with their units, denominators, time points and uncertainty intervals where reported. Preserve the distinction between adjusted and unadjusted estimates. Record important negative findings and harms, not only favorable results. Keep source statements separate from reviewer interpretation, and retain page, table or figure locations so each entry can be checked.

#Check accuracy and preserve uncertainty

The proposed process would use a tested extraction form and independent checking of key entries where feasible. Disagreements would be resolved with documented reasons. Mark unavailable information as not reported, rather than assuming that a procedure was not performed. Any calculated value should be labeled and its derivation recorded.

Assess limitations separately from extracting results, using criteria appropriate to the study design. Summaries should distinguish observed findings, author claims and reviewer judgments. Avoid pooling incompatible outcomes or populations without justification. Date the extraction, preserve amendments and state any limits in searching, access or checking that could affect the resulting account.

#Common misunderstandings

A structured extraction method can make evidence easier to inspect, but it does not automatically make the underlying research reliable. A completed record may still describe a small study, an uncertain finding or a result that applies only to a narrow group.

Extracting a finding is also different from endorsing it. The record should distinguish what researchers measured from how they interpreted the results. An association does not, by itself, establish that one factor caused another, and a statistically significant result is not necessarily large enough to matter in everyday care.

Missing information should not be treated as evidence that nothing happened. For example, a paper that does not report harms cannot establish that an intervention had none. Likewise, several papers may describe the same participants rather than independent studies. Recording these relationships helps prevent double-counting and keeps apparent agreement from looking stronger than it is.

#Questions worth asking a clinician

• How would this proposed extraction method preserve each source’s identity, study population, setting, and exact location of reported results?
• How would you distinguish findings reported by study authors from your interpretations and calculations derived from their data?
• How would you record missing study details and explain how those gaps limit interpretation of the findings?
• What checks would you use to catch extraction errors, and how would you document disagreements and their resolution?
• How would you distinguish planned extraction and checking procedures from completed work, and disclose limitations of any resulting review?