Build and Verify a Research Evidence Table With Elicit
Use an evidence table to compare studies with the same review question. Decide which population, outcome and time point each column should capture before you extract data across the full set.
Worked example
Sample evidence row for an employee onboarding study
Illustrative data only Enrolled = 96 Analyzed at 30 days = 87 Outcome = Time to independently complete the first task Time point = 30 days Reported median = 14 minutes across the 87 analyzed participants Outcome source = Table 2, page 6 Missing data status = Nine enrolled participants absent from the analysis
Extraction instruction for one column
Extract the number of participants whose first independent-task outcome was analyzed at 30 days. Give the analysis denominator and the page or table. If the paper provides only the number enrolled or a different time point, mark the requested value unavailable.
Steps
- Collect papers with their title, DOI and full text where you have access. Track which entries have abstract-only access because results details may be missing from the abstract.
- Create a separate column for each field. Specify the denominator and time point when they affect the meaning, such as employees analyzed for the month-one outcome.
- Pilot the columns on a few papers, including one with a missing outcome and one with results in a table. In Elicit's review workflow, open the supporting quote for each cell and find it in the paper.
- Export the checked rows with page or table references and an explicit status for unreported or unavailable values. Keep different measures and time points in separate columns.
What to check
- Cross-check group totals against each table's header and footnotes. A per-arm sample size can be mistaken for the whole study's sample size.
- Read the context around quoted evidence. Check that the table and column refer to the requested outcome and time point.
- Keep a note when you correct a value so a second reviewer can retrace the change.